AI/rocm-systems
2
0
Garpu 0
Kode Masalah Tarik Permintaan Actions Packages Projects Rilis Wiki Kegiatan

2.17.1-1

Telusuri Sumber 
Add new NVLS algorithm for allreduce using NVLink SHARP (intra-node only).
Add new config options: cgaClusterSize, minCTAs, maxCTAs, netName.
Enable LL128 when we use PXN to close rings.
NVTX3 includes update.
Fix crash when one CollNet (SHARP) rail fails to initialize.


[ROCm/rccl commit: 5d3ab08b69]
This commit is contained in:
Sylvain Jeaugey
2023-02-27 02:48:21 -08:00
orang tua e6e8f2555c
melakukan 8dcf8e8720
72 mengubah file dengan 4541 tambahan dan 2391 penghapusan
Download Patch File Download Diff File Expand all files Collapse all files
projects/rccl/src/include/nvtx.h
+8 -8
Melihat File
@@ -7,12 +7,12 @@
#ifndef NCCL_NVTX_H_
#define NCCL_NVTX_H_
#include "nvtx3.hpp"
#include "nvtx3/nvtx3.hpp"
#if __cpp_constexpr >= 201304L && !defined(NVTX3_RELAXED_CONSTEXPR)
#define NVTX3_RELAXED_CONSTEXPR constexpr
#if __cpp_constexpr >= 201304L && !defined(NVTX3_CONSTEXPR_IF_CPP14)
#define NVTX3_CONSTEXPR_IF_CPP14 constexpr
#else
#define NVTX3_RELAXED_CONSTEXPR
#define NVTX3_CONSTEXPR_IF_CPP14
#endif
// Define all NCCL-provided static schema IDs here (avoid duplicates).
@@ -37,7 +37,7 @@ struct nccl_domain{static constexpr char const* name{"NCCL"};};
class payload_schema {
public:
NVTX3_RELAXED_CONSTEXPR explicit payload_schema(const nvtxPayloadSchemaEntry_t entries[], size_t numEntries, const uint64_t schemaId, const char* schemaName = nullptr) noexcept
explicit payload_schema(const nvtxPayloadSchemaEntry_t entries[], size_t numEntries, const uint64_t schemaId, const char* schemaName = nullptr) noexcept
{
schema_attr.name = schemaName;
schema_attr.entries = entries;
@@ -74,11 +74,11 @@ class payload_schema {
#define NVTX3_FUNC_WITH_PARAMS(ID, S, P) \
static const payload_schema schema{S, std::extent<decltype(S)>::value, \
NVTX_PAYLOAD_ENTRY_TYPE_SCHEMA_ID_STATIC_START + NVTX_SID_##ID, #ID}; \
static ::nvtx3::v1::registered_string<nccl_domain> const nvtx3_func_name__{__func__}; \
static ::nvtx3::v1::registered_string_in<nccl_domain> const nvtx3_func_name__{__func__}; \
nvtxPayloadData_t nvtx3_bpl__[] = { \
{NVTX_PAYLOAD_ENTRY_TYPE_SCHEMA_ID_STATIC_START + NVTX_SID_##ID, sizeof(P), &(P)}}; \
::nvtx3::v1::event_attributes nvtx3_func_attr__{nvtx3_func_name__, nvtx3_bpl__}; \
::nvtx3::v1::domain_thread_range<nccl_domain> const nvtx3_range__{nvtx3_func_attr__};
::nvtx3::v1::event_attributes const nvtx3_func_attr__{nvtx3_func_name__, nvtx3_bpl__}; \
::nvtx3::v1::scoped_range_in<nccl_domain> const nvtx3_range__{nvtx3_func_attr__};
extern void initNvtxRegisteredEnums();