[AArch64] Fix op mask detection in performZExtDeinterleaveShuffleCombine

[davemgreen]

Given a zext from an extract vector, with a shuffle mask like <4, 0, 0, 4> we
would previously recognize the top half as a deinterleave. In order to convert
into a uzp we should have been checking that the bottom half is also poison.

Fixes #125989

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: David Green (davemgreen)

Changes

Given a zext from an extract vector, with a shuffle mask like <4, 0, 0, 4> we
would previously recognize the top half as a deinterleave. In order to convert
into a uzp we should have been checking that the bottom half is also poison.

Fixes #125989

---

Full diff: https://github.com/llvm/llvm-project/pull/126054.diff

2 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64ISelLowering.cpp (+2)
• (modified) llvm/test/CodeGen/AArch64/zext-shuffle.ll (+143)

diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
index 8617377ffc55b5..d31ed4a1159eec 100644
--- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -22332,6 +22332,8 @@ static SDValue performZExtDeinterleaveShuffleCombine(SDNode *N,
   if (!IsDeInterleave)
     IsUndefDeInterleave =
         Shuffle->getOperand(1).isUndef() &&
+        all_of(Shuffle->getMask().slice(ExtOffset, VT.getVectorNumElements() / 2),
+               [](int M) { return M < 0; }) &&
         ShuffleVectorInst::isDeInterleaveMaskOfFactor(
             Shuffle->getMask().slice(ExtOffset + VT.getVectorNumElements() / 2,
                                      VT.getVectorNumElements() / 2),
diff --git a/llvm/test/CodeGen/AArch64/zext-shuffle.ll b/llvm/test/CodeGen/AArch64/zext-shuffle.ll
index 2965996ddcb026..20d2071d7fe54d 100644
--- a/llvm/test/CodeGen/AArch64/zext-shuffle.ll
+++ b/llvm/test/CodeGen/AArch64/zext-shuffle.ll
@@ -543,3 +543,146 @@ define <8 x double> @uitofp_load_fadd(ptr %p) {
     ret <8 x double> %c
 }
 
+define <4 x i32> @isUndefDeInterleave_b0(<8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: isUndefDeInterleave_b0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    movi v1.2d, #0x00ffff0000ffff
+; CHECK-NEXT:    uzp1 v0.4s, v0.4s, v0.4s
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 0, i32 4, i32 poison, i32 poison, i32 poison, i32 poison>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define <4 x i32> @isUndefDeInterleave_b1(<8 x i16> %a) {
+; CHECK-LABEL: isUndefDeInterleave_b1:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uzp1 v0.4s, v0.4s, v0.4s
+; CHECK-NEXT:    ushr v0.4s, v0.4s, #16
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 1, i32 5, i32 poison, i32 poison, i32 poison, i32 poison>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define <4 x i32> @isUndefDeInterleave_b2(<8 x i16> %a) {
+; CHECK-LABEL: isUndefDeInterleave_b2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    movi v1.2d, #0x00ffff0000ffff
+; CHECK-NEXT:    uzp2 v0.4s, v0.4s, v0.4s
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 2, i32 6, i32 poison, i32 poison, i32 poison, i32 poison>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define <4 x i32> @isUndefDeInterleave_b3(<8 x i16> %a) {
+; CHECK-LABEL: isUndefDeInterleave_b3:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uzp2 v0.4s, v0.4s, v0.4s
+; CHECK-NEXT:    ushr v0.4s, v0.4s, #16
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 3, i32 7, i32 poison, i32 poison, i32 poison, i32 poison>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define <4 x i32> @isUndefDeInterleave_t0(<8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: isUndefDeInterleave_t0:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    movi v1.2d, #0x00ffff0000ffff
+; CHECK-NEXT:    uzp1 v0.4s, v0.4s, v0.4s
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 0, i32 4>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define <4 x i32> @isUndefDeInterleave_t1(<8 x i16> %a) {
+; CHECK-LABEL: isUndefDeInterleave_t1:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uzp1 v0.4s, v0.4s, v0.4s
+; CHECK-NEXT:    ushr v0.4s, v0.4s, #16
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 1, i32 5>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define <4 x i32> @isUndefDeInterleave_t2(<8 x i16> %a) {
+; CHECK-LABEL: isUndefDeInterleave_t2:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    movi v1.2d, #0x00ffff0000ffff
+; CHECK-NEXT:    uzp2 v0.4s, v0.4s, v0.4s
+; CHECK-NEXT:    and v0.16b, v0.16b, v1.16b
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 2, i32 6>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define <4 x i32> @isUndefDeInterleave_t3(<8 x i16> %a) {
+; CHECK-LABEL: isUndefDeInterleave_t3:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    uzp2 v0.4s, v0.4s, v0.4s
+; CHECK-NEXT:    ushr v0.4s, v0.4s, #16
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 3, i32 7>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define <4 x i32> @isUndefDeInterleave_b0_bad(<8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: isUndefDeInterleave_b0_bad:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x8, .LCPI40_0
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI40_0]
+; CHECK-NEXT:    tbl v0.16b, { v0.16b }, v1.16b
+; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 4, i32 4, i32 0, i32 4, i32 poison, i32 poison, i32 poison, i32 poison>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define <4 x i32> @isUndefDeInterleave_t1_bad(<8 x i16> %a) {
+; CHECK-LABEL: isUndefDeInterleave_t1_bad:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    adrp x8, .LCPI41_0
+; CHECK-NEXT:    ldr q1, [x8, :lo12:.LCPI41_0]
+; CHECK-NEXT:    tbl v0.16b, { v0.16b }, v1.16b
+; CHECK-NEXT:    ushll2 v0.4s, v0.8h, #0
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %a, <8 x i16> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 4, i32 4, i32 1, i32 5>
+  %s2 = shufflevector <8 x i16> %2, <8 x i16> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+  %3 = zext <4 x i16> %s2 to <4 x i32>
+  ret <4 x i32> %3
+}
+
+define i16 @undeftop(<8 x i16> %0) {
+; CHECK-LABEL: undeftop:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    dup v0.8h, v0.h[4]
+; CHECK-NEXT:    uaddl v0.4s, v0.4h, v0.4h
+; CHECK-NEXT:    xtn v0.4h, v0.4s
+; CHECK-NEXT:    umov w0, v0.h[0]
+; CHECK-NEXT:    ret
+  %2 = shufflevector <8 x i16> %0, <8 x i16> zeroinitializer, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 9, i32 7, i32 5, i32 3>
+  %3 = zext <8 x i16> %2 to <8 x i64>
+  %new0 = add <8 x i64> %3, %3
+  %last = trunc <8 x i64> %new0 to <8 x i16>
+  %4 = extractelement <8 x i16> %last, i32 0
+  ret i16 %4
+}

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[SamTebbs33]

Nice one.