[llvm] [DAG] SelectionDAG::canCreateUndefOrPoison - Mark AVGFLOORS and AVGCEILS as safe (PR #148191)

[via llvm-commits]

aabhinavg1 wrote:

> Could you add a tests to llvm/test/CodeGen/AArch64/hadd-combine.ll that use the aarch64.neon.uhadd/urhadd intrinsics? They should be converted to these avg nodes so can be tested directly.


Hi @davemgreen,

Thanks for the suggestion!

I've drafted the following test cases in llvm/test/CodeGen/AArch64/hadd-combine.ll to exercise the aarch64.neon.uhadd and aarch64.neon.urhadd intrinsics:

```cpp
; RUN: llc < %s -mtriple=aarch64-none-linux-gnu -mattr=+neon | FileCheck %s

declare <8 x i8> @llvm.aarch64.neon.uhadd.v8i8(<8 x i8>, <8 x i8>)
declare <8 x i8> @llvm.aarch64.neon.urhadd.v8i8(<8 x i8>, <8 x i8>)
declare <16 x i8> @llvm.aarch64.neon.uhadd.v16i8(<16 x i8>, <16 x i8>)
declare <16 x i8> @llvm.aarch64.neon.urhadd.v16i8(<16 x i8>, <16 x i8>)

; CHECK-LABEL: test_uhadd_v8i8:
; CHECK: uhadd v0.8b, v0.8b, v1.8b
define <8 x i8> @test_uhadd_v8i8(<8 x i8> %a, <8 x i8> %b) {
  %res = call <8 x i8> @llvm.aarch64.neon.uhadd.v8i8(<8 x i8> %a, <8 x i8> %b)
  ret <8 x i8> %res
}

; CHECK-LABEL: test_urhadd_v8i8:
; CHECK: urhadd v0.8b, v0.8b, v1.8b
define <8 x i8> @test_urhadd_v8i8(<8 x i8> %a, <8 x i8> %b) {
  %res = call <8 x i8> @llvm.aarch64.neon.urhadd.v8i8(<8 x i8> %a, <8 x i8> %b)
  ret <8 x i8> %res
}

; CHECK-LABEL: test_uhadd_v16i8:
; CHECK: uhadd v0.16b, v0.16b, v1.16b
define <16 x i8> @test_uhadd_v16i8(<16 x i8> %a, <16 x i8> %b) {
  %res = call <16 x i8> @llvm.aarch64.neon.uhadd.v16i8(<16 x i8> %a, <16 x i8> %b)
  ret <16 x i8> %res
}

; CHECK-LABEL: test_urhadd_v16i8:
; CHECK: urhadd v0.16b, v0.16b, v1.16b
define <16 x i8> @test_urhadd_v16i8(<16 x i8> %a, <16 x i8> %b) {
  %res = call <16 x i8> @llvm.aarch64.neon.urhadd.v16i8(<16 x i8> %a, <16 x i8> %b)
  ret <16 x i8> %res
}
```

Would this be acceptable for verifying that the uhadd/urhadd intrinsics get correctly lowered into the corresponding DAG avg patterns?


https://github.com/llvm/llvm-project/pull/148191