mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2026-10-23 21:28:17 +08:00
add and update samples for CUDA 11.5
This commit is contained in:
Rutwik Choughule
18
Samples/cuHook/.vscode/c_cpp_properties.json
vendored
Normal file
18
Samples/cuHook/.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/cuHook/.vscode/extensions.json
vendored
Normal file
7
Samples/cuHook/.vscode/extensions.json
vendored
Normal file
@@ -0,0 +1,7 @@
|
||||
{
|
||||
"recommendations": [
|
||||
"nvidia.nsight-vscode-edition",
|
||||
"ms-vscode.cpptools",
|
||||
"ms-vscode.makefile-tools"
|
||||
]
|
||||
}
|
||||
10
Samples/cuHook/.vscode/launch.json
vendored
Normal file
10
Samples/cuHook/.vscode/launch.json
vendored
Normal file
@@ -0,0 +1,10 @@
|
||||
{
|
||||
"configurations": [
|
||||
{
|
||||
"name": "CUDA C++: Launch",
|
||||
"type": "cuda-gdb",
|
||||
"request": "launch",
|
||||
"program": "${workspaceFolder}/cuHook"
|
||||
}
|
||||
]
|
||||
}
|
||||
15
Samples/cuHook/.vscode/tasks.json
vendored
Normal file
15
Samples/cuHook/.vscode/tasks.json
vendored
Normal file
@@ -0,0 +1,15 @@
|
||||
{
|
||||
"version": "2.0.0",
|
||||
"tasks": [
|
||||
{
|
||||
"label": "sample",
|
||||
"type": "shell",
|
||||
"command": "make dbg=1",
|
||||
"problemMatcher": ["$nvcc"],
|
||||
"group": {
|
||||
"kind": "build",
|
||||
"isDefault": true
|
||||
}
|
||||
}
|
||||
]
|
||||
}
|
||||
462
Samples/cuHook/Makefile
Normal file
462
Samples/cuHook/Makefile
Normal file
@@ -0,0 +1,462 @@
|
||||
################################################################################
|
||||
# Copyright (c) 2019, 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 += --qpp-config 5.4.0,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))
|
||||
|
||||
UBUNTU = $(shell lsb_release -i -s 2>/dev/null | grep -i ubuntu)
|
||||
|
||||
SAMPLE_ENABLED := 1
|
||||
|
||||
# This sample is not supported on Mac OSX
|
||||
ifeq ($(TARGET_OS),darwin)
|
||||
$(info >>> WARNING - cuHook is not supported on Mac OSX - waiving sample <<<)
|
||||
SAMPLE_ENABLED := 0
|
||||
endif
|
||||
|
||||
# This sample is not supported on ARMv7
|
||||
ifeq ($(TARGET_ARCH),armv7l)
|
||||
$(info >>> WARNING - cuHook is not supported on ARMv7 - waiving sample <<<)
|
||||
SAMPLE_ENABLED := 0
|
||||
endif
|
||||
|
||||
# This sample is not supported on aarch64
|
||||
ifeq ($(TARGET_ARCH),aarch64)
|
||||
$(info >>> WARNING - cuHook is not supported on aarch64 - waiving sample <<<)
|
||||
SAMPLE_ENABLED := 0
|
||||
endif
|
||||
|
||||
# This sample is not supported on sbsa
|
||||
ifeq ($(TARGET_ARCH),sbsa)
|
||||
$(info >>> WARNING - cuHook is not supported on sbsa - 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 :=
|
||||
|
||||
################################################################################
|
||||
|
||||
#Detect if installed version of GCC supports required C++11
|
||||
ifeq ($(TARGET_OS),linux)
|
||||
empty :=
|
||||
space := $(empty) $(empty)
|
||||
GCCVERSIONSTRING := $(shell expr `$(HOST_COMPILER) -dumpversion`)
|
||||
#Create version number without "."
|
||||
GCCVERSION := $(shell expr `echo $(GCCVERSIONSTRING)` | cut -f1 -d.)
|
||||
GCCVERSION += $(shell expr `echo $(GCCVERSIONSTRING)` | cut -f2 -d.)
|
||||
GCCVERSION += $(shell expr `echo $(GCCVERSIONSTRING)` | cut -f3 -d.)
|
||||
# Make sure the version number has at least 3 decimals
|
||||
GCCVERSION += 00
|
||||
# Remove spaces from the version number
|
||||
GCCVERSION := $(subst $(space),$(empty),$(GCCVERSION))
|
||||
#$(warning $(GCCVERSION))
|
||||
|
||||
IS_MIN_VERSION := $(shell expr `echo $(GCCVERSION)` \>= 51000)
|
||||
|
||||
ifeq ($(IS_MIN_VERSION), 1)
|
||||
$(info >>> GCC Version is greater or equal to 5.1.0 <<<)
|
||||
else
|
||||
$(info >>> Waiving build. Minimum GCC version required is 5.1.0<<<)
|
||||
SAMPLE_ENABLED := 0
|
||||
endif
|
||||
endif
|
||||
|
||||
# Gencode arguments
|
||||
ifeq ($(TARGET_ARCH),$(filter $(TARGET_ARCH),armv7l aarch64))
|
||||
SMS ?= 35 37 50 52 60 61 70 72 75 80 86
|
||||
else
|
||||
SMS ?= 35 37 50 52 60 61 70 75 80 86
|
||||
endif
|
||||
|
||||
ifeq ($(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
|
||||
|
||||
ifeq ($(TARGET_OS),darwin)
|
||||
ALL_LDFLAGS += -Xcompiler -F/Library/Frameworks -Xlinker -framework -Xlinker CUDA
|
||||
else
|
||||
ifeq ($(TARGET_ARCH),x86_64)
|
||||
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/lib64/stubs
|
||||
CUDA_SEARCH_PATH += $(CUDA_PATH)/targets/x86_64-linux/lib/stubs
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-linux)
|
||||
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/armv7-linux-gnueabihf/lib/stubs
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-linux)
|
||||
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-linux/lib/stubs
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),sbsa-linux)
|
||||
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/sbsa-linux/lib/stubs
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-android)
|
||||
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/armv7-linux-androideabi/lib/stubs
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-android)
|
||||
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-linux-androideabi/lib/stubs
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),armv7l-qnx)
|
||||
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/ARMv7-linux-QNX/lib/stubs
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_ARCH)-$(TARGET_OS),aarch64-qnx)
|
||||
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/aarch64-qnx/lib/stubs
|
||||
ifdef TARGET_OVERRIDE
|
||||
CUDA_SEARCH_PATH := $(CUDA_PATH)/targets/$(TARGET_OVERRIDE)/lib/stubs
|
||||
endif
|
||||
endif
|
||||
|
||||
ifeq ($(TARGET_ARCH),ppc64le)
|
||||
CUDA_SEARCH_PATH ?= $(CUDA_PATH)/targets/ppc64le-linux/lib/stubs
|
||||
endif
|
||||
|
||||
ifeq ($(HOST_ARCH),ppc64le)
|
||||
CUDA_SEARCH_PATH += $(CUDA_PATH)/lib64/stubs
|
||||
endif
|
||||
|
||||
CUDALIB ?= $(shell find -L $(CUDA_SEARCH_PATH) -maxdepth 1 -name libcuda.so 2> /dev/null)
|
||||
ifeq ("$(CUDALIB)","")
|
||||
$(info >>> WARNING - libcuda.so not found, CUDA Driver is not installed. Please re-install the driver. <<<)
|
||||
SAMPLE_ENABLED := 0
|
||||
else
|
||||
CUDALIB := $(shell echo $(CUDALIB) | sed "s/ .*//" | sed "s/\/libcuda.so//" )
|
||||
LIBRARIES += -L$(CUDALIB) -lcuda
|
||||
endif
|
||||
endif
|
||||
|
||||
ALL_CCFLAGS += --threads 0 --std=c++11
|
||||
|
||||
LIBSIZE :=
|
||||
ifneq ($(TARGET_OS),darwin)
|
||||
ifeq ($(TARGET_SIZE),64)
|
||||
LIBSIZE := 64
|
||||
endif
|
||||
endif
|
||||
|
||||
LIBRARIES += -L$(CUDA_PATH)/lib$(LIBSIZE) -ldl
|
||||
|
||||
ifeq ($(SAMPLE_ENABLED),0)
|
||||
EXEC ?= @echo "[@]"
|
||||
endif
|
||||
|
||||
################################################################################
|
||||
|
||||
# Target rules
|
||||
all: build
|
||||
|
||||
build: cuHook libcuhook.so.1
|
||||
|
||||
check.deps:
|
||||
ifeq ($(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
|
||||
|
||||
cuHook.o:cuHook.cpp
|
||||
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) $(GENCODE_FLAGS) -o $@ -c $<
|
||||
|
||||
libcuhook.o:libcuhook.cpp
|
||||
$(EXEC) $(NVCC) $(INCLUDES) $(ALL_CCFLAGS) --compiler-options '-fPIC' $(GENCODE_FLAGS) -o $@ -c $<
|
||||
|
||||
libcuhook.so.1:libcuhook.o
|
||||
$(EXEC) $(NVCC) -shared $(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)
|
||||
|
||||
cuHook: cuHook.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
|
||||
$(EXEC) LD_PRELOAD=./libcuhook.so.1 ./cuHook
|
||||
|
||||
clean:
|
||||
rm -f cuHook cuHook.o libcuhook.o libcuhook.so.1
|
||||
rm -rf ../../bin/$(TARGET_ARCH)/$(TARGET_OS)/$(BUILD_TYPE)/cuHook
|
||||
|
||||
clobber: clean
|
||||
55
Samples/cuHook/README.md
Normal file
55
Samples/cuHook/README.md
Normal file
@@ -0,0 +1,55 @@
|
||||
# cuHook - CUDA Interception Library
|
||||
|
||||
## Description
|
||||
|
||||
This sample demonstrates how to build and use an intercept library with CUDA. The library has to be loaded via LD_PRELOAD, e.g. LD_PRELOAD=<full_path>/libcuhook.so.1 ./cuHook
|
||||
|
||||
## Key Concepts
|
||||
|
||||
## 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 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)
|
||||
|
||||
## Supported OSes
|
||||
|
||||
Linux
|
||||
|
||||
## Supported CPU Architecture
|
||||
|
||||
x86_64, ppc64le
|
||||
|
||||
## CUDA APIs involved
|
||||
|
||||
## Prerequisites
|
||||
|
||||
Download and install the [CUDA Toolkit 11.5](https://developer.nvidia.com/cuda-downloads) for your corresponding platform.
|
||||
|
||||
## Build and Run
|
||||
|
||||
### 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.
|
||||
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/>
|
||||
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)
|
||||
|
||||
123
Samples/cuHook/cuHook.cpp
Normal file
123
Samples/cuHook/cuHook.cpp
Normal file
@@ -0,0 +1,123 @@
|
||||
/* Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
|
||||
*
|
||||
* Redistribution and use in source and binary forms, with or without
|
||||
* modification, are permitted provided that the following conditions
|
||||
* are met:
|
||||
* * Redistributions of source code must retain the above copyright
|
||||
* notice, this list of conditions and the following disclaimer.
|
||||
* * Redistributions in binary form must reproduce the above copyright
|
||||
* notice, this list of conditions and the following disclaimer in the
|
||||
* documentation and/or other materials provided with the distribution.
|
||||
* * Neither the name of NVIDIA CORPORATION nor the names of its
|
||||
* contributors may be used to endorse or promote products derived
|
||||
* from this software without specific prior written permission.
|
||||
*
|
||||
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
|
||||
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
||||
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
||||
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
|
||||
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
||||
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
||||
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
||||
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
|
||||
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
||||
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
||||
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
*/
|
||||
|
||||
// This sample demonstrates how to use the CUDA hook library to receive
|
||||
// callbacks
|
||||
|
||||
#include <cuda.h>
|
||||
#include <stdio.h>
|
||||
#include <dlfcn.h>
|
||||
|
||||
#include "libcuhook.h"
|
||||
|
||||
#define ASSERT_COND(x, msg) \
|
||||
do { \
|
||||
if (!(x)) { \
|
||||
fprintf(stderr, "Error: Condition (%s) failed at %s:%d\n", #x, __FILE__, \
|
||||
__LINE__); \
|
||||
fprintf(stderr, "cuHook sample failed (%s)\n", msg); \
|
||||
exit(1); \
|
||||
} \
|
||||
} while (0)
|
||||
|
||||
/*
|
||||
** Example of how to use the CUDA Interception Library, libcuhook.so
|
||||
** The library has to be loaded via LD_PRELOAD, e.g.
|
||||
*LD_PRELOAD=<full_path>/libcuhook.so.1 ./cuHook
|
||||
*/
|
||||
|
||||
static int allocation_cb = 0;
|
||||
static int free_cb = 0;
|
||||
static int destroy_ctx_cb = 0;
|
||||
|
||||
CUresult device_allocation_callback(CUdeviceptr *dptr, size_t bytesize) {
|
||||
fprintf(stdout, "Received memory allocation callback!\n");
|
||||
allocation_cb++;
|
||||
return CUDA_SUCCESS;
|
||||
}
|
||||
|
||||
CUresult device_free_callback(CUdeviceptr dptr) {
|
||||
fprintf(stdout, "Received memory de-allocation callback!\n");
|
||||
free_cb++;
|
||||
return CUDA_SUCCESS;
|
||||
}
|
||||
|
||||
CUresult destroy_context_callback(CUcontext ctx) {
|
||||
fprintf(stdout, "Received context destroy event!\n");
|
||||
destroy_ctx_cb++;
|
||||
return CUDA_SUCCESS;
|
||||
}
|
||||
|
||||
int main() {
|
||||
int count;
|
||||
CUcontext ctx;
|
||||
|
||||
count = 0;
|
||||
|
||||
cuInit(0);
|
||||
cuDeviceGetCount(&count);
|
||||
ASSERT_COND(count > 0, "No suitable devices found");
|
||||
|
||||
// Load the cudaHookRegisterCallback symbol using the default library search
|
||||
// order. If we found the symbol, then the hooking library has been loaded
|
||||
fnCuHookRegisterCallback cuHook =
|
||||
(fnCuHookRegisterCallback)dlsym(RTLD_DEFAULT, "cuHookRegisterCallback");
|
||||
// ASSERT_COND(cuHook, dlerror());
|
||||
if (cuHook) {
|
||||
// CUDA Runtime symbols cannot be hooked but the underlying driver ones
|
||||
// _can_. Example:
|
||||
// - cudaFree() will trigger cuMemFree
|
||||
// - cudaDeviceReset() will trigger a context change and you would need to
|
||||
// intercept cuCtxGetCurrent/cuCtxSetCurrent
|
||||
cuHook(CU_HOOK_MEM_ALLOC, POST_CALL_HOOK,
|
||||
(void *)device_allocation_callback);
|
||||
cuHook(CU_HOOK_MEM_FREE, PRE_CALL_HOOK, (void *)device_free_callback);
|
||||
cuHook(CU_HOOK_CTX_DESTROY, POST_CALL_HOOK,
|
||||
(void *)destroy_context_callback);
|
||||
}
|
||||
|
||||
cuCtxCreate(&ctx, 0, 0);
|
||||
{
|
||||
CUresult status;
|
||||
CUdeviceptr dptr;
|
||||
|
||||
status = cuMemAlloc(&dptr, 1024);
|
||||
ASSERT_COND(status == CUDA_SUCCESS, "cuMemAlloc call failed");
|
||||
|
||||
status = cuMemFree(dptr);
|
||||
ASSERT_COND(status == CUDA_SUCCESS, "cuMemFree call failed");
|
||||
}
|
||||
cuCtxDestroy(ctx);
|
||||
|
||||
ASSERT_COND(allocation_cb == 1, "Didn't receive the allocation callback");
|
||||
ASSERT_COND(free_cb == 1, "Didn't receive the free callback");
|
||||
ASSERT_COND(destroy_ctx_cb == 1,
|
||||
"Didn't receive the destroy context callback");
|
||||
|
||||
fprintf(stdout, "Sample finished successfully.\n");
|
||||
return (0);
|
||||
}
|
||||
188
Samples/cuHook/libcuhook.cpp
Normal file
188
Samples/cuHook/libcuhook.cpp
Normal file
@@ -0,0 +1,188 @@
|
||||
/* Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved.
|
||||
*
|
||||
* Redistribution and use in source and binary forms, with or without
|
||||
* modification, are permitted provided that the following conditions
|
||||
* are met:
|
||||
* * Redistributions of source code must retain the above copyright
|
||||
* notice, this list of conditions and the following disclaimer.
|
||||
* * Redistributions in binary form must reproduce the above copyright
|
||||
* notice, this list of conditions and the following disclaimer in the
|
||||
* documentation and/or other materials provided with the distribution.
|
||||
* * Neither the name of NVIDIA CORPORATION nor the names of its
|
||||
* contributors may be used to endorse or promote products derived
|
||||
* from this software without specific prior written permission.
|
||||
*
|
||||
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
|
||||
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
||||
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
|
||||
* PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
|
||||
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
|
||||
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
|
||||
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
|
||||
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
|
||||
* OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
||||
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
||||
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
||||
*/
|
||||
|
||||
// This sample demonstrates a simple library to interpose CUDA symbols
|
||||
|
||||
#define __USE_GNU
|
||||
#include <dlfcn.h>
|
||||
#include <stdio.h>
|
||||
#include <string.h>
|
||||
#include <unistd.h>
|
||||
#include <stdlib.h>
|
||||
|
||||
#include <cuda.h>
|
||||
#include "libcuhook.h"
|
||||
|
||||
// For interposing dlsym(). See elf/dl-libc.c for the internal dlsym interface
|
||||
// function For interposing dlopen(). Sell elf/dl-lib.c for the internal
|
||||
// dlopen_mode interface function
|
||||
extern "C" {
|
||||
void* __libc_dlsym(void* map, const char* name);
|
||||
}
|
||||
extern "C" {
|
||||
void* __libc_dlopen_mode(const char* name, int mode);
|
||||
}
|
||||
|
||||
// We need to give the pre-processor a chance to replace a function, such as:
|
||||
// cuMemAlloc => cuMemAlloc_v2
|
||||
#define STRINGIFY(x) #x
|
||||
#define CUDA_SYMBOL_STRING(x) STRINGIFY(x)
|
||||
|
||||
// We need to interpose dlsym since anyone using dlopen+dlsym to get the CUDA
|
||||
// driver symbols will bypass the hooking mechanism (this includes the CUDA
|
||||
// runtime). Its tricky though, since if we replace the real dlsym with ours, we
|
||||
// can't dlsym() the real dlsym. To get around that, call the 'private' libc
|
||||
// interface called __libc_dlsym to get the real dlsym.
|
||||
typedef void* (*fnDlsym)(void*, const char*);
|
||||
|
||||
static void* real_dlsym(void* handle, const char* symbol) {
|
||||
static fnDlsym internal_dlsym = (fnDlsym)__libc_dlsym(
|
||||
__libc_dlopen_mode("libdl.so.2", RTLD_LAZY), "dlsym");
|
||||
return (*internal_dlsym)(handle, symbol);
|
||||
}
|
||||
|
||||
// Main structure that gets initialized at library load time
|
||||
// Choose a unique name, or it can clash with other preloaded libraries.
|
||||
struct cuHookInfo {
|
||||
void* handle;
|
||||
void* preHooks[CU_HOOK_SYMBOLS];
|
||||
void* postHooks[CU_HOOK_SYMBOLS];
|
||||
|
||||
// Debugging/Stats Info
|
||||
int bDebugEnabled;
|
||||
int hookedFunctionCalls[CU_HOOK_SYMBOLS];
|
||||
|
||||
cuHookInfo() {
|
||||
const char* envHookDebug;
|
||||
|
||||
// Check environment for CU_HOOK_DEBUG to facilitate debugging
|
||||
envHookDebug = getenv("CU_HOOK_DEBUG");
|
||||
if (envHookDebug && envHookDebug[0] == '1') {
|
||||
bDebugEnabled = 1;
|
||||
fprintf(stderr, "* %6d >> CUDA HOOK Library loaded.\n", getpid());
|
||||
}
|
||||
}
|
||||
|
||||
~cuHookInfo() {
|
||||
if (bDebugEnabled) {
|
||||
pid_t pid = getpid();
|
||||
// You can gather statistics, timings, etc.
|
||||
fprintf(stderr, "* %6d >> CUDA HOOK Library Unloaded - Statistics:\n",
|
||||
pid);
|
||||
fprintf(stderr, "* %6d >> %20s ... %d\n", pid,
|
||||
CUDA_SYMBOL_STRING(cuMemAlloc),
|
||||
hookedFunctionCalls[CU_HOOK_MEM_ALLOC]);
|
||||
fprintf(stderr, "* %6d >> %20s ... %d\n", pid,
|
||||
CUDA_SYMBOL_STRING(cuMemFree),
|
||||
hookedFunctionCalls[CU_HOOK_MEM_FREE]);
|
||||
fprintf(stderr, "* %6d >> %20s ... %d\n", pid,
|
||||
CUDA_SYMBOL_STRING(cuCtxGetCurrent),
|
||||
hookedFunctionCalls[CU_HOOK_CTX_GET_CURRENT]);
|
||||
fprintf(stderr, "* %6d >> %20s ... %d\n", pid,
|
||||
CUDA_SYMBOL_STRING(cuCtxSetCurrent),
|
||||
hookedFunctionCalls[CU_HOOK_CTX_SET_CURRENT]);
|
||||
fprintf(stderr, "* %6d >> %20s ... %d\n", pid,
|
||||
CUDA_SYMBOL_STRING(cuCtxDestroy),
|
||||
hookedFunctionCalls[CU_HOOK_CTX_DESTROY]);
|
||||
}
|
||||
if (handle) {
|
||||
dlclose(handle);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
static struct cuHookInfo cuhl;
|
||||
|
||||
// Exposed API
|
||||
void cuHookRegisterCallback(HookSymbols symbol, HookTypes type,
|
||||
void* callback) {
|
||||
if (type == PRE_CALL_HOOK) {
|
||||
cuhl.preHooks[symbol] = callback;
|
||||
} else if (type == POST_CALL_HOOK) {
|
||||
cuhl.postHooks[symbol] = callback;
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
** Interposed Functions
|
||||
*/
|
||||
void* dlsym(void* handle, const char* symbol) {
|
||||
// Early out if not a CUDA driver symbol
|
||||
if (strncmp(symbol, "cu", 2) != 0) {
|
||||
return (real_dlsym(handle, symbol));
|
||||
}
|
||||
|
||||
if (strcmp(symbol, CUDA_SYMBOL_STRING(cuMemAlloc)) == 0) {
|
||||
return (void*)(&cuMemAlloc);
|
||||
} else if (strcmp(symbol, CUDA_SYMBOL_STRING(cuMemFree)) == 0) {
|
||||
return (void*)(&cuMemFree);
|
||||
} else if (strcmp(symbol, CUDA_SYMBOL_STRING(cuCtxGetCurrent)) == 0) {
|
||||
return (void*)(&cuCtxGetCurrent);
|
||||
} else if (strcmp(symbol, CUDA_SYMBOL_STRING(cuCtxSetCurrent)) == 0) {
|
||||
return (void*)(&cuCtxSetCurrent);
|
||||
} else if (strcmp(symbol, CUDA_SYMBOL_STRING(cuCtxDestroy)) == 0) {
|
||||
return (void*)(&cuCtxDestroy);
|
||||
}
|
||||
return (real_dlsym(handle, symbol));
|
||||
}
|
||||
|
||||
/*
|
||||
** If the user of this library does not wish to include CUDA specific
|
||||
*code/headers in the code,
|
||||
** then all the parameters can be changed and/or simply casted before calling
|
||||
*the callback.
|
||||
*/
|
||||
#define CU_HOOK_GENERATE_INTERCEPT(hooksymbol, funcname, params, ...) \
|
||||
CUresult CUDAAPI funcname params { \
|
||||
static void* real_func = \
|
||||
(void*)real_dlsym(RTLD_NEXT, CUDA_SYMBOL_STRING(funcname)); \
|
||||
CUresult result = CUDA_SUCCESS; \
|
||||
\
|
||||
if (cuhl.bDebugEnabled) { \
|
||||
cuhl.hookedFunctionCalls[hooksymbol]++; \
|
||||
} \
|
||||
if (cuhl.preHooks[hooksymbol]) { \
|
||||
((CUresult CUDAAPI(*) params)cuhl.preHooks[hooksymbol])(__VA_ARGS__); \
|
||||
} \
|
||||
result = ((CUresult CUDAAPI(*) params)real_func)(__VA_ARGS__); \
|
||||
if (cuhl.postHooks[hooksymbol] && result == CUDA_SUCCESS) { \
|
||||
((CUresult CUDAAPI(*) params)cuhl.postHooks[hooksymbol])(__VA_ARGS__); \
|
||||
} \
|
||||
return (result); \
|
||||
}
|
||||
|
||||
CU_HOOK_GENERATE_INTERCEPT(CU_HOOK_MEM_ALLOC, cuMemAlloc,
|
||||
(CUdeviceptr * dptr, size_t bytesize), dptr,
|
||||
bytesize)
|
||||
CU_HOOK_GENERATE_INTERCEPT(CU_HOOK_MEM_FREE, cuMemFree, (CUdeviceptr dptr),
|
||||
dptr)
|
||||
CU_HOOK_GENERATE_INTERCEPT(CU_HOOK_CTX_GET_CURRENT, cuCtxGetCurrent,
|
||||
(CUcontext * pctx), pctx)
|
||||
CU_HOOK_GENERATE_INTERCEPT(CU_HOOK_CTX_SET_CURRENT, cuCtxSetCurrent,
|
||||
(CUcontext ctx), ctx)
|
||||
CU_HOOK_GENERATE_INTERCEPT(CU_HOOK_CTX_DESTROY, cuCtxDestroy, (CUcontext ctx),
|
||||
ctx)
|
||||
63
Samples/cuHook/libcuhook.h
Normal file
63
Samples/cuHook/libcuhook.h
Normal file
@@ -0,0 +1,63 @@
|
||||
/* Copyright (c) 2021, 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 _CUHOOK_H_
|
||||
#define _CUHOOK_H_
|
||||
|
||||
typedef enum HookTypesEnum {
|
||||
PRE_CALL_HOOK,
|
||||
POST_CALL_HOOK,
|
||||
CU_HOOK_TYPES,
|
||||
} HookTypes;
|
||||
|
||||
typedef enum HookSymbolsEnum {
|
||||
CU_HOOK_MEM_ALLOC,
|
||||
CU_HOOK_MEM_FREE,
|
||||
CU_HOOK_CTX_GET_CURRENT,
|
||||
CU_HOOK_CTX_SET_CURRENT,
|
||||
CU_HOOK_CTX_DESTROY,
|
||||
CU_HOOK_SYMBOLS,
|
||||
} HookSymbols;
|
||||
|
||||
// One and only function to call to register a callback
|
||||
// You need to dlsym this symbol in your application and call it to register
|
||||
// callbacks
|
||||
typedef void (*fnCuHookRegisterCallback)(HookSymbols symbol, HookTypes type,
|
||||
void* callback);
|
||||
extern "C" {
|
||||
void cuHookRegisterCallback(HookSymbols symbol, HookTypes type, void* callback);
|
||||
}
|
||||
|
||||
// In case you want to intercept, the callbacks need the same type/parameters as
|
||||
// the real functions
|
||||
typedef CUresult CUDAAPI (*fnMemAlloc)(CUdeviceptr* dptr, size_t bytesize);
|
||||
typedef CUresult CUDAAPI (*fnMemFree)(CUdeviceptr dptr);
|
||||
typedef CUresult CUDAAPI (*fnCtxGetCurrent)(CUcontext* pctx);
|
||||
typedef CUresult CUDAAPI (*fnCtxSetCurrent)(CUcontext ctx);
|
||||
typedef CUresult CUDAAPI (*fnCtxDestroy)(CUcontext ctx);
|
||||
|
||||
#endif /* _CUHOOK_H_ */
|
||||
Reference in New Issue
Block a user