[MPS] Fix bfloat to complex casts #137070
Conversation
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/137070
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit ebf87c7 with merge base dfe1d45 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
pytorch-bot
bot
added
ciflow/mps
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
| if (dstComplex) { | ||
| // TODO: Document why explicit cast is needed only for bfloat types | ||
| if (dtypeSrc == "bfloat") { | ||
| return dtypeDst + "(float(x), 0.0)"; |
There was a problem hiding this comment.
| return dtypeDst + "(float(x), 0.0)"; | |
| return dtypeDst + "(float(x), 0.0)"; |
| return "bfloat(x)"; | ||
| } | ||
| return "(x)"; | ||
| return dtypeSrc == "bfloat" ? dtypeDst + "(x)" : "(x)"; |
There was a problem hiding this comment.
| return dtypeSrc == "bfloat" ? dtypeDst + "(x)" : "(x)"; | |
| return dtypeSrc == "bfloat" ? dtypeDst + "(x)" : "(x)"; |
For Metal cast ops to comple, one need to explicitly cast to/from `bfloat` unlike for other dtypes Tested in #136987 [ghstack-poisoned]
| const bool srcComplex = dtypeSrc[dtypeSrc.size() - 1] == '2'; | ||
| const bool dstComplex = dtypeDst[dtypeDst.size() - 1] == '2'; | ||
| if (dstComplex) { | ||
| // TODO: Document why explicit cast is needed only for bfloat types |
There was a problem hiding this comment.
I guess this should be done before merging?
There was a problem hiding this comment.
I just copy-n-pasted TODO from another line. I guess I'll need a help from @kulinseth because I could not find anything in the https://developer.apple.com/metal/Metal-Shading-Language-Specification.pdf that talks about it
|
@pytorchbot merge -f "Lint + MPS tests are green" |
Merge startedYour change will be merged immediately since you used the force (-f) flag, bypassing any CI checks (ETA: 1-5 minutes). Please use Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
By even further reducing precisions of imprecise FP16 ops, introducing new BF16_LOW_PRECISION_OPS category and marking BF16 tests as xfail for `divfloor_rounding`, `floor_divide` and `remainder`. I guess the nature of low-precision results, is that MPSGraph, unlike the rest of the PyTorch does not do accumulation over fp32 for reduction operations Pull Request resolved: #136987 Approved by: https://github.com/albanD ghstack dependencies: #137070
For Metal cast ops to comple, one need to explicitly cast to/from `bfloat` unlike for other dtypes Tested in pytorch#136987 Pull Request resolved: pytorch#137070 Approved by: https://github.com/Skylion007
By even further reducing precisions of imprecise FP16 ops, introducing new BF16_LOW_PRECISION_OPS category and marking BF16 tests as xfail for `divfloor_rounding`, `floor_divide` and `remainder`. I guess the nature of low-precision results, is that MPSGraph, unlike the rest of the PyTorch does not do accumulation over fp32 for reduction operations Pull Request resolved: pytorch#136987 Approved by: https://github.com/albanD ghstack dependencies: pytorch#137070
4 participants
Stack from ghstack (oldest at bottom):
For Metal cast ops to comple, one need to explicitly cast to/from
bfloatunlike for other dtypesTested in #136987