projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 5d24089) | patch
[SLP] use reduction kind's opcode to create new instructions; NFC
authorSanjay Patel <spatel@rotateright.com>
Wed, 6 Jan 2021 19:10:40 +0000 (14:10 -0500)
committerSanjay Patel <spatel@rotateright.com>
Wed, 6 Jan 2021 19:37:44 +0000 (14:37 -0500)
commit4c022b5a41dee998ae50cdad4e8b6548acbeee9f
tree5fb9b182a4b357f4edfd0462767c12bbfd8d6b49tree | snapshot
parent5d24089a7001e9fb4c0e665e93312916d88aaef9commit | diff
[SLP] use reduction kind's opcode to create new instructions; NFC

Similar to 5a1d31a28 -
This should be no-functional-change because the reduction kind
opcodes are 1-for-1 mappings to the instructions we are matching
as reductions. But we want to remove the need for the
`OperationData` opcode field because that does not work when
we start matching intrinsics (eg, maxnum) as reduction candidates.
llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp diff | blob | history
Domain: System / Toolchain;