[cfe-dev] clang/Linux/x86-32 status on the LLVM test suite

[Daniel Dunbar]

Hi all,

I just finished going through the test results for running clang on
the LLVM test suite on Linux/x86-32 and thought I would share the
current results.

There are 12 total failures:

Presumed miscompilations:
--
MultiSource/Applications/oggenc/oggenc
  PR3480: clang miscompiles MultiSource/Applications/oggenc?
MultiSource/Benchmarks/Prolangs-C/cdecl/cdecl
  PR3481: clang miscompiles MultiSource/Benchmarks/Prolangs-C/cdecl
MultiSource/Benchmarks/VersaBench/bmm/bmm
  PR3482: clang miscompiles MultiSource/Benchmarks/VersaBench/bmm
MultiSource/Benchmarks/VersaBench/dbms/dbms
  PR3483: clang miscompiles MultiSource/Benchmarks/VersaBench/dbms

Other:
--
MultiSource/Benchmarks/ASC_Sequoia/CrystalMk/CrystalMk
SingleSource/Benchmarks/BenchmarkGame/spectral-norm
SingleSource/Regression/C/2003-05-22-VarSizeArray
  PR3454: error: cannot codegen this break/continue/return inside
scope with VLA yet

MultiSource/Benchmarks/McCat/18-imp/imp
  PR2821: clang rejects ANSI C style prototype + K&R style definition

SingleSource/UnitTests/2006-01-23-UnionInit
  PR2561: flexible array member initializers generate error

SingleSource/Benchmarks/BenchmarkGame/fannkuch
  both gcc/llc timeout? not sure what the deal is here.
SingleSource/UnitTests/2007-04-25-weak
  no attribute weak support
SingleSource/UnitTests/Threads/tls
  no TLS support
--

I used:  CCC_ADD_ARGS=-ccc-echo make LLVMGXX=false LLVMGCC="$(which
xcc) -std=gnu89 -m32" TEST=nightly DISABLE_CBE=1 DISABLE_JIT=1 report
to run the test suite; xcc is just a symlink to the new ccc. I hand
filtered out the failures due to faking LLVMGXX.

 - Daniel