[SelectionDAG] Add support for vector demandedelts in MUL opcodes

llvm-svn: 286471

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index e272315..5ad4dcb 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -2166,11 +2166,13 @@ void SelectionDAG::computeKnownBits(SDValue Op, APInt &KnownZero,
     break;
   }
   case ISD::MUL: {
-    computeKnownBits(Op.getOperand(1), KnownZero, KnownOne, Depth+1);
-    computeKnownBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1);
+    computeKnownBits(Op.getOperand(1), KnownZero, KnownOne, DemandedElts,
+                     Depth + 1);
+    computeKnownBits(Op.getOperand(0), KnownZero2, KnownOne2, DemandedElts,
+                     Depth + 1);
 
     // If low bits are zero in either operand, output low known-0 bits.
-    // Also compute a conserative estimate for high known-0 bits.
+    // Also compute a conservative estimate for high known-0 bits.
     // More trickiness is possible, but this is sufficient for the
     // interesting case of alignment computation.
     KnownOne.clearAllBits();

diff --git a/llvm/test/CodeGen/X86/known-bits-vector.ll b/llvm/test/CodeGen/X86/known-bits-vector.ll
index 3306313..f1b757f 100644 (file)
--- a/llvm/test/CodeGen/X86/known-bits-vector.ll
+++ b/llvm/test/CodeGen/X86/known-bits-vector.ll
@@ -173,18 +173,12 @@ define <4 x i32> @knownbits_mask_ashr_shuffle_lshr(<4 x i32> %a0) nounwind {
 define <4 x i32> @knownbits_mask_mul_shuffle_shl(<4 x i32> %a0, <4 x i32> %a1) nounwind {
 ; X32-LABEL: knownbits_mask_mul_shuffle_shl:
 ; X32:       # BB#0:
-; X32-NEXT:    vpand {{\.LCPI.*}}, %xmm0, %xmm0
-; X32-NEXT:    vpmulld %xmm0, %xmm1, %xmm0
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3]
-; X32-NEXT:    vpslld $22, %xmm0, %xmm0
+; X32-NEXT:    vxorps %xmm0, %xmm0, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: knownbits_mask_mul_shuffle_shl:
 ; X64:       # BB#0:
-; X64-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; X64-NEXT:    vpmulld %xmm0, %xmm1, %xmm0
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3]
-; X64-NEXT:    vpslld $22, %xmm0, %xmm0
+; X64-NEXT:    vxorps %xmm0, %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = and <4 x i32> %a0, <i32 -65536, i32 -7, i32 -7, i32 -65536>
   %2 = mul <4 x i32> %a1, %1