mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2026-10-25 22:28:17 +08:00
Updating samples for CUDA 12.5
This commit is contained in:
Rob Nertney
@@ -10,3 +10,5 @@ This sample demonstrates Matrix Transpose. Different performance are shown to a
|
||||
### [UnifiedMemoryPerf](./UnifiedMemoryPerf)
|
||||
This sample demonstrates the performance comparision using matrix multiplication kernel of Unified Memory with/without hints and other types of memory like zero copy buffers, pageable, pagelocked memory performing synchronous and Asynchronous transfers on a single GPU.
|
||||
|
||||
### [cudaGraphsPerfScaling](./cudaGraphsPerfScaling)
|
||||
This sample demonstrates the performance characteristics of cuda graphs. It is focused on how the apis scale with graph size.
|
||||
|
||||
@@ -169,6 +169,19 @@ CCFLAGS :=
|
||||
LDFLAGS :=
|
||||
|
||||
# build flags
|
||||
|
||||
# Link flag for customized HOST_COMPILER with gcc realpath
|
||||
GCC_PATH := $(shell which gcc)
|
||||
ifeq ($(CUSTOM_HOST_COMPILER),1)
|
||||
ifneq ($(filter /%,$(HOST_COMPILER)),)
|
||||
ifneq ($(findstring gcc,$(HOST_COMPILER)),)
|
||||
ifneq ($(GCC_PATH),$(HOST_COMPILER))
|
||||
LDFLAGS += -lstdc++
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_OS),darwin)
|
||||
LDFLAGS += -rpath $(CUDA_PATH)/lib
|
||||
CCFLAGS += -arch $(HOST_ARCH)
|
||||
|
||||
@@ -28,7 +28,7 @@ cudaMemcpy, cudaStreamDestroy, cudaMemPrefetchAsync, cudaFree, cudaMallocHost, c
|
||||
|
||||
## Prerequisites
|
||||
|
||||
Download and install the [CUDA Toolkit 12.4](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
|
||||
Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
|
||||
Make sure the dependencies mentioned in [Dependencies]() section above are installed.
|
||||
|
||||
## Build and Run
|
||||
|
||||
@@ -38,7 +38,7 @@
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
@@ -110,6 +110,6 @@
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
||||
@@ -34,7 +34,7 @@
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
@@ -106,6 +106,6 @@
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
||||
@@ -34,7 +34,7 @@
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
@@ -106,6 +106,6 @@
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
||||
@@ -169,6 +169,19 @@ CCFLAGS :=
|
||||
LDFLAGS :=
|
||||
|
||||
# build flags
|
||||
|
||||
# Link flag for customized HOST_COMPILER with gcc realpath
|
||||
GCC_PATH := $(shell which gcc)
|
||||
ifeq ($(CUSTOM_HOST_COMPILER),1)
|
||||
ifneq ($(filter /%,$(HOST_COMPILER)),)
|
||||
ifneq ($(findstring gcc,$(HOST_COMPILER)),)
|
||||
ifneq ($(GCC_PATH),$(HOST_COMPILER))
|
||||
LDFLAGS += -lstdc++
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_OS),darwin)
|
||||
LDFLAGS += -rpath $(CUDA_PATH)/lib
|
||||
CCFLAGS += -arch $(HOST_ARCH)
|
||||
|
||||
@@ -27,7 +27,7 @@ cudaMemcpy, cudaFree, cudaDeviceSynchronize, cudaMemset, cudaMalloc, cudaGetDevi
|
||||
|
||||
## Prerequisites
|
||||
|
||||
Download and install the [CUDA Toolkit 12.4](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
|
||||
Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
|
||||
|
||||
## Build and Run
|
||||
|
||||
|
||||
@@ -38,7 +38,7 @@
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
@@ -107,6 +107,6 @@
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
||||
@@ -34,7 +34,7 @@
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
@@ -103,6 +103,6 @@
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
||||
@@ -34,7 +34,7 @@
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
@@ -103,6 +103,6 @@
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
||||
18
Samples/6_Performance/cudaGraphsPerfScaling/.vscode/c_cpp_properties.json
vendored
Normal file
18
Samples/6_Performance/cudaGraphsPerfScaling/.vscode/c_cpp_properties.json
vendored
Normal file
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"configurations": [
|
||||
{
|
||||
"name": "Linux",
|
||||
"includePath": [
|
||||
"${workspaceFolder}/**",
|
||||
"${workspaceFolder}/../../../Common"
|
||||
],
|
||||
"defines": [],
|
||||
"compilerPath": "/usr/local/cuda/bin/nvcc",
|
||||
"cStandard": "gnu17",
|
||||
"cppStandard": "gnu++14",
|
||||
"intelliSenseMode": "linux-gcc-x64",
|
||||
"configurationProvider": "ms-vscode.makefile-tools"
|
||||
}
|
||||
],
|
||||
"version": 4
|
||||
}
|
||||
7
Samples/6_Performance/cudaGraphsPerfScaling/.vscode/extensions.json
vendored
Normal file
7
Samples/6_Performance/cudaGraphsPerfScaling/.vscode/extensions.json
vendored
Normal file
@@ -0,0 +1,7 @@
|
||||
{
|
||||
"recommendations": [
|
||||
"nvidia.nsight-vscode-edition",
|
||||
"ms-vscode.cpptools",
|
||||
"ms-vscode.makefile-tools"
|
||||
]
|
||||
}
|
||||
10
Samples/6_Performance/cudaGraphsPerfScaling/.vscode/launch.json
vendored
Normal file
10
Samples/6_Performance/cudaGraphsPerfScaling/.vscode/launch.json
vendored
Normal file
@@ -0,0 +1,10 @@
|
||||
{
|
||||
"configurations": [
|
||||
{
|
||||
"name": "CUDA C++: Launch",
|
||||
"type": "cuda-gdb",
|
||||
"request": "launch",
|
||||
"program": "${workspaceFolder}/cudaGraphsPerfScaling"
|
||||
}
|
||||
]
|
||||
}
|
||||
15
Samples/6_Performance/cudaGraphsPerfScaling/.vscode/tasks.json
vendored
Normal file
15
Samples/6_Performance/cudaGraphsPerfScaling/.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
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
363
Samples/6_Performance/cudaGraphsPerfScaling/Makefile
Normal file
363
Samples/6_Performance/cudaGraphsPerfScaling/Makefile
Normal file
@@ -0,0 +1,363 @@
|
||||
################################################################################
|
||||
# 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 $(TARGET_OS) for TARGET_OS!)
|
||||
endif
|
||||
|
||||
# host compiler
|
||||
ifdef HOST_COMPILER
|
||||
CUSTOM_HOST_COMPILER = 1
|
||||
endif
|
||||
|
||||
ifeq ($(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
|
||||
|
||||
# Link flag for customized HOST_COMPILER with gcc realpath
|
||||
GCC_PATH := $(shell which gcc)
|
||||
ifeq ($(CUSTOM_HOST_COMPILER),1)
|
||||
ifneq ($(filter /%,$(HOST_COMPILER)),)
|
||||
ifneq ($(findstring gcc,$(HOST_COMPILER)),)
|
||||
ifneq ($(GCC_PATH),$(HOST_COMPILER))
|
||||
LDFLAGS += -lstdc++
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
|
||||
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
|
||||
|
||||
# This sample is not supported on Mac OSX
|
||||
ifeq ($(TARGET_OS),darwin)
|
||||
$(info >>> WARNING - cudaGraphsPerfScaling is not supported on Mac OSX - waiving sample <<<)
|
||||
SAMPLE_ENABLED := 0
|
||||
endif
|
||||
|
||||
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 $(TARGET_ARCH),armv7l aarch64 sbsa))
|
||||
SMS ?= 53 61 70 72 75 80 86 87 90
|
||||
else
|
||||
SMS ?= 50 52 60 61 70 75 80 86 89 90
|
||||
endif
|
||||
|
||||
ifeq ($(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 += --std=c++11 --threads 0
|
||||
|
||||
ifeq ($(SAMPLE_ENABLED),0)
|
||||
EXEC ?= @echo "[@]"
|
||||
endif
|
||||
|
||||
################################################################################
|
||||
|
||||
# Target rules
|
||||
all: build
|
||||
|
||||
build: cudaGraphsPerfScaling
|
||||
|
||||
check.deps:
|
||||
ifeq ($(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
|
||||
|
||||
cudaGraphPerfScaling.o:cudaGraphPerfScaling.cu
|
||||
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
|
||||
|
||||
cudaGraphsPerfScaling: cudaGraphPerfScaling.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
|
||||
$(EXEC) ./cudaGraphsPerfScaling
|
||||
|
||||
testrun: build
|
||||
|
||||
clean:
|
||||
rm -f cudaGraphsPerfScaling cudaGraphPerfScaling.o
|
||||
rm -rf ../../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/cudaGraphsPerfScaling
|
||||
|
||||
clobber: clean
|
||||
70
Samples/6_Performance/cudaGraphsPerfScaling/README.md
Normal file
70
Samples/6_Performance/cudaGraphsPerfScaling/README.md
Normal file
@@ -0,0 +1,70 @@
|
||||
# cudaGraphsPerfScaling - Cuda Graphs Perf Scaling
|
||||
|
||||
## Description
|
||||
|
||||
A simple program for characterizing cuda graph api performance with different sized graphs.
|
||||
|
||||
## Key Concepts
|
||||
|
||||
Performance Strategies
|
||||
|
||||
## Supported SM Architectures
|
||||
|
||||
[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) [SM 8.9 ](https://developer.nvidia.com/cuda-gpus) [SM 9.0 ](https://developer.nvidia.com/cuda-gpus)
|
||||
|
||||
## Supported OSes
|
||||
|
||||
Linux, Windows
|
||||
|
||||
## Supported CPU Architecture
|
||||
|
||||
x86_64, armv7l
|
||||
|
||||
## CUDA APIs involved
|
||||
|
||||
### [CUDA Runtime API](http://docs.nvidia.com/cuda/cuda-runtime-api/index.html)
|
||||
cudaStreamBeginCapture, cudaGraphInstantiate, cudaGraphLaunch, cudaGraphUpload
|
||||
|
||||
## Prerequisites
|
||||
|
||||
Download and install the [CUDA Toolkit 12.5](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, 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=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)
|
||||
|
||||
@@ -0,0 +1,434 @@
|
||||
/* Copyright (c) 2024, 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 is a simple application showing the performance characteristics of cudaGraphs.
|
||||
*/
|
||||
|
||||
#define USE_NVTX
|
||||
|
||||
#include <cstdio>
|
||||
#include <cuda_runtime.h>
|
||||
#include <vector>
|
||||
#include <chrono>
|
||||
|
||||
typedef volatile int LatchType;
|
||||
|
||||
std::chrono::time_point<std::chrono::high_resolution_clock> getCpuTime()
|
||||
{
|
||||
return std::chrono::high_resolution_clock::now();
|
||||
}
|
||||
|
||||
template <typename T>
|
||||
float getMicroSecondDuration(T start, T end)
|
||||
{
|
||||
return std::chrono::duration_cast<std::chrono::nanoseconds>(end-start).count() *.001f;
|
||||
}
|
||||
|
||||
float getAsyncMicroSecondDuration(cudaEvent_t start, cudaEvent_t end)
|
||||
{
|
||||
float ms;
|
||||
cudaEventElapsedTime(&ms, start, end);
|
||||
return ms*1000;
|
||||
}
|
||||
|
||||
#ifdef USE_NVTX
|
||||
#include <nvtx3/nvToolsExt.h>
|
||||
|
||||
class Tracer {
|
||||
public:
|
||||
Tracer(const char* name) {
|
||||
nvtxRangePushA(name);
|
||||
}
|
||||
~Tracer() {
|
||||
nvtxRangePop();
|
||||
}
|
||||
};
|
||||
#define RANGE(name) Tracer uniq_name_using_macros(name);
|
||||
#define RANGE_PUSH(name) nvtxRangePushA(name)
|
||||
#define RANGE_POP() nvtxRangePop();
|
||||
#else
|
||||
#define RANGE(name)
|
||||
#endif
|
||||
|
||||
std::vector<cudaStream_t> stream;
|
||||
cudaEvent_t event[1];
|
||||
cudaEvent_t timingEvent[2];
|
||||
|
||||
struct hostData {
|
||||
long long timeElapsed;
|
||||
bool timeoutDetected;
|
||||
long long timeElapsed2;
|
||||
bool timeoutDetected2;
|
||||
LatchType latch;
|
||||
LatchType latch2;
|
||||
};
|
||||
|
||||
struct hostData *hostData;
|
||||
|
||||
__global__ void empty()
|
||||
{
|
||||
}
|
||||
|
||||
// Function to read the GPU nanosecond timer in a kernel
|
||||
__device__ __forceinline__ unsigned long long __globaltimer() {
|
||||
unsigned long long globaltimer;
|
||||
asm volatile ("mov.u64 %0, %globaltimer;" : "=l"(globaltimer));
|
||||
return globaltimer;
|
||||
}
|
||||
|
||||
__global__ void delay(long long ticks)
|
||||
{
|
||||
long long endTime = clock64() + ticks;
|
||||
while (clock64() < endTime);
|
||||
}
|
||||
|
||||
__global__ void waitWithTimeout(long long nanoseconds, bool* timeoutDetected, long long *timeElapsed, LatchType* latch)
|
||||
{
|
||||
long long startTime = __globaltimer();
|
||||
long long endTime = startTime + nanoseconds;
|
||||
long long time = 0;
|
||||
do {
|
||||
time = __globaltimer();
|
||||
} while (time < endTime && (latch == NULL || *latch == 0));
|
||||
if (timeElapsed != NULL) {
|
||||
*timeElapsed = time - startTime;
|
||||
}
|
||||
if (timeoutDetected) {
|
||||
// report timeout if latch not detected
|
||||
*timeoutDetected = (latch == NULL || *latch == 0);
|
||||
}
|
||||
}
|
||||
|
||||
__global__ void preUploadAnnotation()
|
||||
{
|
||||
}
|
||||
|
||||
__global__ void postUploadAnnotation()
|
||||
{
|
||||
}
|
||||
|
||||
cudaGraph_t createParallelChain(int length, int width, bool singleEntry = false)
|
||||
{
|
||||
RANGE_PUSH(__func__);
|
||||
RANGE("capture");
|
||||
cudaGraph_t graph;
|
||||
cudaStreamBeginCapture(stream[0], cudaStreamCaptureModeGlobal);
|
||||
int streamIdx = 0;
|
||||
if (singleEntry) {
|
||||
empty<<<1,1,0,stream[streamIdx]>>>();
|
||||
}
|
||||
|
||||
cudaEventRecord(event[0], stream[0]);
|
||||
for (int i = 1; i < width; i++) {
|
||||
cudaStreamWaitEvent(stream[i], event[0]);
|
||||
}
|
||||
|
||||
for (int i = 0; i < width; i++) {
|
||||
streamIdx = i;
|
||||
for (int j = 0; j < length; j++) {
|
||||
empty<<<1,1,0,stream[streamIdx]>>>();
|
||||
}
|
||||
}
|
||||
|
||||
for (int i = 1; i < width; i++) {
|
||||
cudaEventRecord(event[0], stream[i]);
|
||||
cudaStreamWaitEvent(stream[0], event[0]);
|
||||
}
|
||||
|
||||
cudaStreamEndCapture(stream[0], &graph);
|
||||
return graph;
|
||||
}
|
||||
|
||||
std::vector<const char*> metricName;
|
||||
std::vector<float> metricValue;
|
||||
|
||||
int counter2 = 0;
|
||||
void runDemo(cudaGraph_t graph, int length, int width)
|
||||
{
|
||||
cudaGraphExec_t graphExec;
|
||||
{
|
||||
auto start = getCpuTime();
|
||||
cudaGraphInstantiateWithFlags(&graphExec, graph, 0);
|
||||
auto end = getCpuTime();
|
||||
metricName.push_back("instantiation");
|
||||
metricValue.push_back(getMicroSecondDuration(start, end));
|
||||
}
|
||||
{
|
||||
RANGE("launch including upload");
|
||||
auto start = getCpuTime();
|
||||
cudaGraphLaunch(graphExec, stream[0]);
|
||||
auto apiReturn = getCpuTime();
|
||||
cudaStreamSynchronize(stream[0]);
|
||||
auto streamSync = getCpuTime();
|
||||
metricName.push_back("first_launch_api");
|
||||
metricValue.push_back(getMicroSecondDuration(start, apiReturn));
|
||||
metricName.push_back("first_launch_total");
|
||||
metricValue.push_back(getMicroSecondDuration(start, streamSync));
|
||||
}
|
||||
{
|
||||
RANGE("repeat lauch in empty stream");
|
||||
auto start = getCpuTime();
|
||||
cudaGraphLaunch(graphExec, stream[0]);
|
||||
auto apiReturn = getCpuTime();
|
||||
cudaStreamSynchronize(stream[0]);
|
||||
auto streamSync = getCpuTime();
|
||||
metricName.push_back("repeat_launch_api");
|
||||
metricValue.push_back(getMicroSecondDuration(start, apiReturn));
|
||||
metricName.push_back("repeat_launch_total");
|
||||
metricValue.push_back(getMicroSecondDuration(start, streamSync));
|
||||
}
|
||||
{
|
||||
// re-instantiating the exec to simulate first launch into a busy stream.
|
||||
cudaGraphExecDestroy(graphExec);
|
||||
cudaGraphInstantiateWithFlags(&graphExec, graph, 0);
|
||||
|
||||
long long maxTimeoutNanoSeconds = 4000 + 500*length*width;
|
||||
waitWithTimeout<<<1,1,0,stream[0]>>>(maxTimeoutNanoSeconds, &hostData->timeoutDetected, &hostData->timeElapsed, &hostData->latch);
|
||||
|
||||
RANGE("launch including upload in busy stream");
|
||||
cudaEventRecord(timingEvent[0], stream[0]);
|
||||
cudaGraphLaunch(graphExec, stream[0]);
|
||||
cudaEventRecord(timingEvent[1], stream[0]);
|
||||
|
||||
hostData->latch = 1;
|
||||
cudaStreamSynchronize(stream[0]);
|
||||
|
||||
metricName.push_back("first_launch_device");
|
||||
metricValue.push_back(getAsyncMicroSecondDuration(timingEvent[0], timingEvent[1]));
|
||||
metricName.push_back("blockingKernelTimeoutDetected");
|
||||
metricValue.push_back(hostData->timeoutDetected);
|
||||
hostData->latch = 0;
|
||||
hostData->timeoutDetected = 0;
|
||||
}
|
||||
{
|
||||
RANGE("repeat lauch in busy stream");
|
||||
long long maxTimeoutNanoSeconds = 4000 + 500*length*width;
|
||||
waitWithTimeout<<<1,1,0,stream[0]>>>(maxTimeoutNanoSeconds, &hostData->timeoutDetected, &hostData->timeElapsed, &hostData->latch);
|
||||
cudaEventRecord(timingEvent[0], stream[0]);
|
||||
cudaGraphLaunch(graphExec, stream[0]);
|
||||
cudaEventRecord(timingEvent[1], stream[0]);
|
||||
|
||||
hostData->latch = 1;
|
||||
cudaStreamSynchronize(stream[0]);
|
||||
|
||||
metricName.push_back("repeat_launch_device");
|
||||
metricValue.push_back(getAsyncMicroSecondDuration(timingEvent[0], timingEvent[1]));
|
||||
metricName.push_back("blockingKernelTimeoutDetected");
|
||||
metricValue.push_back(hostData->timeoutDetected);
|
||||
hostData->latch = 0;
|
||||
hostData->timeoutDetected = 0;
|
||||
}
|
||||
{
|
||||
// re-instantiating the exec to provide upload with work to do.
|
||||
cudaGraphExecDestroy(graphExec);
|
||||
cudaGraphInstantiateWithFlags(&graphExec, graph, 0);
|
||||
long long maxTimeoutNanoSeconds = 4000 + 1000*length*width;
|
||||
waitWithTimeout<<<1,1,0,stream[0]>>>(maxTimeoutNanoSeconds, &hostData->timeoutDetected2, &hostData->timeElapsed2, &hostData->latch2);
|
||||
maxTimeoutNanoSeconds = 2000 + 500*length*width;
|
||||
waitWithTimeout<<<1,1,0,stream[1]>>>(maxTimeoutNanoSeconds, &hostData->timeoutDetected, &hostData->timeElapsed, &hostData->latch);
|
||||
|
||||
RANGE("uploading a graph off of the critical path");
|
||||
preUploadAnnotation<<<1,1,0,stream[1]>>>();
|
||||
cudaEventRecord(timingEvent[0], stream[0]);
|
||||
auto start = getCpuTime();
|
||||
cudaGraphUpload(graphExec, stream[1]);
|
||||
auto apiReturn = getCpuTime();
|
||||
cudaEventRecord(event[0],stream[1]);
|
||||
cudaEventRecord(timingEvent[1], stream[0]);
|
||||
postUploadAnnotation<<<1,1,0,stream[1]>>>();
|
||||
|
||||
hostData->latch = 1; // release the blocking kernel for the upload
|
||||
cudaStreamWaitEvent(stream[0],event[0]);
|
||||
cudaGraphLaunch(graphExec, stream[0]);
|
||||
cudaEventSynchronize(event[0]); // upload done, similuate critical path being ready for the graph to run by the release of the second latch
|
||||
|
||||
hostData->latch2 = 1; // release the work
|
||||
cudaStreamSynchronize(stream[0]);
|
||||
|
||||
metricName.push_back("upload_api_time");
|
||||
metricValue.push_back(getMicroSecondDuration(start, apiReturn));
|
||||
metricName.push_back("updoad_device_time");
|
||||
metricValue.push_back(getAsyncMicroSecondDuration(timingEvent[0], timingEvent[1]));
|
||||
metricName.push_back("blockingKernelTimeoutDetected");
|
||||
metricValue.push_back(hostData->timeoutDetected);
|
||||
|
||||
hostData->latch = 0;
|
||||
hostData->latch2 = 0;
|
||||
hostData->timeoutDetected = 0;
|
||||
hostData->timeoutDetected2 = 0;
|
||||
}
|
||||
cudaGraphExecDestroy(graphExec);
|
||||
cudaGraphDestroy(graph);
|
||||
RANGE_POP();
|
||||
}
|
||||
|
||||
void usage() {
|
||||
printf("programName [outputFmt] [numTrials] [length] [width] [pattern] [stride] [maxLength] \n");
|
||||
printf("\toutputFmt - program output, default=3 (see below)\n");
|
||||
printf("\tnumTrials (per length)\n");
|
||||
printf("\tstarting length of the topology\n");
|
||||
printf("\twidth - width of the graph topology\n");
|
||||
printf("\tpattern - Structure of graph, default=0 (see below)\n");
|
||||
printf("\tstride - how to grow the length between each set of trials \n");
|
||||
printf("\tmaxLength - maximum lenght to try \n");
|
||||
printf("\n");
|
||||
printf("outputFmt can be:\n");
|
||||
printf("\t0: this help message\n");
|
||||
printf("\t1: csv data headers\n");
|
||||
printf("\t2: per trial csv data\n");
|
||||
printf("\t3: csv data & headers\n");
|
||||
printf("\t4: csv data is printed and trials are averaged for each length\n");
|
||||
printf("\t5: csv data is printed and trials are averaged for each length and headers are printed\n");
|
||||
printf("\n");
|
||||
printf("Pattern can be:\n");
|
||||
printf("\t0: No interconnect between branches\n");
|
||||
printf("\t1: Adds an extra root node before the initial fork\n");
|
||||
}
|
||||
|
||||
int main(int argc, char **argv)
|
||||
{
|
||||
if(argc < 1) {
|
||||
usage();
|
||||
return 0;
|
||||
}
|
||||
|
||||
int numTrials=1, length=20, width=1, outputFmt=3, pattern=0, stride = 1;
|
||||
if(argc > 1) outputFmt = atoi(argv[1]);
|
||||
if(argc > 2) numTrials = atoi(argv[2]);
|
||||
if(argc > 3) length= atoi(argv[3]);
|
||||
if(argc > 4) width= atoi(argv[4]);
|
||||
if(argc > 5) pattern = atoi(argv[5]);
|
||||
if(argc > 6) stride = atoi(argv[6]);
|
||||
int maxLength = length;
|
||||
if(argc > 7) maxLength = atoi(argv[7]);
|
||||
if (maxLength < length) {
|
||||
maxLength = length;
|
||||
}
|
||||
|
||||
if((outputFmt & 4) && (outputFmt & 2)) {
|
||||
printf("printing average and all samples doesn't make sense\n");
|
||||
}
|
||||
|
||||
if(length == 0 ||
|
||||
width == 0 ||
|
||||
outputFmt == 0 ||
|
||||
outputFmt > 5 ||
|
||||
pattern > 1)
|
||||
{
|
||||
usage();
|
||||
return 0;
|
||||
}
|
||||
|
||||
bool singleEntry = (pattern == 1);
|
||||
|
||||
cudaGraph_t graph;
|
||||
|
||||
cudaFree(0);
|
||||
cudaMallocHost(&hostData, sizeof(*hostData));
|
||||
stream.resize(width);
|
||||
for (int i = 0; i < width; i++)
|
||||
{
|
||||
cudaStreamCreate(&stream[i]);
|
||||
}
|
||||
|
||||
cudaEventCreate(&event[0], cudaEventDisableTiming);
|
||||
cudaEventCreate(&timingEvent[0], 0);
|
||||
cudaEventCreate(&timingEvent[1], 0);
|
||||
|
||||
{
|
||||
RANGE("warmup");
|
||||
for (int i = 0; i < width; i++)
|
||||
{
|
||||
empty<<<1,1,0,stream[i]>>>();
|
||||
}
|
||||
cudaStreamSynchronize(stream[0]);
|
||||
|
||||
auto start = getCpuTime();
|
||||
graph = createParallelChain(length, width, singleEntry);
|
||||
auto end = getCpuTime();
|
||||
metricValue.push_back(getMicroSecondDuration(start, end));
|
||||
metricName.push_back("capture");
|
||||
runDemo(graph, length, width);
|
||||
}
|
||||
|
||||
if (outputFmt & 1) {
|
||||
printf("length, width, pattern, ");
|
||||
for (int i = 0; i < metricName.size(); i++) {
|
||||
printf("%s, ", metricName[i]);
|
||||
}
|
||||
printf("\r\n");
|
||||
}
|
||||
|
||||
if (!(outputFmt & 6)) {
|
||||
printf("skipping trials since no output is expected\n");
|
||||
return;
|
||||
}
|
||||
|
||||
std::vector<double> metricTotal;
|
||||
metricTotal.resize(metricValue.size());
|
||||
|
||||
while (length <= maxLength) {
|
||||
for (int i = 0; i < numTrials; i++) {
|
||||
metricName.clear();
|
||||
metricValue.clear();
|
||||
auto start = getCpuTime();
|
||||
graph = createParallelChain(length, width, singleEntry);
|
||||
auto end = getCpuTime();
|
||||
metricValue.push_back(getMicroSecondDuration(start, end));
|
||||
|
||||
runDemo(graph, length, width);
|
||||
|
||||
if (outputFmt & 2) {
|
||||
printf("%d, %d, %d, ",length, width, pattern);
|
||||
for (int i = 0; i < metricValue.size(); i++) {
|
||||
printf("%0.3f, ", metricValue[i]);
|
||||
}
|
||||
printf("\r\n");
|
||||
}
|
||||
if (outputFmt & 4) {
|
||||
for (int i = 0; i < metricTotal.size(); i++) {
|
||||
metricTotal[i] += metricValue[i];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if (outputFmt & 4) {
|
||||
printf("%d, %d, %d, ",length, width, pattern);
|
||||
for (int i = 0; i < metricTotal.size(); i++) {
|
||||
printf("%0.3f, ", metricTotal[i]/numTrials);
|
||||
metricTotal[i] = 0;
|
||||
}
|
||||
printf("\r\n");
|
||||
}
|
||||
|
||||
length += stride;
|
||||
}
|
||||
|
||||
printf("\n");
|
||||
}
|
||||
|
||||
@@ -0,0 +1,20 @@
|
||||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2017
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "cudaGraphsPerfScaling", "cudaGraphsPerfScaling_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 +1,112 @@
|
||||
<?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>cudaGraphsPerfScaling_vs2017</RootNamespace>
|
||||
<ProjectName>cudaGraphsPerfScaling</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 12.5.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)/cudaGraphsPerfScaling.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration>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;compute_89,sm_89;compute_90,sm_90;</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="cudaGraphPerfScaling.cu" />
|
||||
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.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}") = "cudaGraphsPerfScaling", "cudaGraphsPerfScaling_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 +1,108 @@
|
||||
<?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>cudaGraphsPerfScaling_vs2019</RootNamespace>
|
||||
<ProjectName>cudaGraphsPerfScaling</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="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(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>$(OutDir)/cudaGraphsPerfScaling.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration>compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;compute_89,sm_89;compute_90,sm_90;</CodeGeneration>
|
||||
<AdditionalOptions>-Xcompiler "/wd 4819" --threads 0 </AdditionalOptions>
|
||||
<Include>./;../../../Common</Include>
|
||||
<Defines>WIN32</Defines>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemDefinitionGroup Condition="'$(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="'$(Configuration)'=='Release'">
|
||||
<ClCompile>
|
||||
<Optimization>MaxSpeed</Optimization>
|
||||
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<GenerateDebugInformation>false</GenerateDebugInformation>
|
||||
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<Runtime>MT</Runtime>
|
||||
<TargetMachinePlatform>64</TargetMachinePlatform>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<CudaCompile Include="cudaGraphPerfScaling.cu" />
|
||||
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
@@ -0,0 +1,20 @@
|
||||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2022
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "cudaGraphsPerfScaling", "cudaGraphsPerfScaling_vs2022.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
EndProject
|
||||
Global
|
||||
GlobalSection(SolutionConfigurationPlatforms) = preSolution
|
||||
Debug|x64 = Debug|x64
|
||||
Release|x64 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(ProjectConfigurationPlatforms) = postSolution
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(SolutionProperties) = preSolution
|
||||
HideSolutionNode = FALSE
|
||||
EndGlobalSection
|
||||
EndGlobal
|
||||
@@ -0,0 +1,108 @@
|
||||
<?xml version="1.0" encoding="utf-8"?>
|
||||
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
|
||||
<PropertyGroup>
|
||||
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
|
||||
</PropertyGroup>
|
||||
<ItemGroup Label="ProjectConfigurations">
|
||||
<ProjectConfiguration Include="Debug|x64">
|
||||
<Configuration>Debug</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
<ProjectConfiguration Include="Release|x64">
|
||||
<Configuration>Release</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
</ItemGroup>
|
||||
<PropertyGroup Label="Globals">
|
||||
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
|
||||
<RootNamespace>cudaGraphsPerfScaling_vs2022</RootNamespace>
|
||||
<ProjectName>cudaGraphsPerfScaling</ProjectName>
|
||||
<CudaToolkitCustomDir />
|
||||
</PropertyGroup>
|
||||
<Import Project="$(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 12.5.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)/cudaGraphsPerfScaling.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration>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;compute_89,sm_89;compute_90,sm_90;</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="cudaGraphPerfScaling.cu" />
|
||||
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
@@ -0,0 +1,17 @@
|
||||
GPU=$1
|
||||
DRIVER_VERSION=$2
|
||||
BINARY=./cudaGraphsPerfScaling
|
||||
datadir=PERF_DATA
|
||||
|
||||
suffix=$DRIVER_VERSION
|
||||
prefix=$GPU
|
||||
mkdir -p $datadir
|
||||
|
||||
trials=600
|
||||
|
||||
width=1
|
||||
nvidia-smi > $datadir/${prefix}_info_${suffix}.txt
|
||||
$BINARY 5 $trials 1 $width 0 1 256 > $datadir/${prefix}_${width}_small_${suffix}.csv
|
||||
$BINARY 5 $trials 1 $width 0 32 2048 > $datadir/${prefix}_${width}_large_${suffix}.csv
|
||||
width=4
|
||||
$BINARY 5 $trials 1 $width 0 1 256 > $datadir/${prefix}_${width}_small_${suffix}.csv
|
||||
@@ -169,6 +169,19 @@ CCFLAGS :=
|
||||
LDFLAGS :=
|
||||
|
||||
# build flags
|
||||
|
||||
# Link flag for customized HOST_COMPILER with gcc realpath
|
||||
GCC_PATH := $(shell which gcc)
|
||||
ifeq ($(CUSTOM_HOST_COMPILER),1)
|
||||
ifneq ($(filter /%,$(HOST_COMPILER)),)
|
||||
ifneq ($(findstring gcc,$(HOST_COMPILER)),)
|
||||
ifneq ($(GCC_PATH),$(HOST_COMPILER))
|
||||
LDFLAGS += -lstdc++
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_OS),darwin)
|
||||
LDFLAGS += -rpath $(CUDA_PATH)/lib
|
||||
CCFLAGS += -arch $(HOST_ARCH)
|
||||
|
||||
@@ -27,7 +27,7 @@ cudaMemcpy, cudaMalloc, cudaFree, cudaGetLastError, cudaEventSynchronize, cudaEv
|
||||
|
||||
## Prerequisites
|
||||
|
||||
Download and install the [CUDA Toolkit 12.4](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
|
||||
Download and install the [CUDA Toolkit 12.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
|
||||
|
||||
## Build and Run
|
||||
|
||||
|
||||
@@ -38,7 +38,7 @@
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
@@ -107,6 +107,6 @@
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
||||
@@ -34,7 +34,7 @@
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
@@ -103,6 +103,6 @@
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
||||
@@ -34,7 +34,7 @@
|
||||
</PropertyGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
|
||||
<ImportGroup Label="ExtensionSettings">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.props" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.props" />
|
||||
</ImportGroup>
|
||||
<ImportGroup Label="PropertySheets">
|
||||
<Import Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" />
|
||||
@@ -103,6 +103,6 @@
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.4.targets" />
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 12.5.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
|
||||
Reference in New Issue
Block a user