From b2edee4693113411e4e0207374048cebf6809547 Mon Sep 17 00:00:00 2001
From: Ronak Chauhan <RonakNilesh.Chauhan@amd.com>
Date: Tue, 16 Jun 2020 02:49:12 -0400
Subject: [PATCH] Support passing macros to hipLaunchKernelGGL

This makes hipLaunchKernelGGL take a variable argument list, that will be
expanded before being fed to hipLaunchKernelGGLInternal.

Change-Id: Id76e2bf91acd5d68f56a24fc39f219f2eeb06d33


[ROCm/hip commit: 961717879dfb46f15510d8c973ac0ab32f866f4e]
---
 .../hip/include/hip/hcc_detail/hip_runtime.h    |  6 ++++--
 .../hip/include/hip/nvcc_detail/hip_runtime.h   |  6 ++++--
 projects/hip/tests/src/kernel/hipLaunchParm.cpp | 17 +++++++++++++++++
 3 files changed, 25 insertions(+), 4 deletions(-)

diff --git a/projects/hip/include/hip/hcc_detail/hip_runtime.h b/projects/hip/include/hip/hcc_detail/hip_runtime.h
index 117af3e1fa..b022b8fcdb 100644
--- a/projects/hip/include/hip/hcc_detail/hip_runtime.h
+++ b/projects/hip/include/hip/hcc_detail/hip_runtime.h
@@ -435,10 +435,12 @@ void hipLaunchKernelGGL(F kernel, const dim3& numBlocks, const dim3& dimBlocks,
     hipLaunchKernel(k, numBlocks, dimBlocks, _Args, sharedMemBytes, stream);
 }
 #else
-#define hipLaunchKernelGGL(kernelName, numblocks, numthreads, memperblock, streamId, ...)          \
+#define hipLaunchKernelGGLInternal(kernelName, numBlocks, numThreads, memPerBlock, streamId, ...)  \
     do {                                                                                           \
-        kernelName<<<(numblocks), (numthreads), (memperblock), (streamId)>>>(__VA_ARGS__);         \
+        kernelName<<<(numBlocks), (numThreads), (memPerBlock), (streamId)>>>(__VA_ARGS__);         \
     } while (0)
+
+#define hipLaunchKernelGGL(...)  hipLaunchKernelGGLInternal(__VA_ARGS__)
 #endif
 
 #include <hip/hip_runtime_api.h>
diff --git a/projects/hip/include/hip/nvcc_detail/hip_runtime.h b/projects/hip/include/hip/nvcc_detail/hip_runtime.h
index d159e116c0..8443e3eb2a 100644
--- a/projects/hip/include/hip/nvcc_detail/hip_runtime.h
+++ b/projects/hip/include/hip/nvcc_detail/hip_runtime.h
@@ -31,11 +31,13 @@ THE SOFTWARE.
 
 typedef int hipLaunchParm;
 
-#define hipLaunchKernelGGL(kernelName, numblocks, numthreads, memperblock, streamId, ...)          \
+#define hipLaunchKernelGGLInternal(kernelName, numBlocks, numThreads, memPerBlock, streamId, ...)  \
     do {                                                                                           \
-        kernelName<<<numblocks, numthreads, memperblock, streamId>>>(__VA_ARGS__);                 \
+        kernelName<<<numBlocks, numThreads, memPerBlock, streamId>>>(__VA_ARGS__);                 \
     } while (0)
 
+#define hipLaunchKernelGGL(...)  hipLaunchKernelGGLInternal(__VA_ARGS__)
+
 #define hipReadModeElementType cudaReadModeElementType
 
 #ifdef __CUDA_ARCH__
diff --git a/projects/hip/tests/src/kernel/hipLaunchParm.cpp b/projects/hip/tests/src/kernel/hipLaunchParm.cpp
index 797bc84ced..992ac5a18f 100644
--- a/projects/hip/tests/src/kernel/hipLaunchParm.cpp
+++ b/projects/hip/tests/src/kernel/hipLaunchParm.cpp
@@ -571,6 +571,10 @@ __global__ void hipLaunchKernelStructFunc21(
 
 __global__ void vAdd(float* a) {}
 
+template<class T1, class T2>
+__global__ void myKernel(T1 a, T2 b) {}
+
+
 //---
 // Some wrapper macro for testing:
 #define WRAP(...) __VA_ARGS__
@@ -913,6 +917,19 @@ int main() {
     hipLaunchKernelGGL(HIP_KERNEL_NAME(vAdd), dim3(1024), 1, 0, 0, Ad);
     hipLaunchKernelGGL(HIP_KERNEL_NAME(vAdd), dim3(1024), dim3(1), 0, 0, Ad);
 
+    // Test: Passing macro to hipLaunchKernelGGL
+#define KERNEL_CONFIG  dim3(1024), dim3(1), 0, 0
+    hipLaunchKernelGGL(HIP_KERNEL_NAME(vAdd), KERNEL_CONFIG, Ad);
+
+    // Test: Same thing with templates:
+    // Additional parenthesis required case of templates with multiple type paramters.
+    int a;
+    float b;
+    hipLaunchKernelGGL(HIP_KERNEL_NAME((myKernel<int, float>)), KERNEL_CONFIG, a, b);
+
+#define TYPE_PARAM_CONFIG int, float
+    hipLaunchKernelGGL(HIP_KERNEL_NAME((myKernel<TYPE_PARAM_CONFIG>)), KERNEL_CONFIG, a, b);
+
     // Test: Passing hipLaunchKernelGGL inside another macro:
     float e0;
     MY_LAUNCH_MACRO(hipLaunchKernelGGL(vAdd, dim3(1024),