[X86] PromoteMaskArithmetic - oneuse vector_extend_inreg nodes can be safely promoted on AVX2+ targets

[RKSimon]

Allows us to extend a result back to the largest type after we've handled mask logic using vXi1 result types from different source vector widths (e.g. v8i32 and v8i8)

Fixes #157382

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Simon Pilgrim (RKSimon)

Changes

Allows us to extend a result back to the largest type after we've handled mask logic using vXi1 result types from different source vector widths (e.g. v8i32 and v8i8)

Fixes #157411

---

Full diff: https://github.com/llvm/llvm-project/pull/157425.diff

2 Files Affected:

• (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+15-7)
• (modified) llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-sext.ll (+2-5)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index cdc97faf394ca..af624e381be5a 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -50936,10 +50936,12 @@ static SDValue combineAndShuffleNot(SDNode *N, SelectionDAG &DAG,
 // Given a target type \p VT, we generate
 //   or (and x, y), (xor z, zext(build_vector (constants)))
 // given x, y and z are of type \p VT. We can do so, if operands are either
-// truncates from VT types, the second operand is a vector of constants or can
-// be recursively promoted.
+// truncates from VT types, the second operand is a vector of constants, can
+// be recursively promoted or is an existing extension we can extend further.
 static SDValue PromoteMaskArithmetic(SDValue N, const SDLoc &DL, EVT VT,
-                                     SelectionDAG &DAG, unsigned Depth) {
+                                     SelectionDAG &DAG,
+                                     const X86Subtarget &Subtarget,
+                                     unsigned Depth) {
   // Limit recursion to avoid excessive compile times.
   if (Depth >= SelectionDAG::MaxRecursionDepth)
     return SDValue();
@@ -50954,7 +50956,8 @@ static SDValue PromoteMaskArithmetic(SDValue N, const SDLoc &DL, EVT VT,
   if (!TLI.isOperationLegalOrPromote(N.getOpcode(), VT))
     return SDValue();
 
-  if (SDValue NN0 = PromoteMaskArithmetic(N0, DL, VT, DAG, Depth + 1))
+  if (SDValue NN0 =
+          PromoteMaskArithmetic(N0, DL, VT, DAG, Subtarget, Depth + 1))
     N0 = NN0;
   else {
     // The left side has to be a trunc.
@@ -50968,14 +50971,19 @@ static SDValue PromoteMaskArithmetic(SDValue N, const SDLoc &DL, EVT VT,
     N0 = N0.getOperand(0);
   }
 
-  if (SDValue NN1 = PromoteMaskArithmetic(N1, DL, VT, DAG, Depth + 1))
+  if (SDValue NN1 =
+          PromoteMaskArithmetic(N1, DL, VT, DAG, Subtarget, Depth + 1))
     N1 = NN1;
   else {
-    // The right side has to be a 'trunc' or a (foldable) constant.
+    // The right side has to be a 'trunc', a (foldable) constant or an
+    // existing extension we can extend further.
     bool RHSTrunc = N1.getOpcode() == ISD::TRUNCATE &&
                     N1.getOperand(0).getValueType() == VT;
     if (RHSTrunc)
       N1 = N1.getOperand(0);
+    else if (ISD::isExtVecInRegOpcode(N1.getOpcode()) && VT.is256BitVector() &&
+             Subtarget.hasInt256() && N1.hasOneUse())
+      N1 = DAG.getNode(N1.getOpcode(), DL, VT, N1.getOperand(0));
     else if (SDValue Cst =
                  DAG.FoldConstantArithmetic(ISD::ZERO_EXTEND, DL, VT, {N1}))
       N1 = Cst;
@@ -51005,7 +51013,7 @@ static SDValue PromoteMaskArithmetic(SDValue N, const SDLoc &DL,
   EVT NarrowVT = Narrow.getValueType();
 
   // Generate the wide operation.
-  SDValue Op = PromoteMaskArithmetic(Narrow, DL, VT, DAG, 0);
+  SDValue Op = PromoteMaskArithmetic(Narrow, DL, VT, DAG, Subtarget, 0);
   if (!Op)
     return SDValue();
   switch (N.getOpcode()) {
diff --git a/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-sext.ll b/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-sext.ll
index 9e1686e19ce1b..474be4465d9b7 100644
--- a/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-sext.ll
+++ b/llvm/test/CodeGen/X86/bitcast-int-to-vector-bool-sext.ll
@@ -733,11 +733,8 @@ define <8 x i32> @PR157382(ptr %p0, ptr %p1, ptr %p2) {
 ; AVX2-NEXT:    vpcmpeqb %xmm3, %xmm2, %xmm2
 ; AVX2-NEXT:    vpcmpeqd %xmm3, %xmm3, %xmm3
 ; AVX2-NEXT:    vpxor %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpmovsxbw %xmm2, %xmm2
-; AVX2-NEXT:    vextracti128 $1, %ymm1, %xmm3
-; AVX2-NEXT:    vpackssdw %xmm3, %xmm1, %xmm1
-; AVX2-NEXT:    vpor %xmm2, %xmm1, %xmm1
-; AVX2-NEXT:    vpmovsxwd %xmm1, %ymm1
+; AVX2-NEXT:    vpmovsxbd %xmm2, %ymm2
+; AVX2-NEXT:    vpor %ymm2, %ymm1, %ymm1
 ; AVX2-NEXT:    vpand %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    retq
 ;

[phoebewang]

Fixes #157411

Do you mean #157382?

[RKSimon]

Fixes #157411

Do you mean #157382?

Yes, sorry - cut+pasta :(

[phoebewang]

LGTM.