[X86][AVX512] createVariablePermute - Non-VLX targets can widen v4i64/v8f64 variable...

Permutes in the upper elements will be undefined, but they will be discarded anyway.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327238 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index eafa118..55c28c3 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -8048,9 +8048,19 @@ SDValue createVariablePermute(MVT VT, SDValue SrcVec, SDValue IndicesVec,
     break;
   case MVT::v4i64:
   case MVT::v4f64:
-    if (Subtarget.hasVLX())
+    if (Subtarget.hasAVX512()) {
+      if (!Subtarget.hasVLX()) {
+        MVT WidenSrcVT = MVT::getVectorVT(VT.getScalarType(), 8);
+        SrcVec = widenSubVector(WidenSrcVT, SrcVec, false, Subtarget, DAG,
+                                SDLoc(SrcVec));
+        IndicesVec = widenSubVector(MVT::v8i64, IndicesVec, false, Subtarget,
+                                    DAG, SDLoc(IndicesVec));
+        SDValue Res = createVariablePermute(WidenSrcVT, SrcVec, IndicesVec, DL,
+                                            DAG, Subtarget);
+        return extract256BitVector(Res, 0, DAG, DL);
+      }
       Opcode = X86ISD::VPERMV;
-    else if (Subtarget.hasXOP()) {
+    } else if (Subtarget.hasXOP()) {
       SrcVec = DAG.getBitcast(MVT::v4f64, SrcVec);
       SDValue LoLo =
           DAG.getVectorShuffle(MVT::v4f64, DL, SrcVec, SrcVec, {0, 1, 0, 1});

diff --git a/test/CodeGen/X86/var-permute-256.ll b/test/CodeGen/X86/var-permute-256.ll
index 8538ed7..87cd091 100644 (file)
--- a/test/CodeGen/X86/var-permute-256.ll
+++ b/test/CodeGen/X86/var-permute-256.ll
@@ -66,28 +66,18 @@ define <4 x i64> @var_shuffle_v4i64(<4 x i64> %v, <4 x i64> %indices) nounwind {
 ;
 ; AVX512F-LABEL: var_shuffle_v4i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2,2,2,2]
-; AVX512F-NEXT:    vpmuludq %ymm2, %ymm1, %ymm2
-; AVX512F-NEXT:    vpsrlq $32, %ymm1, %ymm3
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [8589934594,8589934594,8589934594,8589934594]
-; AVX512F-NEXT:    vpmuludq %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpaddq %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllq $32, %ymm2, %ymm2
-; AVX512F-NEXT:    vpmuludq %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [4294967296,4294967296,4294967296,4294967296]
-; AVX512F-NEXT:    vpaddq %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpaddq %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: var_shuffle_v4i64:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512DQ-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [8589934594,8589934594,8589934594,8589934594]
-; AVX512DQ-NEXT:    vpmullq %zmm2, %zmm1, %zmm1
-; AVX512DQ-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4294967296,4294967296,4294967296,4294967296]
-; AVX512DQ-NEXT:    vpaddq %ymm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512DQ-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512DQ-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; AVX512DQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_shuffle_v4i64:
@@ -1377,28 +1367,18 @@ define <4 x double> @var_shuffle_v4f64(<4 x double> %v, <4 x i64> %indices) noun
 ;
 ; AVX512F-LABEL: var_shuffle_v4f64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2,2,2,2]
-; AVX512F-NEXT:    vpmuludq %ymm2, %ymm1, %ymm2
-; AVX512F-NEXT:    vpsrlq $32, %ymm1, %ymm3
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [8589934594,8589934594,8589934594,8589934594]
-; AVX512F-NEXT:    vpmuludq %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpaddq %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllq $32, %ymm2, %ymm2
-; AVX512F-NEXT:    vpmuludq %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [4294967296,4294967296,4294967296,4294967296]
-; AVX512F-NEXT:    vpaddq %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpaddq %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512F-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: var_shuffle_v4f64:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512DQ-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [8589934594,8589934594,8589934594,8589934594]
-; AVX512DQ-NEXT:    vpmullq %zmm2, %zmm1, %zmm1
-; AVX512DQ-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4294967296,4294967296,4294967296,4294967296]
-; AVX512DQ-NEXT:    vpaddq %ymm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512DQ-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
+; AVX512DQ-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; AVX512DQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_shuffle_v4f64:
@@ -1558,30 +1538,18 @@ define <4 x i64> @var_shuffle_v4i64_from_v2i64(<2 x i64> %v, <4 x i64> %indices)
 ;
 ; AVX512F-LABEL: var_shuffle_v4i64_from_v2i64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2,2,2,2]
-; AVX512F-NEXT:    vpmuludq %ymm2, %ymm1, %ymm2
-; AVX512F-NEXT:    vpsrlq $32, %ymm1, %ymm3
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [8589934594,8589934594,8589934594,8589934594]
-; AVX512F-NEXT:    vpmuludq %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpaddq %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllq $32, %ymm2, %ymm2
-; AVX512F-NEXT:    vpmuludq %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [4294967296,4294967296,4294967296,4294967296]
-; AVX512F-NEXT:    vpaddq %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpaddq %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: var_shuffle_v4i64_from_v2i64:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX512DQ-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [8589934594,8589934594,8589934594,8589934594]
-; AVX512DQ-NEXT:    vpmullq %zmm2, %zmm1, %zmm1
-; AVX512DQ-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4294967296,4294967296,4294967296,4294967296]
-; AVX512DQ-NEXT:    vpaddq %ymm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512DQ-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; AVX512DQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_shuffle_v4i64_from_v2i64:
@@ -2806,30 +2774,18 @@ define <4 x double> @var_shuffle_v4f64_from_v2f64(<2 x double> %v, <4 x i64> %in
 ;
 ; AVX512F-LABEL: var_shuffle_v4f64_from_v2f64:
 ; AVX512F:       # %bb.0:
-; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2,2,2,2]
-; AVX512F-NEXT:    vpmuludq %ymm2, %ymm1, %ymm2
-; AVX512F-NEXT:    vpsrlq $32, %ymm1, %ymm3
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm4 = [8589934594,8589934594,8589934594,8589934594]
-; AVX512F-NEXT:    vpmuludq %ymm4, %ymm3, %ymm3
-; AVX512F-NEXT:    vpaddq %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpsllq $32, %ymm2, %ymm2
-; AVX512F-NEXT:    vpmuludq %ymm4, %ymm1, %ymm1
-; AVX512F-NEXT:    vpbroadcastq {{.*#+}} ymm3 = [4294967296,4294967296,4294967296,4294967296]
-; AVX512F-NEXT:    vpaddq %ymm3, %ymm2, %ymm2
-; AVX512F-NEXT:    vpaddq %ymm2, %ymm1, %ymm1
-; AVX512F-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512F-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
+; AVX512F-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512F-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; AVX512F-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512F-NEXT:    retq
 ;
 ; AVX512DQ-LABEL: var_shuffle_v4f64_from_v2f64:
 ; AVX512DQ:       # %bb.0:
 ; AVX512DQ-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
-; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX512DQ-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [8589934594,8589934594,8589934594,8589934594]
-; AVX512DQ-NEXT:    vpmullq %zmm2, %zmm1, %zmm1
-; AVX512DQ-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [4294967296,4294967296,4294967296,4294967296]
-; AVX512DQ-NEXT:    vpaddq %ymm2, %ymm1, %ymm1
-; AVX512DQ-NEXT:    vpermd %ymm0, %ymm1, %ymm0
+; AVX512DQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
+; AVX512DQ-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
+; AVX512DQ-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
 ; AVX512DQ-NEXT:    retq
 ;
 ; AVX512VL-LABEL: var_shuffle_v4f64_from_v2f64: