derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
31 lines
1.7 KiB
ReStructuredText
31 lines
1.7 KiB
ReStructuredText
.. meta::
|
|
:description: MI200 MFMA behavior specifics
|
|
:keywords: MIOpen, ROCm, API, documentation, MI200, MFMA
|
|
|
|
***********************************************************************************
|
|
MI200 matrix fused multiply-add (MFMA) behavior specifics
|
|
***********************************************************************************
|
|
|
|
On the MI200, ``MFMA_F16``, ``MFMA_BF16``, and ``MFMA_BF16_1K`` flush subnormal input/output data to
|
|
zero. This behavior might affect the convolution operation in certain workloads due to the limited
|
|
exponent range of the half-precision floating-point datatypes.
|
|
|
|
MIOpen offers an alternate implementation for the half-precision datatype via conversion instructions
|
|
to utilize the larger exponent range of the ``BFloat16`` data type, albeit with reduced accuracy. The following
|
|
caveats apply to this alternate implementation:
|
|
|
|
* It's disabled by default in the forward convolution operations.
|
|
|
|
* It's enabled by default in the backward data and backward weights convolution operations.
|
|
|
|
* You can override the default MIOpen behavior by using the ``miopenSetConvolutionAttribute`` API
|
|
call. To use the alternate implementation, pass the convolution descriptor for the appropriate
|
|
convolution operation and the ``MIOPEN_CONVOLUTION_ATTRIB_FP16_ALT_IMPL`` convolution
|
|
attribute (with a non-zero value).
|
|
|
|
* You can also override the behavior using the
|
|
``MIOPEN_DEBUG_CONVOLUTION_ATTRIB_FP16_ALT_IMPL`` environment variable. When set to ``1``,
|
|
``MIOPEN_DEBUG_CONVOLUTION_ATTRIB_FP16_ALT_IMPL`` engages the alternate implementation.
|
|
When set to ``0``, it's disabled. This environment variable impacts the convolution
|
|
operation in all directions.
|