[LLVMdev] RFC: Tail call optimization X86 

Arnold Schwaighofer arnold.schwaighofer at gmail.com
Sun Sep 23 09:57:07 PDT 2007


Okay I finished testing (TEST=nightly report):

Results with my optimization not enabled in test-suite look great:

darwin-x86 (r42159):
*passed llvm/test (except for arm failures and one x86
   failure that where also in the vanilla version)
*same number of failures for vanilla version and my
   patch version in test-suite

gentoo/x86 in a parallels virtual machine (really
sloooow, that was an earlier revision):
*passed llvm/test
*same number of failures for vanilla version and my
   patched version in test-suite

gentoo/amd64 (r42159):
*passed llvm/test (except for arm failures and one x86
   failure that where also in the vanilla version)
*llvm-gcc compiled but test-suite failed with glibc double
   free error (also in vanilla version)

gentoo/amd64/32bit chroot (r42159):
*passed llvm/test (except for arm failures and one x86
   failure that where also in the vanilla version)
*could not get llvm-gcc to compile most likely due to
   my strange configuration ;)

Results with my optimization enabled in test-suite look not so bad:
darwin-x86 (r42244):
*passed all llvm/test tests
*test-suite: When turning tail call optimization on in LLC_BETA
   the 'SPASS' test fails (other than those that also fail in the
   regular llc - vanilla version) on darwin-x86

Of course there is always a party of bugs lurking around somewhere ;)

In order to test the optimization i have added some tests to llvm/
test/CodeGen/X86. Most of those tests finally compile the .s file
generated by llc with gcc and execute the resulting executable.
This won't work for non x86 architectures. My question is now what
should i add to the XFAIL (ppc, sparc ?) line to exclude those. Or
should those tests go into the test-suite? Maybe i also should
decrease their quantity?

regards arnold


-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: report.nightly.txt
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20070923/48e2ca7e/attachment.txt>
-------------- next part --------------




More information about the llvm-dev mailing list