[PATCH] D15537: limit the number of instructions per block examined by dead store elimination

Bob Haarman via llvm-commits llvm-commits at lists.llvm.org
Fri Feb 5 11:00:01 PST 2016

On 2016-01-05 15:40 , Bruno Cardoso Lopes wrote:
> Hi Bob,
> Thanks for taking a look at this. Is there a testcase you can share?
http://reviews.llvm.org/P1649 (warning: it's about 5MB) is what I used 
to test with. It's a preprossed version of 

I compile it with
   clang++ -nostdinc -std=c++14 -O3 -Wno-deprecated 
-Wno-invalid-offsetof -S -emit-llvm -o constants.ll constants.ii

With Clang 67e7611e1ae0b3cccaaaa0f22c2185abac5d956e and LLVM 
b4612d88320c2624fff969237a4477c0719f5bc7, this takes about 94 seconds on 
my machine.

With the patch applied to LLVM, it takes about 27 seconds to compile.

The generated LLVM IR is identical, except for the embedded LLVM commit 

   -- Bob

More information about the llvm-commits mailing list