Regression.CodeGen.Generic.2004-04-09-SameValueCoalescing.llx and the
code size problem.
This bug prevented us from doing most register coallesces.
llvm-svn: 16031
- Implement parsing of the .output={bytecode|assembly} item.
- Drop parsing support for translator.optimizes, translator.groks_dash_O,
optimizer.groks_dash_O, translator.output_is_asm, optimizer.output_is_asm
- Add parsing support for translator.output and optimizer.output
- Add optimizer.required parsing support
- Add linker.libs and linker.libpaths parsing support
- Fix error messages to list correct set of tokens expected.
- Rename FileReadable -> FileIsReadable (changed in CompilerDriver.cpp)
llvm-svn: 16023
- Drop name tokens not used any more (GROKS_DASH_O)
- Rename OUTPUT_IS_ASM as OUTPUT and allow "bytecode" and "assembly" as
values of it.
- Simplify handleContext (now handleNameContext) arguments and ensure that
it returns OPTION tokens exactly as scanned.
llvm-svn: 16021
- Renamed FileReadable -> FileIsReadable
- Remove support for GROKS_DASH_O; deferr it to %opts% substitution
- Require a -o option except in case of linking where it defaults to a.out
- Fix problem processing *.o and *.bc files.
- Ignore optimization phase if -emit-raw-code
- Correct the assembly generation phase to support -emit-native-code
- Implement dependent library lookup
llvm-svn: 16020
with emitting .xwords when not on an 8-byte boundary (.xword 0 is not the
same as 8 .byte 0's). Because we do not know when or when we are not aligned,
just emit bytes like the old V9 asmprinter did.
llvm-svn: 16006
- Rearrange output order to make more sense
- Make only the function level output count as "detailed"
- Output dump output directly to stream, don't buffer it.
- Fix counting of block sizes
- Implement new handlers for number of types, dependent libs, target triple
- Compute the size of block headers.
llvm-svn: 15985
- Pass the output stream to the analyzer so it can write its output there
directly instead of buffering it.
- Don't pass a boolean to ParseBytecode because its not needed any more.
llvm-svn: 15983
Remove the "processFunctions" boolean from ParseBytecode as it is no
longer needed. This is part of avoiding double reading of functions
when analyzing bytecode.
llvm-svn: 15982
- provide the correct conversion for ModuleBlockID in read_block (a potential
bug but not actually exploited because reading module block ids doesn't
use read_block).
- install support for handleTypeList handler
- install support for handleDependentLibrary handler
- install support for handleTargetTriple handler
- clean up comments, output strings,
- ensure that processing function arguments doesn't SIGSEGV if one of the
arguments is a null pointer (yeah, it happened .. weird)
- prepare for version 5 bytecode by documenting what will change.
llvm-svn: 15981
