release/18.x: [AArch64] Remove invalid uabdl patterns. (#89272)

[RSilicon]

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

(cherry picked from commit 851462f)

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: AtariDreams (AtariDreams)

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

(cherry picked from commit 851462f)

---

Full diff: https://github.com/llvm/llvm-project/pull/89380.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 03baa7497615e3..ac61dd8745d4e6 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -4885,19 +4885,9 @@ 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 fe4da2e7cf36b5..89c8d540b97e04 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
+}
+

[RSilicon]

@arsenm @nikic What do you think about this?

[davemgreen]

I think this should be OK for the branch, if it is wanted. It should be a safe commit to backport, considering it just removes some invalid patterns. LGTM.

[tstellar]

Hi @AtariDreams (or anyone else). If you would like to add a note about this fix in the release notes (completely optional). Please reply to this comment with a one or two sentence description of the fix. When you are done, please add the release:note label to this PR.

[RSilicon]

I would add "Fixed issue where on AArch64 some patterns would be miscompiled."