feat(hygon): enable RCCL collectives

CMakeLists.txt

@@ -463,8 +463,8 @@ if(WITH_OMPI OR WITH_MPICH)
 endif()
 
 if(WITH_NCCL)
-    if (NOT WITH_NVIDIA AND NOT WITH_ILUVATAR)
-        message(FATAL_ERROR "NCCL backend requires NVIDIA or Iluvatar GPU support. Please enable `WITH_NVIDIA` or `WITH_ILUVATAR`.")
+    if (NOT WITH_NVIDIA AND NOT WITH_ILUVATAR AND NOT WITH_HYGON)
+        message(FATAL_ERROR "NCCL-compatible backend requires NVIDIA, Iluvatar, or Hygon GPU support.")
     endif()
 
     find_library(NCCL_LIB NAMES nccl HINTS ${_NCCL_HINTS} PATH_SUFFIXES lib lib64 REQUIRED)

README.md

@@ -355,7 +355,7 @@ export LD_LIBRARY_PATH=${INFINI_INSTALL}/lib:$LD_LIBRARY_PATH
 |---------|---------------|----------------------|---------------|
 | **OpenMPI** | Full | `WITH_OMPI=ON` | The default backend. Requires the OpenMPI development package.|
 | **MPICH** | Full | `WITH_MPICH=ON` | Requires the MPICH development package.|
-| **NCCL** | Partial | `WITH_NCCL=ON` | Requires NVIDIA or Iluvatar NCCL. Currently available when `WITH_NVIDIA=ON` or `WITH_ILUVATAR=ON`.|
+| **NCCL-compatible** | Partial | `WITH_NCCL=ON` | Uses NCCL on NVIDIA/Iluvatar and the NCCL-compatible RCCL interface on Hygon. |
 | **MCCL** | Partial | `WITH_MCCL=ON` | Requires MetaX or Moore MCCL. Currently available when `WITH_METAX=ON` or `WITH_MOORE=ON`.|
 
 </details>

src/CMakeLists.txt

@@ -251,6 +251,9 @@ if(WITH_NCCL)
     target_sources(infiniccl PRIVATE ${NCCL_SRCS})
     target_include_directories(infiniccl PRIVATE ${NCCL_INC})
     target_link_libraries(infiniccl PRIVATE ${NCCL_LIB})
+    if(WITH_HYGON)
+        target_compile_definitions(infiniccl PUBLIC INFINI_CCL_USE_RCCL=1)
+    endif()
 endif()
 
 # MCCL

src/backend_device_map.h

@@ -21,6 +21,10 @@ template <>
 struct IsSupportedCombination<BackendType::kNccl, Device::Type::kIluvatar>
     : std::true_type {};
 
+template <>
+struct IsSupportedCombination<BackendType::kNccl, Device::Type::kHygon>
+    : std::true_type {};
+
 template <>
 struct IsSupportedCombination<BackendType::kMccl, Device::Type::kMetax>
     : std::true_type {};

src/backends/ccl/nccl/api.h

@@ -1,11 +1,10 @@
 #ifndef INFINI_CCL_BACKENDS_CCL_NCCL_API_H_
 #define INFINI_CCL_BACKENDS_CCL_NCCL_API_H_
 
-#include <nccl.h>
-
 #include <cstddef>
 
 #include "backends/ccl/common/api.h"
+#include "backends/ccl/nccl/vendor.h"
 #include "logging.h"
 #include "return_status_impl.h"
 #include "runtime.h"

src/backends/ccl/nccl/checks.h

@@ -1,10 +1,9 @@
 #ifndef INFINI_CCL_BACKENDS_CCL_NCCL_CHECKS_H_
 #define INFINI_CCL_BACKENDS_CCL_NCCL_CHECKS_H_
 
-#include <nccl.h>
-
 #include <iostream>
 
+#include "backends/ccl/nccl/vendor.h"
 #include "return_status_impl.h"
 
 #define INFINI_CHECK_NCCL(result) \

src/backends/ccl/nccl/hygon/api.h

@@ -0,0 +1,15 @@
+#ifndef INFINI_CCL_BACKENDS_CCL_NCCL_HYGON_API_H_
+#define INFINI_CCL_BACKENDS_CCL_NCCL_HYGON_API_H_
+
+#include "backends/ccl/nccl/api.h"
+#include "devices/hygon/runtime_.h"
+
+namespace infini::ccl {
+
+template <>
+struct CclApi<BackendType::kNccl, Device::Type::kHygon>
+    : NcclApi<Device::Type::kHygon> {};
+
+}  // namespace infini::ccl
+
+#endif  // INFINI_CCL_BACKENDS_CCL_NCCL_HYGON_API_H_

src/backends/ccl/nccl/type_map.h

@@ -1,16 +1,15 @@
 #ifndef INFINI_CCL_BACKENDS_CCL_NCCL_TYPE_MAP_H_
 #define INFINI_CCL_BACKENDS_CCL_NCCL_TYPE_MAP_H_
 
-#include <nccl.h>
-
 #include "backends/ccl/common/api.h"
+#include "backends/ccl/nccl/vendor.h"
 #include "comm_impl.h"
 #include "data_type_impl.h"
 #include "logging.h"
 
 namespace infini::ccl {
 
-#if defined(__CUDA_BF16_TYPES_EXIST__)
+#if defined(__CUDA_BF16_TYPES_EXIST__) || defined(RCCL_BFLOAT16)
 constexpr ncclDataType_t kNcclBFloat16Val = ncclBfloat16;
 #else
 constexpr ncclDataType_t kNcclBFloat16Val = ncclNumTypes;

src/backends/ccl/nccl/vendor.h

@@ -0,0 +1,10 @@
+#ifndef INFINI_CCL_BACKENDS_CCL_NCCL_VENDOR_H_
+#define INFINI_CCL_BACKENDS_CCL_NCCL_VENDOR_H_
+
+#if defined(INFINI_CCL_USE_RCCL)
+#include <rccl/rccl.h>
+#else
+#include <nccl.h>
+#endif
+
+#endif  // INFINI_CCL_BACKENDS_CCL_NCCL_VENDOR_H_