projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 7b17f4e) | patch
Add new operations affine.dma_start and affine.dma_wait which take affine maps for...
authorAndy Davis <andydavis@google.com>
Fri, 28 Jun 2019 20:31:31 +0000 (13:31 -0700)
committerA. Unique TensorFlower <gardener@tensorflow.org>
Fri, 28 Jun 2019 20:31:57 +0000 (13:31 -0700)
commit6c68596aeedbdcd61570dffcfa0e4cd4e3c5489c
treeb544d8d05e2bd1e2ad8863d5110167fe85efece3tree | snapshot
parent7b17f4e64778513f5b8c4c8a3572e7b1c744f278commit | diff
Add new operations affine.dma_start and affine.dma_wait which take affine maps for indexing memrefs by construction.
These ops are analogues of the current standard ops dma_start/wait, with the exception that the memref operands are affine expressions of loop IVs and symbols (analogous to affine.load/store).
The addition of these operations will enable changes to affine transformation and analysis passes which operate on memory dereferencing operations.

PiperOrigin-RevId: 255658382
mlir/include/mlir/AffineOps/AffineOps.h diff | blob | history
mlir/lib/AffineOps/AffineOps.cpp diff | blob | history
mlir/test/AffineOps/dma.mlir [new file with mode: 0644] blob
Domain: System / Toolchain;