buffer_view CTS by alan-baker · Pull Request #4652 · gpuweb/cts

alan-baker · 2026-05-30T20:09:52Z

buffer_view validation and execution CTS tests.

Issue: #

Requirements for PR author:

All missing test coverage is tracked with "TODO" or .unimplemented().
New helpers are /** documented */ and new helper files are found in helper_index.txt.
Test behaves as expected in a WebGPU implementation. (If not passing, explain above.)
Test have be tested with compatibility mode validation enabled and behave as expected. (If not passing, explain above.)

Requirements for reviewer sign-off:

Tests are properly located.
Test descriptions are accurate and complete.
Tests provide complete coverage (including validation control cases). Missing coverage MUST be covered by TODOs.
Tests avoid over-parameterization (see case count report).

When landing this PR, be sure to make any necessary issue status updates.

* Validation of buffer type and related builtin functions

* Tests for * bufferLength * bufferView * bufferArrayView

github-actions · 2026-05-30T20:16:43Z

Results for build job (at dd3cfe0):

+webgpu:api,validation,dispatch:shader_required_buffer_size:* - 19 cases, 76 subcases (~4/case)
+webgpu:shader,execution,expression,call,builtin,bufferArrayView:array_length:* - 78 cases, 832 subcases (~11/case)
+webgpu:shader,execution,expression,call,builtin,bufferArrayView:array_length,functions:* - 78 cases, 1170 subcases (~15/case)
+webgpu:shader,execution,expression,call,builtin,bufferArrayView:read_layout:* - 35 cases, 604 subcases (~17/case)
+webgpu:shader,execution,expression,call,builtin,bufferArrayView:write_layout:* - 35 cases, 604 subcases (~17/case)
+webgpu:shader,execution,expression,call,builtin,bufferArrayView:read:* - 4 cases, 744 subcases (~186/case)
+webgpu:shader,execution,expression,call,builtin,bufferArrayView:write:* - 4 cases, 1224 subcases (~306/case)
+webgpu:shader,execution,expression,call,builtin,bufferLength:sized_buffer:* - 5 cases, 120 subcases (~24/case)
+webgpu:shader,execution,expression,call,builtin,bufferLength:unsized_buffer:* - 5 cases, 80 subcases (~16/case)
+webgpu:shader,execution,expression,call,builtin,bufferLength:max_size_buffer:* - 6 cases, 6 subcases (~1/case)
+webgpu:shader,execution,expression,call,builtin,bufferView:array_length:* - 78 cases, 832 subcases (~11/case)
+webgpu:shader,execution,expression,call,builtin,bufferView:array_length,functions:* - 78 cases, 1542 subcases (~20/case)
+webgpu:shader,execution,expression,call,builtin,bufferView:read_layout:* - 35 cases, 604 subcases (~17/case)
+webgpu:shader,execution,expression,call,builtin,bufferView:write_layout:* - 35 cases, 604 subcases (~17/case)
+webgpu:shader,execution,expression,call,builtin,bufferView:read:* - 4 cases, 856 subcases (~214/case)
+webgpu:shader,execution,expression,call,builtin,bufferView:write:* - 4 cases, 1224 subcases (~306/case)
-webgpu:shader,validation,decl,let:type:* - 21 cases, 21 subcases (~1/case)
-webgpu:shader,validation,decl,let:initializer:* - 10 cases, 10 subcases (~1/case)
 [snip - full report in action logs]
-TOTAL: 281361 cases, 2323132 subcases
+TOTAL: 282684 cases, 2338877 subcases

dneto0

only partway reviewed

dneto0 · 2026-06-24T19:27:35Z

+    });
+
+    const buffer = t.createBufferTracked({
+      size: t.params.valid ? testcase.size : testcase.size - 4,


I had to look up why this uses "- 4".
It comes from the fact that the buffer binding size in createBindGroup must be a multiple of 4. That's just a familiarity thing for me.

No request for change.

dneto0 · 2026-06-24T19:50:43Z

+    binding_type: 'read-only-storage',
+    requires: ['buffer_view'],
+  },
+};


This needs test cases for:

offset is an override expression

with initializer

with pipeline-time-supplied initializer value

size parameter is an override expression.

if so, then use that size.

otherwise use MinTypeSize. (See my question / comment at https://github.com/gpuweb/gpuweb/pull/6291/changes#r3469837734 )

uh, I think my request is nonsense because the size param has to be a const-expression. (!)

Alan reminded me that the case here is using an override expression for the function arguments; not for the declarations of the buffers

dneto0 · 2026-06-24T19:54:32Z

+    requires: ['buffer_view'],
+  },
+};
+


Also, the bufferArrayView cases less rich than the bufferView cases.

Interesting ones missing for bufferArrayView

compute max value across multiple uses in a shader

multiple uses in the shader, but only pay attention to the ones statically used by the shader in the pipeline

dneto0 · 2026-06-24T19:58:46Z

+): boolean {
+  if (size === 0) {
+    // bufferView case
+    return offset + (type.arrayOffset ?? 0) + type.stride < bufferSize;


Should this be less-than-or-equal?
Also on line 45.

dneto0 · 2026-06-24T20:25:22Z

+  }
+}
+
+export function calculateArrayLength(


Worth saying this returns the element count. That's a bit less ambiguous

dneto0 · 2026-06-24T20:48:40Z

+    decls: `@group(0) @binding(0) var<storage> b : buffer;`,
+  },
+  buffer_sized: {
+    code: `let x = buffer<128> = b;`,


typo: first = should be :

dneto0 · 2026-06-24T20:52:33Z

+    code: `let x : ptr<uniform, buffer<64>> = &b;`,
+    valid: false,
+    decls: `@group(0) @binding(0) var<uniform> b : buffer<128>;`,
+  },


another case:

workgroup sized buffer with override size, assign its address to ptr-workgr-sized-buffer but mismatched override identifier.

dneto0

Finished reviewing. This is a very impressive suite. Thanks for creating it!

I also suggest adding additional validation test cases for arrayLength, i.e. in src/webgpu/shader/validation/expression/call/arrayLength. Sized and unsized buffer types should be tested.

dneto0 · 2026-06-26T16:01:53Z

+    code: `const x = 0; alias T = buffer<x>;`,
+    valid: false,
+  },
+  const_functio: {


nit: typo in key name

dneto0 · 2026-06-26T16:11:07Z

+      expected = t.params.aspace === 'workgroup';
+    }
+    t.expectCompileResult(expected, wgsl);
+  });


Could add a case testing pointer types.

alias PT = ptr<address_space, T>;

(with shenanigans for read-only/read_write access modes)

dneto0 · 2026-06-26T16:12:37Z

+    binding_type: 'read-only-storage',
+    requires: ['buffer_view'],
+  },
+};


Alan reminded me that the case here is using an override expression for the function arguments; not for the declarations of the buffers

dneto0 · 2026-06-26T16:19:02Z

+    let mvar = '';
+    let fvar = '';
+    switch (t.params.aspace) {
+      case 'function':


now that immedates have landed, please add that case. Should always fail, so no need to skip if immediates are not supported. :-)

dneto0 · 2026-06-26T16:32:46Z

    matches: ['ptrWorkgroupOverrideExpr'],
    needsUnrestrictedPointerParameters: true,
  },
+


this is a glorious set of tests. I love how explicit these are.

dneto0 · 2026-06-26T17:24:54Z

+    t.expectCompileResult(type === Type.u32, wgsl);
+  });
+
+g.test('data_type')


Please add a tests:

calling bufferLength on other types: e .g. scalar, array, ptr-to-scalar, ptr-to-array. (don't have to do too much here)

wrong number of parameters: 0, and 2.

dneto0 · 2026-06-26T17:51:20Z

+      t.params.bufferSize / 2
+    }>>, gidx : u32) -> u32 {
+  return unsizedLength(p, gidx);
+}


This is a great test.

Suggest another scenario:

call to sized-N param that calls into sized-N/2 param --> yields N/2

dneto0 · 2026-06-26T18:00:38Z

+fn main() {
+  let val = ${testcase.f16 ? 'f16' : testcase.f32 ? 'f32' : 'u32'}(in);
+  let ptr = ${call_expr};
+  (*${t.params.assign === 'call' ? call_expr : 'ptr'})${testcase.access} = val;


Really great to see both 'let' and 'call' variations here!

dneto0 · 2026-06-26T18:07:28Z

+}
+`;
+
+    runReadWriteTest(false, t, wgsl, ele_ty, ty, t.params.aspace, t.params.offset, bufferSize);


Very impressed and happy with the creation and use of readWriteTest to clarify so much of the mechanics!

dneto0 · 2026-06-26T18:22:13Z

+      t.params.bufferSize / 2
+    }>>, gidx : u32) -> u32 {
+  return unsizedLength(p, gidx);
+}


Similar to arrayView, suggest another scenario:

call to sized-N param that calls into sized-N/2 param --> yields N/2

alan-baker added 2 commits May 27, 2026 09:23

Validation tests for buffer_view

755c414

* Validation of buffer type and related builtin functions

Execution tests for buffer_view

dd3cfe0

* Tests for * bufferLength * bufferView * bufferArrayView

alan-baker requested review from dneto0 and jrprice May 30, 2026 20:09

alan-baker mentioned this pull request Jun 11, 2026

Add buffer_view language feature gpuweb/gpuweb#6291

Open

alan-baker marked this pull request as ready for review June 17, 2026 15:20

mwyrzykowski approved these changes Jun 23, 2026

View reviewed changes

dneto0 requested changes Jun 25, 2026

View reviewed changes

dneto0 requested changes Jun 26, 2026

View reviewed changes

Uh oh!

Conversation

alan-baker commented May 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented May 30, 2026

Uh oh!

dneto0 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dneto0 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

alan-baker commented May 30, 2026 •

edited

Loading