diff options
| author | Elena Demikhovsky <elena.demikhovsky@intel.com> | 2013-10-27 08:18:37 +0000 |
|---|---|---|
| committer | Elena Demikhovsky <elena.demikhovsky@intel.com> | 2013-10-27 08:18:37 +0000 |
| commit | 199c8235551c37b8a51a44c080f7fae1b7ab8884 (patch) | |
| tree | 5ce70ec8d52c2492ee62734ce2521070fd6f2e2b /llvm/lib | |
| parent | 6822ecb08707ead61098021bb6764be1a39b23ba (diff) | |
| download | bcm5719-llvm-199c8235551c37b8a51a44c080f7fae1b7ab8884.tar.gz bcm5719-llvm-199c8235551c37b8a51a44c080f7fae1b7ab8884.zip | |
AVX-512: PMIN/PMAX intrinsics and patterns
Patch by Cameron McInally <cameron.mcinally@nyu.edu>
llvm-svn: 193497
Diffstat (limited to 'llvm/lib')
| -rw-r--r-- | llvm/lib/Target/X86/X86ISelLowering.cpp | 18 | ||||
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrAVX512.td | 28 |
2 files changed, 45 insertions, 1 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 0d7818943c8..5dbef0f6fc3 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -11173,24 +11173,32 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) { case Intrinsic::x86_avx2_pmaxu_b: case Intrinsic::x86_avx2_pmaxu_w: case Intrinsic::x86_avx2_pmaxu_d: + case Intrinsic::x86_avx512_pmaxu_d: + case Intrinsic::x86_avx512_pmaxu_q: case Intrinsic::x86_sse2_pminu_b: case Intrinsic::x86_sse41_pminuw: case Intrinsic::x86_sse41_pminud: case Intrinsic::x86_avx2_pminu_b: case Intrinsic::x86_avx2_pminu_w: case Intrinsic::x86_avx2_pminu_d: + case Intrinsic::x86_avx512_pminu_d: + case Intrinsic::x86_avx512_pminu_q: case Intrinsic::x86_sse41_pmaxsb: case Intrinsic::x86_sse2_pmaxs_w: case Intrinsic::x86_sse41_pmaxsd: case Intrinsic::x86_avx2_pmaxs_b: case Intrinsic::x86_avx2_pmaxs_w: case Intrinsic::x86_avx2_pmaxs_d: + case Intrinsic::x86_avx512_pmaxs_d: + case Intrinsic::x86_avx512_pmaxs_q: case Intrinsic::x86_sse41_pminsb: case Intrinsic::x86_sse2_pmins_w: case Intrinsic::x86_sse41_pminsd: case Intrinsic::x86_avx2_pmins_b: case Intrinsic::x86_avx2_pmins_w: - case Intrinsic::x86_avx2_pmins_d: { + case Intrinsic::x86_avx2_pmins_d: + case Intrinsic::x86_avx512_pmins_d: + case Intrinsic::x86_avx512_pmins_q: { unsigned Opcode; switch (IntNo) { default: llvm_unreachable("Impossible intrinsic"); // Can't reach here. @@ -11200,6 +11208,8 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) { case Intrinsic::x86_avx2_pmaxu_b: case Intrinsic::x86_avx2_pmaxu_w: case Intrinsic::x86_avx2_pmaxu_d: + case Intrinsic::x86_avx512_pmaxu_d: + case Intrinsic::x86_avx512_pmaxu_q: Opcode = X86ISD::UMAX; break; case Intrinsic::x86_sse2_pminu_b: @@ -11208,6 +11218,8 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) { case Intrinsic::x86_avx2_pminu_b: case Intrinsic::x86_avx2_pminu_w: case Intrinsic::x86_avx2_pminu_d: + case Intrinsic::x86_avx512_pminu_d: + case Intrinsic::x86_avx512_pminu_q: Opcode = X86ISD::UMIN; break; case Intrinsic::x86_sse41_pmaxsb: @@ -11216,6 +11228,8 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) { case Intrinsic::x86_avx2_pmaxs_b: case Intrinsic::x86_avx2_pmaxs_w: case Intrinsic::x86_avx2_pmaxs_d: + case Intrinsic::x86_avx512_pmaxs_d: + case Intrinsic::x86_avx512_pmaxs_q: Opcode = X86ISD::SMAX; break; case Intrinsic::x86_sse41_pminsb: @@ -11224,6 +11238,8 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) { case Intrinsic::x86_avx2_pmins_b: case Intrinsic::x86_avx2_pmins_w: case Intrinsic::x86_avx2_pmins_d: + case Intrinsic::x86_avx512_pmins_d: + case Intrinsic::x86_avx512_pmins_q: Opcode = X86ISD::SMIN; break; } diff --git a/llvm/lib/Target/X86/X86InstrAVX512.td b/llvm/lib/Target/X86/X86InstrAVX512.td index ed90a4bca3b..1ac563010fd 100644 --- a/llvm/lib/Target/X86/X86InstrAVX512.td +++ b/llvm/lib/Target/X86/X86InstrAVX512.td @@ -1599,6 +1599,34 @@ defm VPMULUDQZ : avx512_binop_rm2<0xF4, "vpmuludq", v8i64, v16i32, def : Pat<(v8i64 (X86pmuludq (v16i32 VR512:$src1), (v16i32 VR512:$src2))), (VPMULUDQZrr VR512:$src1, VR512:$src2)>; +defm VPMAXUDZ : avx512_binop_rm<0x3F, "vpmaxud", X86umax, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + T8, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPMAXUQZ : avx512_binop_rm<0x3F, "vpmaxuq", X86umax, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>, + T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VPMAXSDZ : avx512_binop_rm<0x3D, "vpmaxsd", X86smax, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPMAXSQZ : avx512_binop_rm<0x3D, "vpmaxsq", X86smax, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>, + T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VPMINUDZ : avx512_binop_rm<0x3B, "vpminud", X86umin, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + T8, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPMINUQZ : avx512_binop_rm<0x3B, "vpminuq", X86umin, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>, + T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + +defm VPMINSDZ : avx512_binop_rm<0x39, "vpminsd", X86smin, v16i32, VR512, memopv16i32, + i512mem, loadi32, i32mem, "{1to16}", SSE_INTALU_ITINS_P, 1>, + T8, EVEX_V512, EVEX_CD8<32, CD8VF>; +defm VPMINSQZ : avx512_binop_rm<0x39, "vpminsq", X86smin, v8i64, VR512, memopv8i64, + i512mem, loadi64, i64mem, "{1to8}", SSE_INTALU_ITINS_P, 0>, + T8, EVEX_V512, VEX_W, EVEX_CD8<64, CD8VF>; + //===----------------------------------------------------------------------===// // AVX-512 - Unpack Instructions //===----------------------------------------------------------------------===// |

