[GlobalISel] Fold bitcast(undef) to undef.

[davemgreen]

As with other operations such as trunc and fp converts, it should be valid to convert bitcast(undef) to undef.

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

@llvm/pr-subscribers-llvm-globalisel

Author: David Green (davemgreen)

Changes

As with other operations such as trunc and fp converts, it should be valid to convert bitcast(undef) to undef.

---

Full diff: https://github.com/llvm/llvm-project/pull/111491.diff

2 Files Affected:

• (modified) llvm/include/llvm/Target/GlobalISel/Combine.td (+1-1)
• (modified) llvm/test/CodeGen/AArch64/concat-vector.ll (+4-15)

diff --git a/llvm/include/llvm/Target/GlobalISel/Combine.td b/llvm/include/llvm/Target/GlobalISel/Combine.td
index f838c6e62a2ce3..77cb4370b54664 100644
--- a/llvm/include/llvm/Target/GlobalISel/Combine.td
+++ b/llvm/include/llvm/Target/GlobalISel/Combine.td
@@ -420,7 +420,7 @@ def unary_undef_to_zero: GICombineRule<
 // replaced with undef.
 def propagate_undef_any_op: GICombineRule<
   (defs root:$root),
-  (match (wip_match_opcode G_ADD, G_FPTOSI, G_FPTOUI, G_SUB, G_XOR, G_TRUNC):$root,
+  (match (wip_match_opcode G_ADD, G_FPTOSI, G_FPTOUI, G_SUB, G_XOR, G_TRUNC, G_BITCAST):$root,
          [{ return Helper.matchAnyExplicitUseIsUndef(*${root}); }]),
   (apply [{ Helper.replaceInstWithUndef(*${root}); }])>;
 
diff --git a/llvm/test/CodeGen/AArch64/concat-vector.ll b/llvm/test/CodeGen/AArch64/concat-vector.ll
index eee917e8acb0d7..d800b2549cf223 100644
--- a/llvm/test/CodeGen/AArch64/concat-vector.ll
+++ b/llvm/test/CodeGen/AArch64/concat-vector.ll
@@ -189,9 +189,6 @@ define <8 x i16> @concat_v8s16_v2s16(ptr %ptr) {
 ; CHECK-GI-NEXT:    xtn v0.4h, v0.4s
 ; CHECK-GI-NEXT:    fmov w8, s0
 ; CHECK-GI-NEXT:    mov v0.s[0], w8
-; CHECK-GI-NEXT:    mov v0.s[1], w8
-; CHECK-GI-NEXT:    mov v0.s[2], w8
-; CHECK-GI-NEXT:    mov v0.s[3], w8
 ; CHECK-GI-NEXT:    ret
     %a = load <2 x i16>, ptr %ptr
     %b = shufflevector <2 x i16> %a, <2 x i16> %a, <8 x i32> <i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
@@ -199,18 +196,10 @@ define <8 x i16> @concat_v8s16_v2s16(ptr %ptr) {
 }
 
 define <16 x i8> @concat_v16s8_v4s8(ptr %ptr) {
-; CHECK-SD-LABEL: concat_v16s8_v4s8:
-; CHECK-SD:       // %bb.0:
-; CHECK-SD-NEXT:    ldr s0, [x0]
-; CHECK-SD-NEXT:    ret
-;
-; CHECK-GI-LABEL: concat_v16s8_v4s8:
-; CHECK-GI:       // %bb.0:
-; CHECK-GI-NEXT:    ldr s0, [x0]
-; CHECK-GI-NEXT:    mov v0.s[1], w8
-; CHECK-GI-NEXT:    mov v0.s[2], w8
-; CHECK-GI-NEXT:    mov v0.s[3], w8
-; CHECK-GI-NEXT:    ret
+; CHECK-LABEL: concat_v16s8_v4s8:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    ldr s0, [x0]
+; CHECK-NEXT:    ret
     %a = load <4 x i8>, ptr %ptr
     %b = shufflevector <4 x i8> %a, <4 x i8> %a, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
     ret <16 x i8> %b

[arsenm]

I'm surprised this is the only test change