maggicl/HPC
Archived
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

hw1: ex2 optimization

Browse source
This commit is contained in:
Claudio Maggioni 2022-10-03 21:02:08 +02:00
parent df1cb3cf01
commit 86f9021ce5
3 changed files with 14 additions and 1978710 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1978702
Project1/project_1_maggioni_claudio/matmult/aaa.txt
View file

File diff suppressed because it is too large Load diff

18
Project1/project_1_maggioni_claudio/matmult/dgemm-blocked.c
View file

@ -17,17 +17,17 @@ LDLIBS = -lrt -Wl,--start-group $(MKLROOT)/lib/intel64/libmkl_intel_lp64.a $(MKL
const char* dgemm_desc = "Block-based dgemm."; const char* dgemm_desc = "Block-based dgemm.";
const int block_size = 18; const int block_size = 27;
inline int min(int a, int b) { inline int min(int a, int b) {
return a < b ? a : b; return a < b ? a : b;
} }
inline void naivemm(int r_min, int r_max, int k_min, int k_max, int c_min, int c_max, int n, double* A, double* B, double* C_temp) { inline void naivemm(int r_min, int r_max, int k_min, int k_max, int c_min, int c_max, int n, double* A_row, double* B, double* C_temp) {
for (int i = r_min, ii = 0; i < r_max; ++i, ++ii) { for (int i = r_min, ii = 0; i < r_max; ++i, ++ii) {
for (int j = c_min, jj = 0; j < c_max; ++j, ++jj) { for (int j = c_min, jj = 0; j < c_max; ++j, ++jj) {
for (int k = k_min; k < k_max; k++) { for (int k = k_min; k < k_max; k++) {
C_temp[ii + jj * block_size] += A[i + k * n] * B[k + j * n]; C_temp[ii + jj * block_size] += A_row[i * n + k] * B[k + j * n];
} }
} }
} }
@ -44,8 +44,18 @@ inline void store_c(double* C, double* C_temp, int r_min, int r_max, int c_min,
* where A, B, and C are lda-by-lda matrices stored in column-major format. * where A, B, and C are lda-by-lda matrices stored in column-major format.
* On exit, A and B maintain their input values. */ * On exit, A and B maintain their input values. */
void square_dgemm(int n, double* A, double* B, double* C) { void square_dgemm(int n, double* A, double* B, double* C) {
double A_row[n * n];
double C_temp[block_size * block_size]; double C_temp[block_size * block_size];
for (int m = 0; m < n; ++m) {
double row_tmp[n];
memcpy(row_tmp, A + m * n, n * sizeof(double));
for (int l = 0; l < n; ++l) {
A_row[l * n + m] = row_tmp[l];
}
}
for (int i = 0; i < n; i += block_size) { for (int i = 0; i < n; i += block_size) {
int i_next = min(i + block_size, n); int i_next = min(i + block_size, n);
@ -56,7 +66,7 @@ void square_dgemm(int n, double* A, double* B, double* C) {
for (int k = 0; k < n; k += block_size) { for (int k = 0; k < n; k += block_size) {
int k_next = min(k + block_size, n); int k_next = min(k + block_size, n);
naivemm(i, i_next, k, k_next, j, j_next, n, A, B, C_temp); naivemm(i, i_next, k, k_next, j, j_next, n, A_row, B, C_temp);
} }
store_c(C, C_temp, i, i_next, j, j_next, n); store_c(C, C_temp, i, i_next, j, j_next, n);

4
Project1/project_1_maggioni_claudio/matmult/timing_basic_dgemm.data
View file

@ -1,4 +0,0 @@
#Description: Naive, three-loop dgemm.
Size: 31 Mflop/s: 2431.2 Percentage: 6.61
#Average percentage of Peak = 6.60652