[PATCH] D70157: Align branches within 32-Byte boundary

[Chandler Carruth via Phabricator via llvm-commits]

chandlerc added a reviewer: chandlerc.
chandlerc added a comment.

Thanks for sending this out!

I've made a minor comment on the flag structure, but my bigger question would be: can you summarize the performance hit and code size hit you've seen across some benchmarks? SPEC and the LLVM test suite would be nice, and maybe code size alone for some large binary like Chrome, Firefox, or Safari?

I'd be particularly interested in comparing the performance & code size hit incurred by the suggested "fused,jcc,jmp" set compared to all (adding call, ret, and indirect). If there is a big drop in either, I'd love to know which of these incurs the large drop.

Thanks,
-Chandler



================
Comment at: llvm/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp:88
+    SmallVector<StringRef, 6> BranchTypes;
+    StringRef(Val).split(BranchTypes, '-', -1, false);
+    for (auto BranchType : BranchTypes) {
----------------
I feel like a comma-separated list would be a bit more clear...


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D70157/new/

https://reviews.llvm.org/D70157