Fusion wgpu compilation cache #1069
Merged
Conversation
This file contains 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
Codecov ReportAttention:
Additional details and impacted files@@ Coverage Diff @@
## main #1069 +/- ##
==========================================
+ Coverage 85.55% 85.66% +0.11%
==========================================
Files 508 509 +1
Lines 53910 54126 +216
==========================================
+ Hits 46122 46368 +246
+ Misses 7788 7758 -30 ☔ View full report in Codecov by Sentry. |
louisfd
requested changes
Dec 15, 2023
burn-wgpu/src/codegen/kernel.rs
Outdated
| impl ElemWiseKernelCodegen<CompilationPhase> { | ||
| /// Compile the kernel into a [compute shader](ComputeShader). | ||
| pub fn compile(self) -> ComputeShader { | ||
| let mut inputs = Vec::with_capacity(self.input_bindings.len()); |
There was a problem hiding this comment.
inputs and outpus can be directly self.input_bindings and self.output_bindings
burn-wgpu/src/codegen/shader.rs
Outdated
| pub enum Visibility { | ||
| Read, | ||
| ReadWrite, | ||
| } | ||
|
|
||
| #[derive(Debug, Clone, Hash, PartialEq, Eq, Copy)] | ||
| #[derive(Debug, Clone, PartialEq, Eq, Copy)] | ||
| pub enum Elem { | ||
| F32, | ||
| #[allow(dead_code)] |
burn-wgpu/src/fusion/cache.rs
Outdated
| items: HashSet<String>, | ||
| } | ||
|
|
||
| pub enum CachedComputeShader { |
burn-wgpu/src/fusion/cache.rs
Outdated
| fn source(&self) -> SourceTemplate { | ||
| match self { | ||
| CachedComputeShader::Cached(_) => { | ||
| panic!("NoSource compute shader should only be used by a higher level cache.") |
| inputs, | ||
| outputs, | ||
| locals, | ||
| operators: self.operators.clone(), | ||
| scalars_f32: self.scalars_f32, | ||
| device: self.device.clone(), | ||
| cache: KernelCache::default(), |
| @@ -1,21 +0,0 @@ | |||
| @group(0) | |||
There was a problem hiding this comment.
Don't forget to delete safe tanh file
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Changes
Add a cache step where the compilation of fused kernels are only executed once, then the kernel ID is used without any overhead.
Refactor the way we create kernels with fusion into a DSL so that we can use it to create WGSL shaders easily, not just when using fusion. I migrated the unary kernels to use the new way, but following PRs will migrate more kernels and remove a lot of code.
Drop
I removed the clamp max and min functions in Fusion as well as in WGPU, since we need to reduce the number of operations that can be fused to their basic operations. Long story short, they will be fused, so we don't need specific kernels, and having those functions will actually break the fusing stream, so they actually hurt performance as well as add complexity.