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/2_Concepts_and_Techniques/threadFenceReduction/.vscode/c_cpp_properties.json
vendored
Normal file
18
Samples/2_Concepts_and_Techniques/threadFenceReduction/.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/2_Concepts_and_Techniques/threadFenceReduction/.vscode/extensions.json
vendored
Normal file
7
Samples/2_Concepts_and_Techniques/threadFenceReduction/.vscode/extensions.json
vendored
Normal file
@@ -0,0 +1,7 @@
|
||||
{
|
||||
"recommendations": [
|
||||
"nvidia.nsight-vscode-edition",
|
||||
"ms-vscode.cpptools",
|
||||
"ms-vscode.makefile-tools"
|
||||
]
|
||||
}
|
||||
10
Samples/2_Concepts_and_Techniques/threadFenceReduction/.vscode/launch.json
vendored
Normal file
10
Samples/2_Concepts_and_Techniques/threadFenceReduction/.vscode/launch.json
vendored
Normal file
@@ -0,0 +1,10 @@
|
||||
{
|
||||
"configurations": [
|
||||
{
|
||||
"name": "CUDA C++: Launch",
|
||||
"type": "cuda-gdb",
|
||||
"request": "launch",
|
||||
"program": "${workspaceFolder}/threadFenceReduction"
|
||||
}
|
||||
]
|
||||
}
|
||||
15
Samples/2_Concepts_and_Techniques/threadFenceReduction/.vscode/tasks.json
vendored
Normal file
15
Samples/2_Concepts_and_Techniques/threadFenceReduction/.vscode/tasks.json
vendored
Normal file
@@ -0,0 +1,15 @@
|
||||
{
|
||||
"version": "2.0.0",
|
||||
"tasks": [
|
||||
{
|
||||
"label": "sample",
|
||||
"type": "shell",
|
||||
"command": "make dbg=1",
|
||||
"problemMatcher": ["$nvcc"],
|
||||
"group": {
|
||||
"kind": "build",
|
||||
"isDefault": true
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
340
Samples/2_Concepts_and_Techniques/threadFenceReduction/Makefile
Normal file
340
Samples/2_Concepts_and_Techniques/threadFenceReduction/Makefile
Normal file
@@ -0,0 +1,340 @@
|
||||
################################################################################
|
||||
# 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 ($(SAMPLE_ENABLED),0)
|
||||
EXEC ?= @echo "[@]"
|
||||
endif
|
||||
|
||||
################################################################################
|
||||
|
||||
# Target rules
|
||||
all: build
|
||||
|
||||
build: threadFenceReduction
|
||||
|
||||
check.deps:
|
||||
ifeq ($(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
|
||||
|
||||
threadFenceReduction.o:threadFenceReduction.cu
|
||||
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
|
||||
|
||||
threadFenceReduction: threadFenceReduction.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
|
||||
$(EXEC) ./threadFenceReduction
|
||||
|
||||
testrun: build
|
||||
|
||||
clean:
|
||||
rm -f threadFenceReduction threadFenceReduction.o
|
||||
rm -rf ../../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/threadFenceReduction
|
||||
|
||||
clobber: clean
|
||||
@@ -0,0 +1,79 @@
|
||||
<?xml version="1.0" encoding="UTF-8"?>
|
||||
<!DOCTYPE entry SYSTEM "SamplesInfo.dtd">
|
||||
<entry>
|
||||
<name>threadFenceReduction</name>
|
||||
<cuda_api_list>
|
||||
<toolkit>cudaFree</toolkit>
|
||||
<toolkit>cudaDeviceSynchronize</toolkit>
|
||||
<toolkit>cudaMalloc</toolkit>
|
||||
<toolkit>cudaMemcpy</toolkit>
|
||||
<toolkit>cudaGetDeviceProperties</toolkit>
|
||||
</cuda_api_list>
|
||||
<description><![CDATA[This sample shows how to perform a reduction operation on an array of values using the thread Fence intrinsic to produce a single value in a single kernel (as opposed to two or more kernel calls as shown in the "reduction" CUDA Sample). Single-pass reduction requires global atomic instructions (Compute Capability 2.0 or later) and the _threadfence() intrinsic (CUDA 2.2 or later).]]></description>
|
||||
<devicecompilation>whole</devicecompilation>
|
||||
<includepaths>
|
||||
<path>./</path>
|
||||
<path>../</path>
|
||||
<path>../../../Common</path>
|
||||
</includepaths>
|
||||
<keyconcepts>
|
||||
<concept level="basic">Cooperative Groups</concept>
|
||||
<concept level="advanced">Data-Parallel Algorithms</concept>
|
||||
<concept level="advanced">Performance Strategies</concept>
|
||||
</keyconcepts>
|
||||
<keywords>
|
||||
<keyword>reduction</keyword>
|
||||
</keywords>
|
||||
<libraries>
|
||||
</libraries>
|
||||
<librarypaths>
|
||||
</librarypaths>
|
||||
<nsight_eclipse>true</nsight_eclipse>
|
||||
<primary_file>threadFenceReduction.cu</primary_file>
|
||||
<scopes>
|
||||
<scope>1:CUDA Advanced Topics</scope>
|
||||
<scope>1:Data-Parallel Algorithms</scope>
|
||||
<scope>1:Performance Strategies</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>
|
||||
<include>all</include>
|
||||
</supported_sm_architectures>
|
||||
<title>threadFenceReduction</title>
|
||||
<type>exe</type>
|
||||
</entry>
|
||||
@@ -0,0 +1,70 @@
|
||||
# threadFenceReduction - threadFenceReduction
|
||||
|
||||
## Description
|
||||
|
||||
This sample shows how to perform a reduction operation on an array of values using the thread Fence intrinsic to produce a single value in a single kernel (as opposed to two or more kernel calls as shown in the "reduction" CUDA Sample). Single-pass reduction requires global atomic instructions (Compute Capability 2.0 or later) and the _threadfence() intrinsic (CUDA 2.2 or later).
|
||||
|
||||
## Key Concepts
|
||||
|
||||
Cooperative Groups, Data-Parallel Algorithms, Performance Strategies
|
||||
|
||||
## 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)
|
||||
cudaFree, cudaDeviceSynchronize, cudaMalloc, cudaMemcpy, cudaGetDeviceProperties
|
||||
|
||||
## 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)
|
||||
|
||||
@@ -0,0 +1,465 @@
|
||||
/* 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.
|
||||
*/
|
||||
|
||||
/*
|
||||
Parallel reduction
|
||||
|
||||
This sample shows how to perform a reduction operation on an array of values
|
||||
to produce a single value in a single kernel (as opposed to two or more
|
||||
kernel calls as shown in the "reduction" CUDA Sample). Single-pass
|
||||
reduction requires global atomic instructions (Compute Capability 1.1 or
|
||||
later) and the __threadfence() intrinsic (CUDA 2.2 or later).
|
||||
|
||||
Reductions are a very common computation in parallel algorithms. Any time
|
||||
an array of values needs to be reduced to a single value using a binary
|
||||
associative operator, a reduction can be used. Example applications include
|
||||
statistics computations such as mean and standard deviation, and image
|
||||
processing applications such as finding the total luminance of an
|
||||
image.
|
||||
|
||||
This code performs sum reductions, but any associative operator such as
|
||||
min() or max() could also be used.
|
||||
|
||||
It assumes the input size is a power of 2.
|
||||
|
||||
COMMAND LINE ARGUMENTS
|
||||
|
||||
"--shmoo": Test performance for 1 to 32M elements with each of the
|
||||
7 different kernels
|
||||
"--n=<N>": Specify the number of elements to reduce (default 1048576)
|
||||
"--threads=<N>": Specify the number of threads per block (default 128)
|
||||
"--maxblocks=<N>": Specify the maximum number of thread blocks to launch
|
||||
(kernel 6 only, default 64)
|
||||
"--cpufinal": Read back the per-block results and do final sum of block
|
||||
sums on CPU (default false)
|
||||
"--cputhresh=<N>": The threshold of number of blocks sums below which to
|
||||
perform a CPU final reduction (default 1)
|
||||
"--multipass": Use a multipass reduction instead of a single-pass reduction
|
||||
*/
|
||||
|
||||
// includes, system
|
||||
#include <stdlib.h>
|
||||
#include <stdio.h>
|
||||
#include <string.h>
|
||||
#include <math.h>
|
||||
|
||||
// includes, project
|
||||
#include <helper_functions.h>
|
||||
#include <helper_cuda.h>
|
||||
|
||||
#define VERSION_MAJOR (CUDART_VERSION / 1000)
|
||||
#define VERSION_MINOR (CUDART_VERSION % 100) / 10
|
||||
|
||||
const char *sSDKsample = "threadFenceReduction";
|
||||
|
||||
#if CUDART_VERSION >= 2020
|
||||
#include "threadFenceReduction_kernel.cuh"
|
||||
#else
|
||||
#pragma comment(user, "CUDA 2.2 is required to build for threadFenceReduction")
|
||||
#endif
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// declaration, forward
|
||||
bool runTest(int argc, char **argv);
|
||||
|
||||
extern "C" {
|
||||
void reduce(int size, int threads, int blocks, float *d_idata, float *d_odata);
|
||||
void reduceSinglePass(int size, int threads, int blocks, float *d_idata,
|
||||
float *d_odata);
|
||||
}
|
||||
|
||||
#if CUDART_VERSION < 2020
|
||||
void reduce(int size, int threads, int blocks, float *d_idata, float *d_odata) {
|
||||
printf("reduce(), compiler not supported, aborting tests\n");
|
||||
}
|
||||
|
||||
void reduceSinglePass(int size, int threads, int blocks, float *d_idata,
|
||||
float *d_odata) {
|
||||
printf("reduceSinglePass(), compiler not supported, aborting tests\n");
|
||||
}
|
||||
#endif
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// Program main
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
int main(int argc, char **argv) {
|
||||
cudaDeviceProp deviceProp;
|
||||
deviceProp.major = 0;
|
||||
deviceProp.minor = 0;
|
||||
int dev;
|
||||
|
||||
printf("%s Starting...\n\n", sSDKsample);
|
||||
|
||||
dev = findCudaDevice(argc, (const char **)argv);
|
||||
|
||||
checkCudaErrors(cudaGetDeviceProperties(&deviceProp, dev));
|
||||
|
||||
printf("GPU Device supports SM %d.%d compute capability\n\n",
|
||||
deviceProp.major, deviceProp.minor);
|
||||
|
||||
bool bTestResult = false;
|
||||
|
||||
#if CUDART_VERSION >= 2020
|
||||
bTestResult = runTest(argc, argv);
|
||||
#else
|
||||
print_NVCC_min_spec(sSDKsample, "2.2", "Version 185");
|
||||
exit(EXIT_SUCCESS);
|
||||
#endif
|
||||
|
||||
exit(bTestResult ? EXIT_SUCCESS : EXIT_FAILURE);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
//! Compute sum reduction on CPU
|
||||
//! We use Kahan summation for an accurate sum of large arrays.
|
||||
//! http://en.wikipedia.org/wiki/Kahan_summation_algorithm
|
||||
//!
|
||||
//! @param data pointer to input data
|
||||
//! @param size number of input data elements
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
template <class T>
|
||||
T reduceCPU(T *data, int size) {
|
||||
T sum = data[0];
|
||||
T c = (T)0.0;
|
||||
|
||||
for (int i = 1; i < size; i++) {
|
||||
T y = data[i] - c;
|
||||
T t = sum + y;
|
||||
c = (t - sum) - y;
|
||||
sum = t;
|
||||
}
|
||||
|
||||
return sum;
|
||||
}
|
||||
|
||||
unsigned int nextPow2(unsigned int x) {
|
||||
--x;
|
||||
x |= x >> 1;
|
||||
x |= x >> 2;
|
||||
x |= x >> 4;
|
||||
x |= x >> 8;
|
||||
x |= x >> 16;
|
||||
return ++x;
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// Compute the number of threads and blocks to use for the reduction
|
||||
// We set threads / block to the minimum of maxThreads and n/2.
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
void getNumBlocksAndThreads(int n, int maxBlocks, int maxThreads, int &blocks,
|
||||
int &threads) {
|
||||
if (n == 1) {
|
||||
threads = 1;
|
||||
blocks = 1;
|
||||
} else {
|
||||
threads = (n < maxThreads * 2) ? nextPow2(n / 2) : maxThreads;
|
||||
blocks = max(1, n / (threads * 2));
|
||||
}
|
||||
|
||||
blocks = min(maxBlocks, blocks);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// This function performs a reduction of the input data multiple times and
|
||||
// measures the average reduction time.
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
float benchmarkReduce(int n, int numThreads, int numBlocks, int maxThreads,
|
||||
int maxBlocks, int testIterations, bool multiPass,
|
||||
bool cpuFinalReduction, int cpuFinalThreshold,
|
||||
StopWatchInterface *timer, float *h_odata, float *d_idata,
|
||||
float *d_odata) {
|
||||
float gpu_result = 0;
|
||||
bool bNeedReadback = true;
|
||||
cudaError_t error;
|
||||
|
||||
for (int i = 0; i < testIterations; ++i) {
|
||||
gpu_result = 0;
|
||||
unsigned int retCnt = 0;
|
||||
error = setRetirementCount(retCnt);
|
||||
checkCudaErrors(error);
|
||||
|
||||
cudaDeviceSynchronize();
|
||||
sdkStartTimer(&timer);
|
||||
|
||||
if (multiPass) {
|
||||
// execute the kernel
|
||||
reduce(n, numThreads, numBlocks, d_idata, d_odata);
|
||||
|
||||
// check if kernel execution generated an error
|
||||
getLastCudaError("Kernel execution failed");
|
||||
|
||||
if (cpuFinalReduction) {
|
||||
// sum partial sums from each block on CPU
|
||||
// copy result from device to host
|
||||
error = cudaMemcpy(h_odata, d_odata, numBlocks * sizeof(float),
|
||||
cudaMemcpyDeviceToHost);
|
||||
checkCudaErrors(error);
|
||||
|
||||
for (int i = 0; i < numBlocks; i++) {
|
||||
gpu_result += h_odata[i];
|
||||
}
|
||||
|
||||
bNeedReadback = false;
|
||||
} else {
|
||||
// sum partial block sums on GPU
|
||||
int s = numBlocks;
|
||||
|
||||
while (s > cpuFinalThreshold) {
|
||||
int threads = 0, blocks = 0;
|
||||
getNumBlocksAndThreads(s, maxBlocks, maxThreads, blocks, threads);
|
||||
|
||||
reduce(s, threads, blocks, d_odata, d_odata);
|
||||
|
||||
s = s / (threads * 2);
|
||||
}
|
||||
|
||||
if (s > 1) {
|
||||
// copy result from device to host
|
||||
error = cudaMemcpy(h_odata, d_odata, s * sizeof(float),
|
||||
cudaMemcpyDeviceToHost);
|
||||
checkCudaErrors(error);
|
||||
|
||||
for (int i = 0; i < s; i++) {
|
||||
gpu_result += h_odata[i];
|
||||
}
|
||||
|
||||
bNeedReadback = false;
|
||||
}
|
||||
}
|
||||
} else {
|
||||
getLastCudaError("Kernel execution failed");
|
||||
|
||||
// execute the kernel
|
||||
reduceSinglePass(n, numThreads, numBlocks, d_idata, d_odata);
|
||||
|
||||
// check if kernel execution generated an error
|
||||
getLastCudaError("Kernel execution failed");
|
||||
}
|
||||
|
||||
cudaDeviceSynchronize();
|
||||
sdkStopTimer(&timer);
|
||||
}
|
||||
|
||||
if (bNeedReadback) {
|
||||
// copy final sum from device to host
|
||||
error =
|
||||
cudaMemcpy(&gpu_result, d_odata, sizeof(float), cudaMemcpyDeviceToHost);
|
||||
checkCudaErrors(error);
|
||||
}
|
||||
|
||||
return gpu_result;
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// This function calls benchmarkReduce multiple times for a range of array sizes
|
||||
// and prints a report in CSV (comma-separated value) format that can be used
|
||||
// for generating a "shmoo" plot showing the performance for each kernel
|
||||
// variation over a wide range of input sizes.
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
void shmoo(int minN, int maxN, int maxThreads, int maxBlocks) {
|
||||
// create random input data on CPU
|
||||
unsigned int bytes = maxN * sizeof(float);
|
||||
|
||||
float *h_idata = (float *)malloc(bytes);
|
||||
|
||||
for (int i = 0; i < maxN; i++) {
|
||||
// Keep the numbers small so we don't get truncation error in the sum
|
||||
h_idata[i] = (rand() & 0xFF) / (float)RAND_MAX;
|
||||
}
|
||||
|
||||
int maxNumBlocks = min(65535, maxN / maxThreads);
|
||||
|
||||
// allocate mem for the result on host side
|
||||
float *h_odata = (float *)malloc(maxNumBlocks * sizeof(float));
|
||||
|
||||
// allocate device memory and data
|
||||
float *d_idata = NULL;
|
||||
float *d_odata = NULL;
|
||||
|
||||
checkCudaErrors(cudaMalloc((void **)&d_idata, bytes));
|
||||
checkCudaErrors(cudaMalloc((void **)&d_odata, maxNumBlocks * sizeof(float)));
|
||||
|
||||
// copy data directly to device memory
|
||||
checkCudaErrors(cudaMemcpy(d_idata, h_idata, bytes, cudaMemcpyHostToDevice));
|
||||
checkCudaErrors(cudaMemcpy(d_odata, h_idata, maxNumBlocks * sizeof(float),
|
||||
cudaMemcpyHostToDevice));
|
||||
|
||||
// warm-up
|
||||
reduce(maxN, maxThreads, maxNumBlocks, d_idata, d_odata);
|
||||
int testIterations = 100;
|
||||
|
||||
StopWatchInterface *timer = NULL;
|
||||
sdkCreateTimer(&timer);
|
||||
|
||||
// print headers
|
||||
printf("N, %d blocks one pass, %d blocks multipass\n", maxBlocks, maxBlocks);
|
||||
|
||||
for (int i = minN; i <= maxN; i *= 2) {
|
||||
printf("%d, ", i);
|
||||
|
||||
for (int multiPass = 0; multiPass <= 1; multiPass++) {
|
||||
sdkResetTimer(&timer);
|
||||
int numBlocks = 0;
|
||||
int numThreads = 0;
|
||||
getNumBlocksAndThreads(i, maxBlocks, maxThreads, numBlocks, numThreads);
|
||||
|
||||
benchmarkReduce(i, numThreads, numBlocks, maxThreads, maxBlocks,
|
||||
testIterations, multiPass == 1, false, 1, timer, h_odata,
|
||||
d_idata, d_odata);
|
||||
|
||||
float reduceTime = sdkGetAverageTimerValue(&timer);
|
||||
printf("%f%s", reduceTime, multiPass == 0 ? ", " : "\n");
|
||||
}
|
||||
}
|
||||
|
||||
printf("\n");
|
||||
|
||||
// cleanup
|
||||
sdkDeleteTimer(&timer);
|
||||
free(h_idata);
|
||||
free(h_odata);
|
||||
|
||||
cudaFree(d_idata);
|
||||
cudaFree(d_odata);
|
||||
}
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// The main function which runs the reduction test.
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
bool runTest(int argc, char **argv) {
|
||||
int size = 1 << 20; // number of elements to reduce
|
||||
int maxThreads = 128; // number of threads per block
|
||||
int maxBlocks = 64;
|
||||
bool cpuFinalReduction = false;
|
||||
int cpuFinalThreshold = 1;
|
||||
bool multipass = false;
|
||||
bool bTestResult = false;
|
||||
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "n")) {
|
||||
size = getCmdLineArgumentInt(argc, (const char **)argv, "n");
|
||||
}
|
||||
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "threads")) {
|
||||
maxThreads = getCmdLineArgumentInt(argc, (const char **)argv, "threads");
|
||||
}
|
||||
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "maxblocks")) {
|
||||
maxBlocks = getCmdLineArgumentInt(argc, (const char **)argv, "maxblocks");
|
||||
}
|
||||
|
||||
printf("%d elements\n", size);
|
||||
printf("%d threads (max)\n", maxThreads);
|
||||
|
||||
cpuFinalReduction = checkCmdLineFlag(argc, (const char **)argv, "cpufinal");
|
||||
multipass = checkCmdLineFlag(argc, (const char **)argv, "multipass");
|
||||
|
||||
if (checkCmdLineFlag(argc, (const char **)argv, "cputhresh")) {
|
||||
cpuFinalThreshold =
|
||||
getCmdLineArgumentInt(argc, (const char **)argv, "cputhresh");
|
||||
}
|
||||
|
||||
bool runShmoo = checkCmdLineFlag(argc, (const char **)argv, "shmoo");
|
||||
|
||||
if (runShmoo) {
|
||||
shmoo(1, 33554432, maxThreads, maxBlocks);
|
||||
} else {
|
||||
// create random input data on CPU
|
||||
unsigned int bytes = size * sizeof(float);
|
||||
|
||||
float *h_idata = (float *)malloc(bytes);
|
||||
|
||||
for (int i = 0; i < size; i++) {
|
||||
// Keep the numbers small so we don't get truncation error in the sum
|
||||
h_idata[i] = (rand() & 0xFF) / (float)RAND_MAX;
|
||||
}
|
||||
|
||||
int numBlocks = 0;
|
||||
int numThreads = 0;
|
||||
getNumBlocksAndThreads(size, maxBlocks, maxThreads, numBlocks, numThreads);
|
||||
|
||||
if (numBlocks == 1) {
|
||||
cpuFinalThreshold = 1;
|
||||
}
|
||||
|
||||
// allocate mem for the result on host side
|
||||
float *h_odata = (float *)malloc(numBlocks * sizeof(float));
|
||||
|
||||
printf("%d blocks\n", numBlocks);
|
||||
|
||||
// allocate device memory and data
|
||||
float *d_idata = NULL;
|
||||
float *d_odata = NULL;
|
||||
|
||||
checkCudaErrors(cudaMalloc((void **)&d_idata, bytes));
|
||||
checkCudaErrors(cudaMalloc((void **)&d_odata, numBlocks * sizeof(float)));
|
||||
|
||||
// copy data directly to device memory
|
||||
checkCudaErrors(
|
||||
cudaMemcpy(d_idata, h_idata, bytes, cudaMemcpyHostToDevice));
|
||||
checkCudaErrors(cudaMemcpy(d_odata, h_idata, numBlocks * sizeof(float),
|
||||
cudaMemcpyHostToDevice));
|
||||
|
||||
// warm-up
|
||||
reduce(size, numThreads, numBlocks, d_idata, d_odata);
|
||||
int testIterations = 100;
|
||||
|
||||
StopWatchInterface *timer = 0;
|
||||
sdkCreateTimer(&timer);
|
||||
|
||||
float gpu_result = 0;
|
||||
|
||||
gpu_result =
|
||||
benchmarkReduce(size, numThreads, numBlocks, maxThreads, maxBlocks,
|
||||
testIterations, multipass, cpuFinalReduction,
|
||||
cpuFinalThreshold, timer, h_odata, d_idata, d_odata);
|
||||
|
||||
float reduceTime = sdkGetAverageTimerValue(&timer);
|
||||
printf("Average time: %f ms\n", reduceTime);
|
||||
printf("Bandwidth: %f GB/s\n\n",
|
||||
(size * sizeof(int)) / (reduceTime * 1.0e6));
|
||||
|
||||
// compute reference solution
|
||||
float cpu_result = reduceCPU<float>(h_idata, size);
|
||||
|
||||
printf("GPU result = %0.12f\n", gpu_result);
|
||||
printf("CPU result = %0.12f\n", cpu_result);
|
||||
|
||||
double threshold = 1e-8 * size;
|
||||
double diff = abs((double)gpu_result - (double)cpu_result);
|
||||
bTestResult = (diff < threshold);
|
||||
|
||||
// cleanup
|
||||
sdkDeleteTimer(&timer);
|
||||
|
||||
free(h_idata);
|
||||
free(h_odata);
|
||||
cudaFree(d_idata);
|
||||
cudaFree(d_odata);
|
||||
}
|
||||
|
||||
return bTestResult;
|
||||
}
|
||||
@@ -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 __REDUCTION_H__
|
||||
#define __REDUCTION_H__
|
||||
|
||||
cudaError_t setRetirementCount(int retCnt);
|
||||
|
||||
#endif
|
||||
@@ -0,0 +1,440 @@
|
||||
/* 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.
|
||||
*/
|
||||
|
||||
/*
|
||||
Parallel reduction kernels
|
||||
*/
|
||||
|
||||
#ifndef _REDUCE_KERNEL_H_
|
||||
#define _REDUCE_KERNEL_H_
|
||||
|
||||
#include <cuda_runtime_api.h>
|
||||
#include <cooperative_groups.h>
|
||||
|
||||
namespace cg = cooperative_groups;
|
||||
|
||||
/*
|
||||
Parallel sum reduction using shared memory
|
||||
- takes log(n) steps for n input elements
|
||||
- uses n/2 threads
|
||||
- only works for power-of-2 arrays
|
||||
|
||||
This version adds multiple elements per thread sequentially. This reduces
|
||||
the overall cost of the algorithm while keeping the work complexity O(n) and
|
||||
the step complexity O(log n). (Brent's Theorem optimization)
|
||||
|
||||
See the CUDA SDK "reduction" sample for more information.
|
||||
*/
|
||||
|
||||
template <unsigned int blockSize>
|
||||
__device__ void reduceBlock(volatile float *sdata, float mySum,
|
||||
const unsigned int tid, cg::thread_block cta) {
|
||||
cg::thread_block_tile<32> tile32 = cg::tiled_partition<32>(cta);
|
||||
sdata[tid] = mySum;
|
||||
cg::sync(tile32);
|
||||
|
||||
const int VEC = 32;
|
||||
const int vid = tid & (VEC - 1);
|
||||
|
||||
float beta = mySum;
|
||||
float temp;
|
||||
|
||||
for (int i = VEC / 2; i > 0; i >>= 1) {
|
||||
if (vid < i) {
|
||||
temp = sdata[tid + i];
|
||||
beta += temp;
|
||||
sdata[tid] = beta;
|
||||
}
|
||||
cg::sync(tile32);
|
||||
}
|
||||
cg::sync(cta);
|
||||
|
||||
if (cta.thread_rank() == 0) {
|
||||
beta = 0;
|
||||
for (int i = 0; i < blockDim.x; i += VEC) {
|
||||
beta += sdata[i];
|
||||
}
|
||||
sdata[0] = beta;
|
||||
}
|
||||
cg::sync(cta);
|
||||
}
|
||||
|
||||
template <unsigned int blockSize, bool nIsPow2>
|
||||
__device__ void reduceBlocks(const float *g_idata, float *g_odata,
|
||||
unsigned int n, cg::thread_block cta) {
|
||||
extern __shared__ float sdata[];
|
||||
|
||||
// perform first level of reduction,
|
||||
// reading from global memory, writing to shared memory
|
||||
unsigned int tid = threadIdx.x;
|
||||
unsigned int i = blockIdx.x * (blockSize * 2) + threadIdx.x;
|
||||
unsigned int gridSize = blockSize * 2 * gridDim.x;
|
||||
float mySum = 0;
|
||||
|
||||
// we reduce multiple elements per thread. The number is determined by the
|
||||
// number of active thread blocks (via gridDim). More blocks will result
|
||||
// in a larger gridSize and therefore fewer elements per thread
|
||||
while (i < n) {
|
||||
mySum += g_idata[i];
|
||||
|
||||
// ensure we don't read out of bounds -- this is optimized away for powerOf2
|
||||
// sized arrays
|
||||
if (nIsPow2 || i + blockSize < n) mySum += g_idata[i + blockSize];
|
||||
|
||||
i += gridSize;
|
||||
}
|
||||
|
||||
// do reduction in shared mem
|
||||
reduceBlock<blockSize>(sdata, mySum, tid, cta);
|
||||
|
||||
// write result for this block to global mem
|
||||
if (tid == 0) g_odata[blockIdx.x] = sdata[0];
|
||||
}
|
||||
|
||||
template <unsigned int blockSize, bool nIsPow2>
|
||||
__global__ void reduceMultiPass(const float *g_idata, float *g_odata,
|
||||
unsigned int n) {
|
||||
// Handle to thread block group
|
||||
cg::thread_block cta = cg::this_thread_block();
|
||||
reduceBlocks<blockSize, nIsPow2>(g_idata, g_odata, n, cta);
|
||||
}
|
||||
|
||||
// Global variable used by reduceSinglePass to count how many blocks have
|
||||
// finished
|
||||
__device__ unsigned int retirementCount = 0;
|
||||
|
||||
cudaError_t setRetirementCount(int retCnt) {
|
||||
return cudaMemcpyToSymbol(retirementCount, &retCnt, sizeof(unsigned int), 0,
|
||||
cudaMemcpyHostToDevice);
|
||||
}
|
||||
|
||||
// This reduction kernel reduces an arbitrary size array in a single kernel
|
||||
// invocation It does so by keeping track of how many blocks have finished.
|
||||
// After each thread block completes the reduction of its own block of data, it
|
||||
// "takes a ticket" by atomically incrementing a global counter. If the ticket
|
||||
// value is equal to the number of thread blocks, then the block holding the
|
||||
// ticket knows that it is the last block to finish. This last block is
|
||||
// responsible for summing the results of all the other blocks.
|
||||
//
|
||||
// In order for this to work, we must be sure that before a block takes a
|
||||
// ticket, all of its memory transactions have completed. This is what
|
||||
// __threadfence() does -- it blocks until the results of all outstanding memory
|
||||
// transactions within the calling thread are visible to all other threads.
|
||||
//
|
||||
// For more details on the reduction algorithm (notably the multi-pass
|
||||
// approach), see the "reduction" sample in the CUDA SDK.
|
||||
template <unsigned int blockSize, bool nIsPow2>
|
||||
__global__ void reduceSinglePass(const float *g_idata, float *g_odata,
|
||||
unsigned int n) {
|
||||
// Handle to thread block group
|
||||
cg::thread_block cta = cg::this_thread_block();
|
||||
//
|
||||
// PHASE 1: Process all inputs assigned to this block
|
||||
//
|
||||
|
||||
reduceBlocks<blockSize, nIsPow2>(g_idata, g_odata, n, cta);
|
||||
|
||||
//
|
||||
// PHASE 2: Last block finished will process all partial sums
|
||||
//
|
||||
|
||||
if (gridDim.x > 1) {
|
||||
const unsigned int tid = threadIdx.x;
|
||||
__shared__ bool amLast;
|
||||
extern float __shared__ smem[];
|
||||
|
||||
// wait until all outstanding memory instructions in this thread are
|
||||
// finished
|
||||
__threadfence();
|
||||
|
||||
// Thread 0 takes a ticket
|
||||
if (tid == 0) {
|
||||
unsigned int ticket = atomicInc(&retirementCount, gridDim.x);
|
||||
// If the ticket ID is equal to the number of blocks, we are the last
|
||||
// block!
|
||||
amLast = (ticket == gridDim.x - 1);
|
||||
}
|
||||
|
||||
cg::sync(cta);
|
||||
|
||||
// The last block sums the results of all other blocks
|
||||
if (amLast) {
|
||||
int i = tid;
|
||||
float mySum = 0;
|
||||
|
||||
while (i < gridDim.x) {
|
||||
mySum += g_odata[i];
|
||||
i += blockSize;
|
||||
}
|
||||
|
||||
reduceBlock<blockSize>(smem, mySum, tid, cta);
|
||||
|
||||
if (tid == 0) {
|
||||
g_odata[0] = smem[0];
|
||||
|
||||
// reset retirement count so that next run succeeds
|
||||
retirementCount = 0;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
bool isPow2(unsigned int x) { return ((x & (x - 1)) == 0); }
|
||||
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
// Wrapper function for kernel launch
|
||||
////////////////////////////////////////////////////////////////////////////////
|
||||
extern "C" void reduce(int size, int threads, int blocks, float *d_idata,
|
||||
float *d_odata) {
|
||||
dim3 dimBlock(threads, 1, 1);
|
||||
dim3 dimGrid(blocks, 1, 1);
|
||||
int smemSize =
|
||||
(threads <= 32) ? 2 * threads * sizeof(float) : threads * sizeof(float);
|
||||
|
||||
// choose which of the optimized versions of reduction to launch
|
||||
if (isPow2(size)) {
|
||||
switch (threads) {
|
||||
case 512:
|
||||
reduceMultiPass<512, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 256:
|
||||
reduceMultiPass<256, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 128:
|
||||
reduceMultiPass<128, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 64:
|
||||
reduceMultiPass<64, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 32:
|
||||
reduceMultiPass<32, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 16:
|
||||
reduceMultiPass<16, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 8:
|
||||
reduceMultiPass<8, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 4:
|
||||
reduceMultiPass<4, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 2:
|
||||
reduceMultiPass<2, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 1:
|
||||
reduceMultiPass<1, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
}
|
||||
} else {
|
||||
switch (threads) {
|
||||
case 512:
|
||||
reduceMultiPass<512, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 256:
|
||||
reduceMultiPass<256, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 128:
|
||||
reduceMultiPass<128, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 64:
|
||||
reduceMultiPass<64, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 32:
|
||||
reduceMultiPass<32, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 16:
|
||||
reduceMultiPass<16, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 8:
|
||||
reduceMultiPass<8, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 4:
|
||||
reduceMultiPass<4, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 2:
|
||||
reduceMultiPass<2, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 1:
|
||||
reduceMultiPass<1, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
extern "C" void reduceSinglePass(int size, int threads, int blocks,
|
||||
float *d_idata, float *d_odata) {
|
||||
dim3 dimBlock(threads, 1, 1);
|
||||
dim3 dimGrid(blocks, 1, 1);
|
||||
int smemSize = threads * sizeof(float);
|
||||
|
||||
// choose which of the optimized versions of reduction to launch
|
||||
if (isPow2(size)) {
|
||||
switch (threads) {
|
||||
case 512:
|
||||
reduceSinglePass<512, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 256:
|
||||
reduceSinglePass<256, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 128:
|
||||
reduceSinglePass<128, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 64:
|
||||
reduceSinglePass<64, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 32:
|
||||
reduceSinglePass<32, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 16:
|
||||
reduceSinglePass<16, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 8:
|
||||
reduceSinglePass<8, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 4:
|
||||
reduceSinglePass<4, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 2:
|
||||
reduceSinglePass<2, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 1:
|
||||
reduceSinglePass<1, true>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
}
|
||||
} else {
|
||||
switch (threads) {
|
||||
case 512:
|
||||
reduceSinglePass<512, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 256:
|
||||
reduceSinglePass<256, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 128:
|
||||
reduceSinglePass<128, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 64:
|
||||
reduceSinglePass<64, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 32:
|
||||
reduceSinglePass<32, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 16:
|
||||
reduceSinglePass<16, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 8:
|
||||
reduceSinglePass<8, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 4:
|
||||
reduceSinglePass<4, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 2:
|
||||
reduceSinglePass<2, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
|
||||
case 1:
|
||||
reduceSinglePass<1, false>
|
||||
<<<dimGrid, dimBlock, smemSize>>>(d_idata, d_odata, size);
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif // #ifndef _REDUCE_KERNEL_H_
|
||||
@@ -0,0 +1,20 @@
|
||||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2017
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "threadFenceReduction", "threadFenceReduction_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>threadFenceReduction_vs2017</RootNamespace>
|
||||
<ProjectName>threadFenceReduction</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)/threadFenceReduction.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="threadFenceReduction.cu" />
|
||||
<ClInclude Include="threadFenceReduction.h" />
|
||||
<None Include="threadFenceReduction_kernel.cuh" />
|
||||
</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}") = "threadFenceReduction", "threadFenceReduction_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>threadFenceReduction_vs2019</RootNamespace>
|
||||
<ProjectName>threadFenceReduction</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>$(OutDir)/threadFenceReduction.exe</OutputFile>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<CodeGeneration>compute_35,sm_35;compute_37,sm_37;compute_50,sm_50;compute_52,sm_52;compute_60,sm_60;compute_61,sm_61;compute_70,sm_70;compute_75,sm_75;compute_80,sm_80;compute_86,sm_86;</CodeGeneration>
|
||||
<AdditionalOptions>-Xcompiler "/wd 4819" --threads 0 </AdditionalOptions>
|
||||
<Include>./;../../../Common</Include>
|
||||
<Defines>WIN32</Defines>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemDefinitionGroup Condition="'$(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="'$(Configuration)'=='Release'">
|
||||
<ClCompile>
|
||||
<Optimization>MaxSpeed</Optimization>
|
||||
<RuntimeLibrary>MultiThreaded</RuntimeLibrary>
|
||||
</ClCompile>
|
||||
<Link>
|
||||
<GenerateDebugInformation>false</GenerateDebugInformation>
|
||||
<LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
|
||||
</Link>
|
||||
<CudaCompile>
|
||||
<Runtime>MT</Runtime>
|
||||
<TargetMachinePlatform>64</TargetMachinePlatform>
|
||||
</CudaCompile>
|
||||
</ItemDefinitionGroup>
|
||||
<ItemGroup>
|
||||
<CudaCompile Include="threadFenceReduction.cu" />
|
||||
<ClInclude Include="threadFenceReduction.h" />
|
||||
<None Include="threadFenceReduction_kernel.cuh" />
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 11.6.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
@@ -0,0 +1,20 @@
|
||||
|
||||
Microsoft Visual Studio Solution File, Format Version 12.00
|
||||
# Visual Studio 2022
|
||||
Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "threadFenceReduction", "threadFenceReduction_vs2022.vcxproj", "{997E0757-EA74-4A4E-A0FC-47D8C8831A15}"
|
||||
EndProject
|
||||
Global
|
||||
GlobalSection(SolutionConfigurationPlatforms) = preSolution
|
||||
Debug|x64 = Debug|x64
|
||||
Release|x64 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(ProjectConfigurationPlatforms) = postSolution
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.ActiveCfg = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Debug|x64.Build.0 = Debug|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.ActiveCfg = Release|x64
|
||||
{997E0757-EA74-4A4E-A0FC-47D8C8831A15}.Release|x64.Build.0 = Release|x64
|
||||
EndGlobalSection
|
||||
GlobalSection(SolutionProperties) = preSolution
|
||||
HideSolutionNode = FALSE
|
||||
EndGlobalSection
|
||||
EndGlobal
|
||||
@@ -0,0 +1,109 @@
|
||||
<?xml version="1.0" encoding="utf-8"?>
|
||||
<Project DefaultTargets="Build" ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
|
||||
<PropertyGroup>
|
||||
<CUDAPropsPath Condition="'$(CUDAPropsPath)'==''">$(VCTargetsPath)\BuildCustomizations</CUDAPropsPath>
|
||||
</PropertyGroup>
|
||||
<ItemGroup Label="ProjectConfigurations">
|
||||
<ProjectConfiguration Include="Debug|x64">
|
||||
<Configuration>Debug</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
<ProjectConfiguration Include="Release|x64">
|
||||
<Configuration>Release</Configuration>
|
||||
<Platform>x64</Platform>
|
||||
</ProjectConfiguration>
|
||||
</ItemGroup>
|
||||
<PropertyGroup Label="Globals">
|
||||
<ProjectGuid>{997E0757-EA74-4A4E-A0FC-47D8C8831A15}</ProjectGuid>
|
||||
<RootNamespace>threadFenceReduction_vs2022</RootNamespace>
|
||||
<ProjectName>threadFenceReduction</ProjectName>
|
||||
<CudaToolkitCustomDir />
|
||||
</PropertyGroup>
|
||||
<Import Project="$(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)/threadFenceReduction.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="threadFenceReduction.cu" />
|
||||
<ClInclude Include="threadFenceReduction.h" />
|
||||
<None Include="threadFenceReduction_kernel.cuh" />
|
||||
</ItemGroup>
|
||||
<Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
|
||||
<ImportGroup Label="ExtensionTargets">
|
||||
<Import Project="$(CUDAPropsPath)\CUDA 11.6.targets" />
|
||||
</ImportGroup>
|
||||
</Project>
|
||||
Reference in New Issue
Block a user