Add mechanism for remapping device-specific module imports #4539
Conversation
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
Contributor
Author
|
I initially tried to tackle #4509 with a purely "userspace" solution by changing |
7 tasks
Contributor
Author
|
fixed test |
This is motivated by triton-lang#4509. The crux of the problem is that the Triton code generator needs to inspect a function's arguments / attributes / types in order to determine how it should be called. This meant that "implementation details" like whether a function is a builtin needed to be exposed in the "interface" `tl.extra.libdevice` module, instead of just residing in `tl.extra.cuda.libdevice`. Moreover, this meant that libdevice functions marked as @core.extern in the interface could not be implemented via JitFunctions. Allowing each backend to provide its own module map solves this problem as the code generator can inspect the actual function implementation.
ThomasRaoux
approved these changes
Aug 22, 2024
Jokeren
pushed a commit
that referenced
this pull request
Aug 24, 2024
This is motivated by #4509. The crux of the problem is that the Triton code generator needs to inspect a function's arguments / attributes / types in order to determine how it should be called. This meant that "implementation details" like whether a function is a builtin needed to be exposed in the "interface" `tl.extra.libdevice` module, instead of just residing in `tl.extra.cuda.libdevice`. Moreover, this meant that libdevice functions marked as `@core.extern` in the interface could not be implemented via JitFunctions. Allowing each backend to provide its own module map solves this problem as the code generator can inspect the actual function implementation.
pytorchmergebot
pushed a commit
to pytorch/pytorch
that referenced
this pull request
Aug 31, 2024
…#134774) In triton-lang/triton#4539 the `make_ir` API was modified to accept a new `module_map` parameter. Update the Inductor callsite accordingly, preserving backwards compatibility following the existing code. Fixes #134674 Pull Request resolved: #134774 Approved by: https://github.com/EikanWang, https://github.com/zou3519, https://github.com/jansel
tolleybot
pushed a commit
to tolleybot/pytorch
that referenced
this pull request
Sep 14, 2024
…pytorch#134774) In triton-lang/triton#4539 the `make_ir` API was modified to accept a new `module_map` parameter. Update the Inductor callsite accordingly, preserving backwards compatibility following the existing code. Fixes pytorch#134674 Pull Request resolved: pytorch#134774 Approved by: https://github.com/EikanWang, https://github.com/zou3519, https://github.com/jansel
Chao1Han
pushed a commit
to Chao1Han/pytorch
that referenced
this pull request
Sep 20, 2024
…pytorch#134774) In triton-lang/triton#4539 the `make_ir` API was modified to accept a new `module_map` parameter. Update the Inductor callsite accordingly, preserving backwards compatibility following the existing code. Fixes pytorch#134674 Pull Request resolved: pytorch#134774 Approved by: https://github.com/EikanWang, https://github.com/zou3519, https://github.com/jansel
davidberard98
added a commit
to davidberard98/triton
that referenced
this pull request
Nov 6, 2024
Context: In `CodeGenerator.__init__`, globals for a given triton function are modified to handle remapping the libdevice module to cuda or hip (from triton-lang#4539). In particular, this logic: ```python for k, v in gscope.items(): # gscope is a dict of fn.__globals__ ... self.gscope[k] = getattr(module_map[module_name], k) ``` was failing if you do this in the global scope: `from triton.language.extras.libdevice import fast_dividef as my_fast_dividef`.
peterbell10
pushed a commit
that referenced
this pull request
Nov 6, 2024
…5081) Context: in `CodeGenerator.__init__`, globals for a given triton function are modified to handle remapping the libdevice module to cuda or hip (from #4539). In particular, this logic: ```python for k, v in gscope.items(): # gscope is a dict of fn.__globals__ ... self.gscope[k] = getattr(module_map[module_name], k) ``` was failing if you do this in the global scope: `from triton.language.extras.libdevice import fast_dividef as my_fast_dividef`.
bertmaher
pushed a commit
that referenced
this pull request
Nov 6, 2024
…5081) Context: in `CodeGenerator.__init__`, globals for a given triton function are modified to handle remapping the libdevice module to cuda or hip (from #4539). In particular, this logic: ```python for k, v in gscope.items(): # gscope is a dict of fn.__globals__ ... self.gscope[k] = getattr(module_map[module_name], k) ``` was failing if you do this in the global scope: `from triton.language.extras.libdevice import fast_dividef as my_fast_dividef`.
Luosuu
pushed a commit
to Luosuu/triton
that referenced
this pull request
Nov 13, 2024
…riton-lang#5081) Context: in `CodeGenerator.__init__`, globals for a given triton function are modified to handle remapping the libdevice module to cuda or hip (from triton-lang#4539). In particular, this logic: ```python for k, v in gscope.items(): # gscope is a dict of fn.__globals__ ... self.gscope[k] = getattr(module_map[module_name], k) ``` was failing if you do this in the global scope: `from triton.language.extras.libdevice import fast_dividef as my_fast_dividef`.
bertmaher
pushed a commit
to bertmaher/triton
that referenced
this pull request
Dec 10, 2024
…ng#4539) This is motivated by triton-lang#4509. The crux of the problem is that the Triton code generator needs to inspect a function's arguments / attributes / types in order to determine how it should be called. This meant that "implementation details" like whether a function is a builtin needed to be exposed in the "interface" `tl.extra.libdevice` module, instead of just residing in `tl.extra.cuda.libdevice`. Moreover, this meant that libdevice functions marked as `@core.extern` in the interface could not be implemented via JitFunctions. Allowing each backend to provide its own module map solves this problem as the code generator can inspect the actual function implementation.
liuyunqi20
pushed a commit
to flagos-ai/FlagTree
that referenced
this pull request
Oct 21, 2025
…5081) Context: in `CodeGenerator.__init__`, globals for a given triton function are modified to handle remapping the libdevice module to cuda or hip (from triton-lang/triton#4539). In particular, this logic: ```python for k, v in gscope.items(): # gscope is a dict of fn.__globals__ ... self.gscope[k] = getattr(module_map[module_name], k) ``` was failing if you do this in the global scope: `from triton.language.extras.libdevice import fast_dividef as my_fast_dividef`.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
This is motivated by #4509. The crux of the problem is that the Triton code generator needs to inspect a function's arguments / attributes / types in order to determine how it should be called. This meant that "implementation details" like whether a function is a builtin needed to be exposed in the "interface"
tl.extra.libdevicemodule, instead of just residing intl.extra.cuda.libdevice. Moreover, this meant that libdevice functions marked as@core.externin the interface could not be implemented via JitFunctions.Allowing each backend to provide its own module map solves this problem as the code generator can inspect the actual function implementation.