Frost Mitchell issues

Results 4 issues of


                                            Frost Mitchell

[inductor][cpp] Add BMM kernel template for autotuning

This PR adds the Cpp template for BMM, for FP32, FP16, and BF16. See #125683 for more background. 1. Adds `CppBmmTemplate` class which inherits from `CppPackedGemmTemplate`. Given a number of...

oncall: distributed

module: cpu

module: mkldnn

open source

module: amp (automated mixed precision)

NNC

ciflow/trunk

release notes: quantization

release notes: releng

module: inductor

module: dynamo

ciflow/inductor

module: distributed_checkpoint

Including XPU and CUDA in ProfilerActivity causes XPU profiling to be ignored

### 🐛 Describe the bug When using the Pytorch profiler, the current doc page shows a method of profiling on accelerators by including both `ProfilerActivity.CUDA` and `ProfilerActivity.XPU` in the activities...

oncall: profiler

module: xpu

[Don't merge] [Kineto][submodule] Update kineto pin for XPU toggle feature

Part of #154898 Update kineto submodule Summary: We add the toggleCollectionDynamic functionality to XPUPTI in Kineto, so profiler can be enabled/disabled dynamically.

open source

ciflow/trunk

topic: not user facing

ciflow/xpu

Add XPU, CPU Autocast for torchvision ops

This PR enables autocast for ops that do not yet have autocast support (`roi_pool`, `ps_roi_align`, `ps_roi_pool`, `deform_conv2d`) Tests have been updated to check autocast on CPU. In regards to the...

cla signed