3 Commits

Author SHA1 Message Date
Matt Arsenault
a15bb16493 AMDGPU: Make frame register caller preserved
Using SplitCSR for the frame register was very broken. Often
the copies in the prolog and epilog were optimized out, in addition
to them being inserted after the true prolog where the FP
was clobbered.

I have a hacky solution which works that continues to use
split CSR, but for now this is simpler and will get to working
programs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@313274 91177308-0d34-0410-b5e6-96231b3b80d8
2017-09-14 17:14:57 +00:00
Matt Arsenault
1a6aed20fa IPRA: Don't assume called function is first call operand
Fixes not finding the called global for AMDGPU
call pseudoinstructions, which prevented IPRA
from doing much.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@311637 91177308-0d34-0410-b5e6-96231b3b80d8
2017-08-24 07:55:15 +00:00
Matt Arsenault
e1f2d6cc65 IPRA: Don't crash on null getCallPreservedMask
Kernels aren't callable, so they don't have a call preserved mask.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@310172 91177308-0d34-0410-b5e6-96231b3b80d8
2017-08-05 07:50:18 +00:00