[llvm] r350473 - [X86] Allow LowerTRUNCATE to use PACKUS/PACKSS for v16i16->v16i8 truncate when -mprefer-vector-width-256 is in effect and BWI is not available.

[Craig Topper via llvm-commits]

Author: ctopper
Date: Sat Jan  5 10:48:11 2019
New Revision: 350473

URL: http://llvm.org/viewvc/llvm-project?rev=350473&view=rev
Log:
[X86] Allow LowerTRUNCATE to use PACKUS/PACKSS for v16i16->v16i8 truncate when -mprefer-vector-width-256 is in effect and BWI is not available.

Modified:
    llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
    llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-extend.ll
    llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-shuffle.ll

Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=350473&r1=350472&r2=350473&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Sat Jan  5 10:48:11 2019
@@ -17949,9 +17949,10 @@ static SDValue truncateVectorWithPACK(un
                                       const X86Subtarget &Subtarget) {
   assert((Opcode == X86ISD::PACKSS || Opcode == X86ISD::PACKUS) &&
          "Unexpected PACK opcode");
+  assert(DstVT.isVector() && "VT not a vector?");
 
   // Requires SSE2 but AVX512 has fast vector truncate.
-  if (!Subtarget.hasSSE2() || Subtarget.hasAVX512() || !DstVT.isVector())
+  if (!Subtarget.hasSSE2())
     return SDValue();
 
   EVT SrcVT = In.getValueType();
@@ -36899,6 +36900,7 @@ static SDValue combineTruncateWithSat(SD
       return DAG.getNode(X86ISD::VTRUNCUS, DL, VT, USatVal);
   }
   if (VT.isVector() && isPowerOf2_32(VT.getVectorNumElements()) &&
+      !Subtarget.hasAVX512() &&
       (SVT == MVT::i8 || SVT == MVT::i16) &&
       (InSVT == MVT::i16 || InSVT == MVT::i32)) {
     if (auto USatVal = detectSSatPattern(In, VT, true)) {

Modified: llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-extend.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-extend.ll?rev=350473&r1=350472&r2=350473&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-extend.ll (original)
+++ llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-extend.ll Sat Jan  5 10:48:11 2019
@@ -48,11 +48,9 @@ define <16 x i8> @testv16i1_sext_v16i8(<
 ; AVX256-NEXT:    vpcmpeqd %ymm0, %ymm0, %ymm0
 ; AVX256-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k2} {z}
 ; AVX256-NEXT:    vpmovdw %ymm1, %xmm1
-; AVX256-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX256-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
 ; AVX256-NEXT:    vpmovdw %ymm0, %xmm0
-; AVX256-NEXT:    vpsrlw $8, %xmm0, %xmm0
-; AVX256-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX256-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX256-NEXT:    vzeroupper
 ; AVX256-NEXT:    retq
 ;

Modified: llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-shuffle.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-shuffle.ll?rev=350473&r1=350472&r2=350473&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-shuffle.ll (original)
+++ llvm/trunk/test/CodeGen/X86/prefer-avx256-mask-shuffle.ll Sat Jan  5 10:48:11 2019
@@ -34,11 +34,9 @@ define <16 x i1> @shuf16i1_3_6_22_12_3_7
 ; AVX256VL-NEXT:    kshiftrw $8, %k0, %k2
 ; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k2} {z}
 ; AVX256VL-NEXT:    vpmovdw %ymm1, %xmm1
-; AVX256VL-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
 ; AVX256VL-NEXT:    vpmovdw %ymm0, %xmm0
-; AVX256VL-NEXT:    vpsrlw $8, %xmm0, %xmm0
-; AVX256VL-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX256VL-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX256VL-NEXT:    vzeroupper
 ; AVX256VL-NEXT:    retq
 ;
@@ -169,11 +167,9 @@ define <32 x i1> @shuf32i1_3_6_22_12_3_7
 ; AVX256VL-NEXT:    kshiftrw $8, %k0, %k2
 ; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm1 {%k2} {z}
 ; AVX256VL-NEXT:    vpmovdw %ymm1, %xmm1
-; AVX256VL-NEXT:    vpsrlw $8, %xmm1, %xmm1
 ; AVX256VL-NEXT:    vmovdqa32 %ymm0, %ymm0 {%k1} {z}
 ; AVX256VL-NEXT:    vpmovdw %ymm0, %xmm0
-; AVX256VL-NEXT:    vpsrlw $8, %xmm0, %xmm0
-; AVX256VL-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
+; AVX256VL-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
 ; AVX256VL-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
 ; AVX256VL-NEXT:    retq
 ;