Ben Ashbaugh

icon indicating an email [email protected]

icon location I work on GPUs for Intel, but the projects on my Github are my own. Interested in parallel and heterogeneous compute.

Results 432 comments of Ben Ashbaugh

add test for a NULL local work size and a required work-group size

Note, for 3D ND-ranges, the kernel currently requres a 2x3x4 work-group size. This is _probably_ supported by most devices, but it is greater than the minimum required by the spec....

add test for a NULL local work size and a required work-group size

Meh, I just added the checks, they weren't too bad.

add test for a NULL local work size and a required work-group size

Merging as discussed in the November 4th teleconference.

SVM tests where the size is zero

I'll keep this a draft while we determine which of these tests are valid and which aren't.

SVM tests where the size is zero

I'm going to add a couple of additional SVM-related "size is zero" tests: * [x] Calling clSetKernelExecInfo with CL_KERNEL_EXEC_INFO_SVM_PTRS with an empty set of SVM pointers. * This should definitely...

SVM tests where the size is zero

Discussed in the July 8th teleconference. One additional case to add: * [x] Calling clEnqueueSVMFree to asynchronously free a NULL pointer.

SVM tests where the size is zero

Maybe another interesting case? * [x] clEnqueueSVMMigrateMem with a NULL pointer and a size equal to zero.

SVM tests where the size is zero

> clEnqueueSVMMigrateMem with a NULL pointer and a size equal to zero. Observation: All of the implementations I've tested consider this to be an error: `CL_INVALID_VALUE`.

SVM tests where the size is zero

CC @karolherbst

SVM tests where the size is zero

Discussed in the November 4th memory subgroup. * We are going to change these tests to only run when the `cl_khr_unified_svm` extension is supported. This means that shipping implementations remain...