From: Craig Topper <craig.topper@gmail.com>
Date: Tue, 19 Jul 2016 02:00:38 +0000 (+0000)
Subject: [AVX512] Give priority to EVEX encoded PSHUFB over the VEX versions.
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=d6ca1dc45e0c037f6d01cf21261f3a6fde6cdd6a;p=platform%2Fupstream%2Fllvm.git

[AVX512] Give priority to EVEX encoded PSHUFB over the VEX versions.

llvm-svn: 275942
---

diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index 3686bbe..3972fa9 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -5552,6 +5552,14 @@ multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
        Sched<[Sched.Folded, ReadAfterLd]>;
 }
 
+let ImmT = NoImm, Predicates = [HasAVX, NoVLX_Or_NoBWI] in {
+let isCommutable = 0 in {
+  defm VPSHUFB    : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
+                                  loadv2i64, i128mem,
+                                  SSE_PSHUFB, 0>, VEX_4V;
+}
+}
+
 let ImmT = NoImm, Predicates = [HasAVX] in {
 let isCommutable = 0 in {
   defm VPHADDW    : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v8i16, VR128,
@@ -5575,9 +5583,6 @@ let isCommutable = 0 in {
   defm VPSIGND    : SS3I_binop_rm_int<0x0A, "vpsignd",
                                       int_x86_ssse3_psign_d_128,
                                       SSE_PSIGN, loadv2i64, 0>, VEX_4V;
-  defm VPSHUFB    : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v16i8, VR128,
-                                  loadv2i64, i128mem,
-                                  SSE_PSHUFB, 0>, VEX_4V;
   defm VPHADDSW   : SS3I_binop_rm_int<0x03, "vphaddsw",
                                       int_x86_ssse3_phadd_sw_128,
                                       SSE_PHADDSUBSW, loadv2i64, 0>, VEX_4V;
@@ -5593,6 +5598,14 @@ defm VPMULHRSW    : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
                                       SSE_PMULHRSW, loadv2i64, 0>, VEX_4V;
 }
 
+let ImmT = NoImm, Predicates = [HasAVX2, NoVLX_Or_NoBWI] in {
+let isCommutable = 0 in {
+  defm VPSHUFBY   : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
+                                  loadv4i64, i256mem,
+                                  SSE_PSHUFB, 0>, VEX_4V, VEX_L;
+}
+}
+
 let ImmT = NoImm, Predicates = [HasAVX2] in {
 let isCommutable = 0 in {
   defm VPHADDWY   : SS3I_binop_rm<0x01, "vphaddw", X86hadd, v16i16, VR256,
@@ -5613,9 +5626,6 @@ let isCommutable = 0 in {
                                         WriteVecALU>, VEX_4V, VEX_L;
   defm VPSIGNDY   : SS3I_binop_rm_int_y<0x0A, "vpsignd", int_x86_avx2_psign_d,
                                         WriteVecALU>, VEX_4V, VEX_L;
-  defm VPSHUFBY   : SS3I_binop_rm<0x00, "vpshufb", X86pshufb, v32i8, VR256,
-                                  loadv4i64, i256mem,
-                                  SSE_PSHUFB, 0>, VEX_4V, VEX_L;
   defm VPHADDSW   : SS3I_binop_rm_int_y<0x03, "vphaddsw",
                                         int_x86_avx2_phadd_sw,
                                         WriteVecALU>, VEX_4V, VEX_L;
diff --git a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
index 534d5c8..cf1b15e 100644
--- a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
@@ -4727,7 +4727,7 @@ define <16 x i8>@test_int_x86_avx512_mask_pshuf_b_128(<16 x i8> %x0, <16 x i8> %
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovw %edi, %k1 ## encoding: [0xc5,0xf8,0x92,0xcf]
 ; CHECK-NEXT:    vpshufb %xmm1, %xmm0, %xmm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x09,0x00,0xd1]
-; CHECK-NEXT:    vpshufb %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe2,0x79,0x00,0xc1]
+; CHECK-NEXT:    vpshufb %xmm1, %xmm0, %xmm0 ## encoding: [0x62,0xf2,0x7d,0x08,0x00,0xc1]
 ; CHECK-NEXT:    vpaddb %xmm0, %xmm2, %xmm0 ## encoding: [0x62,0xf1,0x6d,0x08,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <16 x i8> @llvm.x86.avx512.mask.pshuf.b.128(<16 x i8> %x0, <16 x i8> %x1, <16 x i8> %x2, i16 %x3)
@@ -4743,7 +4743,7 @@ define <32 x i8>@test_int_x86_avx512_mask_pshuf_b_256(<32 x i8> %x0, <32 x i8> %
 ; CHECK:       ## BB#0:
 ; CHECK-NEXT:    kmovd %edi, %k1 ## encoding: [0xc5,0xfb,0x92,0xcf]
 ; CHECK-NEXT:    vpshufb %ymm1, %ymm0, %ymm2 {%k1} ## encoding: [0x62,0xf2,0x7d,0x29,0x00,0xd1]
-; CHECK-NEXT:    vpshufb %ymm1, %ymm0, %ymm0 ## encoding: [0xc4,0xe2,0x7d,0x00,0xc1]
+; CHECK-NEXT:    vpshufb %ymm1, %ymm0, %ymm0 ## encoding: [0x62,0xf2,0x7d,0x28,0x00,0xc1]
 ; CHECK-NEXT:    vpaddb %ymm0, %ymm2, %ymm0 ## encoding: [0x62,0xf1,0x6d,0x28,0xfc,0xc0]
 ; CHECK-NEXT:    retq ## encoding: [0xc3]
   %res = call <32 x i8> @llvm.x86.avx512.mask.pshuf.b.256(<32 x i8> %x0, <32 x i8> %x1, <32 x i8> %x2, i32 %x3)