projects / oota-llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 6178e5f) | patch
Correct cost model for vector shift on AVX2
authorMichael Liao <michael.liao@intel.com>
Wed, 20 Mar 2013 22:01:10 +0000 (22:01 +0000)
committerMichael Liao <michael.liao@intel.com>
Wed, 20 Mar 2013 22:01:10 +0000 (22:01 +0000)
commitf74e9bf650d7c40d595d3bb60e3c901e2bccec4b
tree26fc6ab996a2aa6e0516091cb0f5bee4483168c7tree | snapshot
parent6178e5f50c0c8be26913cd93238a5035a39cdf37commit | diff
Correct cost model for vector shift on AVX2

- After moving logic recognizing vector shift with scalar amount from
  DAG combining into DAG lowering, we declare to customize all vector
  shifts even vector shift on AVX is legal. As a result, the cost model
  needs special tuning to identify these legal cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177586 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/X86/X86TargetTransformInfo.cpp diff | blob | history
test/Analysis/CostModel/X86/arith.ll diff | blob | history
C/C++ LLVM-based compilers that forbids OOTA behaviors