TREX/qmckl
1
0
Fork 0
mirror of https://github.com/TREX-CoE/qmckl.git synced 2025-01-03 18:16:28 +01:00
Code Issues Projects Releases Wiki Activity

Update names

Browse Source
This commit is contained in:
Anthony Scemama 2021-04-02 12:04:24 +02:00
parent 249c145d0c
commit fbfe558937
1 changed files with 83 additions and 58 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

141
src/qmckl_tile.org
View File

@ -76,19 +76,20 @@ MunitResult test_<<filename()>>() {
tiles is fixed at compile-time to increase performance. It is tiles is fixed at compile-time to increase performance. It is
defined as $2^s$: defined as $2^s$:
| s | tile size | | s | tile size |
|----+-----------| |---+-----------|
| 2 | 4 | | 2 | 4 |
| 3 | 8 | | 3 | 8 |
| 4 | 16 | | 4 | 16 |
| 55 | 32 | | 5 | 32 |
| 6 | 64 | | 6 | 64 |
| 7 | 128 | | 7 | 128 |
#+begin_src c :tangle (eval h_private_type) #+begin_src c :tangle (eval h_private_type)
#define TILE_SIZE_SHIFT 3 #define TILE_SIZE_SHIFT 3
#define TILE_SIZE 8 #define TILE_SIZE 8
#define VEC_SIZE 8
#+end_src #+end_src
@ -96,8 +97,8 @@ MunitResult test_<<filename()>>() {
#+begin_src c #+begin_src c
typedef struct $T$_tile_struct { typedef struct $T$_tile_struct {
$T$ element[TILE_SIZE][TILE_SIZE]; $T$ element[TILE_SIZE][TILE_SIZE];
int32_t is_null; int64_t is_null;
int32_t padding; int64_t padding[VEC_SIZE-1];
} $T$_tile_struct; } $T$_tile_struct;
#+end_src #+end_src
@ -109,8 +110,10 @@ typedef struct $T$_tile_struct {
#+begin_src c #+begin_src c
typedef struct $T$_tiled_matrix { typedef struct $T$_tiled_matrix {
$T$_tile_struct** tile; $T$_tile_struct** tile;
size_t rows; size_t n_row;
size_t cols; size_t n_col;
size_t n_tile_row;
size_t n_tile_col;
} $T$_tiled_matrix; } $T$_tiled_matrix;
#+end_src #+end_src
@ -120,16 +123,37 @@ typedef struct $T$_tiled_matrix {
#+begin_src c #+begin_src c
qmckl_exit_code $T$_tiled_matrix_init (qmckl_context context, qmckl_exit_code $T$_tiled_matrix_init (qmckl_context context,
$T$_tiled_matrix* m, $T$_tiled_matrix* m,
size_t rows, size_t n_tile_row,
size_t cols); size_t n_tile_col);
#+end_src #+end_src
#+NAME: init_c #+NAME: init_c
#+begin_src c #+begin_src c
qmckl_exit_code $T$_tiled_matrix_init (qmckl_context context, qmckl_exit_code $T$_tiled_matrix_init (qmckl_context context,
$T$_tiled_matrix* m, $T$_tiled_matrix* m,
size_t rows, size_t n_tile_row,
size_t cols) { size_t n_tile_col) {
if (qmckl_context_check(context) == QMCKL_NULL_CONTEXT) {
} $T$_tiled_matrix;
#+end_src
When a tiled matrix is initialized, it is set to zero.
#+NAME: init_hpf
#+begin_src c
qmckl_exit_code $T$_tiled_matrix_init (qmckl_context context,
$T$_tiled_matrix* m,
size_t n_tile_row,
size_t n_tile_col);
#+end_src
#+NAME: init_c
#+begin_src c
qmckl_exit_code $T$_tiled_matrix_init (qmckl_context context,
$T$_tiled_matrix* m,
size_t n_tile_row,
size_t n_tile_col) {
if (qmckl_context_check(context) == QMCKL_NULL_CONTEXT) { if (qmckl_context_check(context) == QMCKL_NULL_CONTEXT) {
return QMCKL_INVALID_CONTEXT; return QMCKL_INVALID_CONTEXT;
@ -142,14 +166,14 @@ qmckl_exit_code $T$_tiled_matrix_init (qmckl_context context,
NULL); NULL);
} }
if (rows == (size_t) 0) { if (n_tile_row == (size_t) 0) {
return qmckl_failwith(context, return qmckl_failwith(context,
QMCKL_INVALID_ARG_3, QMCKL_INVALID_ARG_3,
"$T$_tiled_matrix_init", "$T$_tiled_matrix_init",
NULL); NULL);
} }
if (cols == (size_t) 0) { if (n_tile_col == (size_t) 0) {
return qmckl_failwith(context, return qmckl_failwith(context,
QMCKL_INVALID_ARG_4, QMCKL_INVALID_ARG_4,
"$T$_tiled_matrix_init", "$T$_tiled_matrix_init",
@ -157,19 +181,19 @@ qmckl_exit_code $T$_tiled_matrix_init (qmckl_context context,
} }
qmckl_memory_info_struct info = qmckl_memory_info_struct_zero; qmckl_memory_info_struct info = qmckl_memory_info_struct_zero;
size_t n = rows * cols; size_t n = n_tile_row * n_tile_col;
/* Check overflow */ /* Check overflow */
if (n/cols != rows if (n/n_tile_col != n_tile_row
|| n > SIZE_MAX / sizeof($T$_tile_struct) ) { || n > SIZE_MAX / sizeof($T$_tile_struct) ) {
return qmckl_failwith(context, return qmckl_failwith(context,
QMCKL_ALLOCATION_FAILED, QMCKL_ALLOCATION_FAILED,
"$T$_tiled_matrix_init", "$T$_tiled_matrix_init",
"rows * cols overflows" ); "n_tile_row * n_tile_col overflows" );
} }
/* Allocate array of column pointers */ /* Allocate array of column pointers */
info.size = cols * sizeof($T$_tile_struct*) ; info.size = n_tile_col * sizeof($T$_tile_struct*) ;
m->tile = ($T$_tile_struct**) qmckl_malloc(context, info); m->tile = ($T$_tile_struct**) qmckl_malloc(context, info);
if (m->tile == NULL) { if (m->tile == NULL) {
@ -192,15 +216,16 @@ qmckl_exit_code $T$_tiled_matrix_init (qmckl_context context,
} }
/* Compute array of pointers to the 1st element of columns */ /* Compute array of pointers to the 1st element of columns */
for (size_t i=1 ; i<cols ; ++i) { for (size_t i=1 ; i<n_tile_col ; ++i) {
m->tile[i] = m->tile[i-1] + rows; m->tile[i] = m->tile[i-1] + n_tile_row;
} }
m->rows = rows; m->n_tile_row = n_tile_row;
m->cols = cols; m->n_tile_col = n_tile_col;
return QMCKL_SUCCESS; return QMCKL_SUCCESS;
} }
#+end_src #+end_src
* Write templates * Write templates
@ -251,8 +276,8 @@ return '\n'.join( [ ""
typedef struct float_tiled_matrix { typedef struct float_tiled_matrix {
float_tile_struct** tile; float_tile_struct** tile;
size_t rows; size_t n_tile_row;
size_t cols; size_t n_tile_col;
} float_tiled_matrix; } float_tiled_matrix;
@ -265,8 +290,8 @@ return '\n'.join( [ ""
typedef struct double_tiled_matrix { typedef struct double_tiled_matrix {
double_tile_struct** tile; double_tile_struct** tile;
size_t rows; size_t n_tile_row;
size_t cols; size_t n_tile_col;
} double_tiled_matrix; } double_tiled_matrix;
@ -276,15 +301,15 @@ return '\n'.join( [ ""
qmckl_exit_code float_tiled_matrix_init (qmckl_context context, qmckl_exit_code float_tiled_matrix_init (qmckl_context context,
float_tiled_matrix* m, float_tiled_matrix* m,
size_t rows, size_t n_tile_row,
size_t cols); size_t n_tile_col);
qmckl_exit_code double_tiled_matrix_init (qmckl_context context, qmckl_exit_code double_tiled_matrix_init (qmckl_context context,
double_tiled_matrix* m, double_tiled_matrix* m,
size_t rows, size_t n_tile_row,
size_t cols); size_t n_tile_col);
#+end_src #+end_src
@ -293,8 +318,8 @@ return '\n'.join( [ ""
qmckl_exit_code float_tiled_matrix_init (qmckl_context context, qmckl_exit_code float_tiled_matrix_init (qmckl_context context,
float_tiled_matrix* m, float_tiled_matrix* m,
size_t rows, size_t n_tile_row,
size_t cols) { size_t n_tile_col) {
if (qmckl_context_check(context) == QMCKL_NULL_CONTEXT) { if (qmckl_context_check(context) == QMCKL_NULL_CONTEXT) {
return QMCKL_INVALID_CONTEXT; return QMCKL_INVALID_CONTEXT;
@ -307,14 +332,14 @@ return '\n'.join( [ ""
NULL); NULL);
} }
if (rows == (size_t) 0) { if (n_tile_row == (size_t) 0) {
return qmckl_failwith(context, return qmckl_failwith(context,
QMCKL_INVALID_ARG_3, QMCKL_INVALID_ARG_3,
"float_tiled_matrix_init", "float_tiled_matrix_init",
NULL); NULL);
} }
if (cols == (size_t) 0) { if (n_tile_col == (size_t) 0) {
return qmckl_failwith(context, return qmckl_failwith(context,
QMCKL_INVALID_ARG_4, QMCKL_INVALID_ARG_4,
"float_tiled_matrix_init", "float_tiled_matrix_init",
@ -322,19 +347,19 @@ return '\n'.join( [ ""
} }
qmckl_memory_info_struct info = qmckl_memory_info_struct_zero; qmckl_memory_info_struct info = qmckl_memory_info_struct_zero;
size_t n = rows * cols; size_t n = n_tile_row * n_tile_col;
/* Check overflow */ /* Check overflow */
if (n/cols != rows if (n/n_tile_col != n_tile_row
|| n > SIZE_MAX / sizeof(float_tile_struct) ) { || n > SIZE_MAX / sizeof(float_tile_struct) ) {
return qmckl_failwith(context, return qmckl_failwith(context,
QMCKL_ALLOCATION_FAILED, QMCKL_ALLOCATION_FAILED,
"float_tiled_matrix_init", "float_tiled_matrix_init",
"rows * cols overflows" ); "n_tile_row * n_tile_col overflows" );
} }
/* Allocate array of column pointers */ /* Allocate array of column pointers */
info.size = cols * sizeof(float_tile_struct*) ; info.size = n_tile_col * sizeof(float_tile_struct*) ;
m->tile = (float_tile_struct**) qmckl_malloc(context, info); m->tile = (float_tile_struct**) qmckl_malloc(context, info);
if (m->tile == NULL) { if (m->tile == NULL) {
@ -357,12 +382,12 @@ return '\n'.join( [ ""
} }
/* Compute array of pointers to the 1st element of columns */ /* Compute array of pointers to the 1st element of columns */
for (size_t i=1 ; i<cols ; ++i) { for (size_t i=1 ; i<n_tile_col ; ++i) {
m->tile[i] = m->tile[i-1] + rows; m->tile[i] = m->tile[i-1] + n_tile_row;
} }
m->rows = rows; m->n_tile_row = n_tile_row;
m->cols = cols; m->n_tile_col = n_tile_col;
return QMCKL_SUCCESS; return QMCKL_SUCCESS;
} }
@ -371,8 +396,8 @@ return '\n'.join( [ ""
qmckl_exit_code double_tiled_matrix_init (qmckl_context context, qmckl_exit_code double_tiled_matrix_init (qmckl_context context,
double_tiled_matrix* m, double_tiled_matrix* m,
size_t rows, size_t n_tile_row,
size_t cols) { size_t n_tile_col) {
if (qmckl_context_check(context) == QMCKL_NULL_CONTEXT) { if (qmckl_context_check(context) == QMCKL_NULL_CONTEXT) {
return QMCKL_INVALID_CONTEXT; return QMCKL_INVALID_CONTEXT;
@ -385,14 +410,14 @@ return '\n'.join( [ ""
NULL); NULL);
} }
if (rows == (size_t) 0) { if (n_tile_row == (size_t) 0) {
return qmckl_failwith(context, return qmckl_failwith(context,
QMCKL_INVALID_ARG_3, QMCKL_INVALID_ARG_3,
"double_tiled_matrix_init", "double_tiled_matrix_init",
NULL); NULL);
} }
if (cols == (size_t) 0) { if (n_tile_col == (size_t) 0) {
return qmckl_failwith(context, return qmckl_failwith(context,
QMCKL_INVALID_ARG_4, QMCKL_INVALID_ARG_4,
"double_tiled_matrix_init", "double_tiled_matrix_init",
@ -400,19 +425,19 @@ return '\n'.join( [ ""
} }
qmckl_memory_info_struct info = qmckl_memory_info_struct_zero; qmckl_memory_info_struct info = qmckl_memory_info_struct_zero;
size_t n = rows * cols; size_t n = n_tile_row * n_tile_col;
/* Check overflow */ /* Check overflow */
if (n/cols != rows if (n/n_tile_col != n_tile_row
|| n > SIZE_MAX / sizeof(double_tile_struct) ) { || n > SIZE_MAX / sizeof(double_tile_struct) ) {
return qmckl_failwith(context, return qmckl_failwith(context,
QMCKL_ALLOCATION_FAILED, QMCKL_ALLOCATION_FAILED,
"double_tiled_matrix_init", "double_tiled_matrix_init",
"rows * cols overflows" ); "n_tile_row * n_tile_col overflows" );
} }
/* Allocate array of column pointers */ /* Allocate array of column pointers */
info.size = cols * sizeof(double_tile_struct*) ; info.size = n_tile_col * sizeof(double_tile_struct*) ;
m->tile = (double_tile_struct**) qmckl_malloc(context, info); m->tile = (double_tile_struct**) qmckl_malloc(context, info);
if (m->tile == NULL) { if (m->tile == NULL) {
@ -435,12 +460,12 @@ return '\n'.join( [ ""
} }
/* Compute array of pointers to the 1st element of columns */ /* Compute array of pointers to the 1st element of columns */
for (size_t i=1 ; i<cols ; ++i) { for (size_t i=1 ; i<n_tile_col ; ++i) {
m->tile[i] = m->tile[i-1] + rows; m->tile[i] = m->tile[i-1] + n_tile_row;
} }
m->rows = rows; m->n_tile_row = n_tile_row;
m->cols = cols; m->n_tile_col = n_tile_col;
return QMCKL_SUCCESS; return QMCKL_SUCCESS;
} }