mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2026-10-23 21:28:17 +08:00
add and update samples for CUDA 11.6
This commit is contained in:
Rutwik Choughule
18
Samples/3_CUDA_Features/jacobiCudaGraphs/.vscode/c_cpp_properties.json
vendored
Normal file
18
Samples/3_CUDA_Features/jacobiCudaGraphs/.vscode/c_cpp_properties.json
vendored
Normal file
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"configurations": [
|
||||
{
|
||||
"name": "Linux",
|
||||
"includePath": [
|
||||
"${workspaceFolder}/**",
|
||||
"${workspaceFolder}/../../../Common"
|
||||
],
|
||||
"defines": [],
|
||||
"compilerPath": "/usr/local/cuda/bin/nvcc",
|
||||
"cStandard": "gnu17",
|
||||
"cppStandard": "gnu++14",
|
||||
"intelliSenseMode": "linux-gcc-x64",
|
||||
"configurationProvider": "ms-vscode.makefile-tools"
|
||||
}
|
||||
],
|
||||
"version": 4
|
||||
}
|
||||
7
Samples/3_CUDA_Features/jacobiCudaGraphs/.vscode/extensions.json
vendored
Normal file
7
Samples/3_CUDA_Features/jacobiCudaGraphs/.vscode/extensions.json
vendored
Normal file
@@ -0,0 +1,7 @@
|
||||
{
|
||||
"recommendations": [
|
||||
"nvidia.nsight-vscode-edition",
|
||||
"ms-vscode.cpptools",
|
||||
"ms-vscode.makefile-tools"
|
||||
]
|
||||
}
|
||||
10
Samples/3_CUDA_Features/jacobiCudaGraphs/.vscode/launch.json
vendored
Normal file
10
Samples/3_CUDA_Features/jacobiCudaGraphs/.vscode/launch.json
vendored
Normal file
@@ -0,0 +1,10 @@
|
||||
{
|
||||
"configurations": [
|
||||
{
|
||||
"name": "CUDA C++: Launch",
|
||||
"type": "cuda-gdb",
|
||||
"request": "launch",
|
||||
"program": "${workspaceFolder}/jacobiCudaGraphs"
|
||||
}
|
||||
]
|
||||
}
|
||||
15
Samples/3_CUDA_Features/jacobiCudaGraphs/.vscode/tasks.json
vendored
Normal file
15
Samples/3_CUDA_Features/jacobiCudaGraphs/.vscode/tasks.json
vendored
Normal file
@@ -0,0 +1,15 @@
|
||||
{
|
||||
"version": "2.0.0",
|
||||
"tasks": [
|
||||
{
|
||||
"label": "sample",
|
||||
"type": "shell",
|
||||
"command": "make dbg=1",
|
||||
"problemMatcher": ["$nvcc"],
|
||||
"group": {
|
||||
"kind": "build",
|
||||
"isDefault": true
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
343
Samples/3_CUDA_Features/jacobiCudaGraphs/Makefile
Normal file
343
Samples/3_CUDA_Features/jacobiCudaGraphs/Makefile
Normal file
@@ -0,0 +1,343 @@
|
||||
################################################################################
|
||||
# Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
|
||||
#
|
||||
# Redistribution and use in source and binary forms, with or without
|
||||
# modification, are permitted provided that the following conditions
|
||||
# are met:
|
||||
# * Redistributions of source code must retain the above copyright
|
||||
# notice, this list of conditions and the following disclaimer.
|
||||
# * Redistributions in binary form must reproduce the above copyright
|
||||
# notice, this list of conditions and the following disclaimer in the
|
||||
# documentation and/or other materials provided with the distribution.
|
||||
# * Neither the name of NVIDIA CORPORATION nor the names of its
|
||||
# contributors may be used to endorse or promote products derived
|
||||
# from this software without specific prior written permission.
|
||||
#
|
||||
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
|
||||
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
||||
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
||||
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
|
||||
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
||||
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
||||
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
||||
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
|
||||
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
||||
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
||||
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
#
|
||||
################################################################################
|
||||
#
|
||||
# Makefile project only supported on Mac OS X and Linux Platforms)
|
||||
#
|
||||
################################################################################
|
||||
|
||||
# Location of the CUDA Toolkit
|
||||
CUDA_PATH ?= /usr/local/cuda
|
||||
|
||||
##############################
|
||||
# start deprecated interface #
|
||||
##############################
|
||||
ifeq ($(x86_64),1)
|
||||
$(info WARNING - x86_64 variable has been deprecated)
|
||||
$(info WARNING - please use TARGET_ARCH=x86_64 instead)
|
||||
TARGET_ARCH ?= x86_64
|
||||
endif
|
||||
ifeq ($(ARMv7),1)
|
||||
$(info WARNING - ARMv7 variable has been deprecated)
|
||||
$(info WARNING - please use TARGET_ARCH=armv7l instead)
|
||||
TARGET_ARCH ?= armv7l
|
||||
endif
|
||||
ifeq ($(aarch64),1)
|
||||
$(info WARNING - aarch64 variable has been deprecated)
|
||||
$(info WARNING - please use TARGET_ARCH=aarch64 instead)
|
||||
TARGET_ARCH ?= aarch64
|
||||
endif
|
||||
ifeq ($(ppc64le),1)
|
||||
$(info WARNING - ppc64le variable has been deprecated)
|
||||
$(info WARNING - please use TARGET_ARCH=ppc64le instead)
|
||||
TARGET_ARCH ?= ppc64le
|
||||
endif
|
||||
ifneq ($(GCC),)
|
||||
$(info WARNING - GCC variable has been deprecated)
|
||||
$(info WARNING - please use HOST_COMPILER=$(GCC) instead)
|
||||
HOST_COMPILER ?= $(GCC)
|
||||
endif
|
||||
ifneq ($(abi),)
|
||||
$(error ERROR - abi variable has been removed)
|
||||
endif
|
||||
############################
|
||||
# end deprecated interface #
|
||||
############################
|
||||
|
||||
# architecture
|
||||
HOST_ARCH := $(shell uname -m)
|
||||
TARGET_ARCH ?= $(HOST_ARCH)
|
||||
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le armv7l))
|
||||
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
|
||||
ifneq (,$(filter $(TARGET_ARCH),x86_64 aarch64 sbsa ppc64le))
|
||||
TARGET_SIZE := 64
|
||||
else ifneq (,$(filter $(TARGET_ARCH),armv7l))
|
||||
TARGET_SIZE := 32
|
||||
endif
|
||||
else
|
||||
TARGET_SIZE := $(shell getconf LONG_BIT)
|
||||
endif
|
||||
else
|
||||
$(error ERROR - unsupported value $(TARGET_ARCH) for TARGET_ARCH!)
|
||||
endif
|
||||
|
||||
# sbsa and aarch64 systems look similar. Need to differentiate them at host level for now.
|
||||
ifeq ($(HOST_ARCH),aarch64)
|
||||
ifeq ($(CUDA_PATH)/targets/sbsa-linux,$(shell ls -1d $(CUDA_PATH)/targets/sbsa-linux 2>/dev/null))
|
||||
HOST_ARCH := sbsa
|
||||
TARGET_ARCH := sbsa
|
||||
endif
|
||||
endif
|
||||
|
||||
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
|
||||
ifeq (,$(filter $(HOST_ARCH)-$(TARGET_ARCH),aarch64-armv7l x86_64-armv7l x86_64-aarch64 x86_64-sbsa x86_64-ppc64le))
|
||||
$(error ERROR - cross compiling from $(HOST_ARCH) to $(TARGET_ARCH) is not supported!)
|
||||
endif
|
||||
endif
|
||||
|
||||
# When on native aarch64 system with userspace of 32-bit, change TARGET_ARCH to armv7l
|
||||
ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_SIZE),aarch64-aarch64-32)
|
||||
TARGET_ARCH = armv7l
|
||||
endif
|
||||
|
||||
# operating system
|
||||
HOST_OS := $(shell uname -s 2>/dev/null | tr "[:upper:]" "[:lower:]")
|
||||
TARGET_OS ?= $(HOST_OS)
|
||||
ifeq (,$(filter $(TARGET_OS),linux darwin qnx android))
|
||||
$(error ERROR - unsupported value $(TARGET_OS) for TARGET_OS!)
|
||||
endif
|
||||
|
||||
# host compiler
|
||||
ifeq ($(TARGET_OS),darwin)
|
||||
ifeq ($(shell expr `xcodebuild -version | grep -i xcode | awk '{print $$2}' | cut -d'.' -f1` \>= 5),1)
|
||||
HOST_COMPILER ?= clang++
|
||||
endif
|
||||
else ifneq ($(TARGET_ARCH),$(HOST_ARCH))
|
||||
ifeq ($(HOST_ARCH)-$(TARGET_ARCH),x86_64-armv7l)
|
||||
ifeq ($(TARGET_OS),linux)
|
||||
HOST_COMPILER ?= arm-linux-gnueabihf-g++
|
||||
else ifeq ($(TARGET_OS),qnx)
|
||||
ifeq ($(QNX_HOST),)
|
||||
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
|
||||
endif
|
||||
ifeq ($(QNX_TARGET),)
|
||||
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
|
||||
endif
|
||||
export QNX_HOST
|
||||
export QNX_TARGET
|
||||
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/arm-unknown-nto-qnx6.6.0eabi-g++
|
||||
else ifeq ($(TARGET_OS),android)
|
||||
HOST_COMPILER ?= arm-linux-androideabi-g++
|
||||
endif
|
||||
else ifeq ($(TARGET_ARCH),aarch64)
|
||||
ifeq ($(TARGET_OS), linux)
|
||||
HOST_COMPILER ?= aarch64-linux-gnu-g++
|
||||
else ifeq ($(TARGET_OS),qnx)
|
||||
ifeq ($(QNX_HOST),)
|
||||
$(error ERROR - QNX_HOST must be passed to the QNX host toolchain)
|
||||
endif
|
||||
ifeq ($(QNX_TARGET),)
|
||||
$(error ERROR - QNX_TARGET must be passed to the QNX target toolchain)
|
||||
endif
|
||||
export QNX_HOST
|
||||
export QNX_TARGET
|
||||
HOST_COMPILER ?= $(QNX_HOST)/usr/bin/q++
|
||||
else ifeq ($(TARGET_OS), android)
|
||||
HOST_COMPILER ?= aarch64-linux-android-clang++
|
||||
endif
|
||||
else ifeq ($(TARGET_ARCH),sbsa)
|
||||
HOST_COMPILER ?= aarch64-linux-gnu-g++
|
||||
else ifeq ($(TARGET_ARCH),ppc64le)
|
||||
HOST_COMPILER ?= powerpc64le-linux-gnu-g++
|
||||
endif
|
||||
endif
|
||||
HOST_COMPILER ?= g++
|
||||
NVCC := $(CUDA_PATH)/bin/nvcc -ccbin $(HOST_COMPILER)
|
||||
|
||||
# internal flags
|
||||
NVCCFLAGS := -m${TARGET_SIZE}
|
||||
CCFLAGS :=
|
||||
LDFLAGS :=
|
||||
|
||||
# build flags
|
||||
ifeq ($(TARGET_OS),darwin)
|
||||
LDFLAGS += -rpath $(CUDA_PATH)/lib
|
||||
CCFLAGS += -arch $(HOST_ARCH)
|
||||
else ifeq ($(HOST_ARCH)-$(TARGET_ARCH)-$(TARGET_OS),x86_64-armv7l-linux)
|
||||
LDFLAGS += --dynamic-linker=/lib/ld-linux-armhf.so.3
|
||||
CCFLAGS += -mfloat-abi=hard
|
||||
else ifeq ($(TARGET_OS),android)
|
||||
LDFLAGS += -pie
|
||||
CCFLAGS += -fpie -fpic -fexceptions
|
||||
endif
|
||||
|
||||
ifneq ($(TARGET_ARCH),$(HOST_ARCH))
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
|
||||
ifneq ($(TARGET_FS),)
|
||||
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
|
||||
ifeq ($(GCCVERSIONLTEQ46),1)
|
||||
CCFLAGS += --sysroot=$(TARGET_FS)
|
||||
endif
|
||||
LDFLAGS += --sysroot=$(TARGET_FS)
|
||||
LDFLAGS += -rpath-link=$(TARGET_FS)/lib
|
||||
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib
|
||||
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/arm-linux-gnueabihf
|
||||
endif
|
||||
endif
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
|
||||
ifneq ($(TARGET_FS),)
|
||||
GCCVERSIONLTEQ46 := $(shell expr `$(HOST_COMPILER) -dumpversion` \<= 4.6)
|
||||
ifeq ($(GCCVERSIONLTEQ46),1)
|
||||
CCFLAGS += --sysroot=$(TARGET_FS)
|
||||
endif
|
||||
LDFLAGS += --sysroot=$(TARGET_FS)
|
||||
LDFLAGS += -rpath-link=$(TARGET_FS)/lib -L$(TARGET_FS)/lib
|
||||
LDFLAGS += -rpath-link=$(TARGET_FS)/lib/aarch64-linux-gnu -L$(TARGET_FS)/lib/aarch64-linux-gnu
|
||||
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib -L$(TARGET_FS)/usr/lib
|
||||
LDFLAGS += -rpath-link=$(TARGET_FS)/usr/lib/aarch64-linux-gnu -L$(TARGET_FS)/usr/lib/aarch64-linux-gnu
|
||||
LDFLAGS += --unresolved-symbols=ignore-in-shared-libs
|
||||
CCFLAGS += -isystem=$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include -I$(TARGET_FS)/usr/include/libdrm
|
||||
CCFLAGS += -isystem=$(TARGET_FS)/usr/include/aarch64-linux-gnu -I$(TARGET_FS)/usr/include/aarch64-linux-gnu
|
||||
endif
|
||||
endif
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
|
||||
NVCCFLAGS += -D_QNX_SOURCE
|
||||
NVCCFLAGS += --qpp-config 8.3.0,gcc_ntoaarch64le
|
||||
CCFLAGS += -DWIN_INTERFACE_CUSTOM -I/usr/include/aarch64-qnx-gnu
|
||||
LDFLAGS += -lsocket
|
||||
LDFLAGS += -L/usr/lib/aarch64-qnx-gnu
|
||||
CCFLAGS += "-Wl\,-rpath-link\,/usr/lib/aarch64-qnx-gnu"
|
||||
ifdef TARGET_OVERRIDE
|
||||
LDFLAGS += -lslog2
|
||||
endif
|
||||
|
||||
ifneq ($(TARGET_FS),)
|
||||
LDFLAGS += -L$(TARGET_FS)/usr/lib
|
||||
CCFLAGS += "-Wl\,-rpath-link\,$(TARGET_FS)/usr/lib"
|
||||
LDFLAGS += -L$(TARGET_FS)/usr/libnvidia
|
||||
CCFLAGS += "-Wl\,-rpath-link\,$(TARGET_FS)/usr/libnvidia"
|
||||
CCFLAGS += -I$(TARGET_FS)/../include
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
|
||||
ifdef TARGET_OVERRIDE # cuda toolkit targets override
|
||||
NVCCFLAGS += -target-dir $(TARGET_OVERRIDE)
|
||||
endif
|
||||
|
||||
# Install directory of different arch
|
||||
CUDA_INSTALL_TARGET_DIR :=
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
|
||||
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-gnueabihf/
|
||||
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
|
||||
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux/
|
||||
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
|
||||
CUDA_INSTALL_TARGET_DIR = targets/sbsa-linux/
|
||||
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
|
||||
CUDA_INSTALL_TARGET_DIR = targets/armv7-linux-androideabi/
|
||||
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
|
||||
CUDA_INSTALL_TARGET_DIR = targets/aarch64-linux-androideabi/
|
||||
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
|
||||
CUDA_INSTALL_TARGET_DIR = targets/ARMv7-linux-QNX/
|
||||
else ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
|
||||
CUDA_INSTALL_TARGET_DIR = targets/aarch64-qnx/
|
||||
else ifeq ($(TARGET_ARCH),ppc64le)
|
||||
CUDA_INSTALL_TARGET_DIR = targets/ppc64le-linux/
|
||||
endif
|
||||
|
||||
# Debug build flags
|
||||
ifeq ($(dbg),1)
|
||||
NVCCFLAGS += -g -G
|
||||
BUILD_TYPE := debug
|
||||
else
|
||||
BUILD_TYPE := release
|
||||
endif
|
||||
|
||||
ALL_CCFLAGS :=
|
||||
ALL_CCFLAGS += $(NVCCFLAGS)
|
||||
ALL_CCFLAGS += $(EXTRA_NVCCFLAGS)
|
||||
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(CCFLAGS))
|
||||
ALL_CCFLAGS += $(addprefix -Xcompiler ,$(EXTRA_CCFLAGS))
|
||||
|
||||
SAMPLE_ENABLED := 1
|
||||
|
||||
ALL_LDFLAGS :=
|
||||
ALL_LDFLAGS += $(ALL_CCFLAGS)
|
||||
ALL_LDFLAGS += $(addprefix -Xlinker ,$(LDFLAGS))
|
||||
ALL_LDFLAGS += $(addprefix -Xlinker ,$(EXTRA_LDFLAGS))
|
||||
|
||||
# Common includes and paths for CUDA
|
||||
INCLUDES := -I../../../Common
|
||||
LIBRARIES :=
|
||||
|
||||
################################################################################
|
||||
|
||||
# Gencode arguments
|
||||
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64 sbsa))
|
||||
SMS ?= 53 61 70 72 75 80 86 87
|
||||
else
|
||||
SMS ?= 35 37 50 52 60 61 70 75 80 86
|
||||
endif
|
||||
|
||||
ifeq ($(SMS),)
|
||||
$(info >>> WARNING - no SM architectures have been specified - waiving sample <<<)
|
||||
SAMPLE_ENABLED := 0
|
||||
endif
|
||||
|
||||
ifeq ($(GENCODE_FLAGS),)
|
||||
# Generate SASS code for each SM architecture listed in $(SMS)
|
||||
$(foreach sm,$(SMS),$(eval GENCODE_FLAGS += -gencode arch=compute_$(sm),code=sm_$(sm)))
|
||||
|
||||
# Generate PTX code from the highest SM architecture in $(SMS) to guarantee forward-compatibility
|
||||
HIGHEST_SM := $(lastword $(sort $(SMS)))
|
||||
ifneq ($(HIGHEST_SM),)
|
||||
GENCODE_FLAGS += -gencode arch=compute_$(HIGHEST_SM),code=compute_$(HIGHEST_SM)
|
||||
endif
|
||||
endif
|
||||
|
||||
ALL_CCFLAGS += --threads 0 --std=c++11
|
||||
|
||||
ifeq ($(SAMPLE_ENABLED),0)
|
||||
EXEC ?= @echo "[@]"
|
||||
endif
|
||||
|
||||
################################################################################
|
||||
|
||||
# Target rules
|
||||
all: build
|
||||
|
||||
build: jacobiCudaGraphs
|
||||
|
||||
check.deps:
|
||||
ifeq ($(SAMPLE_ENABLED),0)
|
||||
@echo "Sample will be waived due to the above missing dependencies"
|
||||
else
|
||||
@echo "Sample is ready - all dependencies have been met"
|
||||
endif
|
||||
|
||||
jacobi.o:jacobi.cu
|
||||
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
|
||||
|
||||
main.o:main.cpp
|
||||
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
|
||||
|
||||
jacobiCudaGraphs: jacobi.o main.o
|
||||
$(EXEC) $(NVCC) $(ALL_LDFLAGS) $(GENCODE_FLAGS) -o $@ $+ $(LIBRARIES)
|
||||
$(EXEC) mkdir -p ../../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
|
||||
$(EXEC) cp $@ ../../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)
|
||||
|
||||
run: build
|
||||
$(EXEC) ./jacobiCudaGraphs
|
||||
|
||||
testrun: build
|
||||
|
||||
clean:
|
||||
rm -f jacobiCudaGraphs jacobi.o main.o
|
||||
rm -rf ../../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/jacobiCudaGraphs
|
||||
|
||||
clobber: clean
|
||||
97
Samples/3_CUDA_Features/jacobiCudaGraphs/NsightEclipse.xml
Normal file
97
Samples/3_CUDA_Features/jacobiCudaGraphs/NsightEclipse.xml
Normal file
@@ -0,0 +1,97 @@
|
||||
<?xml version="1.0" encoding="UTF-8"?>
|
||||
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
|
||||
<entry>
|
||||
<name>jacobiCudaGraphs</name>
|
||||
<cuda_api_list>
|
||||
<toolkit>cudaGraphAddMemsetNode</toolkit>
|
||||
<toolkit>cudaStreamCreateWithFlags</toolkit>
|
||||
<toolkit>cudaMemcpyAsync</toolkit>
|
||||
<toolkit>cudaMallocHost</toolkit>
|
||||
<toolkit>cudaPitchedPtr</toolkit>
|
||||
<toolkit>cudaGraphCreate</toolkit>
|
||||
<toolkit>cudaMalloc</toolkit>
|
||||
<toolkit>cudaPos</toolkit>
|
||||
<toolkit>cudaGraphAddMemcpyNode</toolkit>
|
||||
<toolkit>cudaStreamEndCapture</toolkit>
|
||||
<toolkit>cudaGraphExecDestroy</toolkit>
|
||||
<toolkit>cudaStreamBeginCapture</toolkit>
|
||||
<toolkit>cudaGraphExecKernelNodeSetParams</toolkit>
|
||||
<toolkit>cudaStreamSynchronize</toolkit>
|
||||
<toolkit>cudaGraphLaunch</toolkit>
|
||||
<toolkit>cudaFree</toolkit>
|
||||
<toolkit>cudaGraphInstantiate</toolkit>
|
||||
<toolkit>cudaExtent</toolkit>
|
||||
<toolkit>cudaMemsetAsync</toolkit>
|
||||
<toolkit>cudaFreeHost</toolkit>
|
||||
<toolkit>cudaGraphAddKernelNode</toolkit>
|
||||
<toolkit>cudaGraphExecUpdate</toolkit>
|
||||
</cuda_api_list>
|
||||
<description><![CDATA[Demonstrates Instantiated CUDA Graph Update with Jacobi Iterative Method using cudaGraphExecKernelNodeSetParams() and cudaGraphExecUpdate() approach.]]></description>
|
||||
<devicecompilation>whole</devicecompilation>
|
||||
<includepaths>
|
||||
<path>./</path>
|
||||
<path>../</path>
|
||||
<path>../../../Common</path>
|
||||
</includepaths>
|
||||
<keyconcepts>
|
||||
<concept level="basic">CUDA Graphs</concept>
|
||||
<concept level="basic">Stream Capture</concept>
|
||||
<concept level="basic">Instantiated CUDA Graph Update</concept>
|
||||
<concept level="basic">Cooperative Groups</concept>
|
||||
</keyconcepts>
|
||||
<keywords>
|
||||
<keyword>CUDA</keyword>
|
||||
<keyword>GPGPU</keyword>
|
||||
<keyword>CUDA Graphs</keyword>
|
||||
</keywords>
|
||||
<libraries>
|
||||
</libraries>
|
||||
<librarypaths>
|
||||
</librarypaths>
|
||||
<nsight_eclipse>true</nsight_eclipse>
|
||||
<primary_file>main.cpp</primary_file>
|
||||
<scopes>
|
||||
<scope>1:CUDA</scope>
|
||||
</scopes>
|
||||
<sm-arch>sm35</sm-arch>
|
||||
<sm-arch>sm37</sm-arch>
|
||||
<sm-arch>sm50</sm-arch>
|
||||
<sm-arch>sm52</sm-arch>
|
||||
<sm-arch>sm53</sm-arch>
|
||||
<sm-arch>sm60</sm-arch>
|
||||
<sm-arch>sm61</sm-arch>
|
||||
<sm-arch>sm70</sm-arch>
|
||||
<sm-arch>sm72</sm-arch>
|
||||
<sm-arch>sm75</sm-arch>
|
||||
<sm-arch>sm80</sm-arch>
|
||||
<sm-arch>sm86</sm-arch>
|
||||
<sm-arch>sm87</sm-arch>
|
||||
<supported_envs>
|
||||
<env>
|
||||
<arch>x86_64</arch>
|
||||
<platform>linux</platform>
|
||||
</env>
|
||||
<env>
|
||||
<platform>windows7</platform>
|
||||
</env>
|
||||
<env>
|
||||
<arch>x86_64</arch>
|
||||
<platform>macosx</platform>
|
||||
</env>
|
||||
<env>
|
||||
<arch>arm</arch>
|
||||
</env>
|
||||
<env>
|
||||
<arch>sbsa</arch>
|
||||
</env>
|
||||
<env>
|
||||
<arch>ppc64le</arch>
|
||||
<platform>linux</platform>
|
||||
</env>
|
||||
</supported_envs>
|
||||
<supported_sm_architectures>
|
||||
<from>3.5</from>
|
||||
</supported_sm_architectures>
|
||||
<title>Jacobi CUDA Graphs</title>
|
||||
<type>exe</type>
|
||||
</entry>
|
||||
70
Samples/3_CUDA_Features/jacobiCudaGraphs/README.md
Normal file
70
Samples/3_CUDA_Features/jacobiCudaGraphs/README.md
Normal file
@@ -0,0 +1,70 @@
|
||||
# jacobiCudaGraphs - Jacobi CUDA Graphs
|
||||
|
||||
## Description
|
||||
|
||||
Demonstrates Instantiated CUDA Graph Update with Jacobi Iterative Method using cudaGraphExecKernelNodeSetParams() and cudaGraphExecUpdate() approach.
|
||||
|
||||
## Key Concepts
|
||||
|
||||
CUDA Graphs, Stream Capture, Instantiated CUDA Graph Update, Cooperative Groups
|
||||
|
||||
## Supported SM Architectures
|
||||
|
||||
[SM 3.5 ](https://developer.nvidia.com/cuda-gpus) [SM 3.7 ](https://developer.nvidia.com/cuda-gpus) [SM 5.0 ](https://developer.nvidia.com/cuda-gpus) [SM 5.2 ](https://developer.nvidia.com/cuda-gpus) [SM 5.3 ](https://developer.nvidia.com/cuda-gpus) [SM 6.0 ](https://developer.nvidia.com/cuda-gpus) [SM 6.1 ](https://developer.nvidia.com/cuda-gpus) [SM 7.0 ](https://developer.nvidia.com/cuda-gpus) [SM 7.2 ](https://developer.nvidia.com/cuda-gpus) [SM 7.5 ](https://developer.nvidia.com/cuda-gpus) [SM 8.0 ](https://developer.nvidia.com/cuda-gpus) [SM 8.6 ](https://developer.nvidia.com/cuda-gpus) [SM 8.7 ](https://developer.nvidia.com/cuda-gpus)
|
||||
|
||||
## Supported OSes
|
||||
|
||||
Linux, Windows
|
||||
|
||||
## Supported CPU Architecture
|
||||
|
||||
x86_64, ppc64le, armv7l
|
||||
|
||||
## CUDA APIs involved
|
||||
|
||||
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
|
||||
cudaGraphAddMemsetNode, cudaStreamCreateWithFlags, cudaMemcpyAsync, cudaMallocHost, cudaPitchedPtr, cudaGraphCreate, cudaMalloc, cudaPos, cudaGraphAddMemcpyNode, cudaStreamEndCapture, cudaGraphExecDestroy, cudaStreamBeginCapture, cudaGraphExecKernelNodeSetParams, cudaStreamSynchronize, cudaGraphLaunch, cudaFree, cudaGraphInstantiate, cudaExtent, cudaMemsetAsync, cudaFreeHost, cudaGraphAddKernelNode, cudaGraphExecUpdate
|
||||
|
||||
## Prerequisites
|
||||
|
||||
Download and install the [CUDA Toolkit 11.6](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
|
||||
|
||||
## Build and Run
|
||||
|
||||
### Windows
|
||||
The Windows samples are built using the Visual Studio IDE. Solution files (.sln) are provided for each supported version of Visual Studio, using the format:
|
||||
```
|
||||
*_vs<version>.sln - for Visual Studio <version>
|
||||
```
|
||||
Each individual sample has its own set of solution files in its directory:
|
||||
|
||||
To build/examine all the samples at once, the complete solution files should be used. To build/examine a single sample, the individual sample solution files should be used.
|
||||
> **Note:** Some samples require that the Microsoft DirectX SDK (June 2010 or newer) be installed and that the VC++ directory paths are properly set up (**Tools > Options...**). Check DirectX Dependencies section for details."
|
||||
|
||||
### Linux
|
||||
The Linux samples are built using makefiles. To use the makefiles, change the current directory to the sample directory you wish to build, and run make:
|
||||
```
|
||||
$ cd <sample_dir>
|
||||
$ make
|
||||
```
|
||||
The samples makefiles can take advantage of certain options:
|
||||
* **TARGET_ARCH=<arch>** - cross-compile targeting a specific architecture. Allowed architectures are x86_64, ppc64le, armv7l.
|
||||
By default, TARGET_ARCH is set to HOST_ARCH. On a x86_64 machine, not setting TARGET_ARCH is the equivalent of setting TARGET_ARCH=x86_64.<br/>
|
||||
`$ make TARGET_ARCH=x86_64` <br/> `$ make TARGET_ARCH=ppc64le` <br/> `$ make TARGET_ARCH=armv7l` <br/>
|
||||
See [here](http://docs.nvidia.com/cuda/cuda-samples/index.html#cross-samples) for more details.
|
||||
* **dbg=1** - build with debug symbols
|
||||
```
|
||||
$ make dbg=1
|
||||
```
|
||||
* **SMS="A B ..."** - override the SM architectures for which the sample will be built, where `"A B ..."` is a space-delimited list of SM architectures. For example, to generate SASS for SM 50 and SM 60, use `SMS="50 60"`.
|
||||
```
|
||||
$ make SMS="50 60"
|
||||
```
|
||||
|
||||
* **HOST_COMPILER=<host_compiler>** - override the default g++ host compiler. See the [Linux Installation Guide](http://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#system-requirements) for a list of supported host compilers.
|
||||
```
|
||||
$ make HOST_COMPILER=g++
|
||||
```
|
||||
|
||||
## References (for more details)
|
||||
|
||||
393
Samples/3_CUDA_Features/jacobiCudaGraphs/jacobi.cu
Normal file
393
Samples/3_CUDA_Features/jacobiCudaGraphs/jacobi.cu
Normal file
@@ -0,0 +1,393 @@
|
||||
/* Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
|
||||
*
|
||||
* Redistribution and use in source and binary forms, with or without
|
||||
* modification, are permitted provided that the following conditions
|
||||
* are met:
|
||||
* * Redistributions of source code must retain the above copyright
|
||||
* notice, this list of conditions and the following disclaimer.
|
||||
* * Redistributions in binary form must reproduce the above copyright
|
||||
* notice, this list of conditions and the following disclaimer in the
|
||||
* documentation and/or other materials provided with the distribution.
|
||||
* * Neither the name of NVIDIA CORPORATION nor the names of its
|
||||
* contributors may be used to endorse or promote products derived
|
||||
* from this software without specific prior written permission.
|
||||
*
|
||||
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
|
||||
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
||||
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
||||
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
|
||||
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
||||
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
||||
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
||||
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
|
||||
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
||||
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
||||
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
*/
|
||||
|
||||
#include <cooperative_groups.h>
|
||||
#include <cuda_runtime.h>
|
||||
#include <helper_cuda.h>
|
||||
#include <vector>
|
||||
#include "jacobi.h"
|
||||
|
||||
namespace cg = cooperative_groups;
|
||||
|
||||
// 8 Rows of square-matrix A processed by each CTA.
|
||||
// This can be max 32 and only power of 2 (i.e., 2/4/8/16/32).
|
||||
#define ROWS_PER_CTA 8
|
||||
|
||||
#if !defined(__CUDA_ARCH__) || __CUDA_ARCH__ >= 600
|
||||
#else
|
||||
__device__ double atomicAdd(double *address, double val) {
|
||||
unsigned long long int *address_as_ull = (unsigned long long int *)address;
|
||||
unsigned long long int old = *address_as_ull, assumed;
|
||||
|
||||
do {
|
||||
assumed = old;
|
||||
old = atomicCAS(address_as_ull, assumed,
|
||||
__double_as_longlong(val + __longlong_as_double(assumed)));
|
||||
|
||||
// Note: uses integer comparison to avoid hang in case of NaN (since NaN !=
|
||||
// NaN)
|
||||
} while (assumed != old);
|
||||
|
||||
return __longlong_as_double(old);
|
||||
}
|
||||
#endif
|
||||
|
||||
static __global__ void JacobiMethod(const float *A, const double *b,
|
||||
const float conv_threshold, double *x,
|
||||
double *x_new, double *sum) {
|
||||
// Handle to thread block group
|
||||
cg::thread_block cta = cg::this_thread_block();
|
||||
__shared__ double x_shared[N_ROWS]; // N_ROWS == n
|
||||
__shared__ double b_shared[ROWS_PER_CTA + 1];
|
||||
|
||||
for (int i = threadIdx.x; i < N_ROWS; i += blockDim.x) {
|
||||
x_shared[i] = x[i];
|
||||
}
|
||||
|
||||
if (threadIdx.x < ROWS_PER_CTA) {
|
||||
int k = threadIdx.x;
|
||||
for (int i = k + (blockIdx.x * ROWS_PER_CTA);
|
||||
(k < ROWS_PER_CTA) && (i < N_ROWS);
|
||||
k += ROWS_PER_CTA, i += ROWS_PER_CTA) {
|
||||
b_shared[i % (ROWS_PER_CTA + 1)] = b[i];
|
||||
}
|
||||
}
|
||||
|
||||
cg::sync(cta);
|
||||
|
||||
cg::thread_block_tile<32> tile32 = cg::tiled_partition<32>(cta);
|
||||
|
||||
for (int k = 0, i = blockIdx.x * ROWS_PER_CTA;
|
||||
(k < ROWS_PER_CTA) && (i < N_ROWS); k++, i++) {
|
||||
double rowThreadSum = 0.0;
|
||||
for (int j = threadIdx.x; j < N_ROWS; j += blockDim.x) {
|
||||
rowThreadSum += (A[i * N_ROWS + j] * x_shared[j]);
|
||||
}
|
||||
|
||||
for (int offset = tile32.size() / 2; offset > 0; offset /= 2) {
|
||||
rowThreadSum += tile32.shfl_down(rowThreadSum, offset);
|
||||
}
|
||||
|
||||
if (tile32.thread_rank() == 0) {
|
||||
atomicAdd(&b_shared[i % (ROWS_PER_CTA + 1)], -rowThreadSum);
|
||||
}
|
||||
}
|
||||
|
||||
cg::sync(cta);
|
||||
|
||||
if (threadIdx.x < ROWS_PER_CTA) {
|
||||
cg::thread_block_tile<ROWS_PER_CTA> tile8 =
|
||||
cg::tiled_partition<ROWS_PER_CTA>(cta);
|
||||
double temp_sum = 0.0;
|
||||
|
||||
int k = threadIdx.x;
|
||||
|
||||
for (int i = k + (blockIdx.x * ROWS_PER_CTA);
|
||||
(k < ROWS_PER_CTA) && (i < N_ROWS);
|
||||
k += ROWS_PER_CTA, i += ROWS_PER_CTA) {
|
||||
double dx = b_shared[i % (ROWS_PER_CTA + 1)];
|
||||
dx /= A[i * N_ROWS + i];
|
||||
|
||||
x_new[i] = (x_shared[i] + dx);
|
||||
temp_sum += fabs(dx);
|
||||
}
|
||||
|
||||
for (int offset = tile8.size() / 2; offset > 0; offset /= 2) {
|
||||
temp_sum += tile8.shfl_down(temp_sum, offset);
|
||||
}
|
||||
|
||||
if (tile8.thread_rank() == 0) {
|
||||
atomicAdd(sum, temp_sum);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// Thread block size for finalError kernel should be multiple of 32
|
||||
static __global__ void finalError(double *x, double *g_sum) {
|
||||
// Handle to thread block group
|
||||
cg::thread_block cta = cg::this_thread_block();
|
||||
extern __shared__ double warpSum[];
|
||||
double sum = 0.0;
|
||||
|
||||
int globalThreadId = blockIdx.x * blockDim.x + threadIdx.x;
|
||||
|
||||
for (int i = globalThreadId; i < N_ROWS; i += blockDim.x * gridDim.x) {
|
||||
double d = x[i] - 1.0;
|
||||
sum += fabs(d);
|
||||
}
|
||||
|
||||
cg::thread_block_tile<32> tile32 = cg::tiled_partition<32>(cta);
|
||||
|
||||
for (int offset = tile32.size() / 2; offset > 0; offset /= 2) {
|
||||
sum += tile32.shfl_down(sum, offset);
|
||||
}
|
||||
|
||||
if (tile32.thread_rank() == 0) {
|
||||
warpSum[threadIdx.x / warpSize] = sum;
|
||||
}
|
||||
|
||||
cg::sync(cta);
|
||||
|
||||
double blockSum = 0.0;
|
||||
if (threadIdx.x < (blockDim.x / warpSize)) {
|
||||
blockSum = warpSum[threadIdx.x];
|
||||
}
|
||||
|
||||
if (threadIdx.x < 32) {
|
||||
for (int offset = tile32.size() / 2; offset > 0; offset /= 2) {
|
||||
blockSum += tile32.shfl_down(blockSum, offset);
|
||||
}
|
||||
if (tile32.thread_rank() == 0) {
|
||||
atomicAdd(g_sum, blockSum);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
double JacobiMethodGpuCudaGraphExecKernelSetParams(
|
||||
const float *A, const double *b, const float conv_threshold,
|
||||
const int max_iter, double *x, double *x_new, cudaStream_t stream) {
|
||||
// CTA size
|
||||
dim3 nthreads(256, 1, 1);
|
||||
// grid size
|
||||
dim3 nblocks((N_ROWS / ROWS_PER_CTA) + 2, 1, 1);
|
||||
cudaGraph_t graph;
|
||||
cudaGraphExec_t graphExec = NULL;
|
||||
|
||||
double sum = 0.0;
|
||||
double *d_sum = NULL;
|
||||
checkCudaErrors(cudaMalloc(&d_sum, sizeof(double)));
|
||||
|
||||
std::vector<cudaGraphNode_t> nodeDependencies;
|
||||
cudaGraphNode_t memcpyNode, jacobiKernelNode, memsetNode;
|
||||
cudaMemcpy3DParms memcpyParams = {0};
|
||||
cudaMemsetParams memsetParams = {0};
|
||||
|
||||
memsetParams.dst = (void *)d_sum;
|
||||
memsetParams.value = 0;
|
||||
memsetParams.pitch = 0;
|
||||
// elementSize can be max 4 bytes, so we take sizeof(float) and width=2
|
||||
memsetParams.elementSize = sizeof(float);
|
||||
memsetParams.width = 2;
|
||||
memsetParams.height = 1;
|
||||
|
||||
checkCudaErrors(cudaGraphCreate(&graph, 0));
|
||||
checkCudaErrors(
|
||||
cudaGraphAddMemsetNode(&memsetNode, graph, NULL, 0, &memsetParams));
|
||||
nodeDependencies.push_back(memsetNode);
|
||||
|
||||
cudaKernelNodeParams NodeParams0, NodeParams1;
|
||||
NodeParams0.func = (void *)JacobiMethod;
|
||||
NodeParams0.gridDim = nblocks;
|
||||
NodeParams0.blockDim = nthreads;
|
||||
NodeParams0.sharedMemBytes = 0;
|
||||
void *kernelArgs0[6] = {(void *)&A, (void *)&b, (void *)&conv_threshold,
|
||||
(void *)&x, (void *)&x_new, (void *)&d_sum};
|
||||
NodeParams0.kernelParams = kernelArgs0;
|
||||
NodeParams0.extra = NULL;
|
||||
|
||||
checkCudaErrors(
|
||||
cudaGraphAddKernelNode(&jacobiKernelNode, graph, nodeDependencies.data(),
|
||||
nodeDependencies.size(), &NodeParams0));
|
||||
|
||||
nodeDependencies.clear();
|
||||
nodeDependencies.push_back(jacobiKernelNode);
|
||||
|
||||
memcpyParams.srcArray = NULL;
|
||||
memcpyParams.srcPos = make_cudaPos(0, 0, 0);
|
||||
memcpyParams.srcPtr = make_cudaPitchedPtr(d_sum, sizeof(double), 1, 1);
|
||||
memcpyParams.dstArray = NULL;
|
||||
memcpyParams.dstPos = make_cudaPos(0, 0, 0);
|
||||
memcpyParams.dstPtr = make_cudaPitchedPtr(&sum, sizeof(double), 1, 1);
|
||||
memcpyParams.extent = make_cudaExtent(sizeof(double), 1, 1);
|
||||
memcpyParams.kind = cudaMemcpyDeviceToHost;
|
||||
|
||||
checkCudaErrors(
|
||||
cudaGraphAddMemcpyNode(&memcpyNode, graph, nodeDependencies.data(),
|
||||
nodeDependencies.size(), &memcpyParams));
|
||||
|
||||
checkCudaErrors(cudaGraphInstantiate(&graphExec, graph, NULL, NULL, 0));
|
||||
|
||||
NodeParams1.func = (void *)JacobiMethod;
|
||||
NodeParams1.gridDim = nblocks;
|
||||
NodeParams1.blockDim = nthreads;
|
||||
NodeParams1.sharedMemBytes = 0;
|
||||
void *kernelArgs1[6] = {(void *)&A, (void *)&b, (void *)&conv_threshold,
|
||||
(void *)&x_new, (void *)&x, (void *)&d_sum};
|
||||
NodeParams1.kernelParams = kernelArgs1;
|
||||
NodeParams1.extra = NULL;
|
||||
|
||||
int k = 0;
|
||||
for (k = 0; k < max_iter; k++) {
|
||||
checkCudaErrors(cudaGraphExecKernelNodeSetParams(
|
||||
graphExec, jacobiKernelNode,
|
||||
((k & 1) == 0) ? &NodeParams0 : &NodeParams1));
|
||||
checkCudaErrors(cudaGraphLaunch(graphExec, stream));
|
||||
checkCudaErrors(cudaStreamSynchronize(stream));
|
||||
|
||||
if (sum <= conv_threshold) {
|
||||
checkCudaErrors(cudaMemsetAsync(d_sum, 0, sizeof(double), stream));
|
||||
nblocks.x = (N_ROWS / nthreads.x) + 1;
|
||||
size_t sharedMemSize = ((nthreads.x / 32) + 1) * sizeof(double);
|
||||
if ((k & 1) == 0) {
|
||||
finalError<<<nblocks, nthreads, sharedMemSize, stream>>>(x_new, d_sum);
|
||||
} else {
|
||||
finalError<<<nblocks, nthreads, sharedMemSize, stream>>>(x, d_sum);
|
||||
}
|
||||
|
||||
checkCudaErrors(cudaMemcpyAsync(&sum, d_sum, sizeof(double),
|
||||
cudaMemcpyDeviceToHost, stream));
|
||||
checkCudaErrors(cudaStreamSynchronize(stream));
|
||||
printf("GPU iterations : %d\n", k + 1);
|
||||
printf("GPU error : %.3e\n", sum);
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
checkCudaErrors(cudaFree(d_sum));
|
||||
return sum;
|
||||
}
|
||||
|
||||
double JacobiMethodGpuCudaGraphExecUpdate(const float *A, const double *b,
|
||||
const float conv_threshold,
|
||||
const int max_iter, double *x,
|
||||
double *x_new, cudaStream_t stream) {
|
||||
// CTA size
|
||||
dim3 nthreads(256, 1, 1);
|
||||
// grid size
|
||||
dim3 nblocks((N_ROWS / ROWS_PER_CTA) + 2, 1, 1);
|
||||
cudaGraph_t graph;
|
||||
cudaGraphExec_t graphExec = NULL;
|
||||
|
||||
double sum = 0.0;
|
||||
double *d_sum;
|
||||
checkCudaErrors(cudaMalloc(&d_sum, sizeof(double)));
|
||||
|
||||
int k = 0;
|
||||
for (k = 0; k < max_iter; k++) {
|
||||
checkCudaErrors(
|
||||
cudaStreamBeginCapture(stream, cudaStreamCaptureModeGlobal));
|
||||
checkCudaErrors(cudaMemsetAsync(d_sum, 0, sizeof(double), stream));
|
||||
if ((k & 1) == 0) {
|
||||
JacobiMethod<<<nblocks, nthreads, 0, stream>>>(A, b, conv_threshold, x,
|
||||
x_new, d_sum);
|
||||
} else {
|
||||
JacobiMethod<<<nblocks, nthreads, 0, stream>>>(A, b, conv_threshold,
|
||||
x_new, x, d_sum);
|
||||
}
|
||||
checkCudaErrors(cudaMemcpyAsync(&sum, d_sum, sizeof(double),
|
||||
cudaMemcpyDeviceToHost, stream));
|
||||
checkCudaErrors(cudaStreamEndCapture(stream, &graph));
|
||||
|
||||
if (graphExec == NULL) {
|
||||
checkCudaErrors(cudaGraphInstantiate(&graphExec, graph, NULL, NULL, 0));
|
||||
} else {
|
||||
cudaGraphExecUpdateResult updateResult_out;
|
||||
checkCudaErrors(
|
||||
cudaGraphExecUpdate(graphExec, graph, NULL, &updateResult_out));
|
||||
if (updateResult_out != cudaGraphExecUpdateSuccess) {
|
||||
if (graphExec != NULL) {
|
||||
checkCudaErrors(cudaGraphExecDestroy(graphExec));
|
||||
}
|
||||
printf("k = %d graph update failed with error - %d\n", k,
|
||||
updateResult_out);
|
||||
checkCudaErrors(cudaGraphInstantiate(&graphExec, graph, NULL, NULL, 0));
|
||||
}
|
||||
}
|
||||
checkCudaErrors(cudaGraphLaunch(graphExec, stream));
|
||||
checkCudaErrors(cudaStreamSynchronize(stream));
|
||||
|
||||
if (sum <= conv_threshold) {
|
||||
checkCudaErrors(cudaMemsetAsync(d_sum, 0, sizeof(double), stream));
|
||||
nblocks.x = (N_ROWS / nthreads.x) + 1;
|
||||
size_t sharedMemSize = ((nthreads.x / 32) + 1) * sizeof(double);
|
||||
if ((k & 1) == 0) {
|
||||
finalError<<<nblocks, nthreads, sharedMemSize, stream>>>(x_new, d_sum);
|
||||
} else {
|
||||
finalError<<<nblocks, nthreads, sharedMemSize, stream>>>(x, d_sum);
|
||||
}
|
||||
|
||||
checkCudaErrors(cudaMemcpyAsync(&sum, d_sum, sizeof(double),
|
||||
cudaMemcpyDeviceToHost, stream));
|
||||
checkCudaErrors(cudaStreamSynchronize(stream));
|
||||
printf("GPU iterations : %d\n", k + 1);
|
||||
printf("GPU error : %.3e\n", sum);
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
checkCudaErrors(cudaFree(d_sum));
|
||||
return sum;
|
||||
}
|
||||
|
||||
double JacobiMethodGpu(const float *A, const double *b,
|
||||
const float conv_threshold, const int max_iter,
|
||||
double *x, double *x_new, cudaStream_t stream) {
|
||||
// CTA size
|
||||
dim3 nthreads(256, 1, 1);
|
||||
// grid size
|
||||
dim3 nblocks((N_ROWS / ROWS_PER_CTA) + 2, 1, 1);
|
||||
|
||||
double sum = 0.0;
|
||||
double *d_sum;
|
||||
checkCudaErrors(cudaMalloc(&d_sum, sizeof(double)));
|
||||
int k = 0;
|
||||
|
||||
for (k = 0; k < max_iter; k++) {
|
||||
checkCudaErrors(cudaMemsetAsync(d_sum, 0, sizeof(double), stream));
|
||||
if ((k & 1) == 0) {
|
||||
JacobiMethod<<<nblocks, nthreads, 0, stream>>>(A, b, conv_threshold, x,
|
||||
x_new, d_sum);
|
||||
} else {
|
||||
JacobiMethod<<<nblocks, nthreads, 0, stream>>>(A, b, conv_threshold,
|
||||
x_new, x, d_sum);
|
||||
}
|
||||
checkCudaErrors(cudaMemcpyAsync(&sum, d_sum, sizeof(double),
|
||||
cudaMemcpyDeviceToHost, stream));
|
||||
checkCudaErrors(cudaStreamSynchronize(stream));
|
||||
|
||||
if (sum <= conv_threshold) {
|
||||
checkCudaErrors(cudaMemsetAsync(d_sum, 0, sizeof(double), stream));
|
||||
nblocks.x = (N_ROWS / nthreads.x) + 1;
|
||||
size_t sharedMemSize = ((nthreads.x / 32) + 1) * sizeof(double);
|
||||
if ((k & 1) == 0) {
|
||||
finalError<<<nblocks, nthreads, sharedMemSize, stream>>>(x_new, d_sum);
|
||||
} else {
|
||||
finalError<<<nblocks, nthreads, sharedMemSize, stream>>>(x, d_sum);
|
||||
}
|
||||
|
||||
checkCudaErrors(cudaMemcpyAsync(&sum, d_sum, sizeof(double),
|
||||
cudaMemcpyDeviceToHost, stream));
|
||||
checkCudaErrors(cudaStreamSynchronize(stream));
|
||||
printf("GPU iterations : %d\n", k + 1);
|
||||
printf("GPU error : %.3e\n", sum);
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
checkCudaErrors(cudaFree(d_sum));
|
||||
return sum;
|
||||
}
|
||||
33
Samples/3_CUDA_Features/jacobiCudaGraphs/jacobi.h
Normal file
33
Samples/3_CUDA_Features/jacobiCudaGraphs/jacobi.h
Normal file
@@ -0,0 +1,33 @@
|
||||
/* Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
|
||||
*
|
||||
* Redistribution and use in source and binary forms, with or without
|
||||
* modification, are permitted provided that the following conditions
|
||||
* are met:
|
||||
* * Redistributions of source code must retain the above copyright
|
||||
* notice, this list of conditions and the following disclaimer.
|
||||
* * Redistributions in binary form must reproduce the above copyright
|
||||
* notice, this list of conditions and the following disclaimer in the
|
||||
* documentation and/or other materials provided with the distribution.
|
||||
* * Neither the name of NVIDIA CORPORATION nor the names of its
|
||||
* contributors may be used to endorse or promote products derived
|
||||
* from this software without specific prior written permission.
|
||||
*
|
||||
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
|
||||
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
||||
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
||||
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
|
||||
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
||||
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
||||
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
||||
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
|
||||
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
||||
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
||||
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
*/
|
||||
|
||||
#ifndef JACOBI_H
|
||||
#define JACOBI_H
|
||||
|
||||
#define N_ROWS 512
|
||||
|
||||
#endif
|
||||
@@ -0,0 +1,20 @@
|
||||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2017
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "jacobiCudaGraphs", "jacobiCudaGraphs_vs2017.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
EndProject
|
||||
Global
|
||||
GlobalSection(SolutionConfigurationPlatforms) = preSolution
|
||||
Debug|x64 = Debug|x64
|
||||
Release|x64 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(ProjectConfigurationPlatforms) = postSolution
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(SolutionProperties) = preSolution
|
||||
HideSolutionNode = FALSE
|
||||
EndGlobalSection
|
||||
EndGlobal
|
||||
@@ -0,0 +1,113 @@
|
||||
<?xml version="1.0" encoding="utf-8"?>
|
||||
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
|
||||
<PropertyGroup>
|
||||
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
|
||||
</PropertyGroup>
|
||||
<ItemGroup Label="ProjectConfigurations">
|
||||
<ProjectConfiguration Include="Debug|x64">
|
||||
<Configuration>Debug</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
<ProjectConfiguration Include="Release|x64">
|
||||
<Configuration>Release</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
</ItemGroup>
|
||||
<PropertyGroup Label="Globals">
|
||||
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
|
||||
<RootNamespace>jacobiCudaGraphs_vs2017</RootNamespace>
|
||||
<ProjectName>jacobiCudaGraphs</ProjectName>
|
||||
<CudaToolkitCustomDir />
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(WindowsTargetPlatformVersion)'==''">
|
||||
<LatestTargetPlatformVersion>$([Microsoft.Build.Utilities.ToolLocationHelper]::GetLatestSDKTargetPlatformVersion('Windows', '10.0'))</LatestTargetPlatformVersion>
|
||||
<WindowsTargetPlatformVersion Condition="'$(WindowsTargetPlatformVersion)' == ''">$(LatestTargetPlatformVersion)</WindowsTargetPlatformVersion>
|
||||
<TargetPlatformVersion>$(WindowsTargetPlatformVersion)</TargetPlatformVersion>
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
|
||||
<PropertyGroup>
|
||||
<ConfigurationType>Application</ConfigurationType>
|
||||
<CharacterSet>MultiByte</CharacterSet>
|
||||
<PlatformToolset>v141</PlatformToolset>
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
|
||||
<UseDebugLibraries>true</UseDebugLibraries>
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(Configuration)'=='Release'">
|
||||
<WholeProgramOptimization>true</WholeProgramOptimization>
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 11.6.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
</ImportGroup>
|
||||
<PropertyGroup Label="UserMacros" />
|
||||
<PropertyGroup>
|
||||
<IntDir>$(Platform)/$(Configuration)/</IntDir>
|
||||
<IncludePath>$(IncludePath)</IncludePath>
|
||||
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
|
||||
<CodeAnalysisRules />
|
||||
<CodeAnalysisRuleAssemblies />
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(Platform)'=='x64'">
|
||||
<OutDir>../../../bin/win64/$(Configuration)/</OutDir>
|
||||
</PropertyGroup>
|
||||
<ItemDefinitionGroup>
|
||||
<ClCompile>
|
||||
<WarningLevel>Level3</WarningLevel>
|
||||
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../../Common;</AdditionalIncludeDirectories>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<SubSystem>Console</SubSystem>
|
||||
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
|
||||
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
|
||||
<OutputFile>$(OutDir)/jacobiCudaGraphs.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;</CodeGeneration>
|
||||
<AdditionalOptions>-Xcompiler "/wd 4819" --threads 0 </AdditionalOptions>
|
||||
<Include>./;../../../Common</Include>
|
||||
<Defines>WIN32</Defines>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
|
||||
<ClCompile>
|
||||
<Optimization>Disabled</Optimization>
|
||||
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<GenerateDebugInformation>true</GenerateDebugInformation>
|
||||
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<Runtime>MTd</Runtime>
|
||||
<TargetMachinePlatform>64</TargetMachinePlatform>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
|
||||
<ClCompile>
|
||||
<Optimization>MaxSpeed</Optimization>
|
||||
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<GenerateDebugInformation>false</GenerateDebugInformation>
|
||||
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<Runtime>MT</Runtime>
|
||||
<TargetMachinePlatform>64</TargetMachinePlatform>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<CudaCompile Include="jacobi.cu" />
|
||||
<ClCompile Include="main.cpp" />
|
||||
<ClInclude Include="jacobi.h" />
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 11.6.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
@@ -0,0 +1,20 @@
|
||||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2019
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "jacobiCudaGraphs", "jacobiCudaGraphs_vs2019.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
EndProject
|
||||
Global
|
||||
GlobalSection(SolutionConfigurationPlatforms) = preSolution
|
||||
Debug|x64 = Debug|x64
|
||||
Release|x64 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(ProjectConfigurationPlatforms) = postSolution
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(SolutionProperties) = preSolution
|
||||
HideSolutionNode = FALSE
|
||||
EndGlobalSection
|
||||
EndGlobal
|
||||
@@ -0,0 +1,109 @@
|
||||
<?xml version="1.0" encoding="utf-8"?>
|
||||
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
|
||||
<PropertyGroup>
|
||||
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
|
||||
</PropertyGroup>
|
||||
<ItemGroup Label="ProjectConfigurations">
|
||||
<ProjectConfiguration Include="Debug|x64">
|
||||
<Configuration>Debug</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
<ProjectConfiguration Include="Release|x64">
|
||||
<Configuration>Release</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
</ItemGroup>
|
||||
<PropertyGroup Label="Globals">
|
||||
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
|
||||
<RootNamespace>jacobiCudaGraphs_vs2019</RootNamespace>
|
||||
<ProjectName>jacobiCudaGraphs</ProjectName>
|
||||
<CudaToolkitCustomDir />
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
|
||||
<PropertyGroup>
|
||||
<ConfigurationType>Application</ConfigurationType>
|
||||
<CharacterSet>MultiByte</CharacterSet>
|
||||
<PlatformToolset>v142</PlatformToolset>
|
||||
<WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
|
||||
<UseDebugLibraries>true</UseDebugLibraries>
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(Configuration)'=='Release'">
|
||||
<WholeProgramOptimization>true</WholeProgramOptimization>
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 11.6.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
</ImportGroup>
|
||||
<PropertyGroup Label="UserMacros" />
|
||||
<PropertyGroup>
|
||||
<IntDir>$(Platform)/$(Configuration)/</IntDir>
|
||||
<IncludePath>$(IncludePath)</IncludePath>
|
||||
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
|
||||
<CodeAnalysisRules />
|
||||
<CodeAnalysisRuleAssemblies />
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(Platform)'=='x64'">
|
||||
<OutDir>../../../bin/win64/$(Configuration)/</OutDir>
|
||||
</PropertyGroup>
|
||||
<ItemDefinitionGroup>
|
||||
<ClCompile>
|
||||
<WarningLevel>Level3</WarningLevel>
|
||||
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../../Common;</AdditionalIncludeDirectories>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<SubSystem>Console</SubSystem>
|
||||
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
|
||||
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
|
||||
<OutputFile>$(OutDir)/jacobiCudaGraphs.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;</CodeGeneration>
|
||||
<AdditionalOptions>-Xcompiler "/wd 4819" --threads 0 </AdditionalOptions>
|
||||
<Include>./;../../../Common</Include>
|
||||
<Defines>WIN32</Defines>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
|
||||
<ClCompile>
|
||||
<Optimization>Disabled</Optimization>
|
||||
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<GenerateDebugInformation>true</GenerateDebugInformation>
|
||||
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<Runtime>MTd</Runtime>
|
||||
<TargetMachinePlatform>64</TargetMachinePlatform>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
|
||||
<ClCompile>
|
||||
<Optimization>MaxSpeed</Optimization>
|
||||
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<GenerateDebugInformation>false</GenerateDebugInformation>
|
||||
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<Runtime>MT</Runtime>
|
||||
<TargetMachinePlatform>64</TargetMachinePlatform>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<CudaCompile Include="jacobi.cu" />
|
||||
<ClCompile Include="main.cpp" />
|
||||
<ClInclude Include="jacobi.h" />
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 11.6.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
@@ -0,0 +1,20 @@
|
||||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2022
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "jacobiCudaGraphs", "jacobiCudaGraphs_vs2022.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
EndProject
|
||||
Global
|
||||
GlobalSection(SolutionConfigurationPlatforms) = preSolution
|
||||
Debug|x64 = Debug|x64
|
||||
Release|x64 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(ProjectConfigurationPlatforms) = postSolution
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(SolutionProperties) = preSolution
|
||||
HideSolutionNode = FALSE
|
||||
EndGlobalSection
|
||||
EndGlobal
|
||||
@@ -0,0 +1,109 @@
|
||||
<?xml version="1.0" encoding="utf-8"?>
|
||||
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
|
||||
<PropertyGroup>
|
||||
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
|
||||
</PropertyGroup>
|
||||
<ItemGroup Label="ProjectConfigurations">
|
||||
<ProjectConfiguration Include="Debug|x64">
|
||||
<Configuration>Debug</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
<ProjectConfiguration Include="Release|x64">
|
||||
<Configuration>Release</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
</ItemGroup>
|
||||
<PropertyGroup Label="Globals">
|
||||
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
|
||||
<RootNamespace>jacobiCudaGraphs_vs2022</RootNamespace>
|
||||
<ProjectName>jacobiCudaGraphs</ProjectName>
|
||||
<CudaToolkitCustomDir />
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
|
||||
<PropertyGroup>
|
||||
<ConfigurationType>Application</ConfigurationType>
|
||||
<CharacterSet>MultiByte</CharacterSet>
|
||||
<PlatformToolset>v143</PlatformToolset>
|
||||
<WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(Configuration)'=='Debug'">
|
||||
<UseDebugLibraries>true</UseDebugLibraries>
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(Configuration)'=='Release'">
|
||||
<WholeProgramOptimization>true</WholeProgramOptimization>
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 11.6.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
</ImportGroup>
|
||||
<PropertyGroup Label="UserMacros" />
|
||||
<PropertyGroup>
|
||||
<IntDir>$(Platform)/$(Configuration)/</IntDir>
|
||||
<IncludePath>$(IncludePath)</IncludePath>
|
||||
<CodeAnalysisRuleSet>AllRules.ruleset</CodeAnalysisRuleSet>
|
||||
<CodeAnalysisRules />
|
||||
<CodeAnalysisRuleAssemblies />
|
||||
</PropertyGroup>
|
||||
<PropertyGroup Condition="'$(Platform)'=='x64'">
|
||||
<OutDir>../../../bin/win64/$(Configuration)/</OutDir>
|
||||
</PropertyGroup>
|
||||
<ItemDefinitionGroup>
|
||||
<ClCompile>
|
||||
<WarningLevel>Level3</WarningLevel>
|
||||
<PreprocessorDefinitions>WIN32;_MBCS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
|
||||
<AdditionalIncludeDirectories>./;$(CudaToolkitDir)/include;../../../Common;</AdditionalIncludeDirectories>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<SubSystem>Console</SubSystem>
|
||||
<AdditionalDependencies>cudart_static.lib;kernel32.lib;user32.lib;gdi32.lib;winspool.lib;comdlg32.lib;advapi32.lib;shell32.lib;ole32.lib;oleaut32.lib;uuid.lib;odbc32.lib;odbccp32.lib;%(AdditionalDependencies)</AdditionalDependencies>
|
||||
<AdditionalLibraryDirectories>$(CudaToolkitLibDir);</AdditionalLibraryDirectories>
|
||||
<OutputFile>$(OutDir)/jacobiCudaGraphs.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;</CodeGeneration>
|
||||
<AdditionalOptions>-Xcompiler "/wd 4819" --threads 0 </AdditionalOptions>
|
||||
<Include>./;../../../Common</Include>
|
||||
<Defines>WIN32</Defines>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemDefinitionGroup Condition="'$(Configuration)'=='Debug'">
|
||||
<ClCompile>
|
||||
<Optimization>Disabled</Optimization>
|
||||
<RuntimeLibrary>MultiThreadedDebug</RuntimeLibrary>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<GenerateDebugInformation>true</GenerateDebugInformation>
|
||||
<LinkTimeCodeGeneration>Default</LinkTimeCodeGeneration>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<Runtime>MTd</Runtime>
|
||||
<TargetMachinePlatform>64</TargetMachinePlatform>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemDefinitionGroup Condition="'$(Configuration)'=='Release'">
|
||||
<ClCompile>
|
||||
<Optimization>MaxSpeed</Optimization>
|
||||
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<GenerateDebugInformation>false</GenerateDebugInformation>
|
||||
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<Runtime>MT</Runtime>
|
||||
<TargetMachinePlatform>64</TargetMachinePlatform>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<CudaCompile Include="jacobi.cu" />
|
||||
<ClCompile Include="main.cpp" />
|
||||
<ClInclude Include="jacobi.h" />
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 11.6.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
215
Samples/3_CUDA_Features/jacobiCudaGraphs/main.cpp
Normal file
215
Samples/3_CUDA_Features/jacobiCudaGraphs/main.cpp
Normal file
@@ -0,0 +1,215 @@
|
||||
/* Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
|
||||
*
|
||||
* Redistribution and use in source and binary forms, with or without
|
||||
* modification, are permitted provided that the following conditions
|
||||
* are met:
|
||||
* * Redistributions of source code must retain the above copyright
|
||||
* notice, this list of conditions and the following disclaimer.
|
||||
* * Redistributions in binary form must reproduce the above copyright
|
||||
* notice, this list of conditions and the following disclaimer in the
|
||||
* documentation and/or other materials provided with the distribution.
|
||||
* * Neither the name of NVIDIA CORPORATION nor the names of its
|
||||
* contributors may be used to endorse or promote products derived
|
||||
* from this software without specific prior written permission.
|
||||
*
|
||||
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
|
||||
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
||||
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
||||
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
|
||||
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
||||
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
||||
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
||||
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
|
||||
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
||||
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
||||
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
*/
|
||||
|
||||
// This sample demonstrates Instantiated CUDA Graph Update
|
||||
// with Jacobi Iterative Method in 3 different methods:
|
||||
// 1 - JacobiMethodGpuCudaGraphExecKernelSetParams() - CUDA Graph with
|
||||
// cudaGraphExecKernelNodeSetParams() 2 - JacobiMethodGpuCudaGraphExecUpdate() -
|
||||
// CUDA Graph with cudaGraphExecUpdate() 3 - JacobiMethodGpu() - Non CUDA Graph
|
||||
// method
|
||||
|
||||
// Jacobi method on a linear system A*x = b,
|
||||
// where A is diagonally dominant and the exact solution consists
|
||||
// of all ones.
|
||||
// The dimension N_ROWS is included in jacobi.h
|
||||
|
||||
#include <cuda_runtime.h>
|
||||
#include <helper_cuda.h>
|
||||
#include <helper_timer.h>
|
||||
#include <math.h>
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
#include "jacobi.h"
|
||||
|
||||
// Run the Jacobi method for A*x = b on GPU with CUDA Graph -
|
||||
// cudaGraphExecKernelNodeSetParams().
|
||||
extern double JacobiMethodGpuCudaGraphExecKernelSetParams(
|
||||
const float *A, const double *b, const float conv_threshold,
|
||||
const int max_iter, double *x, double *x_new, cudaStream_t stream);
|
||||
|
||||
// Run the Jacobi method for A*x = b on GPU with Instantiated CUDA Graph Update
|
||||
// API - cudaGraphExecUpdate().
|
||||
extern double JacobiMethodGpuCudaGraphExecUpdate(
|
||||
const float *A, const double *b, const float conv_threshold,
|
||||
const int max_iter, double *x, double *x_new, cudaStream_t stream);
|
||||
|
||||
// Run the Jacobi method for A*x = b on GPU without CUDA Graph.
|
||||
extern double JacobiMethodGpu(const float *A, const double *b,
|
||||
const float conv_threshold, const int max_iter,
|
||||
double *x, double *x_new, cudaStream_t stream);
|
||||
|
||||
// creates N_ROWS x N_ROWS matrix A with N_ROWS+1 on the diagonal and 1
|
||||
// elsewhere. The elements of the right hand side b all equal 2*n, hence the
|
||||
// exact solution x to A*x = b is a vector of ones.
|
||||
void createLinearSystem(float *A, double *b);
|
||||
|
||||
// Run the Jacobi method for A*x = b on CPU.
|
||||
void JacobiMethodCPU(float *A, double *b, float conv_threshold, int max_iter,
|
||||
int *numit, double *x);
|
||||
|
||||
int main(int argc, char **argv) {
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "help")) {
|
||||
printf("Command line: jacobiCudaGraphs [-option]\n");
|
||||
printf("Valid options:\n");
|
||||
printf(
|
||||
"-gpumethod=<0,1 or 2> : 0 - [Default] "
|
||||
"JacobiMethodGpuCudaGraphExecKernelSetParams\n");
|
||||
printf(" : 1 - JacobiMethodGpuCudaGraphExecUpdate\n");
|
||||
printf(" : 2 - JacobiMethodGpu - Non CUDA Graph\n");
|
||||
printf("-device=device_num : cuda device id");
|
||||
printf("-help : Output a help message\n");
|
||||
exit(EXIT_SUCCESS);
|
||||
}
|
||||
|
||||
int gpumethod = 0;
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "gpumethod")) {
|
||||
gpumethod = getCmdLineArgumentInt(argc, (const char **)argv, "gpumethod");
|
||||
|
||||
if (gpumethod < 0 || gpumethod > 2) {
|
||||
printf("Error: gpumethod must be 0 or 1 or 2, gpumethod=%d is invalid\n",
|
||||
gpumethod);
|
||||
exit(EXIT_SUCCESS);
|
||||
}
|
||||
}
|
||||
|
||||
int dev = findCudaDevice(argc, (const char **)argv);
|
||||
|
||||
double *b = NULL;
|
||||
float *A = NULL;
|
||||
checkCudaErrors(cudaMallocHost(&b, N_ROWS * sizeof(double)));
|
||||
memset(b, 0, N_ROWS * sizeof(double));
|
||||
checkCudaErrors(cudaMallocHost(&A, N_ROWS * N_ROWS * sizeof(float)));
|
||||
memset(A, 0, N_ROWS * N_ROWS * sizeof(float));
|
||||
|
||||
createLinearSystem(A, b);
|
||||
double *x = NULL;
|
||||
// start with array of all zeroes
|
||||
x = (double *)calloc(N_ROWS, sizeof(double));
|
||||
|
||||
float conv_threshold = 1.0e-2;
|
||||
int max_iter = 4 * N_ROWS * N_ROWS;
|
||||
int cnt = 0;
|
||||
|
||||
// create timer
|
||||
StopWatchInterface *timerCPU = NULL, *timerGpu = NULL;
|
||||
sdkCreateTimer(&timerCPU);
|
||||
|
||||
sdkStartTimer(&timerCPU);
|
||||
JacobiMethodCPU(A, b, conv_threshold, max_iter, &cnt, x);
|
||||
|
||||
double sum = 0.0;
|
||||
// Compute error
|
||||
for (int i = 0; i < N_ROWS; i++) {
|
||||
double d = x[i] - 1.0;
|
||||
sum += fabs(d);
|
||||
}
|
||||
sdkStopTimer(&timerCPU);
|
||||
printf("CPU iterations : %d\n", cnt);
|
||||
printf("CPU error : %.3e\n", sum);
|
||||
printf("CPU Processing time: %f (ms)\n", sdkGetTimerValue(&timerCPU));
|
||||
|
||||
float *d_A;
|
||||
double *d_b, *d_x, *d_x_new;
|
||||
cudaStream_t stream1;
|
||||
checkCudaErrors(cudaStreamCreateWithFlags(&stream1, cudaStreamNonBlocking));
|
||||
checkCudaErrors(cudaMalloc(&d_b, sizeof(double) * N_ROWS));
|
||||
checkCudaErrors(cudaMalloc(&d_A, sizeof(float) * N_ROWS * N_ROWS));
|
||||
checkCudaErrors(cudaMalloc(&d_x, sizeof(double) * N_ROWS));
|
||||
checkCudaErrors(cudaMalloc(&d_x_new, sizeof(double) * N_ROWS));
|
||||
|
||||
checkCudaErrors(cudaMemsetAsync(d_x, 0, sizeof(double) * N_ROWS, stream1));
|
||||
checkCudaErrors(
|
||||
cudaMemsetAsync(d_x_new, 0, sizeof(double) * N_ROWS, stream1));
|
||||
checkCudaErrors(cudaMemcpyAsync(d_A, A, sizeof(float) * N_ROWS * N_ROWS,
|
||||
cudaMemcpyHostToDevice, stream1));
|
||||
checkCudaErrors(cudaMemcpyAsync(d_b, b, sizeof(double) * N_ROWS,
|
||||
cudaMemcpyHostToDevice, stream1));
|
||||
|
||||
sdkCreateTimer(&timerGpu);
|
||||
sdkStartTimer(&timerGpu);
|
||||
|
||||
double sumGPU = 0.0;
|
||||
if (gpumethod == 0) {
|
||||
sumGPU = JacobiMethodGpuCudaGraphExecKernelSetParams(
|
||||
d_A, d_b, conv_threshold, max_iter, d_x, d_x_new, stream1);
|
||||
} else if (gpumethod == 1) {
|
||||
sumGPU = JacobiMethodGpuCudaGraphExecUpdate(
|
||||
d_A, d_b, conv_threshold, max_iter, d_x, d_x_new, stream1);
|
||||
} else if (gpumethod == 2) {
|
||||
sumGPU = JacobiMethodGpu(d_A, d_b, conv_threshold, max_iter, d_x, d_x_new,
|
||||
stream1);
|
||||
}
|
||||
|
||||
sdkStopTimer(&timerGpu);
|
||||
printf("GPU Processing time: %f (ms)\n", sdkGetTimerValue(&timerGpu));
|
||||
|
||||
checkCudaErrors(cudaFree(d_b));
|
||||
checkCudaErrors(cudaFree(d_A));
|
||||
checkCudaErrors(cudaFree(d_x));
|
||||
checkCudaErrors(cudaFree(d_x_new));
|
||||
|
||||
checkCudaErrors(cudaFreeHost(A));
|
||||
checkCudaErrors(cudaFreeHost(b));
|
||||
|
||||
printf("&&&& jacobiCudaGraphs %s\n",
|
||||
(fabs(sum - sumGPU) < conv_threshold) ? "PASSED" : "FAILED");
|
||||
|
||||
return (fabs(sum - sumGPU) < conv_threshold) ? EXIT_SUCCESS : EXIT_FAILURE;
|
||||
}
|
||||
|
||||
void createLinearSystem(float *A, double *b) {
|
||||
int i, j;
|
||||
for (i = 0; i < N_ROWS; i++) {
|
||||
b[i] = 2.0 * N_ROWS;
|
||||
for (j = 0; j < N_ROWS; j++) A[i * N_ROWS + j] = 1.0;
|
||||
A[i * N_ROWS + i] = N_ROWS + 1.0;
|
||||
}
|
||||
}
|
||||
|
||||
void JacobiMethodCPU(float *A, double *b, float conv_threshold, int max_iter,
|
||||
int *num_iter, double *x) {
|
||||
double *x_new;
|
||||
x_new = (double *)calloc(N_ROWS, sizeof(double));
|
||||
int k;
|
||||
|
||||
for (k = 0; k < max_iter; k++) {
|
||||
double sum = 0.0;
|
||||
for (int i = 0; i < N_ROWS; i++) {
|
||||
double temp_dx = b[i];
|
||||
for (int j = 0; j < N_ROWS; j++) temp_dx -= A[i * N_ROWS + j] * x[j];
|
||||
temp_dx /= A[i * N_ROWS + i];
|
||||
x_new[i] += temp_dx;
|
||||
sum += fabs(temp_dx);
|
||||
}
|
||||
|
||||
for (int i = 0; i < N_ROWS; i++) x[i] = x_new[i];
|
||||
|
||||
if (sum <= conv_threshold) break;
|
||||
}
|
||||
*num_iter = k + 1;
|
||||
free(x_new);
|
||||
}
|
||||
Reference in New Issue
Block a user