[PATCH] D50850: clang: Add triples support for MIPS r6

[Simon Atanasyan via Phabricator via cfe-commits]

atanasyan added a comment.

In https://reviews.llvm.org/D50850#1250285, @wzssyqa wrote:

> This is really for Clang. I guess you mean that compiler-rt directory also need to be patched.


If you take a look at the previous version of this patch https://reviews.llvm.org/D50850?id=167419, you see that it changed LLVM files. Probably you attached another diff.

As to `compiler-rt` - the Phabricator replace `R6` symbols by the `https://reviews.llvm.org/source/compiler-rt/` links. So my statement was "Could you attach an actual patch brings `R6` support to the Clang driver?".



================
Comment at: lib/Driver/ToolChains/Gnu.cpp:2093
     BiarchTripleAliases.append(begin(MIPSELTriples), end(MIPSELTriples));
-    BiarchTripleAliases.append(begin(MIPSTriples), end(MIPSTriples));
+    BiarchLibDirs.append(begin(MIPSN32ELLibDirs), end(MIPSN32ELLibDirs));
+    BiarchTripleAliases.append(begin(MIPSN32ELTriples), end(MIPSN32ELTriples));
----------------
wzssyqa wrote:
> atanasyan wrote:
> > Ditto
> As a question: why  MIPSTriples here?
> the above mips64 lines don't include any EL Triples.
I do not remember exact answer, although I might be an author if this code. Maybe it handle some complicated directories tree from multi-arch toolchains. Are all tests passed if you remove this line?


================
Comment at: lib/Driver/ToolChains/Gnu.cpp:2437
     if (getTriple().getEnvironment() == llvm::Triple::GNUABI64 ||
-        getTriple().isAndroid() ||
-        getTriple().isOSFreeBSD() ||
+        getTriple().getEnvironment() == llvm::Triple::GNUABIN32 ||
+        getTriple().isAndroid() || getTriple().isOSFreeBSD() ||
----------------
wzssyqa wrote:
> atanasyan wrote:
> > Are you sure that integrated LLVM assembler is ready to support N32 code generation? Anyway this change is for  a separate patch.
> I didn't have so many test, while helloworld programs really works.
> 
> 
You created a patch that teaches the Clang driver to understand (pass arguments to backend, find includes and libraries etc) N32 ABI better. Now I can compile "hello world". And after that you decided that LLVM backend does not have any MIPS N32 ABI related problems. I think we can enable the integrated assembler when a) it's possible to recursively build Clang with N32 ABI, b) all tests from LLVM test suite (https://llvm.org/docs/TestSuiteGuide.html) are passed in N32 ABI mode, c) we check that LLVM produces correct DWARF for N32.


https://reviews.llvm.org/D50850