X86: Add patterns for MULHU/MULHS of v8i16 and v16i16.

This gets us pretty code for divs of i16 vectors. Turn the existing
intrinsics into the corresponding nodes.

llvm-svn: 207317

3 files changed, 79 insertions, 4 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 9a419bb5269..ed2d7e6e8c0 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -941,6 +941,8 @@ void X86TargetLowering::resetOperationActions() {
     setOperationAction(ISD::MUL,                MVT::v4i32, Custom);
     setOperationAction(ISD::MUL,                MVT::v2i64, Custom);
     setOperationAction(ISD::UMUL_LOHI,          MVT::v4i32, Custom);
+    setOperationAction(ISD::MULHU,              MVT::v8i16, Legal);
+    setOperationAction(ISD::MULHS,              MVT::v8i16, Legal);
     setOperationAction(ISD::SUB,                MVT::v16i8, Legal);
     setOperationAction(ISD::SUB,                MVT::v8i16, Legal);
     setOperationAction(ISD::SUB,                MVT::v4i32, Legal);
@@ -1225,6 +1227,8 @@ void X86TargetLowering::resetOperationActions() {
       // Don't lower v32i8 because there is no 128-bit byte mul
 
       setOperationAction(ISD::UMUL_LOHI,       MVT::v8i32, Custom);
+      setOperationAction(ISD::MULHU,           MVT::v16i16, Legal);
+      setOperationAction(ISD::MULHS,           MVT::v16i16, Legal);
 
       setOperationAction(ISD::VSELECT,         MVT::v32i8, Legal);
     } else {
@@ -11725,6 +11729,16 @@ static SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) {
     return DAG.getNode(X86ISD::PMULUDQ, dl, Op.getValueType(),
                        Op.getOperand(1), Op.getOperand(2));
 
+  case Intrinsic::x86_sse2_pmulhu_w:
+  case Intrinsic::x86_avx2_pmulhu_w:
+    return DAG.getNode(ISD::MULHU, dl, Op.getValueType(),
+                       Op.getOperand(1), Op.getOperand(2));
+
+  case Intrinsic::x86_sse2_pmulh_w:
+  case Intrinsic::x86_avx2_pmulh_w:
+    return DAG.getNode(ISD::MULHS, dl, Op.getValueType(),
+                       Op.getOperand(1), Op.getOperand(2));
+
   // SSE2/AVX2 sub with unsigned saturation intrinsics
   case Intrinsic::x86_sse2_psubus_b:
   case Intrinsic::x86_sse2_psubus_w:
diff --git a/llvm/lib/Target/X86/X86InstrSSE.td b/llvm/lib/Target/X86/X86InstrSSE.td
index 3b0cc6bff2f..eb036f23eb7 100644
--- a/llvm/lib/Target/X86/X86InstrSSE.td
+++ b/llvm/lib/Target/X86/X86InstrSSE.td
@@ -4081,6 +4081,10 @@ defm PADDQ   : PDI_binop_all<0xD4, "paddq", add, v2i64, v4i64,
                              SSE_INTALUQ_ITINS_P, 1>;
 defm PMULLW  : PDI_binop_all<0xD5, "pmullw", mul, v8i16, v16i16,
                              SSE_INTMUL_ITINS_P, 1>;
+defm PMULHUW : PDI_binop_all<0xE4, "pmulhuw", mulhu, v8i16, v16i16,
+                             SSE_INTMUL_ITINS_P, 1>;
+defm PMULHW  : PDI_binop_all<0xE5, "pmulhw", mulhs, v8i16, v16i16,
+                             SSE_INTMUL_ITINS_P, 1>;
 defm PSUBB   : PDI_binop_all<0xF8, "psubb", sub, v16i8, v32i8,
                              SSE_INTALU_ITINS_P, 0>;
 defm PSUBW   : PDI_binop_all<0xF9, "psubw", sub, v8i16, v16i16,
@@ -4115,10 +4119,6 @@ defm PADDUSB : PDI_binop_all_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
                                  int_x86_avx2_paddus_b, SSE_INTALU_ITINS_P, 1>;
 defm PADDUSW : PDI_binop_all_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
                                  int_x86_avx2_paddus_w, SSE_INTALU_ITINS_P, 1>;
-defm PMULHUW : PDI_binop_all_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
-                                 int_x86_avx2_pmulhu_w, SSE_INTMUL_ITINS_P, 1>;
-defm PMULHW  : PDI_binop_all_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
-                                 int_x86_avx2_pmulh_w, SSE_INTMUL_ITINS_P, 1>;
 defm PMADDWD : PDI_binop_all_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
                                  int_x86_avx2_pmadd_wd, SSE_PMADD, 1>;
 defm PAVGB   : PDI_binop_all_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
diff --git a/llvm/test/CodeGen/X86/vector-idiv.ll b/llvm/test/CodeGen/X86/vector-idiv.ll
index 5b8153a9685..0d43b96163d 100644
--- a/llvm/test/CodeGen/X86/vector-idiv.ll
+++ b/llvm/test/CodeGen/X86/vector-idiv.ll
@@ -42,4 +42,65 @@ define <8 x i32> @test2(<8 x i32> %a) {
 ; AVX: vpsrld $2
 }
 
+define <8 x i16> @test3(<8 x i16> %a) {
+  %div = udiv <8 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+  ret <8 x i16> %div
+
+; SSE-LABEL: test3:
+; SSE: pmulhuw
+; SSE: psubw
+; SSE: psrlw $1
+; SSE: paddw
+; SSE: psrlw $2
+
+; AVX-LABEL: test3:
+; AVX: vpmulhuw
+; AVX: vpsubw
+; AVX: vpsrlw $1
+; AVX: vpaddw
+; AVX: vpsrlw $2
+}
+
+define <16 x i16> @test4(<16 x i16> %a) {
+  %div = udiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7,i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7,i16 7, i16 7, i16 7, i16 7>
+  ret <16 x i16> %div
+
+; AVX-LABEL: test4:
+; AVX: vpmulhuw
+; AVX: vpsubw
+; AVX: vpsrlw $1
+; AVX: vpaddw
+; AVX: vpsrlw $2
+; AVX-NOT: vpmulhuw
+}
+
+define <8 x i16> @test5(<8 x i16> %a) {
+  %div = sdiv <8 x i16> %a, <i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7>
+  ret <8 x i16> %div
+
+; SSE-LABEL: test5:
+; SSE: pmulhw
+; SSE: psrlw $15
+; SSE: psraw $1
+; SSE: paddw
+
+; AVX-LABEL: test5:
+; AVX: vpmulhw
+; AVX: vpsrlw $15
+; AVX: vpsraw $1
+; AVX: vpaddw
+}
+
+define <16 x i16> @test6(<16 x i16> %a) {
+  %div = sdiv <16 x i16> %a, <i16 7, i16 7, i16 7, i16 7,i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7, i16 7,i16 7, i16 7, i16 7, i16 7>
+  ret <16 x i16> %div
+
+; AVX-LABEL: test6:
+; AVX: vpmulhw
+; AVX: vpsrlw $15
+; AVX: vpsraw $1
+; AVX: vpaddw
+; AVX-NOT: vpmulhw
+}
+
 ; TODO: sdiv -> pmuldq