//************************************************************************** // 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 #include #include //-------------------------------------------------------------------------- // 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. int i, j, k; /*547287 for ( i = coreid*lda/ncores; i < (coreid+1)*lda/ncores; i++ ) { for ( j = 0; j < lda; j++ ) { int aIndex = j*lda; int cIndex = i + aIndex; C[cIndex] += A[aIndex] * B[i]; C[cIndex] += A[aIndex + 1] * B[1*lda + i]; C[cIndex] += A[aIndex + 2] * B[2*lda + i]; C[cIndex] += A[aIndex + 3] * B[3*lda + i]; C[cIndex] += A[aIndex + 4] * B[4*lda + i]; C[cIndex] += A[aIndex + 5] * B[5*lda + i]; C[cIndex] += A[aIndex + 6] * B[6*lda + i]; C[cIndex] += A[aIndex + 7] * B[7*lda + i]; C[cIndex] += A[aIndex + 8] * B[8*lda + i]; C[cIndex] += A[aIndex + 9] * B[9*lda + i]; C[cIndex] += A[aIndex + 10] * B[10*lda + i]; C[cIndex] += A[aIndex + 11] * B[11*lda + i]; C[cIndex] += A[aIndex + 12] * B[12*lda + i]; C[cIndex] += A[aIndex + 13] * B[13*lda + i]; C[cIndex] += A[aIndex + 14] * B[14*lda + i]; C[cIndex] += A[aIndex + 15] * B[15*lda + i]; C[cIndex] += A[aIndex + 16] * B[16*lda + i]; C[cIndex] += A[aIndex + 17] * B[17*lda + i]; C[cIndex] += A[aIndex + 18] * B[18*lda + i]; C[cIndex] += A[aIndex + 19] * B[19*lda + i]; C[cIndex] += A[aIndex + 20] * B[20*lda + i]; C[cIndex] += A[aIndex + 21] * B[21*lda + i]; C[cIndex] += A[aIndex + 22] * B[22*lda + i]; C[cIndex] += A[aIndex + 23] * B[23*lda + i]; C[cIndex] += A[aIndex + 24] * B[24*lda + i]; C[cIndex] += A[aIndex + 25] * B[25*lda + i]; C[cIndex] += A[aIndex + 26] * B[26*lda + i]; C[cIndex] += A[aIndex + 27] * B[27*lda + i]; C[cIndex] += A[aIndex + 28] * B[28*lda + i]; C[cIndex] += A[aIndex + 29] * B[29*lda + i]; C[cIndex] += A[aIndex + 30] * B[30*lda + i]; C[cIndex] += A[aIndex + 31] * B[31*lda + i]; } } */ //492827 /* for ( i = coreid*lda/ncores; i < (coreid+1)*lda/ncores; i++ ) { for ( j = 0; j < lda; j++ ) { int aIndex = j*lda; int cIndex = i + aIndex; for ( k = 0; k < lda; k++) { C[cIndex] += A[aIndex + k] * B[k*lda + i]; /* C[cIndex] += A[aIndex + k+1] * B[(k+1)*lda + i]; C[cIndex] += A[aIndex + k+2] * B[(k+2)*lda + i]; C[cIndex] += A[aIndex + k+3] * B[(k+3)*lda + i]; C[cIndex] += A[aIndex + k+4] * B[(k+4)*lda + i]; C[cIndex] += A[aIndex + k+5] * B[(k+5)*lda + i]; C[cIndex] += A[aIndex + k+6] * B[(k+6)*lda + i]; C[cIndex] += A[aIndex + k+7] * B[(k+7)*lda + i]; C[cIndex] += A[aIndex + k+8] * B[(k+8)*lda + i]; C[cIndex] += A[aIndex + k+9] * B[(k+9)*lda + i]; C[cIndex] += A[aIndex + k+10] * B[(k+10)*lda + i]; C[cIndex] += A[aIndex + k+11] * B[(k+11)*lda + i]; C[cIndex] += A[aIndex + k+12] * B[(k+12)*lda + i]; C[cIndex] += A[aIndex + k+13] * B[(k+13)*lda + i]; C[cIndex] += A[aIndex + k+14] * B[(k+14)*lda + i]; C[cIndex] += A[aIndex + k+15] * B[(k+15)*lda + i];*/ /* } } }*/ /* //326378 data_t bTrans[1024]; for (int counti = 0; counti < 32; counti++) { for (int countj = 0; countj < 32; countj++) { *(bTrans + counti + countj*lda) = *(B + countj + counti*lda); } } int BLOCKSIZE = 8; for ( i = 0; i < lda; i+=BLOCKSIZE ) { for ( int iTemp = i; iTemp < i + BLOCKSIZE; iTemp++ ) { int iFlag = iTemp*lda; for ( j = coreid*lda/ncores; j < (coreid+1)*lda/ncores; j++ ) { int jFlag = j*lda; int cLoc = jFlag+iTemp; for ( k = 0; k < lda; k+=8) { *(C+cLoc) += *(A+jFlag+k) * *(bTrans+iFlag+k); *(C+cLoc) += *(A+jFlag+k+1) * *(bTrans+iFlag+k+1); *(C+cLoc) += *(A+jFlag+k+2) * *(bTrans+iFlag+k+2); *(C+cLoc) += *(A+jFlag+k+3) * *(bTrans+iFlag+k+3); *(C+cLoc) += *(A+jFlag+k+4) * *(bTrans+iFlag+k+4); *(C+cLoc) += *(A+jFlag+k+5) * *(bTrans+iFlag+k+5); *(C+cLoc) += *(A+jFlag+k+6) * *(bTrans+iFlag+k+6); *(C+cLoc) += *(A+jFlag+k+7) * *(bTrans+iFlag+k+7); } } } }*/ data_t bTrans[1024]; for (int counti = 0; counti < 32; counti++) { for (int countj = 0; countj < 32; countj++) { *(bTrans + counti + countj*lda) = *(B + countj + counti*lda); } } int BLOCKSIZE = 8; for ( j = 0; j < lda; j++ ) { //for ( int jTemp = j; jTemp < j + BLOCKSIZE; jTemp++ ) { int jFlag = j*lda; for ( i = coreid*lda/ncores; i < (coreid+1)*lda/ncores; i+=BLOCKSIZE ) { for ( int iTemp = i; iTemp < i + BLOCKSIZE; iTemp++ ) { int iFlag = iTemp*lda; int cLoc = jFlag+iTemp; for ( k = 0; k < lda; k+=16) { *(C+cLoc) += *(A+jFlag+k) * *(bTrans+iFlag+k); *(C+cLoc) += *(A+jFlag+k+1) * *(bTrans+iFlag+k+1); *(C+cLoc) += *(A+jFlag+k+2) * *(bTrans+iFlag+k+2); *(C+cLoc) += *(A+jFlag+k+3) * *(bTrans+iFlag+k+3); *(C+cLoc) += *(A+jFlag+k+4) * *(bTrans+iFlag+k+4); *(C+cLoc) += *(A+jFlag+k+5) * *(bTrans+iFlag+k+5); *(C+cLoc) += *(A+jFlag+k+6) * *(bTrans+iFlag+k+6); *(C+cLoc) += *(A+jFlag+k+7) * *(bTrans+iFlag+k+7); *(C+cLoc) += *(A+jFlag+k+8) * *(bTrans+iFlag+k+8); *(C+cLoc) += *(A+jFlag+k+9) * *(bTrans+iFlag+k+9); *(C+cLoc) += *(A+jFlag+k+10) * *(bTrans+iFlag+k+10); *(C+cLoc) += *(A+jFlag+k+11) * *(bTrans+iFlag+k+11); *(C+cLoc) += *(A+jFlag+k+12) * *(bTrans+iFlag+k+12); *(C+cLoc) += *(A+jFlag+k+13) * *(bTrans+iFlag+k+13); *(C+cLoc) += *(A+jFlag+k+14) * *(bTrans+iFlag+k+14); *(C+cLoc) += *(A+jFlag+k+15) * *(bTrans+iFlag+k+15); } } } //} } } //-------------------------------------------------------------------------- // 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); }