[InstCombine][X86] Add multiply-by-one handling for MULH/PMULHU/PMULHRS intrinsics

MULH/PMULHU simplifies to ASHR/ZERO as they just become a SEXT/ZEXT sign-splat instruction

PMULHRS doesn't simplify as much so I've not attempted to fold it.

3 files changed, 22 insertions, 18 deletions

diff --git a/llvm/lib/Target/X86/X86InstCombineIntrinsic.cpp b/llvm/lib/Target/X86/X86InstCombineIntrinsic.cpp
index 6d4734d..163584b 100644
--- a/llvm/lib/Target/X86/X86InstCombineIntrinsic.cpp
+++ b/llvm/lib/Target/X86/X86InstCombineIntrinsic.cpp
@@ -521,6 +521,16 @@ static Value *simplifyX86pmulh(IntrinsicInst &II,
   if (isa<ConstantAggregateZero>(Arg0) || isa<ConstantAggregateZero>(Arg1))
     return ConstantAggregateZero::get(ResTy);
 
+  // Multiply by one.
+  if (!IsRounding) {
+    if (match(Arg0, PatternMatch::m_One()))
+      return IsSigned ? Builder.CreateAShr(Arg1, 15)
+                      : ConstantAggregateZero::get(ResTy);
+    if (match(Arg1, PatternMatch::m_One()))
+      return IsSigned ? Builder.CreateAShr(Arg0, 15)
+                      : ConstantAggregateZero::get(ResTy);
+  }
+
   // Constant folding.
   if (!isa<Constant>(Arg0) || !isa<Constant>(Arg1))
     return nullptr;
diff --git a/llvm/test/Transforms/InstCombine/X86/x86-pmulh.ll b/llvm/test/Transforms/InstCombine/X86/x86-pmulh.ll
index 699a3c9..185ab46 100644
--- a/llvm/test/Transforms/InstCombine/X86/x86-pmulh.ll
+++ b/llvm/test/Transforms/InstCombine/X86/x86-pmulh.ll
@@ -111,7 +111,7 @@ define <32 x i16> @zero_pmulh_512_commute(<32 x i16> %a0) {
 
 define <8 x i16> @one_pmulh_128(<8 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulh_128(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> [[A0:%.*]], <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
+; CHECK-NEXT:    [[TMP1:%.*]] = ashr <8 x i16> [[A0:%.*]], <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
 ; CHECK-NEXT:    ret <8 x i16> [[TMP1]]
 ;
   %1 = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> %a0, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
@@ -120,7 +120,7 @@ define <8 x i16> @one_pmulh_128(<8 x i16> %a0) {
 
 define <8 x i16> @one_pmulh_128_commute(<8 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulh_128_commute(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <8 x i16> [[A0:%.*]])
+; CHECK-NEXT:    [[TMP1:%.*]] = ashr <8 x i16> [[A0:%.*]], <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
 ; CHECK-NEXT:    ret <8 x i16> [[TMP1]]
 ;
   %1 = call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <8 x i16> %a0)
@@ -129,7 +129,7 @@ define <8 x i16> @one_pmulh_128_commute(<8 x i16> %a0) {
 
 define <16 x i16> @one_pmulh_256(<16 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulh_256(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i16> @llvm.x86.avx2.pmulh.w(<16 x i16> [[A0:%.*]], <16 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
+; CHECK-NEXT:    [[TMP1:%.*]] = ashr <16 x i16> [[A0:%.*]], <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
 ; CHECK-NEXT:    ret <16 x i16> [[TMP1]]
 ;
   %1 = call <16 x i16> @llvm.x86.avx2.pmulh.w(<16 x i16> %a0, <16 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
@@ -138,7 +138,7 @@ define <16 x i16> @one_pmulh_256(<16 x i16> %a0) {
 
 define <16 x i16> @one_pmulh_256_commute(<16 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulh_256_commute(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i16> @llvm.x86.avx2.pmulh.w(<16 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <16 x i16> [[A0:%.*]])
+; CHECK-NEXT:    [[TMP1:%.*]] = ashr <16 x i16> [[A0:%.*]], <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
 ; CHECK-NEXT:    ret <16 x i16> [[TMP1]]
 ;
   %1 = call <16 x i16> @llvm.x86.avx2.pmulh.w(<16 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <16 x i16> %a0)
@@ -147,7 +147,7 @@ define <16 x i16> @one_pmulh_256_commute(<16 x i16> %a0) {
 
 define <32 x i16> @one_pmulh_512(<32 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulh_512(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> [[A0:%.*]], <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
+; CHECK-NEXT:    [[TMP1:%.*]] = ashr <32 x i16> [[A0:%.*]], <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
 ; CHECK-NEXT:    ret <32 x i16> [[TMP1]]
 ;
   %1 = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> %a0, <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
@@ -156,7 +156,7 @@ define <32 x i16> @one_pmulh_512(<32 x i16> %a0) {
 
 define <32 x i16> @one_pmulh_512_commute(<32 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulh_512_commute(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <32 x i16> [[A0:%.*]])
+; CHECK-NEXT:    [[TMP1:%.*]] = ashr <32 x i16> [[A0:%.*]], <i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15>
 ; CHECK-NEXT:    ret <32 x i16> [[TMP1]]
 ;
   %1 = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <32 x i16> %a0)
diff --git a/llvm/test/Transforms/InstCombine/X86/x86-pmulhu.ll b/llvm/test/Transforms/InstCombine/X86/x86-pmulhu.ll
index e970ae6..b18833f 100644
--- a/llvm/test/Transforms/InstCombine/X86/x86-pmulhu.ll
+++ b/llvm/test/Transforms/InstCombine/X86/x86-pmulhu.ll
@@ -111,8 +111,7 @@ define <32 x i16> @zero_pmulhu_512_commute(<32 x i16> %a0) {
 
 define <8 x i16> @one_pmulhu_128(<8 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulhu_128(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16> [[A0:%.*]], <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
-; CHECK-NEXT:    ret <8 x i16> [[TMP1]]
+; CHECK-NEXT:    ret <8 x i16> zeroinitializer
 ;
   %1 = call <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16> %a0, <8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
   ret <8 x i16> %1
@@ -120,8 +119,7 @@ define <8 x i16> @one_pmulhu_128(<8 x i16> %a0) {
 
 define <8 x i16> @one_pmulhu_128_commute(<8 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulhu_128_commute(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <8 x i16> [[A0:%.*]])
-; CHECK-NEXT:    ret <8 x i16> [[TMP1]]
+; CHECK-NEXT:    ret <8 x i16> zeroinitializer
 ;
   %1 = call <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <8 x i16> %a0)
   ret <8 x i16> %1
@@ -129,8 +127,7 @@ define <8 x i16> @one_pmulhu_128_commute(<8 x i16> %a0) {
 
 define <16 x i16> @one_pmulhu_256(<16 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulhu_256(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i16> @llvm.x86.avx2.pmulhu.w(<16 x i16> [[A0:%.*]], <16 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
-; CHECK-NEXT:    ret <16 x i16> [[TMP1]]
+; CHECK-NEXT:    ret <16 x i16> zeroinitializer
 ;
   %1 = call <16 x i16> @llvm.x86.avx2.pmulhu.w(<16 x i16> %a0, <16 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
   ret <16 x i16> %1
@@ -138,8 +135,7 @@ define <16 x i16> @one_pmulhu_256(<16 x i16> %a0) {
 
 define <16 x i16> @one_pmulhu_256_commute(<16 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulhu_256_commute(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <16 x i16> @llvm.x86.avx2.pmulhu.w(<16 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <16 x i16> [[A0:%.*]])
-; CHECK-NEXT:    ret <16 x i16> [[TMP1]]
+; CHECK-NEXT:    ret <16 x i16> zeroinitializer
 ;
   %1 = call <16 x i16> @llvm.x86.avx2.pmulhu.w(<16 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <16 x i16> %a0)
   ret <16 x i16> %1
@@ -147,8 +143,7 @@ define <16 x i16> @one_pmulhu_256_commute(<16 x i16> %a0) {
 
 define <32 x i16> @one_pmulhu_512(<32 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulhu_512(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> [[A0:%.*]], <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
-; CHECK-NEXT:    ret <32 x i16> [[TMP1]]
+; CHECK-NEXT:    ret <32 x i16> zeroinitializer
 ;
   %1 = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> %a0, <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>)
   ret <32 x i16> %1
@@ -156,8 +151,7 @@ define <32 x i16> @one_pmulhu_512(<32 x i16> %a0) {
 
 define <32 x i16> @one_pmulhu_512_commute(<32 x i16> %a0) {
 ; CHECK-LABEL: @one_pmulhu_512_commute(
-; CHECK-NEXT:    [[TMP1:%.*]] = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <32 x i16> [[A0:%.*]])
-; CHECK-NEXT:    ret <32 x i16> [[TMP1]]
+; CHECK-NEXT:    ret <32 x i16> zeroinitializer
 ;
   %1 = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1>, <32 x i16> %a0)
   ret <32 x i16> %1