[CUDA] Use appropriate return code for out of registers kernel launch

Georgi Mirazchiyski · Georgi Mirazchiyski · commit 78a71f3aa7a6 · 2024-03-28T14:26:59.000Z
Change the returned error code for exiting the kernel launch entry point in CUDA
when exceeding the maximum available registers for execution on the SM.
Previously we were returning a misleading error code.
diff --git a/source/adapters/cuda/enqueue.cpp b/source/adapters/cuda/enqueue.cpp
@@ -237,7 +237,7 @@ setKernelParams(const ur_context_handle_t Context,
 
         if (hasExceededMaxRegistersPerBlock(Device, Kernel,
                                             KernelLocalWorkGroupSize)) {
-          return UR_RESULT_ERROR_INVALID_WORK_GROUP_SIZE;
+          return UR_RESULT_ERROR_OUT_OF_RESOURCES;
         }
       } else {
         guessLocalWorkSize(Device, ThreadsPerBlock, GlobalWorkSize, WorkDim,

Original file line number	Diff line number	Diff line change
`@@ -237,7 +237,7 @@ setKernelParams(const ur_context_handle_t Context,`
`237`	`237`
`238`	`238`	`if (hasExceededMaxRegistersPerBlock(Device, Kernel,`
`239`	`239`	`KernelLocalWorkGroupSize)) {`
`240`		`- return UR_RESULT_ERROR_INVALID_WORK_GROUP_SIZE;`
	`240`	`+ return UR_RESULT_ERROR_OUT_OF_RESOURCES;`
`241`	`241`	`}`
`242`	`242`	`} else {`
`243`	`243`	`guessLocalWorkSize(Device, ThreadsPerBlock, GlobalWorkSize, WorkDim,`