Ben Ashbaugh
Ben Ashbaugh
See: https://github.com/KhronosGroup/OpenCL-Docs/issues/1467#issuecomment-3417603695 We should: 1. Clearly document which appendices are normative and which are informative. Are all appendices informative? 2. Move the normative text from Appendix C out of the...
Currently, in section 3.3.9.1. Atomic Operations: https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_API.html#_atomic_operations We have the line: > [Device-side enqueue](https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_API.html#device-side-enqueue) defines the enumerated type memory_order. This is confusing and misleading, though: the device-side enqueue section does...
The Unified SVM specification currently includes a capability to indicate that this type of SVM has a "single address space" capability: `CL_SVM_CAPABILITY_SINGLE_ADDRESS_SPACE_KHR`. This capability indicate that SVM allocations are from...
This is a PR to partially address the way we document error conditions, see https://github.com/KhronosGroup/OpenCL-Docs/issues/1320 and in particular https://github.com/KhronosGroup/OpenCL-Docs/issues/1320#issuecomment-2745919727. It contains a subset of the changes from https://github.com/KhronosGroup/OpenCL-Docs/pull/1399, specifically for...
#1444 updated the XML file to add `cl_icd_dispatch` as a requirement for the `cl_loader_layers` extension. This seemed like the right thing to do - see https://github.com/KhronosGroup/OpenCL-Docs/pull/1444/files#r2283135496 - but it causes...
This issue is to track discussion regarding adding a "concurrent access atom size" to the unified SVM extension. Specifically, this is a proposed resolution to the issue recorded in the...
Found while refactoring the error conditions for clCreateImageWithProperties: There is currently an error condition that is specified to return `CL_INVALID_IMAGE_FORMAT_DESCRIPTOR`: >if an image is created from a buffer and the...
OpenCL contraction behavior is complicated, especially the interaction between `#pragma OPENCL FP_CONTRACT` and the `-cl-mad-enable` build option, and the subtle difference between an `fma` and a `mad`. We should try...
Found while considering memory objects created from SVM allocations for the unified SVM extension: When a memory object is created from an SVM allocation and `CL_MEM_USE_HOST_PTR`, must the passed-in _host_ptr_...
Currently, the spec is unclear whether zero should be allowed for the `size` argument to `clEnqueueSVMMemFill` and `clEnqueueSVMMemcpy`. For the cl_khr_unified_svm extension, we are leaning towards allowing the `size` to...