🏡 index : cold-air-networks/mirror-webgpu-cts.git

export const description = `
Atomically read, and and store value.

* Load the original value pointed to by atomic_ptr.
* Obtains a new value by anding with the value v.
* Store the new value using atomic_ptr.

Returns the original value stored in the atomic object.
`;

import { makeTestGroup } from '../../../../../../../common/framework/test_group.js';
import { keysOf } from '../../../../../../../common/util/data_tables.js';
import { AllFeaturesMaxLimitsGPUTest } from '../../../../../../gpu_test.js';

import {
  dispatchSizes,
  workgroupSizes,
  runStorageVariableTest,
  runWorkgroupVariableTest,
  kMapId,
  typedArrayCtor,
} from './harness.js';

export const g = makeTestGroup(AllFeaturesMaxLimitsGPUTest);

g.test('and_storage')
  .specURL('https://www.w3.org/TR/WGSL/#atomic-rmw')
  .desc(
    `
AS is storage or workgroup
T is i32 or u32

fn atomicAnd(atomic_ptr: ptr<AS, atomic<T>, read_write>, v: T) -> T
`
  )
  .params(u =>
    u
      .combine('workgroupSize', workgroupSizes)
      .combine('dispatchSize', dispatchSizes)
      .combine('mapId', keysOf(kMapId))
      .combine('scalarType', ['u32', 'i32'] as const)
  )
  .fn(t => {
    const numInvocations = t.params.workgroupSize * t.params.dispatchSize;

    // Allocate an output buffer with bitsize of max invocations plus 1 for validation
    const bufferNumElements = Math.max(1, numInvocations / 32) + 1;

    // Start with all bits high, then using atomicAnd to set mapped global id bit off.
    // Note: Both WGSL and JS will shift left 1 by id modulo 32.
    const initValue = 0xffffffff;

    const scalarType = t.params.scalarType;
    const mapId = kMapId[t.params.mapId];
    const extra = mapId.wgsl(numInvocations); // Defines map_id()
    const op = `
      let i = map_id(u32(id));
      atomicAnd(&output[i / 32], ~(${scalarType}(1) << i))
    `;

    const expected = new (typedArrayCtor(scalarType))(bufferNumElements).fill(initValue);
    for (let id = 0; id < numInvocations; ++id) {
      const i = mapId.f(id, numInvocations);
      expected[Math.floor(i / 32)] &= ~(1 << i);
    }

    runStorageVariableTest({
      t,
      workgroupSize: t.params.workgroupSize,
      dispatchSize: t.params.dispatchSize,
      bufferNumElements,
      initValue,
      op,
      expected,
      extra,
    });
  });

g.test('and_workgroup')
  .specURL('https://www.w3.org/TR/WGSL/#atomic-rmw')
  .desc(
    `
AS is storage or workgroup
T is i32 or u32

fn atomicAnd(atomic_ptr: ptr<AS, atomic<T>, read_write>, v: T) -> T
`
  )
  .params(u =>
    u
      .combine('workgroupSize', workgroupSizes)
      .combine('dispatchSize', dispatchSizes)
      .combine('mapId', keysOf(kMapId))
      .combine('scalarType', ['u32', 'i32'] as const)
  )
  .fn(t => {
    const numInvocations = t.params.workgroupSize;

    // Allocate workgroup array with bitsize of max invocations plus 1 for validation
    const wgNumElements = Math.max(1, numInvocations / 32) + 1;

    // Start with all bits high, then using atomicAnd to set mapped global id bit off.
    // Note: Both WGSL and JS will shift left 1 by id modulo 32.
    const initValue = 0xffffffff;

    const scalarType = t.params.scalarType;
    const mapId = kMapId[t.params.mapId];
    const extra = mapId.wgsl(numInvocations); // Defines map_id()
    const op = `
      let i = map_id(u32(id));
      atomicAnd(&wg[i / 32], ~(${scalarType}(1) << i))
    `;

    const expected = new (typedArrayCtor(scalarType))(wgNumElements * t.params.dispatchSize).fill(
      initValue
    );
    for (let d = 0; d < t.params.dispatchSize; ++d) {
      for (let id = 0; id < numInvocations; ++id) {
        const wg = expected.subarray(d * wgNumElements);
        const i = mapId.f(id, numInvocations);
        wg[Math.floor(i / 32)] &= ~(1 << i);
      }
    }

    runWorkgroupVariableTest({
      t,
      workgroupSize: t.params.workgroupSize,
      dispatchSize: t.params.dispatchSize,
      wgNumElements,
      initValue,
      op,
      expected,
      extra,
    });
  });