projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 5ab1a8a) | patch
[NFC][AArch64]Precommit test cases to show ORR is better when one operand is a shift...
authorMingming Liu <mingmingl@google.com>
Wed, 9 Nov 2022 06:28:07 +0000 (22:28 -0800)
committerMingming Liu <mingmingl@google.com>
Fri, 11 Nov 2022 22:01:02 +0000 (14:01 -0800)
commit3acbaddd43d8340257ad79a3514865c9758612c2
tree21d07b13481b40040c48270777acab126a53172ctree | snapshot
parent5ab1a8aef454b82c409587aa849f4a1d2f1a86f9commit | diff
[NFC][AArch64]Precommit test cases to show ORR is better when one operand is a shift of the other operand

In `bfi-not-orr` tests, bfi/bfxil are better since they simplifies away two instructions (extracting bits into destination directly)

In `orr-not-bfi` tests, orr is better since both orr and bfm would simplify away one instruction (the shl node), orr has higher throughput and shorter latency than bfm.
llvm/test/CodeGen/AArch64/bitfield-insert.ll diff | blob | history
Domain: System / Toolchain;