[AArch64] Remove invalid uabdl patterns.

[davemgreen]

These were added in https://reviews.llvm.org/D14208, which look like they attempt to detect abs from xor+add+ashr. They do not appear to be detecting the correct value for the src input though, which I think is intended to be the sub(zext, zext) part of the pattern. We have pattens from abs now, so the old invalid patterns can be removed.

Fixes #88784

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: David Green (davemgreen)

Changes

These were added in https://reviews.llvm.org/D14208, which look like they attempt to detect abs from xor+add+ashr. They do not appear to be detecting the correct value for the src input though, which I think is intended to be the sub(zext, zext) part of the pattern. We have pattens from abs now, so the old invalid patterns can be removed.

Fixes #88784

---

Full diff: https://github.com/llvm/llvm-project/pull/89272.diff

2 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64InstrInfo.td (-10)
• (modified) llvm/test/CodeGen/AArch64/arm64-vabs.ll (+48)

diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td index 3bf90778363c6c..be53be782077b7 100644 --- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td +++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td @@ -4954,19 +4954,9 @@ defm UABDL : SIMDLongThreeVectorBHSabdl<1, 0b0111, "uabdl", def : Pat<(abs (v8i16 (sub (zext (v8i8 V64:$opA)), (zext (v8i8 V64:$opB))))), (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>; -def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))), - (v8i16 (add (sub (zext (v8i8 V64:$opA)), - (zext (v8i8 V64:$opB))), - (AArch64vashr v8i16:$src, (i32 15))))), - (UABDLv8i8_v8i16 V64:$opA, V64:$opB)>; def : Pat<(abs (v8i16 (sub (zext (extract_high_v16i8 (v16i8 V128:$opA))), (zext (extract_high_v16i8 (v16i8 V128:$opB)))))), (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>; -def : Pat<(xor (v8i16 (AArch64vashr v8i16:$src, (i32 15))), - (v8i16 (add (sub (zext (extract_high_v16i8 (v16i8 V128:$opA))), - (zext (extract_high_v16i8 (v16i8 V128:$opB)))), - (AArch64vashr v8i16:$src, (i32 15))))), - (UABDLv16i8_v8i16 V128:$opA, V128:$opB)>; def : Pat<(abs (v4i32 (sub (zext (v4i16 V64:$opA)), (zext (v4i16 V64:$opB))))), (UABDLv4i16_v4i32 V64:$opA, V64:$opB)>; diff --git a/llvm/test/CodeGen/AArch64/arm64-vabs.ll b/llvm/test/CodeGen/AArch64/arm64-vabs.ll index 5b45ba2552cefd..d64327656a9e01 100644 --- a/llvm/test/CodeGen/AArch64/arm64-vabs.ll +++ b/llvm/test/CodeGen/AArch64/arm64-vabs.ll @@ -1848,3 +1848,51 @@ define <2 x i128> @uabd_i64(<2 x i64> %a, <2 x i64> %b) { %absel = select <2 x i1> %abcmp, <2 x i128> %ababs, <2 x i128> %abdiff ret <2 x i128> %absel } + +define <8 x i16> @pr88784(<8 x i8> %l0, <8 x i8> %l1, <8 x i16> %l2) { +; CHECK-SD-LABEL: pr88784: +; CHECK-SD: // %bb.0: +; CHECK-SD-NEXT: usubl.8h v0, v0, v1 +; CHECK-SD-NEXT: cmlt.8h v1, v2, #0 +; CHECK-SD-NEXT: ssra.8h v0, v2, #15 +; CHECK-SD-NEXT: eor.16b v0, v1, v0 +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: pr88784: +; CHECK-GI: // %bb.0: +; CHECK-GI-NEXT: usubl.8h v0, v0, v1 +; CHECK-GI-NEXT: sshr.8h v1, v2, #15 +; CHECK-GI-NEXT: ssra.8h v0, v2, #15 +; CHECK-GI-NEXT: eor.16b v0, v1, v0 +; CHECK-GI-NEXT: ret + %l4 = zext <8 x i8> %l0 to <8 x i16> + %l5 = ashr <8 x i16> %l2, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15> + %l6 = zext <8 x i8> %l1 to <8 x i16> + %l7 = sub <8 x i16> %l4, %l6 + %l8 = add <8 x i16> %l5, %l7 + %l9 = xor <8 x i16> %l5, %l8 + ret <8 x i16> %l9 +} + +define <8 x i16> @pr88784_fixed(<8 x i8> %l0, <8 x i8> %l1, <8 x i16> %l2) { +; CHECK-SD-LABEL: pr88784_fixed: +; CHECK-SD: // %bb.0: +; CHECK-SD-NEXT: uabdl.8h v0, v0, v1 +; CHECK-SD-NEXT: ret +; +; CHECK-GI-LABEL: pr88784_fixed: +; CHECK-GI: // %bb.0: +; CHECK-GI-NEXT: usubl.8h v0, v0, v1 +; CHECK-GI-NEXT: sshr.8h v1, v0, #15 +; CHECK-GI-NEXT: ssra.8h v0, v0, #15 +; CHECK-GI-NEXT: eor.16b v0, v1, v0 +; CHECK-GI-NEXT: ret + %l4 = zext <8 x i8> %l0 to <8 x i16> + %l6 = zext <8 x i8> %l1 to <8 x i16> + %l7 = sub <8 x i16> %l4, %l6 + %l5 = ashr <8 x i16> %l7, <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15> + %l8 = add <8 x i16> %l5, %l7 + %l9 = xor <8 x i16> %l5, %l8 + ret <8 x i16> %l9 +} + 

[efriedma-quic]

LGTM

[AZero13]

We should backport this too methinks