diff options
Diffstat (limited to 'mt/br_matmul/matmul_mi.c')
-rwxr-xr-x | mt/br_matmul/matmul_mi.c | 283 |
1 files changed, 283 insertions, 0 deletions
diff --git a/mt/br_matmul/matmul_mi.c b/mt/br_matmul/matmul_mi.c new file mode 100755 index 0000000..5ca1dbe --- /dev/null +++ b/mt/br_matmul/matmul_mi.c @@ -0,0 +1,283 @@ +//************************************************************************** +// Multi-threaded Matrix Multiply benchmark +//-------------------------------------------------------------------------- +// TA : Christopher Celio +// Student: Benjamin Han +// +// +// 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 printArray( 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)) verify(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 j2, i2, k2, j, i, k; + int tmpC00, tmpC01, tmpC02, tmpC03, tmpC04, tmpC05, tmpC06, tmpC07; + int tmpC10, tmpC11, tmpC12, tmpC13, tmpC14, tmpC15, tmpC16, tmpC17; + int jBLOCK = 32; + int iBLOCK = 16; + int kBLOCK = 32; + static __thread int tB[4096]; //__thread + int startInd = 0; + int endInd = lda >> 1; + if (coreid == 1) { + startInd = lda >> 1; + endInd = lda; + } + + //tranpose B (block?) + for (i = 0; i < lda; i += 2) { + for (j = startInd; j < endInd; j += 2) { + tB[j*lda + i] = B[i*lda + j]; + tB[(j + 1)*lda + i] = B[i*lda + j + 1]; + tB[j*lda + i + 1] = B[(i + 1)*lda + j]; + tB[(j + 1)*lda + i + 1] = B[(i + 1)*lda + j + 1]; + } + } + barrier(); + + // compute C[j*n + i] += A[j*n + k] + Btranspose[i*n + k] + for ( j2 = 0; j2 < lda; j2 += jBLOCK ) + for ( i2 = startInd; i2 < endInd; i2 += iBLOCK ) + for ( j = j2; j < j2 + jBLOCK; j += 2 ) + for ( k2 = 0; k2 < lda; k2 += kBLOCK ) + for ( i = i2; i < i2 + iBLOCK; i += 4) { + tmpC00 = C[j*lda + i + 0]; tmpC10 = C[(j + 1)*lda + i + 0]; + tmpC01 = C[j*lda + i + 1]; tmpC11 = C[(j + 1)*lda + i + 1]; + tmpC02 = C[j*lda + i + 2]; tmpC12 = C[(j + 1)*lda + i + 2]; + tmpC03 = C[j*lda + i + 3]; tmpC13 = C[(j + 1)*lda + i + 3]; + //tmpC04 = C[j*lda + i + 4]; tmpC14 = C[(j + 1)*lda + i + 4]; + //tmpC05 = C[j*lda + i + 5]; tmpC15 = C[(j + 1)*lda + i + 5]; + //tmpC06 = C[j*lda + i + 6]; tmpC16 = C[(j + 1)*lda + i + 6]; + //tmpC07 = C[j*lda + i + 7]; tmpC17 = C[(j + 1)*lda + i + 7]; + for ( k = k2; k < k2 + kBLOCK; k += 4) { + tmpC00 += A[j*lda + k] * tB[(i + 0)*lda + k]; + tmpC01 += A[j*lda + k] * tB[(i + 1)*lda + k]; + tmpC02 += A[j*lda + k] * tB[(i + 2)*lda + k]; + tmpC03 += A[j*lda + k] * tB[(i + 3)*lda + k]; + //tmpC04 += A[j*lda + k] * tB[(i + 4)*lda + k]; + //tmpC05 += A[j*lda + k] * tB[(i + 5)*lda + k]; + //tmpC06 += A[j*lda + k] * tB[(i + 6)*lda + k]; + //tmpC07 += A[j*lda + k] * tB[(i + 7)*lda + k]; + tmpC10 += A[(j + 1)*lda + k] * tB[(i + 0)*lda + k]; + tmpC11 += A[(j + 1)*lda + k] * tB[(i + 1)*lda + k]; + tmpC12 += A[(j + 1)*lda + k] * tB[(i + 2)*lda + k]; + tmpC13 += A[(j + 1)*lda + k] * tB[(i + 3)*lda + k]; + //tmpC14 += A[(j + 1)*lda + k] * tB[(i + 4)*lda + k]; + //tmpC15 += A[(j + 1)*lda + k] * tB[(i + 5)*lda + k]; + //tmpC16 += A[(j + 1)*lda + k] * tB[(i + 6)*lda + k]; + //tmpC17 += A[(j + 1)*lda + k] * tB[(i + 7)*lda + k]; + + tmpC00 += A[j*lda + k + 1] * tB[(i + 0)*lda + k + 1]; + tmpC01 += A[j*lda + k + 1] * tB[(i + 1)*lda + k + 1]; + tmpC02 += A[j*lda + k + 1] * tB[(i + 2)*lda + k + 1]; + tmpC03 += A[j*lda + k + 1] * tB[(i + 3)*lda + k + 1]; + //tmpC04 += A[j*lda + k + 1] * tB[(i + 4)*lda + k + 1]; + //tmpC05 += A[j*lda + k + 1] * tB[(i + 5)*lda + k + 1]; + //tmpC06 += A[j*lda + k + 1] * tB[(i + 6)*lda + k + 1]; + //tmpC07 += A[j*lda + k + 1] * tB[(i + 7)*lda + k + 1]; + tmpC10 += A[(j + 1)*lda + k + 1] * tB[(i + 0)*lda + k + 1]; + tmpC11 += A[(j + 1)*lda + k + 1] * tB[(i + 1)*lda + k + 1]; + tmpC12 += A[(j + 1)*lda + k + 1] * tB[(i + 2)*lda + k + 1]; + tmpC13 += A[(j + 1)*lda + k + 1] * tB[(i + 3)*lda + k + 1]; + //tmpC14 += A[(j + 1)*lda + k + 1] * tB[(i + 4)*lda + k + 1]; + //tmpC15 += A[(j + 1)*lda + k + 1] * tB[(i + 5)*lda + k + 1]; + //tmpC16 += A[(j + 1)*lda + k + 1] * tB[(i + 6)*lda + k + 1]; + //tmpC17 += A[(j + 1)*lda + k + 1] * tB[(i + 7)*lda + k + 1]; + + tmpC00 += A[j*lda + k + 2] * tB[(i + 0)*lda + k + 2]; + tmpC01 += A[j*lda + k + 2] * tB[(i + 1)*lda + k + 2]; + tmpC02 += A[j*lda + k + 2] * tB[(i + 2)*lda + k + 2]; + tmpC03 += A[j*lda + k + 2] * tB[(i + 3)*lda + k + 2]; + //tmpC04 += A[j*lda + k + 2] * tB[(i + 4)*lda + k + 2]; + //tmpC05 += A[j*lda + k + 2] * tB[(i + 5)*lda + k + 2]; + //tmpC06 += A[j*lda + k + 2] * tB[(i + 6)*lda + k + 2]; + //tmpC07 += A[j*lda + k + 2] * tB[(i + 7)*lda + k + 2]; + tmpC10 += A[(j + 1)*lda + k + 2] * tB[(i + 0)*lda + k + 2]; + tmpC11 += A[(j + 1)*lda + k + 2] * tB[(i + 1)*lda + k + 2]; + tmpC12 += A[(j + 1)*lda + k + 2] * tB[(i + 2)*lda + k + 2]; + tmpC13 += A[(j + 1)*lda + k + 2] * tB[(i + 3)*lda + k + 2]; + //tmpC14 += A[(j + 1)*lda + k + 2] * tB[(i + 4)*lda + k + 2]; + //tmpC15 += A[(j + 1)*lda + k + 2] * tB[(i + 5)*lda + k + 2]; + //tmpC16 += A[(j + 1)*lda + k + 2] * tB[(i + 6)*lda + k + 2]; + //tmpC17 += A[(j + 1)*lda + k + 2] * tB[(i + 7)*lda + k + 2]; + + tmpC00 += A[j*lda + k + 3] * tB[(i + 0)*lda + k + 3]; + tmpC01 += A[j*lda + k + 3] * tB[(i + 1)*lda + k + 3]; + tmpC02 += A[j*lda + k + 3] * tB[(i + 2)*lda + k + 3]; + tmpC03 += A[j*lda + k + 3] * tB[(i + 3)*lda + k + 3]; + //tmpC04 += A[j*lda + k + 3] * tB[(i + 4)*lda + k + 3]; + //tmpC05 += A[j*lda + k + 3] * tB[(i + 5)*lda + k + 3]; + //tmpC06 += A[j*lda + k + 3] * tB[(i + 6)*lda + k + 3]; + //tmpC07 += A[j*lda + k + 3] * tB[(i + 7)*lda + k + 3]; + tmpC10 += A[(j + 1)*lda + k + 3] * tB[(i + 0)*lda + k + 3]; + tmpC11 += A[(j + 1)*lda + k + 3] * tB[(i + 1)*lda + k + 3]; + tmpC12 += A[(j + 1)*lda + k + 3] * tB[(i + 2)*lda + k + 3]; + tmpC13 += A[(j + 1)*lda + k + 3] * tB[(i + 3)*lda + k + 3]; + //tmpC14 += A[(j + 1)*lda + k + 3] * tB[(i + 4)*lda + k + 3]; + //tmpC15 += A[(j + 1)*lda + k + 3] * tB[(i + 5)*lda + k + 3]; + //tmpC16 += A[(j + 1)*lda + k + 3] * tB[(i + 6)*lda + k + 3]; + //tmpC17 += A[(j + 1)*lda + k + 3] * tB[(i + 7)*lda + k + 3]; + } + C[j*lda + i + 0] = tmpC00; C[(j + 1)*lda + i + 0] = tmpC10; + C[j*lda + i + 1] = tmpC01; C[(j + 1)*lda + i + 1] = tmpC11; + C[j*lda + i + 2] = tmpC02; C[(j + 1)*lda + i + 2] = tmpC12; + C[j*lda + i + 3] = tmpC03; C[(j + 1)*lda + i + 3] = tmpC13; + //C[j*lda + i + 4] = tmpC04; C[(j + 1)*lda + i + 4] = tmpC14; + //C[j*lda + i + 5] = tmpC05; C[(j + 1)*lda + i + 5] = tmpC15; + //C[j*lda + i + 6] = tmpC06; C[(j + 1)*lda + i + 6] = tmpC16; + //C[j*lda + i + 7] = tmpC07; C[(j + 1)*lda + i + 7] = tmpC17; + } +} + +//-------------------------------------------------------------------------- +// 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(); +// stats(matmul_naive(DIM_SIZE, input1_data, input2_data, results_data); barrier()); +// +// +// // verify +// verify(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(); + + + // Execute your faster matmul + barrier(); + stats(matmul(DIM_SIZE, input1_data, input2_data, results_data); barrier()); + +#ifdef DEBUG + printArray("results:", ARRAY_SIZE, results_data); + printArray("verify :", ARRAY_SIZE, verify_data); +#endif + + // verify + verify(ARRAY_SIZE, results_data, verify_data); + barrier(); + + exit(0); +} + |