AVX-512: PMIN/PMAX intrinsics and patterns

Patch by Cameron McInally <cameron.mcinally@nyu.edu>


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@193497 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 45 insertions, 1 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 0d78189..5dbef0f 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/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/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index ed90a4b..1ac5630 100644
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/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
 //===----------------------------------------------------------------------===//