Add getMemOpBaseRegImmOfsWidth to enable determining independence during MiSched.
Details
Diff Detail
Event Timeline
Adding simple test case to verify effect on scheduling DAG due to areMemAccessesTriviallyDisjoint change.
@mcrosier: Yes indeed, funny things can happen if these are wrong :). Added a IR test to test some of the different code paths in areMemAccessesTriviallyDisjoint.
| lib/Target/Lanai/LanaiInstrInfo.cpp | ||
|---|---|---|
| 110 | std::min? | |
| 111 | std::max? | |
| 113 | Please clang-format this. | |
| 363 | clang-format this | |
| 369–388 | clang-format this too. | |
| 398–412 | And this. | |
| 412 | Extra semicolon. | |
| test/CodeGen/Lanai/lanai-misched-trivial-disjoint.ll | ||
| 2 | I'd recommend piping the file in via stdin: llc < %s -mtriple=lanai-unknown-unknown -debug-only=misched -o /dev/null 2>&1 This prevents the path to the file from showing up in any of LLVM's output. Occasionally, this causes problems if content that a CHECK line is searching for can be found inside of a path. | |
| 7 | I'd remove this comment | |
Updated, thanks
| test/CodeGen/Lanai/lanai-misched-trivial-disjoint.ll | ||
|---|---|---|
| 2 | Fixed thanks. Didn't think about check matching on the path. | |
std::min?