maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[X86] AVX512: Allow writemask argument in vpermt* intrinsics 

llvm-svn: 212223
This commit is contained in:
Adam Nemet 2014-07-02 21:26:01 +00:00
parent efe9c98a16
commit 11dd5cf9f1
2 changed files with 22 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
llvm/lib/Target/X86/X86InstrAVX512.td
View File

@ -693,22 +693,32 @@ defm VPERMI2PD : avx512_perm_3src<0x77, "vpermi2pd", VR512, memopv8f64,
multiclass avx512_perm_table_3src<bits<8> opc, string Suffix, RegisterClass RC, multiclass avx512_perm_table_3src<bits<8> opc, string Suffix, RegisterClass RC,
PatFrag mem_frag, X86MemOperand x86memop, PatFrag mem_frag, X86MemOperand x86memop,
SDNode OpNode, ValueType OpVT, RegisterClass KRC> : SDNode OpNode, ValueType OpVT, RegisterClass KRC,
ValueType MaskVT, RegisterClass MRC> :
avx512_perm_3src<opc, "vpermt2"##Suffix, RC, mem_frag, x86memop, OpNode, avx512_perm_3src<opc, "vpermt2"##Suffix, RC, mem_frag, x86memop, OpNode,
OpVT, KRC> { OpVT, KRC> {
def : Pat<(OpVT (!cast<Intrinsic>("int_x86_avx512_mask_vpermt_"##Suffix##"_512") def : Pat<(OpVT (!cast<Intrinsic>("int_x86_avx512_mask_vpermt_"##Suffix##"_512")
VR512:$idx, VR512:$src1, VR512:$src2, -1)), VR512:$idx, VR512:$src1, VR512:$src2, -1)),
(!cast<Instruction>(NAME#rr) VR512:$src1, VR512:$idx, VR512:$src2)>; (!cast<Instruction>(NAME#rr) VR512:$src1, VR512:$idx, VR512:$src2)>;
def : Pat<(OpVT (!cast<Intrinsic>("int_x86_avx512_mask_vpermt_"##Suffix##"_512")
VR512:$idx, VR512:$src1, VR512:$src2, MRC:$mask)),
(!cast<Instruction>(NAME#rrk) VR512:$src1,
(MaskVT (COPY_TO_REGCLASS MRC:$mask, KRC)), VR512:$idx, VR512:$src2)>;
} }
defm VPERMT2D : avx512_perm_table_3src<0x7E, "d", VR512, memopv16i32, i512mem, defm VPERMT2D : avx512_perm_table_3src<0x7E, "d", VR512, memopv16i32, i512mem,
X86VPermv3, v16i32, VK16WM>, EVEX_V512, EVEX_CD8<32, CD8VF>; X86VPermv3, v16i32, VK16WM, v16i1, GR16>,
EVEX_V512, EVEX_CD8<32, CD8VF>;
defm VPERMT2Q : avx512_perm_table_3src<0x7E, "q", VR512, memopv8i64, i512mem, defm VPERMT2Q : avx512_perm_table_3src<0x7E, "q", VR512, memopv8i64, i512mem,
X86VPermv3, v8i64, VK8WM>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; X86VPermv3, v8i64, VK8WM, v8i1, GR8>,
EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
defm VPERMT2PS : avx512_perm_table_3src<0x7F, "ps", VR512, memopv16f32, i512mem, defm VPERMT2PS : avx512_perm_table_3src<0x7F, "ps", VR512, memopv16f32, i512mem,
X86VPermv3, v16f32, VK16WM>, EVEX_V512, EVEX_CD8<32, CD8VF>; X86VPermv3, v16f32, VK16WM, v16i1, GR16>,
EVEX_V512, EVEX_CD8<32, CD8VF>;
defm VPERMT2PD : avx512_perm_table_3src<0x7F, "pd", VR512, memopv8f64, i512mem, defm VPERMT2PD : avx512_perm_table_3src<0x7F, "pd", VR512, memopv8f64, i512mem,
X86VPermv3, v8f64, VK8WM>, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; X86VPermv3, v8f64, VK8WM, v8i1, GR8>,
EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>;
//===----------------------------------------------------------------------===// //===----------------------------------------------------------------------===//
// AVX-512 - BLEND using mask // AVX-512 - BLEND using mask

7
llvm/test/CodeGen/X86/avx512-intrinsics.ll
View File

@ -594,6 +594,13 @@ define <16 x float> @test_vpermt2ps(<16 x float>%x, <16 x float>%y, <16 x i32>%p
ret <16 x float> %res ret <16 x float> %res
} }
define <16 x float> @test_vpermt2ps_mask(<16 x float>%x, <16 x float>%y, <16 x i32>%perm, i16 %mask) {
; CHECK-LABEL: test_vpermt2ps_mask:
; CHECK: vpermt2ps %zmm1, %zmm2, %zmm0 {%k1} ## encoding: [0x62,0xf2,0x6d,0x49,0x7f,0xc1]
%res = call <16 x float> @llvm.x86.avx512.mask.vpermt.ps.512(<16 x i32>%perm, <16 x float>%x, <16 x float>%y, i16 %mask)
ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.mask.vpermt.ps.512(<16 x i32>, <16 x float>, <16 x float>, i16) declare <16 x float> @llvm.x86.avx512.mask.vpermt.ps.512(<16 x i32>, <16 x float>, <16 x float>, i16)
define <8 x i64> @test_vmovntdqa(i8 *%x) { define <8 x i64> @test_vmovntdqa(i8 *%x) {