diff options
Diffstat (limited to 'mt/ab_matmul/ab_matmul.c')
| -rwxr-xr-x | mt/ab_matmul/ab_matmul.c | 246 |
1 files changed, 0 insertions, 246 deletions
diff --git a/mt/ab_matmul/ab_matmul.c b/mt/ab_matmul/ab_matmul.c deleted file mode 100755 index 6530a5d..0000000 --- a/mt/ab_matmul/ab_matmul.c +++ /dev/null @@ -1,246 +0,0 @@ -//************************************************************************** -// Multi-threaded Matrix Multiply benchmark -//-------------------------------------------------------------------------- -// TA : Christopher Celio -// Student: -// -// -// This benchmark multiplies two 2-D arrays together and writes the results to -// a third vector. The input data (and reference data) should be generated -// using the matmul_gendata.pl perl script and dumped to a file named -// dataset.h. - - -// print out arrays, etc. -//#define DEBUG - -//-------------------------------------------------------------------------- -// Includes - -#include <string.h> -#include <stdlib.h> -#include <stdio.h> - - -//-------------------------------------------------------------------------- -// Input/Reference Data - -typedef float data_t; -#include "dataset.h" - - -//-------------------------------------------------------------------------- -// Basic Utilities and Multi-thread Support - -__thread unsigned long coreid; -unsigned long ncores; - -#include "util.h" - -#define stringify_1(s) #s -#define stringify(s) stringify_1(s) -#define stats(code) do { \ - unsigned long _c = -rdcycle(), _i = -rdinstret(); \ - code; \ - _c += rdcycle(), _i += rdinstret(); \ - if (coreid == 0) \ - printf("%s: %ld cycles, %ld.%ld cycles/iter, %ld.%ld CPI\n", \ - stringify(code), _c, _c/DIM_SIZE/DIM_SIZE/DIM_SIZE, 10*_c/DIM_SIZE/DIM_SIZE/DIM_SIZE%10, _c/_i, 10*_c/_i%10); \ - } while(0) - - -//-------------------------------------------------------------------------- -// Helper functions - -void printArrayMT( char name[], int n, data_t arr[] ) -{ - int i; - if (coreid != 0) - return; - - printf( " %10s :", name ); - for ( i = 0; i < n; i++ ) - printf( " %3ld ", (long) arr[i] ); - printf( "\n" ); -} - -void __attribute__((noinline)) verifyMT(size_t n, const data_t* test, const data_t* correct) -{ - if (coreid != 0) - return; - - size_t i; - for (i = 0; i < n; i++) - { - if (test[i] != correct[i]) - { - printf("FAILED test[%d]= %3ld, correct[%d]= %3ld\n", - i, (long)test[i], i, (long)correct[i]); - exit(-1); - } - } - - return; -} - -//-------------------------------------------------------------------------- -// matmul function - -// single-thread, naive version -void __attribute__((noinline)) matmul_naive(const int lda, const data_t A[], const data_t B[], data_t C[] ) -{ - int i, j, k; - - if (coreid > 0) - return; - - for ( i = 0; i < lda; i++ ) - for ( j = 0; j < lda; j++ ) - { - for ( k = 0; k < lda; k++ ) - { - C[i + j*lda] += A[j*lda + k] * B[k*lda + i]; - } - } - -} - - - -void __attribute__((noinline)) matmul(const int lda, const data_t A[], const data_t B[], data_t C[] ) -{ - - // ***************************** // - // **** ADD YOUR CODE HERE ***** // - // ***************************** // - // - // feel free to make a separate function for MI and MSI versions. - - // I think I've got a way for this to not need the "shared" state to work nicely, so no MSI version - int i, j, k, lda_over_2; - lda_over_2 = lda/2; - - if(coreid > 1) - return; - // left side of c - if(coreid == 0) - { - // first half of topleft corner - for(i = 0; i < lda_over_2; i++) { - for(j = 0; j < lda_over_2; j++) { - for(k = 0; k < lda_over_2; k++) { - C[i*lda + j] += A[i*lda + k]*B[k*lda + j]; - } - } - } - // second half of topleft corner - for(i = 0; i < lda_over_2; i++) { - for(j = 0; j < lda_over_2; j++) { - for(k = lda_over_2; k < lda; k++) { - C[i*lda + j] += A[i*lda + k]*B[k*lda + j]; - } - } - } - // second half of bottomleft corner - for(i = lda_over_2; i < lda; i++) { - for(j = 0; j < lda_over_2; j++) { - for(k = lda_over_2; k < lda; k++) { - C[i*lda + j] += A[i*lda + k]*B[k*lda + j]; - } - } - } - // first half of bottomleft corner - for(i = lda_over_2; i < lda; i++) { - for(j = 0; j < lda_over_2; j++) { - for(k = 0; k < lda_over_2; k++) { - C[i*lda + j] += A[i*lda + k]*B[k*lda + j]; - } - } - } - } - else // coreid == 1 - { - // first half of bottomright corner - for(i = lda_over_2; i < lda; i++) { - for(j = lda_over_2; j < lda; j++) { - for(k = 0; k < lda_over_2; k++) { - C[i*lda + j] += A[i*lda + k]*B[k*lda + j]; - } - } - } - // second half of bottomright corner - for(i = lda_over_2; i < lda; i++) { - for(j = lda_over_2; j < lda; j++) { - for(k = lda_over_2; k < lda; k++) { - C[i*lda + j] += A[i*lda + k]*B[k*lda + j]; - } - } - } - // second half of topright corner - for(i = 0; i < lda_over_2; i++) { - for(j = lda_over_2; j < lda; j++) { - for(k = lda_over_2; k < lda; k++) { - C[i*lda + j] += A[i*lda + k]*B[k*lda + j]; - } - } - } - // first half of topright corner - for(i = 0; i < lda_over_2; i++) { - for(j = lda_over_2; j < lda; j++) { - for(k = 0; k < lda_over_2; k++) { - C[i*lda + j] += A[i*lda + k]*B[k*lda + j]; - } - } - } - } - - return; -} - -//-------------------------------------------------------------------------- -// Main -// -// all threads start executing thread_entry(). Use their "coreid" to -// differentiate between threads (each thread is running on a separate core). - -void thread_entry(int cid, int nc) -{ - coreid = cid; - ncores = nc; - - // static allocates data in the binary, which is visible to both threads - static data_t results_data[ARRAY_SIZE]; - - -// // Execute the provided, naive matmul -// barrier(nc); -// stats(matmul_naive(DIM_SIZE, input1_data, input2_data, results_data); barrier(nc)); -// -// -// // verify -// verifyMT(ARRAY_SIZE, results_data, verify_data); -// -// // clear results from the first trial -// size_t i; -// if (coreid == 0) -// for (i=0; i < ARRAY_SIZE; i++) -// results_data[i] = 0; -// barrier(nc); - - - // Execute your faster matmul - barrier(nc); - stats(matmul(DIM_SIZE, input1_data, input2_data, results_data); barrier(nc)); - -#ifdef DEBUG - printArrayMT("results:", ARRAY_SIZE, results_data); - printArrayMT("verify :", ARRAY_SIZE, verify_data); -#endif - - // verify - verifyMT(ARRAY_SIZE, results_data, verify_data); - barrier(nc); - - exit(0); -} - |