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

2.10.3-1

Browse Source 
Add support for bfloat16.
Add ncclAvg reduction operation.
Improve performance for aggregated operations.
Improve performance for tree.
Improve network error reporting.
Add NCCL_NET parameter to force a specific network.
Add NCCL_IB_QPS_PER_CONNECTION parameter to split IB traffic onto multiple queue pairs.
Fix topology detection error in WSL2.
Fix proxy memory elements affinity (improve alltoall performance).
Fix graph search on cubemesh topologies.
Fix hang in cubemesh during NVB connections.
This commit is contained in:
Ke Wen
2021-07-08 14:12:04 -07:00
parent 3fec2fa5ee
commit 7e51592129
52 changed files with 3496 additions and 2469 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/include/collectives.h
+19 -4
View File
@@ -1,5 +1,5 @@
/*************************************************************************
* Copyright (c) 2017-2020, NVIDIA CORPORATION. All rights reserved.
* Copyright (c) 2017-2021, NVIDIA CORPORATION. All rights reserved.
*
* See LICENSE.txt for license information
************************************************************************/
@@ -21,8 +21,8 @@
/* Declare all collective operations */
#define DECL5(func, algo, proto, redop, type) \
extern __device__ void NCCL_FUNC_NAME(func, algo, proto, redop, type)(struct ncclWorkElem* args); \
extern __global__ void NCCL_KERN_NAME(func, algo, proto, redop, type)(struct ncclWorkElem c); \
extern __device__ void NCCL_FUNC_NAME(func, algo, proto, redop, type)(); \
extern __global__ void NCCL_KERN_NAME(func, algo, proto, redop, type)(ncclWorkElem c); \
#define DECL4(func, algo, redop, type) \
DECL5(func, algo, SIMPLE, redop, type) \
@@ -34,6 +34,19 @@
DECL4(func, TREE, redop, type) \
DECL4(func, COLLNET, redop, type)
#if defined(__CUDA_BF16_TYPES_EXIST__)
#define DECL2(func, redop) \
DECL3(func, redop, int8_t) \
DECL3(func, redop, uint8_t) \
DECL3(func, redop, int32_t) \
DECL3(func, redop, uint32_t) \
DECL3(func, redop, int64_t) \
DECL3(func, redop, uint64_t) \
DECL3(func, redop, half) \
DECL3(func, redop, float) \
DECL3(func, redop, double) \
DECL3(func, redop, __nv_bfloat16)
#else
#define DECL2(func, redop) \
DECL3(func, redop, int8_t) \
DECL3(func, redop, uint8_t) \
@@ -44,12 +57,14 @@
DECL3(func, redop, half) \
DECL3(func, redop, float) \
DECL3(func, redop, double)
#endif
#define DECL(func) \
DECL2(func, Sum) \
DECL2(func, Prod) \
DECL2(func, Min) \
DECL2(func, Max)
DECL2(func, Max) \
DECL2(func, Avg)
#define DECL_ALL \
DECL2(Broadcast, Sum) \