release/18.x: [AArch64] Only apply bool vector bitcast opt if result is scalar (#81256)

[llvmbot]

Backport 92d7992

Requested by: @dianqk

[llvmbot]

@lawben What do you think about merging this PR to the release branch?

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: None (llvmbot)

Changes

Backport 92d7992

Requested by: @DianQK

---

Full diff: https://github.com/llvm/llvm-project/pull/81454.diff

2 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64ISelLowering.cpp (+2-1)
• (modified) llvm/test/CodeGen/AArch64/vec-combine-compare-to-bitmask.ll (+28)

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index e97f5e32201488..2d6b5bc983e739 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -24419,7 +24419,8 @@ void AArch64TargetLowering::ReplaceBITCASTResults(
     return;
   }
 
-  if (SrcVT.isVector() && SrcVT.getVectorElementType() == MVT::i1)
+  if (SrcVT.isVector() && SrcVT.getVectorElementType() == MVT::i1 &&
+      !VT.isVector())
     return replaceBoolVectorBitcast(N, Results, DAG);
 
   if (VT != MVT::i16 || (SrcVT != MVT::f16 && SrcVT != MVT::bf16))
diff --git a/llvm/test/CodeGen/AArch64/vec-combine-compare-to-bitmask.ll b/llvm/test/CodeGen/AArch64/vec-combine-compare-to-bitmask.ll
index 1b22e2f900ddb7..557aa010b3a7d9 100644
--- a/llvm/test/CodeGen/AArch64/vec-combine-compare-to-bitmask.ll
+++ b/llvm/test/CodeGen/AArch64/vec-combine-compare-to-bitmask.ll
@@ -489,3 +489,31 @@ define i6 @no_combine_illegal_num_elements(<6 x i32> %vec) {
   %bitmask = bitcast <6 x i1> %cmp_result to i6
   ret i6 %bitmask
 }
+
+; Only apply the combine when casting a vector to a scalar.
+define <2 x i8> @vector_to_vector_cast(<16 x i1> %arg) nounwind {
+; CHECK-LABEL: vector_to_vector_cast:
+; CHECK:       ; %bb.0:
+; CHECK-NEXT:    sub sp, sp, #16
+; CHECK-NEXT:    shl.16b v0, v0, #7
+; CHECK-NEXT:  Lloh36:
+; CHECK-NEXT:    adrp x8, lCPI20_0@PAGE
+; CHECK-NEXT:  Lloh37:
+; CHECK-NEXT:    ldr q1, [x8, lCPI20_0@PAGEOFF]
+; CHECK-NEXT:    add x8, sp, #14
+; CHECK-NEXT:    cmlt.16b v0, v0, #0
+; CHECK-NEXT:    and.16b v0, v0, v1
+; CHECK-NEXT:    ext.16b v1, v0, v0, #8
+; CHECK-NEXT:    zip1.16b v0, v0, v1
+; CHECK-NEXT:    addv.8h h0, v0
+; CHECK-NEXT:    str h0, [sp, #14]
+; CHECK-NEXT:    ld1.b { v0 }[0], [x8]
+; CHECK-NEXT:    orr x8, x8, #0x1
+; CHECK-NEXT:    ld1.b { v0 }[4], [x8]
+; CHECK-NEXT:    ; kill: def $d0 killed $d0 killed $q0
+; CHECK-NEXT:    add sp, sp, #16
+; CHECK-NEXT:    ret
+; CHECK-NEXT:    .loh AdrpLdr Lloh36, Lloh37
+  %bc = bitcast <16 x i1> %arg to <2 x i8>
+  ret <2 x i8> %bc
+}

[davemgreen]

Looks good to get onto the branch to me

[lawben]

Sorry, was offline a few days while traveling. I'm reviewing this on my phone right now, but it looks like it's just the bug fix, so LGTM :) thanks for fixing and back porting