replica
/
hanchenye-llvm-project
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[PATCH] Fix VGATHER* operand constraints 

Add earlyclobber constaints to prevent input register being allocated as
the output register because, according to Intel spec [1], "If any pair
of the index, mask, or destination registers are the same, this
instruction results a UD fault."

---
[1] http://software.intel.com/sites/default/files/319433-014.pdf

llvm-svn: 183327
This commit is contained in:
Michael Liao 2013-06-05 18:12:26 +00:00
parent 4e75df1560
commit 00b20cc924
3 changed files with 23 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
llvm/lib/Target/X86/X86ISelDAGToDAG.cpp
View File

@ -2041,6 +2041,8 @@ SDNode *X86DAGToDAGISel::Select(SDNode *Node) {
case Intrinsic::x86_avx2_gather_d_d_256:
case Intrinsic::x86_avx2_gather_q_d:
case Intrinsic::x86_avx2_gather_q_d_256: {
if (!Subtarget->hasAVX2())
break;
unsigned Opc;
switch (IntNo) {
default: llvm_unreachable("Impossible intrinsic");

4
llvm/lib/Target/X86/X86InstrSSE.td
View File

@ -8372,7 +8372,9 @@ multiclass avx2_gather<bits<8> opc, string OpcodeStr, RegisterClass RC256,
[]>, VEX_4VOp3, VEX_L;
}
let mayLoad = 1, Constraints = "$src1 = $dst, $mask = $mask_wb" in {
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>;

18
llvm/test/CodeGen/X86/avx2-gather.ll Normal file
View File

@ -0,0 +1,18 @@
; RUN: not llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=core-avx2 | FileCheck %s
declare <4 x float> @llvm.x86.avx2.gather.d.ps(<4 x float>, i8*,
<4 x i32>, <4 x float>, i8) nounwind readonly
define <4 x float> @test_x86_avx2_gather_d_ps(i8* %a1,
<4 x i32> %idx, <4 x float> %mask) {
%res = call <4 x float> @llvm.x86.avx2.gather.d.ps(<4 x float> undef,
i8* %a1, <4 x i32> %idx, <4 x float> %mask, i8 2) ;
ret <4 x float> %res
}
; CHECK: test_x86_avx2_gather_d_ps
; CHECK: vgatherdps
; CHECK-NOT: [[DST]]
; CHECK: [[DST:%xmm[0-9]+]]{{$}}
; CHECK: ret