[PATCH] D27203: [X86] Prefer reduced width multiplication over pmulld on Silvermont

[Zvi Rackover via Phabricator via llvm-commits]

zvi added inline comments.


================
Comment at: lib/Target/X86/X86Subtarget.cpp:232
+
+  assert((!isPMULLDSlow() || (isPMULLDSlow() && hasSSE41())) &&
+         "Feature Slow PMULLD can only be set on a subtarget with SSE4.1");
----------------
mkuper wrote:
> (!isPMULLDSlow() || hasSSE41()) is good enough, logically speaking.
Good catch


================
Comment at: test/CodeGen/X86/slow-pmulld.ll:1
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i386-unknown-unknown -mcpu=silvermont | FileCheck %s --check-prefix=CHECK32
----------------
mkuper wrote:
> Please add a check that we *do* generate a pmulld for non-slow targets.
> (Or, if we already have such a test, merge this one into that).
Ok, I will add tests for SSE4.1 targets w/o the slowpmulld feature


Repository:
  rL LLVM

https://reviews.llvm.org/D27203