ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-24 23:05:54 +00:00
Code Issues Packages Projects Releases Wiki Activity

Added bwd data v3r1 v4r1, tweaking v1 (#10)

Browse Source 
* Added bwd data v3r1: breaking down compute into a series of load balanced GEMM, and launch in a single kernel
* Added bwd data v4r1: like v3r1, but launch GEMMs in multiple kernels
* Tweaked v1r1  and v1r2 (atomic) on AMD GPU

[ROCm/composable_kernel commit: c5da0377fb]
This commit is contained in:
Chao Liu
2020-01-20 10:20:03 -06:00
committed by GitHub
parent 24f7d66609
commit 7c9100b53f
43 changed files with 2123 additions and 452 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
driver/include/device.hpp
View File

@@ -30,33 +30,81 @@ struct KernelTimer
std::unique_ptr<KernelTimerImpl> impl;
};
#if CK_DEVICE_BACKEND_AMD
using device_stream_t = hipStream_t;
template <typename... Args, typename F>
float launch_kernel(F kernel, dim3 grid_dim, dim3 block_dim, std::size_t lds_byte, Args... args)
void launch_kernel(F kernel,
dim3 grid_dim,
dim3 block_dim,
std::size_t lds_byte,
hipStream_t stream_id,
Args... args)
{
hipLaunchKernelGGL(kernel, grid_dim, block_dim, lds_byte, stream_id, args...);
}
template <typename... Args, typename F>
float launch_and_time_kernel(F kernel,
dim3 grid_dim,
dim3 block_dim,
std::size_t lds_byte,
hipStream_t stream_id,
Args... args)
{
KernelTimer timer;
#if CK_DEVICE_BACKEND_AMD
timer.Start();
hipLaunchKernelGGL(kernel, grid_dim, block_dim, lds_byte, 0, args...);
hipLaunchKernelGGL(kernel, grid_dim, block_dim, lds_byte, stream_id, args...);
timer.End();
hipGetErrorString(hipGetLastError());
return timer.GetElapsedTime();
}
#elif CK_DEVICE_BACKEND_NVIDIA
using device_stream_t = cudaStream_t;
template <typename... Args, typename F>
void launch_kernel(F kernel,
dim3 grid_dim,
dim3 block_dim,
std::size_t lds_byte,
cudaStream_t stream_id,
Args... args)
{
const void* f = reinterpret_cast<const void*>(kernel);
void* p_args[] = {&args...};
cudaError_t error = cudaLaunchKernel(f, grid_dim, block_dim, p_args, lds_byte, stream_id);
}
template <typename... Args, typename F>
float launch_and_time_kernel(F kernel,
dim3 grid_dim,
dim3 block_dim,
std::size_t lds_byte,
cudaStream_t stream_id,
Args... args)
{
KernelTimer timer;
const void* f = reinterpret_cast<const void*>(kernel);
void* p_args[] = {&args...};
timer.Start();
cudaError_t error = cudaLaunchKernel(f, grid_dim, block_dim, p_args, lds_byte, 0);
cudaError_t error = cudaLaunchKernel(f, grid_dim, block_dim, p_args, lds_byte, stream_id);
timer.End();
checkCudaErrors(error);
#endif
return timer.GetElapsedTime();
}
#endif
#endif