polymer
polymer copied to clipboard
Bridging polyhedral analysis tools to the MLIR framework
I wonder if I can specify the number of cores as one of the optimization options when using the parallel polymer. I only see this option in PlutoTransform.cc ``` Option...
I meet this error when I tried to optimize the matmul example. ``` Dep type: 4 polymer-opt: osl_pluto.c:1562: deps_read: Assertion `0 && "unexpected dependence type"' failed. PLEASE submit a bug...
Does this script is still valid? It checks out version of Polygeist from polygeist-version.txt which seems to be incompatible and miss some required headers so I'm getting the error ```...
```mlir func private @S0(%A: memref, %t: index, %i: index) attributes {scop.stmt} { %cst = constant 0.25: f32 %cst2 = constant 2.0 : f32 %0 = affine.load %A[%t mod 2, %i...
Hi, after building the polymer-opt with the following commit, I try to execute (matmul.mlir is from /test/archive/polymer-translate/export-scop/matmul.mlir) ./bin/polymer-opt -pluto-opt matmul.mlir It promotes the following error: ./matmul.mlir:4:8: error: custom op 'alloc'...
- [x] Include `isl` as a submodule - [x] Add its build step in CMake - [x] Ask Pluto to use this new `isl` version instead of its bundled one
This PR tries to migrate things from 26122d9
This will be useful for further optimisation concentrating on the point loop bands. ```sh cd build ./bin/polymer-opt ../example/polybench/EXTRALARGE/gemm/gemm.mlir \ -reg2mem \ -extract-scop-stmt \ -pluto-opt \ -canonicalize \ -annotate-point-loops \ -extract-point-loops...
Currently covered by 25afb44482845e39c5157145fc98a249b22ff381 Approach: IVs directly used by a scop.stmt caller are point loop IVs. This is assumed by the Pluto algorithm. - [ ] Can it work before...
## if-else block ### Resolve the domain for the else block This should be fixed in the upstream (`getIndexSet` method) ### multiple ifs and for-within-if See the `export-scop/load-store-if.mlir` test case.