[ValueTracking][X86] computeKnownBitsFromOperator - add PMULH/PMULHU intrinsics mulhs/mulhu known bits handling.

These map directly to the KnownBits implementations.

Author: RKSimon

llvm/lib/Analysis/ValueTracking.cpp

@@ -1739,6 +1739,20 @@ static void computeKnownBitsFromOperator(const Operator *I,
         Known &= Known2.anyextOrTrunc(BitWidth);
         break;
       }
+      case Intrinsic::x86_sse2_pmulh_w:
+      case Intrinsic::x86_avx2_pmulh_w:
+      case Intrinsic::x86_avx512_pmulh_w_512:
+        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
+        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
+        Known = KnownBits::mulhs(Known, Known2);
+        break;
+      case Intrinsic::x86_sse2_pmulhu_w:
+      case Intrinsic::x86_avx2_pmulhu_w:
+      case Intrinsic::x86_avx512_pmulhu_w_512:
+        computeKnownBits(I->getOperand(0), DemandedElts, Known, Depth + 1, Q);
+        computeKnownBits(I->getOperand(1), DemandedElts, Known2, Depth + 1, Q);
+        Known = KnownBits::mulhu(Known, Known2);
+        break;
       case Intrinsic::x86_sse42_crc32_64_64:
         Known.Zero.setBitsFrom(32);
         break;

llvm/test/Transforms/InstCombine/X86/x86-pmulh.ll

@@ -240,11 +240,7 @@ define <32 x i16> @elts_pmulh_512(<32 x i16> %a0, <32 x i16> %a1) {
 
 define <8 x i16> @known_pmulh_128(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2) {
 ; CHECK-LABEL: @known_pmulh_128(
-; CHECK-NEXT:    [[X0:%.*]] = lshr <8 x i16> [[A0:%.*]], <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
-; CHECK-NEXT:    [[X1:%.*]] = and <8 x i16> [[A1:%.*]], <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; CHECK-NEXT:    [[M:%.*]] = tail call <8 x i16> @llvm.x86.sse2.pmulh.w(<8 x i16> [[X0]], <8 x i16> [[X1]])
-; CHECK-NEXT:    [[R:%.*]] = add <8 x i16> [[M]], [[A2:%.*]]
-; CHECK-NEXT:    ret <8 x i16> [[R]]
+; CHECK-NEXT:    ret <8 x i16> [[A2:%.*]]
 ;
   %x0 = lshr <8 x i16> %a0, <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   %x1 = and <8 x i16> %a1, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -255,11 +251,7 @@ define <8 x i16> @known_pmulh_128(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2) {
 
 define <16 x i16> @known_pmulh_256(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> %a2) {
 ; CHECK-LABEL: @known_pmulh_256(
-; CHECK-NEXT:    [[X0:%.*]] = lshr <16 x i16> [[A0:%.*]], <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
-; CHECK-NEXT:    [[X1:%.*]] = and <16 x i16> [[A1:%.*]], <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; CHECK-NEXT:    [[M:%.*]] = tail call <16 x i16> @llvm.x86.avx2.pmulh.w(<16 x i16> [[X0]], <16 x i16> [[X1]])
-; CHECK-NEXT:    [[R:%.*]] = add <16 x i16> [[M]], [[A2:%.*]]
-; CHECK-NEXT:    ret <16 x i16> [[R]]
+; CHECK-NEXT:    ret <16 x i16> [[A2:%.*]]
 ;
   %x0 = lshr <16 x i16> %a0, <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   %x1 = and <16 x i16> %a1, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -270,11 +262,7 @@ define <16 x i16> @known_pmulh_256(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> %a
 
 define <32 x i16> @known_pmulh_512(<32 x i16> %a0, <32 x i16> %a1, <32 x i16> %a2) {
 ; CHECK-LABEL: @known_pmulh_512(
-; CHECK-NEXT:    [[X0:%.*]] = lshr <32 x i16> [[A0:%.*]], <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
-; CHECK-NEXT:    [[X1:%.*]] = and <32 x i16> [[A1:%.*]], <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; CHECK-NEXT:    [[M:%.*]] = tail call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> [[X0]], <32 x i16> [[X1]])
-; CHECK-NEXT:    [[R:%.*]] = add <32 x i16> [[M]], [[A2:%.*]]
-; CHECK-NEXT:    ret <32 x i16> [[R]]
+; CHECK-NEXT:    ret <32 x i16> [[A2:%.*]]
 ;
   %x0 = lshr <32 x i16> %a0, <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   %x1 = and <32 x i16> %a1, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>

llvm/test/Transforms/InstCombine/X86/x86-pmulhu.ll

@@ -234,11 +234,7 @@ define <32 x i16> @elts_pmulhu_512(<32 x i16> %a0, <32 x i16> %a1) {
 
 define <8 x i16> @known_pmulhu_128(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2) {
 ; CHECK-LABEL: @known_pmulhu_128(
-; CHECK-NEXT:    [[X0:%.*]] = lshr <8 x i16> [[A0:%.*]], <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
-; CHECK-NEXT:    [[X1:%.*]] = and <8 x i16> [[A1:%.*]], <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; CHECK-NEXT:    [[M:%.*]] = tail call <8 x i16> @llvm.x86.sse2.pmulhu.w(<8 x i16> [[X0]], <8 x i16> [[X1]])
-; CHECK-NEXT:    [[R:%.*]] = add <8 x i16> [[M]], [[A2:%.*]]
-; CHECK-NEXT:    ret <8 x i16> [[R]]
+; CHECK-NEXT:    ret <8 x i16> [[A2:%.*]]
 ;
   %x0 = lshr <8 x i16> %a0, <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   %x1 = and <8 x i16> %a1, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -249,11 +245,7 @@ define <8 x i16> @known_pmulhu_128(<8 x i16> %a0, <8 x i16> %a1, <8 x i16> %a2)
 
 define <16 x i16> @known_pmulhu_256(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> %a2) {
 ; CHECK-LABEL: @known_pmulhu_256(
-; CHECK-NEXT:    [[X0:%.*]] = lshr <16 x i16> [[A0:%.*]], <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
-; CHECK-NEXT:    [[X1:%.*]] = and <16 x i16> [[A1:%.*]], <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; CHECK-NEXT:    [[M:%.*]] = tail call <16 x i16> @llvm.x86.avx2.pmulhu.w(<16 x i16> [[X0]], <16 x i16> [[X1]])
-; CHECK-NEXT:    [[R:%.*]] = add <16 x i16> [[M]], [[A2:%.*]]
-; CHECK-NEXT:    ret <16 x i16> [[R]]
+; CHECK-NEXT:    ret <16 x i16> [[A2:%.*]]
 ;
   %x0 = lshr <16 x i16> %a0, <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   %x1 = and <16 x i16> %a1, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
@@ -264,11 +256,7 @@ define <16 x i16> @known_pmulhu_256(<16 x i16> %a0, <16 x i16> %a1, <16 x i16> %
 
 define <32 x i16> @known_pmulhu_512(<32 x i16> %a0, <32 x i16> %a1, <32 x i16> %a2) {
 ; CHECK-LABEL: @known_pmulhu_512(
-; CHECK-NEXT:    [[X0:%.*]] = lshr <32 x i16> [[A0:%.*]], <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
-; CHECK-NEXT:    [[X1:%.*]] = and <32 x i16> [[A1:%.*]], <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>
-; CHECK-NEXT:    [[M:%.*]] = tail call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> [[X0]], <32 x i16> [[X1]])
-; CHECK-NEXT:    [[R:%.*]] = add <32 x i16> [[M]], [[A2:%.*]]
-; CHECK-NEXT:    ret <32 x i16> [[R]]
+; CHECK-NEXT:    ret <32 x i16> [[A2:%.*]]
 ;
   %x0 = lshr <32 x i16> %a0, <i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15, i16 8, i16 9, i16 10, i16 11, i16 12, i16 13, i16 14, i16 15>
   %x1 = and <32 x i16> %a1, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3>