CPU Optimizations for FP8 #2559
+500
−138
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -76,8 +76,6 @@ def get_tensor_device(tensor: torch.Tensor) -> int: | |
| The order of attributes checked is important to also | ||
| minimize overhead. | ||
| """ | ||
| if hasattr(tensor, "_rowwise_data") and tensor._rowwise_data is not None: | ||
| return tensor._rowwise_data.device.index | ||
| if hasattr(tensor, "_columnwise_data") and tensor._columnwise_data is not None: | ||
|
|
@@ -86,6 +84,8 @@ def get_tensor_device(tensor: torch.Tensor) -> int: | |
| return tensor._data.device.index | ||
| if hasattr(tensor, "_transpose") and tensor._transpose is not None: | ||
| return tensor._transpose.device.index | ||
| if hasattr(tensor, "device"): | ||
| return tensor.device.index | ||
| return torch.cuda.current_device() | ||
|
|
||
|
|
||
|
|
||
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
|
|
@@ -35,9 +35,9 @@ PyTypeObject *Float8BlockwiseQuantizerClass = nullptr; | |
| PyTypeObject *NVFP4TensorPythonClass = nullptr; | ||
| PyTypeObject *NVFP4TensorStoragePythonClass = nullptr; | ||
| PyTypeObject *NVFP4QuantizerClass = nullptr; | ||
| std::once_flag extension_init_flag; | ||
|
|
||
| void init_float8_extension() { | ||
| auto fp8_module = py::module_::import("transformer_engine.pytorch.tensor.float8_tensor"); | ||
| Float8QuantizerClass = | ||
| reinterpret_cast<PyTypeObject *>(PyObject_GetAttrString(fp8_module.ptr(), "Float8Quantizer")); | ||
|
|
@@ -54,7 +54,6 @@ void init_float8_extension() { | |
| } | ||
|
|
||
| void init_mxfp8_extension() { | ||
| auto fp8_module = py::module_::import("transformer_engine.pytorch.tensor.mxfp8_tensor"); | ||
| MXFP8QuantizerClass = | ||
| reinterpret_cast<PyTypeObject *>(PyObject_GetAttrString(fp8_module.ptr(), "MXFP8Quantizer")); | ||
|
|
@@ -69,7 +68,6 @@ void init_mxfp8_extension() { | |
| } | ||
|
|
||
| void init_float8blockwise_extension() { | ||
| auto fp8_module = | ||
| py::module_::import("transformer_engine.pytorch.tensor.float8_blockwise_tensor"); | ||
| auto fp8_base_module = py::module_::import( | ||
|
|
@@ -90,7 +88,6 @@ void init_float8blockwise_extension() { | |
| } | ||
|
|
||
| void init_nvfp4_extensions() { | ||
| auto nvfp4_module = py::module_::import("transformer_engine.pytorch.tensor.nvfp4_tensor"); | ||
| NVFP4QuantizerClass = reinterpret_cast<PyTypeObject *>( | ||
| PyObject_GetAttrString(nvfp4_module.ptr(), "NVFP4Quantizer")); | ||
|
|
@@ -105,10 +102,12 @@ void init_nvfp4_extensions() { | |
| } | ||
|
|
||
| void init_extension() { | ||
| std::call_once(extension_init_flag, []() { | ||
| init_float8_extension(); | ||
| init_mxfp8_extension(); | ||
| init_float8blockwise_extension(); | ||
| init_nvfp4_extensions(); | ||
| }); | ||
vthumbe1503 marked this conversation as resolved.
Show resolved
Hide resolved
|
||
| } | ||
|
Comment on lines
+105
to
111
Contributor
There was a problem hiding this comment. using if these functions are ever called directly (not through |
||
|
|
||
| } // namespace transformer_engine::pytorch | ||
|
|
||
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
caching
nvte_is_non_tn_fp8_gemm_supported()result avoids redundant calls throughout the GEMM configuration for both A and B matrices. Clean optimization with proper scopingNote: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!