Chlo decompose patterns

[linuxlonelyeagle]

I had added Chlo conversion to Stablehlo pass, and Canonicalize @GleasonK .

[linuxlonelyeagle]

Here's some more trivia, should I fix them?

/my_stablehlo/stablehlo/transforms/StablehloCanonicalize.cpp:45:13: warning: comparison of integers of different signs: 'std::tuple_element<0, std::tuple<unsigned long, const long long &>>::type' (aka '__type_pack_element<0UL, unsigned long, const long long &>') and 'const long long' [-Wsign-compare]
    if (idx != value) {
        ~~~ ^  ~~~~~
/my_stablehlo/stablehlo/transforms/StablehloCanonicalize.cpp:692:53: warning: comparison of integers of different signs: 'size_t' (aka 'unsigned long') and 'int64_t' (aka 'long long') [-Wsign-compare]
        op.getKnownNonexpandingDimensions()->size() != resultType.getRank()) {
        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ^  ~~~~~~~~~~~~~~~~~~~~
/my_stablehlo/stablehlo/transforms/StablehloCanonicalize.cpp:53:13: warning: unused function 'isIotaRange' [-Wunused-function]
static bool isIotaRange(ElementsAttr attr) {

[GleasonK]

Thanks for upstreaming these passes!! 🚀

Minor feedback on file renaming / pass naming, and lets split off the canonicalize pass to a separate PR and give it a simple test point to make sure its working, keep this PR specific to CHLO decomposition. I can add more tests once these are submitted.

[GleasonK]

Here's some more trivia, should I fix them?

/my_stablehlo/stablehlo/transforms/StablehloCanonicalize.cpp:45:13: warning: comparison of integers of different signs: 'std::tuple_element<0, std::tuple<unsigned long, const long long &>>::type' (aka '__type_pack_element<0UL, unsigned long, const long long &>') and 'const long long' [-Wsign-compare]
    if (idx != value) {
        ~~~ ^  ~~~~~
/my_stablehlo/stablehlo/transforms/StablehloCanonicalize.cpp:692:53: warning: comparison of integers of different signs: 'size_t' (aka 'unsigned long') and 'int64_t' (aka 'long long') [-Wsign-compare]
        op.getKnownNonexpandingDimensions()->size() != resultType.getRank()) {
        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ^  ~~~~~~~~~~~~~~~~~~~~
/my_stablehlo/stablehlo/transforms/StablehloCanonicalize.cpp:53:13: warning: unused function 'isIotaRange' [-Wunused-function]
static bool isIotaRange(ElementsAttr attr) {

As mentioned above, lets split canonicalize to a separate PR. Then yes, it would be great to fix these if they're simple.

[GleasonK]

Looks like we need to address those compilation warnings you mentioned to get CI working. If you can get CMakeBuild working, I'll send you a patch to make the Bazel CI work. Holding off on sending that patch until existing feedback is addressed / CMake is working to avoid any merge conflicts.

[linuxlonelyeagle]

Looks like we need to address those compilation warnings you mentioned to get CI working. If you can get CMakeBuild working, I'll send you a patch to make the Bazel CI work. Holding off on sending that patch until existing feedback is addressed / CMake is working to avoid any merge conflicts.

ok,It may be a little late to address this problem,I read the error reported by ci and I know what the problem is.

[GleasonK]

LGTM! Thanks again for doing this! I'll add some extra test cases and might so some additional refactoring as I start trying to integrate this pass elsewhere. In the near future I plan to migrate the CHLO->MHLO pass to use CHLO->StableHLO->MHLO, this PR was an important first step. 🚀

[linuxlonelyeagle]

LGTM! Thanks again for doing this! I'll add some extra test cases and might so some additional refactoring as I start trying to integrate this pass elsewhere. In the near future I plan to migrate the CHLO->MHLO pass to use CHLO->StableHLO->MHLO, this PR was an important first step. 🚀

I appreciate everything you do. I hope to continue working together in the future 😉.