root / BLAS / xGEMM / Makefile @ 264
Historique | Voir | Annoter | Télécharger (3,97 ko)
1 |
# Makefile designed for Debian Buster |
---|---|
2 |
|
3 |
SOURCE=xGEMM.c |
4 |
|
5 |
CC=gcc |
6 |
FC=gfortran |
7 |
CFLAGS=-Wall -O3 -ffast-math |
8 |
LDFLAGS=-lm |
9 |
|
10 |
CLBLASINC=/usr/include |
11 |
CLBLASLIB=/usr/lib/x86_64-linux-gnu |
12 |
|
13 |
CUDALIB?=/usr/lib/x86_64-linux-gnu/ |
14 |
CUDAINC?=/usr/include |
15 |
CUDASRC?=/usr/share/doc/nvidia-cuda-toolkit/examples/libcublas |
16 |
THUNKING=fortran_thunking.c |
17 |
CUDASRCINC=fortran_common.h |
18 |
|
19 |
GSLINC=/usr/include/gsl |
20 |
|
21 |
# OPENBLAS=/usr/lib/x86_64-linux-gnu |
22 |
# ATLAS=/usr/lib/x86_64-linux-gnu |
23 |
|
24 |
ACML=/opt/acml |
25 |
ACMLINC=$(ACML)/gfortran64_mp/include |
26 |
ACMLLIB=$(ACML)/gfortran64_mp/lib |
27 |
|
28 |
#EXECUTABLE=gsl cblas fblas openblas clblas cublas acml thunking |
29 |
EXECUTABLE=gsl cblas fblas openblas clblas cublas thunking |
30 |
|
31 |
# FORMAT define the precision of Floating Point numbers |
32 |
# FP32 for simple of 32 bits size |
33 |
# FP64 for double of 64 bits size |
34 |
FORMAT=FP64 |
35 |
#FORMAT=FP32 |
36 |
|
37 |
#DIRECTIVES=-D$(FORMAT) -DPRINT -DUNIT |
38 |
#DIRECTIVES=-D$(FORMAT) -DUNIT -DRESULTS -DQUIET |
39 |
DIRECTIVES=-DUNIT -DQUIET |
40 |
|
41 |
all: $(EXECUTABLE) |
42 |
|
43 |
cblas: $(SOURCE) |
44 |
|
45 |
$(CC) $(CFLAGS) $(DIRECTIVES) -DFP32 -DCBLAS $(LDFLAGS) \ |
46 |
$(SOURCE) -latlas -lcblas -o $(SOURCE:.c=)_SP_$@ |
47 |
|
48 |
$(CC) $(CFLAGS) $(DIRECTIVES) -DFP64 -DCBLAS $(LDFLAGS) \ |
49 |
$(SOURCE) -latlas -lcblas -o $(SOURCE:.c=)_DP_$@ |
50 |
|
51 |
openblas: $(SOURCE) |
52 |
|
53 |
$(CC) $(CFLAGS) $(DIRECTIVES) -DFP32 -DCBLAS $(LDFLAGS) \ |
54 |
$(SOURCE) -lopenblas -lpthread -o $(SOURCE:.c=)_SP_$@ |
55 |
|
56 |
$(CC) $(CFLAGS) $(DIRECTIVES) -DFP64 -DCBLAS $(LDFLAGS) \ |
57 |
$(SOURCE) -lopenblas -lpthread -o $(SOURCE:.c=)_DP_$@ |
58 |
|
59 |
acml: $(SOURCE) |
60 |
|
61 |
$(CC) -I$(ACMLINC) $(CFLAGS) $(DIRECTIVES) -DFP32 -DACML $(LDFLAGS) \ |
62 |
$(SOURCE) -L$(ACMLLIB) -lacml_mp \ |
63 |
-lgomp -lgfortran -lpthread -o $(SOURCE:.c=)_SP_$@ |
64 |
|
65 |
$(CC) -I$(ACMLINC) $(CFLAGS) $(DIRECTIVES) -DFP64 -DACML $(LDFLAGS) \ |
66 |
$(SOURCE) -L$(ACMLLIB) -lacml_mp \ |
67 |
-lgomp -lgfortran -lpthread -o $(SOURCE:.c=)_DP_$@ |
68 |
|
69 |
fblas: $(SOURCE) |
70 |
|
71 |
# ATLAS version |
72 |
# $(CC) -L$(ATLAS) $(CFLAGS) $(DIRECTIVES) -DFP32 -DFBLAS $(LDFLAGS) \ |
73 |
# $(SOURCE) -latlas -lf77blas -o $(SOURCE:.c=)_SP_$@ |
74 |
|
75 |
# $(CC) -L$(ATLAS) $(CFLAGS) $(DIRECTIVES) -DFP64 -DFBLAS $(LDFLAGS) \ |
76 |
# $(SOURCE) -latlas -lf77blas -o $(SOURCE:.c=)_DP_$@ |
77 |
|
78 |
# OpenBLAS version |
79 |
$(CC) $(CFLAGS) $(DIRECTIVES) -DFP32 -DFBLAS $(LDFLAGS) \ |
80 |
$(SOURCE) -lopenblas -lpthread -o $(SOURCE:.c=)_SP_$@ |
81 |
|
82 |
$(CC) $(CFLAGS) $(DIRECTIVES) -DFP64 -DFBLAS $(LDFLAGS) \ |
83 |
$(SOURCE) -lopenblas -lpthread -o $(SOURCE:.c=)_DP_$@ |
84 |
|
85 |
|
86 |
gsl: $(SOURCE) |
87 |
|
88 |
$(CC) -I$(GSLINC) $(CFLAGS) $(DIRECTIVES) -DFP32 \ |
89 |
-DGSL $(LDFLAGS) \ |
90 |
$(SOURCE) -lgslcblas -o $(SOURCE:.c=)_SP_$@ |
91 |
|
92 |
$(CC) -I$(GSLINC) $(CFLAGS) $(DIRECTIVES) -DFP64 \ |
93 |
-DGSL $(LDFLAGS) \ |
94 |
$(SOURCE) -lgslcblas -o $(SOURCE:.c=)_DP_$@ |
95 |
|
96 |
clblas: $(SOURCE) |
97 |
|
98 |
$(CC) -I$(CLBLASINC) -L$(CLBLASLIB) $(CFLAGS) -DFP32 \ |
99 |
-DCLBLAS $(LDFLAGS) \ |
100 |
$(DIRECTIVES) $(SOURCE) -lclBLAS -lOpenCL -o $(SOURCE:.c=)_SP_$@ |
101 |
|
102 |
$(CC) -I$(CLBLASINC) -L$(CLBLASLIB) $(CFLAGS) -DFP64 \ |
103 |
-DCLBLAS $(LDFLAGS) \ |
104 |
$(DIRECTIVES) $(SOURCE) -lclBLAS -lOpenCL -o $(SOURCE:.c=)_DP_$@ |
105 |
|
106 |
cublas: $(SOURCE) |
107 |
|
108 |
$(CC) -I$(CUDAINC) -L$(CUDALIB) $(CFLAGS) -DFP32 \ |
109 |
-DCUBLAS $(LDFLAGS) \ |
110 |
$(DIRECTIVES) $(SOURCE) -lcublas -o $(SOURCE:.c=)_SP_$@ |
111 |
|
112 |
$(CC) -I$(CUDAINC) -L$(CUDALIB) $(CFLAGS) -DFP64 \ |
113 |
-DCUBLAS $(LDFLAGS) \ |
114 |
$(DIRECTIVES) $(SOURCE) -lcublas -o $(SOURCE:.c=)_DP_$@ |
115 |
|
116 |
thunking: $(SOURCE) |
117 |
|
118 |
# Copy of source for thunking CUBLAS approach |
119 |
cp $(CUDASRC)/$(THUNKING) $(CUDASRC)/$(THUNKING:.c=.h) \ |
120 |
$(CUDASRC)/$(CUDASRCINC) . |
121 |
# Patch Thunking prototypes to compile on Debian Lenny |
122 |
$(CC) -DCUBLAS_GFORTRAN -I$(CUDAINC) $(CFLAGS) -c $(THUNKING) |
123 |
|
124 |
$(CC) -DCUBLAS_GFORTRAN \ |
125 |
-I$(CUDAINC) -L$(CUDALIB) $(CFLAGS) -DFP32 -DTHUNKING \ |
126 |
$(LDFLAGS) $(DIRECTIVES) $(SOURCE) \ |
127 |
$(THUNKING:.c=.o) -lcublas -o $(SOURCE:.c=)_SP_$@ |
128 |
|
129 |
$(CC) -DCUBLAS_GFORTRAN \ |
130 |
-I$(CUDAINC) -L$(CUDALIB) $(CFLAGS) -DFP64 -DTHUNKING \ |
131 |
$(LDFLAGS) $(DIRECTIVES) $(SOURCE) \ |
132 |
$(THUNKING:.c=.o) -lcublas -o $(SOURCE:.c=)_DP_$@ |
133 |
|
134 |
clean: $(SOURCE) |
135 |
find . -name "$(SOURCE:.c=)_*" -exec rm {} \; |
136 |
find . -name "$(THUNKING:.c=)*" -exec rm {} \; |
137 |
find . -name "*~" -exec rm {} \; |
138 |
find . -name "$(CUDASRCINC)" -exec rm {} \; |
139 |
|
140 |
check: $(EXECUTABLE) |
141 |
|
142 |
$(SOURCE:.c=)_SP_$(EXECUTABLE) 2 1 |
143 |
|