webGPU resources #180

bmschmidt · 2024-12-18T21:52:19Z

Important

Adds WebGPU buffer management, TinyForest implementation, and compute shader preparation for enhanced GPU operations.

WebGPU Buffer Management:
- Introduces WebGPUBufferSet class in buffertools.ts for managing GPU buffers with methods like set() and register().
- Adds createSingletonBuffer() function for creating disposable GPU buffers.
TinyForest Implementation:
- Adds TinyForest class in forests.ts extending StatefulGPU for managing forest parameters and GPU buffers.
- Implements countPipeline() and chooseNextFeatures() for compute pipeline setup and feature selection.
Compute Shader Preparation:
- Adds prepareComputeShader() function in lib.ts for setting up compute shaders with buffer bindings.
- Introduces create_hamming_transform() for creating transformations using Hamming distance in scatterplots.

^{This description was created by}^{for 83f360f. You can customize this summary. It will automatically update as commits are pushed.}

bmschmidt · 2024-12-18T21:52:33Z

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

extensions to webGPU strategies for hamming distance #183
working hamming distance transformation #182
webGPU pipelines #181
webGPU resources #180 👈 (View in Graphite)
refactor buffer storage 4 #179
refactor buffer storage 3 #178
refactor buffer storage 2a #177
refactor buffer storage part 1 #176
webGPU imports #175
allow setting label font #174
refactor manifests to be lazier #173 : 1 other dependent PR (#195 )
uv for quadfeather #154
cleaner deeptable instantiation pattern #172
update evaluation set size on sortedDataSelection #192 : 1 other dependent PR (#193 )
no linting in ci #194 : 1 other dependent PR (#81 )
main

This stack of pull requests is managed by Graphite. Learn more about stacking.

graphite-app · 2024-12-18T21:53:11Z

src/webGPU/lib.ts

+			})
+		);
+
+		passEncoder.dispatchWorkgroups(size / 4 / 64);


The workgroup dispatch calculation size / 4 / 64 should be wrapped in Math.ceil() to prevent dropping elements when the size is not perfectly divisible by the workgroup size. Without rounding up, the last partial workgroup may be skipped, leaving some elements unprocessed.

Spotted by Graphite Reviewer

Is this helpful? React 👍 or 👎 to let us know.

graphite-app · 2024-12-18T21:53:12Z

src/webGPU/forests.ts

+      @compute @workgroup_size(64)
+      //TODOD HERE
+      ` });


The shader module is missing its main function implementation after the @compute workgroup declaration. The code as written will fail to compile. Consider implementing the core compute shader logic here before merging.

Spotted by Graphite Reviewer

Is this helpful? React 👍 or 👎 to let us know.

ellipsis-dev

Caution

Changes requested ❌

Reviewed everything up to 83f360f in 2 minutes and 19 seconds. Click for details.

Reviewed 509 lines of code in 3 files
Skipped 0 files when reviewing.
Skipped posting 3 draft comments. View those below.
Modify your settings and rules to customize what types of comments Ellipsis leaves. And don't forget to react with 👍 or 👎 to teach Ellipsis.

1. src/webGPU/forests.ts:77

Draft comment:
Typo in shader comment: 'TODOD' should likely be 'TODO'.
Reason this comment was not posted:
Comment did not seem useful. Confidence is useful = 70% <= threshold 70% This comment points out a typo in a comment within the code. Correcting typos in comments can improve code readability and maintainability, which aligns with the rule of using descriptive variable and constant names. However, it doesn't directly impact the functionality or performance of the code. Still, it's a minor but valid suggestion for code quality improvement.

2. src/webGPU/buffertools.ts:79

Draft comment:
Typo: In the comment on line 78, "cast it to uint32array" should use the correct capitalization as "Uint32Array".
Reason this comment was not posted:
Decided after close inspection that this draft comment was likely wrong and/or not actionable: usefulness confidence = 10% vs. threshold = 70% While technically correct that Uint32Array is the proper capitalization, this is just about the text in a code comment, not about the actual code functionality. The code itself on line 80 uses the correct Uint32Array capitalization. Comments are meant to convey meaning, and the lowercase version still effectively communicates the intent. This feels like an extremely minor nitpick that doesn't improve code quality. The proper capitalization could be considered part of maintaining technical accuracy in documentation. Consistent and correct terminology usage can help prevent confusion. While technical accuracy is good, this change would have no impact on code functionality or clarity - the meaning is clear either way. This falls under the rule of not making purely informative comments that don't require code changes. Delete this comment as it's an extremely minor documentation nitpick that doesn't impact code quality or functionality.

3. src/webGPU/forests.ts:77

Draft comment:
Typo: 'TODOD HERE' seems to be a mistake. It should probably be 'TODO HERE'.
Reason this comment was not posted:
Decided after close inspection that this draft comment was likely wrong and/or not actionable: usefulness confidence = 10% vs. threshold = 70% While this is technically a typo, it's in a TODO comment that's clearly marking incomplete code. The typo doesn't affect functionality and the entire shader module is incomplete. The comment is very minor and doesn't address any substantive issues. According to the rules, we should not make comments that are obvious or unimportant. The typo could be a sign of rushed or careless code writing, which might warrant extra attention to detail in review. TODOs in committed code could be worth flagging. While attention to detail is important, this is a trivial typo in a comment marking clearly incomplete code. The TODO itself indicates this is work in progress, and nitpicking spelling in such contexts isn't productive. Delete this comment as it points out an unimportant typo in a TODO marker for incomplete code, which doesn't warrant a formal review comment.

Workflow ID: wflow_mwTP0x2HA2ohU6b0

^{You can customize}^{by changing your verbosity settings, reacting with 👍 or 👎, replying to comments, or adding code review rules.}

ellipsis-dev · 2025-05-23T18:51:07Z

src/webGPU/buffertools.ts

+	private async passThroughStagingBuffer(values: Uint32Array, bufferLocation: WebGPUBufferLocation) {
+		// WebGPU 
+		const { buffer, offset, paddedSize } = bufferLocation;
+		while (this.stagingBuffer.mapState !== 'unmapped') {


Busy loop waiting for stagingBuffer to be unmapped may block the main thread. Consider using a queuing mechanism to manage multiple mapping requests.

ellipsis-dev · 2025-05-23T18:51:07Z

src/webGPU/buffertools.ts

+		// If it's a typed array, we can just copy it directly.
+		// cast it to uint32array
+		const v2 = value;
+		const data = new Uint32Array(v2.buffer, v2.byteOffset, v2.byteLength / 4);


Casting all TypedArray values to Uint32Array might be unsafe for non-32‐bit arrays. Ensure that the input type is appropriate.

ellipsis-dev · 2025-05-23T18:51:07Z

src/webGPU/buffertools.ts

+					buffer: this.buffers[0],
+					offset: this.pointer,
+					stride: 4, // meaningless here.
+					byte_size: this.buffer_size - this.pointer,


Inconsistent naming: byte_size uses snake_case while paddedSize is camelCase. Please standardize new code to camelCase.

Suggested change

byte_size: this.buffer_size - this.pointer,

byteSize: this.buffer_size - this.pointer,

ellipsis-dev · 2025-05-23T18:51:07Z

src/webGPU/forests.ts

+  }
+
+  //@ts-expect-error foo
+  private chooseNextFeatures(n = 32) {


Remove debugging console.log and either use the parameter n in chooseNextFeatures or remove it if unnecessary.

ellipsis-dev · 2025-05-23T18:51:07Z

src/webGPU/lib.ts

+}
+
+// hide the state in a global variable.
+const dumb: StatefulGPU[] = [];


Rename global variable dumb to a more descriptive name (e.g., gpuInstances) for clarity.

ellipsis-dev · 2025-05-23T18:51:08Z

src/webGPU/lib.ts

+	const defs = makeShaderDataDefinitions(shaderCode);
+	const myUniformValues = makeStructuredView(defs.uniforms.myUniforms);
+	myUniformValues.set({
+		objectSize: embeddingSize / 32


Clarify the significance of dividing embeddingSize by 32 when setting objectSize in the uniform. A comment explaining this rationale would aid maintainability.

Suggested change

objectSize: embeddingSize / 32

objectSize: embeddingSize / 32 // embeddingSize is in bits, but objectSize is in 32-bit words

This was referenced Dec 18, 2024

refactor buffer storage 4 #179

Open

webGPU pipelines #181

Draft

graphite-app bot reviewed Dec 18, 2024

View reviewed changes

This was referenced Dec 20, 2024

working hamming distance transformation #182

Draft

extensions to webGPU strategies for hamming distance #183

Draft

bmschmidt force-pushed the 12-17-refactor_buffer_storage_split_split_split branch from 7237d17 to 4bb8cf9 Compare May 23, 2025 18:29

bmschmidt force-pushed the 12-17-webgpu_resources branch from 1e1ae8e to 830811c Compare May 23, 2025 18:29

bmschmidt force-pushed the 12-17-refactor_buffer_storage_split_split_split branch from 4bb8cf9 to 61337d0 Compare May 23, 2025 18:37

bmschmidt force-pushed the 12-17-webgpu_resources branch from 830811c to a23a2c3 Compare May 23, 2025 18:37

bmschmidt force-pushed the 12-17-refactor_buffer_storage_split_split_split branch from 61337d0 to eef33f1 Compare May 23, 2025 18:40

bmschmidt force-pushed the 12-17-webgpu_resources branch from a23a2c3 to b9dc76b Compare May 23, 2025 18:40

bmschmidt force-pushed the 12-17-refactor_buffer_storage_split_split_split branch from eef33f1 to 38c301d Compare May 23, 2025 18:46

bmschmidt force-pushed the 12-17-webgpu_resources branch from b9dc76b to 83f360f Compare May 23, 2025 18:46

bmschmidt marked this pull request as ready for review May 23, 2025 18:48

ellipsis-dev bot reviewed May 23, 2025

View reviewed changes

webGPU resources

07bc4b3

bmschmidt force-pushed the 12-17-refactor_buffer_storage_split_split_split branch from 38c301d to 169b197 Compare May 23, 2025 19:12

bmschmidt force-pushed the 12-17-webgpu_resources branch from 83f360f to 07bc4b3 Compare May 23, 2025 19:12

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

webGPU resources #180

webGPU resources #180

Uh oh!

bmschmidt commented Dec 18, 2024 •

edited by ellipsis-dev bot

Loading

Uh oh!

bmschmidt commented Dec 18, 2024 •

edited

Loading

Uh oh!

graphite-app bot Dec 18, 2024

Uh oh!

graphite-app bot Dec 18, 2024

Uh oh!

ellipsis-dev bot left a comment

Uh oh!

ellipsis-dev bot May 23, 2025

Uh oh!

ellipsis-dev bot May 23, 2025

Uh oh!

ellipsis-dev bot May 23, 2025

Uh oh!

ellipsis-dev bot May 23, 2025

Uh oh!

ellipsis-dev bot May 23, 2025

Uh oh!

ellipsis-dev bot May 23, 2025

Uh oh!

Uh oh!

	byte_size: this.buffer_size - this.pointer,
	byteSize: this.buffer_size - this.pointer,

	objectSize: embeddingSize / 32
	objectSize: embeddingSize / 32 // embeddingSize is in bits, but objectSize is in 32-bit words

webGPU resources #180

Are you sure you want to change the base?

webGPU resources #180

Uh oh!

Conversation

bmschmidt commented Dec 18, 2024 • edited by ellipsis-dev bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

bmschmidt commented Dec 18, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

graphite-app bot Dec 18, 2024

Choose a reason for hiding this comment

Uh oh!

graphite-app bot Dec 18, 2024

Choose a reason for hiding this comment

Uh oh!

ellipsis-dev bot left a comment

Choose a reason for hiding this comment

Uh oh!

ellipsis-dev bot May 23, 2025

Choose a reason for hiding this comment

Uh oh!

ellipsis-dev bot May 23, 2025

Choose a reason for hiding this comment

Uh oh!

ellipsis-dev bot May 23, 2025

Choose a reason for hiding this comment

Uh oh!

ellipsis-dev bot May 23, 2025

Choose a reason for hiding this comment

Uh oh!

ellipsis-dev bot May 23, 2025

Choose a reason for hiding this comment

Uh oh!

ellipsis-dev bot May 23, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

bmschmidt commented Dec 18, 2024 •

edited by ellipsis-dev bot

Loading

bmschmidt commented Dec 18, 2024 •

edited

Loading