diff options
author | Chris Lattner <sabre@nondot.org> | 2003-12-01 07:05:31 +0000 |
---|---|---|
committer | Chris Lattner <sabre@nondot.org> | 2003-12-01 07:05:31 +0000 |
commit | 8d1dbd2134d2bcd83f63c504c1f8f019c0342f92 (patch) | |
tree | dfdab9d1fae48af6cbe07615745c6120c7d17524 /lib/Bytecode/Reader | |
parent | d540e2c9956a7fca733af7db6e3bbcb29aedbb4f (diff) |
Emit & read more compressed bytecode by not emitting a bytecodeblock for
each basic block in function. Instead, just emit a stream of instructions,
chopping up basic blocks based on when we find terminator instructions. This
saves a fairly substantial chunk of bytecode space. In stripped, sample
cases, for example, we get this reduction in size:
197.parser: 163036 -> 137180: 18.8% reduction
254.gap : 844936 -> 689392: 22.6%
255.vortex: 621724 -> 528444: 17.7%
...
Not bad for something this simple. :) Note that this doesn't require a new
bytecode version number at all, though version 1.1 should not need to support
the old format.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@10280 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Bytecode/Reader')
-rw-r--r-- | lib/Bytecode/Reader/Reader.cpp | 42 | ||||
-rw-r--r-- | lib/Bytecode/Reader/ReaderInternals.h | 4 |
2 files changed, 44 insertions, 2 deletions
diff --git a/lib/Bytecode/Reader/Reader.cpp b/lib/Bytecode/Reader/Reader.cpp index ed8f9e804c..097d548b98 100644 --- a/lib/Bytecode/Reader/Reader.cpp +++ b/lib/Bytecode/Reader/Reader.cpp @@ -188,7 +188,8 @@ Constant *BytecodeParser::getConstantValue(unsigned TypeSlot, unsigned Slot) { } } - +/// ParseBasicBlock - In LLVM 1.0 bytecode files, we used to output one +/// basicblock at a time. This method reads in one of the basicblock packets. BasicBlock *BytecodeParser::ParseBasicBlock(const unsigned char *&Buf, const unsigned char *EndBuf, unsigned BlockNo) { @@ -207,6 +208,38 @@ BasicBlock *BytecodeParser::ParseBasicBlock(const unsigned char *&Buf, return BB; } + +/// ParseInstructionList - Parse all of the BasicBlock's & Instruction's in the +/// body of a function. In post 1.0 bytecode files, we no longer emit basic +/// block individually, in order to avoid per-basic-block overhead. +unsigned BytecodeParser::ParseInstructionList(Function *F, + const unsigned char *&Buf, + const unsigned char *EndBuf) { + unsigned BlockNo = 0; + std::vector<unsigned> Args; + + while (Buf < EndBuf) { + BasicBlock *BB; + if (ParsedBasicBlocks.size() == BlockNo) + ParsedBasicBlocks.push_back(BB = new BasicBlock()); + else if (ParsedBasicBlocks[BlockNo] == 0) + BB = ParsedBasicBlocks[BlockNo] = new BasicBlock(); + else + BB = ParsedBasicBlocks[BlockNo]; + ++BlockNo; + F->getBasicBlockList().push_back(BB); + + // Read instructions into this basic block until we get to a terminator + while (Buf < EndBuf && !BB->getTerminator()) + ParseInstruction(Buf, EndBuf, Args, BB); + + if (!BB->getTerminator()) + throw std::string("Non-terminated basic block found!"); + } + + return BlockNo; +} + void BytecodeParser::ParseSymbolTable(const unsigned char *&Buf, const unsigned char *EndBuf, SymbolTable *ST, @@ -345,6 +378,13 @@ void BytecodeParser::materializeFunction(Function* F) { break; } + case BytecodeFormat::InstructionList: { + BCR_TRACE(2, "BLOCK BytecodeFormat::InstructionList: {\n"); + if (BlockNum) throw std::string("Already parsed basic blocks!"); + BlockNum = ParseInstructionList(F, Buf, Buf+Size); + break; + } + case BytecodeFormat::SymbolTable: BCR_TRACE(2, "BLOCK BytecodeFormat::SymbolTable: {\n"); ParseSymbolTable(Buf, Buf+Size, &F->getSymbolTable(), F); diff --git a/lib/Bytecode/Reader/ReaderInternals.h b/lib/Bytecode/Reader/ReaderInternals.h index aea45c2bed..fd0a1ed008 100644 --- a/lib/Bytecode/Reader/ReaderInternals.h +++ b/lib/Bytecode/Reader/ReaderInternals.h @@ -162,7 +162,9 @@ private: BasicBlock *ParseBasicBlock(const unsigned char *&Buf, const unsigned char *End, unsigned BlockNo); - + unsigned ParseInstructionList(Function *F, const unsigned char *&Buf, + const unsigned char *EndBuf); + void ParseInstruction(const unsigned char *&Buf, const unsigned char *End, std::vector<unsigned> &Args, BasicBlock *BB); |