From 025a3d857a3649dff84f4c3cd02e8e6132b24613 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Mon, 1 Feb 2016 22:05:50 +0000
Subject: [PATCH] [X86][AVX512] Add support for AVX512 VINSERTPS shuffle
 decoding

llvm-svn: 259420
---
 .../lib/Target/X86/InstPrinter/X86InstComments.cpp |  2 +
 .../CodeGen/X86/merge-consecutive-loads-128.ll     | 44 +++++-----------------
 .../CodeGen/X86/merge-consecutive-loads-256.ll     |  2 +-
 3 files changed, 13 insertions(+), 35 deletions(-)

diff --git a/llvm/lib/Target/X86/InstPrinter/X86InstComments.cpp b/llvm/lib/Target/X86/InstPrinter/X86InstComments.cpp
index 73f654c..9dd9144 100644
--- a/llvm/lib/Target/X86/InstPrinter/X86InstComments.cpp
+++ b/llvm/lib/Target/X86/InstPrinter/X86InstComments.cpp
@@ -278,10 +278,12 @@ bool llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
 
   case X86::INSERTPSrr:
   case X86::VINSERTPSrr:
+  case X86::VINSERTPSzrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     // FALL THROUGH.
   case X86::INSERTPSrm:
   case X86::VINSERTPSrm:
+  case X86::VINSERTPSzrm:
     DestName = getRegName(MI->getOperand(0).getReg());
     Src1Name = getRegName(MI->getOperand(1).getReg());
     if (MI->getOperand(MI->getNumOperands() - 1).isImm())
diff --git a/llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll b/llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll
index d2d5bcb..81aba5e 100644
--- a/llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll
+++ b/llvm/test/CodeGen/X86/merge-consecutive-loads-128.ll
@@ -188,23 +188,11 @@ define <4 x float> @merge_4f32_f32_012u(float* %ptr) nounwind uwtable noinline s
 ; SSE41-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
 ; SSE41-NEXT:    retq
 ;
-; AVX1-LABEL: merge_4f32_f32_012u:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX1-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: merge_4f32_f32_012u:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX2-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
-; AVX2-NEXT:    retq
-;
-; AVX512F-LABEL: merge_4f32_f32_012u:
-; AVX512F:       # BB#0:
-; AVX512F-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX512F-NEXT:    vinsertps $32, 8(%rdi), %xmm0, %xmm0
-; AVX512F-NEXT:    retq
+; AVX-LABEL: merge_4f32_f32_012u:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX-NEXT:    retq
   %ptr0 = getelementptr inbounds float, float* %ptr, i64 0
   %ptr1 = getelementptr inbounds float, float* %ptr, i64 1
   %ptr2 = getelementptr inbounds float, float* %ptr, i64 2
@@ -234,23 +222,11 @@ define <4 x float> @merge_4f32_f32_019u(float* %ptr) nounwind uwtable noinline s
 ; SSE41-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
 ; SSE41-NEXT:    retq
 ;
-; AVX1-LABEL: merge_4f32_f32_019u:
-; AVX1:       # BB#0:
-; AVX1-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX1-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
-; AVX1-NEXT:    retq
-;
-; AVX2-LABEL: merge_4f32_f32_019u:
-; AVX2:       # BB#0:
-; AVX2-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX2-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
-; AVX2-NEXT:    retq
-;
-; AVX512F-LABEL: merge_4f32_f32_019u:
-; AVX512F:       # BB#0:
-; AVX512F-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX512F-NEXT:    vinsertps $32, 36(%rdi), %xmm0, %xmm0
-; AVX512F-NEXT:    retq
+; AVX-LABEL: merge_4f32_f32_019u:
+; AVX:       # BB#0:
+; AVX-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
+; AVX-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],mem[0],xmm0[3]
+; AVX-NEXT:    retq
   %ptr0 = getelementptr inbounds float, float* %ptr, i64 0
   %ptr1 = getelementptr inbounds float, float* %ptr, i64 1
   %ptr2 = getelementptr inbounds float, float* %ptr, i64 9
diff --git a/llvm/test/CodeGen/X86/merge-consecutive-loads-256.ll b/llvm/test/CodeGen/X86/merge-consecutive-loads-256.ll
index 4f3427a..efcc323 100644
--- a/llvm/test/CodeGen/X86/merge-consecutive-loads-256.ll
+++ b/llvm/test/CodeGen/X86/merge-consecutive-loads-256.ll
@@ -319,7 +319,7 @@ define <8 x float> @merge_8f32_f32_1u3u5zu8(float* %ptr) nounwind uwtable noinli
 ; AVX512F-NEXT:    vmovss 20(%rdi), %xmm1
 ; AVX512F-NEXT:    vshufps {{.*#+}} xmm0 = xmm1[0,1],xmm0[1,0]
 ; AVX512F-NEXT:    vmovss 4(%rdi), %xmm1
-; AVX512F-NEXT:    vinsertps $32, 12(%rdi), %xmm1, %xmm1
+; AVX512F-NEXT:    vinsertps {{.*#+}} xmm1 = xmm1[0,1],mem[0],xmm1[3]
 ; AVX512F-NEXT:    vinserti128 $1, %xmm0, %ymm1, %ymm0
 ; AVX512F-NEXT:    retq
   %ptr0 = getelementptr inbounds float, float* %ptr, i64 1
-- 
2.7.4