[X86] Only fold AND/ANDNP back to VSELECT if we know the predicated mask select is legal (#156663)

By only checking type legality we didn't account for 128/256-bit ops
being run on non-AVX512VL targets, or vXi8/i16 ops being run on
non-AVX512BW targets

This check is cropping up in several places now and I intend to hoist it
out into a common helper, but this initial fix needs to be as clean as
possible to be back ported to 21.X

Fixes #156256

(cherry picked from commit 86879d4)

Author: RKSimon

llvm/lib/Target/X86/X86ISelLowering.cpp

@@ -51775,6 +51775,8 @@ static SDValue combineAnd(SDNode *N, SelectionDAG &DAG,
     SDValue X, Y;
     EVT CondVT = VT.changeVectorElementType(MVT::i1);
     if (TLI.isTypeLegal(VT) && TLI.isTypeLegal(CondVT) &&
+        (VT.is512BitVector() || Subtarget.hasVLX()) &&
+        (VT.getScalarSizeInBits() >= 32 || Subtarget.hasBWI()) &&
         sd_match(N, m_And(m_Value(X),
                           m_OneUse(m_SExt(m_AllOf(
                               m_Value(Y), m_SpecificVT(CondVT),
@@ -55329,6 +55331,8 @@ static SDValue combineAndnp(SDNode *N, SelectionDAG &DAG,
     SDValue Src = N0.getOperand(0);
     EVT SrcVT = Src.getValueType();
     if (Src.getOpcode() == ISD::SETCC && SrcVT.getScalarType() == MVT::i1 &&
+        (VT.is512BitVector() || Subtarget.hasVLX()) &&
+        (VT.getScalarSizeInBits() >= 32 || Subtarget.hasBWI()) &&
         TLI.isTypeLegal(SrcVT) && N0.hasOneUse() && Src.hasOneUse())
       return DAG.getSelect(DL, VT, DAG.getNOT(DL, Src, SrcVT), N1,
                            getZeroVector(VT, Subtarget, DAG, DL));

llvm/test/CodeGen/X86/pr156256.ll

@@ -0,0 +1,25 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefix=AVX512
+; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefix=AVX512VL
+
+define <16 x i16> @PR156256(<16 x i32> %a, <16 x i32> %b) {
+; AVX512-LABEL: PR156256:
+; AVX512:       # %bb.0:
+; AVX512-NEXT:    vpcmpnleud %zmm1, %zmm0, %k0
+; AVX512-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512-NEXT:    vpand {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX512-NEXT:    retq
+;
+; AVX512VL-LABEL: PR156256:
+; AVX512VL:       # %bb.0:
+; AVX512VL-NEXT:    vpcmpnleud %zmm1, %zmm0, %k0
+; AVX512VL-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512VL-NEXT:    vpmovdw %zmm0, %ymm0
+; AVX512VL-NEXT:    vpandd {{\.?LCPI[0-9]+_[0-9]+}}(%rip){1to8}, %ymm0, %ymm0
+; AVX512VL-NEXT:    retq
+  %icmp = icmp ugt <16 x i32> %a, %b
+  %sext = sext <16 x i1> %icmp to <16 x i16>
+  %and = and <16 x i16> %sext, splat (i16 16256)
+  ret <16 x i16> %and
+}