don't use cblas anymore, but the same blas interface as plearn.

odelalleau · Jul 7, 2009 · 48f7a12 · 48f7a12
1 parent f176ec4
commit 48f7a12
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 11 deletions.
diff --git a/speedtest/Makefile b/speedtest/Makefile
@@ -3,7 +3,8 @@ CFLAGS=-Wall -fPIC -O3
 LIBCBLAS=~bastienf/.NOBACKUP/CBLAS/lib/${OSARCH}/libcblas.a
 
 LIBBLAS=-lblas ${LIBCBLAS}
-LIBGOTO=-lgoto -lgfortran ${LIBCBLAS}
+#LIBGOTO=-lgoto -lgfortran ${LIBCBLAS}
+LIBGOTO=-lgoto -lgfortran -I~/PLearn
 LIBATLAS=-lcblas -lf77blas -latlas -lg2c
 LIBMKL=-lmkl -lvml ${LIBCBLAS}
 LIBMKLPT=-lmkl -lvml -lguide -lpthread ${LIBCBLAS}
@@ -12,7 +13,8 @@ ACMLBASE=/u/lisa/local/acml-3-6-1beta/gfortran64
 ACMLBASE=/u/lisa/local/acml-3-6-0/gfortran32
 LIBACML= ${LIBCBLAS} ${ACMLBASE}/lib/libacml.a -lgfortran
 LIBACMLPT= ${LIBCBLAS} ${ACMLBASE}_mp/lib/libacml_mp.a -lg2c -lgfortran
-LIBNVIDIA=-L/u/bastienf/NVIDIA_CUDA_SDK/cuda/lib -L/u/bastienf/NVIDIA_CUDA_SDK/lib/ -I/u/bastienf/NVIDIA_CUDA_SDK/cuda/include/ -lcuda -lcudart -lGL -lGLU -lcublas -l cutil
+LIBNVIDIA=-L/u/bastienf/NVIDIA_CUDA_SDK/cuda/lib -L/u/bastienf/NVIDIA_CUDA_SDK/lib/ -I/u/bastienf/NVIDIA_CUDA_SDK/cuda/include/ -lcuda -lcudart -lGL -lGLU -lcublas 
+#-l cutil
 
 all: xgemm
 
@@ -85,6 +87,12 @@ xgemm-sgemm-nvidia: xgemm.c
 xgemm-sgemm-nvidia-compare: xgemm.c
 	${CC} ${CFLAGS} -m32 -o $@ -DUSEFLOAT $< ${LIBNVIDIA} -DNVIDIA -DCOMPARE
 
+xgemm-sgemm-nvidia-64: xgemm.c
+	${CC} ${CFLAGS} -m64 -o $@ -DUSEFLOAT $<  ${LIBNVIDIA} -DNVIDIA
+
+xgemm-sgemm-nvidia-compare-64: xgemm.c
+	${CC} ${CFLAGS} -m64 -o $@ -DUSEFLOAT $< ${LIBNVIDIA} -DNVIDIA -DCOMPARE
+
 xgemv: xgemv-blas xgemv-blas-compare xgemv-cxgemm xgemv-goto xgemv-goto-compare xgemv-nvidia xgemv-nvidia-compare xgemv-sgemv-goto xgemv-dgemv-goto xgemv-sgemv-blas xgemv-dgemv-blas
 	true
 

diff --git a/speedtest/xgemm.c b/speedtest/xgemm.c
@@ -9,15 +9,16 @@
   /* Includes, cuda */
   #include <cublas.h>
 #else
-  /* Includes, cblas */
-  #include <gsl/gsl_cblas.h>
+#include <plearn/math/blas_proto.h>
 #endif
 #ifdef USEDOUBLE
-  typedef double real;
-  #define cblas_xgemm cblas_dgemm
+//  typedef double real;
+#define real double
+  #define xgemm_ dgemm_
 #elif USEFLOAT
-  typedef float real;
-  #define cblas_xgemm cblas_sgemm
+//  typedef float real;
+#define real float
+  #define xgemm_ sgemm_
 #else
   #error "USEDOUBLE or USEFLOAT must be defined"
 #endif
@@ -34,7 +35,7 @@ static void c_xgemm(int M, int N, int K, const real alpha, const real *A,
   int k;
   for (i = 0; i < M; ++i) {
     for (j = 0; j < N; ++j) {
-      float prod = 0;
+      real prod = 0;
       for (k = 0; k < K; ++k) {
 	prod += A[i * K + k] * B[k * N + j];
       }
@@ -146,7 +147,7 @@ int main(int argc, char** argv)
       c_xgemm(M,N,K, alpha, h_A, h_B, beta, h_C);
     h_C_ref = h_C;
     /* Allocate host memory for reading back the result from device memory */
-    h_C = (float*)malloc(NC * sizeof(h_C[0]));
+    h_C = (real*)malloc(NC * sizeof(h_C[0]));
     if (h_C == 0) {
         fprintf (stderr, "!!!! host memory allocation error (C)\n");
         return EXIT_FAILURE;
@@ -174,8 +175,10 @@ int main(int argc, char** argv)
     for (int i=0;i<NBITER;i++)
       c_xgemm(M,N,K, alpha, h_A, h_B, beta, h_C);
 #else
+    char transa='N', transb='N';
     for (int i=0;i<NBITER;i++)
-      cblas_xgemm(CblasRowMajor, CblasNoTrans, CblasNoTrans, M,N,K, alpha, h_A, K, h_B, N, beta, h_C, N);
+      sgemm_(&transb, &transa, &N, &M, &K, &alpha, h_B, &N, h_A, &K, &beta, h_C, &N);
+
 #endif
 #ifdef COMPARE
     /* Check result against reference */