Dave Airlie

why no semaphores? events are a bit fine grained

3

While level 0 has fences and events it doesn't appear to have any semaphore support. Events can be a messy solution from a scheduler point of view, as they pretty...

This isn't really an issue, it's more of an FYI to level0 developers. In the piglit open source testing project we have a bunch of CL program tests. These tests...

kernel arguments and group size thread safety recommendations

5

The spec doesn't seem to explicitly say this but kernel objects are currently defined should be generated from modules per command recording thread. I wonder if it would be better...

command list are per device not per queue

3

Why are command lists per device not per queue? If you have multi-queue hardware with say separate compute and copy queues, you can't record the final command buffer until submit...

image formats should be kept compatible with vulkan

1

I'm not sure why level 0 invents a new set of image parameters, it probably makes more sense to just import the vulkan image format list and use that as...

device memory allocations required to have mappings/be dereferenceable?

2

Looking at the device memory allocation API, __ze_api_export ze_result_t __zecall zeDriverAllocDeviceMem( ze_driver_handle_t hDriver, ///< [in] handle of the driver instance const ze_device_mem_alloc_desc_t* device_desc, ///< [in] pointer to device mem alloc...

KHR-GL45.texture_view.view_classes accesses unknown memory

1

I built the opengl cts with ASAN support as I was having some wierd issues with running it on llvmpipe swrast but it's uncovering a few things. One of the...

Running with ASAN enabled I got this, looks like releaseXFBVaryingNames needs a delete[] instead of a delete. Test case 'KHR-GL45.gpu_shader_fp64.fp64.varyings'.. ================================================================= ==9303==ERROR: AddressSanitizer: alloc-dealloc-mismatch (operator new [] vs operator delete)...

vulkan backend should use VK_KHR_cooperative_matrix if possible

3

# Feature Description Vulkan has an extension to allow implementations to expose matrix multipliers with certain properties. ([VK_KHR_cooperative_matrix](https://registry.khronos.org/vulkan/specs/1.3-extensions/man/html/VK_KHR_cooperative_matrix.html)) This seems like it should be useful to implement llama.cpp matmul at...

improve 10-bit hevc format handling

2

Attempting a 10-bit HEVC decode on radv results in some issues, I haven't solved then all yet. The first one though is the format picking, I think there is one...