From: Craig Topper <craig.topper@intel.com>
Date: Sat, 5 Jan 2019 18:48:11 +0000 (+0000)
Subject: [X86] Allow LowerTRUNCATE to use PACKUS/PACKSS for v16i16->v16i8 truncate when -mpref... 
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=0716b2a298b4c6894e75ba2dc1f625857b47a48b;p=llvm

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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@350473 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 25a93e94990..4056b4982b0 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -17949,9 +17949,10 @@ static SDValue truncateVectorWithPACK(unsigned Opcode, EVT DstVT, SDValue In,
                                       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(SDValue In, EVT VT, const SDLoc &DL,
       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)) {
diff --git a/test/CodeGen/X86/prefer-avx256-mask-extend.ll b/test/CodeGen/X86/prefer-avx256-mask-extend.ll
index b4f8e5b0b6c..b4d452f2d3e 100644
--- a/test/CodeGen/X86/prefer-avx256-mask-extend.ll
+++ b/test/CodeGen/X86/prefer-avx256-mask-extend.ll
@@ -48,11 +48,9 @@ define <16 x i8> @testv16i1_sext_v16i8(<8 x i32>* %p, <8 x i32>* %q) {
 ; 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
 ;
diff --git a/test/CodeGen/X86/prefer-avx256-mask-shuffle.ll b/test/CodeGen/X86/prefer-avx256-mask-shuffle.ll
index 92f6e271126..7f4480ceb63 100644
--- a/test/CodeGen/X86/prefer-avx256-mask-shuffle.ll
+++ b/test/CodeGen/X86/prefer-avx256-mask-shuffle.ll
@@ -34,11 +34,9 @@ define <16 x i1> @shuf16i1_3_6_22_12_3_7_7_0_3_6_1_13_3_21_7_0(<8 x i32>* %a, <8
 ; 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_7_0_3_6_1_13_3_21_7_0_3_6_22_12_3_7_7_0
 ; 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
 ;