[X86] Reduce math for index calculation for inserting and extracting subvectors and...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@251740 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index e8b1b8bb5d300850152d58a4161d6078b074b687..3910c5b7e65e963bca0cb1ecf4de9800f758a688 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -4373,19 +4373,18 @@ static SDValue ExtractSubVector(SDValue Vec, unsigned IdxVal,
 
   // Extract the relevant vectorWidth bits.  Generate an EXTRACT_SUBVECTOR
   unsigned ElemsPerChunk = vectorWidth / ElVT.getSizeInBits();
+  assert(isPowerOf2_32(ElemsPerChunk) && "Elements per chunk not power of 2");
 
   // This is the index of the first element of the vectorWidth-bit chunk
-  // we want.
-  unsigned NormalizedIdxVal = (((IdxVal * ElVT.getSizeInBits()) / vectorWidth)
-                               * ElemsPerChunk);
+  // we want. Since ElemsPerChunk is a power of 2 just need to clear bits.
+  IdxVal &= ~(ElemsPerChunk - 1);
 
   // If the input is a buildvector just emit a smaller one.
   if (Vec.getOpcode() == ISD::BUILD_VECTOR)
     return DAG.getNode(ISD::BUILD_VECTOR, dl, ResultVT,
-                       makeArrayRef(Vec->op_begin() + NormalizedIdxVal,
-                                    ElemsPerChunk));
+                       makeArrayRef(Vec->op_begin() + IdxVal, ElemsPerChunk));
 
-  SDValue VecIdx = DAG.getIntPtrConstant(NormalizedIdxVal, dl);
+  SDValue VecIdx = DAG.getIntPtrConstant(IdxVal, dl);
   return DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, ResultVT, Vec, VecIdx);
 }
 
@@ -4423,13 +4422,13 @@ static SDValue InsertSubVector(SDValue Result, SDValue Vec,
 
   // Insert the relevant vectorWidth bits.
   unsigned ElemsPerChunk = vectorWidth/ElVT.getSizeInBits();
+  assert(isPowerOf2_32(ElemsPerChunk) && "Elements per chunk not power of 2");
 
   // This is the index of the first element of the vectorWidth-bit chunk
-  // we want.
-  unsigned NormalizedIdxVal = (((IdxVal * ElVT.getSizeInBits())/vectorWidth)
-                               * ElemsPerChunk);
+  // we want. Since ElemsPerChunk is a power of 2 just need to clear bits.
+  IdxVal &= ~(ElemsPerChunk - 1);
 
-  SDValue VecIdx = DAG.getIntPtrConstant(NormalizedIdxVal, dl);
+  SDValue VecIdx = DAG.getIntPtrConstant(IdxVal, dl);
   return DAG.getNode(ISD::INSERT_SUBVECTOR, dl, ResultVT, Result, Vec, VecIdx);
 }
 
@@ -11390,10 +11389,11 @@ X86TargetLowering::LowerEXTRACT_VECTOR_ELT(SDValue Op,
     MVT EltVT = VecVT.getVectorElementType();
 
     unsigned ElemsPerChunk = 128 / EltVT.getSizeInBits();
+    assert(isPowerOf2_32(ElemsPerChunk) && "Elements per chunk not power of 2");
 
-    //if (IdxVal >= NumElems/2)
-    //  IdxVal -= NumElems/2;
-    IdxVal -= (IdxVal/ElemsPerChunk)*ElemsPerChunk;
+    // Find IdxVal modulo ElemsPerChunk. Since ElemsPerChunk is a power of 2
+    // this can be done with a mask.
+    IdxVal &= ElemsPerChunk - 1;
     return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, Op.getValueType(), Vec,
                        DAG.getConstant(IdxVal, dl, MVT::i32));
   }
@@ -11529,7 +11529,9 @@ SDValue X86TargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op,
 
     // Insert the element into the desired chunk.
     unsigned NumEltsIn128 = 128 / EltVT.getSizeInBits();
-    unsigned IdxIn128 = IdxVal - (IdxVal / NumEltsIn128) * NumEltsIn128;
+    assert(isPowerOf2_32(NumEltsIn128));
+    // Since NumEltsIn128 is a power of 2 we can use mask instead of modulo.
+    unsigned IdxIn128 = IdxVal & (NumEltsIn128 - 1);
 
     V = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, V.getValueType(), V, N1,
                     DAG.getConstant(IdxIn128, dl, MVT::i32));