Add all-in-one pass pipeline #75
Conversation
|
WIP: Depending on #70 and support of openmp in gcc. |
Menooker
changed the title
| populateVectorPasses(pm); | ||
| // back-end, arith/math/vector/memref dialects | ||
| populateBufferizationPasses(pm); | ||
| // REMOVE this pass after the TensorPasses are added. Currently we add this |
There was a problem hiding this comment.
Can we add "TODO" to highlight the remaining items to better distinguish with other common comments?
| int main(int argc, char **argv) { | ||
| // keeps GCCPURuntime linked | ||
| gc_runtime_keep_alive = 0; |
There was a problem hiding this comment.
Yes we do, to make sure gc_runtime is linked.
| @@ -0,0 +1,23 @@ | |||
| // RUN: gc-opt %s --gc-cpu-pipeline | gc-cpu-runner -e main -entry-point-result=void | FileCheck %s | |||
There was a problem hiding this comment.
We might need to add some complex example to cover the default pipeline.
There was a problem hiding this comment.
@LongshengDu Please help to add some test cases and check if the pipeline works for the current linalgx and onednngraph dialect. Maybe in another PR, or in this one?
|
@AndreyPavlenko, please take a look |
OK, changed the comments in pipeline to "todo"s.
I have changed most of them to "todo"s. They mark the position where not-yet-implemented MLIR passes should be. The developers of that pass can know where to add to the pipeline in the future. BTW, this PR depends on #79. It contains code of that PR. |
No description provided.