RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2024-11-26 05:00:26 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

If compiling for PPC on an i386 box, the LTO wouldn't get the altivec (and

Browse Source 
other) feature information. The workaround is inelegant and could be cleaned up
if this information were available some other way (say, in the IR).



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@52447 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Bill Wendling 2008-06-18 06:35:30 +00:00
parent 0521e3cdc1
commit 604a818463
2 changed files with 51 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

53
tools/lto2/LTOCodeGenerator.cpp
View File

@ -32,6 +32,7 @@
#include "llvm/Analysis/Verifier.h"
#include "llvm/Analysis/LoadValueNumbering.h"
#include "llvm/CodeGen/FileWriters.h"
#include "llvm/Target/SubtargetFeature.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetMachine.h"
@ -261,8 +262,24 @@ bool LTOCodeGenerator::determineTarget(std::string& errMsg)
*mergedModule, errMsg);
if ( march == NULL )
return true;
std::string features;
_target = march->CtorFn(*mergedModule, features);
// construct LTModule, hand over ownership of module and target
//
// FIXME: This is an inelegant way of specifying the features of a
// subtarget. It would be better if we could encode this information
// into the IR. See <rdar://5972456>.
SubtargetFeatures Features;
std::string FeatureStr;
std::string TargetTriple = _linker.getModule()->getTargetTriple();
if (strncmp(TargetTriple.c_str(), "powerpc-apple-", 14) == 0) {
Features.AddFeature("altivec", true);
} else if (strncmp(TargetTriple.c_str(), "powerpc64-apple-", 16) == 0) {
Features.AddFeature("64bit", true);
Features.AddFeature("altivec", true);
}
_target = march->CtorFn(*mergedModule, Features.getString());
}
return false;
}
@ -302,7 +319,6 @@ void LTOCodeGenerator::applyScopeRestrictions()
}
}
/// Optimize merged modules using various IPO passes
bool LTOCodeGenerator::generateAssemblyCode(std::ostream& out, std::string& errMsg)
{
@ -360,12 +376,9 @@ bool LTOCodeGenerator::generateAssemblyCode(std::ostream& out, std::string& errM
// function pointers. When this happens, we often have to resolve varargs
// calls, etc, so let instcombine do this.
passes.add(createInstructionCombiningPass());
passes.add(createFunctionInliningPass()); // Inline small functions
passes.add(createPruneEHPass()); // Remove dead EH info
passes.add(createGlobalDCEPass()); // Remove dead functions
passes.add(createFunctionInliningPass()); // Inline small functions
passes.add(createPruneEHPass()); // Remove dead EH info
passes.add(createGlobalDCEPass()); // Remove dead functions
// If we didn't decide to inline a function, check to see if we can
// transform it to pass arguments by value instead of by reference.
@ -377,16 +390,14 @@ bool LTOCodeGenerator::generateAssemblyCode(std::ostream& out, std::string& errM
passes.add(createScalarReplAggregatesPass()); // Break up allocas
// Run a few AA driven optimizations here and now, to cleanup the code.
passes.add(createGlobalsModRefPass()); // IP alias analysis
passes.add(createLICMPass()); // Hoist loop invariants
passes.add(createGVNPass()); // Remove common subexprs
passes.add(createMemCpyOptPass()); // Remove dead memcpy's
passes.add(createGlobalsModRefPass()); // IP alias analysis
passes.add(createLICMPass()); // Hoist loop invariants
passes.add(createGVNPass()); // Remove common subexprs
passes.add(createMemCpyOptPass()); // Remove dead memcpy's
passes.add(createDeadStoreEliminationPass()); // Nuke dead stores
// Cleanup and simplify the code after the scalar optimizations.
passes.add(createInstructionCombiningPass());
passes.add(createJumpThreadingPass()); // Thread jumps.
// Delete basic blocks, which optimization passes may have killed...
@ -431,13 +442,13 @@ bool LTOCodeGenerator::generateAssemblyCode(std::ostream& out, std::string& errM
// Run the code generator, and write assembly file
codeGenPasses->doInitialization();
for (Module::iterator it = mergedModule->begin(),
e = mergedModule->end(); it != e; ++it) {
if (!it->isDeclaration())
codeGenPasses->run(*it);
}
codeGenPasses->doFinalization();
for (Module::iterator
it = mergedModule->begin(), e = mergedModule->end(); it != e; ++it)
if (!it->isDeclaration())
codeGenPasses->run(*it);
codeGenPasses->doFinalization();
return false; // success
}

22
tools/lto2/LTOModule.cpp
View File

@ -24,11 +24,11 @@
#include "llvm/Support/MathExtras.h"
#include "llvm/System/Path.h"
#include "llvm/System/Process.h"
#include "llvm/Target/SubtargetFeature.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetMachineRegistry.h"
#include "llvm/Target/TargetAsmInfo.h"
#include <fstream>
using namespace llvm;
@ -125,11 +125,27 @@ LTOModule* LTOModule::makeLTOModule(MemoryBuffer* buffer, std::string& errMsg)
// find machine architecture for this module
const TargetMachineRegistry::entry* march =
TargetMachineRegistry::getClosestStaticTargetForModule(*m, errMsg);
if ( march == NULL )
return NULL;
// construct LTModule, hand over ownership of module and target
std::string features;
TargetMachine* target = march->CtorFn(*m, features);
//
// FIXME: This is an inelegant way of specifying the features of a
// subtarget. It would be better if we could encode this information into
// the IR. See <rdar://5972456>.
SubtargetFeatures Features;
std::string FeatureStr;
const char *TargetTriple = m->getTargetTriple().c_str();
if (strncmp(TargetTriple, "powerpc-apple-", 14) == 0) {
Features.AddFeature("altivec", true);
} else if (strncmp(TargetTriple, "powerpc64-apple-", 16) == 0) {
Features.AddFeature("64bit", true);
Features.AddFeature("altivec", true);
}
TargetMachine* target = march->CtorFn(*m, Features.getString());
return new LTOModule(m.take(), target);
}