[AVX-512] If avx512dq is available use vpmovm2d/vpmovm2q instead of vselect of zeroes...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@291402 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 4ccdece..b41d9f4 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -17206,22 +17206,26 @@ static SDValue LowerSIGN_EXTEND_AVX512(SDValue Op,
   if (NumElts != 8 && NumElts != 16 && !Subtarget.hasBWI())
     return SDValue();
 
-  if (VT.is512BitVector() && InVT.getVectorElementType() != MVT::i1) {
+  if (VT.is512BitVector() && InVTElt != MVT::i1) {
     if (In.getOpcode() == X86ISD::VSEXT || In.getOpcode() == X86ISD::VZEXT)
       return DAG.getNode(In.getOpcode(), dl, VT, In.getOperand(0));
     return DAG.getNode(X86ISD::VSEXT, dl, VT, In);
   }
 
-  assert (InVT.getVectorElementType() == MVT::i1 && "Unexpected vector type");
+  assert (InVTElt == MVT::i1 && "Unexpected vector type");
   MVT ExtVT = MVT::getVectorVT(MVT::getIntegerVT(512/NumElts), NumElts);
-  SDValue NegOne = DAG.getConstant(
-      APInt::getAllOnesValue(ExtVT.getScalarSizeInBits()), dl, ExtVT);
-  SDValue Zero = DAG.getConstant(
-      APInt::getNullValue(ExtVT.getScalarSizeInBits()), dl, ExtVT);
+  SDValue V;
+  if (Subtarget.hasDQI()) {
+    V = DAG.getNode(X86ISD::VSEXT, dl, ExtVT, In);
+    assert(!VT.is512BitVector() && "Unexpected vector type");
+  } else {
+    SDValue NegOne = getOnesVector(ExtVT, Subtarget, DAG, dl);
+    SDValue Zero = getZeroVector(ExtVT, Subtarget, DAG, dl);
+    V = DAG.getNode(ISD::VSELECT, dl, ExtVT, In, NegOne, Zero);
+    if (VT.is512BitVector())
+      return V;
+  }
 
-  SDValue V = DAG.getNode(ISD::VSELECT, dl, ExtVT, In, NegOne, Zero);
-  if (VT.is512BitVector())
-    return V;
   return DAG.getNode(X86ISD::VTRUNC, dl, VT, V);
 }
 

diff --git a/test/CodeGen/X86/vector-compare-results.ll b/test/CodeGen/X86/vector-compare-results.ll
index 17269e1..42e47f3 100644 (file)
--- a/test/CodeGen/X86/vector-compare-results.ll
+++ b/test/CodeGen/X86/vector-compare-results.ll
@@ -605,13 +605,28 @@ define <8 x i1> @test_cmp_v8f64(<8 x double> %a0, <8 x double> %a1) nounwind {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: test_cmp_v8f64:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vcmpltpd %zmm0, %zmm1, %k1
-; AVX512-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
-; AVX512-NEXT:    vmovdqa64 %zmm0, %zmm0 {%k1} {z}
-; AVX512-NEXT:    vpmovqw %zmm0, %xmm0
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: test_cmp_v8f64:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vcmpltpd %zmm0, %zmm1, %k1
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
+; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512DQ-LABEL: test_cmp_v8f64:
+; AVX512DQ:       # BB#0:
+; AVX512DQ-NEXT:    vcmpltpd %zmm0, %zmm1, %k0
+; AVX512DQ-NEXT:    vpmovm2q %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512DQ-NEXT:    retq
+;
+; AVX512BW-LABEL: test_cmp_v8f64:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vcmpltpd %zmm0, %zmm1, %k1
+; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512BW-NEXT:    retq
   %1 = fcmp ogt <8 x double> %a0, %a1
   ret <8 x i1> %1
 }
@@ -652,13 +667,28 @@ define <16 x i1> @test_cmp_v16f32(<16 x float> %a0, <16 x float> %a1) nounwind {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: test_cmp_v16f32:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vcmpltps %zmm0, %zmm1, %k1
-; AVX512-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
-; AVX512-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
-; AVX512-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: test_cmp_v16f32:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vcmpltps %zmm0, %zmm1, %k1
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
+; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512DQ-LABEL: test_cmp_v16f32:
+; AVX512DQ:       # BB#0:
+; AVX512DQ-NEXT:    vcmpltps %zmm0, %zmm1, %k0
+; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQ-NEXT:    retq
+;
+; AVX512BW-LABEL: test_cmp_v16f32:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vcmpltps %zmm0, %zmm1, %k1
+; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512BW-NEXT:    retq
   %1 = fcmp ogt <16 x float> %a0, %a1
   ret <16 x i1> %1
 }
@@ -750,13 +780,28 @@ define <8 x i1> @test_cmp_v8i64(<8 x i64> %a0, <8 x i64> %a1) nounwind {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: test_cmp_v8i64:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
-; AVX512-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
-; AVX512-NEXT:    vmovdqa64 %zmm0, %zmm0 {%k1} {z}
-; AVX512-NEXT:    vpmovqw %zmm0, %xmm0
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: test_cmp_v8i64:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
+; AVX512F-NEXT:    vmovdqa64 %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512DQ-LABEL: test_cmp_v8i64:
+; AVX512DQ:       # BB#0:
+; AVX512DQ-NEXT:    vpcmpgtq %zmm1, %zmm0, %k0
+; AVX512DQ-NEXT:    vpmovm2q %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512DQ-NEXT:    retq
+;
+; AVX512BW-LABEL: test_cmp_v8i64:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpcmpgtq %zmm1, %zmm0, %k1
+; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa64 %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    vpmovqw %zmm0, %xmm0
+; AVX512BW-NEXT:    retq
   %1 = icmp sgt <8 x i64> %a0, %a1
   ret <8 x i1> %1
 }
@@ -800,13 +845,28 @@ define <16 x i1> @test_cmp_v16i32(<16 x i32> %a0, <16 x i32> %a1) nounwind {
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;
-; AVX512-LABEL: test_cmp_v16i32:
-; AVX512:       # BB#0:
-; AVX512-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
-; AVX512-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
-; AVX512-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
-; AVX512-NEXT:    vpmovdb %zmm0, %xmm0
-; AVX512-NEXT:    retq
+; AVX512F-LABEL: test_cmp_v16i32:
+; AVX512F:       # BB#0:
+; AVX512F-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
+; AVX512F-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
+; AVX512F-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
+; AVX512F-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512F-NEXT:    retq
+;
+; AVX512DQ-LABEL: test_cmp_v16i32:
+; AVX512DQ:       # BB#0:
+; AVX512DQ-NEXT:    vpcmpgtd %zmm1, %zmm0, %k0
+; AVX512DQ-NEXT:    vpmovm2d %k0, %zmm0
+; AVX512DQ-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512DQ-NEXT:    retq
+;
+; AVX512BW-LABEL: test_cmp_v16i32:
+; AVX512BW:       # BB#0:
+; AVX512BW-NEXT:    vpcmpgtd %zmm1, %zmm0, %k1
+; AVX512BW-NEXT:    vpternlogd $255, %zmm0, %zmm0, %zmm0
+; AVX512BW-NEXT:    vmovdqa32 %zmm0, %zmm0 {%k1} {z}
+; AVX512BW-NEXT:    vpmovdb %zmm0, %xmm0
+; AVX512BW-NEXT:    retq
   %1 = icmp sgt <16 x i32> %a0, %a1
   ret <16 x i1> %1
 }