projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: c1a2985) | patch
[mlir][linalg][bufferize] Allow non-tensor mappings in BufferizationState
authorMatthias Springer <springerm@google.com>
Mon, 15 Nov 2021 10:34:58 +0000 (19:34 +0900)
committerMatthias Springer <springerm@google.com>
Mon, 15 Nov 2021 10:40:30 +0000 (19:40 +0900)
commit8835a1924e37fe8029f9af7aacd9eb81c2848401
tree4afa68a5781b898cc1e55a72178c7cd1338df1e6tree | snapshot
parentc1a2985d7f4edb0d1ffeda512f84282e60eae677commit | diff
[mlir][linalg][bufferize] Allow non-tensor mappings in BufferizationState

This change makes it possible to set up custom mappings in a PostAnalysisStep. Some users of Comprehensive Bufferize have custom tensor types and it is most convenient to just reuse the same bvm.

Also add some more assertions.

Differential Revision: https://reviews.llvm.org/D113726
mlir/include/mlir/Dialect/Linalg/ComprehensiveBufferize/BufferizableOpInterface.h diff | blob | history
mlir/lib/Dialect/Linalg/ComprehensiveBufferize/BufferizableOpInterface.cpp diff | blob | history
mlir/lib/Dialect/Linalg/ComprehensiveBufferize/ComprehensiveBufferize.cpp diff | blob | history
Domain: System / Toolchain;