#ifndef BITSTREAM_WRITER_H
#define BITSTREAM_WRITER_H
+#include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/SmallVector.h"
#include "llvm/Bitcode/BitCodes.h"
#include <vector>
namespace llvm {
class BitstreamWriter {
- std::vector<unsigned char> &Out;
+ SmallVectorImpl<char> &Out;
/// CurBit - Always between 0 and 31 inclusive, specifies the next bit to use.
unsigned CurBit;
};
std::vector<BlockInfo> BlockInfoRecords;
+ // BackpatchWord - Backpatch a 32-bit word in the output with the specified
+ // value.
+ void BackpatchWord(unsigned ByteNo, unsigned NewWord) {
+ Out[ByteNo++] = (unsigned char)(NewWord >> 0);
+ Out[ByteNo++] = (unsigned char)(NewWord >> 8);
+ Out[ByteNo++] = (unsigned char)(NewWord >> 16);
+ Out[ByteNo ] = (unsigned char)(NewWord >> 24);
+ }
+
+ void WriteByte(unsigned char Value) {
+ Out.push_back(Value);
+ }
+
+ void WriteWord(unsigned Value) {
+ unsigned char Bytes[4] = {
+ (unsigned char)(Value >> 0),
+ (unsigned char)(Value >> 8),
+ (unsigned char)(Value >> 16),
+ (unsigned char)(Value >> 24) };
+ Out.append(&Bytes[0], &Bytes[4]);
+ }
+
+ unsigned GetBufferOffset() const {
+ return Out.size();
+ }
+
+ unsigned GetWordIndex() const {
+ unsigned Offset = GetBufferOffset();
+ assert((Offset & 3) == 0 && "Not 32-bit aligned");
+ return Offset / 4;
+ }
+
public:
- explicit BitstreamWriter(std::vector<unsigned char> &O)
+ explicit BitstreamWriter(SmallVectorImpl<char> &O)
: Out(O), CurBit(0), CurValue(0), CurCodeSize(2) {}
~BitstreamWriter() {
}
}
- std::vector<unsigned char> &getBuffer() { return Out; }
+ /// \brief Retrieve the current position in the stream, in bits.
+ uint64_t GetCurrentBitNo() const { return GetBufferOffset() * 8 + CurBit; }
//===--------------------------------------------------------------------===//
// Basic Primitives for emitting bits to the stream.
//===--------------------------------------------------------------------===//
void Emit(uint32_t Val, unsigned NumBits) {
- assert(NumBits <= 32 && "Invalid value size!");
+ assert(NumBits && NumBits <= 32 && "Invalid value size!");
assert((Val & ~(~0U >> (32-NumBits))) == 0 && "High bits set!");
CurValue |= Val << CurBit;
if (CurBit + NumBits < 32) {
}
// Add the current word.
- unsigned V = CurValue;
- Out.push_back((unsigned char)(V >> 0));
- Out.push_back((unsigned char)(V >> 8));
- Out.push_back((unsigned char)(V >> 16));
- Out.push_back((unsigned char)(V >> 24));
+ WriteWord(CurValue);
if (CurBit)
CurValue = Val >> (32-CurBit);
void FlushToWord() {
if (CurBit) {
- unsigned V = CurValue;
- Out.push_back((unsigned char)(V >> 0));
- Out.push_back((unsigned char)(V >> 8));
- Out.push_back((unsigned char)(V >> 16));
- Out.push_back((unsigned char)(V >> 24));
+ WriteWord(CurValue);
CurBit = 0;
CurValue = 0;
}
}
void EmitVBR(uint32_t Val, unsigned NumBits) {
+ assert(NumBits <= 32 && "Too many bits to emit!");
uint32_t Threshold = 1U << (NumBits-1);
// Emit the bits with VBR encoding, NumBits-1 bits at a time.
}
void EmitVBR64(uint64_t Val, unsigned NumBits) {
+ assert(NumBits <= 32 && "Too many bits to emit!");
if ((uint32_t)Val == Val)
return EmitVBR((uint32_t)Val, NumBits);
- uint64_t Threshold = 1U << (NumBits-1);
+ uint64_t Threshold = uint64_t(1U << (NumBits-1));
// Emit the bits with VBR encoding, NumBits-1 bits at a time.
while (Val >= Threshold) {
Emit(Val, CurCodeSize);
}
- // BackpatchWord - Backpatch a 32-bit word in the output with the specified
- // value.
- void BackpatchWord(unsigned ByteNo, unsigned NewWord) {
- Out[ByteNo++] = (unsigned char)(NewWord >> 0);
- Out[ByteNo++] = (unsigned char)(NewWord >> 8);
- Out[ByteNo++] = (unsigned char)(NewWord >> 16);
- Out[ByteNo ] = (unsigned char)(NewWord >> 24);
- }
-
//===--------------------------------------------------------------------===//
// Block Manipulation
//===--------------------------------------------------------------------===//
EmitVBR(CodeLen, bitc::CodeLenWidth);
FlushToWord();
- unsigned BlockSizeWordLoc = static_cast<unsigned>(Out.size());
+ unsigned BlockSizeWordIndex = GetWordIndex();
unsigned OldCodeSize = CurCodeSize;
// Emit a placeholder, which will be replaced when the block is popped.
// Push the outer block's abbrev set onto the stack, start out with an
// empty abbrev set.
- BlockScope.push_back(Block(OldCodeSize, BlockSizeWordLoc/4));
+ BlockScope.push_back(Block(OldCodeSize, BlockSizeWordIndex));
BlockScope.back().PrevAbbrevs.swap(CurAbbrevs);
// If there is a blockinfo for this BlockID, add all the predefined abbrevs
FlushToWord();
// Compute the size of the block, in words, not counting the size field.
- unsigned SizeInWords= static_cast<unsigned>(Out.size())/4-B.StartSizeWord-1;
+ unsigned SizeInWords = GetWordIndex() - B.StartSizeWord - 1;
unsigned ByteNo = B.StartSizeWord*4;
// Update the block size field in the header of this sub-block.
// Encode the value as we are commanded.
switch (Op.getEncoding()) {
- default: assert(0 && "Unknown encoding!");
+ default: llvm_unreachable("Unknown encoding!");
case BitCodeAbbrevOp::Fixed:
- Emit((unsigned)V, (unsigned)Op.getEncodingData());
+ if (Op.getEncodingData())
+ Emit((unsigned)V, (unsigned)Op.getEncodingData());
break;
case BitCodeAbbrevOp::VBR:
- EmitVBR64(V, (unsigned)Op.getEncodingData());
+ if (Op.getEncodingData())
+ EmitVBR64(V, (unsigned)Op.getEncodingData());
break;
case BitCodeAbbrevOp::Char6:
Emit(BitCodeAbbrevOp::EncodeChar6((char)V), 6);
}
}
-public:
-
- /// EmitRecord - Emit the specified record to the stream, using an abbrev if
- /// we have one to compress the output.
+ /// EmitRecordWithAbbrevImpl - This is the core implementation of the record
+ /// emission code. If BlobData is non-null, then it specifies an array of
+ /// data that should be emitted as part of the Blob or Array operand that is
+ /// known to exist at the end of the record.
template<typename uintty>
- void EmitRecord(unsigned Code, SmallVectorImpl<uintty> &Vals,
- unsigned Abbrev = 0) {
- if (!Abbrev) {
- // If we don't have an abbrev to use, emit this in its fully unabbreviated
- // form.
- EmitCode(bitc::UNABBREV_RECORD);
- EmitVBR(Code, 6);
- EmitVBR(static_cast<uint32_t>(Vals.size()), 6);
- for (unsigned i = 0, e = static_cast<unsigned>(Vals.size()); i != e; ++i)
- EmitVBR64(Vals[i], 6);
- return;
- }
-
+ void EmitRecordWithAbbrevImpl(unsigned Abbrev, SmallVectorImpl<uintty> &Vals,
+ StringRef Blob) {
+ const char *BlobData = Blob.data();
+ unsigned BlobLen = (unsigned) Blob.size();
unsigned AbbrevNo = Abbrev-bitc::FIRST_APPLICATION_ABBREV;
assert(AbbrevNo < CurAbbrevs.size() && "Invalid abbrev #!");
BitCodeAbbrev *Abbv = CurAbbrevs[AbbrevNo];
EmitCode(Abbrev);
- // Insert the code into Vals to treat it uniformly.
- Vals.insert(Vals.begin(), Code);
-
unsigned RecordIdx = 0;
for (unsigned i = 0, e = static_cast<unsigned>(Abbv->getNumOperandInfos());
i != e; ++i) {
assert(RecordIdx < Vals.size() && "Invalid abbrev/record");
EmitAbbreviatedLiteral(Op, Vals[RecordIdx]);
++RecordIdx;
- } else if (Op.getEncoding() != BitCodeAbbrevOp::Array) {
- assert(RecordIdx < Vals.size() && "Invalid abbrev/record");
- EmitAbbreviatedField(Op, Vals[RecordIdx]);
- ++RecordIdx;
- } else {
+ } else if (Op.getEncoding() == BitCodeAbbrevOp::Array) {
// Array case.
assert(i+2 == e && "array op not second to last?");
const BitCodeAbbrevOp &EltEnc = Abbv->getOperandInfo(++i);
+ // If this record has blob data, emit it, otherwise we must have record
+ // entries to encode this way.
+ if (BlobData) {
+ assert(RecordIdx == Vals.size() &&
+ "Blob data and record entries specified for array!");
+ // Emit a vbr6 to indicate the number of elements present.
+ EmitVBR(static_cast<uint32_t>(BlobLen), 6);
+
+ // Emit each field.
+ for (unsigned i = 0; i != BlobLen; ++i)
+ EmitAbbreviatedField(EltEnc, (unsigned char)BlobData[i]);
+
+ // Know that blob data is consumed for assertion below.
+ BlobData = 0;
+ } else {
+ // Emit a vbr6 to indicate the number of elements present.
+ EmitVBR(static_cast<uint32_t>(Vals.size()-RecordIdx), 6);
+
+ // Emit each field.
+ for (unsigned e = Vals.size(); RecordIdx != e; ++RecordIdx)
+ EmitAbbreviatedField(EltEnc, Vals[RecordIdx]);
+ }
+ } else if (Op.getEncoding() == BitCodeAbbrevOp::Blob) {
+ // If this record has blob data, emit it, otherwise we must have record
+ // entries to encode this way.
+
// Emit a vbr6 to indicate the number of elements present.
- EmitVBR(static_cast<uint32_t>(Vals.size()-RecordIdx), 6);
-
- // Emit each field.
- for (; RecordIdx != Vals.size(); ++RecordIdx)
- EmitAbbreviatedField(EltEnc, Vals[RecordIdx]);
+ if (BlobData) {
+ EmitVBR(static_cast<uint32_t>(BlobLen), 6);
+ assert(RecordIdx == Vals.size() &&
+ "Blob data and record entries specified for blob operand!");
+ } else {
+ EmitVBR(static_cast<uint32_t>(Vals.size()-RecordIdx), 6);
+ }
+
+ // Flush to a 32-bit alignment boundary.
+ FlushToWord();
+
+ // Emit each field as a literal byte.
+ if (BlobData) {
+ for (unsigned i = 0; i != BlobLen; ++i)
+ WriteByte((unsigned char)BlobData[i]);
+
+ // Know that blob data is consumed for assertion below.
+ BlobData = 0;
+ } else {
+ for (unsigned e = Vals.size(); RecordIdx != e; ++RecordIdx) {
+ assert(Vals[RecordIdx] < 256 && "Value too large to emit as blob");
+ WriteByte((unsigned char)Vals[RecordIdx]);
+ }
+ }
+
+ // Align end to 32-bits.
+ while (GetBufferOffset() & 3)
+ WriteByte(0);
+ } else { // Single scalar field.
+ assert(RecordIdx < Vals.size() && "Invalid abbrev/record");
+ EmitAbbreviatedField(Op, Vals[RecordIdx]);
+ ++RecordIdx;
}
}
assert(RecordIdx == Vals.size() && "Not all record operands emitted!");
+ assert(BlobData == 0 &&
+ "Blob data specified for record that doesn't use it!");
}
+
+public:
+ /// EmitRecord - Emit the specified record to the stream, using an abbrev if
+ /// we have one to compress the output.
+ template<typename uintty>
+ void EmitRecord(unsigned Code, SmallVectorImpl<uintty> &Vals,
+ unsigned Abbrev = 0) {
+ if (!Abbrev) {
+ // If we don't have an abbrev to use, emit this in its fully unabbreviated
+ // form.
+ EmitCode(bitc::UNABBREV_RECORD);
+ EmitVBR(Code, 6);
+ EmitVBR(static_cast<uint32_t>(Vals.size()), 6);
+ for (unsigned i = 0, e = static_cast<unsigned>(Vals.size()); i != e; ++i)
+ EmitVBR64(Vals[i], 6);
+ return;
+ }
+
+ // Insert the code into Vals to treat it uniformly.
+ Vals.insert(Vals.begin(), Code);
+
+ EmitRecordWithAbbrev(Abbrev, Vals);
+ }
+
+ /// EmitRecordWithAbbrev - Emit a record with the specified abbreviation.
+ /// Unlike EmitRecord, the code for the record should be included in Vals as
+ /// the first entry.
+ template<typename uintty>
+ void EmitRecordWithAbbrev(unsigned Abbrev, SmallVectorImpl<uintty> &Vals) {
+ EmitRecordWithAbbrevImpl(Abbrev, Vals, StringRef());
+ }
+
+ /// EmitRecordWithBlob - Emit the specified record to the stream, using an
+ /// abbrev that includes a blob at the end. The blob data to emit is
+ /// specified by the pointer and length specified at the end. In contrast to
+ /// EmitRecord, this routine expects that the first entry in Vals is the code
+ /// of the record.
+ template<typename uintty>
+ void EmitRecordWithBlob(unsigned Abbrev, SmallVectorImpl<uintty> &Vals,
+ StringRef Blob) {
+ EmitRecordWithAbbrevImpl(Abbrev, Vals, Blob);
+ }
+ template<typename uintty>
+ void EmitRecordWithBlob(unsigned Abbrev, SmallVectorImpl<uintty> &Vals,
+ const char *BlobData, unsigned BlobLen) {
+ return EmitRecordWithAbbrevImpl(Abbrev, Vals, StringRef(BlobData, BlobLen));
+ }
+
+ /// EmitRecordWithArray - Just like EmitRecordWithBlob, works with records
+ /// that end with an array.
+ template<typename uintty>
+ void EmitRecordWithArray(unsigned Abbrev, SmallVectorImpl<uintty> &Vals,
+ StringRef Array) {
+ EmitRecordWithAbbrevImpl(Abbrev, Vals, Array);
+ }
+ template<typename uintty>
+ void EmitRecordWithArray(unsigned Abbrev, SmallVectorImpl<uintty> &Vals,
+ const char *ArrayData, unsigned ArrayLen) {
+ return EmitRecordWithAbbrevImpl(Abbrev, Vals, StringRef(ArrayData,
+ ArrayLen));
+ }
+
//===--------------------------------------------------------------------===//
// Abbrev Emission
//===--------------------------------------------------------------------===//
/// EnterBlockInfoBlock - Start emitting the BLOCKINFO_BLOCK.
void EnterBlockInfoBlock(unsigned CodeWidth) {
EnterSubblock(bitc::BLOCKINFO_BLOCK_ID, CodeWidth);
- BlockInfoCurBID = -1U;
+ BlockInfoCurBID = ~0U;
}
private:
/// SwitchToBlockID - If we aren't already talking about the specified block