Skip to content

Commit e7e8840

Browse files
ihcinihsdkihcinihsdk
committed
Give a first attack on parallel RCM algorithm implementation.
1 parent 5743cb9 commit e7e8840

12 files changed

+2224
-0
lines changed

‎CMakeLists.txt

+37
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,37 @@
1+
CMAKE_MINIMUM_REQUIRED(VERSION 2.8)
2+
PROJECT(rcm)
3+
INCLUDE(cmake/SBELUtils.cmake)
4+
5+
SET(EXECUTABLE_OUTPUT_PATH ${PROJECT_BINARY_DIR}/bin)
6+
7+
enable_cuda_support()
8+
MESSAGE(STATUS "Cuda version: ${CUDA_VERSION}")
9+
10+
SET(RCM_HEADERS
11+
rcm/common.h
12+
rcm/rcm_um.h
13+
rcm/exception.h
14+
rcm/timer.h
15+
)
16+
17+
SET(RCM_CUHEADERS
18+
rcm/device/kernels.cuh
19+
)
20+
21+
SET(MMIO_FILES
22+
mm_io/mm_io.h
23+
mm_io/mm_io.c
24+
)
25+
26+
SOURCE_GROUP("Headers" FILES ${MC64_HEADERS})
27+
SOURCE_GROUP("CUDA Headers" FILES ${MC64_CUHEADERS})
28+
SOURCE_GROUP("MM_IO" FILES ${MMIO_FILES})
29+
30+
INCLUDE_DIRECTORIES(
31+
${CMAKE_SOURCE_DIR}
32+
)
33+
34+
IF(NOT (${CUDA_VERSION} VERSION_LESS "6.0"))
35+
cuda_add_executable(driver_um driver_um.cu ${MC64_HEADERS} ${MC64_CUHEADERS} ${MMIO_FILES})
36+
cuda_add_executable(testing testing.cu ${MC64_HEADERS} ${MC64_CUHEADERS} ${MMIO_FILES})
37+
ENDIF()

‎SBELUtils.cmake

+166
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,166 @@
1+
####################################################
2+
## Only modify if you know what you're doing. ##
3+
####################################################
4+
5+
6+
# Helps Eclipse/CDT find our include directories
7+
set(CMAKE_VERBOSE_MAKEFILE on)
8+
9+
# Detect the bitness of our machine (eg 32- or 64-bit)
10+
# C-equiv: sizeof(void*)
11+
# Alt: 8*sizeof(void*)
12+
math(EXPR CMAKE_ARCH_BITNESS 8*${CMAKE_SIZEOF_VOID_P})
13+
14+
# For non-multi-configuration generators (eg, make, Eclipse)
15+
# The Visual Studio generator creates a single project with all these
16+
set(CMAKE_BUILD_TYPE "Release" CACHE STRING "For single-configuration generators (e.g. make) set the type of build: Release, Debug, RelWithDebugInfo, MinSizeRel")
17+
SET_PROPERTY(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Release" "Debug" "RelWithDebugInfo" "MinSizeRel")
18+
19+
20+
####################################################
21+
## ---------------------------------------------- ##
22+
## - - ##
23+
## - Enable MPI Support - ##
24+
## - - ##
25+
## ---------------------------------------------- ##
26+
####################################################
27+
28+
# Begin configuring MPI options
29+
macro(enable_mpi_support)
30+
31+
find_package("MPI" REQUIRED)
32+
33+
# Add the MPI-specific compiler and linker flags
34+
# Also, search for #includes in MPI's paths
35+
36+
list(APPEND CMAKE_C_COMPILE_FLAGS ${MPI_C_COMPILE_FLAGS})
37+
list(APPEND CMAKE_C_LINK_FLAGS ${MPI_C_LINK_FLAGS})
38+
include_directories(${MPI_C_INCLUDE_PATH})
39+
40+
list(APPEND CMAKE_CXX_COMPILE_FLAGS ${MPI_CXX_COMPILE_FLAGS})
41+
list(APPEND CMAKE_CXX_LINK_FLAGS ${MPI_CXX_LINK_FLAGS})
42+
include_directories(${MPI_CXX_INCLUDE_PATH})
43+
44+
endmacro(enable_mpi_support)
45+
# Done configuring MPI Options
46+
47+
48+
####################################################
49+
## ---------------------------------------------- ##
50+
## - - ##
51+
## - Enable OpenMP Support - ##
52+
## - - ##
53+
## ---------------------------------------------- ##
54+
####################################################
55+
56+
# Begin configuring OpenMP options
57+
macro(enable_openmp_support)
58+
59+
find_package("OpenMP" REQUIRED)
60+
61+
# Add the OpenMP-specific compiler and linker flags
62+
list(APPEND CMAKE_CXX_FLAGS ${OpenMP_CXX_FLAGS})
63+
list(APPEND CMAKE_C_FLAGS ${OpenMP_C_FLAGS})
64+
65+
endmacro(enable_openmp_support)
66+
# Done configuring OpenMP Options
67+
68+
69+
####################################################
70+
## ---------------------------------------------- ##
71+
## - - ##
72+
## - Enable CUDA Support - ##
73+
## - - ##
74+
## ---------------------------------------------- ##
75+
####################################################
76+
77+
# Begin configuring CUDA options
78+
# This is ugly...
79+
macro(enable_cuda_support)
80+
81+
# Hide a number of options from the default CMake screen
82+
mark_as_advanced(CLEAR CUDA_BUILD_CUBIN)
83+
mark_as_advanced(CLEAR CUDA_SDK_ROOT_DIR)
84+
mark_as_advanced(CLEAR CUDA_TOOLKIT_ROOT_DIR)
85+
mark_as_advanced(CLEAR CUDA_VERBOSE_BUILD)
86+
mark_as_advanced(CLEAR CUDA_FAST_MATH)
87+
mark_as_advanced(CLEAR CUDA_USE_CUSTOM_COMPILER)
88+
mark_as_advanced(CLEAR CUDA_VERBOSE_PTX)
89+
mark_as_advanced(CLEAR CUDA_DEVICE_VERSION)
90+
91+
# select Compute Capability
92+
# This needs to be manually updated when devices with new CCs come out
93+
set(CUDA_DEVICE_VERSION "20" CACHE STRING "CUDA Device Version")
94+
set_property(CACHE CUDA_DEVICE_VERSION PROPERTY STRINGS "10" "11" "12" "13" "20" "21" "30" "35")
95+
96+
# Enable fast-math for CUDA (_not_ GCC)
97+
set(CUDA_FAST_MATH TRUE CACHE BOOL "Use Fast Math Operations")
98+
99+
# Tell nvcc to use a separate compiler for non-CUDA code.
100+
# This is useful if you need to use an older of GCC than comes by default
101+
set(CUDA_USE_CUSTOM_COMPILER FALSE CACHE BOOL "Use Custom Compiler")
102+
set(CUDA_CUSTOM_COMPILER "" CACHE STRING "Custom C++ Compiler for CUDA If Needed")
103+
104+
# Shows register usage, etc
105+
set(CUDA_VERBOSE_PTX TRUE CACHE BOOL "Show Verbose Kernel Info During Compilation")
106+
107+
108+
# Let's get going...
109+
find_package("CUDA" REQUIRED)
110+
111+
# Frequently used in the examples
112+
cuda_include_directories(${CUDA_SDK_ROOT_DIR}/common/inc)
113+
cuda_include_directories(${CUDA_SDK_ROOT_DIR}/../shared/inc)
114+
115+
set(CUDA_SDK_LIB_DIR ${CUDA_SDK_ROOT_DIR}/common/lib
116+
${CUDA_SDK_ROOT_DIR}/lib ${CUDA_SDK_ROOT_DIR}/../shared/lib)
117+
118+
# these are no longer needed
119+
# # Find path to shrutil libs, from CUDA SDK
120+
# find_library(LIBSHRUTIL
121+
# NAMES shrUtils${CMAKE_ARCH_BITNESS} shrutil_${CMAKE_SYSTEM_PROCESSOR}
122+
# PATHS ${CUDA_SDK_LIB_DIR})
123+
# find_library(LIBSHRUTIL_DBG
124+
# NAMES shrUtils${CMAKE_ARCH_BITNESS}D shrutil_${CMAKE_SYSTEM_PROCESSOR}D
125+
# PATHS ${CUDA_SDK_LIB_DIR})
126+
#
127+
# # Find path to cutil libs, from CUDA SDK
128+
# find_library(LIBCUTIL
129+
# NAMES cutil${CMAKE_ARCH_BITNESS} cutil_${CMAKE_SYSTEM_PROCESSOR}
130+
# PATHS ${CUDA_SDK_LIB_DIR})
131+
# find_library(LIBCUTIL_DBG
132+
# NAMES cutil${arch}D cutil_${CMAKE_SYSTEM_PROCESSOR}D
133+
# PATHS ${CUDA_SDK_LIB_DIR})
134+
135+
# Set custom compiler flags
136+
set(CUDA_NVCC_FLAGS "" CACHE STRING "" FORCE)
137+
138+
if(CUDA_USE_CUSTOM_COMPILER)
139+
mark_as_advanced(CLEAR CUDA_CUSTOM_COMPILER)
140+
list(APPEND CUDA_NVCC_FLAGS "-ccbin=${CUDA_CUSTOM_COMPILER}")
141+
else()
142+
mark_as_advanced(FORCE CUDA_CUSTOM_COMPILER)
143+
endif()
144+
145+
# Macro for setting the Compute Capability
146+
macro(set_compute_capability cc)
147+
list(APPEND CUDA_NVCC_FLAGS "-gencode=arch=compute_${cc},code=sm_${cc}")
148+
list(APPEND CUDA_NVCC_FLAGS "-gencode=arch=compute_${cc},code=compute_${cc}")
149+
endmacro(set_compute_capability)
150+
151+
# Tell nvcc to compile for the selected Compute Capability
152+
# This can also be called from the main CMakeLists.txt to enable
153+
# support for additional CCs
154+
set_compute_capability(${CUDA_DEVICE_VERSION})
155+
156+
# Enable fast-math if selected
157+
if(CUDA_FAST_MATH)
158+
list(APPEND CUDA_NVCC_FLAGS "-use_fast_math")
159+
endif()
160+
161+
# Enable verbose compile if selected
162+
if(CUDA_VERBOSE_PTX)
163+
list(APPEND CUDA_NVCC_FLAGS "--ptxas-options=-v")
164+
endif()
165+
endmacro(enable_cuda_support)
166+
# Done configuring CUDA options

‎cmake/SBELUtils.cmake

+166
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,166 @@
1+
####################################################
2+
## Only modify if you know what you're doing. ##
3+
####################################################
4+
5+
6+
# Helps Eclipse/CDT find our include directories
7+
set(CMAKE_VERBOSE_MAKEFILE on)
8+
9+
# Detect the bitness of our machine (eg 32- or 64-bit)
10+
# C-equiv: sizeof(void*)
11+
# Alt: 8*sizeof(void*)
12+
math(EXPR CMAKE_ARCH_BITNESS 8*${CMAKE_SIZEOF_VOID_P})
13+
14+
# For non-multi-configuration generators (eg, make, Eclipse)
15+
# The Visual Studio generator creates a single project with all these
16+
set(CMAKE_BUILD_TYPE "Release" CACHE STRING "For single-configuration generators (e.g. make) set the type of build: Release, Debug, RelWithDebugInfo, MinSizeRel")
17+
SET_PROPERTY(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Release" "Debug" "RelWithDebugInfo" "MinSizeRel")
18+
19+
20+
####################################################
21+
## ---------------------------------------------- ##
22+
## - - ##
23+
## - Enable MPI Support - ##
24+
## - - ##
25+
## ---------------------------------------------- ##
26+
####################################################
27+
28+
# Begin configuring MPI options
29+
macro(enable_mpi_support)
30+
31+
find_package("MPI" REQUIRED)
32+
33+
# Add the MPI-specific compiler and linker flags
34+
# Also, search for #includes in MPI's paths
35+
36+
list(APPEND CMAKE_C_COMPILE_FLAGS ${MPI_C_COMPILE_FLAGS})
37+
list(APPEND CMAKE_C_LINK_FLAGS ${MPI_C_LINK_FLAGS})
38+
include_directories(${MPI_C_INCLUDE_PATH})
39+
40+
list(APPEND CMAKE_CXX_COMPILE_FLAGS ${MPI_CXX_COMPILE_FLAGS})
41+
list(APPEND CMAKE_CXX_LINK_FLAGS ${MPI_CXX_LINK_FLAGS})
42+
include_directories(${MPI_CXX_INCLUDE_PATH})
43+
44+
endmacro(enable_mpi_support)
45+
# Done configuring MPI Options
46+
47+
48+
####################################################
49+
## ---------------------------------------------- ##
50+
## - - ##
51+
## - Enable OpenMP Support - ##
52+
## - - ##
53+
## ---------------------------------------------- ##
54+
####################################################
55+
56+
# Begin configuring OpenMP options
57+
macro(enable_openmp_support)
58+
59+
find_package("OpenMP" REQUIRED)
60+
61+
# Add the OpenMP-specific compiler and linker flags
62+
list(APPEND CMAKE_CXX_FLAGS ${OpenMP_CXX_FLAGS})
63+
list(APPEND CMAKE_C_FLAGS ${OpenMP_C_FLAGS})
64+
65+
endmacro(enable_openmp_support)
66+
# Done configuring OpenMP Options
67+
68+
69+
####################################################
70+
## ---------------------------------------------- ##
71+
## - - ##
72+
## - Enable CUDA Support - ##
73+
## - - ##
74+
## ---------------------------------------------- ##
75+
####################################################
76+
77+
# Begin configuring CUDA options
78+
# This is ugly...
79+
macro(enable_cuda_support)
80+
81+
# Hide a number of options from the default CMake screen
82+
mark_as_advanced(CLEAR CUDA_BUILD_CUBIN)
83+
mark_as_advanced(CLEAR CUDA_SDK_ROOT_DIR)
84+
mark_as_advanced(CLEAR CUDA_TOOLKIT_ROOT_DIR)
85+
mark_as_advanced(CLEAR CUDA_VERBOSE_BUILD)
86+
mark_as_advanced(CLEAR CUDA_FAST_MATH)
87+
mark_as_advanced(CLEAR CUDA_USE_CUSTOM_COMPILER)
88+
mark_as_advanced(CLEAR CUDA_VERBOSE_PTX)
89+
mark_as_advanced(CLEAR CUDA_DEVICE_VERSION)
90+
91+
# select Compute Capability
92+
# This needs to be manually updated when devices with new CCs come out
93+
set(CUDA_DEVICE_VERSION "20" CACHE STRING "CUDA Device Version")
94+
set_property(CACHE CUDA_DEVICE_VERSION PROPERTY STRINGS "10" "11" "12" "13" "20" "21" "30" "35")
95+
96+
# Enable fast-math for CUDA (_not_ GCC)
97+
set(CUDA_FAST_MATH TRUE CACHE BOOL "Use Fast Math Operations")
98+
99+
# Tell nvcc to use a separate compiler for non-CUDA code.
100+
# This is useful if you need to use an older of GCC than comes by default
101+
set(CUDA_USE_CUSTOM_COMPILER FALSE CACHE BOOL "Use Custom Compiler")
102+
set(CUDA_CUSTOM_COMPILER "" CACHE STRING "Custom C++ Compiler for CUDA If Needed")
103+
104+
# Shows register usage, etc
105+
set(CUDA_VERBOSE_PTX TRUE CACHE BOOL "Show Verbose Kernel Info During Compilation")
106+
107+
108+
# Let's get going...
109+
find_package("CUDA" REQUIRED)
110+
111+
# Frequently used in the examples
112+
cuda_include_directories(${CUDA_SDK_ROOT_DIR}/common/inc)
113+
cuda_include_directories(${CUDA_SDK_ROOT_DIR}/../shared/inc)
114+
115+
set(CUDA_SDK_LIB_DIR ${CUDA_SDK_ROOT_DIR}/common/lib
116+
${CUDA_SDK_ROOT_DIR}/lib ${CUDA_SDK_ROOT_DIR}/../shared/lib)
117+
118+
# these are no longer needed
119+
# # Find path to shrutil libs, from CUDA SDK
120+
# find_library(LIBSHRUTIL
121+
# NAMES shrUtils${CMAKE_ARCH_BITNESS} shrutil_${CMAKE_SYSTEM_PROCESSOR}
122+
# PATHS ${CUDA_SDK_LIB_DIR})
123+
# find_library(LIBSHRUTIL_DBG
124+
# NAMES shrUtils${CMAKE_ARCH_BITNESS}D shrutil_${CMAKE_SYSTEM_PROCESSOR}D
125+
# PATHS ${CUDA_SDK_LIB_DIR})
126+
#
127+
# # Find path to cutil libs, from CUDA SDK
128+
# find_library(LIBCUTIL
129+
# NAMES cutil${CMAKE_ARCH_BITNESS} cutil_${CMAKE_SYSTEM_PROCESSOR}
130+
# PATHS ${CUDA_SDK_LIB_DIR})
131+
# find_library(LIBCUTIL_DBG
132+
# NAMES cutil${arch}D cutil_${CMAKE_SYSTEM_PROCESSOR}D
133+
# PATHS ${CUDA_SDK_LIB_DIR})
134+
135+
# Set custom compiler flags
136+
set(CUDA_NVCC_FLAGS "" CACHE STRING "" FORCE)
137+
138+
if(CUDA_USE_CUSTOM_COMPILER)
139+
mark_as_advanced(CLEAR CUDA_CUSTOM_COMPILER)
140+
list(APPEND CUDA_NVCC_FLAGS "-ccbin=${CUDA_CUSTOM_COMPILER}")
141+
else()
142+
mark_as_advanced(FORCE CUDA_CUSTOM_COMPILER)
143+
endif()
144+
145+
# Macro for setting the Compute Capability
146+
macro(set_compute_capability cc)
147+
list(APPEND CUDA_NVCC_FLAGS "-gencode=arch=compute_${cc},code=sm_${cc}")
148+
list(APPEND CUDA_NVCC_FLAGS "-gencode=arch=compute_${cc},code=compute_${cc}")
149+
endmacro(set_compute_capability)
150+
151+
# Tell nvcc to compile for the selected Compute Capability
152+
# This can also be called from the main CMakeLists.txt to enable
153+
# support for additional CCs
154+
set_compute_capability(${CUDA_DEVICE_VERSION})
155+
156+
# Enable fast-math if selected
157+
if(CUDA_FAST_MATH)
158+
list(APPEND CUDA_NVCC_FLAGS "-use_fast_math")
159+
endif()
160+
161+
# Enable verbose compile if selected
162+
if(CUDA_VERBOSE_PTX)
163+
list(APPEND CUDA_NVCC_FLAGS "--ptxas-options=-v")
164+
endif()
165+
endmacro(enable_cuda_support)
166+
# Done configuring CUDA options

0 commit comments

Comments
 (0)