diff --git a/llvm/lib/Target/AMDGPU/AMDGPULowerBufferFatPointers.cpp b/llvm/lib/Target/AMDGPU/AMDGPULowerBufferFatPointers.cpp
index 86b2c4f78fc3e..608b43b59eed3 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULowerBufferFatPointers.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULowerBufferFatPointers.cpp
@@ -45,16 +45,16 @@
 //
 // This pass proceeds in three main phases:
 //
-// ## Rewriting loads and stores of p7
+// ## Rewriting loads and stores of p7 and memcpy()-like handling
 //
 // The first phase is to rewrite away all loads and stors of `ptr addrspace(7)`,
 // including aggregates containing such pointers, to ones that use `i160`. This
-// is handled by `StoreFatPtrsAsIntsVisitor` , which visits loads, stores, and
-// allocas and, if the loaded or stored type contains `ptr addrspace(7)`,
-// rewrites that type to one where the p7s are replaced by i160s, copying other
-// parts of aggregates as needed. In the case of a store, each pointer is
-// `ptrtoint`d to i160 before storing, and load integers are `inttoptr`d back.
-// This same transformation is applied to vectors of pointers.
+// is handled by `StoreFatPtrsAsIntsAndExpandMemcpyVisitor` , which visits
+// loads, stores, and allocas and, if the loaded or stored type contains `ptr
+// addrspace(7)`, rewrites that type to one where the p7s are replaced by i160s,
+// copying other parts of aggregates as needed. In the case of a store, each
+// pointer is `ptrtoint`d to i160 before storing, and load integers are
+// `inttoptr`d back. This same transformation is applied to vectors of pointers.
 //
 // Such a transformation allows the later phases of the pass to not need
 // to handle buffer fat pointers moving to and from memory, where we load
@@ -66,6 +66,10 @@
 // Atomics operations on `ptr addrspace(7)` values are not suppported, as the
 // hardware does not include a 160-bit atomic.
 //
+// In order to save on O(N) work and to ensure that the contents type
+// legalizer correctly splits up wide loads, also unconditionally lower
+// memcpy-like intrinsics into loops here.
+//
 // ## Buffer contents type legalization
 //
 // The underlying buffer intrinsics only support types up to 128 bits long,
@@ -231,20 +235,24 @@
 #include "llvm/IR/InstIterator.h"
 #include "llvm/IR/InstVisitor.h"
 #include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/IntrinsicsAMDGPU.h"
 #include "llvm/IR/Metadata.h"
 #include "llvm/IR/Operator.h"
 #include "llvm/IR/PatternMatch.h"
 #include "llvm/IR/ReplaceConstant.h"
+#include "llvm/IR/ValueHandle.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Pass.h"
+#include "llvm/Support/AMDGPUAddrSpace.h"
 #include "llvm/Support/Alignment.h"
 #include "llvm/Support/AtomicOrdering.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Transforms/Utils/Cloning.h"
 #include "llvm/Transforms/Utils/Local.h"
+#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
 #include "llvm/Transforms/Utils/ValueMapper.h"
 
 #define DEBUG_TYPE "amdgpu-lower-buffer-fat-pointers"
@@ -431,14 +439,16 @@ namespace {
 /// marshalling costs when reading or storing these values, but since placing
 /// such pointers into memory is an uncommon operation at best, we feel that
 /// this cost is acceptable for better performance in the common case.
-class StoreFatPtrsAsIntsVisitor
-    : public InstVisitor<StoreFatPtrsAsIntsVisitor, bool> {
+class StoreFatPtrsAsIntsAndExpandMemcpyVisitor
+    : public InstVisitor<StoreFatPtrsAsIntsAndExpandMemcpyVisitor, bool> {
   BufferFatPtrToIntTypeMap *TypeMap;
 
   ValueToValueMapTy ConvertedForStore;
 
   IRBuilder<> IRB;
 
+  const TargetMachine *TM;
+
   // Convert all the buffer fat pointers within the input value to inttegers
   // so that it can be stored in memory.
   Value *fatPtrsToInts(Value *V, Type *From, Type *To, const Twine &Name);
@@ -448,8 +458,10 @@ class StoreFatPtrsAsIntsVisitor
   Value *intsToFatPtrs(Value *V, Type *From, Type *To, const Twine &Name);
 
 public:
-  StoreFatPtrsAsIntsVisitor(BufferFatPtrToIntTypeMap *TypeMap, LLVMContext &Ctx)
-      : TypeMap(TypeMap), IRB(Ctx) {}
+  StoreFatPtrsAsIntsAndExpandMemcpyVisitor(BufferFatPtrToIntTypeMap *TypeMap,
+                                           LLVMContext &Ctx,
+                                           const TargetMachine *TM)
+      : TypeMap(TypeMap), IRB(Ctx), TM(TM) {}
   bool processFunction(Function &F);
 
   bool visitInstruction(Instruction &I) { return false; }
@@ -457,11 +469,16 @@ class StoreFatPtrsAsIntsVisitor
   bool visitLoadInst(LoadInst &LI);
   bool visitStoreInst(StoreInst &SI);
   bool visitGetElementPtrInst(GetElementPtrInst &I);
+
+  bool visitMemCpyInst(MemCpyInst &MCI);
+  bool visitMemMoveInst(MemMoveInst &MMI);
+  bool visitMemSetInst(MemSetInst &MSI);
+  bool visitMemSetPatternInst(MemSetPatternInst &MSPI);
 };
 } // namespace
 
-Value *StoreFatPtrsAsIntsVisitor::fatPtrsToInts(Value *V, Type *From, Type *To,
-                                                const Twine &Name) {
+Value *StoreFatPtrsAsIntsAndExpandMemcpyVisitor::fatPtrsToInts(
+    Value *V, Type *From, Type *To, const Twine &Name) {
   if (From == To)
     return V;
   ValueToValueMapTy::iterator Find = ConvertedForStore.find(V);
@@ -498,8 +515,8 @@ Value *StoreFatPtrsAsIntsVisitor::fatPtrsToInts(Value *V, Type *From, Type *To,
   return Ret;
 }
 
-Value *StoreFatPtrsAsIntsVisitor::intsToFatPtrs(Value *V, Type *From, Type *To,
-                                                const Twine &Name) {
+Value *StoreFatPtrsAsIntsAndExpandMemcpyVisitor::intsToFatPtrs(
+    Value *V, Type *From, Type *To, const Twine &Name) {
   if (From == To)
     return V;
   if (isBufferFatPtrOrVector(To)) {
@@ -531,18 +548,25 @@ Value *StoreFatPtrsAsIntsVisitor::intsToFatPtrs(Value *V, Type *From, Type *To,
   return Ret;
 }
 
-bool StoreFatPtrsAsIntsVisitor::processFunction(Function &F) {
+bool StoreFatPtrsAsIntsAndExpandMemcpyVisitor::processFunction(Function &F) {
   bool Changed = false;
-  // The visitors will mutate GEPs and allocas, but will push loads and stores
-  // to the worklist to avoid invalidation.
+  // Process memcpy-like instructions after the main iteration because they can
+  // invalidate iterators.
+  SmallVector<WeakTrackingVH> CanBecomeLoops;
   for (Instruction &I : make_early_inc_range(instructions(F))) {
-    Changed |= visit(I);
+    if (isa<MemTransferInst, MemSetInst, MemSetPatternInst>(I))
+      CanBecomeLoops.push_back(&I);
+    else
+      Changed |= visit(I);
+  }
+  for (WeakTrackingVH VH : make_early_inc_range(CanBecomeLoops)) {
+    Changed |= visit(cast<Instruction>(VH));
   }
   ConvertedForStore.clear();
   return Changed;
 }
 
-bool StoreFatPtrsAsIntsVisitor::visitAllocaInst(AllocaInst &I) {
+bool StoreFatPtrsAsIntsAndExpandMemcpyVisitor::visitAllocaInst(AllocaInst &I) {
   Type *Ty = I.getAllocatedType();
   Type *NewTy = TypeMap->remapType(Ty);
   if (Ty == NewTy)
@@ -551,7 +575,8 @@ bool StoreFatPtrsAsIntsVisitor::visitAllocaInst(AllocaInst &I) {
   return true;
 }
 
-bool StoreFatPtrsAsIntsVisitor::visitGetElementPtrInst(GetElementPtrInst &I) {
+bool StoreFatPtrsAsIntsAndExpandMemcpyVisitor::visitGetElementPtrInst(
+    GetElementPtrInst &I) {
   Type *Ty = I.getSourceElementType();
   Type *NewTy = TypeMap->remapType(Ty);
   if (Ty == NewTy)
@@ -563,7 +588,7 @@ bool StoreFatPtrsAsIntsVisitor::visitGetElementPtrInst(GetElementPtrInst &I) {
   return true;
 }
 
-bool StoreFatPtrsAsIntsVisitor::visitLoadInst(LoadInst &LI) {
+bool StoreFatPtrsAsIntsAndExpandMemcpyVisitor::visitLoadInst(LoadInst &LI) {
   Type *Ty = LI.getType();
   Type *IntTy = TypeMap->remapType(Ty);
   if (Ty == IntTy)
@@ -581,7 +606,7 @@ bool StoreFatPtrsAsIntsVisitor::visitLoadInst(LoadInst &LI) {
   return true;
 }
 
-bool StoreFatPtrsAsIntsVisitor::visitStoreInst(StoreInst &SI) {
+bool StoreFatPtrsAsIntsAndExpandMemcpyVisitor::visitStoreInst(StoreInst &SI) {
   Value *V = SI.getValueOperand();
   Type *Ty = V->getType();
   Type *IntTy = TypeMap->remapType(Ty);
@@ -597,6 +622,47 @@ bool StoreFatPtrsAsIntsVisitor::visitStoreInst(StoreInst &SI) {
   return true;
 }
 
+bool StoreFatPtrsAsIntsAndExpandMemcpyVisitor::visitMemCpyInst(
+    MemCpyInst &MCI) {
+  // TODO: Allow memcpy.p7.p3 as a synonym for the direct-to-LDS copy, which'll
+  // need loop expansion here.
+  if (MCI.getSourceAddressSpace() != AMDGPUAS::BUFFER_FAT_POINTER &&
+      MCI.getDestAddressSpace() != AMDGPUAS::BUFFER_FAT_POINTER)
+    return false;
+  llvm::expandMemCpyAsLoop(&MCI,
+                           TM->getTargetTransformInfo(*MCI.getFunction()));
+  MCI.eraseFromParent();
+  return true;
+}
+
+bool StoreFatPtrsAsIntsAndExpandMemcpyVisitor::visitMemMoveInst(
+    MemMoveInst &MMI) {
+  if (MMI.getSourceAddressSpace() != AMDGPUAS::BUFFER_FAT_POINTER &&
+      MMI.getDestAddressSpace() != AMDGPUAS::BUFFER_FAT_POINTER)
+    return false;
+  report_fatal_error(
+      "memmove() on buffer descriptors is not implemented because pointer "
+      "comparison on buffer descriptors isn't implemented\n");
+}
+
+bool StoreFatPtrsAsIntsAndExpandMemcpyVisitor::visitMemSetInst(
+    MemSetInst &MSI) {
+  if (MSI.getDestAddressSpace() != AMDGPUAS::BUFFER_FAT_POINTER)
+    return false;
+  llvm::expandMemSetAsLoop(&MSI);
+  MSI.eraseFromParent();
+  return true;
+}
+
+bool StoreFatPtrsAsIntsAndExpandMemcpyVisitor::visitMemSetPatternInst(
+    MemSetPatternInst &MSPI) {
+  if (MSPI.getDestAddressSpace() != AMDGPUAS::BUFFER_FAT_POINTER)
+    return false;
+  llvm::expandMemSetPatternAsLoop(&MSPI);
+  MSPI.eraseFromParent();
+  return true;
+}
+
 namespace {
 /// Convert loads/stores of types that the buffer intrinsics can't handle into
 /// one ore more such loads/stores that consist of legal types.
@@ -1127,6 +1193,7 @@ bool LegalizeBufferContentTypesVisitor::visitStoreInst(StoreInst &SI) {
 
 bool LegalizeBufferContentTypesVisitor::processFunction(Function &F) {
   bool Changed = false;
+  // Note, memory transfer intrinsics won't
   for (Instruction &I : make_early_inc_range(instructions(F))) {
     Changed |= visit(I);
   }
@@ -2084,6 +2151,12 @@ static bool isRemovablePointerIntrinsic(Intrinsic::ID IID) {
   case Intrinsic::invariant_end:
   case Intrinsic::launder_invariant_group:
   case Intrinsic::strip_invariant_group:
+  case Intrinsic::memcpy:
+  case Intrinsic::memcpy_inline:
+  case Intrinsic::memmove:
+  case Intrinsic::memset:
+  case Intrinsic::memset_inline:
+  case Intrinsic::experimental_memset_pattern:
     return true;
   }
 }
@@ -2353,7 +2426,8 @@ bool AMDGPULowerBufferFatPointers::run(Module &M, const TargetMachine &TM) {
         /*RemoveDeadConstants=*/false, /*IncludeSelf=*/true);
   }
 
-  StoreFatPtrsAsIntsVisitor MemOpsRewrite(&IntTM, M.getContext());
+  StoreFatPtrsAsIntsAndExpandMemcpyVisitor MemOpsRewrite(&IntTM, M.getContext(),
+                                                         &TM);
   LegalizeBufferContentTypesVisitor BufferContentsTypeRewrite(DL,
                                                               M.getContext());
   for (Function &F : M.functions()) {
diff --git a/llvm/test/CodeGen/AMDGPU/buffer-fat-pointers-memcpy.ll b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointers-memcpy.ll
new file mode 100644
index 0000000000000..8e023723ec25c
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/buffer-fat-pointers-memcpy.ll
@@ -0,0 +1,1345 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefix=SDAG-GFX942 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck -check-prefix=SDAG-GFX1100 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx942 -global-isel=1 -global-isel-abort=2 < %s | FileCheck -check-prefix=GISEL-GFX942 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1100 -global-isel=1 -global-isel-abort=2 < %s | FileCheck -check-prefix=GISEL-GFX1100 %s
+
+; Note: if you're adding tests here, also add them to
+; lower-buffer-fat-pointers-mem-transfer.ll to verify the IR produced by
+; the lowering.
+;
+; This file is a sanity check to make sure that the code generated
+; for buffer-related memcpy() calls turns into something reasonable in
+; the backend, despite the wide intermediate vectors
+
+target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8"
+target triple = "amdgcn--"
+
+;; memcpy
+
+declare void @llvm.memcpy.p7.p7.i32(ptr addrspace(7), ptr addrspace(7), i32, i1)
+
+define amdgpu_kernel void @memcpy_known(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; SDAG-LABEL: memcpy_known:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT:    s_mov_b32 s7, s24
+; SDAG-NEXT:    s_mov_b32 s6, s23
+; SDAG-NEXT:    s_mov_b32 s5, s22
+; SDAG-NEXT:    s_mov_b32 s4, s21
+; SDAG-NEXT:    s_mov_b32 s8, 0
+; SDAG-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v42, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v43, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v44, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v45, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v46, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v47, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v56, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v57, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v58, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v59, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v60, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v61, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v62, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v63, off, s[0:3], s32 ; 4-byte Folded Spill
+; SDAG-NEXT:  .LBB0_1: ; %load-store-loop
+; SDAG-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-NEXT:    s_add_i32 s9, s20, s8
+; SDAG-NEXT:    v_mov_b32_e32 v60, s9
+; SDAG-NEXT:    buffer_load_dwordx4 v[0:3], v60, s[16:19], 0 offen
+; SDAG-NEXT:    s_add_i32 s9, s25, s8
+; SDAG-NEXT:    s_addk_i32 s8, 0x100
+; SDAG-NEXT:    s_cmpk_lt_u32 s8, 0x2000
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dword v0, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v2, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_load_dwordx4 v[0:3], v60, s[16:19], 0 offen offset:16
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dword v0, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v2, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:92 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_load_dwordx4 v[8:11], v60, s[16:19], 0 offen offset:32
+; SDAG-NEXT:    buffer_load_dwordx4 v[12:15], v60, s[16:19], 0 offen offset:48
+; SDAG-NEXT:    buffer_load_dwordx4 v[16:19], v60, s[16:19], 0 offen offset:64
+; SDAG-NEXT:    buffer_load_dwordx4 v[20:23], v60, s[16:19], 0 offen offset:80
+; SDAG-NEXT:    buffer_load_dwordx4 v[24:27], v60, s[16:19], 0 offen offset:96
+; SDAG-NEXT:    buffer_load_dwordx4 v[28:31], v60, s[16:19], 0 offen offset:112
+; SDAG-NEXT:    buffer_load_dwordx4 v[32:35], v60, s[16:19], 0 offen offset:128
+; SDAG-NEXT:    buffer_load_dwordx4 v[36:39], v60, s[16:19], 0 offen offset:144
+; SDAG-NEXT:    buffer_load_dwordx4 v[48:51], v60, s[16:19], 0 offen offset:160
+; SDAG-NEXT:    buffer_load_dwordx4 v[52:55], v60, s[16:19], 0 offen offset:176
+; SDAG-NEXT:    buffer_load_dwordx4 v[40:43], v60, s[16:19], 0 offen offset:192
+; SDAG-NEXT:    buffer_load_dwordx4 v[44:47], v60, s[16:19], 0 offen offset:208
+; SDAG-NEXT:    buffer_load_dwordx4 v[56:59], v60, s[16:19], 0 offen offset:224
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    buffer_load_dwordx4 v[60:63], v60, s[16:19], 0 offen offset:240
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; SDAG-NEXT:    v_mov_b32_e32 v0, s9
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dwordx4 v[1:4], v0, s[4:7], 0 offen
+; SDAG-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:88 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:92 ; 4-byte Folded Reload
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dwordx4 v[1:4], v0, s[4:7], 0 offen offset:16
+; SDAG-NEXT:    buffer_store_dwordx4 v[8:11], v0, s[4:7], 0 offen offset:32
+; SDAG-NEXT:    buffer_store_dwordx4 v[12:15], v0, s[4:7], 0 offen offset:48
+; SDAG-NEXT:    buffer_store_dwordx4 v[16:19], v0, s[4:7], 0 offen offset:64
+; SDAG-NEXT:    buffer_store_dwordx4 v[20:23], v0, s[4:7], 0 offen offset:80
+; SDAG-NEXT:    buffer_store_dwordx4 v[24:27], v0, s[4:7], 0 offen offset:96
+; SDAG-NEXT:    buffer_store_dwordx4 v[28:31], v0, s[4:7], 0 offen offset:112
+; SDAG-NEXT:    buffer_store_dwordx4 v[32:35], v0, s[4:7], 0 offen offset:128
+; SDAG-NEXT:    buffer_store_dwordx4 v[36:39], v0, s[4:7], 0 offen offset:144
+; SDAG-NEXT:    buffer_store_dwordx4 v[48:51], v0, s[4:7], 0 offen offset:160
+; SDAG-NEXT:    buffer_store_dwordx4 v[52:55], v0, s[4:7], 0 offen offset:176
+; SDAG-NEXT:    buffer_store_dwordx4 v[40:43], v0, s[4:7], 0 offen offset:192
+; SDAG-NEXT:    buffer_store_dwordx4 v[44:47], v0, s[4:7], 0 offen offset:208
+; SDAG-NEXT:    buffer_store_dwordx4 v[56:59], v0, s[4:7], 0 offen offset:224
+; SDAG-NEXT:    buffer_store_dwordx4 v[60:63], v0, s[4:7], 0 offen offset:240
+; SDAG-NEXT:    s_cbranch_scc1 .LBB0_1
+; SDAG-NEXT:  ; %bb.2: ; %memcpy-split
+; SDAG-NEXT:    buffer_load_dword v63, off, s[0:3], s32 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v62, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v61, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v60, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v59, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v58, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v57, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v56, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v47, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v46, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v45, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v44, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v43, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v42, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: memcpy_known:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT:    s_mov_b32 s8, 0
+; GISEL-NEXT:    s_mov_b32 s4, s21
+; GISEL-NEXT:    s_mov_b32 s5, s22
+; GISEL-NEXT:    s_mov_b32 s6, s23
+; GISEL-NEXT:    s_mov_b32 s7, s24
+; GISEL-NEXT:    v_mov_b32_e32 v0, 0x2000
+; GISEL-NEXT:    v_mov_b32_e32 v1, s8
+; GISEL-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v42, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v43, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v44, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v45, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v46, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v56, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v57, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v58, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v59, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v60, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v61, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v62, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v63, off, s[0:3], s32 ; 4-byte Folded Spill
+; GISEL-NEXT:  .LBB0_1: ; %load-store-loop
+; GISEL-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-NEXT:    v_add_u32_e32 v46, s20, v1
+; GISEL-NEXT:    buffer_load_dwordx4 v[2:5], v46, s[16:19], 0 offen
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dword v2, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v5, off, s[0:3], s32 offset:88 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_load_dwordx4 v[6:9], v46, s[16:19], 0 offen offset:16
+; GISEL-NEXT:    buffer_load_dwordx4 v[10:13], v46, s[16:19], 0 offen offset:32
+; GISEL-NEXT:    buffer_load_dwordx4 v[14:17], v46, s[16:19], 0 offen offset:48
+; GISEL-NEXT:    buffer_load_dwordx4 v[18:21], v46, s[16:19], 0 offen offset:64
+; GISEL-NEXT:    buffer_load_dwordx4 v[22:25], v46, s[16:19], 0 offen offset:80
+; GISEL-NEXT:    buffer_load_dwordx4 v[26:29], v46, s[16:19], 0 offen offset:96
+; GISEL-NEXT:    buffer_load_dwordx4 v[30:33], v46, s[16:19], 0 offen offset:112
+; GISEL-NEXT:    buffer_load_dwordx4 v[34:37], v46, s[16:19], 0 offen offset:128
+; GISEL-NEXT:    buffer_load_dwordx4 v[48:51], v46, s[16:19], 0 offen offset:144
+; GISEL-NEXT:    buffer_load_dwordx4 v[52:55], v46, s[16:19], 0 offen offset:160
+; GISEL-NEXT:    buffer_load_dwordx4 v[38:41], v46, s[16:19], 0 offen offset:176
+; GISEL-NEXT:    buffer_load_dwordx4 v[42:45], v46, s[16:19], 0 offen offset:192
+; GISEL-NEXT:    buffer_load_dwordx4 v[56:59], v46, s[16:19], 0 offen offset:208
+; GISEL-NEXT:    buffer_load_dwordx4 v[60:63], v46, s[16:19], 0 offen offset:224
+; GISEL-NEXT:    buffer_load_dwordx4 v[2:5], v46, s[16:19], 0 offen offset:240
+; GISEL-NEXT:    v_add_u32_e32 v46, s25, v1
+; GISEL-NEXT:    v_add_u32_e32 v1, 0x100, v1
+; GISEL-NEXT:    v_cmp_lt_u32_e32 vcc, v1, v0
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dword v2, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v5, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v5, off, s[0:3], s32 offset:88 ; 4-byte Folded Reload
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dwordx4 v[2:5], v46, s[4:7], 0 offen
+; GISEL-NEXT:    buffer_store_dwordx4 v[6:9], v46, s[4:7], 0 offen offset:16
+; GISEL-NEXT:    buffer_store_dwordx4 v[10:13], v46, s[4:7], 0 offen offset:32
+; GISEL-NEXT:    buffer_store_dwordx4 v[14:17], v46, s[4:7], 0 offen offset:48
+; GISEL-NEXT:    buffer_store_dwordx4 v[18:21], v46, s[4:7], 0 offen offset:64
+; GISEL-NEXT:    buffer_store_dwordx4 v[22:25], v46, s[4:7], 0 offen offset:80
+; GISEL-NEXT:    buffer_store_dwordx4 v[26:29], v46, s[4:7], 0 offen offset:96
+; GISEL-NEXT:    buffer_store_dwordx4 v[30:33], v46, s[4:7], 0 offen offset:112
+; GISEL-NEXT:    buffer_store_dwordx4 v[34:37], v46, s[4:7], 0 offen offset:128
+; GISEL-NEXT:    buffer_store_dwordx4 v[48:51], v46, s[4:7], 0 offen offset:144
+; GISEL-NEXT:    buffer_store_dwordx4 v[52:55], v46, s[4:7], 0 offen offset:160
+; GISEL-NEXT:    buffer_store_dwordx4 v[38:41], v46, s[4:7], 0 offen offset:176
+; GISEL-NEXT:    buffer_store_dwordx4 v[42:45], v46, s[4:7], 0 offen offset:192
+; GISEL-NEXT:    buffer_store_dwordx4 v[56:59], v46, s[4:7], 0 offen offset:208
+; GISEL-NEXT:    buffer_store_dwordx4 v[60:63], v46, s[4:7], 0 offen offset:224
+; GISEL-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v5, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dwordx4 v[2:5], v46, s[4:7], 0 offen offset:240
+; GISEL-NEXT:    s_cbranch_vccnz .LBB0_1
+; GISEL-NEXT:  ; %bb.2: ; %memcpy-split
+; GISEL-NEXT:    buffer_load_dword v63, off, s[0:3], s32 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v62, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v61, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v60, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v59, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v58, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v57, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v56, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v46, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v45, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v44, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v43, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v42, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_setpc_b64 s[30:31]
+; SDAG-GFX942-LABEL: memcpy_known:
+; SDAG-GFX942:       ; %bb.3:
+; SDAG-GFX942-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x0
+; SDAG-GFX942-NEXT:    s_load_dword s12, s[4:5], 0x10
+; SDAG-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX942-NEXT:    s_branch .LBB0_0
+; SDAG-GFX942-NEXT:    .p2align 8
+; SDAG-GFX942-NEXT:  ; %bb.4:
+; SDAG-GFX942-NEXT:  .LBB0_0:
+; SDAG-GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; SDAG-GFX942-NEXT:    s_load_dword s17, s[4:5], 0x34
+; SDAG-GFX942-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x44
+; SDAG-GFX942-NEXT:    s_load_dword s12, s[4:5], 0x54
+; SDAG-GFX942-NEXT:    s_mov_b32 s16, 0
+; SDAG-GFX942-NEXT:    s_mov_b32 s5, s16
+; SDAG-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX942-NEXT:    s_mov_b32 s4, s3
+; SDAG-GFX942-NEXT:    s_or_b64 s[6:7], s[4:5], s[16:17]
+; SDAG-GFX942-NEXT:    s_mov_b32 s17, s2
+; SDAG-GFX942-NEXT:    s_mov_b32 s2, s1
+; SDAG-GFX942-NEXT:    s_mov_b32 s3, s16
+; SDAG-GFX942-NEXT:    s_or_b64 s[4:5], s[2:3], s[16:17]
+; SDAG-GFX942-NEXT:    s_mov_b32 s17, s12
+; SDAG-GFX942-NEXT:    s_mov_b32 s2, s11
+; SDAG-GFX942-NEXT:    s_or_b64 s[14:15], s[2:3], s[16:17]
+; SDAG-GFX942-NEXT:    s_mov_b32 s17, s10
+; SDAG-GFX942-NEXT:    s_mov_b32 s2, s9
+; SDAG-GFX942-NEXT:    s_or_b64 s[12:13], s[2:3], s[16:17]
+; SDAG-GFX942-NEXT:  .LBB0_1: ; %load-store-loop
+; SDAG-GFX942-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX942-NEXT:    s_add_i32 s1, s0, s16
+; SDAG-GFX942-NEXT:    v_mov_b32_e32 v60, s1
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[8:11], v60, s[4:7], 0 offen
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[4:7], v60, s[4:7], 0 offen offset:16
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[12:15], v60, s[4:7], 0 offen offset:32
+; SDAG-GFX942-NEXT:    s_add_i32 s2, s8, s16
+; SDAG-GFX942-NEXT:    v_mov_b32_e32 v0, s2
+; SDAG-GFX942-NEXT:    s_addk_i32 s16, 0x100
+; SDAG-GFX942-NEXT:    s_cmpk_lt_u32 s16, 0x2000
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a0, v15 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a1, v14 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a2, v13 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a3, v12 ; Reload Reuse
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[12:15], v60, s[4:7], 0 offen offset:48
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[16:19], v60, s[4:7], 0 offen offset:64
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[20:23], v60, s[4:7], 0 offen offset:80
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[24:27], v60, s[4:7], 0 offen offset:96
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[28:31], v60, s[4:7], 0 offen offset:112
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[32:35], v60, s[4:7], 0 offen offset:128
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[36:39], v60, s[4:7], 0 offen offset:144
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[40:43], v60, s[4:7], 0 offen offset:160
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[44:47], v60, s[4:7], 0 offen offset:176
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[48:51], v60, s[4:7], 0 offen offset:192
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[52:55], v60, s[4:7], 0 offen offset:208
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[56:59], v60, s[4:7], 0 offen offset:224
+; SDAG-GFX942-NEXT:    s_nop 0
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[60:63], v60, s[4:7], 0 offen offset:240
+; SDAG-GFX942-NEXT:    s_nop 0
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[8:11], v0, s[12:15], 0 offen
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[4:7], v0, s[12:15], 0 offen offset:16
+; SDAG-GFX942-NEXT:    s_nop 1
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v5, a0 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v4, a1 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v3, a2 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v2, a3 ; Reload Reuse
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v0, s[12:15], 0 offen offset:32
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[12:15], v0, s[12:15], 0 offen offset:48
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[16:19], v0, s[12:15], 0 offen offset:64
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[20:23], v0, s[12:15], 0 offen offset:80
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[24:27], v0, s[12:15], 0 offen offset:96
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[28:31], v0, s[12:15], 0 offen offset:112
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[32:35], v0, s[12:15], 0 offen offset:128
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[36:39], v0, s[12:15], 0 offen offset:144
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[40:43], v0, s[12:15], 0 offen offset:160
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[44:47], v0, s[12:15], 0 offen offset:176
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[48:51], v0, s[12:15], 0 offen offset:192
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[52:55], v0, s[12:15], 0 offen offset:208
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[56:59], v0, s[12:15], 0 offen offset:224
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[60:63], v0, s[12:15], 0 offen offset:240
+; SDAG-GFX942-NEXT:    s_cbranch_scc1 .LBB0_1
+; SDAG-GFX942-NEXT:  ; %bb.2: ; %memcpy-split
+; SDAG-GFX942-NEXT:    s_endpgm
+;
+; SDAG-GFX1100-LABEL: memcpy_known:
+; SDAG-GFX1100:       ; %bb.0:
+; SDAG-GFX1100-NEXT:    s_clause 0x3
+; SDAG-GFX1100-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; SDAG-GFX1100-NEXT:    s_load_b32 s17, s[4:5], 0x34
+; SDAG-GFX1100-NEXT:    s_load_b128 s[8:11], s[4:5], 0x44
+; SDAG-GFX1100-NEXT:    s_load_b32 s18, s[4:5], 0x54
+; SDAG-GFX1100-NEXT:    s_mov_b32 s16, 0
+; SDAG-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; SDAG-GFX1100-NEXT:    s_mov_b32 s5, s16
+; SDAG-GFX1100-NEXT:    s_mov_b32 s13, s16
+; SDAG-GFX1100-NEXT:    s_mov_b32 s15, s16
+; SDAG-GFX1100-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX1100-NEXT:    s_mov_b32 s4, s3
+; SDAG-GFX1100-NEXT:    s_mov_b32 s12, s1
+; SDAG-GFX1100-NEXT:    s_or_b64 s[6:7], s[4:5], s[16:17]
+; SDAG-GFX1100-NEXT:    s_mov_b32 s17, s2
+; SDAG-GFX1100-NEXT:    s_mov_b32 s14, s11
+; SDAG-GFX1100-NEXT:    s_or_b64 s[4:5], s[12:13], s[16:17]
+; SDAG-GFX1100-NEXT:    s_mov_b32 s17, s18
+; SDAG-GFX1100-NEXT:    s_mov_b32 s2, s9
+; SDAG-GFX1100-NEXT:    s_or_b64 s[14:15], s[14:15], s[16:17]
+; SDAG-GFX1100-NEXT:    s_mov_b32 s17, s10
+; SDAG-GFX1100-NEXT:    s_mov_b32 s3, s16
+; SDAG-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; SDAG-GFX1100-NEXT:    s_or_b64 s[12:13], s[2:3], s[16:17]
+; SDAG-GFX1100-NEXT:  .LBB0_1: ; %load-store-loop
+; SDAG-GFX1100-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX1100-NEXT:    s_add_i32 s1, s0, s16
+; SDAG-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; SDAG-GFX1100-NEXT:    v_mov_b32_e32 v60, s1
+; SDAG-GFX1100-NEXT:    s_add_i32 s1, s8, s16
+; SDAG-GFX1100-NEXT:    s_addk_i32 s16, 0x100
+; SDAG-GFX1100-NEXT:    v_mov_b32_e32 v64, s1
+; SDAG-GFX1100-NEXT:    s_cmpk_lt_u32 s16, 0x2000
+; SDAG-GFX1100-NEXT:    s_clause 0xf
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[0:3], v60, s[4:7], 0 offen
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[4:7], v60, s[4:7], 0 offen offset:16
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[8:11], v60, s[4:7], 0 offen offset:32
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[12:15], v60, s[4:7], 0 offen offset:48
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[16:19], v60, s[4:7], 0 offen offset:64
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[20:23], v60, s[4:7], 0 offen offset:80
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[24:27], v60, s[4:7], 0 offen offset:96
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[28:31], v60, s[4:7], 0 offen offset:112
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[32:35], v60, s[4:7], 0 offen offset:128
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[36:39], v60, s[4:7], 0 offen offset:144
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[40:43], v60, s[4:7], 0 offen offset:160
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[44:47], v60, s[4:7], 0 offen offset:176
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[48:51], v60, s[4:7], 0 offen offset:192
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[52:55], v60, s[4:7], 0 offen offset:208
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[56:59], v60, s[4:7], 0 offen offset:224
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[60:63], v60, s[4:7], 0 offen offset:240
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[0:3], v64, s[12:15], 0 offen
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(14)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[4:7], v64, s[12:15], 0 offen offset:16
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(13)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[8:11], v64, s[12:15], 0 offen offset:32
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(12)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[12:15], v64, s[12:15], 0 offen offset:48
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(11)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[16:19], v64, s[12:15], 0 offen offset:64
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(10)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[20:23], v64, s[12:15], 0 offen offset:80
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(9)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[24:27], v64, s[12:15], 0 offen offset:96
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(8)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[28:31], v64, s[12:15], 0 offen offset:112
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(7)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[32:35], v64, s[12:15], 0 offen offset:128
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(6)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[36:39], v64, s[12:15], 0 offen offset:144
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(5)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[40:43], v64, s[12:15], 0 offen offset:160
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(4)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[44:47], v64, s[12:15], 0 offen offset:176
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(3)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[48:51], v64, s[12:15], 0 offen offset:192
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(2)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[52:55], v64, s[12:15], 0 offen offset:208
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(1)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[56:59], v64, s[12:15], 0 offen offset:224
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[60:63], v64, s[12:15], 0 offen offset:240
+; SDAG-GFX1100-NEXT:    s_cbranch_scc1 .LBB0_1
+; SDAG-GFX1100-NEXT:  ; %bb.2: ; %memcpy-split
+; SDAG-GFX1100-NEXT:    s_endpgm
+;
+; GISEL-GFX942-LABEL: memcpy_known:
+; GISEL-GFX942:       ; %bb.0:
+; GISEL-GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GISEL-GFX942-NEXT:    s_load_dword s11, s[4:5], 0x34
+; GISEL-GFX942-NEXT:    s_mov_b32 s7, 0
+; GISEL-GFX942-NEXT:    s_load_dwordx4 s[12:15], s[4:5], 0x44
+; GISEL-GFX942-NEXT:    s_mov_b32 s8, s7
+; GISEL-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s1
+; GISEL-GFX942-NEXT:    s_mov_b32 s9, s2
+; GISEL-GFX942-NEXT:    s_or_b64 s[8:9], s[6:7], s[8:9]
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s3
+; GISEL-GFX942-NEXT:    s_load_dword s3, s[4:5], 0x54
+; GISEL-GFX942-NEXT:    s_mov_b32 s10, s7
+; GISEL-GFX942-NEXT:    s_or_b64 s[10:11], s[6:7], s[10:11]
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s13
+; GISEL-GFX942-NEXT:    s_mov_b32 s4, s7
+; GISEL-GFX942-NEXT:    s_mov_b32 s5, s14
+; GISEL-GFX942-NEXT:    s_mov_b32 s16, 0
+; GISEL-GFX942-NEXT:    s_or_b64 s[4:5], s[6:7], s[4:5]
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s15
+; GISEL-GFX942-NEXT:    s_mov_b32 s2, s7
+; GISEL-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX942-NEXT:    s_or_b64 s[6:7], s[6:7], s[2:3]
+; GISEL-GFX942-NEXT:    v_mov_b32_e32 v0, 0x2000
+; GISEL-GFX942-NEXT:    v_mov_b32_e32 v1, s16
+; GISEL-GFX942-NEXT:  .LBB0_1: ; %load-store-loop
+; GISEL-GFX942-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX942-NEXT:    v_add_u32_e32 v62, s0, v1
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[2:5], v62, s[8:11], 0 offen
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[6:9], v62, s[8:11], 0 offen offset:16
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[10:13], v62, s[8:11], 0 offen offset:32
+; GISEL-GFX942-NEXT:    v_add_u32_e32 v63, s12, v1
+; GISEL-GFX942-NEXT:    v_add_u32_e32 v1, 0x100, v1
+; GISEL-GFX942-NEXT:    v_cmp_lt_u32_e32 vcc, v1, v0
+; GISEL-GFX942-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a0, v13 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a1, v12 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a2, v11 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a3, v10 ; Reload Reuse
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[14:17], v62, s[8:11], 0 offen offset:48
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[18:21], v62, s[8:11], 0 offen offset:64
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[22:25], v62, s[8:11], 0 offen offset:80
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[26:29], v62, s[8:11], 0 offen offset:96
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[30:33], v62, s[8:11], 0 offen offset:112
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[34:37], v62, s[8:11], 0 offen offset:128
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[38:41], v62, s[8:11], 0 offen offset:144
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[42:45], v62, s[8:11], 0 offen offset:160
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[46:49], v62, s[8:11], 0 offen offset:176
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[50:53], v62, s[8:11], 0 offen offset:192
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[54:57], v62, s[8:11], 0 offen offset:208
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[58:61], v62, s[8:11], 0 offen offset:224
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[10:13], v62, s[8:11], 0 offen offset:240
+; GISEL-GFX942-NEXT:    s_nop 0
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v63, s[4:7], 0 offen
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[6:9], v63, s[4:7], 0 offen offset:16
+; GISEL-GFX942-NEXT:    s_waitcnt vmcnt(2)
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a4, v13 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v5, a0 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v4, a1 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v3, a2 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v2, a3 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a5, v12 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a6, v11 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a7, v10 ; Reload Reuse
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v63, s[4:7], 0 offen offset:32
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[14:17], v63, s[4:7], 0 offen offset:48
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[18:21], v63, s[4:7], 0 offen offset:64
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[22:25], v63, s[4:7], 0 offen offset:80
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[26:29], v63, s[4:7], 0 offen offset:96
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[30:33], v63, s[4:7], 0 offen offset:112
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[34:37], v63, s[4:7], 0 offen offset:128
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[38:41], v63, s[4:7], 0 offen offset:144
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[42:45], v63, s[4:7], 0 offen offset:160
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[46:49], v63, s[4:7], 0 offen offset:176
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[50:53], v63, s[4:7], 0 offen offset:192
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[54:57], v63, s[4:7], 0 offen offset:208
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[58:61], v63, s[4:7], 0 offen offset:224
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v5, a4 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v4, a5 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v3, a6 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v2, a7 ; Reload Reuse
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v63, s[4:7], 0 offen offset:240
+; GISEL-GFX942-NEXT:    s_cbranch_vccnz .LBB0_1
+; GISEL-GFX942-NEXT:  ; %bb.2: ; %memcpy-split
+; GISEL-GFX942-NEXT:    s_endpgm
+;
+; GISEL-GFX1100-LABEL: memcpy_known:
+; GISEL-GFX1100:       ; %bb.0:
+; GISEL-GFX1100-NEXT:    s_clause 0x3
+; GISEL-GFX1100-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GISEL-GFX1100-NEXT:    s_load_b128 s[8:11], s[4:5], 0x44
+; GISEL-GFX1100-NEXT:    s_load_b32 s7, s[4:5], 0x34
+; GISEL-GFX1100-NEXT:    s_load_b32 s15, s[4:5], 0x54
+; GISEL-GFX1100-NEXT:    s_mov_b32 s17, 0
+; GISEL-GFX1100-NEXT:    s_mov_b32 s12, 0
+; GISEL-GFX1100-NEXT:    s_mov_b32 s4, s17
+; GISEL-GFX1100-NEXT:    s_mov_b32 s6, s17
+; GISEL-GFX1100-NEXT:    v_mov_b32_e32 v0, s12
+; GISEL-GFX1100-NEXT:    s_mov_b32 s14, s17
+; GISEL-GFX1100-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX1100-NEXT:    s_mov_b32 s16, s1
+; GISEL-GFX1100-NEXT:    s_mov_b32 s5, s2
+; GISEL-GFX1100-NEXT:    s_mov_b32 s2, s17
+; GISEL-GFX1100-NEXT:    s_or_b64 s[4:5], s[16:17], s[4:5]
+; GISEL-GFX1100-NEXT:    s_mov_b32 s16, s3
+; GISEL-GFX1100-NEXT:    s_mov_b32 s3, s10
+; GISEL-GFX1100-NEXT:    s_or_b64 s[6:7], s[16:17], s[6:7]
+; GISEL-GFX1100-NEXT:    s_mov_b32 s16, s9
+; GISEL-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GISEL-GFX1100-NEXT:    s_or_b64 s[12:13], s[16:17], s[2:3]
+; GISEL-GFX1100-NEXT:    s_mov_b32 s16, s11
+; GISEL-GFX1100-NEXT:    s_or_b64 s[14:15], s[16:17], s[14:15]
+; GISEL-GFX1100-NEXT:  .LBB0_1: ; %load-store-loop
+; GISEL-GFX1100-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX1100-NEXT:    v_add_nc_u32_e32 v61, s0, v0
+; GISEL-GFX1100-NEXT:    v_add_nc_u32_e32 v65, s8, v0
+; GISEL-GFX1100-NEXT:    v_add_nc_u32_e32 v0, 0x100, v0
+; GISEL-GFX1100-NEXT:    s_clause 0xf
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[1:4], v61, s[4:7], 0 offen
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[5:8], v61, s[4:7], 0 offen offset:16
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[9:12], v61, s[4:7], 0 offen offset:32
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[13:16], v61, s[4:7], 0 offen offset:48
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[17:20], v61, s[4:7], 0 offen offset:64
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[21:24], v61, s[4:7], 0 offen offset:80
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[25:28], v61, s[4:7], 0 offen offset:96
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[29:32], v61, s[4:7], 0 offen offset:112
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[33:36], v61, s[4:7], 0 offen offset:128
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[37:40], v61, s[4:7], 0 offen offset:144
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[41:44], v61, s[4:7], 0 offen offset:160
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[45:48], v61, s[4:7], 0 offen offset:176
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[49:52], v61, s[4:7], 0 offen offset:192
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[53:56], v61, s[4:7], 0 offen offset:208
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[57:60], v61, s[4:7], 0 offen offset:224
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[61:64], v61, s[4:7], 0 offen offset:240
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(15)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[1:4], v65, s[12:15], 0 offen
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(14)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[5:8], v65, s[12:15], 0 offen offset:16
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(13)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[9:12], v65, s[12:15], 0 offen offset:32
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(12)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[13:16], v65, s[12:15], 0 offen offset:48
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(11)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[17:20], v65, s[12:15], 0 offen offset:64
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(10)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[21:24], v65, s[12:15], 0 offen offset:80
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(9)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[25:28], v65, s[12:15], 0 offen offset:96
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(8)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[29:32], v65, s[12:15], 0 offen offset:112
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(7)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[33:36], v65, s[12:15], 0 offen offset:128
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(6)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[37:40], v65, s[12:15], 0 offen offset:144
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(5)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[41:44], v65, s[12:15], 0 offen offset:160
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(4)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[45:48], v65, s[12:15], 0 offen offset:176
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(3)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[49:52], v65, s[12:15], 0 offen offset:192
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(2)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[53:56], v65, s[12:15], 0 offen offset:208
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(1)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[57:60], v65, s[12:15], 0 offen offset:224
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[61:64], v65, s[12:15], 0 offen offset:240
+; GISEL-GFX1100-NEXT:    v_cmp_gt_u32_e32 vcc_lo, 0x2000, v0
+; GISEL-GFX1100-NEXT:    s_cbranch_vccnz .LBB0_1
+; GISEL-GFX1100-NEXT:  ; %bb.2: ; %memcpy-split
+; GISEL-GFX1100-NEXT:    s_endpgm
+  call void @llvm.memcpy.p7.p7.i32(ptr addrspace(7) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 8192, i1 false)
+  ret void
+}
+
+define amdgpu_kernel void @memcpy_known_medium(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; SDAG-LABEL: memcpy_known_medium:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT:    s_mov_b32 s7, s24
+; SDAG-NEXT:    s_mov_b32 s6, s23
+; SDAG-NEXT:    s_mov_b32 s5, s22
+; SDAG-NEXT:    s_mov_b32 s4, s21
+; SDAG-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v42, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v43, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v44, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v45, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v56, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v57, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v58, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v59, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v60, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v61, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v62, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v63, off, s[0:3], s32 ; 4-byte Folded Spill
+; SDAG-NEXT:  .LBB1_1: ; %load-store-loop
+; SDAG-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-NEXT:    v_add_u32_e32 v45, s20, v0
+; SDAG-NEXT:    buffer_load_dwordx4 v[1:4], v45, s[16:19], 0 offen
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    buffer_store_dword v2, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_load_dwordx4 v[5:8], v45, s[16:19], 0 offen offset:16
+; SDAG-NEXT:    buffer_load_dwordx4 v[9:12], v45, s[16:19], 0 offen offset:32
+; SDAG-NEXT:    buffer_load_dwordx4 v[13:16], v45, s[16:19], 0 offen offset:48
+; SDAG-NEXT:    buffer_load_dwordx4 v[17:20], v45, s[16:19], 0 offen offset:64
+; SDAG-NEXT:    buffer_load_dwordx4 v[21:24], v45, s[16:19], 0 offen offset:80
+; SDAG-NEXT:    buffer_load_dwordx4 v[25:28], v45, s[16:19], 0 offen offset:96
+; SDAG-NEXT:    buffer_load_dwordx4 v[29:32], v45, s[16:19], 0 offen offset:112
+; SDAG-NEXT:    buffer_load_dwordx4 v[33:36], v45, s[16:19], 0 offen offset:128
+; SDAG-NEXT:    buffer_load_dwordx4 v[48:51], v45, s[16:19], 0 offen offset:144
+; SDAG-NEXT:    buffer_load_dwordx4 v[52:55], v45, s[16:19], 0 offen offset:160
+; SDAG-NEXT:    buffer_load_dwordx4 v[37:40], v45, s[16:19], 0 offen offset:176
+; SDAG-NEXT:    buffer_load_dwordx4 v[41:44], v45, s[16:19], 0 offen offset:192
+; SDAG-NEXT:    buffer_load_dwordx4 v[56:59], v45, s[16:19], 0 offen offset:208
+; SDAG-NEXT:    buffer_load_dwordx4 v[60:63], v45, s[16:19], 0 offen offset:224
+; SDAG-NEXT:    buffer_load_dwordx4 v[1:4], v45, s[16:19], 0 offen offset:240
+; SDAG-NEXT:    v_add_u32_e32 v45, s25, v0
+; SDAG-NEXT:    v_add_co_u32_e32 v0, vcc, 0x100, v0
+; SDAG-NEXT:    s_and_b64 vcc, exec, vcc
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    buffer_store_dword v2, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; SDAG-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; SDAG-NEXT:    s_nop 0
+; SDAG-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dwordx4 v[1:4], v45, s[4:7], 0 offen
+; SDAG-NEXT:    buffer_store_dwordx4 v[5:8], v45, s[4:7], 0 offen offset:16
+; SDAG-NEXT:    buffer_store_dwordx4 v[9:12], v45, s[4:7], 0 offen offset:32
+; SDAG-NEXT:    buffer_store_dwordx4 v[13:16], v45, s[4:7], 0 offen offset:48
+; SDAG-NEXT:    buffer_store_dwordx4 v[17:20], v45, s[4:7], 0 offen offset:64
+; SDAG-NEXT:    buffer_store_dwordx4 v[21:24], v45, s[4:7], 0 offen offset:80
+; SDAG-NEXT:    buffer_store_dwordx4 v[25:28], v45, s[4:7], 0 offen offset:96
+; SDAG-NEXT:    buffer_store_dwordx4 v[29:32], v45, s[4:7], 0 offen offset:112
+; SDAG-NEXT:    buffer_store_dwordx4 v[33:36], v45, s[4:7], 0 offen offset:128
+; SDAG-NEXT:    buffer_store_dwordx4 v[48:51], v45, s[4:7], 0 offen offset:144
+; SDAG-NEXT:    buffer_store_dwordx4 v[52:55], v45, s[4:7], 0 offen offset:160
+; SDAG-NEXT:    buffer_store_dwordx4 v[37:40], v45, s[4:7], 0 offen offset:176
+; SDAG-NEXT:    buffer_store_dwordx4 v[41:44], v45, s[4:7], 0 offen offset:192
+; SDAG-NEXT:    buffer_store_dwordx4 v[56:59], v45, s[4:7], 0 offen offset:208
+; SDAG-NEXT:    buffer_store_dwordx4 v[60:63], v45, s[4:7], 0 offen offset:224
+; SDAG-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dwordx4 v[1:4], v45, s[4:7], 0 offen offset:240
+; SDAG-NEXT:    s_cbranch_vccnz .LBB1_1
+; SDAG-NEXT:  ; %bb.2: ; %memcpy-split
+; SDAG-NEXT:    buffer_load_dword v63, off, s[0:3], s32 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v62, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v61, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v60, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v59, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v58, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v57, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v56, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v45, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v44, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v43, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v42, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; SDAG-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: memcpy_known_medium:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT:    s_mov_b32 s8, 0
+; GISEL-NEXT:    s_mov_b32 s4, s21
+; GISEL-NEXT:    s_mov_b32 s5, s22
+; GISEL-NEXT:    s_mov_b32 s6, s23
+; GISEL-NEXT:    s_mov_b32 s7, s24
+; GISEL-NEXT:    v_mov_b32_e32 v0, s8
+; GISEL-NEXT:    buffer_store_dword v40, off, s[0:3], s32 offset:52 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v41, off, s[0:3], s32 offset:48 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v42, off, s[0:3], s32 offset:44 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v43, off, s[0:3], s32 offset:40 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v44, off, s[0:3], s32 offset:36 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v45, off, s[0:3], s32 offset:32 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v56, off, s[0:3], s32 offset:28 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v57, off, s[0:3], s32 offset:24 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v58, off, s[0:3], s32 offset:20 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v59, off, s[0:3], s32 offset:16 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v60, off, s[0:3], s32 offset:12 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v61, off, s[0:3], s32 offset:8 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v62, off, s[0:3], s32 offset:4 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v63, off, s[0:3], s32 ; 4-byte Folded Spill
+; GISEL-NEXT:  .LBB1_1: ; %load-store-loop
+; GISEL-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-NEXT:    v_add_u32_e32 v45, s20, v0
+; GISEL-NEXT:    buffer_load_dwordx4 v[1:4], v45, s[16:19], 0 offen
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:72 ; 4-byte Folded Spill
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    buffer_store_dword v2, off, s[0:3], s32 offset:76 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:80 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:84 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_load_dwordx4 v[5:8], v45, s[16:19], 0 offen offset:16
+; GISEL-NEXT:    buffer_load_dwordx4 v[9:12], v45, s[16:19], 0 offen offset:32
+; GISEL-NEXT:    buffer_load_dwordx4 v[13:16], v45, s[16:19], 0 offen offset:48
+; GISEL-NEXT:    buffer_load_dwordx4 v[17:20], v45, s[16:19], 0 offen offset:64
+; GISEL-NEXT:    buffer_load_dwordx4 v[21:24], v45, s[16:19], 0 offen offset:80
+; GISEL-NEXT:    buffer_load_dwordx4 v[25:28], v45, s[16:19], 0 offen offset:96
+; GISEL-NEXT:    buffer_load_dwordx4 v[29:32], v45, s[16:19], 0 offen offset:112
+; GISEL-NEXT:    buffer_load_dwordx4 v[33:36], v45, s[16:19], 0 offen offset:128
+; GISEL-NEXT:    buffer_load_dwordx4 v[48:51], v45, s[16:19], 0 offen offset:144
+; GISEL-NEXT:    buffer_load_dwordx4 v[52:55], v45, s[16:19], 0 offen offset:160
+; GISEL-NEXT:    buffer_load_dwordx4 v[37:40], v45, s[16:19], 0 offen offset:176
+; GISEL-NEXT:    buffer_load_dwordx4 v[41:44], v45, s[16:19], 0 offen offset:192
+; GISEL-NEXT:    buffer_load_dwordx4 v[56:59], v45, s[16:19], 0 offen offset:208
+; GISEL-NEXT:    buffer_load_dwordx4 v[60:63], v45, s[16:19], 0 offen offset:224
+; GISEL-NEXT:    buffer_load_dwordx4 v[1:4], v45, s[16:19], 0 offen offset:240
+; GISEL-NEXT:    v_add_u32_e32 v45, s25, v0
+; GISEL-NEXT:    v_add_co_u32_e32 v0, vcc, 0x100, v0
+; GISEL-NEXT:    s_xor_b64 s[8:9], vcc, -1
+; GISEL-NEXT:    s_xor_b64 s[8:9], s[8:9], -1
+; GISEL-NEXT:    s_and_b64 vcc, s[8:9], exec
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dword v1, off, s[0:3], s32 offset:56 ; 4-byte Folded Spill
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    buffer_store_dword v2, off, s[0:3], s32 offset:60 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v3, off, s[0:3], s32 offset:64 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_store_dword v4, off, s[0:3], s32 offset:68 ; 4-byte Folded Spill
+; GISEL-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:72 ; 4-byte Folded Reload
+; GISEL-NEXT:    s_nop 0
+; GISEL-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:76 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:80 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:84 ; 4-byte Folded Reload
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dwordx4 v[1:4], v45, s[4:7], 0 offen
+; GISEL-NEXT:    buffer_store_dwordx4 v[5:8], v45, s[4:7], 0 offen offset:16
+; GISEL-NEXT:    buffer_store_dwordx4 v[9:12], v45, s[4:7], 0 offen offset:32
+; GISEL-NEXT:    buffer_store_dwordx4 v[13:16], v45, s[4:7], 0 offen offset:48
+; GISEL-NEXT:    buffer_store_dwordx4 v[17:20], v45, s[4:7], 0 offen offset:64
+; GISEL-NEXT:    buffer_store_dwordx4 v[21:24], v45, s[4:7], 0 offen offset:80
+; GISEL-NEXT:    buffer_store_dwordx4 v[25:28], v45, s[4:7], 0 offen offset:96
+; GISEL-NEXT:    buffer_store_dwordx4 v[29:32], v45, s[4:7], 0 offen offset:112
+; GISEL-NEXT:    buffer_store_dwordx4 v[33:36], v45, s[4:7], 0 offen offset:128
+; GISEL-NEXT:    buffer_store_dwordx4 v[48:51], v45, s[4:7], 0 offen offset:144
+; GISEL-NEXT:    buffer_store_dwordx4 v[52:55], v45, s[4:7], 0 offen offset:160
+; GISEL-NEXT:    buffer_store_dwordx4 v[37:40], v45, s[4:7], 0 offen offset:176
+; GISEL-NEXT:    buffer_store_dwordx4 v[41:44], v45, s[4:7], 0 offen offset:192
+; GISEL-NEXT:    buffer_store_dwordx4 v[56:59], v45, s[4:7], 0 offen offset:208
+; GISEL-NEXT:    buffer_store_dwordx4 v[60:63], v45, s[4:7], 0 offen offset:224
+; GISEL-NEXT:    buffer_load_dword v1, off, s[0:3], s32 offset:56 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v2, off, s[0:3], s32 offset:60 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v3, off, s[0:3], s32 offset:64 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v4, off, s[0:3], s32 offset:68 ; 4-byte Folded Reload
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dwordx4 v[1:4], v45, s[4:7], 0 offen offset:240
+; GISEL-NEXT:    s_cbranch_vccnz .LBB1_1
+; GISEL-NEXT:  ; %bb.2: ; %memcpy-split
+; GISEL-NEXT:    buffer_load_dword v63, off, s[0:3], s32 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v62, off, s[0:3], s32 offset:4 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v61, off, s[0:3], s32 offset:8 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v60, off, s[0:3], s32 offset:12 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v59, off, s[0:3], s32 offset:16 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v58, off, s[0:3], s32 offset:20 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v57, off, s[0:3], s32 offset:24 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v56, off, s[0:3], s32 offset:28 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v45, off, s[0:3], s32 offset:32 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v44, off, s[0:3], s32 offset:36 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v43, off, s[0:3], s32 offset:40 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v42, off, s[0:3], s32 offset:44 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v41, off, s[0:3], s32 offset:48 ; 4-byte Folded Reload
+; GISEL-NEXT:    buffer_load_dword v40, off, s[0:3], s32 offset:52 ; 4-byte Folded Reload
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_setpc_b64 s[30:31]
+; SDAG-GFX942-LABEL: memcpy_known_medium:
+; SDAG-GFX942:       ; %bb.3:
+; SDAG-GFX942-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x0
+; SDAG-GFX942-NEXT:    s_load_dword s12, s[4:5], 0x10
+; SDAG-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX942-NEXT:    s_branch .LBB1_0
+; SDAG-GFX942-NEXT:    .p2align 8
+; SDAG-GFX942-NEXT:  ; %bb.4:
+; SDAG-GFX942-NEXT:  .LBB1_0:
+; SDAG-GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; SDAG-GFX942-NEXT:    s_load_dword s13, s[4:5], 0x34
+; SDAG-GFX942-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x44
+; SDAG-GFX942-NEXT:    s_load_dword s14, s[4:5], 0x54
+; SDAG-GFX942-NEXT:    s_mov_b32 s12, 0
+; SDAG-GFX942-NEXT:    s_mov_b32 s5, s12
+; SDAG-GFX942-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX942-NEXT:    s_mov_b32 s4, s3
+; SDAG-GFX942-NEXT:    s_or_b64 s[6:7], s[4:5], s[12:13]
+; SDAG-GFX942-NEXT:    s_mov_b32 s13, s2
+; SDAG-GFX942-NEXT:    s_mov_b32 s2, s1
+; SDAG-GFX942-NEXT:    s_mov_b32 s3, s12
+; SDAG-GFX942-NEXT:    s_or_b64 s[4:5], s[2:3], s[12:13]
+; SDAG-GFX942-NEXT:    s_mov_b32 s13, s14
+; SDAG-GFX942-NEXT:    s_mov_b32 s2, s11
+; SDAG-GFX942-NEXT:    s_or_b64 s[14:15], s[2:3], s[12:13]
+; SDAG-GFX942-NEXT:    s_mov_b32 s13, s10
+; SDAG-GFX942-NEXT:    s_mov_b32 s2, s9
+; SDAG-GFX942-NEXT:    s_or_b64 s[12:13], s[2:3], s[12:13]
+; SDAG-GFX942-NEXT:  .LBB1_1: ; %load-store-loop
+; SDAG-GFX942-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX942-NEXT:    v_add_u32_e32 v1, s0, v0
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[2:5], v1, s[4:7], 0 offen
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[6:9], v1, s[4:7], 0 offen offset:16
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[10:13], v1, s[4:7], 0 offen offset:32
+; SDAG-GFX942-NEXT:    v_add_u32_e32 v62, s8, v0
+; SDAG-GFX942-NEXT:    v_add_co_u32_e32 v0, vcc, 0x100, v0
+; SDAG-GFX942-NEXT:    s_and_b64 vcc, exec, vcc
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a0, v13 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a1, v12 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a2, v11 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a3, v10 ; Reload Reuse
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[14:17], v1, s[4:7], 0 offen offset:48
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[18:21], v1, s[4:7], 0 offen offset:64
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[22:25], v1, s[4:7], 0 offen offset:80
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[26:29], v1, s[4:7], 0 offen offset:96
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[30:33], v1, s[4:7], 0 offen offset:112
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[34:37], v1, s[4:7], 0 offen offset:128
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[38:41], v1, s[4:7], 0 offen offset:144
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[42:45], v1, s[4:7], 0 offen offset:160
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[46:49], v1, s[4:7], 0 offen offset:176
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[50:53], v1, s[4:7], 0 offen offset:192
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[54:57], v1, s[4:7], 0 offen offset:208
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[58:61], v1, s[4:7], 0 offen offset:224
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[10:13], v1, s[4:7], 0 offen offset:240
+; SDAG-GFX942-NEXT:    s_nop 0
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v62, s[12:15], 0 offen
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[6:9], v62, s[12:15], 0 offen offset:16
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(2)
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a4, v13 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v5, a0 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v4, a1 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v3, a2 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v2, a3 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a5, v12 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a6, v11 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_write_b32 a7, v10 ; Reload Reuse
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v62, s[12:15], 0 offen offset:32
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[14:17], v62, s[12:15], 0 offen offset:48
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[18:21], v62, s[12:15], 0 offen offset:64
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[22:25], v62, s[12:15], 0 offen offset:80
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[26:29], v62, s[12:15], 0 offen offset:96
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[30:33], v62, s[12:15], 0 offen offset:112
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[34:37], v62, s[12:15], 0 offen offset:128
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[38:41], v62, s[12:15], 0 offen offset:144
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[42:45], v62, s[12:15], 0 offen offset:160
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[46:49], v62, s[12:15], 0 offen offset:176
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[50:53], v62, s[12:15], 0 offen offset:192
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[54:57], v62, s[12:15], 0 offen offset:208
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[58:61], v62, s[12:15], 0 offen offset:224
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v5, a4 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v4, a5 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v3, a6 ; Reload Reuse
+; SDAG-GFX942-NEXT:    v_accvgpr_read_b32 v2, a7 ; Reload Reuse
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v62, s[12:15], 0 offen offset:240
+; SDAG-GFX942-NEXT:    s_cbranch_vccnz .LBB1_1
+; SDAG-GFX942-NEXT:  ; %bb.2: ; %memcpy-split
+; SDAG-GFX942-NEXT:    s_endpgm
+;
+; SDAG-GFX1100-LABEL: memcpy_known_medium:
+; SDAG-GFX1100:       ; %bb.0:
+; SDAG-GFX1100-NEXT:    s_clause 0x3
+; SDAG-GFX1100-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; SDAG-GFX1100-NEXT:    s_load_b32 s13, s[4:5], 0x34
+; SDAG-GFX1100-NEXT:    s_load_b128 s[8:11], s[4:5], 0x44
+; SDAG-GFX1100-NEXT:    s_load_b32 s18, s[4:5], 0x54
+; SDAG-GFX1100-NEXT:    s_mov_b32 s12, 0
+; SDAG-GFX1100-NEXT:    v_mov_b32_e32 v0, 0
+; SDAG-GFX1100-NEXT:    s_mov_b32 s5, s12
+; SDAG-GFX1100-NEXT:    s_mov_b32 s15, s12
+; SDAG-GFX1100-NEXT:    s_mov_b32 s17, s12
+; SDAG-GFX1100-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX1100-NEXT:    s_mov_b32 s4, s3
+; SDAG-GFX1100-NEXT:    s_mov_b32 s14, s1
+; SDAG-GFX1100-NEXT:    s_or_b64 s[6:7], s[4:5], s[12:13]
+; SDAG-GFX1100-NEXT:    s_mov_b32 s13, s2
+; SDAG-GFX1100-NEXT:    s_mov_b32 s16, s11
+; SDAG-GFX1100-NEXT:    s_or_b64 s[4:5], s[14:15], s[12:13]
+; SDAG-GFX1100-NEXT:    s_mov_b32 s13, s18
+; SDAG-GFX1100-NEXT:    s_mov_b32 s2, s9
+; SDAG-GFX1100-NEXT:    s_or_b64 s[14:15], s[16:17], s[12:13]
+; SDAG-GFX1100-NEXT:    s_mov_b32 s13, s10
+; SDAG-GFX1100-NEXT:    s_mov_b32 s3, s12
+; SDAG-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; SDAG-GFX1100-NEXT:    s_or_b64 s[12:13], s[2:3], s[12:13]
+; SDAG-GFX1100-NEXT:  .LBB1_1: ; %load-store-loop
+; SDAG-GFX1100-NEXT:    ; =>This Inner Loop Header: Depth=1
+; SDAG-GFX1100-NEXT:    v_add_nc_u32_e32 v61, s0, v0
+; SDAG-GFX1100-NEXT:    v_add_nc_u32_e32 v65, s8, v0
+; SDAG-GFX1100-NEXT:    v_add_co_u32 v0, s1, 0x100, v0
+; SDAG-GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; SDAG-GFX1100-NEXT:    s_and_b32 vcc_lo, exec_lo, s1
+; SDAG-GFX1100-NEXT:    s_clause 0xf
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[1:4], v61, s[4:7], 0 offen
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[5:8], v61, s[4:7], 0 offen offset:16
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[9:12], v61, s[4:7], 0 offen offset:32
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[13:16], v61, s[4:7], 0 offen offset:48
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[17:20], v61, s[4:7], 0 offen offset:64
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[21:24], v61, s[4:7], 0 offen offset:80
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[25:28], v61, s[4:7], 0 offen offset:96
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[29:32], v61, s[4:7], 0 offen offset:112
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[33:36], v61, s[4:7], 0 offen offset:128
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[37:40], v61, s[4:7], 0 offen offset:144
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[41:44], v61, s[4:7], 0 offen offset:160
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[45:48], v61, s[4:7], 0 offen offset:176
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[49:52], v61, s[4:7], 0 offen offset:192
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[53:56], v61, s[4:7], 0 offen offset:208
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[57:60], v61, s[4:7], 0 offen offset:224
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[61:64], v61, s[4:7], 0 offen offset:240
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(15)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[1:4], v65, s[12:15], 0 offen
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(14)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[5:8], v65, s[12:15], 0 offen offset:16
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(13)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[9:12], v65, s[12:15], 0 offen offset:32
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(12)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[13:16], v65, s[12:15], 0 offen offset:48
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(11)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[17:20], v65, s[12:15], 0 offen offset:64
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(10)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[21:24], v65, s[12:15], 0 offen offset:80
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(9)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[25:28], v65, s[12:15], 0 offen offset:96
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(8)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[29:32], v65, s[12:15], 0 offen offset:112
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(7)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[33:36], v65, s[12:15], 0 offen offset:128
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(6)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[37:40], v65, s[12:15], 0 offen offset:144
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(5)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[41:44], v65, s[12:15], 0 offen offset:160
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(4)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[45:48], v65, s[12:15], 0 offen offset:176
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(3)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[49:52], v65, s[12:15], 0 offen offset:192
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(2)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[53:56], v65, s[12:15], 0 offen offset:208
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(1)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[57:60], v65, s[12:15], 0 offen offset:224
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[61:64], v65, s[12:15], 0 offen offset:240
+; SDAG-GFX1100-NEXT:    s_cbranch_vccnz .LBB1_1
+; SDAG-GFX1100-NEXT:  ; %bb.2: ; %memcpy-split
+; SDAG-GFX1100-NEXT:    s_endpgm
+;
+; GISEL-GFX942-LABEL: memcpy_known_medium:
+; GISEL-GFX942:       ; %bb.0:
+; GISEL-GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GISEL-GFX942-NEXT:    s_load_dword s11, s[4:5], 0x34
+; GISEL-GFX942-NEXT:    s_mov_b32 s7, 0
+; GISEL-GFX942-NEXT:    s_load_dwordx4 s[12:15], s[4:5], 0x44
+; GISEL-GFX942-NEXT:    s_mov_b32 s8, s7
+; GISEL-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s1
+; GISEL-GFX942-NEXT:    s_mov_b32 s9, s2
+; GISEL-GFX942-NEXT:    s_or_b64 s[8:9], s[6:7], s[8:9]
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s3
+; GISEL-GFX942-NEXT:    s_load_dword s3, s[4:5], 0x54
+; GISEL-GFX942-NEXT:    s_mov_b32 s10, s7
+; GISEL-GFX942-NEXT:    s_or_b64 s[10:11], s[6:7], s[10:11]
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s13
+; GISEL-GFX942-NEXT:    s_mov_b32 s4, s7
+; GISEL-GFX942-NEXT:    s_mov_b32 s5, s14
+; GISEL-GFX942-NEXT:    s_mov_b32 s16, 0
+; GISEL-GFX942-NEXT:    s_or_b64 s[4:5], s[6:7], s[4:5]
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s15
+; GISEL-GFX942-NEXT:    s_mov_b32 s2, s7
+; GISEL-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX942-NEXT:    s_or_b64 s[6:7], s[6:7], s[2:3]
+; GISEL-GFX942-NEXT:    v_mov_b32_e32 v0, s16
+; GISEL-GFX942-NEXT:  .LBB1_1: ; %load-store-loop
+; GISEL-GFX942-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX942-NEXT:    v_add_u32_e32 v1, s0, v0
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[2:5], v1, s[8:11], 0 offen
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[6:9], v1, s[8:11], 0 offen offset:16
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[10:13], v1, s[8:11], 0 offen offset:32
+; GISEL-GFX942-NEXT:    v_add_u32_e32 v62, s12, v0
+; GISEL-GFX942-NEXT:    v_add_co_u32_e32 v0, vcc, 0x100, v0
+; GISEL-GFX942-NEXT:    s_xor_b64 s[2:3], vcc, -1
+; GISEL-GFX942-NEXT:    s_xor_b64 s[2:3], s[2:3], -1
+; GISEL-GFX942-NEXT:    s_and_b64 vcc, s[2:3], exec
+; GISEL-GFX942-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a0, v13 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a1, v12 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a2, v11 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a3, v10 ; Reload Reuse
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[14:17], v1, s[8:11], 0 offen offset:48
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[18:21], v1, s[8:11], 0 offen offset:64
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[22:25], v1, s[8:11], 0 offen offset:80
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[26:29], v1, s[8:11], 0 offen offset:96
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[30:33], v1, s[8:11], 0 offen offset:112
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[34:37], v1, s[8:11], 0 offen offset:128
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[38:41], v1, s[8:11], 0 offen offset:144
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[42:45], v1, s[8:11], 0 offen offset:160
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[46:49], v1, s[8:11], 0 offen offset:176
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[50:53], v1, s[8:11], 0 offen offset:192
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[54:57], v1, s[8:11], 0 offen offset:208
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[58:61], v1, s[8:11], 0 offen offset:224
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[10:13], v1, s[8:11], 0 offen offset:240
+; GISEL-GFX942-NEXT:    s_nop 0
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v62, s[4:7], 0 offen
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[6:9], v62, s[4:7], 0 offen offset:16
+; GISEL-GFX942-NEXT:    s_waitcnt vmcnt(2)
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a4, v13 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v5, a0 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v4, a1 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v3, a2 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v2, a3 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a5, v12 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a6, v11 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_write_b32 a7, v10 ; Reload Reuse
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v62, s[4:7], 0 offen offset:32
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[14:17], v62, s[4:7], 0 offen offset:48
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[18:21], v62, s[4:7], 0 offen offset:64
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[22:25], v62, s[4:7], 0 offen offset:80
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[26:29], v62, s[4:7], 0 offen offset:96
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[30:33], v62, s[4:7], 0 offen offset:112
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[34:37], v62, s[4:7], 0 offen offset:128
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[38:41], v62, s[4:7], 0 offen offset:144
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[42:45], v62, s[4:7], 0 offen offset:160
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[46:49], v62, s[4:7], 0 offen offset:176
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[50:53], v62, s[4:7], 0 offen offset:192
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[54:57], v62, s[4:7], 0 offen offset:208
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[58:61], v62, s[4:7], 0 offen offset:224
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v5, a4 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v4, a5 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v3, a6 ; Reload Reuse
+; GISEL-GFX942-NEXT:    v_accvgpr_read_b32 v2, a7 ; Reload Reuse
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[2:5], v62, s[4:7], 0 offen offset:240
+; GISEL-GFX942-NEXT:    s_cbranch_vccnz .LBB1_1
+; GISEL-GFX942-NEXT:  ; %bb.2: ; %memcpy-split
+; GISEL-GFX942-NEXT:    s_endpgm
+;
+; GISEL-GFX1100-LABEL: memcpy_known_medium:
+; GISEL-GFX1100:       ; %bb.0:
+; GISEL-GFX1100-NEXT:    s_clause 0x3
+; GISEL-GFX1100-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GISEL-GFX1100-NEXT:    s_load_b128 s[8:11], s[4:5], 0x44
+; GISEL-GFX1100-NEXT:    s_load_b32 s7, s[4:5], 0x34
+; GISEL-GFX1100-NEXT:    s_load_b32 s15, s[4:5], 0x54
+; GISEL-GFX1100-NEXT:    s_mov_b32 s17, 0
+; GISEL-GFX1100-NEXT:    s_mov_b32 s12, 0
+; GISEL-GFX1100-NEXT:    s_mov_b32 s4, s17
+; GISEL-GFX1100-NEXT:    s_mov_b32 s6, s17
+; GISEL-GFX1100-NEXT:    v_mov_b32_e32 v0, s12
+; GISEL-GFX1100-NEXT:    s_mov_b32 s14, s17
+; GISEL-GFX1100-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX1100-NEXT:    s_mov_b32 s16, s1
+; GISEL-GFX1100-NEXT:    s_mov_b32 s5, s2
+; GISEL-GFX1100-NEXT:    s_mov_b32 s2, s17
+; GISEL-GFX1100-NEXT:    s_or_b64 s[4:5], s[16:17], s[4:5]
+; GISEL-GFX1100-NEXT:    s_mov_b32 s16, s3
+; GISEL-GFX1100-NEXT:    s_mov_b32 s3, s10
+; GISEL-GFX1100-NEXT:    s_or_b64 s[6:7], s[16:17], s[6:7]
+; GISEL-GFX1100-NEXT:    s_mov_b32 s16, s9
+; GISEL-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GISEL-GFX1100-NEXT:    s_or_b64 s[12:13], s[16:17], s[2:3]
+; GISEL-GFX1100-NEXT:    s_mov_b32 s16, s11
+; GISEL-GFX1100-NEXT:    s_or_b64 s[14:15], s[16:17], s[14:15]
+; GISEL-GFX1100-NEXT:  .LBB1_1: ; %load-store-loop
+; GISEL-GFX1100-NEXT:    ; =>This Inner Loop Header: Depth=1
+; GISEL-GFX1100-NEXT:    v_add_nc_u32_e32 v61, s0, v0
+; GISEL-GFX1100-NEXT:    v_add_nc_u32_e32 v65, s8, v0
+; GISEL-GFX1100-NEXT:    v_add_co_u32 v0, s1, 0x100, v0
+; GISEL-GFX1100-NEXT:    s_delay_alu instid0(VALU_DEP_1)
+; GISEL-GFX1100-NEXT:    s_xor_b32 s1, s1, -1
+; GISEL-GFX1100-NEXT:    s_clause 0xf
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[1:4], v61, s[4:7], 0 offen
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[5:8], v61, s[4:7], 0 offen offset:16
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[9:12], v61, s[4:7], 0 offen offset:32
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[13:16], v61, s[4:7], 0 offen offset:48
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[17:20], v61, s[4:7], 0 offen offset:64
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[21:24], v61, s[4:7], 0 offen offset:80
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[25:28], v61, s[4:7], 0 offen offset:96
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[29:32], v61, s[4:7], 0 offen offset:112
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[33:36], v61, s[4:7], 0 offen offset:128
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[37:40], v61, s[4:7], 0 offen offset:144
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[41:44], v61, s[4:7], 0 offen offset:160
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[45:48], v61, s[4:7], 0 offen offset:176
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[49:52], v61, s[4:7], 0 offen offset:192
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[53:56], v61, s[4:7], 0 offen offset:208
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[57:60], v61, s[4:7], 0 offen offset:224
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[61:64], v61, s[4:7], 0 offen offset:240
+; GISEL-GFX1100-NEXT:    s_xor_b32 s1, s1, -1
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(15)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[1:4], v65, s[12:15], 0 offen
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(14)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[5:8], v65, s[12:15], 0 offen offset:16
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(13)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[9:12], v65, s[12:15], 0 offen offset:32
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(12)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[13:16], v65, s[12:15], 0 offen offset:48
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(11)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[17:20], v65, s[12:15], 0 offen offset:64
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(10)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[21:24], v65, s[12:15], 0 offen offset:80
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(9)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[25:28], v65, s[12:15], 0 offen offset:96
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(8)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[29:32], v65, s[12:15], 0 offen offset:112
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(7)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[33:36], v65, s[12:15], 0 offen offset:128
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(6)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[37:40], v65, s[12:15], 0 offen offset:144
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(5)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[41:44], v65, s[12:15], 0 offen offset:160
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(4)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[45:48], v65, s[12:15], 0 offen offset:176
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(3)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[49:52], v65, s[12:15], 0 offen offset:192
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(2)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[53:56], v65, s[12:15], 0 offen offset:208
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(1)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[57:60], v65, s[12:15], 0 offen offset:224
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[61:64], v65, s[12:15], 0 offen offset:240
+; GISEL-GFX1100-NEXT:    s_and_b32 vcc_lo, exec_lo, s1
+; GISEL-GFX1100-NEXT:    s_cbranch_vccnz .LBB1_1
+; GISEL-GFX1100-NEXT:  ; %bb.2: ; %memcpy-split
+; GISEL-GFX1100-NEXT:    s_endpgm
+  call void @llvm.memcpy.p7.p7.i32(ptr addrspace(7) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 256, i1 false)
+  ret void
+}
+
+define amdgpu_kernel void @memcpy_known_small(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; SDAG-LABEL: memcpy_known_small:
+; SDAG:       ; %bb.0:
+; SDAG-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; SDAG-NEXT:    v_mov_b32_e32 v4, s20
+; SDAG-NEXT:    buffer_load_dwordx4 v[0:3], v4, s[16:19], 0 offen
+; SDAG-NEXT:    s_mov_b32 s7, s24
+; SDAG-NEXT:    s_mov_b32 s6, s23
+; SDAG-NEXT:    s_mov_b32 s5, s22
+; SDAG-NEXT:    s_mov_b32 s4, s21
+; SDAG-NEXT:    v_mov_b32_e32 v5, s25
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dwordx4 v[0:3], v5, s[4:7], 0 offen
+; SDAG-NEXT:    buffer_load_dwordx4 v[0:3], v4, s[16:19], 0 offen offset:16
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    buffer_store_dwordx4 v[0:3], v5, s[4:7], 0 offen offset:16
+; SDAG-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-NEXT:    s_setpc_b64 s[30:31]
+;
+; GISEL-LABEL: memcpy_known_small:
+; GISEL:       ; %bb.0:
+; GISEL-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
+; GISEL-NEXT:    v_mov_b32_e32 v4, s20
+; GISEL-NEXT:    buffer_load_dwordx4 v[0:3], v4, s[16:19], 0 offen
+; GISEL-NEXT:    s_mov_b32 s4, s21
+; GISEL-NEXT:    s_mov_b32 s5, s22
+; GISEL-NEXT:    s_mov_b32 s6, s23
+; GISEL-NEXT:    s_mov_b32 s7, s24
+; GISEL-NEXT:    v_mov_b32_e32 v5, s25
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dwordx4 v[0:3], v5, s[4:7], 0 offen
+; GISEL-NEXT:    buffer_load_dwordx4 v[0:3], v4, s[16:19], 0 offen offset:16
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    buffer_store_dwordx4 v[0:3], v5, s[4:7], 0 offen offset:16
+; GISEL-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-NEXT:    s_setpc_b64 s[30:31]
+; SDAG-GFX942-LABEL: memcpy_known_small:
+; SDAG-GFX942:       ; %bb.1:
+; SDAG-GFX942-NEXT:    s_load_dwordx4 s[8:11], s[4:5], 0x0
+; SDAG-GFX942-NEXT:    s_load_dword s12, s[4:5], 0x10
+; SDAG-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX942-NEXT:    s_branch .LBB2_0
+; SDAG-GFX942-NEXT:    .p2align 8
+; SDAG-GFX942-NEXT:  ; %bb.2:
+; SDAG-GFX942-NEXT:  .LBB2_0:
+; SDAG-GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; SDAG-GFX942-NEXT:    s_load_dword s13, s[4:5], 0x34
+; SDAG-GFX942-NEXT:    s_mov_b32 s12, 0
+; SDAG-GFX942-NEXT:    s_mov_b32 s7, s12
+; SDAG-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX942-NEXT:    s_mov_b32 s6, s3
+; SDAG-GFX942-NEXT:    s_or_b64 s[10:11], s[6:7], s[12:13]
+; SDAG-GFX942-NEXT:    s_mov_b32 s13, s2
+; SDAG-GFX942-NEXT:    s_mov_b32 s2, s1
+; SDAG-GFX942-NEXT:    s_mov_b32 s3, s12
+; SDAG-GFX942-NEXT:    s_or_b64 s[8:9], s[2:3], s[12:13]
+; SDAG-GFX942-NEXT:    v_mov_b32_e32 v4, s0
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[0:3], v4, s[8:11], 0 offen
+; SDAG-GFX942-NEXT:    s_load_dword s13, s[4:5], 0x54
+; SDAG-GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x44
+; SDAG-GFX942-NEXT:    s_mov_b32 s5, s12
+; SDAG-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX942-NEXT:    s_mov_b32 s4, s3
+; SDAG-GFX942-NEXT:    s_or_b64 s[6:7], s[4:5], s[12:13]
+; SDAG-GFX942-NEXT:    s_mov_b32 s13, s2
+; SDAG-GFX942-NEXT:    s_mov_b32 s2, s1
+; SDAG-GFX942-NEXT:    s_mov_b32 s3, s12
+; SDAG-GFX942-NEXT:    s_or_b64 s[4:5], s[2:3], s[12:13]
+; SDAG-GFX942-NEXT:    v_mov_b32_e32 v5, s0
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[0:3], v5, s[4:7], 0 offen
+; SDAG-GFX942-NEXT:    buffer_load_dwordx4 v[0:3], v4, s[8:11], 0 offen offset:16
+; SDAG-GFX942-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX942-NEXT:    buffer_store_dwordx4 v[0:3], v5, s[4:7], 0 offen offset:16
+; SDAG-GFX942-NEXT:    s_endpgm
+;
+; SDAG-GFX1100-LABEL: memcpy_known_small:
+; SDAG-GFX1100:       ; %bb.0:
+; SDAG-GFX1100-NEXT:    s_clause 0x1
+; SDAG-GFX1100-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; SDAG-GFX1100-NEXT:    s_load_b32 s13, s[4:5], 0x34
+; SDAG-GFX1100-NEXT:    s_mov_b32 s12, 0
+; SDAG-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; SDAG-GFX1100-NEXT:    s_mov_b32 s7, s12
+; SDAG-GFX1100-NEXT:    s_mov_b32 s9, s12
+; SDAG-GFX1100-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX1100-NEXT:    s_mov_b32 s6, s3
+; SDAG-GFX1100-NEXT:    v_mov_b32_e32 v4, s0
+; SDAG-GFX1100-NEXT:    s_mov_b32 s8, s1
+; SDAG-GFX1100-NEXT:    s_or_b64 s[10:11], s[6:7], s[12:13]
+; SDAG-GFX1100-NEXT:    s_mov_b32 s13, s2
+; SDAG-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; SDAG-GFX1100-NEXT:    s_or_b64 s[8:9], s[8:9], s[12:13]
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[0:3], v4, s[8:11], 0 offen
+; SDAG-GFX1100-NEXT:    s_clause 0x1
+; SDAG-GFX1100-NEXT:    s_load_b32 s13, s[4:5], 0x54
+; SDAG-GFX1100-NEXT:    s_load_b128 s[0:3], s[4:5], 0x44
+; SDAG-GFX1100-NEXT:    s_mov_b32 s5, s12
+; SDAG-GFX1100-NEXT:    s_waitcnt lgkmcnt(0)
+; SDAG-GFX1100-NEXT:    v_mov_b32_e32 v5, s0
+; SDAG-GFX1100-NEXT:    s_mov_b32 s4, s3
+; SDAG-GFX1100-NEXT:    s_mov_b32 s3, s12
+; SDAG-GFX1100-NEXT:    s_or_b64 s[6:7], s[4:5], s[12:13]
+; SDAG-GFX1100-NEXT:    s_mov_b32 s13, s2
+; SDAG-GFX1100-NEXT:    s_mov_b32 s2, s1
+; SDAG-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; SDAG-GFX1100-NEXT:    s_or_b64 s[4:5], s[2:3], s[12:13]
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[0:3], v5, s[4:7], 0 offen
+; SDAG-GFX1100-NEXT:    buffer_load_b128 v[0:3], v4, s[8:11], 0 offen offset:16
+; SDAG-GFX1100-NEXT:    s_waitcnt vmcnt(0)
+; SDAG-GFX1100-NEXT:    buffer_store_b128 v[0:3], v5, s[4:7], 0 offen offset:16
+; SDAG-GFX1100-NEXT:    s_endpgm
+;
+; GISEL-GFX942-LABEL: memcpy_known_small:
+; GISEL-GFX942:       ; %bb.0:
+; GISEL-GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x24
+; GISEL-GFX942-NEXT:    s_load_dword s11, s[4:5], 0x34
+; GISEL-GFX942-NEXT:    s_mov_b32 s7, 0
+; GISEL-GFX942-NEXT:    s_mov_b32 s8, s7
+; GISEL-GFX942-NEXT:    s_mov_b32 s10, s7
+; GISEL-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s1
+; GISEL-GFX942-NEXT:    s_mov_b32 s9, s2
+; GISEL-GFX942-NEXT:    s_or_b64 s[8:9], s[6:7], s[8:9]
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s3
+; GISEL-GFX942-NEXT:    s_or_b64 s[10:11], s[6:7], s[10:11]
+; GISEL-GFX942-NEXT:    v_mov_b32_e32 v4, s0
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[0:3], v4, s[8:11], 0 offen
+; GISEL-GFX942-NEXT:    s_load_dwordx4 s[0:3], s[4:5], 0x44
+; GISEL-GFX942-NEXT:    s_load_dword s13, s[4:5], 0x54
+; GISEL-GFX942-NEXT:    s_mov_b32 s4, s7
+; GISEL-GFX942-NEXT:    s_mov_b32 s12, s7
+; GISEL-GFX942-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s1
+; GISEL-GFX942-NEXT:    s_mov_b32 s5, s2
+; GISEL-GFX942-NEXT:    s_or_b64 s[4:5], s[6:7], s[4:5]
+; GISEL-GFX942-NEXT:    s_mov_b32 s6, s3
+; GISEL-GFX942-NEXT:    s_or_b64 s[6:7], s[6:7], s[12:13]
+; GISEL-GFX942-NEXT:    v_mov_b32_e32 v5, s0
+; GISEL-GFX942-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[0:3], v5, s[4:7], 0 offen
+; GISEL-GFX942-NEXT:    buffer_load_dwordx4 v[0:3], v4, s[8:11], 0 offen offset:16
+; GISEL-GFX942-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX942-NEXT:    buffer_store_dwordx4 v[0:3], v5, s[4:7], 0 offen offset:16
+; GISEL-GFX942-NEXT:    s_endpgm
+;
+; GISEL-GFX1100-LABEL: memcpy_known_small:
+; GISEL-GFX1100:       ; %bb.0:
+; GISEL-GFX1100-NEXT:    s_clause 0x1
+; GISEL-GFX1100-NEXT:    s_load_b128 s[0:3], s[4:5], 0x24
+; GISEL-GFX1100-NEXT:    s_load_b32 s7, s[4:5], 0x34
+; GISEL-GFX1100-NEXT:    s_mov_b32 s13, 0
+; GISEL-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GISEL-GFX1100-NEXT:    s_mov_b32 s8, s13
+; GISEL-GFX1100-NEXT:    s_mov_b32 s6, s13
+; GISEL-GFX1100-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX1100-NEXT:    s_mov_b32 s12, s1
+; GISEL-GFX1100-NEXT:    s_mov_b32 s9, s2
+; GISEL-GFX1100-NEXT:    v_mov_b32_e32 v4, s0
+; GISEL-GFX1100-NEXT:    s_or_b64 s[0:1], s[12:13], s[8:9]
+; GISEL-GFX1100-NEXT:    s_mov_b32 s12, s3
+; GISEL-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1)
+; GISEL-GFX1100-NEXT:    s_or_b64 s[2:3], s[12:13], s[6:7]
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[0:3], v4, s[0:3], 0 offen
+; GISEL-GFX1100-NEXT:    s_clause 0x1
+; GISEL-GFX1100-NEXT:    s_load_b128 s[8:11], s[4:5], 0x44
+; GISEL-GFX1100-NEXT:    s_load_b32 s7, s[4:5], 0x54
+; GISEL-GFX1100-NEXT:    s_mov_b32 s4, s13
+; GISEL-GFX1100-NEXT:    s_waitcnt lgkmcnt(0)
+; GISEL-GFX1100-NEXT:    v_mov_b32_e32 v5, s8
+; GISEL-GFX1100-NEXT:    s_mov_b32 s12, s9
+; GISEL-GFX1100-NEXT:    s_mov_b32 s5, s10
+; GISEL-GFX1100-NEXT:    s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1)
+; GISEL-GFX1100-NEXT:    s_or_b64 s[4:5], s[12:13], s[4:5]
+; GISEL-GFX1100-NEXT:    s_mov_b32 s12, s11
+; GISEL-GFX1100-NEXT:    s_or_b64 s[6:7], s[12:13], s[6:7]
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[0:3], v5, s[4:7], 0 offen
+; GISEL-GFX1100-NEXT:    buffer_load_b128 v[0:3], v4, s[0:3], 0 offen offset:16
+; GISEL-GFX1100-NEXT:    s_waitcnt vmcnt(0)
+; GISEL-GFX1100-NEXT:    buffer_store_b128 v[0:3], v5, s[4:7], 0 offen offset:16
+; GISEL-GFX1100-NEXT:    s_endpgm
+  call void @llvm.memcpy.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 32, i1 false)
+  ret void
+}
diff --git a/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-mem-transfer.ll b/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-mem-transfer.ll
new file mode 100644
index 0000000000000..e6c2d1907068f
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/lower-buffer-fat-pointers-mem-transfer.ll
@@ -0,0 +1,1730 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S -mcpu=gfx900 -amdgpu-lower-buffer-fat-pointers < %s | FileCheck %s
+; RUN: opt -S -mcpu=gfx900 -passes=amdgpu-lower-buffer-fat-pointers < %s | FileCheck %s
+
+target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-p7:160:256:256:32-p8:128:128-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7:8"
+target triple = "amdgcn--"
+
+;; memcpy
+
+declare void @llvm.memcpy.p7.p7.i32(ptr addrspace(7), ptr addrspace(7), i32, i1)
+declare void @llvm.memcpy.p1.p7.i32(ptr addrspace(1), ptr addrspace(7), i32, i1)
+declare void @llvm.memcpy.p7.p1.i32(ptr addrspace(7), ptr addrspace(1), i32, i1)
+declare void @llvm.memcpy.p7.p7.i64(ptr addrspace(7), ptr addrspace(7), i64, i1)
+declare void @llvm.memcpy.p3.p7.i32(ptr addrspace(3), ptr addrspace(7), i32, i1)
+
+define void @memcpy_known(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTOFF_0:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[TMP1]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_0:%.*]] = shufflevector <4 x i32> [[DOTOFF_0]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_0:%.*]] = shufflevector <64 x i32> poison, <64 x i32> [[DOTEXT_0]], <64 x i32> <i32 64, i32 65, i32 66, i32 67, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_16:%.*]] = add nuw i32 [[TMP1]], 16
+; CHECK-NEXT:    [[DOTOFF_16:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_4:%.*]] = shufflevector <4 x i32> [[DOTOFF_16]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_4:%.*]] = shufflevector <64 x i32> [[DOTPARTS_0]], <64 x i32> [[DOTEXT_4]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 64, i32 65, i32 66, i32 67, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_32:%.*]] = add nuw i32 [[TMP1]], 32
+; CHECK-NEXT:    [[DOTOFF_32:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_8:%.*]] = shufflevector <4 x i32> [[DOTOFF_32]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_8:%.*]] = shufflevector <64 x i32> [[DOTPARTS_4]], <64 x i32> [[DOTEXT_8]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 64, i32 65, i32 66, i32 67, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_48:%.*]] = add nuw i32 [[TMP1]], 48
+; CHECK-NEXT:    [[DOTOFF_48:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_12:%.*]] = shufflevector <4 x i32> [[DOTOFF_48]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_12:%.*]] = shufflevector <64 x i32> [[DOTPARTS_8]], <64 x i32> [[DOTEXT_12]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 64, i32 65, i32 66, i32 67, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_64:%.*]] = add nuw i32 [[TMP1]], 64
+; CHECK-NEXT:    [[DOTOFF_64:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_64]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_16:%.*]] = shufflevector <4 x i32> [[DOTOFF_64]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_16:%.*]] = shufflevector <64 x i32> [[DOTPARTS_12]], <64 x i32> [[DOTEXT_16]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_80:%.*]] = add nuw i32 [[TMP1]], 80
+; CHECK-NEXT:    [[DOTOFF_80:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_80]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_20:%.*]] = shufflevector <4 x i32> [[DOTOFF_80]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_20:%.*]] = shufflevector <64 x i32> [[DOTPARTS_16]], <64 x i32> [[DOTEXT_20]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 64, i32 65, i32 66, i32 67, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_96:%.*]] = add nuw i32 [[TMP1]], 96
+; CHECK-NEXT:    [[DOTOFF_96:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_96]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_24:%.*]] = shufflevector <4 x i32> [[DOTOFF_96]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_24:%.*]] = shufflevector <64 x i32> [[DOTPARTS_20]], <64 x i32> [[DOTEXT_24]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 64, i32 65, i32 66, i32 67, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_112:%.*]] = add nuw i32 [[TMP1]], 112
+; CHECK-NEXT:    [[DOTOFF_112:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_112]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_28:%.*]] = shufflevector <4 x i32> [[DOTOFF_112]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_28:%.*]] = shufflevector <64 x i32> [[DOTPARTS_24]], <64 x i32> [[DOTEXT_28]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 64, i32 65, i32 66, i32 67, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_128:%.*]] = add nuw i32 [[TMP1]], 128
+; CHECK-NEXT:    [[DOTOFF_128:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_128]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_32:%.*]] = shufflevector <4 x i32> [[DOTOFF_128]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_32:%.*]] = shufflevector <64 x i32> [[DOTPARTS_28]], <64 x i32> [[DOTEXT_32]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 64, i32 65, i32 66, i32 67, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_144:%.*]] = add nuw i32 [[TMP1]], 144
+; CHECK-NEXT:    [[DOTOFF_144:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_144]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_36:%.*]] = shufflevector <4 x i32> [[DOTOFF_144]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_36:%.*]] = shufflevector <64 x i32> [[DOTPARTS_32]], <64 x i32> [[DOTEXT_36]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 64, i32 65, i32 66, i32 67, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_160:%.*]] = add nuw i32 [[TMP1]], 160
+; CHECK-NEXT:    [[DOTOFF_160:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_160]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_40:%.*]] = shufflevector <4 x i32> [[DOTOFF_160]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_40:%.*]] = shufflevector <64 x i32> [[DOTPARTS_36]], <64 x i32> [[DOTEXT_40]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 64, i32 65, i32 66, i32 67, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_176:%.*]] = add nuw i32 [[TMP1]], 176
+; CHECK-NEXT:    [[DOTOFF_176:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_176]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_44:%.*]] = shufflevector <4 x i32> [[DOTOFF_176]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_44:%.*]] = shufflevector <64 x i32> [[DOTPARTS_40]], <64 x i32> [[DOTEXT_44]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 64, i32 65, i32 66, i32 67, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_192:%.*]] = add nuw i32 [[TMP1]], 192
+; CHECK-NEXT:    [[DOTOFF_192:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_192]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_48:%.*]] = shufflevector <4 x i32> [[DOTOFF_192]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_48:%.*]] = shufflevector <64 x i32> [[DOTPARTS_44]], <64 x i32> [[DOTEXT_48]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 64, i32 65, i32 66, i32 67, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_208:%.*]] = add nuw i32 [[TMP1]], 208
+; CHECK-NEXT:    [[DOTOFF_208:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_208]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_52:%.*]] = shufflevector <4 x i32> [[DOTOFF_208]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_52:%.*]] = shufflevector <64 x i32> [[DOTPARTS_48]], <64 x i32> [[DOTEXT_52]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 64, i32 65, i32 66, i32 67, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_224:%.*]] = add nuw i32 [[TMP1]], 224
+; CHECK-NEXT:    [[DOTOFF_224:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_224]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_56:%.*]] = shufflevector <4 x i32> [[DOTOFF_224]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_56:%.*]] = shufflevector <64 x i32> [[DOTPARTS_52]], <64 x i32> [[DOTEXT_56]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 64, i32 65, i32 66, i32 67, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_240:%.*]] = add nuw i32 [[TMP1]], 240
+; CHECK-NEXT:    [[DOTOFF_240:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_240]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_60:%.*]] = shufflevector <4 x i32> [[DOTOFF_240]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <64 x i32> [[DOTPARTS_56]], <64 x i32> [[DOTEXT_60]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 64, i32 65, i32 66, i32 67>
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[DST_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTSLICE_0:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_0]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[TMP3]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_4:%.*]] = add nuw i32 [[TMP3]], 16
+; CHECK-NEXT:    [[DOTSLICE_4:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_4]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_4]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_8:%.*]] = add nuw i32 [[TMP3]], 32
+; CHECK-NEXT:    [[DOTSLICE_8:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_8]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_8]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_12:%.*]] = add nuw i32 [[TMP3]], 48
+; CHECK-NEXT:    [[DOTSLICE_12:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_12]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_12]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_16:%.*]] = add nuw i32 [[TMP3]], 64
+; CHECK-NEXT:    [[DOTSLICE_16:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 16, i32 17, i32 18, i32 19>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_16]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_20:%.*]] = add nuw i32 [[TMP3]], 80
+; CHECK-NEXT:    [[DOTSLICE_20:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_20]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_20]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_24:%.*]] = add nuw i32 [[TMP3]], 96
+; CHECK-NEXT:    [[DOTSLICE_24:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 24, i32 25, i32 26, i32 27>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_24]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_24]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_28:%.*]] = add nuw i32 [[TMP3]], 112
+; CHECK-NEXT:    [[DOTSLICE_28:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_28]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_28]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_32:%.*]] = add nuw i32 [[TMP3]], 128
+; CHECK-NEXT:    [[DOTSLICE_32:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 32, i32 33, i32 34, i32 35>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_32]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_36:%.*]] = add nuw i32 [[TMP3]], 144
+; CHECK-NEXT:    [[DOTSLICE_36:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 36, i32 37, i32 38, i32 39>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_36]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_36]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_40:%.*]] = add nuw i32 [[TMP3]], 160
+; CHECK-NEXT:    [[DOTSLICE_40:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 40, i32 41, i32 42, i32 43>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_40]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_40]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_44:%.*]] = add nuw i32 [[TMP3]], 176
+; CHECK-NEXT:    [[DOTSLICE_44:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 44, i32 45, i32 46, i32 47>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_44]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_44]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_48:%.*]] = add nuw i32 [[TMP3]], 192
+; CHECK-NEXT:    [[DOTSLICE_48:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 48, i32 49, i32 50, i32 51>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_48]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_52:%.*]] = add nuw i32 [[TMP3]], 208
+; CHECK-NEXT:    [[DOTSLICE_52:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 52, i32 53, i32 54, i32 55>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_52]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_52]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_56:%.*]] = add nuw i32 [[TMP3]], 224
+; CHECK-NEXT:    [[DOTSLICE_56:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 56, i32 57, i32 58, i32 59>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_56]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_56]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_60:%.*]] = add nuw i32 [[TMP3]], 240
+; CHECK-NEXT:    [[DOTSLICE_60:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_60]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_60]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p7.p7.i32(ptr addrspace(7) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 8192, i1 false)
+  ret void
+}
+
+define void @memcpy_known_small(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_small(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP1]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DST_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = add nuw i32 [[SRC_OFF]], 16
+; CHECK-NEXT:    [[TMP3:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4:%.*]] = add nuw i32 [[DST_OFF]], 16
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP3]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP4]], i32 0, i32 0)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 32, i1 false)
+  ret void
+}
+
+define void @memcpy_known_byte(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_byte(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call i8 @llvm.amdgcn.raw.ptr.buffer.load.i8(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 [[TMP1]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DST_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 1, i1 false)
+  ret void
+}
+
+define void @memcpy_known_tail(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_tail(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call i64 @llvm.amdgcn.raw.ptr.buffer.load.i64(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i64(i64 [[TMP1]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DST_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = add nuw i32 [[SRC_OFF]], 8
+; CHECK-NEXT:    [[TMP3:%.*]] = call i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4:%.*]] = add nuw i32 [[DST_OFF]], 8
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 [[TMP3]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP4]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP5:%.*]] = add nuw i32 [[SRC_OFF]], 12
+; CHECK-NEXT:    [[TMP6:%.*]] = call i16 @llvm.amdgcn.raw.ptr.buffer.load.i16(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP5]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP7:%.*]] = add nuw i32 [[DST_OFF]], 12
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i16(i16 [[TMP6]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP7]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP8:%.*]] = add nuw i32 [[SRC_OFF]], 14
+; CHECK-NEXT:    [[TMP9:%.*]] = call i8 @llvm.amdgcn.raw.ptr.buffer.load.i8(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP8]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP10:%.*]] = add nuw i32 [[DST_OFF]], 14
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 [[TMP9]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP10]], i32 0, i32 0)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 15, i1 false)
+  ret void
+}
+
+define void @memcpy_known_i64(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_i64(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i64 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[LOOP_INDEX_C:%.*]] = trunc i64 [[LOOP_INDEX]] to i32
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX_C]]
+; CHECK-NEXT:    [[DOTOFF_0:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP1]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_0:%.*]] = shufflevector <4 x i32> [[DOTOFF_0]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_0:%.*]] = shufflevector <64 x i32> poison, <64 x i32> [[DOTEXT_0]], <64 x i32> <i32 64, i32 65, i32 66, i32 67, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_16:%.*]] = add nuw i32 [[TMP1]], 16
+; CHECK-NEXT:    [[DOTOFF_16:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_4:%.*]] = shufflevector <4 x i32> [[DOTOFF_16]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_4:%.*]] = shufflevector <64 x i32> [[DOTPARTS_0]], <64 x i32> [[DOTEXT_4]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 64, i32 65, i32 66, i32 67, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_32:%.*]] = add nuw i32 [[TMP1]], 32
+; CHECK-NEXT:    [[DOTOFF_32:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_8:%.*]] = shufflevector <4 x i32> [[DOTOFF_32]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_8:%.*]] = shufflevector <64 x i32> [[DOTPARTS_4]], <64 x i32> [[DOTEXT_8]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 64, i32 65, i32 66, i32 67, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_48:%.*]] = add nuw i32 [[TMP1]], 48
+; CHECK-NEXT:    [[DOTOFF_48:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_12:%.*]] = shufflevector <4 x i32> [[DOTOFF_48]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_12:%.*]] = shufflevector <64 x i32> [[DOTPARTS_8]], <64 x i32> [[DOTEXT_12]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 64, i32 65, i32 66, i32 67, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_64:%.*]] = add nuw i32 [[TMP1]], 64
+; CHECK-NEXT:    [[DOTOFF_64:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_64]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_16:%.*]] = shufflevector <4 x i32> [[DOTOFF_64]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_16:%.*]] = shufflevector <64 x i32> [[DOTPARTS_12]], <64 x i32> [[DOTEXT_16]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_80:%.*]] = add nuw i32 [[TMP1]], 80
+; CHECK-NEXT:    [[DOTOFF_80:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_80]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_20:%.*]] = shufflevector <4 x i32> [[DOTOFF_80]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_20:%.*]] = shufflevector <64 x i32> [[DOTPARTS_16]], <64 x i32> [[DOTEXT_20]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 64, i32 65, i32 66, i32 67, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_96:%.*]] = add nuw i32 [[TMP1]], 96
+; CHECK-NEXT:    [[DOTOFF_96:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_96]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_24:%.*]] = shufflevector <4 x i32> [[DOTOFF_96]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_24:%.*]] = shufflevector <64 x i32> [[DOTPARTS_20]], <64 x i32> [[DOTEXT_24]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 64, i32 65, i32 66, i32 67, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_112:%.*]] = add nuw i32 [[TMP1]], 112
+; CHECK-NEXT:    [[DOTOFF_112:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_112]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_28:%.*]] = shufflevector <4 x i32> [[DOTOFF_112]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_28:%.*]] = shufflevector <64 x i32> [[DOTPARTS_24]], <64 x i32> [[DOTEXT_28]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 64, i32 65, i32 66, i32 67, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_128:%.*]] = add nuw i32 [[TMP1]], 128
+; CHECK-NEXT:    [[DOTOFF_128:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_128]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_32:%.*]] = shufflevector <4 x i32> [[DOTOFF_128]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_32:%.*]] = shufflevector <64 x i32> [[DOTPARTS_28]], <64 x i32> [[DOTEXT_32]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 64, i32 65, i32 66, i32 67, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_144:%.*]] = add nuw i32 [[TMP1]], 144
+; CHECK-NEXT:    [[DOTOFF_144:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_144]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_36:%.*]] = shufflevector <4 x i32> [[DOTOFF_144]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_36:%.*]] = shufflevector <64 x i32> [[DOTPARTS_32]], <64 x i32> [[DOTEXT_36]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 64, i32 65, i32 66, i32 67, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_160:%.*]] = add nuw i32 [[TMP1]], 160
+; CHECK-NEXT:    [[DOTOFF_160:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_160]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_40:%.*]] = shufflevector <4 x i32> [[DOTOFF_160]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_40:%.*]] = shufflevector <64 x i32> [[DOTPARTS_36]], <64 x i32> [[DOTEXT_40]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 64, i32 65, i32 66, i32 67, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_176:%.*]] = add nuw i32 [[TMP1]], 176
+; CHECK-NEXT:    [[DOTOFF_176:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_176]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_44:%.*]] = shufflevector <4 x i32> [[DOTOFF_176]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_44:%.*]] = shufflevector <64 x i32> [[DOTPARTS_40]], <64 x i32> [[DOTEXT_44]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 64, i32 65, i32 66, i32 67, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_192:%.*]] = add nuw i32 [[TMP1]], 192
+; CHECK-NEXT:    [[DOTOFF_192:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_192]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_48:%.*]] = shufflevector <4 x i32> [[DOTOFF_192]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_48:%.*]] = shufflevector <64 x i32> [[DOTPARTS_44]], <64 x i32> [[DOTEXT_48]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 64, i32 65, i32 66, i32 67, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_208:%.*]] = add nuw i32 [[TMP1]], 208
+; CHECK-NEXT:    [[DOTOFF_208:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_208]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_52:%.*]] = shufflevector <4 x i32> [[DOTOFF_208]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_52:%.*]] = shufflevector <64 x i32> [[DOTPARTS_48]], <64 x i32> [[DOTEXT_52]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 64, i32 65, i32 66, i32 67, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_224:%.*]] = add nuw i32 [[TMP1]], 224
+; CHECK-NEXT:    [[DOTOFF_224:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_224]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_56:%.*]] = shufflevector <4 x i32> [[DOTOFF_224]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_56:%.*]] = shufflevector <64 x i32> [[DOTPARTS_52]], <64 x i32> [[DOTEXT_56]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 64, i32 65, i32 66, i32 67, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_240:%.*]] = add nuw i32 [[TMP1]], 240
+; CHECK-NEXT:    [[DOTOFF_240:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_240]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_60:%.*]] = shufflevector <4 x i32> [[DOTOFF_240]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <64 x i32> [[DOTPARTS_56]], <64 x i32> [[DOTEXT_60]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 64, i32 65, i32 66, i32 67>
+; CHECK-NEXT:    [[LOOP_INDEX_C1:%.*]] = trunc i64 [[LOOP_INDEX]] to i32
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[DST_OFF]], [[LOOP_INDEX_C1]]
+; CHECK-NEXT:    [[DOTSLICE_0:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_0]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP3]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_4:%.*]] = add nuw i32 [[TMP3]], 16
+; CHECK-NEXT:    [[DOTSLICE_4:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_4]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_4]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_8:%.*]] = add nuw i32 [[TMP3]], 32
+; CHECK-NEXT:    [[DOTSLICE_8:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_8]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_8]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_12:%.*]] = add nuw i32 [[TMP3]], 48
+; CHECK-NEXT:    [[DOTSLICE_12:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_12]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_12]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_16:%.*]] = add nuw i32 [[TMP3]], 64
+; CHECK-NEXT:    [[DOTSLICE_16:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 16, i32 17, i32 18, i32 19>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_16]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_20:%.*]] = add nuw i32 [[TMP3]], 80
+; CHECK-NEXT:    [[DOTSLICE_20:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_20]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_20]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_24:%.*]] = add nuw i32 [[TMP3]], 96
+; CHECK-NEXT:    [[DOTSLICE_24:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 24, i32 25, i32 26, i32 27>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_24]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_24]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_28:%.*]] = add nuw i32 [[TMP3]], 112
+; CHECK-NEXT:    [[DOTSLICE_28:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_28]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_28]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_32:%.*]] = add nuw i32 [[TMP3]], 128
+; CHECK-NEXT:    [[DOTSLICE_32:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 32, i32 33, i32 34, i32 35>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_32]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_36:%.*]] = add nuw i32 [[TMP3]], 144
+; CHECK-NEXT:    [[DOTSLICE_36:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 36, i32 37, i32 38, i32 39>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_36]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_36]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_40:%.*]] = add nuw i32 [[TMP3]], 160
+; CHECK-NEXT:    [[DOTSLICE_40:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 40, i32 41, i32 42, i32 43>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_40]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_40]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_44:%.*]] = add nuw i32 [[TMP3]], 176
+; CHECK-NEXT:    [[DOTSLICE_44:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 44, i32 45, i32 46, i32 47>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_44]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_44]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_48:%.*]] = add nuw i32 [[TMP3]], 192
+; CHECK-NEXT:    [[DOTSLICE_48:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 48, i32 49, i32 50, i32 51>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_48]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_52:%.*]] = add nuw i32 [[TMP3]], 208
+; CHECK-NEXT:    [[DOTSLICE_52:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 52, i32 53, i32 54, i32 55>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_52]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_52]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_56:%.*]] = add nuw i32 [[TMP3]], 224
+; CHECK-NEXT:    [[DOTSLICE_56:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 56, i32 57, i32 58, i32 59>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_56]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_56]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_60:%.*]] = add nuw i32 [[TMP3]], 240
+; CHECK-NEXT:    [[DOTSLICE_60:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_60]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_60]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4]] = add i64 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i64 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p7.p7.i64(ptr addrspace(7) %dst, ptr addrspace(7) %src, i64 8192, i1 false)
+  ret void
+}
+
+define void @memcpy_known_i32_volatile(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_i32_volatile(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP1]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DST_OFF]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    [[TMP2:%.*]] = add nuw i32 [[SRC_OFF]], 16
+; CHECK-NEXT:    [[TMP3:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP2]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    [[TMP4:%.*]] = add nuw i32 [[DST_OFF]], 16
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP3]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP4]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 32, i1 true)
+  ret void
+}
+
+define void @memcpy_unknown(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst, i32 inreg %length) {
+; CHECK-LABEL: define void @memcpy_unknown(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]], i32 inreg [[LENGTH:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[LENGTH]], 15
+; CHECK-NEXT:    [[TMP2:%.*]] = sub i32 [[LENGTH]], [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne i32 [[TMP2]], 0
+; CHECK-NEXT:    br i1 [[TMP3]], label %[[LOOP_MEMCPY_EXPANSION:.*]], label %[[LOOP_MEMCPY_RESIDUAL_HEADER:.*]]
+; CHECK:       [[LOOP_MEMCPY_EXPANSION]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP7:%.*]], %[[LOOP_MEMCPY_EXPANSION]] ]
+; CHECK-NEXT:    [[TMP4:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP4]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP6:%.*]] = add i32 [[DST_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP5]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP6]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP7]] = add i32 [[LOOP_INDEX]], 16
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp ult i32 [[TMP7]], [[TMP2]]
+; CHECK-NEXT:    br i1 [[TMP8]], label %[[LOOP_MEMCPY_EXPANSION]], label %[[LOOP_MEMCPY_RESIDUAL_HEADER]]
+; CHECK:       [[LOOP_MEMCPY_RESIDUAL:.*]]:
+; CHECK-NEXT:    [[RESIDUAL_LOOP_INDEX:%.*]] = phi i32 [ 0, %[[LOOP_MEMCPY_RESIDUAL_HEADER]] ], [ [[TMP13:%.*]], %[[LOOP_MEMCPY_RESIDUAL]] ]
+; CHECK-NEXT:    [[TMP9:%.*]] = add i32 [[TMP2]], [[RESIDUAL_LOOP_INDEX]]
+; CHECK-NEXT:    [[TMP10:%.*]] = add i32 [[SRC_OFF]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = call i8 @llvm.amdgcn.raw.ptr.buffer.load.i8(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP10]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[DST_OFF]], [[TMP9]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 [[TMP11]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP12]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP13]] = add i32 [[RESIDUAL_LOOP_INDEX]], 1
+; CHECK-NEXT:    [[TMP14:%.*]] = icmp ult i32 [[TMP13]], [[TMP1]]
+; CHECK-NEXT:    br i1 [[TMP14]], label %[[LOOP_MEMCPY_RESIDUAL]], label %[[POST_LOOP_MEMCPY_EXPANSION:.*]]
+; CHECK:       [[POST_LOOP_MEMCPY_EXPANSION]]:
+; CHECK-NEXT:    ret void
+; CHECK:       [[LOOP_MEMCPY_RESIDUAL_HEADER]]:
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne i32 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[TMP15]], label %[[LOOP_MEMCPY_RESIDUAL]], label %[[POST_LOOP_MEMCPY_EXPANSION]]
+;
+  call void @llvm.memcpy.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 %length, i1 false)
+  ret void
+}
+
+define void @memcpy_known_p1_to_p7(ptr addrspace(1) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_p1_to_p7(
+; CHECK-SAME: ptr addrspace(1) inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[SRC]], i32 [[LOOP_INDEX]]
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i32>, ptr addrspace(1) [[TMP1]], align 16
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[DST_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTSLICE_0:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_0]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[TMP3]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_4:%.*]] = add nuw i32 [[TMP3]], 16
+; CHECK-NEXT:    [[DOTSLICE_4:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_4]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_4]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_8:%.*]] = add nuw i32 [[TMP3]], 32
+; CHECK-NEXT:    [[DOTSLICE_8:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_8]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_8]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_12:%.*]] = add nuw i32 [[TMP3]], 48
+; CHECK-NEXT:    [[DOTSLICE_12:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_12]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_12]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_16:%.*]] = add nuw i32 [[TMP3]], 64
+; CHECK-NEXT:    [[DOTSLICE_16:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 16, i32 17, i32 18, i32 19>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_16]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_20:%.*]] = add nuw i32 [[TMP3]], 80
+; CHECK-NEXT:    [[DOTSLICE_20:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_20]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_20]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_24:%.*]] = add nuw i32 [[TMP3]], 96
+; CHECK-NEXT:    [[DOTSLICE_24:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 24, i32 25, i32 26, i32 27>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_24]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_24]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_28:%.*]] = add nuw i32 [[TMP3]], 112
+; CHECK-NEXT:    [[DOTSLICE_28:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_28]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_28]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_32:%.*]] = add nuw i32 [[TMP3]], 128
+; CHECK-NEXT:    [[DOTSLICE_32:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 32, i32 33, i32 34, i32 35>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_32]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_36:%.*]] = add nuw i32 [[TMP3]], 144
+; CHECK-NEXT:    [[DOTSLICE_36:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 36, i32 37, i32 38, i32 39>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_36]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_36]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_40:%.*]] = add nuw i32 [[TMP3]], 160
+; CHECK-NEXT:    [[DOTSLICE_40:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 40, i32 41, i32 42, i32 43>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_40]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_40]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_44:%.*]] = add nuw i32 [[TMP3]], 176
+; CHECK-NEXT:    [[DOTSLICE_44:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 44, i32 45, i32 46, i32 47>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_44]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_44]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_48:%.*]] = add nuw i32 [[TMP3]], 192
+; CHECK-NEXT:    [[DOTSLICE_48:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 48, i32 49, i32 50, i32 51>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_48]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_52:%.*]] = add nuw i32 [[TMP3]], 208
+; CHECK-NEXT:    [[DOTSLICE_52:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 52, i32 53, i32 54, i32 55>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_52]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_52]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_56:%.*]] = add nuw i32 [[TMP3]], 224
+; CHECK-NEXT:    [[DOTSLICE_56:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 56, i32 57, i32 58, i32 59>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_56]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_56]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_60:%.*]] = add nuw i32 [[TMP3]], 240
+; CHECK-NEXT:    [[DOTSLICE_60:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_60]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_60]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p7.p1.i32(ptr addrspace(7) noundef nonnull align 16 %dst, ptr addrspace(1) noundef nonnull align 16 %src, i32 8192, i1 false)
+  ret void
+}
+
+define void @memcpy_known_p7_to_p1(ptr addrspace(7) inreg %src, ptr addrspace(1) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_p7_to_p1(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], ptr addrspace(1) inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTOFF_0:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[TMP1]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_0:%.*]] = shufflevector <4 x i32> [[DOTOFF_0]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_0:%.*]] = shufflevector <64 x i32> poison, <64 x i32> [[DOTEXT_0]], <64 x i32> <i32 64, i32 65, i32 66, i32 67, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_16:%.*]] = add nuw i32 [[TMP1]], 16
+; CHECK-NEXT:    [[DOTOFF_16:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_4:%.*]] = shufflevector <4 x i32> [[DOTOFF_16]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_4:%.*]] = shufflevector <64 x i32> [[DOTPARTS_0]], <64 x i32> [[DOTEXT_4]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 64, i32 65, i32 66, i32 67, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_32:%.*]] = add nuw i32 [[TMP1]], 32
+; CHECK-NEXT:    [[DOTOFF_32:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_8:%.*]] = shufflevector <4 x i32> [[DOTOFF_32]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_8:%.*]] = shufflevector <64 x i32> [[DOTPARTS_4]], <64 x i32> [[DOTEXT_8]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 64, i32 65, i32 66, i32 67, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_48:%.*]] = add nuw i32 [[TMP1]], 48
+; CHECK-NEXT:    [[DOTOFF_48:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_12:%.*]] = shufflevector <4 x i32> [[DOTOFF_48]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_12:%.*]] = shufflevector <64 x i32> [[DOTPARTS_8]], <64 x i32> [[DOTEXT_12]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 64, i32 65, i32 66, i32 67, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_64:%.*]] = add nuw i32 [[TMP1]], 64
+; CHECK-NEXT:    [[DOTOFF_64:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_64]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_16:%.*]] = shufflevector <4 x i32> [[DOTOFF_64]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_16:%.*]] = shufflevector <64 x i32> [[DOTPARTS_12]], <64 x i32> [[DOTEXT_16]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_80:%.*]] = add nuw i32 [[TMP1]], 80
+; CHECK-NEXT:    [[DOTOFF_80:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_80]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_20:%.*]] = shufflevector <4 x i32> [[DOTOFF_80]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_20:%.*]] = shufflevector <64 x i32> [[DOTPARTS_16]], <64 x i32> [[DOTEXT_20]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 64, i32 65, i32 66, i32 67, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_96:%.*]] = add nuw i32 [[TMP1]], 96
+; CHECK-NEXT:    [[DOTOFF_96:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_96]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_24:%.*]] = shufflevector <4 x i32> [[DOTOFF_96]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_24:%.*]] = shufflevector <64 x i32> [[DOTPARTS_20]], <64 x i32> [[DOTEXT_24]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 64, i32 65, i32 66, i32 67, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_112:%.*]] = add nuw i32 [[TMP1]], 112
+; CHECK-NEXT:    [[DOTOFF_112:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_112]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_28:%.*]] = shufflevector <4 x i32> [[DOTOFF_112]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_28:%.*]] = shufflevector <64 x i32> [[DOTPARTS_24]], <64 x i32> [[DOTEXT_28]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 64, i32 65, i32 66, i32 67, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_128:%.*]] = add nuw i32 [[TMP1]], 128
+; CHECK-NEXT:    [[DOTOFF_128:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_128]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_32:%.*]] = shufflevector <4 x i32> [[DOTOFF_128]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_32:%.*]] = shufflevector <64 x i32> [[DOTPARTS_28]], <64 x i32> [[DOTEXT_32]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 64, i32 65, i32 66, i32 67, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_144:%.*]] = add nuw i32 [[TMP1]], 144
+; CHECK-NEXT:    [[DOTOFF_144:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_144]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_36:%.*]] = shufflevector <4 x i32> [[DOTOFF_144]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_36:%.*]] = shufflevector <64 x i32> [[DOTPARTS_32]], <64 x i32> [[DOTEXT_36]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 64, i32 65, i32 66, i32 67, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_160:%.*]] = add nuw i32 [[TMP1]], 160
+; CHECK-NEXT:    [[DOTOFF_160:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_160]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_40:%.*]] = shufflevector <4 x i32> [[DOTOFF_160]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_40:%.*]] = shufflevector <64 x i32> [[DOTPARTS_36]], <64 x i32> [[DOTEXT_40]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 64, i32 65, i32 66, i32 67, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_176:%.*]] = add nuw i32 [[TMP1]], 176
+; CHECK-NEXT:    [[DOTOFF_176:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_176]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_44:%.*]] = shufflevector <4 x i32> [[DOTOFF_176]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_44:%.*]] = shufflevector <64 x i32> [[DOTPARTS_40]], <64 x i32> [[DOTEXT_44]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 64, i32 65, i32 66, i32 67, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_192:%.*]] = add nuw i32 [[TMP1]], 192
+; CHECK-NEXT:    [[DOTOFF_192:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_192]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_48:%.*]] = shufflevector <4 x i32> [[DOTOFF_192]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_48:%.*]] = shufflevector <64 x i32> [[DOTPARTS_44]], <64 x i32> [[DOTEXT_48]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 64, i32 65, i32 66, i32 67, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_208:%.*]] = add nuw i32 [[TMP1]], 208
+; CHECK-NEXT:    [[DOTOFF_208:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_208]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_52:%.*]] = shufflevector <4 x i32> [[DOTOFF_208]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_52:%.*]] = shufflevector <64 x i32> [[DOTPARTS_48]], <64 x i32> [[DOTEXT_52]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 64, i32 65, i32 66, i32 67, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_224:%.*]] = add nuw i32 [[TMP1]], 224
+; CHECK-NEXT:    [[DOTOFF_224:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_224]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_56:%.*]] = shufflevector <4 x i32> [[DOTOFF_224]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_56:%.*]] = shufflevector <64 x i32> [[DOTPARTS_52]], <64 x i32> [[DOTEXT_56]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 64, i32 65, i32 66, i32 67, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_240:%.*]] = add nuw i32 [[TMP1]], 240
+; CHECK-NEXT:    [[DOTOFF_240:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_240]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_60:%.*]] = shufflevector <4 x i32> [[DOTOFF_240]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <64 x i32> [[DOTPARTS_56]], <64 x i32> [[DOTEXT_60]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 64, i32 65, i32 66, i32 67>
+; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[DST]], i32 [[LOOP_INDEX]]
+; CHECK-NEXT:    store <64 x i32> [[TMP2]], ptr addrspace(1) [[TMP3]], align 16
+; CHECK-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p1.p7.i32(ptr addrspace(1) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 8192, i1 false)
+  ret void
+}
+
+;; This could be the direct-to-LDS intrinsics in a future patch
+define void @memcpy_known_p7_to_p3(ptr addrspace(7) inreg %src, ptr addrspace(3) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_p7_to_p3(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], ptr addrspace(3) inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[DST]], i32 0
+; CHECK-NEXT:    store <4 x i32> [[TMP1]], ptr addrspace(3) [[TMP2]], align 16
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p3.p7.i32(ptr addrspace(3) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 16, i1 false)
+  ret void
+}
+
+define void @memcpy_known_p7_to_p3_byte(ptr addrspace(7) inreg %src, ptr addrspace(3) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_p7_to_p3_byte(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], ptr addrspace(3) inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[DST]], i32 0
+; CHECK-NEXT:    store i32 [[TMP1]], ptr addrspace(3) [[TMP2]], align 16
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p3.p7.i32(ptr addrspace(3) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 4, i1 false)
+  ret void
+}
+
+define void @memcpy_known_p7_to_p3_long(ptr addrspace(7) inreg %src, ptr addrspace(3) inreg %dst) {
+; CHECK-LABEL: define void @memcpy_known_p7_to_p3_long(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], ptr addrspace(3) inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTOFF_0:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[TMP1]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_0:%.*]] = shufflevector <4 x i32> [[DOTOFF_0]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_0:%.*]] = shufflevector <64 x i32> poison, <64 x i32> [[DOTEXT_0]], <64 x i32> <i32 64, i32 65, i32 66, i32 67, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_16:%.*]] = add nuw i32 [[TMP1]], 16
+; CHECK-NEXT:    [[DOTOFF_16:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_4:%.*]] = shufflevector <4 x i32> [[DOTOFF_16]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_4:%.*]] = shufflevector <64 x i32> [[DOTPARTS_0]], <64 x i32> [[DOTEXT_4]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 64, i32 65, i32 66, i32 67, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_32:%.*]] = add nuw i32 [[TMP1]], 32
+; CHECK-NEXT:    [[DOTOFF_32:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_8:%.*]] = shufflevector <4 x i32> [[DOTOFF_32]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_8:%.*]] = shufflevector <64 x i32> [[DOTPARTS_4]], <64 x i32> [[DOTEXT_8]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 64, i32 65, i32 66, i32 67, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_48:%.*]] = add nuw i32 [[TMP1]], 48
+; CHECK-NEXT:    [[DOTOFF_48:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_12:%.*]] = shufflevector <4 x i32> [[DOTOFF_48]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_12:%.*]] = shufflevector <64 x i32> [[DOTPARTS_8]], <64 x i32> [[DOTEXT_12]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 64, i32 65, i32 66, i32 67, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_64:%.*]] = add nuw i32 [[TMP1]], 64
+; CHECK-NEXT:    [[DOTOFF_64:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_64]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_16:%.*]] = shufflevector <4 x i32> [[DOTOFF_64]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_16:%.*]] = shufflevector <64 x i32> [[DOTPARTS_12]], <64 x i32> [[DOTEXT_16]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_80:%.*]] = add nuw i32 [[TMP1]], 80
+; CHECK-NEXT:    [[DOTOFF_80:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_80]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_20:%.*]] = shufflevector <4 x i32> [[DOTOFF_80]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_20:%.*]] = shufflevector <64 x i32> [[DOTPARTS_16]], <64 x i32> [[DOTEXT_20]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 64, i32 65, i32 66, i32 67, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_96:%.*]] = add nuw i32 [[TMP1]], 96
+; CHECK-NEXT:    [[DOTOFF_96:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_96]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_24:%.*]] = shufflevector <4 x i32> [[DOTOFF_96]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_24:%.*]] = shufflevector <64 x i32> [[DOTPARTS_20]], <64 x i32> [[DOTEXT_24]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 64, i32 65, i32 66, i32 67, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_112:%.*]] = add nuw i32 [[TMP1]], 112
+; CHECK-NEXT:    [[DOTOFF_112:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_112]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_28:%.*]] = shufflevector <4 x i32> [[DOTOFF_112]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_28:%.*]] = shufflevector <64 x i32> [[DOTPARTS_24]], <64 x i32> [[DOTEXT_28]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 64, i32 65, i32 66, i32 67, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_128:%.*]] = add nuw i32 [[TMP1]], 128
+; CHECK-NEXT:    [[DOTOFF_128:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_128]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_32:%.*]] = shufflevector <4 x i32> [[DOTOFF_128]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_32:%.*]] = shufflevector <64 x i32> [[DOTPARTS_28]], <64 x i32> [[DOTEXT_32]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 64, i32 65, i32 66, i32 67, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_144:%.*]] = add nuw i32 [[TMP1]], 144
+; CHECK-NEXT:    [[DOTOFF_144:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_144]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_36:%.*]] = shufflevector <4 x i32> [[DOTOFF_144]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_36:%.*]] = shufflevector <64 x i32> [[DOTPARTS_32]], <64 x i32> [[DOTEXT_36]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 64, i32 65, i32 66, i32 67, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_160:%.*]] = add nuw i32 [[TMP1]], 160
+; CHECK-NEXT:    [[DOTOFF_160:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_160]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_40:%.*]] = shufflevector <4 x i32> [[DOTOFF_160]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_40:%.*]] = shufflevector <64 x i32> [[DOTPARTS_36]], <64 x i32> [[DOTEXT_40]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 64, i32 65, i32 66, i32 67, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_176:%.*]] = add nuw i32 [[TMP1]], 176
+; CHECK-NEXT:    [[DOTOFF_176:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_176]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_44:%.*]] = shufflevector <4 x i32> [[DOTOFF_176]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_44:%.*]] = shufflevector <64 x i32> [[DOTPARTS_40]], <64 x i32> [[DOTEXT_44]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 64, i32 65, i32 66, i32 67, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_192:%.*]] = add nuw i32 [[TMP1]], 192
+; CHECK-NEXT:    [[DOTOFF_192:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_192]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_48:%.*]] = shufflevector <4 x i32> [[DOTOFF_192]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_48:%.*]] = shufflevector <64 x i32> [[DOTPARTS_44]], <64 x i32> [[DOTEXT_48]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 64, i32 65, i32 66, i32 67, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_208:%.*]] = add nuw i32 [[TMP1]], 208
+; CHECK-NEXT:    [[DOTOFF_208:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_208]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_52:%.*]] = shufflevector <4 x i32> [[DOTOFF_208]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_52:%.*]] = shufflevector <64 x i32> [[DOTPARTS_48]], <64 x i32> [[DOTEXT_52]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 64, i32 65, i32 66, i32 67, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_224:%.*]] = add nuw i32 [[TMP1]], 224
+; CHECK-NEXT:    [[DOTOFF_224:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_224]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_56:%.*]] = shufflevector <4 x i32> [[DOTOFF_224]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_56:%.*]] = shufflevector <64 x i32> [[DOTPARTS_52]], <64 x i32> [[DOTEXT_56]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 64, i32 65, i32 66, i32 67, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_240:%.*]] = add nuw i32 [[TMP1]], 240
+; CHECK-NEXT:    [[DOTOFF_240:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_240]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_60:%.*]] = shufflevector <4 x i32> [[DOTOFF_240]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <64 x i32> [[DOTPARTS_56]], <64 x i32> [[DOTEXT_60]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 64, i32 65, i32 66, i32 67>
+; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[DST]], i32 [[LOOP_INDEX]]
+; CHECK-NEXT:    store <64 x i32> [[TMP2]], ptr addrspace(3) [[TMP3]], align 16
+; CHECK-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.p3.p7.i32(ptr addrspace(3) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 8192, i1 false)
+  ret void
+}
+
+;; memcpy.inline
+
+declare void @llvm.memcpy.inline.p7.p7.i32(ptr addrspace(7), ptr addrspace(7), i32, i1)
+declare void @llvm.memcpy.inline.p1.p7.i32(ptr addrspace(1), ptr addrspace(7), i32, i1)
+declare void @llvm.memcpy.inline.p7.p1.i32(ptr addrspace(7), ptr addrspace(1), i32, i1)
+declare void @llvm.memcpy.inline.p7.p7.i64(ptr addrspace(7), ptr addrspace(7), i64, i1)
+declare void @llvm.memcpy.inline.p3.p7.i32(ptr addrspace(3), ptr addrspace(7), i32, i1)
+
+define void @memcpy.inline_known(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTOFF_0:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[TMP1]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_0:%.*]] = shufflevector <4 x i32> [[DOTOFF_0]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_0:%.*]] = shufflevector <64 x i32> poison, <64 x i32> [[DOTEXT_0]], <64 x i32> <i32 64, i32 65, i32 66, i32 67, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_16:%.*]] = add nuw i32 [[TMP1]], 16
+; CHECK-NEXT:    [[DOTOFF_16:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_4:%.*]] = shufflevector <4 x i32> [[DOTOFF_16]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_4:%.*]] = shufflevector <64 x i32> [[DOTPARTS_0]], <64 x i32> [[DOTEXT_4]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 64, i32 65, i32 66, i32 67, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_32:%.*]] = add nuw i32 [[TMP1]], 32
+; CHECK-NEXT:    [[DOTOFF_32:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_8:%.*]] = shufflevector <4 x i32> [[DOTOFF_32]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_8:%.*]] = shufflevector <64 x i32> [[DOTPARTS_4]], <64 x i32> [[DOTEXT_8]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 64, i32 65, i32 66, i32 67, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_48:%.*]] = add nuw i32 [[TMP1]], 48
+; CHECK-NEXT:    [[DOTOFF_48:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_12:%.*]] = shufflevector <4 x i32> [[DOTOFF_48]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_12:%.*]] = shufflevector <64 x i32> [[DOTPARTS_8]], <64 x i32> [[DOTEXT_12]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 64, i32 65, i32 66, i32 67, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_64:%.*]] = add nuw i32 [[TMP1]], 64
+; CHECK-NEXT:    [[DOTOFF_64:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_64]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_16:%.*]] = shufflevector <4 x i32> [[DOTOFF_64]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_16:%.*]] = shufflevector <64 x i32> [[DOTPARTS_12]], <64 x i32> [[DOTEXT_16]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_80:%.*]] = add nuw i32 [[TMP1]], 80
+; CHECK-NEXT:    [[DOTOFF_80:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_80]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_20:%.*]] = shufflevector <4 x i32> [[DOTOFF_80]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_20:%.*]] = shufflevector <64 x i32> [[DOTPARTS_16]], <64 x i32> [[DOTEXT_20]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 64, i32 65, i32 66, i32 67, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_96:%.*]] = add nuw i32 [[TMP1]], 96
+; CHECK-NEXT:    [[DOTOFF_96:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_96]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_24:%.*]] = shufflevector <4 x i32> [[DOTOFF_96]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_24:%.*]] = shufflevector <64 x i32> [[DOTPARTS_20]], <64 x i32> [[DOTEXT_24]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 64, i32 65, i32 66, i32 67, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_112:%.*]] = add nuw i32 [[TMP1]], 112
+; CHECK-NEXT:    [[DOTOFF_112:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_112]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_28:%.*]] = shufflevector <4 x i32> [[DOTOFF_112]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_28:%.*]] = shufflevector <64 x i32> [[DOTPARTS_24]], <64 x i32> [[DOTEXT_28]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 64, i32 65, i32 66, i32 67, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_128:%.*]] = add nuw i32 [[TMP1]], 128
+; CHECK-NEXT:    [[DOTOFF_128:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_128]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_32:%.*]] = shufflevector <4 x i32> [[DOTOFF_128]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_32:%.*]] = shufflevector <64 x i32> [[DOTPARTS_28]], <64 x i32> [[DOTEXT_32]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 64, i32 65, i32 66, i32 67, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_144:%.*]] = add nuw i32 [[TMP1]], 144
+; CHECK-NEXT:    [[DOTOFF_144:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_144]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_36:%.*]] = shufflevector <4 x i32> [[DOTOFF_144]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_36:%.*]] = shufflevector <64 x i32> [[DOTPARTS_32]], <64 x i32> [[DOTEXT_36]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 64, i32 65, i32 66, i32 67, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_160:%.*]] = add nuw i32 [[TMP1]], 160
+; CHECK-NEXT:    [[DOTOFF_160:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_160]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_40:%.*]] = shufflevector <4 x i32> [[DOTOFF_160]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_40:%.*]] = shufflevector <64 x i32> [[DOTPARTS_36]], <64 x i32> [[DOTEXT_40]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 64, i32 65, i32 66, i32 67, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_176:%.*]] = add nuw i32 [[TMP1]], 176
+; CHECK-NEXT:    [[DOTOFF_176:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_176]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_44:%.*]] = shufflevector <4 x i32> [[DOTOFF_176]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_44:%.*]] = shufflevector <64 x i32> [[DOTPARTS_40]], <64 x i32> [[DOTEXT_44]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 64, i32 65, i32 66, i32 67, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_192:%.*]] = add nuw i32 [[TMP1]], 192
+; CHECK-NEXT:    [[DOTOFF_192:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_192]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_48:%.*]] = shufflevector <4 x i32> [[DOTOFF_192]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_48:%.*]] = shufflevector <64 x i32> [[DOTPARTS_44]], <64 x i32> [[DOTEXT_48]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 64, i32 65, i32 66, i32 67, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_208:%.*]] = add nuw i32 [[TMP1]], 208
+; CHECK-NEXT:    [[DOTOFF_208:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_208]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_52:%.*]] = shufflevector <4 x i32> [[DOTOFF_208]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_52:%.*]] = shufflevector <64 x i32> [[DOTPARTS_48]], <64 x i32> [[DOTEXT_52]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 64, i32 65, i32 66, i32 67, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_224:%.*]] = add nuw i32 [[TMP1]], 224
+; CHECK-NEXT:    [[DOTOFF_224:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_224]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_56:%.*]] = shufflevector <4 x i32> [[DOTOFF_224]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_56:%.*]] = shufflevector <64 x i32> [[DOTPARTS_52]], <64 x i32> [[DOTEXT_56]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 64, i32 65, i32 66, i32 67, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_240:%.*]] = add nuw i32 [[TMP1]], 240
+; CHECK-NEXT:    [[DOTOFF_240:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_240]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_60:%.*]] = shufflevector <4 x i32> [[DOTOFF_240]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <64 x i32> [[DOTPARTS_56]], <64 x i32> [[DOTEXT_60]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 64, i32 65, i32 66, i32 67>
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[DST_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTSLICE_0:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_0]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[TMP3]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_4:%.*]] = add nuw i32 [[TMP3]], 16
+; CHECK-NEXT:    [[DOTSLICE_4:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_4]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_4]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_8:%.*]] = add nuw i32 [[TMP3]], 32
+; CHECK-NEXT:    [[DOTSLICE_8:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_8]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_8]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_12:%.*]] = add nuw i32 [[TMP3]], 48
+; CHECK-NEXT:    [[DOTSLICE_12:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_12]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_12]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_16:%.*]] = add nuw i32 [[TMP3]], 64
+; CHECK-NEXT:    [[DOTSLICE_16:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 16, i32 17, i32 18, i32 19>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_16]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_20:%.*]] = add nuw i32 [[TMP3]], 80
+; CHECK-NEXT:    [[DOTSLICE_20:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_20]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_20]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_24:%.*]] = add nuw i32 [[TMP3]], 96
+; CHECK-NEXT:    [[DOTSLICE_24:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 24, i32 25, i32 26, i32 27>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_24]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_24]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_28:%.*]] = add nuw i32 [[TMP3]], 112
+; CHECK-NEXT:    [[DOTSLICE_28:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_28]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_28]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_32:%.*]] = add nuw i32 [[TMP3]], 128
+; CHECK-NEXT:    [[DOTSLICE_32:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 32, i32 33, i32 34, i32 35>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_32]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_36:%.*]] = add nuw i32 [[TMP3]], 144
+; CHECK-NEXT:    [[DOTSLICE_36:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 36, i32 37, i32 38, i32 39>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_36]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_36]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_40:%.*]] = add nuw i32 [[TMP3]], 160
+; CHECK-NEXT:    [[DOTSLICE_40:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 40, i32 41, i32 42, i32 43>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_40]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_40]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_44:%.*]] = add nuw i32 [[TMP3]], 176
+; CHECK-NEXT:    [[DOTSLICE_44:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 44, i32 45, i32 46, i32 47>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_44]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_44]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_48:%.*]] = add nuw i32 [[TMP3]], 192
+; CHECK-NEXT:    [[DOTSLICE_48:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 48, i32 49, i32 50, i32 51>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_48]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_52:%.*]] = add nuw i32 [[TMP3]], 208
+; CHECK-NEXT:    [[DOTSLICE_52:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 52, i32 53, i32 54, i32 55>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_52]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_52]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_56:%.*]] = add nuw i32 [[TMP3]], 224
+; CHECK-NEXT:    [[DOTSLICE_56:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 56, i32 57, i32 58, i32 59>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_56]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_56]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_60:%.*]] = add nuw i32 [[TMP3]], 240
+; CHECK-NEXT:    [[DOTSLICE_60:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_60]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_60]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p7.p7.i32(ptr addrspace(7) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 8192, i1 false)
+  ret void
+}
+
+define void @memcpy.inline_known_small(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_small(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP1]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DST_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = add nuw i32 [[SRC_OFF]], 16
+; CHECK-NEXT:    [[TMP3:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4:%.*]] = add nuw i32 [[DST_OFF]], 16
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP3]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP4]], i32 0, i32 0)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 32, i1 false)
+  ret void
+}
+
+define void @memcpy.inline_known_byte(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_byte(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call i8 @llvm.amdgcn.raw.ptr.buffer.load.i8(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 [[TMP1]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DST_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 1, i1 false)
+  ret void
+}
+
+define void @memcpy.inline_known_tail(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_tail(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call i64 @llvm.amdgcn.raw.ptr.buffer.load.i64(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i64(i64 [[TMP1]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DST_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = add nuw i32 [[SRC_OFF]], 8
+; CHECK-NEXT:    [[TMP3:%.*]] = call i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4:%.*]] = add nuw i32 [[DST_OFF]], 8
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 [[TMP3]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP4]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP5:%.*]] = add nuw i32 [[SRC_OFF]], 12
+; CHECK-NEXT:    [[TMP6:%.*]] = call i16 @llvm.amdgcn.raw.ptr.buffer.load.i16(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP5]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP7:%.*]] = add nuw i32 [[DST_OFF]], 12
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i16(i16 [[TMP6]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP7]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP8:%.*]] = add nuw i32 [[SRC_OFF]], 14
+; CHECK-NEXT:    [[TMP9:%.*]] = call i8 @llvm.amdgcn.raw.ptr.buffer.load.i8(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP8]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP10:%.*]] = add nuw i32 [[DST_OFF]], 14
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 [[TMP9]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP10]], i32 0, i32 0)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 15, i1 false)
+  ret void
+}
+
+define void @memcpy.inline_known_i64(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_i64(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i64 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[LOOP_INDEX_C:%.*]] = trunc i64 [[LOOP_INDEX]] to i32
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX_C]]
+; CHECK-NEXT:    [[DOTOFF_0:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP1]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_0:%.*]] = shufflevector <4 x i32> [[DOTOFF_0]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_0:%.*]] = shufflevector <64 x i32> poison, <64 x i32> [[DOTEXT_0]], <64 x i32> <i32 64, i32 65, i32 66, i32 67, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_16:%.*]] = add nuw i32 [[TMP1]], 16
+; CHECK-NEXT:    [[DOTOFF_16:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_4:%.*]] = shufflevector <4 x i32> [[DOTOFF_16]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_4:%.*]] = shufflevector <64 x i32> [[DOTPARTS_0]], <64 x i32> [[DOTEXT_4]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 64, i32 65, i32 66, i32 67, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_32:%.*]] = add nuw i32 [[TMP1]], 32
+; CHECK-NEXT:    [[DOTOFF_32:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_8:%.*]] = shufflevector <4 x i32> [[DOTOFF_32]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_8:%.*]] = shufflevector <64 x i32> [[DOTPARTS_4]], <64 x i32> [[DOTEXT_8]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 64, i32 65, i32 66, i32 67, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_48:%.*]] = add nuw i32 [[TMP1]], 48
+; CHECK-NEXT:    [[DOTOFF_48:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_12:%.*]] = shufflevector <4 x i32> [[DOTOFF_48]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_12:%.*]] = shufflevector <64 x i32> [[DOTPARTS_8]], <64 x i32> [[DOTEXT_12]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 64, i32 65, i32 66, i32 67, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_64:%.*]] = add nuw i32 [[TMP1]], 64
+; CHECK-NEXT:    [[DOTOFF_64:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_64]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_16:%.*]] = shufflevector <4 x i32> [[DOTOFF_64]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_16:%.*]] = shufflevector <64 x i32> [[DOTPARTS_12]], <64 x i32> [[DOTEXT_16]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_80:%.*]] = add nuw i32 [[TMP1]], 80
+; CHECK-NEXT:    [[DOTOFF_80:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_80]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_20:%.*]] = shufflevector <4 x i32> [[DOTOFF_80]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_20:%.*]] = shufflevector <64 x i32> [[DOTPARTS_16]], <64 x i32> [[DOTEXT_20]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 64, i32 65, i32 66, i32 67, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_96:%.*]] = add nuw i32 [[TMP1]], 96
+; CHECK-NEXT:    [[DOTOFF_96:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_96]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_24:%.*]] = shufflevector <4 x i32> [[DOTOFF_96]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_24:%.*]] = shufflevector <64 x i32> [[DOTPARTS_20]], <64 x i32> [[DOTEXT_24]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 64, i32 65, i32 66, i32 67, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_112:%.*]] = add nuw i32 [[TMP1]], 112
+; CHECK-NEXT:    [[DOTOFF_112:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_112]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_28:%.*]] = shufflevector <4 x i32> [[DOTOFF_112]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_28:%.*]] = shufflevector <64 x i32> [[DOTPARTS_24]], <64 x i32> [[DOTEXT_28]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 64, i32 65, i32 66, i32 67, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_128:%.*]] = add nuw i32 [[TMP1]], 128
+; CHECK-NEXT:    [[DOTOFF_128:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_128]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_32:%.*]] = shufflevector <4 x i32> [[DOTOFF_128]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_32:%.*]] = shufflevector <64 x i32> [[DOTPARTS_28]], <64 x i32> [[DOTEXT_32]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 64, i32 65, i32 66, i32 67, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_144:%.*]] = add nuw i32 [[TMP1]], 144
+; CHECK-NEXT:    [[DOTOFF_144:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_144]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_36:%.*]] = shufflevector <4 x i32> [[DOTOFF_144]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_36:%.*]] = shufflevector <64 x i32> [[DOTPARTS_32]], <64 x i32> [[DOTEXT_36]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 64, i32 65, i32 66, i32 67, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_160:%.*]] = add nuw i32 [[TMP1]], 160
+; CHECK-NEXT:    [[DOTOFF_160:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_160]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_40:%.*]] = shufflevector <4 x i32> [[DOTOFF_160]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_40:%.*]] = shufflevector <64 x i32> [[DOTPARTS_36]], <64 x i32> [[DOTEXT_40]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 64, i32 65, i32 66, i32 67, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_176:%.*]] = add nuw i32 [[TMP1]], 176
+; CHECK-NEXT:    [[DOTOFF_176:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_176]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_44:%.*]] = shufflevector <4 x i32> [[DOTOFF_176]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_44:%.*]] = shufflevector <64 x i32> [[DOTPARTS_40]], <64 x i32> [[DOTEXT_44]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 64, i32 65, i32 66, i32 67, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_192:%.*]] = add nuw i32 [[TMP1]], 192
+; CHECK-NEXT:    [[DOTOFF_192:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_192]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_48:%.*]] = shufflevector <4 x i32> [[DOTOFF_192]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_48:%.*]] = shufflevector <64 x i32> [[DOTPARTS_44]], <64 x i32> [[DOTEXT_48]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 64, i32 65, i32 66, i32 67, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_208:%.*]] = add nuw i32 [[TMP1]], 208
+; CHECK-NEXT:    [[DOTOFF_208:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_208]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_52:%.*]] = shufflevector <4 x i32> [[DOTOFF_208]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_52:%.*]] = shufflevector <64 x i32> [[DOTPARTS_48]], <64 x i32> [[DOTEXT_52]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 64, i32 65, i32 66, i32 67, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_224:%.*]] = add nuw i32 [[TMP1]], 224
+; CHECK-NEXT:    [[DOTOFF_224:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_224]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_56:%.*]] = shufflevector <4 x i32> [[DOTOFF_224]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_56:%.*]] = shufflevector <64 x i32> [[DOTPARTS_52]], <64 x i32> [[DOTEXT_56]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 64, i32 65, i32 66, i32 67, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_240:%.*]] = add nuw i32 [[TMP1]], 240
+; CHECK-NEXT:    [[DOTOFF_240:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[DOTOFF_PTR_240]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_60:%.*]] = shufflevector <4 x i32> [[DOTOFF_240]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <64 x i32> [[DOTPARTS_56]], <64 x i32> [[DOTEXT_60]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 64, i32 65, i32 66, i32 67>
+; CHECK-NEXT:    [[LOOP_INDEX_C1:%.*]] = trunc i64 [[LOOP_INDEX]] to i32
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[DST_OFF]], [[LOOP_INDEX_C1]]
+; CHECK-NEXT:    [[DOTSLICE_0:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_0]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP3]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_4:%.*]] = add nuw i32 [[TMP3]], 16
+; CHECK-NEXT:    [[DOTSLICE_4:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_4]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_4]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_8:%.*]] = add nuw i32 [[TMP3]], 32
+; CHECK-NEXT:    [[DOTSLICE_8:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_8]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_8]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_12:%.*]] = add nuw i32 [[TMP3]], 48
+; CHECK-NEXT:    [[DOTSLICE_12:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_12]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_12]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_16:%.*]] = add nuw i32 [[TMP3]], 64
+; CHECK-NEXT:    [[DOTSLICE_16:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 16, i32 17, i32 18, i32 19>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_16]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_20:%.*]] = add nuw i32 [[TMP3]], 80
+; CHECK-NEXT:    [[DOTSLICE_20:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_20]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_20]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_24:%.*]] = add nuw i32 [[TMP3]], 96
+; CHECK-NEXT:    [[DOTSLICE_24:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 24, i32 25, i32 26, i32 27>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_24]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_24]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_28:%.*]] = add nuw i32 [[TMP3]], 112
+; CHECK-NEXT:    [[DOTSLICE_28:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_28]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_28]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_32:%.*]] = add nuw i32 [[TMP3]], 128
+; CHECK-NEXT:    [[DOTSLICE_32:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 32, i32 33, i32 34, i32 35>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_32]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_36:%.*]] = add nuw i32 [[TMP3]], 144
+; CHECK-NEXT:    [[DOTSLICE_36:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 36, i32 37, i32 38, i32 39>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_36]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_36]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_40:%.*]] = add nuw i32 [[TMP3]], 160
+; CHECK-NEXT:    [[DOTSLICE_40:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 40, i32 41, i32 42, i32 43>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_40]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_40]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_44:%.*]] = add nuw i32 [[TMP3]], 176
+; CHECK-NEXT:    [[DOTSLICE_44:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 44, i32 45, i32 46, i32 47>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_44]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_44]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_48:%.*]] = add nuw i32 [[TMP3]], 192
+; CHECK-NEXT:    [[DOTSLICE_48:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 48, i32 49, i32 50, i32 51>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_48]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_52:%.*]] = add nuw i32 [[TMP3]], 208
+; CHECK-NEXT:    [[DOTSLICE_52:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 52, i32 53, i32 54, i32 55>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_52]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_52]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_56:%.*]] = add nuw i32 [[TMP3]], 224
+; CHECK-NEXT:    [[DOTSLICE_56:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 56, i32 57, i32 58, i32 59>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_56]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_56]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_60:%.*]] = add nuw i32 [[TMP3]], 240
+; CHECK-NEXT:    [[DOTSLICE_60:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_60]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DOTPART_60]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4]] = add i64 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i64 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p7.p7.i64(ptr addrspace(7) %dst, ptr addrspace(7) %src, i64 8192, i1 false)
+  ret void
+}
+
+define void @memcpy.inline_known_i32_volatile(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_i32_volatile(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP1]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[DST_OFF]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    [[TMP2:%.*]] = add nuw i32 [[SRC_OFF]], 16
+; CHECK-NEXT:    [[TMP3:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP2]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    [[TMP4:%.*]] = add nuw i32 [[DST_OFF]], 16
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP3]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP4]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 32, i1 true)
+  ret void
+}
+
+define void @memcpy.inline_unknown(ptr addrspace(7) inreg %src, ptr addrspace(7) inreg %dst, i32 inreg %length) {
+; CHECK-LABEL: define void @memcpy.inline_unknown(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]], i32 inreg [[LENGTH:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = and i32 [[LENGTH]], 15
+; CHECK-NEXT:    [[TMP2:%.*]] = sub i32 [[LENGTH]], [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = icmp ne i32 [[TMP2]], 0
+; CHECK-NEXT:    br i1 [[TMP3]], label %[[LOOP_MEMCPY_EXPANSION:.*]], label %[[LOOP_MEMCPY_RESIDUAL_HEADER:.*]]
+; CHECK:       [[LOOP_MEMCPY_EXPANSION]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP7:%.*]], %[[LOOP_MEMCPY_EXPANSION]] ]
+; CHECK-NEXT:    [[TMP4:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[TMP5:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP4]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP6:%.*]] = add i32 [[DST_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[TMP5]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP6]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP7]] = add i32 [[LOOP_INDEX]], 16
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp ult i32 [[TMP7]], [[TMP2]]
+; CHECK-NEXT:    br i1 [[TMP8]], label %[[LOOP_MEMCPY_EXPANSION]], label %[[LOOP_MEMCPY_RESIDUAL_HEADER]]
+; CHECK:       [[LOOP_MEMCPY_RESIDUAL:.*]]:
+; CHECK-NEXT:    [[RESIDUAL_LOOP_INDEX:%.*]] = phi i32 [ 0, %[[LOOP_MEMCPY_RESIDUAL_HEADER]] ], [ [[TMP13:%.*]], %[[LOOP_MEMCPY_RESIDUAL]] ]
+; CHECK-NEXT:    [[TMP9:%.*]] = add i32 [[TMP2]], [[RESIDUAL_LOOP_INDEX]]
+; CHECK-NEXT:    [[TMP10:%.*]] = add i32 [[SRC_OFF]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = call i8 @llvm.amdgcn.raw.ptr.buffer.load.i8(ptr addrspace(8) align 1 [[SRC_RSRC]], i32 [[TMP10]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP12:%.*]] = add i32 [[DST_OFF]], [[TMP9]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 [[TMP11]], ptr addrspace(8) align 1 [[DST_RSRC]], i32 [[TMP12]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP13]] = add i32 [[RESIDUAL_LOOP_INDEX]], 1
+; CHECK-NEXT:    [[TMP14:%.*]] = icmp ult i32 [[TMP13]], [[TMP1]]
+; CHECK-NEXT:    br i1 [[TMP14]], label %[[LOOP_MEMCPY_RESIDUAL]], label %[[POST_LOOP_MEMCPY_EXPANSION:.*]]
+; CHECK:       [[POST_LOOP_MEMCPY_EXPANSION]]:
+; CHECK-NEXT:    ret void
+; CHECK:       [[LOOP_MEMCPY_RESIDUAL_HEADER]]:
+; CHECK-NEXT:    [[TMP15:%.*]] = icmp ne i32 [[TMP1]], 0
+; CHECK-NEXT:    br i1 [[TMP15]], label %[[LOOP_MEMCPY_RESIDUAL]], label %[[POST_LOOP_MEMCPY_EXPANSION]]
+;
+  call void @llvm.memcpy.inline.p7.p7.i32(ptr addrspace(7) %dst, ptr addrspace(7) %src, i32 %length, i1 false)
+  ret void
+}
+
+define void @memcpy.inline_known_p1_to_p7(ptr addrspace(1) inreg %src, ptr addrspace(7) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_p1_to_p7(
+; CHECK-SAME: ptr addrspace(1) inreg [[SRC:%.*]], { ptr addrspace(8), i32 } inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[DST_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 0
+; CHECK-NEXT:    [[DST_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[DST]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[SRC]], i32 [[LOOP_INDEX]]
+; CHECK-NEXT:    [[TMP2:%.*]] = load <64 x i32>, ptr addrspace(1) [[TMP1]], align 16
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[DST_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTSLICE_0:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_0]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[TMP3]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_4:%.*]] = add nuw i32 [[TMP3]], 16
+; CHECK-NEXT:    [[DOTSLICE_4:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_4]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_4]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_8:%.*]] = add nuw i32 [[TMP3]], 32
+; CHECK-NEXT:    [[DOTSLICE_8:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_8]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_8]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_12:%.*]] = add nuw i32 [[TMP3]], 48
+; CHECK-NEXT:    [[DOTSLICE_12:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_12]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_12]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_16:%.*]] = add nuw i32 [[TMP3]], 64
+; CHECK-NEXT:    [[DOTSLICE_16:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 16, i32 17, i32 18, i32 19>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_16]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_20:%.*]] = add nuw i32 [[TMP3]], 80
+; CHECK-NEXT:    [[DOTSLICE_20:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 20, i32 21, i32 22, i32 23>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_20]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_20]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_24:%.*]] = add nuw i32 [[TMP3]], 96
+; CHECK-NEXT:    [[DOTSLICE_24:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 24, i32 25, i32 26, i32 27>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_24]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_24]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_28:%.*]] = add nuw i32 [[TMP3]], 112
+; CHECK-NEXT:    [[DOTSLICE_28:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 28, i32 29, i32 30, i32 31>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_28]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_28]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_32:%.*]] = add nuw i32 [[TMP3]], 128
+; CHECK-NEXT:    [[DOTSLICE_32:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 32, i32 33, i32 34, i32 35>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_32]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_36:%.*]] = add nuw i32 [[TMP3]], 144
+; CHECK-NEXT:    [[DOTSLICE_36:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 36, i32 37, i32 38, i32 39>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_36]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_36]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_40:%.*]] = add nuw i32 [[TMP3]], 160
+; CHECK-NEXT:    [[DOTSLICE_40:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 40, i32 41, i32 42, i32 43>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_40]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_40]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_44:%.*]] = add nuw i32 [[TMP3]], 176
+; CHECK-NEXT:    [[DOTSLICE_44:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 44, i32 45, i32 46, i32 47>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_44]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_44]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_48:%.*]] = add nuw i32 [[TMP3]], 192
+; CHECK-NEXT:    [[DOTSLICE_48:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 48, i32 49, i32 50, i32 51>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_48]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_52:%.*]] = add nuw i32 [[TMP3]], 208
+; CHECK-NEXT:    [[DOTSLICE_52:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 52, i32 53, i32 54, i32 55>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_52]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_52]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_56:%.*]] = add nuw i32 [[TMP3]], 224
+; CHECK-NEXT:    [[DOTSLICE_56:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 56, i32 57, i32 58, i32 59>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_56]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_56]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTPART_60:%.*]] = add nuw i32 [[TMP3]], 240
+; CHECK-NEXT:    [[DOTSLICE_60:%.*]] = shufflevector <64 x i32> [[TMP2]], <64 x i32> poison, <4 x i32> <i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.v4i32(<4 x i32> [[DOTSLICE_60]], ptr addrspace(8) align 16 [[DST_RSRC]], i32 [[DOTPART_60]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p7.p1.i32(ptr addrspace(7) noundef nonnull align 16 %dst, ptr addrspace(1) noundef nonnull align 16 %src, i32 8192, i1 false)
+  ret void
+}
+
+define void @memcpy.inline_known_p7_to_p1(ptr addrspace(7) inreg %src, ptr addrspace(1) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_p7_to_p1(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], ptr addrspace(1) inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTOFF_0:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[TMP1]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_0:%.*]] = shufflevector <4 x i32> [[DOTOFF_0]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_0:%.*]] = shufflevector <64 x i32> poison, <64 x i32> [[DOTEXT_0]], <64 x i32> <i32 64, i32 65, i32 66, i32 67, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_16:%.*]] = add nuw i32 [[TMP1]], 16
+; CHECK-NEXT:    [[DOTOFF_16:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_4:%.*]] = shufflevector <4 x i32> [[DOTOFF_16]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_4:%.*]] = shufflevector <64 x i32> [[DOTPARTS_0]], <64 x i32> [[DOTEXT_4]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 64, i32 65, i32 66, i32 67, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_32:%.*]] = add nuw i32 [[TMP1]], 32
+; CHECK-NEXT:    [[DOTOFF_32:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_8:%.*]] = shufflevector <4 x i32> [[DOTOFF_32]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_8:%.*]] = shufflevector <64 x i32> [[DOTPARTS_4]], <64 x i32> [[DOTEXT_8]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 64, i32 65, i32 66, i32 67, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_48:%.*]] = add nuw i32 [[TMP1]], 48
+; CHECK-NEXT:    [[DOTOFF_48:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_12:%.*]] = shufflevector <4 x i32> [[DOTOFF_48]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_12:%.*]] = shufflevector <64 x i32> [[DOTPARTS_8]], <64 x i32> [[DOTEXT_12]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 64, i32 65, i32 66, i32 67, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_64:%.*]] = add nuw i32 [[TMP1]], 64
+; CHECK-NEXT:    [[DOTOFF_64:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_64]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_16:%.*]] = shufflevector <4 x i32> [[DOTOFF_64]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_16:%.*]] = shufflevector <64 x i32> [[DOTPARTS_12]], <64 x i32> [[DOTEXT_16]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_80:%.*]] = add nuw i32 [[TMP1]], 80
+; CHECK-NEXT:    [[DOTOFF_80:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_80]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_20:%.*]] = shufflevector <4 x i32> [[DOTOFF_80]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_20:%.*]] = shufflevector <64 x i32> [[DOTPARTS_16]], <64 x i32> [[DOTEXT_20]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 64, i32 65, i32 66, i32 67, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_96:%.*]] = add nuw i32 [[TMP1]], 96
+; CHECK-NEXT:    [[DOTOFF_96:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_96]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_24:%.*]] = shufflevector <4 x i32> [[DOTOFF_96]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_24:%.*]] = shufflevector <64 x i32> [[DOTPARTS_20]], <64 x i32> [[DOTEXT_24]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 64, i32 65, i32 66, i32 67, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_112:%.*]] = add nuw i32 [[TMP1]], 112
+; CHECK-NEXT:    [[DOTOFF_112:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_112]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_28:%.*]] = shufflevector <4 x i32> [[DOTOFF_112]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_28:%.*]] = shufflevector <64 x i32> [[DOTPARTS_24]], <64 x i32> [[DOTEXT_28]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 64, i32 65, i32 66, i32 67, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_128:%.*]] = add nuw i32 [[TMP1]], 128
+; CHECK-NEXT:    [[DOTOFF_128:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_128]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_32:%.*]] = shufflevector <4 x i32> [[DOTOFF_128]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_32:%.*]] = shufflevector <64 x i32> [[DOTPARTS_28]], <64 x i32> [[DOTEXT_32]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 64, i32 65, i32 66, i32 67, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_144:%.*]] = add nuw i32 [[TMP1]], 144
+; CHECK-NEXT:    [[DOTOFF_144:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_144]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_36:%.*]] = shufflevector <4 x i32> [[DOTOFF_144]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_36:%.*]] = shufflevector <64 x i32> [[DOTPARTS_32]], <64 x i32> [[DOTEXT_36]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 64, i32 65, i32 66, i32 67, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_160:%.*]] = add nuw i32 [[TMP1]], 160
+; CHECK-NEXT:    [[DOTOFF_160:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_160]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_40:%.*]] = shufflevector <4 x i32> [[DOTOFF_160]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_40:%.*]] = shufflevector <64 x i32> [[DOTPARTS_36]], <64 x i32> [[DOTEXT_40]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 64, i32 65, i32 66, i32 67, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_176:%.*]] = add nuw i32 [[TMP1]], 176
+; CHECK-NEXT:    [[DOTOFF_176:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_176]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_44:%.*]] = shufflevector <4 x i32> [[DOTOFF_176]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_44:%.*]] = shufflevector <64 x i32> [[DOTPARTS_40]], <64 x i32> [[DOTEXT_44]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 64, i32 65, i32 66, i32 67, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_192:%.*]] = add nuw i32 [[TMP1]], 192
+; CHECK-NEXT:    [[DOTOFF_192:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_192]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_48:%.*]] = shufflevector <4 x i32> [[DOTOFF_192]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_48:%.*]] = shufflevector <64 x i32> [[DOTPARTS_44]], <64 x i32> [[DOTEXT_48]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 64, i32 65, i32 66, i32 67, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_208:%.*]] = add nuw i32 [[TMP1]], 208
+; CHECK-NEXT:    [[DOTOFF_208:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_208]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_52:%.*]] = shufflevector <4 x i32> [[DOTOFF_208]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_52:%.*]] = shufflevector <64 x i32> [[DOTPARTS_48]], <64 x i32> [[DOTEXT_52]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 64, i32 65, i32 66, i32 67, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_224:%.*]] = add nuw i32 [[TMP1]], 224
+; CHECK-NEXT:    [[DOTOFF_224:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_224]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_56:%.*]] = shufflevector <4 x i32> [[DOTOFF_224]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_56:%.*]] = shufflevector <64 x i32> [[DOTPARTS_52]], <64 x i32> [[DOTEXT_56]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 64, i32 65, i32 66, i32 67, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_240:%.*]] = add nuw i32 [[TMP1]], 240
+; CHECK-NEXT:    [[DOTOFF_240:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_240]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_60:%.*]] = shufflevector <4 x i32> [[DOTOFF_240]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <64 x i32> [[DOTPARTS_56]], <64 x i32> [[DOTEXT_60]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 64, i32 65, i32 66, i32 67>
+; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, ptr addrspace(1) [[DST]], i32 [[LOOP_INDEX]]
+; CHECK-NEXT:    store <64 x i32> [[TMP2]], ptr addrspace(1) [[TMP3]], align 16
+; CHECK-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p1.p7.i32(ptr addrspace(1) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 8192, i1 false)
+  ret void
+}
+
+;; This could be the direct-to-LDS intrinsics in a future patch
+define void @memcpy.inline_known_p7_to_p3(ptr addrspace(7) inreg %src, ptr addrspace(3) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_p7_to_p3(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], ptr addrspace(3) inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[DST]], i32 0
+; CHECK-NEXT:    store <4 x i32> [[TMP1]], ptr addrspace(3) [[TMP2]], align 16
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p3.p7.i32(ptr addrspace(3) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 16, i1 false)
+  ret void
+}
+
+define void @memcpy.inline_known_p7_to_p3_byte(ptr addrspace(7) inreg %src, ptr addrspace(3) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_p7_to_p3_byte(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], ptr addrspace(3) inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @llvm.amdgcn.raw.ptr.buffer.load.i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[SRC_OFF]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[DST]], i32 0
+; CHECK-NEXT:    store i32 [[TMP1]], ptr addrspace(3) [[TMP2]], align 16
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p3.p7.i32(ptr addrspace(3) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 4, i1 false)
+  ret void
+}
+
+define void @memcpy.inline_known_p7_to_p3_long(ptr addrspace(7) inreg %src, ptr addrspace(3) inreg %dst) {
+; CHECK-LABEL: define void @memcpy.inline_known_p7_to_p3_long(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[SRC:%.*]], ptr addrspace(3) inreg [[DST:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[SRC_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 0
+; CHECK-NEXT:    [[SRC_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[SRC]], 1
+; CHECK-NEXT:    br label %[[LOAD_STORE_LOOP:.*]]
+; CHECK:       [[LOAD_STORE_LOOP]]:
+; CHECK-NEXT:    [[LOOP_INDEX:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOAD_STORE_LOOP]] ]
+; CHECK-NEXT:    [[TMP1:%.*]] = add i32 [[SRC_OFF]], [[LOOP_INDEX]]
+; CHECK-NEXT:    [[DOTOFF_0:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[TMP1]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_0:%.*]] = shufflevector <4 x i32> [[DOTOFF_0]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_0:%.*]] = shufflevector <64 x i32> poison, <64 x i32> [[DOTEXT_0]], <64 x i32> <i32 64, i32 65, i32 66, i32 67, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_16:%.*]] = add nuw i32 [[TMP1]], 16
+; CHECK-NEXT:    [[DOTOFF_16:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_16]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_4:%.*]] = shufflevector <4 x i32> [[DOTOFF_16]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_4:%.*]] = shufflevector <64 x i32> [[DOTPARTS_0]], <64 x i32> [[DOTEXT_4]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 64, i32 65, i32 66, i32 67, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_32:%.*]] = add nuw i32 [[TMP1]], 32
+; CHECK-NEXT:    [[DOTOFF_32:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_32]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_8:%.*]] = shufflevector <4 x i32> [[DOTOFF_32]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_8:%.*]] = shufflevector <64 x i32> [[DOTPARTS_4]], <64 x i32> [[DOTEXT_8]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 64, i32 65, i32 66, i32 67, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_48:%.*]] = add nuw i32 [[TMP1]], 48
+; CHECK-NEXT:    [[DOTOFF_48:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_48]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_12:%.*]] = shufflevector <4 x i32> [[DOTOFF_48]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_12:%.*]] = shufflevector <64 x i32> [[DOTPARTS_8]], <64 x i32> [[DOTEXT_12]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 64, i32 65, i32 66, i32 67, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_64:%.*]] = add nuw i32 [[TMP1]], 64
+; CHECK-NEXT:    [[DOTOFF_64:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_64]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_16:%.*]] = shufflevector <4 x i32> [[DOTOFF_64]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_16:%.*]] = shufflevector <64 x i32> [[DOTPARTS_12]], <64 x i32> [[DOTEXT_16]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 64, i32 65, i32 66, i32 67, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_80:%.*]] = add nuw i32 [[TMP1]], 80
+; CHECK-NEXT:    [[DOTOFF_80:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_80]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_20:%.*]] = shufflevector <4 x i32> [[DOTOFF_80]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_20:%.*]] = shufflevector <64 x i32> [[DOTPARTS_16]], <64 x i32> [[DOTEXT_20]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 64, i32 65, i32 66, i32 67, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_96:%.*]] = add nuw i32 [[TMP1]], 96
+; CHECK-NEXT:    [[DOTOFF_96:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_96]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_24:%.*]] = shufflevector <4 x i32> [[DOTOFF_96]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_24:%.*]] = shufflevector <64 x i32> [[DOTPARTS_20]], <64 x i32> [[DOTEXT_24]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 64, i32 65, i32 66, i32 67, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_112:%.*]] = add nuw i32 [[TMP1]], 112
+; CHECK-NEXT:    [[DOTOFF_112:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_112]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_28:%.*]] = shufflevector <4 x i32> [[DOTOFF_112]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_28:%.*]] = shufflevector <64 x i32> [[DOTPARTS_24]], <64 x i32> [[DOTEXT_28]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 64, i32 65, i32 66, i32 67, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_128:%.*]] = add nuw i32 [[TMP1]], 128
+; CHECK-NEXT:    [[DOTOFF_128:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_128]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_32:%.*]] = shufflevector <4 x i32> [[DOTOFF_128]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_32:%.*]] = shufflevector <64 x i32> [[DOTPARTS_28]], <64 x i32> [[DOTEXT_32]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 64, i32 65, i32 66, i32 67, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_144:%.*]] = add nuw i32 [[TMP1]], 144
+; CHECK-NEXT:    [[DOTOFF_144:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_144]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_36:%.*]] = shufflevector <4 x i32> [[DOTOFF_144]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_36:%.*]] = shufflevector <64 x i32> [[DOTPARTS_32]], <64 x i32> [[DOTEXT_36]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 64, i32 65, i32 66, i32 67, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_160:%.*]] = add nuw i32 [[TMP1]], 160
+; CHECK-NEXT:    [[DOTOFF_160:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_160]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_40:%.*]] = shufflevector <4 x i32> [[DOTOFF_160]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_40:%.*]] = shufflevector <64 x i32> [[DOTPARTS_36]], <64 x i32> [[DOTEXT_40]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 64, i32 65, i32 66, i32 67, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_176:%.*]] = add nuw i32 [[TMP1]], 176
+; CHECK-NEXT:    [[DOTOFF_176:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_176]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_44:%.*]] = shufflevector <4 x i32> [[DOTOFF_176]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_44:%.*]] = shufflevector <64 x i32> [[DOTPARTS_40]], <64 x i32> [[DOTEXT_44]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 64, i32 65, i32 66, i32 67, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_192:%.*]] = add nuw i32 [[TMP1]], 192
+; CHECK-NEXT:    [[DOTOFF_192:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_192]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_48:%.*]] = shufflevector <4 x i32> [[DOTOFF_192]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_48:%.*]] = shufflevector <64 x i32> [[DOTPARTS_44]], <64 x i32> [[DOTEXT_48]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 64, i32 65, i32 66, i32 67, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_208:%.*]] = add nuw i32 [[TMP1]], 208
+; CHECK-NEXT:    [[DOTOFF_208:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_208]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_52:%.*]] = shufflevector <4 x i32> [[DOTOFF_208]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_52:%.*]] = shufflevector <64 x i32> [[DOTPARTS_48]], <64 x i32> [[DOTEXT_52]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 64, i32 65, i32 66, i32 67, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_224:%.*]] = add nuw i32 [[TMP1]], 224
+; CHECK-NEXT:    [[DOTOFF_224:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_224]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_56:%.*]] = shufflevector <4 x i32> [[DOTOFF_224]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOTPARTS_56:%.*]] = shufflevector <64 x i32> [[DOTPARTS_52]], <64 x i32> [[DOTEXT_56]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 64, i32 65, i32 66, i32 67, i32 60, i32 61, i32 62, i32 63>
+; CHECK-NEXT:    [[DOTOFF_PTR_240:%.*]] = add nuw i32 [[TMP1]], 240
+; CHECK-NEXT:    [[DOTOFF_240:%.*]] = call <4 x i32> @llvm.amdgcn.raw.ptr.buffer.load.v4i32(ptr addrspace(8) align 16 [[SRC_RSRC]], i32 [[DOTOFF_PTR_240]], i32 0, i32 0)
+; CHECK-NEXT:    [[DOTEXT_60:%.*]] = shufflevector <4 x i32> [[DOTOFF_240]], <4 x i32> poison, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <64 x i32> [[DOTPARTS_56]], <64 x i32> [[DOTEXT_60]], <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 64, i32 65, i32 66, i32 67>
+; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i8, ptr addrspace(3) [[DST]], i32 [[LOOP_INDEX]]
+; CHECK-NEXT:    store <64 x i32> [[TMP2]], ptr addrspace(3) [[TMP3]], align 16
+; CHECK-NEXT:    [[TMP4]] = add i32 [[LOOP_INDEX]], 256
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], 8192
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOAD_STORE_LOOP]], label %[[MEMCPY_SPLIT:.*]]
+; CHECK:       [[MEMCPY_SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memcpy.inline.p3.p7.i32(ptr addrspace(3) noundef nonnull align 16 %dst, ptr addrspace(7) noundef nonnull align 16 %src, i32 8192, i1 false)
+  ret void
+}
+
+;; memset
+
+declare void @llvm.memset.p7.i32(ptr addrspace(7), i8, i32, i1)
+declare void @llvm.memset.p7.i64(ptr addrspace(7), i8, i64, i1)
+
+define void @memset_known(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_known(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 8192
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.p7.i32(ptr addrspace(7) noundef nonnull align 16 %ptr, i8 1, i32 8192, i1 false)
+  ret void
+}
+
+define void @memset_known_small(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_known_small(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 32
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 32, i1 false)
+  ret void
+}
+
+define void @memset_known_byte(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_known_byte(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 1
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 1, i1 false)
+  ret void
+}
+
+define void @memset_known_tail(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_known_tail(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 15
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 15, i1 false)
+  ret void
+}
+
+define void @memset_known_i64(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_known_i64(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i64 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[DOTC:%.*]] = trunc i64 [[TMP1]] to i32
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[DOTC]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i64 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i64 [[TMP3]], 8192
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.p7.i64(ptr addrspace(7) %ptr, i8 1, i64 8192, i1 false)
+  ret void
+}
+
+define void @memset_known_i32_volatile(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_known_i32_volatile(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 32
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 32, i1 true)
+  ret void
+}
+
+define void @memset_unknown(ptr addrspace(7) inreg %ptr, i32 inreg %length) {
+; CHECK-LABEL: define void @memset_unknown(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]], i32 inreg [[LENGTH:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 0, [[LENGTH]]
+; CHECK-NEXT:    br i1 [[TMP1]], label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP2:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[PTR_OFF]], [[TMP2]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP3]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4]] = add i32 [[TMP2]], 1
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], [[LENGTH]]
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 %length, i1 false)
+  ret void
+}
+
+;; memset.inline
+
+declare void @llvm.memset.inline.p7.i32(ptr addrspace(7), i8, i32, i1)
+declare void @llvm.memset.inline.p7.i64(ptr addrspace(7), i8, i64, i1)
+
+define void @memset.inline_known(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset.inline_known(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 8192
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.inline.p7.i32(ptr addrspace(7) noundef nonnull align 16 %ptr, i8 1, i32 8192, i1 false)
+  ret void
+}
+
+define void @memset.inline_known_small(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset.inline_known_small(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 32
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.inline.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 32, i1 false)
+  ret void
+}
+
+define void @memset.inline_known_byte(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset.inline_known_byte(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 1
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.inline.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 1, i1 false)
+  ret void
+}
+
+define void @memset.inline_known_tail(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset.inline_known_tail(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 15
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.inline.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 15, i1 false)
+  ret void
+}
+
+define void @memset.inline_known_i64(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset.inline_known_i64(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i64 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[DOTC:%.*]] = trunc i64 [[TMP1]] to i32
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[DOTC]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i64 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i64 [[TMP3]], 8192
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.inline.p7.i64(ptr addrspace(7) %ptr, i8 1, i64 8192, i1 false)
+  ret void
+}
+
+define void @memset.inline_known_i32_volatile(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset.inline_known_i32_volatile(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[TMP1]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 32
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.inline.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 32, i1 true)
+  ret void
+}
+
+define void @memset.inline_unknown(ptr addrspace(7) inreg %ptr, i32 inreg %length) {
+; CHECK-LABEL: define void @memset.inline_unknown(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]], i32 inreg [[LENGTH:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 0, [[LENGTH]]
+; CHECK-NEXT:    br i1 [[TMP1]], label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP2:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[PTR_OFF]], [[TMP2]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i8(i8 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP3]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4]] = add i32 [[TMP2]], 1
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], [[LENGTH]]
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.memset.inline.p7.i32(ptr addrspace(7) %ptr, i8 1, i32 %length, i1 false)
+  ret void
+}
+
+;; memset.pattern
+
+declare void @llvm.experimental.memset.pattern.p7.i32.i32(ptr addrspace(7), i32, i32, i1)
+declare void @llvm.experimental.memset.pattern.p7.i32.i64(ptr addrspace(7), i32, i64, i1)
+
+define void @memset_pattern_known(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_pattern_known(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[DOTIDX:%.*]] = mul nsw i32 [[TMP1]], 4
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[DOTIDX]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 1, ptr addrspace(8) align 4 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 8192
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.experimental.memset.pattern.p7.i32.i32(ptr addrspace(7) noundef nonnull align 16 %ptr, i32 1, i32 8192, i1 false)
+  ret void
+}
+
+define void @memset_pattern_known_small(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_pattern_known_small(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[DOTIDX:%.*]] = mul nsw i32 [[TMP1]], 4
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[DOTIDX]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 32
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.experimental.memset.pattern.p7.i32.i32(ptr addrspace(7) %ptr, i32 1, i32 32, i1 false)
+  ret void
+}
+
+define void @memset_pattern_known_i64(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_pattern_known_i64(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i64 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[DOTC:%.*]] = trunc i64 [[TMP1]] to i32
+; CHECK-NEXT:    [[DOTIDX:%.*]] = mul nsw i32 [[DOTC]], 4
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[DOTIDX]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP3]] = add i64 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i64 [[TMP3]], 8192
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.experimental.memset.pattern.p7.i32.i64(ptr addrspace(7) %ptr, i32 1, i64 8192, i1 false)
+  ret void
+}
+
+define void @memset_pattern_known_i32_volatile(ptr addrspace(7) inreg %ptr) {
+; CHECK-LABEL: define void @memset_pattern_known_i32_volatile(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    br i1 false, label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP1:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP3:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[DOTIDX:%.*]] = mul nsw i32 [[TMP1]], 4
+; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[PTR_OFF]], [[DOTIDX]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP2]], i32 0, i32 -2147483648)
+; CHECK-NEXT:    [[TMP3]] = add i32 [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp ult i32 [[TMP3]], 32
+; CHECK-NEXT:    br i1 [[TMP4]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.experimental.memset.pattern.p7.i32.i32(ptr addrspace(7) %ptr, i32 1, i32 32, i1 true)
+  ret void
+}
+
+define void @memset_pattern_unknown(ptr addrspace(7) inreg %ptr, i32 inreg %length) {
+; CHECK-LABEL: define void @memset_pattern_unknown(
+; CHECK-SAME: { ptr addrspace(8), i32 } inreg [[PTR:%.*]], i32 inreg [[LENGTH:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT:    [[PTR_RSRC:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 0
+; CHECK-NEXT:    [[PTR_OFF:%.*]] = extractvalue { ptr addrspace(8), i32 } [[PTR]], 1
+; CHECK-NEXT:    [[TMP1:%.*]] = icmp eq i32 0, [[LENGTH]]
+; CHECK-NEXT:    br i1 [[TMP1]], label %[[SPLIT:.*]], label %[[LOADSTORELOOP:.*]]
+; CHECK:       [[LOADSTORELOOP]]:
+; CHECK-NEXT:    [[TMP2:%.*]] = phi i32 [ 0, [[TMP0:%.*]] ], [ [[TMP4:%.*]], %[[LOADSTORELOOP]] ]
+; CHECK-NEXT:    [[DOTIDX:%.*]] = mul nsw i32 [[TMP2]], 4
+; CHECK-NEXT:    [[TMP3:%.*]] = add i32 [[PTR_OFF]], [[DOTIDX]]
+; CHECK-NEXT:    call void @llvm.amdgcn.raw.ptr.buffer.store.i32(i32 1, ptr addrspace(8) align 1 [[PTR_RSRC]], i32 [[TMP3]], i32 0, i32 0)
+; CHECK-NEXT:    [[TMP4]] = add i32 [[TMP2]], 1
+; CHECK-NEXT:    [[TMP5:%.*]] = icmp ult i32 [[TMP4]], [[LENGTH]]
+; CHECK-NEXT:    br i1 [[TMP5]], label %[[LOADSTORELOOP]], label %[[SPLIT]]
+; CHECK:       [[SPLIT]]:
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.experimental.memset.pattern.p7.i32.i32(ptr addrspace(7) %ptr, i32 1, i32 %length, i1 false)
+  ret void
+}