[APInt] Add APInt::extractBits() method to extract APInt subrange

The current pattern for extract bits in range is typically:

Mask.lshr(BitOffset).trunc(SubSizeInBits);

Which can be particularly slow for large APInts (MaskSizeInBits > 64) as they require the allocation of memory for the temporary variable.

This is another of the compile time issues identified in PR32037 (see also D30265).

This patch adds the APInt::extractBits() helper method which avoids the temporary memory allocation.

Differential Revision: https://reviews.llvm.org/D30336

llvm-svn: 296141

2 files changed, 7 insertions, 8 deletions

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 618d7a8c561..348f8308099 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -5207,8 +5207,8 @@ static bool getTargetConstantBitsFromNode(SDValue Op, unsigned EltSizeInBits,
     EltBits.resize(NumElts, APInt(EltSizeInBits, 0));
 
     for (unsigned i = 0; i != NumElts; ++i) {
-      APInt UndefEltBits = UndefBits.lshr(i * EltSizeInBits);
-      UndefEltBits = UndefEltBits.zextOrTrunc(EltSizeInBits);
+      unsigned BitOffset = i * EltSizeInBits;
+      APInt UndefEltBits = UndefBits.extractBits(EltSizeInBits, BitOffset);
 
       // Only treat an element as UNDEF if all bits are UNDEF.
       if (UndefEltBits.isAllOnesValue()) {
@@ -5223,7 +5223,7 @@ static bool getTargetConstantBitsFromNode(SDValue Op, unsigned EltSizeInBits,
       if (UndefEltBits.getBoolValue() && !AllowPartialUndefs)
         return false;
 
-      APInt Bits = MaskBits.lshr(i * EltSizeInBits).zextOrTrunc(EltSizeInBits);
+      APInt Bits = MaskBits.extractBits(EltSizeInBits, BitOffset);
       EltBits[i] = Bits.getZExtValue();
     }
     return true;
@@ -6421,7 +6421,7 @@ static Constant *getConstantVector(MVT VT, const APInt &SplatValue,
 
   SmallVector<Constant *, 32> ConstantVec;
   for (unsigned i = 0; i < NumElm; i++) {
-    APInt Val = SplatValue.lshr(ScalarSize * i).trunc(ScalarSize);
+    APInt Val = SplatValue.extractBits(ScalarSize, ScalarSize * i);
     Constant *Const;
     if (VT.isFloatingPoint()) {
       assert((ScalarSize == 32 || ScalarSize == 64) &&
diff --git a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
index 41ad0971461..c24b74f7480 100644
--- a/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
+++ b/llvm/lib/Target/X86/X86ShuffleDecodeConstantPool.cpp
@@ -77,8 +77,8 @@ static bool extractConstantMask(const Constant *C, unsigned MaskEltSizeInBits,
   RawMask.resize(NumMaskElts, 0);
 
   for (unsigned i = 0; i != NumMaskElts; ++i) {
-    APInt EltUndef = UndefBits.lshr(i * MaskEltSizeInBits);
-    EltUndef = EltUndef.zextOrTrunc(MaskEltSizeInBits);
+    unsigned BitOffset = i * MaskEltSizeInBits;
+    APInt EltUndef = UndefBits.extractBits(MaskEltSizeInBits, BitOffset);
 
     // Only treat the element as UNDEF if all bits are UNDEF, otherwise
     // treat it as zero.
@@ -88,8 +88,7 @@ static bool extractConstantMask(const Constant *C, unsigned MaskEltSizeInBits,
       continue;
     }
 
-    APInt EltBits = MaskBits.lshr(i * MaskEltSizeInBits);
-    EltBits = EltBits.zextOrTrunc(MaskEltSizeInBits);
+    APInt EltBits = MaskBits.extractBits(MaskEltSizeInBits, BitOffset);
     RawMask[i] = EltBits.getZExtValue();
   }