mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2025-02-01 05:01:59 +01:00
[APInt] Add APInt::extractBits() method to extract APInt subrange (reapplied)
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: 296272
This commit is contained in:
parent
730c9e7603
commit
879a5b0804
@ -1272,6 +1272,9 @@ public:
|
||||
/// as "bitPosition".
|
||||
void flipBit(unsigned bitPosition);
|
||||
|
||||
/// Return an APInt with the extracted bits [bitPosition,bitPosition+numBits).
|
||||
APInt extractBits(unsigned numBits, unsigned bitPosition) const;
|
||||
|
||||
/// @}
|
||||
/// \name Value Characterization Functions
|
||||
/// @{
|
||||
|
@ -618,6 +618,42 @@ void APInt::flipBit(unsigned bitPosition) {
|
||||
else setBit(bitPosition);
|
||||
}
|
||||
|
||||
APInt APInt::extractBits(unsigned numBits, unsigned bitPosition) const {
|
||||
assert(numBits > 0 && "Can't extract zero bits");
|
||||
assert(bitPosition < BitWidth && (numBits + bitPosition) <= BitWidth &&
|
||||
"Illegal bit extraction");
|
||||
|
||||
if (isSingleWord())
|
||||
return APInt(numBits, VAL >> bitPosition);
|
||||
|
||||
unsigned loBit = whichBit(bitPosition);
|
||||
unsigned loWord = whichWord(bitPosition);
|
||||
unsigned hiWord = whichWord(bitPosition + numBits - 1);
|
||||
|
||||
// Single word result extracting bits from a single word source.
|
||||
if (loWord == hiWord)
|
||||
return APInt(numBits, pVal[loWord] >> loBit);
|
||||
|
||||
// Extracting bits that start on a source word boundary can be done
|
||||
// as a fast memory copy.
|
||||
if (loBit == 0)
|
||||
return APInt(numBits, makeArrayRef(pVal + loWord, 1 + hiWord - loWord));
|
||||
|
||||
// General case - shift + copy source words directly into place.
|
||||
APInt Result(numBits, 0);
|
||||
unsigned NumSrcWords = getNumWords();
|
||||
unsigned NumDstWords = Result.getNumWords();
|
||||
|
||||
for (unsigned word = 0; word < NumDstWords; ++word) {
|
||||
uint64_t w0 = pVal[loWord + word];
|
||||
uint64_t w1 =
|
||||
(loWord + word + 1) < NumSrcWords ? pVal[loWord + word + 1] : 0;
|
||||
Result.pVal[word] = (w0 >> loBit) | (w1 << (APINT_BITS_PER_WORD - loBit));
|
||||
}
|
||||
|
||||
return Result.clearUnusedBits();
|
||||
}
|
||||
|
||||
unsigned APInt::getBitsNeeded(StringRef str, uint8_t radix) {
|
||||
assert(!str.empty() && "Invalid string length");
|
||||
assert((radix == 10 || radix == 8 || radix == 16 || radix == 2 ||
|
||||
|
@ -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) &&
|
||||
|
@ -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();
|
||||
}
|
||||
|
||||
|
@ -1505,3 +1505,19 @@ TEST(APIntTest, reverseBits) {
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
TEST(APIntTest, extractBits) {
|
||||
APInt i32(32, 0x1234567);
|
||||
EXPECT_EQ(0x3456, i32.extractBits(16, 4));
|
||||
|
||||
APInt i257(257, 0xFFFFFFFFFF0000FFull, true);
|
||||
EXPECT_EQ(0xFFu, i257.extractBits(16, 0));
|
||||
EXPECT_EQ((0xFFu >> 1), i257.extractBits(16, 1));
|
||||
EXPECT_EQ(-1, i257.extractBits(32, 64).getSExtValue());
|
||||
EXPECT_EQ(-1, i257.extractBits(128, 128).getSExtValue());
|
||||
EXPECT_EQ(-1, i257.extractBits(66, 191).getSExtValue());
|
||||
EXPECT_EQ(static_cast<int64_t>(0xFFFFFFFFFF80007Full),
|
||||
i257.extractBits(128, 1).getSExtValue());
|
||||
EXPECT_EQ(static_cast<int64_t>(0xFFFFFFFFFF80007Full),
|
||||
i257.extractBits(129, 1).getSExtValue());
|
||||
}
|
||||
|
Loading…
x
Reference in New Issue
Block a user