Workgroup size is written into the kernel. So to properly modelling
vulkan launch, we have to skip local workgroup size for vulkan launch
call op.
Details
Details
Diff Detail
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
mlir/lib/Conversion/GPUToVulkan/ConvertGPULaunchFuncToVulkanLaunchFunc.cpp | ||
---|---|---|
107 | Super Nit : I think this usage is different from the comment on what kVulkanLaunchNumConfigOperands says. The comment implies, off the argument list to the vulkan call how many of the arguments relate to configuration. Here it is used in the context that the first 3 arguments of the arguments to the gpu.launch_func are needed for the vulkan launch. Had to read this a couple of times to understand. Maybe just add a comment that says that the first kVulkanLaunchNumConfigOperands of the gpu.launch_func op are the same as the config operands for the vulkan call. |
mlir/lib/Conversion/GPUToVulkan/ConvertGPULaunchFuncToVulkanLaunchFunc.cpp | ||
---|---|---|
107 | Thanks for review! Sure, will fix it! |
Super Nit : I think this usage is different from the comment on what kVulkanLaunchNumConfigOperands says. The comment implies, off the argument list to the vulkan call how many of the arguments relate to configuration. Here it is used in the context that the first 3 arguments of the arguments to the gpu.launch_func are needed for the vulkan launch. Had to read this a couple of times to understand. Maybe just add a comment that says that the first kVulkanLaunchNumConfigOperands of the gpu.launch_func op are the same as the config operands for the vulkan call.