[WebAssembly] Prototype i64x2.widen_{low,high}_i32x4_{s,u}

tlively · arichardson · commit 0ced42d06293 · 2021-03-25T12:37:39.000Z
As proposed in WebAssembly/simd#290. As usual, these instructions are available only via builtin functions and intrinsics while they are in the prototyping stage. Differential Revision: https://reviews.llvm.org/D90504
diff --git a/clang/include/clang/Basic/BuiltinsWebAssembly.def b/clang/include/clang/Basic/BuiltinsWebAssembly.def
@@ -194,6 +194,11 @@ TARGET_BUILTIN(__builtin_wasm_narrow_u_i8x16_i16x8, "V16UcV8UsV8Us", "nc", "simd
 TARGET_BUILTIN(__builtin_wasm_narrow_s_i16x8_i32x4, "V8sV4iV4i", "nc", "simd128")
 TARGET_BUILTIN(__builtin_wasm_narrow_u_i16x8_i32x4, "V8UsV4UiV4Ui", "nc", "simd128")
 
+TARGET_BUILTIN(__builtin_wasm_widen_low_s_i32x4_i64x2, "V2LLiV4i", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_high_s_i32x4_i64x2, "V2LLiV4i", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_low_u_i32x4_i64x2, "V2LLUiV4Ui", "nc", "simd128")
+TARGET_BUILTIN(__builtin_wasm_widen_high_u_i32x4_i64x2, "V2LLUiV4Ui", "nc", "simd128")
+
 TARGET_BUILTIN(__builtin_wasm_load32_zero, "V4ii*", "n", "simd128")
 TARGET_BUILTIN(__builtin_wasm_load64_zero, "V2LLiLLi*", "n", "simd128")
 
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -17151,6 +17151,29 @@ Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
         CGM.getIntrinsic(IntNo, {ConvertType(E->getType()), Low->getType()});
     return Builder.CreateCall(Callee, {Low, High});
   }
+  case WebAssembly::BI__builtin_wasm_widen_low_s_i32x4_i64x2:
+  case WebAssembly::BI__builtin_wasm_widen_high_s_i32x4_i64x2:
+  case WebAssembly::BI__builtin_wasm_widen_low_u_i32x4_i64x2:
+  case WebAssembly::BI__builtin_wasm_widen_high_u_i32x4_i64x2: {
+    Value *Vec = EmitScalarExpr(E->getArg(0));
+    unsigned IntNo;
+    switch (BuiltinID) {
+    case WebAssembly::BI__builtin_wasm_widen_low_s_i32x4_i64x2:
+      IntNo = Intrinsic::wasm_widen_low_signed;
+      break;
+    case WebAssembly::BI__builtin_wasm_widen_high_s_i32x4_i64x2:
+      IntNo = Intrinsic::wasm_widen_high_signed;
+      break;
+    case WebAssembly::BI__builtin_wasm_widen_low_u_i32x4_i64x2:
+      IntNo = Intrinsic::wasm_widen_low_unsigned;
+      break;
+    case WebAssembly::BI__builtin_wasm_widen_high_u_i32x4_i64x2:
+      IntNo = Intrinsic::wasm_widen_high_unsigned;
+      break;
+    }
+    Function *Callee = CGM.getIntrinsic(IntNo);
+    return Builder.CreateCall(Callee, Vec);
+  }
   case WebAssembly::BI__builtin_wasm_load32_zero: {
     Value *Ptr = EmitScalarExpr(E->getArg(0));
     Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_load32_zero);
diff --git a/clang/test/CodeGen/builtins-wasm.c b/clang/test/CodeGen/builtins-wasm.c
@@ -918,6 +918,30 @@ u16x8 narrow_u_i16x8_i32x4(u32x4 low, u32x4 high) {
   // WEBASSEMBLY: ret
 }
 
+i64x2 widen_low_s_i32x4_i64x2(i32x4 x) {
+  return __builtin_wasm_widen_low_s_i32x4_i64x2(x);
+  // WEBASSEMBLY: call <2 x i64> @llvm.wasm.widen.low.signed(<4 x i32> %x)
+  // WEBASSEMBLY: ret
+}
+
+i64x2 widen_high_s_i32x4_i64x2(i32x4 x) {
+  return __builtin_wasm_widen_high_s_i32x4_i64x2(x);
+  // WEBASSEMBLY: call <2 x i64> @llvm.wasm.widen.high.signed(<4 x i32> %x)
+  // WEBASSEMBLY: ret
+}
+
+u64x2 widen_low_u_i32x4_i64x2(u32x4 x) {
+  return __builtin_wasm_widen_low_u_i32x4_i64x2(x);
+  // WEBASSEMBLY: call <2 x i64> @llvm.wasm.widen.low.unsigned(<4 x i32> %x)
+  // WEBASSEMBLY: ret
+}
+
+u64x2 widen_high_u_i32x4_i64x2(u32x4 x) {
+  return __builtin_wasm_widen_high_u_i32x4_i64x2(x);
+  // WEBASSEMBLY: call <2 x i64> @llvm.wasm.widen.high.unsigned(<4 x i32> %x)
+  // WEBASSEMBLY: ret
+}
+
 i32x4 load32_zero(int *p) {
   return __builtin_wasm_load32_zero(p);
   // WEBASSEMBLY: call <4 x i32> @llvm.wasm.load32.zero(i32* %p)
diff --git a/llvm/include/llvm/IR/IntrinsicsWebAssembly.td b/llvm/include/llvm/IR/IntrinsicsWebAssembly.td
@@ -151,6 +151,7 @@ def int_wasm_dot :
   Intrinsic<[llvm_v4i32_ty],
             [llvm_v8i16_ty, llvm_v8i16_ty],
             [IntrNoMem, IntrSpeculatable]>;
+
 def int_wasm_narrow_signed :
   Intrinsic<[llvm_anyvector_ty],
             [llvm_anyvector_ty, LLVMMatchType<1>],
@@ -159,6 +160,18 @@ def int_wasm_narrow_unsigned :
   Intrinsic<[llvm_anyvector_ty],
             [llvm_anyvector_ty, LLVMMatchType<1>],
             [IntrNoMem, IntrSpeculatable]>;
+
+// TODO: Replace these intrinsics with normal ISel patterns once i32x4 to i64x2
+// widening is merged to the proposal.
+def int_wasm_widen_low_signed :
+  Intrinsic<[llvm_v2i64_ty], [llvm_v4i32_ty], [IntrNoMem, IntrSpeculatable]>;
+def int_wasm_widen_high_signed :
+  Intrinsic<[llvm_v2i64_ty], [llvm_v4i32_ty], [IntrNoMem, IntrSpeculatable]>;
+def int_wasm_widen_low_unsigned :
+  Intrinsic<[llvm_v2i64_ty], [llvm_v4i32_ty], [IntrNoMem, IntrSpeculatable]>;
+def int_wasm_widen_high_unsigned :
+  Intrinsic<[llvm_v2i64_ty], [llvm_v4i32_ty], [IntrNoMem, IntrSpeculatable]>;
+
 def int_wasm_q15mulr_saturate_signed :
   Intrinsic<[llvm_v8i16_ty],
             [llvm_v8i16_ty, llvm_v8i16_ty],
diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td
@@ -1116,6 +1116,15 @@ multiclass SIMDWiden<ValueType vec_t, string vec, ValueType arg_t, string arg,
 defm "" : SIMDWiden<v8i16, "i16x8", v16i8, "i8x16", 135>;
 defm "" : SIMDWiden<v4i32, "i32x4", v8i16, "i16x8", 167>;
 
+defm "" : SIMDConvert<v2i64, v4i32, int_wasm_widen_low_signed,
+                      "i64x2.widen_low_i32x4_s", 199>;
+defm "" : SIMDConvert<v2i64, v4i32, int_wasm_widen_high_signed,
+                      "i64x2.widen_high_i32x4_s", 200>;
+defm "" : SIMDConvert<v2i64, v4i32, int_wasm_widen_low_unsigned,
+                      "i64x2.widen_low_i32x4_u", 201>;
+defm "" : SIMDConvert<v2i64, v4i32, int_wasm_widen_high_unsigned,
+                      "i64x2.widen_high_i32x4_u", 202>;
+
 // Narrowing operations
 multiclass SIMDNarrow<ValueType vec_t, string vec, ValueType arg_t, string arg,
                       bits<32> baseInst> {
diff --git a/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll b/llvm/test/CodeGen/WebAssembly/simd-intrinsics.ll
@@ -528,6 +528,46 @@ define <4 x i32> @trunc_sat_u_v4i32(<4 x float> %x) {
 ; ==============================================================================
 ; 2 x i64
 ; ==============================================================================
+; CHECK-LABEL: widen_low_s_v2i64:
+; SIMD128-NEXT: .functype widen_low_s_v2i64 (v128) -> (v128){{$}}
+; SIMD128-NEXT: i64x2.widen_low_i32x4_s $push[[R:[0-9]+]]=, $0{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <2 x i64> @llvm.wasm.widen.low.signed(<4 x i32>)
+define <2 x i64> @widen_low_s_v2i64(<4 x i32> %x) {
+  %a = call <2 x i64> @llvm.wasm.widen.low.signed(<4 x i32> %x)
+  ret <2 x i64> %a
+}
+
+; CHECK-LABEL: widen_high_s_v2i64:
+; SIMD128-NEXT: .functype widen_high_s_v2i64 (v128) -> (v128){{$}}
+; SIMD128-NEXT: i64x2.widen_high_i32x4_s $push[[R:[0-9]+]]=, $0{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <2 x i64> @llvm.wasm.widen.high.signed(<4 x i32>)
+define <2 x i64> @widen_high_s_v2i64(<4 x i32> %x) {
+  %a = call <2 x i64> @llvm.wasm.widen.high.signed(<4 x i32> %x)
+  ret <2 x i64> %a
+}
+
+; CHECK-LABEL: widen_low_u_v2i64:
+; SIMD128-NEXT: .functype widen_low_u_v2i64 (v128) -> (v128){{$}}
+; SIMD128-NEXT: i64x2.widen_low_i32x4_u $push[[R:[0-9]+]]=, $0{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <2 x i64> @llvm.wasm.widen.low.unsigned(<4 x i32>)
+define <2 x i64> @widen_low_u_v2i64(<4 x i32> %x) {
+  %a = call <2 x i64> @llvm.wasm.widen.low.unsigned(<4 x i32> %x)
+  ret <2 x i64> %a
+}
+
+; CHECK-LABEL: widen_high_u_v2i64:
+; SIMD128-NEXT: .functype widen_high_u_v2i64 (v128) -> (v128){{$}}
+; SIMD128-NEXT: i64x2.widen_high_i32x4_u $push[[R:[0-9]+]]=, $0{{$}}
+; SIMD128-NEXT: return $pop[[R]]{{$}}
+declare <2 x i64> @llvm.wasm.widen.high.unsigned(<4 x i32>)
+define <2 x i64> @widen_high_u_v2i64(<4 x i32> %x) {
+  %a = call <2 x i64> @llvm.wasm.widen.high.unsigned(<4 x i32> %x)
+  ret <2 x i64> %a
+}
+
 ; CHECK-LABEL: extmul_low_s_v2i64:
 ; SIMD128-NEXT: .functype extmul_low_s_v2i64 (v128, v128) -> (v128){{$}}
 ; SIMD128-NEXT: i64x2.extmul_low_i32x4_s $push[[R:[0-9]+]]=, $0, $1{{$}}
diff --git a/llvm/test/MC/WebAssembly/simd-encodings.s b/llvm/test/MC/WebAssembly/simd-encodings.s
@@ -520,6 +520,18 @@ main:
     # CHECK: i64x2.all_true # encoding: [0xfd,0xc3,0x01]
     i64x2.all_true
 
+    # CHECK: i64x2.widen_low_i32x4_s # encoding: [0xfd,0xc7,0x01]
+    i64x2.widen_low_i32x4_s
+
+    # CHECK: i64x2.widen_high_i32x4_s # encoding: [0xfd,0xc8,0x01]
+    i64x2.widen_high_i32x4_s
+
+    # CHECK: i64x2.widen_low_i32x4_u # encoding: [0xfd,0xc9,0x01]
+    i64x2.widen_low_i32x4_u
+
+    # CHECK: i64x2.widen_high_i32x4_u # encoding: [0xfd,0xca,0x01]
+    i64x2.widen_high_i32x4_u
+
     # CHECK: i64x2.shl # encoding: [0xfd,0xcb,0x01]
     i64x2.shl
 
