fanasina/y_project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

another debug with valgrind in fmock and tensor

Browse Source
This commit is contained in:
fanasina
2024-02-12 01:34:08 +01:00
parent 877891987a
commit 60e4b16308
12 changed files with 318 additions and 39 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tensor_t/src/tensor_t/cl_tensor_t.c
+14 -5
View File
@@ -3,6 +3,12 @@
#define MAX_SOURCE_SIZE (0x100000)
#define FREE_dM_S_\
free_dimension(dM0);\
free_dimension(dM1);\
free_dimension(dM);\
free_dimension(dSub0);\
free_dimension(dSub1);\
#define SETUP_cl_KERNEL_(type,file_cl_src,func_cl_name)\
@@ -18,7 +24,7 @@
fprintf(stderr, "Failed to load kernel. \n"); \
exit(1); \
} \
source_str = (char*)malloc(MAX_SOURCE_SIZE); \
source_str = malloc(MAX_SOURCE_SIZE); \
source_size = fread( source_str, 1, MAX_SOURCE_SIZE, fp); \
fclose( fp ); \
\
@@ -85,10 +91,11 @@
char *log = malloc(sizeof(char)*len);\
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, len, log, NULL);\
printf("Error: Failed to build program executable!\n %s \n",log);\
free(log);\
}\
/*/ Create the OpenCL kernel */ \
/*char func_cl_name[250]; sprintf(func_cl_name,"prodTensorLin_%s", #type);*/ \
printf("cl_func_type = %s\n",func_cl_name); \
/*printf("cl_func_type = %s\n",func_cl_name); */ \
cl_kernel kernel = clCreateKernel(program, func_cl_name, &ret); \
@@ -194,7 +201,7 @@ void cl_tensorContractnProd_##type(tensor_##type** MM, tensor_##type *M0, tensor
dimension *dM1 = init_dim(tDk1, contractionNumber);\
dimension *dM0 = init_dim(tDk0, contractionNumber);\
dimension *dM;\
min_dimension(&dM, dM0, dM1);\
min_copy_dimension(&dM, dM0, dM1);\
\
dimension *dd;\
add_dimension(&dd, dSub0, dSub1);\
@@ -230,6 +237,7 @@ void cl_tensorContractnProd_##type(tensor_##type** MM, tensor_##type *M0, tensor
EXEC_cl_KERNEL(type);\
READ_BUF_N_CLEANUP(type)\
\
FREE_dM_S_ \
} \
\
\
@@ -283,7 +291,7 @@ void cl2d_tensorContractnProd_##type(tensor_##type **MM, tensor_##type *M0, tens
dimension *dM1 = init_dim(tDk1, contractionNumber);\
dimension *dM0 = init_dim(tDk0, contractionNumber);\
dimension *dM;\
min_dimension(&dM, dM0, dM1);\
min_copy_dimension(&dM, dM0, dM1);\
\
dimension *dd;\
add_dimension(&dd, dSub0, dSub1);\
@@ -320,7 +328,8 @@ void cl2d_tensorContractnProd_##type(tensor_##type **MM, tensor_##type *M0, tens
/*printf("EXEC_cl_2d_KERNEL(type,%ld,%ld,%ld,%ld)\n",dSub0->rank,dSub1->rank,div0Wsz,div1Wsz);\
*/EXEC_cl_2d_KERNEL(type,dSub0->rank,dSub1->rank,div0Wsz,div1Wsz);\
READ_BUF_N_CLEANUP(type)\
\
FREE_dM_S_ \
\
} \
\
tensor_t/src/tensor_t/tensor_t.c
+93
View File
@@ -506,6 +506,7 @@ void tensorProdThrea2d_##type(tensor_##type **MM, tensor_##type *M0, tensor_##ty
free(thrd);\
free(arg_th);\
} \
\
struct arg_ProdContract_##type{\
type *M0x;\
type *M1x;\
@@ -603,6 +604,98 @@ void tensorContractnProdThread_##type(tensor_##type** MM, tensor_##type *M0, ten
free(arg_th);\
FREE_dM_S_ ; \
}\
\
struct arg_Pro2dContract_##type{\
type *M0x;\
type *M1x;\
type *Mx;\
size_t beginRange;\
size_t endRange;\
size_t dMRank;\
size_t dSub0Rank;\
size_t dSub1Rank;\
};\
void* runPro2dContract_thread_##type(void *arg){\
struct arg_Pro2dContract_##type *arg_t = arg;\
size_t n0_id, n1_id, l;\
for (size_t i = arg_t->beginRange; i < arg_t->endRange; i++) {\
for (size_t j = 0; j < arg_t->dSub1Rank; j++) {\
if(endian)\
l = j + arg_t->dSub1Rank * i;\
else\
l = j * arg_t->dMRank + i;\
arg_t->Mx[l] = 0;\
for (size_t k = 0; k < arg_t->dMRank; k++) {\
if(endian){\
n0_id= i * arg_t->dMRank + k;\
n1_id= j + arg_t->dSub1Rank * k;\
}\
else{\
n0_id= i + arg_t->dSub0Rank * k;\
n1_id= j * arg_t->dMRank + k;\
}\
arg_t->Mx[l] += arg_t->M0x[n0_id] * arg_t->M1x[n1_id];\
}\
}\
}\
}\
/* M[x0,x1,x3..xn] X M[y0,y1,y3..ym] = M[z0,z1...zp] (deep = l > 0) /exists 1<= l<...<l=n / xl = y0,x{l+1}=y1, x{n}=yl et zi=xi i<n-l et zj=y{j-(n-l)} j>=n-l alor p=n+m-2l\
M[x0,x1,x3..xl x{l+1}...xn] X M[xn,x{n-1},x{n-2}...xl y{l+1} ..ym] = M[x0,x1..xly{l+1}...y{n+m-2l}] (deep = l > 0)\
M[[i][j]]=sum_{[k]}M0[[i][k]]*M[[k][j]]*/\
\
void tensorContractnPro2dThread_##type(tensor_##type** MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread) {\
\
size_t len0 = M0->dim->size - contractionNumber;\
size_t len1 = M1->dim->size - contractionNumber;\
\
size_t* tsub0 = malloc(sizeof(size_t) *len0);\
size_t* tsub1 = malloc(sizeof(size_t) *len1);\
size_t* tDk1 = malloc(sizeof(size_t) *contractionNumber);\
size_t* tDk0 = malloc(sizeof(size_t) *contractionNumber);\
subArray(tsub0, M0->dim->perm, 0, len0, 0);\
subArray(tsub1, M1->dim->perm, 0, len1, contractionNumber);\
subArray(tDk1, M1->dim->perm, 0, contractionNumber, 0);\
subArray(tDk0, M0->dim->perm, 0, contractionNumber, len0);\
dimension *dSub0 = init_dim(tsub0, len0);\
dimension *dSub1 = init_dim(tsub1, len1);\
dimension *dM1 = init_dim(tDk1, contractionNumber);\
dimension *dM0 = init_dim(tDk0, contractionNumber);\
dimension *dM;\
min_copy_dimension(&dM, dM0, dM1);\
\
dimension *dd;\
add_dimension(&dd, dSub0, dSub1);\
updateRankDim(dd);\
*MM = CREATE_TENSOR_##type(dd);\
tensor_##type *M= *MM;\
\
\
\
pthread_t *thrd = malloc(nbthread * sizeof(pthread_t));\
struct arg_Pro2dContract_##type **arg_th = malloc( nbthread * sizeof(struct arg_Pro2dContract_##type *));\
\
for(size_t i = 0; i < nbthread; ++i) {\
arg_th[i] = malloc(sizeof(struct arg_Pro2dContract_##type));\
arg_th[i]->M0x=M0->x;\
arg_th[i]->M1x=M1->x;\
arg_th[i]->Mx=M->x;\
arg_th[i]->beginRange = i*(dSub0->rank)/nbthread ;\
arg_th[i]->endRange = (i+1)*(dSub0->rank)/nbthread ;\
arg_th[i]->dSub1Rank = dSub1->rank;\
arg_th[i]->dSub0Rank = dSub0->rank;\
arg_th[i]->dMRank = dM->rank;\
pthread_create(&thrd[i], NULL, runPro2dContract_thread_##type, (void*)arg_th[i]);\
}\
\
for(size_t i=0; i< nbthread; ++i){\
pthread_join(thrd[i], NULL);\
free(arg_th[i]);\
}\
\
free(thrd);\
free(arg_th);\
FREE_dM_S_ ; \
}\
void tensorContractnProdNotOpt_##type(tensor_##type** MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber) {\
/* if (!checkMatchProdtensor(M0->dim, M1->dim, contractionNumber)) {\
prsize_tf("Deep = %d\n", contractionNumber);\
tensor_t/src/tensor_t/tensor_t.h
+1
View File
@@ -29,6 +29,7 @@ void tensorContractnProd_##type(tensor_##type **MM, tensor_##type *M0, tensor_##
void tensorProdThread_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1,size_t nbthread); \
void tensorProdThrea2d_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1,size_t nbthread); \
void tensorContractnProdThread_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread); \
void tensorContractnPro2dThread_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber, size_t nbthread); \
void tensorContractnProdNotOpt_##type(tensor_##type **MM, tensor_##type *M0, tensor_##type *M1, size_t contractionNumber); \
tensor_t/test/is_good.c
+73 -12
View File
@@ -17,7 +17,7 @@
//#include "permutation_t/permutation_t.h"
#include "tensor_t/tensor_t.h"
#define VALGRING_ 1
#define VALGRIND_ 0
TEST(rank){
dimension *D=create_dim(4);
@@ -40,7 +40,7 @@ void print_tensor_float(tensor_TYPE_FLOAT *M, char *msg){
}
LOG("================= %s ===============\n",msg);
#if VALGRING_
#if VALGRIND_
for(size_t i=0; i<M->dim->rank;++i)
LOG("[%ld]: %f ",i,M->x[i]);
@@ -51,7 +51,7 @@ void print_tensor_float(tensor_TYPE_FLOAT *M, char *msg){
void print_tensor_double(tensor_TYPE_DOUBLE *M, char *msg){
LOG("================= %s ===============\n",msg);
#if VALGRING_
#if VALGRIND_
for(size_t i=0; i<M->dim->rank;++i)
LOG("[%ld]: %lf ",i,M->x[i]);
@@ -262,7 +262,7 @@ TEST(tensorSubtail ){
TEST(tensorProd ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(2);
#if VALGRING_
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
@@ -315,7 +315,7 @@ TEST(tensorProd ){
TEST(tensorContractnProd_TYPE_FLOAT ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRING_
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
@@ -377,7 +377,7 @@ TEST(tensorContractnProd_TYPE_FLOAT ){
TEST(tensorContractnProd_TYPE_FLOAT2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRING_
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
@@ -438,7 +438,7 @@ TEST(tensorContractnProd_TYPE_FLOAT2 ){
TEST(tensorContractnProd_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRING_
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
@@ -499,7 +499,7 @@ TEST(tensorContractnProd_TYPE_DOUBLE2 ){
TEST(VStensorContractnProd_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRING_
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
@@ -555,11 +555,72 @@ TEST(VStensorContractnProd_TYPE_DOUBLE2 ){
free_tensor_TYPE_DOUBLE(M1);
}
TEST(Pthread_tensorContractnPro2d_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
d1->perm[0]=2;
d1->perm[1]=3;//3;
d1->perm[2]=8;
#else
d0->perm[0]=125;
d0->perm[1]=52; //3;
d0->perm[2]=63;
d1->perm[0]=52;
d1->perm[1]=63;//3;
d1->perm[2]=154;
#endif
updateRankDim(d0);
updateRankDim(d1);
tensor_TYPE_DOUBLE *M0 = CREATE_TENSOR_TYPE_DOUBLE(d0);
tensor_TYPE_DOUBLE *M1 = CREATE_TENSOR_TYPE_DOUBLE(d1);
LOG("M0->dim->rank = %ld\n",M0->dim->rank);
LOG("M1->dim->rank = %ld\n",M1->dim->rank);
for(size_t i=0; i<M0->dim->rank;++i) M0->x[i]=i*0.1 +1;
for(size_t i=0; i<M1->dim->rank;++i) M1->x[i]=i*0.003 + 2;
//print_tensor_double(M0,"M0");
//print_tensor_double(M1,"M1");
tensor_TYPE_DOUBLE *M;
tensor_TYPE_DOUBLE *MnO;
size_t nbthread = 5;
tensorContractnProd_TYPE_DOUBLE(&M, M0,M1,2);
//print_tensor_double(M,"M");
//cl_tensorContractnProd_TYPE_DOUBLE(&MnO, M0,M1,2);
tensorContractnPro2dThread_TYPE_DOUBLE(&MnO, M0,M1,2,nbthread);
//print_tensor_double(MnO,"MnO");
// for(size_t i=0;i<M->dim->rank;++i)
// EXPECT_EQ_TYPE_DOUBLE(M->x[i],MnO->x[i]);
EXPECT_ARRAY_EQ_TYPE_DOUBLE(M->x,M->dim->rank,MnO->x,MnO->dim->rank);
free_tensor_TYPE_DOUBLE(M);
free_tensor_TYPE_DOUBLE(MnO);
free_tensor_TYPE_DOUBLE(M0);
free_tensor_TYPE_DOUBLE(M1);
}
TEST(Pthread_tensorContractnProd_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRING_
#if VALGRIND_
d0->perm[0]=5;
d0->perm[1]=2; //3;
d0->perm[2]=3;
@@ -624,7 +685,7 @@ TEST(tensorProd_vs ){
dimension *d1=create_dim(2);
#if VALGRING_
#if VALGRIND_
d0->perm[0]=2;
d0->perm[1]=3;
d0->perm[2]=2;
@@ -676,7 +737,7 @@ TEST(tensorProd_vsThread ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(2);
#if VALGRING_
#if VALGRIND_
d0->perm[0]=2;
d0->perm[1]=3;
d0->perm[2]=2;
@@ -734,7 +795,7 @@ TEST(tensorProd_vsThread2d ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(2);
#if VALGRING_
#if VALGRIND_
d0->perm[0]=2;
d0->perm[1]=3;
d0->perm[2]=2;
tensor_t/test_cl/is_good.c
+90 -4
View File
@@ -19,6 +19,7 @@
#include "tensor_t/cl_tensor_t.h"
//#include "tools_t/tools_t.h"
#define VALGRIND 1
TEST(rank){
dimension *D=create_dim(4);
@@ -36,19 +37,22 @@ TEST(rank){
void print_tensor_float(tensor_TYPE_FLOAT *M, char *msg){
LOG("================= %s ===============\n",msg);
#if VALGRIND
for(size_t i=0; i<M->dim->rank;++i)
LOG("[%ld]: %f ",i,M->x[i]);
LOG("%s","\n");
#endif
}
void print_tensor_double(tensor_TYPE_DOUBLE *M, char *msg){
LOG("================= %s ===============\n",msg);
#if VALGRIND
for(size_t i=0; i<M->dim->rank;++i)
LOG("[%ld]: %lf ",i,M->x[i]);
LOG("%s","\n");
#endif
}
extern long int PRECISION_TYPE_FLOAT ;
@@ -302,7 +306,17 @@ TEST(cl_tensorContractnProd_TYPE_DOUBLE2 ){
TEST(tensorContractnProd_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND
d0->perm[0]=12;
d0->perm[1]=5; //3;
d0->perm[2]=6;
d1->perm[0]=5;
d1->perm[1]=6;//3;
d1->perm[2]=14;
#else
d0->perm[0]=125;
d0->perm[1]=52; //3;
d0->perm[2]=63;
@@ -310,7 +324,7 @@ TEST(tensorContractnProd_TYPE_DOUBLE2 ){
d1->perm[0]=52;
d1->perm[1]=63;//3;
d1->perm[2]=54;
#endif
updateRankDim(d0);
updateRankDim(d1);
@@ -352,6 +366,7 @@ TEST(TensorProdCL){
dimension *d0=create_dim(3);
dimension *d1=create_dim(2);
d0->perm[0]=2;
d0->perm[1]=3;
d0->perm[2]=2;
@@ -375,6 +390,7 @@ TEST(TensorProdCL){
print_tensor_float(M1,"M1");
tensor_TYPE_FLOAT *M;
tensor_TYPE_FLOAT *Mn;
@@ -399,7 +415,17 @@ TEST(TensorProdCL){
TEST(VS_thrd_tensorContractnProd_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND
d0->perm[0]=12;
d0->perm[1]=4; //3;
d0->perm[2]=6;
d1->perm[0]=4;
d1->perm[1]=6;//3;
d1->perm[2]=16;
#else
d0->perm[0]=512;
d0->perm[1]=48; //3;
@@ -408,6 +434,7 @@ TEST(VS_thrd_tensorContractnProd_TYPE_DOUBLE2 ){
d1->perm[0]=48;
d1->perm[1]=64;//3;
d1->perm[2]=240;
#endif
updateRankDim(d0);
updateRankDim(d1);
@@ -457,7 +484,17 @@ TEST(VS_thrd_tensorContractnProd_TYPE_DOUBLE2 ){
TEST(VS_thrd_tensorContractnProd_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND
d0->perm[0]=12;
d0->perm[1]=4; //3;
d0->perm[2]=6;
d1->perm[0]=4;
d1->perm[1]=6;//3;
d1->perm[2]=16;
#else
d0->perm[0]=512;
d0->perm[1]=48; //3;
@@ -466,6 +503,9 @@ TEST(VS_thrd_tensorContractnProd_TYPE_DOUBLE2 ){
d1->perm[0]=48;
d1->perm[1]=64;//3;
d1->perm[2]=240;
#endif
updateRankDim(d0);
updateRankDim(d1);
@@ -519,7 +559,17 @@ TEST(VS_thrd_tensorContractnProd_TYPE_DOUBLE2 ){
TEST(VS_thrd_tensorContractnProd_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND
d0->perm[0]=12;
d0->perm[1]=4; //3;
d0->perm[2]=6;
d1->perm[0]=4;
d1->perm[1]=6;//3;
d1->perm[2]=16;
#else
d0->perm[0]=512;
d0->perm[1]=48; //3;
@@ -528,6 +578,9 @@ TEST(VS_thrd_tensorContractnProd_TYPE_DOUBLE2 ){
d1->perm[0]=48;
d1->perm[1]=64;//3;
d1->perm[2]=240;
#endif
updateRankDim(d0);
updateRankDim(d1);
@@ -578,13 +631,24 @@ TEST(VS_thrd_tensorContractnProd_TYPE_DOUBLE2 ){
#if 1
TEST(VScltensorContractnProd_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND
d0->perm[0]=12;
d0->perm[1]=4; //3;
d0->perm[2]=6;
d1->perm[0]=4;
d1->perm[1]=6;//3;
d1->perm[2]=16;
#else
d0->perm[0]=512;
d0->perm[1]=48; //3;
d0->perm[2]=64;
@@ -592,6 +656,8 @@ TEST(VScltensorContractnProd_TYPE_DOUBLE2 ){
d1->perm[0]=48;
d1->perm[1]=64;//3;
d1->perm[2]=240;
#endif
updateRankDim(d0);
updateRankDim(d1);
@@ -636,6 +702,27 @@ TEST(VScltensorContractnProd_TYPE_DOUBLE2 ){
TEST(VScl2dtensorContractnProd_TYPE_DOUBLE2 ){
dimension *d0=create_dim(3);
dimension *d1=create_dim(3);
#if VALGRIND
d0->perm[0]=12;
d0->perm[1]=4; //3;
d0->perm[2]=6;
d1->perm[0]=4;
d1->perm[1]=6;//3;
d1->perm[2]=16;
#else
d0->perm[0]=512;
d0->perm[1]=48; //3;
d0->perm[2]=64;
d1->perm[0]=48;
d1->perm[1]=64;//3;
d1->perm[2]=240;
#endif
d0->perm[0]=512;
d0->perm[1]=48; //3;
@@ -876,7 +963,6 @@ TEST(tensorProd_vs2d_Endian ){
free_tensor_TYPE_FLOAT(M1);
}
#endif
int main(int argc, char **argv){