[Parallel_libs-commits] [PATCH] D24043: [StreamExecutor] Simplify Kernel classes

[James Price via Parallel_libs-commits]

jprice added inline comments.

================
Comment at: streamexecutor/include/streamexecutor/Kernel.h:47
@@ -49,3 +46,3 @@
 ///     using SaxpyKernel =
-///         streamexecutor::TypedKernel<float *, float *, float *>;
+///         streamexecutor::Kernel<float *, float *, float *>;
 ///     } // namespace compiler_cuda_namespace
----------------
Maybe I'm missing something, but should these parameter types actually be `GlobalDeviceMemory<float>`? Otherwise I'm not sure how such a kernel object could be passed to a `thenLaunch()` function with the correct arguments.

================
Comment at: streamexecutor/include/streamexecutor/Kernel.h:59
@@ -61,2 +58,3 @@
 ///     // MultiKernelLoaderSpec instance like SaxpyKernelLoaderSpec below.
 ///     Expected<ccn::SaxpyKernel> MaybeKernel =
+///         Device->createKernel<cnn::SaxpyKernel>(cnn::SaxpyKernelLoaderSpec);
----------------
Missing `std::unique_ptr<>` here?

================
Comment at: streamexecutor/include/streamexecutor/Kernel.h:60
@@ -62,3 +59,3 @@
 ///     Expected<ccn::SaxpyKernel> MaybeKernel =
-///         ccn::SaxpyKernel::create(Device, ccn::SaxpyKernelLoaderSpec);
+///         Device->createKernel<cnn::SaxpyKernel>(cnn::SaxpyKernelLoaderSpec);
 ///     if (!MaybeKernel) { /* Handle error */ }
----------------
cnn -> ccn


https://reviews.llvm.org/D24043