Displaying 2 results from an estimated 2 matches for "bzip2_regression".
2018 Dec 05
2
Strange regalloc behaviour: one more available register causes much worse allocation
...but I'm basically fumbling in the dark because
I don't yet know how or why the register allocator is making the decisions it
does and I thought I'd send an email to see if anyone has any advice.
The problem
-----------
Attached are (zipped, as llvm-dev has a 100kb message limit):
* bzip2_regression.ll (reduced from bzip2 in spec2006 after being compiled with
some patches that I'm working on) which demonstrates the problem.
* 0001-AArch64-JumpTableDest-scratch-register-isn-t-earlycl.patch which causes
the problem.
* without_patch_regalloc.txt, the regalloc debug log for llc -mcpu=c...
2018 Dec 05
3
Strange regalloc behaviour: one more available register causes much worse allocation
...but I'm basically fumbling in the dark because
I don't yet know how or why the register allocator is making the decisions it
does and I thought I'd send an email to see if anyone has any advice.
The problem
-----------
Attached are (zipped, as llvm-dev has a 100kb message limit):
* bzip2_regression.ll (reduced from bzip2 in spec2006 after being compiled with
some patches that I'm working on) which demonstrates the problem.
* 0001-AArch64-JumpTableDest-scratch-register-isn-t-earlycl.patch which causes
the problem.
* without_patch_regalloc.txt, the regalloc debug log for llc -mcpu=c...