From 3ec862b7aed2e529bdd8ef8c060834136e598f06 Mon Sep 17 00:00:00 2001
From: Cameron McInally <cameron.mcinally@nyu.edu>
Date: Tue, 25 Mar 2014 12:36:38 +0000
Subject: [PATCH] Fix AVX2 Gather execution domains.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204713 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86InstrSSE.td           | 14 ++++++++++----
 test/CodeGen/X86/avx2-gather.ll         | 16 ++++++++++++++++
 test/CodeGen/X86/avx2-intrinsics-x86.ll |  2 +-
 3 files changed, 27 insertions(+), 5 deletions(-)

diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index f7aac458be8..f2f39679447 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -8791,12 +8791,18 @@ multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
 let mayLoad = 1, Constraints
   = "@earlyclobber $dst,@earlyclobber $mask_wb, $src1 = $dst, $mask = $mask_wb"
   in {
-  defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
-  defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
-  defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
-  defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
   defm VPGATHERDQ : avx2_gather<0x90, "vpgatherdq", VR256, vx64mem, vx64mem>, VEX_W;
   defm VPGATHERQQ : avx2_gather<0x91, "vpgatherqq", VR256, vx64mem, vy64mem>, VEX_W;
   defm VPGATHERDD : avx2_gather<0x90, "vpgatherdd", VR256, vx32mem, vy32mem>;
   defm VPGATHERQD : avx2_gather<0x91, "vpgatherqd", VR128, vx32mem, vy32mem>;
+
+  let ExeDomain = SSEPackedDouble in {
+    defm VGATHERDPD : avx2_gather<0x92, "vgatherdpd", VR256, vx64mem, vx64mem>, VEX_W;
+    defm VGATHERQPD : avx2_gather<0x93, "vgatherqpd", VR256, vx64mem, vy64mem>, VEX_W;
+  }
+
+  let ExeDomain = SSEPackedSingle in {
+    defm VGATHERDPS : avx2_gather<0x92, "vgatherdps", VR256, vx32mem, vy32mem>;
+    defm VGATHERQPS : avx2_gather<0x93, "vgatherqps", VR128, vx32mem, vy32mem>;
+  }
 }
diff --git a/test/CodeGen/X86/avx2-gather.ll b/test/CodeGen/X86/avx2-gather.ll
index ee50c457fe8..a9ac0258975 100644
--- a/test/CodeGen/X86/avx2-gather.ll
+++ b/test/CodeGen/X86/avx2-gather.ll
@@ -15,4 +15,20 @@ define <4 x float> @test_x86_avx2_gather_d_ps(i8* %a1,
 ; CHECK: vgatherdps
 ; CHECK-NOT: [[DST]]
 ; CHECK: [[DST:%xmm[0-9]+]]{{$}}
+; CHECK: vmovaps
+; CHECK: ret
+
+declare <2 x double> @llvm.x86.avx2.gather.d.pd(<2 x double>, i8*,
+                      <4 x i32>, <2 x double>, i8) nounwind readonly
+
+define <2 x double> @test_x86_avx2_gather_d_pd(i8* %a1,
+                     <4 x i32> %idx, <2 x double> %mask) {
+  %res = call <2 x double> @llvm.x86.avx2.gather.d.pd(<2 x double> undef,
+                            i8* %a1, <4 x i32> %idx, <2 x double> %mask, i8 2) ;
+  ret <2 x double> %res
+}
+
+; CHECK: test_x86_avx2_gather_d_pd
+; CHECK: vgatherdpd
+; CHECK: vmovapd
 ; CHECK: ret
diff --git a/test/CodeGen/X86/avx2-intrinsics-x86.ll b/test/CodeGen/X86/avx2-intrinsics-x86.ll
index a6141b09561..7ee944b7c92 100644
--- a/test/CodeGen/X86/avx2-intrinsics-x86.ll
+++ b/test/CodeGen/X86/avx2-intrinsics-x86.ll
@@ -1142,7 +1142,7 @@ define <8 x float>  @test_gather_mask(<8 x float> %a0, float* %a,
                                       <8 x i32> %idx, <8 x float> %mask,
                                       float* nocapture %out) {
 ; CHECK: test_gather_mask
-; CHECK: vmovdqa %ymm2, [[DEST:%.*]]
+; CHECK: vmovaps %ymm2, [[DEST:%.*]]
 ; CHECK: vgatherdps [[DEST]]
 ;; gather with mask
   %a_i8 = bitcast float* %a to i8*
-- 
2.34.1