[llvm] [AArch64] Avoid ld4 lowering for deinterleave4 feeding uitofp (PR #200763)

Kamlesh Kumar via llvm-commits llvm-commits at lists.llvm.org
Mon Jun 1 06:01:32 PDT 2026


kamleshbhalui wrote:

> > > Can you provide any performance data demonstrating any improvements with this change?
> > 
> > 
> > For imagick_r It improves performance by ~14%.
> 
> Just for reference, this is a SPEC2017 benchmark. Did you find any performance impact in any other benchmarks besides imagick?

yes it is SPEC2017, after the changes other impacted benchmarks are.

regressed
510.parest_r -> -0.77%
527.cam4_r -> -0.41%
511.povray_r -> -0.23%

Some improved
549.fotonik3d_r -> +1.43%
526.blender_r -> +1.09%
508.namd_r -> +0.16% 
503.bwaves_r -> +0.14%

https://github.com/llvm/llvm-project/pull/200763


More information about the llvm-commits mailing list