Limits on maximum buffer size (MaxBufferSize or MaxStorageBufferBindingSize) is not being correctly applied

[rcoreilly]

Limits on maximum buffer size (MaxBufferSize or MaxStorageBufferBindingSize) is not being correctly applied

[teoxoy]

268435456 is the default limit, you need to explicitly request higher limits via required_limits when you create the device.

[teoxoy]

I think that would resolve gfx-rs/wgpu-native#408 as well?

[rcoreilly]

@teoxoy I'll check that out and close the issue if so.

Meanwhile, now that I'm actually requesting larger buffers, it is actually failing to run (after successfully giving me the device), with a "Validation Error" :( I'll try to replicate on the latest compute example in wgpu_native and see how that goes.

[rcoreilly]

I have verified that it appears to be working correctly in the latest wgpu_native compute example. We'll have to update our version to the latest and hopefully that fixes the issue.

Also note the question in the code comments below about how to set the RequiredLimits to all defaults except ones you specifically set -- I could not find that exposed in the C interface?

  WGPUSupportedLimits sup;
  wgpuAdapterGetLimits(adapter, &sup);

  WGPULimits limits; // note: I could not find a call that sets all fields to the default "undefined" value: 0xffffffff
  // so I had to just use all the supported limits.  we have this DefaultLimits() function in our Go wrapper,
  // and I found this method in rust: https://docs.rs/wgpu/latest/wgpu/struct.Limits.html -- but nothing in the .h file.
  // how is anyone supposed to just specify a few limits to override in their requested limits??
  WGPURequiredLimits req;
  req.limits = sup.limits;
  req.limits.maxStorageBufferBindingSize = 0x1FFFFFFF; //  note: it doesn't seem to care about this
  req.limits.maxBufferSize = 0x1FFFFFFF; // this is the only one that mattered, relative to bufSize below.
  WGPUDeviceDescriptor desc;
  desc.requiredLimits = &req;
 
  WGPUDevice device = NULL;
  wgpuAdapterRequestDevice(adapter, &desc, handle_request_device,
                           (void *)&device);
  assert(device);

  WGPUQueue queue = wgpuDeviceGetQueue(device);
  assert(queue);

  WGPUShaderModule shader_module =
      frmwrk_load_shader_module(device, "shader.wgsl");
  assert(shader_module);

  WGPUBuffer staging_buffer = wgpuDeviceCreateBuffer(
      device, &(const WGPUBufferDescriptor){
                  .label = "staging_buffer",
                  .usage = WGPUBufferUsage_MapRead | WGPUBufferUsage_CopyDst,
                  .size = numbers_size,
                  .mappedAtCreation = false,
              });
  assert(staging_buffer);

  uint32_t bufSize = 0xFFFFFFFF; // this fails if > size specified above, works if <

  WGPUBuffer storage_buffer = wgpuDeviceCreateBuffer(
      device, &(const WGPUBufferDescriptor){
                  .label = "storage_buffer",
                  .usage = WGPUBufferUsage_Storage | WGPUBufferUsage_CopyDst |
                           WGPUBufferUsage_CopySrc,
                  .size = bufSize,
                  .mappedAtCreation = false,
              });
  assert(storage_buffer);

[teoxoy]

I would open an issue at https://github.com/webgpu-native/webgpu-headers.