AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

RCCL 2.4 update

Browse Source
This commit is contained in:
Wenkai Du
2019-07-05 15:43:00 -07:00
parent 4d579e51cc
commit f11c8f60cd
95 changed files with 7829 additions and 614 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/collectives/device/reduce.h
+6 -1
View File
@@ -1,5 +1,6 @@
/*************************************************************************
* Copyright (c) 2015-2019, NVIDIA CORPORATION. All rights reserved.
* Modifications Copyright (c) 2019 Advanced Micro Devices, Inc. All rights reserved.
*
* See LICENSE.txt for license information
************************************************************************/
@@ -9,9 +10,10 @@
#include "collectives.h"
template<int UNROLL, class FUNC, typename T>
__attribute__((noinline))
__device__ void ncclReduceRingKernel(struct CollectiveArgs* args) {
const int tid = threadIdx.x;
const int nthreads = blockDim.x - 1;
const int nthreads = blockDim.x;
const int bid = args->bid;
struct ncclDevComm* comm = args->comm;
struct ncclChannel* channel = comm->channels+blockIdx.x;
@@ -48,9 +50,11 @@ __device__ void ncclReduceRingKernel(struct CollectiveArgs* args) {
}
template<int UNROLL, class FUNC, typename T>
__attribute__((noinline))
__device__ void ncclReduceTreeKernel(struct CollectiveArgs* args) { }
template<int UNUSED, class FUNC, typename T>
__attribute__((noinline))
__device__ void ncclReduceRingLLKernel(struct CollectiveArgs* args) {
const int tid = threadIdx.x;
const int bid = args->bid;
@@ -92,4 +96,5 @@ __device__ void ncclReduceRingLLKernel(struct CollectiveArgs* args) {
}
template<int UNUSED, class FUNC, typename T>
__attribute__((noinline))
__device__ void ncclReduceTreeLLKernel(struct CollectiveArgs* args) { }