strisunshinewentingwang/pytorch

Pull Requests Commits

Allow using nvFuser on CUDA extension

zasdfgbnm committed 2 years ago

4318d9d0

adding permutation for dropout (#1691)

jjsjann123 committed 2 years ago

Verified 64191c9e

Fix usage of ArrayRef (#1697)

naoyam committed 2 years ago

Verified 9fd912be

Fix printing of vector (#1693)

naoyam committed 2 years ago

Verified cd52c3c8

Only extract bias out of linear layer if input ndims >= 4 (#1679)

csarofeen committed 2 years ago

Verified 25aad029

Opinfo patch (#1682)

jjsjann123 committed 2 years ago

Verified 04c5ac65

Patch sync insertion for redundant predicated writes (#1684)

shmsong committed 2 years ago

Verified f9132b7e

Use PYTORCH_NVFUSER_DISABLE in tests (#1688)

zasdfgbnm committed 2 years ago

Verified f952365e

Use int rather than int64_t for hoisted index values when the kernel (#1685)

naoyam committed 2 years ago

Verified 1a695ede

Profile ivalue multiple (#1677)

jjsjann123 committed 2 years ago

Verified 078fa19d

Memory format propagation parser patch (#1675)

jjsjann123 committed 2 years ago

Verified f89f2d42

Use fp64 reference for bfloat16 unary op tests (#1674)

shmsong committed 2 years ago

Verified a18e1094

Rework pwise scheduler for better broadcasting perf on TIMM VIT. (#1664)

csarofeen committed 2 years ago

Verified 26fe9c8c

Simultaneous cleanup conflicting. (#1672)

csarofeen committed 2 years ago

Verified 4153390d

Make sure vectorization and inlining is consistent in the reduction/norm kernels. (#1665)

csarofeen committed 2 years ago

Verified 374e2f78

Add debug option PYTORCH_NVFUSER_DUMP=perf_debug_verbose (#1657)

csarofeen committed 2 years ago

Verified 256a2f5e

Small IrCloner fix (#1668)

csarofeen committed 2 years ago

Verified 7c7ea66a

Make sure transform propagate always takes the most replayed replay. (#1662)

csarofeen committed 2 years ago

Verified 440e307d

Add some TIMM micro-benchmarks and some minor benchmark fixes (#1656)

csarofeen committed 2 years ago

Verified 8dd3fb1f

Code clean up and minor fix on dynamic shared alignment (#1670)

shmsong committed 2 years ago

Verified 8d2495b4

Cleanup some warnings. (#1669)

csarofeen committed 2 years ago

Verified 315a8d14

Cleanup heuristic structs. (#1663)

csarofeen committed 2 years ago

Verified 9dddfb18

Avoid redundant writes on global memory. (#1661)

csarofeen committed 2 years ago

Verified f07f3f22

Consolidate all the PYTORCH_NVFUSER_DISABLE_ options like PYTORCH_NVFUSER_DUMP (#1666)

naoyam committed 2 years ago

Verified 8888c631

Emit thread predicates for vectorized instructions. (#1660)

csarofeen committed 2 years ago

Verified c2495b64

Add returns and indenting to LHS of broadcast and reduction ops. (#1658)

csarofeen committed 2 years ago

Verified e42724cf

Print ca map (#1653)

naoyam committed 2 years ago

Verified f2ef94b5

removing workflow pull (#1654)

jjsjann123 committed 2 years ago

Verified 6f7b599d

Quick patch on pred removal pass (#1642)

shmsong committed 2 years ago

Verified 16192567

Merge pull request #1649 from csarofeen/master_bump_0501

csarofeen committed 2 years ago

Verified 5bbbca56

Older