summaryrefslogtreecommitdiff
path: root/src/compiler/codegen/MethodCodegenDriver.cc
diff options
context:
space:
mode:
author Shih-wei Liao <sliao@google.com> 2012-03-10 21:55:51 -0800
committer Shih-wei Liao <sliao@google.com> 2012-03-13 10:26:51 -0700
commitc4c9881e5d22432f3f1a30eeec5aa109dfc08a7d (patch)
tree98f1125db961451699880019c8663d76e7b94d02 /src/compiler/codegen/MethodCodegenDriver.cc
parent1580379bb02ee7866ca94d54851a4d08de12a5a4 (diff)
Fix the LLVM build. Separate out libart-compiler-llvm.so.
After this refactoring: If (!USE_LLVM_COMPILER), I pass the test-art and boot the phone alright. The behavior is the same for non-LLVM builds as before. Multi-target art-compiler shared libraries are generated as before. If (USE_LLVM_COMPILER), I generate libart-compiler-llvm.so successfully. Note that the ideal refactoring for the next step will be to always build art-compiler and LLVM-[arm|mips|x86] shared libraries. Currently, we haven't removed all the ifdefs for libart.so yet, so this is not ideal yet. Reason for not-done-yet is that the next step requires discussions. I can remove all the ifdefs today, but then the (!USE_LLVM_COMPILER) build will incur extra fields and overhead. This refactoring allows us to restore the LLVM build situation to be before the breakage yesterday, so we can continue making forward progress again. One difficulty in the refactoring is that LLVM is not a method compiler in default, unlike (!USE_LLVM_COMPILER). LLVM actually compiles an LLVM module at a time normally. So we need to do more call backs and tell libart-compiler-llvm.so in each invocation where we are in terms of OatCompilationUnit. Another difficulty is that currently our LLVM compiler is not multithreaded and requires locking, unlike (!USE_LLVM_COMPILER). So more callbacks are needed. This will be fixed when we fix the multithreading issue. Change-Id: I93bce21b6d673254188f2a60b1a7f91b508e497f
Diffstat (limited to 'src/compiler/codegen/MethodCodegenDriver.cc')
0 files changed, 0 insertions, 0 deletions