Add support for H100 (sm90).
Make sure NCCL kernel honor user stream priorities.
This commit is contained in:
Sylvain Jeaugey
2022-09-27 02:31:13 -07:00
parent 78313a6d21
commit da8152e57a
22 changed files with 303 additions and 76 deletions
+7 -3
View File
@@ -31,13 +31,17 @@ CUDA8_GENCODE = -gencode=arch=compute_35,code=sm_35 \
-gencode=arch=compute_61,code=sm_61
CUDA9_GENCODE = -gencode=arch=compute_70,code=sm_70
CUDA11_GENCODE = -gencode=arch=compute_80,code=sm_80
CUDA11_8_GENCODE = -gencode=arch=compute_90,code=sm_90
CUDA8_PTX = -gencode=arch=compute_61,code=compute_61
CUDA9_PTX = -gencode=arch=compute_70,code=compute_70
CUDA11_PTX = -gencode=arch=compute_80,code=compute_80
CUDA11_8_PTX = -gencode=arch=compute_90,code=compute_90
# Include Ampere support if we're using CUDA11 or above
ifeq ($(shell test "0$(CUDA_MAJOR)" -ge 11; echo $$?),0)
ifeq ($(shell test "0$(CUDA_MAJOR)" -eq 11 -a "0$(CUDA_MINOR)" -ge 8 -o "0$(CUDA_MAJOR)" -gt 11; echo $$?),0)
# Include Hopper support if we're using CUDA11.8 or above
NVCC_GENCODE ?= $(CUDA8_GENCODE) $(CUDA9_GENCODE) $(CUDA11_GENCODE) $(CUDA11_8_GENCODE) $(CUDA11_8_PTX)
else ifeq ($(shell test "0$(CUDA_MAJOR)" -ge 11; echo $$?),0)
NVCC_GENCODE ?= $(CUDA8_GENCODE) $(CUDA9_GENCODE) $(CUDA11_GENCODE) $(CUDA11_PTX)
# Include Volta support if we're using CUDA9 or above
else ifeq ($(shell test "0$(CUDA_MAJOR)" -ge 9; echo $$?),0)
@@ -45,7 +49,7 @@ else ifeq ($(shell test "0$(CUDA_MAJOR)" -ge 9; echo $$?),0)
else
NVCC_GENCODE ?= $(CUDA8_GENCODE) $(CUDA8_PTX)
endif
#$(info NVCC_GENCODE is ${NVCC_GENCODE})
$(info NVCC_GENCODE is ${NVCC_GENCODE})
CXXFLAGS := -DCUDA_MAJOR=$(CUDA_MAJOR) -DCUDA_MINOR=$(CUDA_MINOR) -fPIC -fvisibility=hidden \
-Wall -Wno-unused-function -Wno-sign-compare -std=c++11 -Wvla \
+2 -2
View File
@@ -1,6 +1,6 @@
##### version
NCCL_MAJOR := 2
NCCL_MINOR := 14
NCCL_PATCH := 3
NCCL_MINOR := 15
NCCL_PATCH := 1
NCCL_SUFFIX :=
PKG_REVISION := 1