[vulkan] Add conform API methods to memory allocator to fix block allocations#8130
Merged
derek-gerstmann merged 7 commits intomainfrom Mar 6, 2024
Merged
[vulkan] Add conform API methods to memory allocator to fix block allocations#8130derek-gerstmann merged 7 commits intomainfrom
derek-gerstmann merged 7 commits intomainfrom
Conversation
added 2 commits
March 1, 2024 08:48
Override conform requests for Vulkan memory allocator Cleanup memory requirement constraints for Vulkan Add conform test cases to block_allocator runtime test.
Contributor
|
Some minor compilation errors need fixing |
steven-johnson
approved these changes
Mar 5, 2024
steven-johnson
pushed a commit
that referenced
this pull request
Jun 7, 2024
…ocations (#8130) * Add conform API methods to block and region allocator classes Override conform requests for Vulkan memory allocator Cleanup memory requirement constraints for Vulkan Add conform test cases to block_allocator runtime test. * Clang format/tidy pas * Fix unsigned int comparisons * Clang format pass * Fix other unsigned int comparisons * Fix mismatched template types for max() * Fix whitespace for clang format --------- Co-authored-by: Derek Gerstmann <dgerstmann@adobe.com>
steven-johnson
added a commit
that referenced
this pull request
Jun 24, 2024
* [vulkan] Fix Vulkan SIMT mappings for GPU loop vars. (#8259) * Fix Vulkan SIMT mappings for GPU loop vars. Previous refactoring accidentally used the fully qualified var name rather than the categorized vulkan intrinsic name. * Avoid formatting the GPU kernel to a string for Vulkan (since it's binary SPIR-V needs to remain intact). --------- Co-authored-by: Derek Gerstmann <dgerstmann@adobe.com> Co-authored-by: Steven Johnson <srj@google.com> * Update CodeGen_Vulkan_Dev.cpp * [vulkan] Add conform API methods to memory allocator to fix block allocations (#8130) * Add conform API methods to block and region allocator classes Override conform requests for Vulkan memory allocator Cleanup memory requirement constraints for Vulkan Add conform test cases to block_allocator runtime test. * Clang format/tidy pas * Fix unsigned int comparisons * Clang format pass * Fix other unsigned int comparisons * Fix mismatched template types for max() * Fix whitespace for clang format --------- Co-authored-by: Derek Gerstmann <dgerstmann@adobe.com> * Backport fixes for Vulkan in src/runtime/internal for allocations. --------- Co-authored-by: Derek Gerstmann <derek.gerstmann@gmail.com> Co-authored-by: Derek Gerstmann <dgerstmann@adobe.com>
1 task
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
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.
As reported in #8079 ...
Vulkan's API permits any arbitrary alignment and padding on every Buffer allocation, which means that the memory requirements have to be queried for every block and region prior to binding device memory.
Previously, we were only querying the requirements on the creation of the block, and assuming that all regions (which used the same device memory pool, and physical allocation) would match. This was an invalid assumption, and was only partially fixed in #8087.
This PR introduces a new
conformcallback API which allows block and region allocations to query the memory requirements before the block allocator actually binds memory and creates/splits regions.I also added more test cases in
/test/runtime/runtime_internal_block_allocatorto verify the size, alignment and offsets are constrained as expected.Fixes #8079