maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
301,036 Commits 23 Branches 0 Tags 2.2 GiB
C++ 39.1%
LLVM 35.7%
C 12.1%
Assembly 8.9%
Python 1.1%
Other 2.7%
Go to file
Craig Topper 06aea1720a [X86] Move promotion of vector and/or/xor from legalization to DAG combine 
Summary:
I've noticed that the bitcasts we introduce for these make computeKnownBits and computeNumSignBits not work well in LegalizeVectorOps. LegalizeVectorOps legalizes bottom up while LegalizeDAG legalizes top down. The bottom up strategy for LegalizeVectorOps means operands are legalized before their uses. So we promote and/or/xor before we legalize the operands that use them making computeKnownBits/computeNumSignBits in places like LowerTruncate suboptimal. I looked at changing LegalizeVectorOps to be top down as well, but that was more disruptive and caused some regressions. I also looked at just moving promotion of binops to LegalizeDAG, but that had a few issues one around matching AND,ANDN,OR into VSELECT because I had to create ANDN as vXi64, but the other nodes hadn't legalized yet, I didn't look too hard at fixing that.

This patch seems to produce better results overall than my other attempts. We now form broadcasts of constants better in some cases. For at least some of them the AND was being introduced in LegalizeDAG, promoted to vXi64, and the BUILD_VECTOR was also legalized there. I think we got bad ordering of that. Now the promotion is out of the legalizer so we handle this better.

In the longer term I think we really should evaluate whether we should be doing this promotion at all. It's really there to reduce isel pattern count, but I'm wondering if we'd be better served just eating the pattern cost or doing C++ based isel for vector and/or/xor in X86ISelDAGToDAG. The masked and/or/xor will definitely be difficult in patterns if a bitcast gets between the vselect and the and/or/xor node. That becomes a lot of permutations to cover.

Reviewers: RKSimon, spatel

Reviewed By: RKSimon

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D53107

llvm-svn: 344487
 		2018-10-15 01:51:58 +00:00
clang [Hexagon] Update tests account for non-hardcoded linker name. 2018-10-14 17:51:36 +00:00
clang-tools-extra Move some helpers from the global namespace into anonymous ones. 2018-10-13 22:18:22 +00:00
compiler-rt [libfuzzer][Windows] Silence linker warning in unittest 2018-10-14 17:07:40 +00:00
debuginfo-tests Revert "(Retry) Add a basic integration test for C++ smart pointers" 2018-08-20 19:53:33 +00:00
libclc configure: Rework support for gfx9+ devices that were added post LLVM 3.9 2018-09-15 22:02:01 +00:00
libcxx One more local type warning removed from the tests. NFC 2018-10-12 21:59:32 +00:00
libcxxabi Override libcxxabi's .clang-format in the demangle directory 2018-10-11 23:30:56 +00:00
libunwind [CMake] Link to compiler-rt if LIBUNWIND_USE_COMPILER_RT is ON. 2018-10-08 18:35:00 +00:00
lld Simplify. NFC. 2018-10-12 22:44:06 +00:00
lldb Fix double import of _lldb module. 2018-10-14 07:24:56 +00:00
llgo Update copyright year to 2018. 2018-06-18 12:22:17 +00:00
llvm [X86] Move promotion of vector and/or/xor from legalization to DAG combine 2018-10-15 01:51:58 +00:00
openmp [OpenMP] Convert KMP_DYNAMIC_LIB to a 0 or 1 guard everywhere 2018-10-05 17:59:39 +00:00
parallel-libs Update copyright year to 2018. 2018-06-18 12:22:17 +00:00
polly Fix another error related to YAML quoting. 2018-10-12 17:28:39 +00:00
README.md
…

README.md

Low Level Virtual Machine (LLVM)

This directory and its subdirectories contain source code for LLVM, a toolkit for the construction of highly optimized compilers, optimizers, and runtime environments.