Fix support for `hipModuleLoad` + `hipModuleLaunchKernel` when used with the `extra` parameter

cpc / hipcl

OpenCL/SPIR-V implementation of HIP

Other

104 stars 12 forks source link

Closed Kerilk closed 3 years ago

Kerilk commented 3 years ago

Hello,

This patch:

allows hipModuleLoad to load a clang offload bundle as produced by clang++'s --cuda-device-only flag.
fixes ClContext::launchWithExtraParams when called with the extra parameter set
fixes ClContext::launchWithExtraParams and ClContext::launchWithKernelParams when used with the null stream.
adds a test for these functionalities except for ClContext::launchWithKernelParams

Thanks,

Colleen and Brice

Kerilk commented 3 years ago

Thanks for the review, it is greatly appreciated. I fixed the style, and responded inline to the flag issue.