[753] | 1 | /* |
---|
| 2 | Author : Shay Gal-On, EEMBC |
---|
| 3 | |
---|
| 4 | This file is part of EEMBC(R) and CoreMark(TM), which are Copyright (C) 2009 |
---|
| 5 | All rights reserved. |
---|
| 6 | |
---|
| 7 | EEMBC CoreMark Software is a product of EEMBC and is provided under the terms of the |
---|
| 8 | CoreMark License that is distributed with the official EEMBC COREMARK Software release. |
---|
| 9 | If you received this EEMBC CoreMark Software without the accompanying CoreMark License, |
---|
| 10 | you must discontinue use and download the official release from www.coremark.org. |
---|
| 11 | |
---|
| 12 | Also, if you are publicly displaying scores generated from the EEMBC CoreMark software, |
---|
| 13 | make sure that you are in compliance with Run and Reporting rules specified in the accompanying readme.txt file. |
---|
| 14 | |
---|
| 15 | EEMBC |
---|
| 16 | 4354 Town Center Blvd. Suite 114-200 |
---|
| 17 | El Dorado Hills, CA, 95762 |
---|
| 18 | */ |
---|
| 19 | #include "coremark.h" |
---|
| 20 | /* |
---|
| 21 | Topic: Description |
---|
| 22 | Matrix manipulation benchmark |
---|
| 23 | |
---|
| 24 | This very simple algorithm forms the basis of many more complex algorithms. |
---|
| 25 | |
---|
| 26 | The tight inner loop is the focus of many optimizations (compiler as well as hardware based) |
---|
| 27 | and is thus relevant for embedded processing. |
---|
| 28 | |
---|
| 29 | The total available data space will be divided to 3 parts: |
---|
| 30 | NxN Matrix A - initialized with small values (upper 3/4 of the bits all zero). |
---|
| 31 | NxN Matrix B - initialized with medium values (upper half of the bits all zero). |
---|
| 32 | NxN Matrix C - used for the result. |
---|
| 33 | |
---|
| 34 | The actual values for A and B must be derived based on input that is not available at compile time. |
---|
| 35 | */ |
---|
| 36 | ee_s16 matrix_test(ee_u32 N, MATRES *C, MATDAT *A, MATDAT *B, MATDAT val); |
---|
| 37 | ee_s16 matrix_sum(ee_u32 N, MATRES *C, MATDAT clipval); |
---|
| 38 | void matrix_mul_const(ee_u32 N, MATRES *C, MATDAT *A, MATDAT val); |
---|
| 39 | void matrix_mul_vect(ee_u32 N, MATRES *C, MATDAT *A, MATDAT *B); |
---|
| 40 | void matrix_mul_matrix(ee_u32 N, MATRES *C, MATDAT *A, MATDAT *B); |
---|
| 41 | void matrix_mul_matrix_bitextract(ee_u32 N, MATRES *C, MATDAT *A, MATDAT *B); |
---|
| 42 | void matrix_add_const(ee_u32 N, MATDAT *A, MATDAT val); |
---|
| 43 | |
---|
| 44 | #define matrix_test_next(x) (x+1) |
---|
| 45 | #define matrix_clip(x,y) ((y) ? (x) & 0x0ff : (x) & 0x0ffff) |
---|
| 46 | #define matrix_big(x) (0xf000 | (x)) |
---|
| 47 | #define bit_extract(x,from,to) (((x)>>(from)) & (~(0xffffffff << (to)))) |
---|
| 48 | |
---|
| 49 | #if CORE_DEBUG |
---|
| 50 | void printmat(MATDAT *A, ee_u32 N, char *name) { |
---|
| 51 | ee_u32 i,j; |
---|
| 52 | ee_printf("Matrix %s [%dx%d]:\n",name,N,N); |
---|
| 53 | for (i=0; i<N; i++) { |
---|
| 54 | for (j=0; j<N; j++) { |
---|
| 55 | if (j!=0) |
---|
| 56 | ee_printf(","); |
---|
| 57 | ee_printf("%d",A[i*N+j]); |
---|
| 58 | } |
---|
| 59 | ee_printf("\n"); |
---|
| 60 | } |
---|
| 61 | } |
---|
| 62 | void printmatC(MATRES *C, ee_u32 N, char *name) { |
---|
| 63 | ee_u32 i,j; |
---|
| 64 | ee_printf("Matrix %s [%dx%d]:\n",name,N,N); |
---|
| 65 | for (i=0; i<N; i++) { |
---|
| 66 | for (j=0; j<N; j++) { |
---|
| 67 | if (j!=0) |
---|
| 68 | ee_printf(","); |
---|
| 69 | ee_printf("%d",C[i*N+j]); |
---|
| 70 | } |
---|
| 71 | ee_printf("\n"); |
---|
| 72 | } |
---|
| 73 | } |
---|
| 74 | #endif |
---|
| 75 | /* Function: core_bench_matrix |
---|
| 76 | Benchmark function |
---|
| 77 | |
---|
| 78 | Iterate <matrix_test> N times, |
---|
| 79 | changing the matrix values slightly by a constant amount each time. |
---|
| 80 | */ |
---|
| 81 | ee_u16 core_bench_matrix(mat_params *p, ee_s16 seed, ee_u16 crc) { |
---|
| 82 | ee_u32 N=p->N; |
---|
| 83 | MATRES *C=p->C; |
---|
| 84 | MATDAT *A=p->A; |
---|
| 85 | MATDAT *B=p->B; |
---|
| 86 | MATDAT val=(MATDAT)seed; |
---|
| 87 | |
---|
| 88 | crc=crc16(matrix_test(N,C,A,B,val),crc); |
---|
| 89 | |
---|
| 90 | return crc; |
---|
| 91 | } |
---|
| 92 | |
---|
| 93 | /* Function: matrix_test |
---|
| 94 | Perform matrix manipulation. |
---|
| 95 | |
---|
| 96 | Parameters: |
---|
| 97 | N - Dimensions of the matrix. |
---|
| 98 | C - memory for result matrix. |
---|
| 99 | A - input matrix |
---|
| 100 | B - operator matrix (not changed during operations) |
---|
| 101 | |
---|
| 102 | Returns: |
---|
| 103 | A CRC value that captures all results calculated in the function. |
---|
| 104 | In particular, crc of the value calculated on the result matrix |
---|
| 105 | after each step by <matrix_sum>. |
---|
| 106 | |
---|
| 107 | Operation: |
---|
| 108 | |
---|
| 109 | 1 - Add a constant value to all elements of a matrix. |
---|
| 110 | 2 - Multiply a matrix by a constant. |
---|
| 111 | 3 - Multiply a matrix by a vector. |
---|
| 112 | 4 - Multiply a matrix by a matrix. |
---|
| 113 | 5 - Add a constant value to all elements of a matrix. |
---|
| 114 | |
---|
| 115 | After the last step, matrix A is back to original contents. |
---|
| 116 | */ |
---|
| 117 | ee_s16 matrix_test(ee_u32 N, MATRES *C, MATDAT *A, MATDAT *B, MATDAT val) { |
---|
| 118 | ee_u16 crc=0; |
---|
| 119 | MATDAT clipval=matrix_big(val); |
---|
| 120 | |
---|
| 121 | matrix_add_const(N,A,val); /* make sure data changes */ |
---|
| 122 | #if CORE_DEBUG |
---|
| 123 | printmat(A,N,"matrix_add_const"); |
---|
| 124 | #endif |
---|
| 125 | matrix_mul_const(N,C,A,val); |
---|
| 126 | crc=crc16(matrix_sum(N,C,clipval),crc); |
---|
| 127 | #if CORE_DEBUG |
---|
| 128 | printmatC(C,N,"matrix_mul_const"); |
---|
| 129 | #endif |
---|
| 130 | matrix_mul_vect(N,C,A,B); |
---|
| 131 | crc=crc16(matrix_sum(N,C,clipval),crc); |
---|
| 132 | #if CORE_DEBUG |
---|
| 133 | printmatC(C,N,"matrix_mul_vect"); |
---|
| 134 | #endif |
---|
| 135 | matrix_mul_matrix(N,C,A,B); |
---|
| 136 | crc=crc16(matrix_sum(N,C,clipval),crc); |
---|
| 137 | #if CORE_DEBUG |
---|
| 138 | printmatC(C,N,"matrix_mul_matrix"); |
---|
| 139 | #endif |
---|
| 140 | matrix_mul_matrix_bitextract(N,C,A,B); |
---|
| 141 | crc=crc16(matrix_sum(N,C,clipval),crc); |
---|
| 142 | #if CORE_DEBUG |
---|
| 143 | printmatC(C,N,"matrix_mul_matrix_bitextract"); |
---|
| 144 | #endif |
---|
| 145 | |
---|
| 146 | matrix_add_const(N,A,-val); /* return matrix to initial value */ |
---|
| 147 | return crc; |
---|
| 148 | } |
---|
| 149 | |
---|
| 150 | /* Function : matrix_init |
---|
| 151 | Initialize the memory block for matrix benchmarking. |
---|
| 152 | |
---|
| 153 | Parameters: |
---|
| 154 | blksize - Size of memory to be initialized. |
---|
| 155 | memblk - Pointer to memory block. |
---|
| 156 | seed - Actual values chosen depend on the seed parameter. |
---|
| 157 | p - pointers to <mat_params> containing initialized matrixes. |
---|
| 158 | |
---|
| 159 | Returns: |
---|
| 160 | Matrix dimensions. |
---|
| 161 | |
---|
| 162 | Note: |
---|
| 163 | The seed parameter MUST be supplied from a source that cannot be determined at compile time |
---|
| 164 | */ |
---|
| 165 | ee_u32 core_init_matrix(ee_u32 blksize, void *memblk, ee_s32 seed, mat_params *p) { |
---|
| 166 | ee_u32 N=0; |
---|
| 167 | MATDAT *A; |
---|
| 168 | MATDAT *B; |
---|
| 169 | ee_s32 order=1; |
---|
| 170 | MATDAT val; |
---|
| 171 | ee_u32 i=0,j=0; |
---|
| 172 | if (seed==0) |
---|
| 173 | seed=1; |
---|
| 174 | while (j<blksize) { |
---|
| 175 | i++; |
---|
| 176 | j=i*i*2*4; |
---|
| 177 | } |
---|
| 178 | N=i-1; |
---|
| 179 | A=(MATDAT *)align_mem(memblk); |
---|
| 180 | B=A+N*N; |
---|
| 181 | |
---|
| 182 | for (i=0; i<N; i++) { |
---|
| 183 | for (j=0; j<N; j++) { |
---|
| 184 | seed = ( ( order * seed ) % 65536 ); |
---|
| 185 | val = (seed + order); |
---|
| 186 | val=matrix_clip(val,0); |
---|
| 187 | B[i*N+j] = val; |
---|
| 188 | val = (val + order); |
---|
| 189 | val=matrix_clip(val,1); |
---|
| 190 | A[i*N+j] = val; |
---|
| 191 | order++; |
---|
| 192 | } |
---|
| 193 | } |
---|
| 194 | |
---|
| 195 | p->A=A; |
---|
| 196 | p->B=B; |
---|
| 197 | p->C=(MATRES *)align_mem(B+N*N); |
---|
| 198 | p->N=N; |
---|
| 199 | #if CORE_DEBUG |
---|
| 200 | printmat(A,N,"A"); |
---|
| 201 | printmat(B,N,"B"); |
---|
| 202 | #endif |
---|
| 203 | return N; |
---|
| 204 | } |
---|
| 205 | |
---|
| 206 | /* Function: matrix_sum |
---|
| 207 | Calculate a function that depends on the values of elements in the matrix. |
---|
| 208 | |
---|
| 209 | For each element, accumulate into a temporary variable. |
---|
| 210 | |
---|
| 211 | As long as this value is under the parameter clipval, |
---|
| 212 | add 1 to the result if the element is bigger then the previous. |
---|
| 213 | |
---|
| 214 | Otherwise, reset the accumulator and add 10 to the result. |
---|
| 215 | */ |
---|
| 216 | ee_s16 matrix_sum(ee_u32 N, MATRES *C, MATDAT clipval) { |
---|
| 217 | MATRES tmp=0,prev=0,cur=0; |
---|
| 218 | ee_s16 ret=0; |
---|
| 219 | ee_u32 i,j; |
---|
| 220 | for (i=0; i<N; i++) { |
---|
| 221 | for (j=0; j<N; j++) { |
---|
| 222 | cur=C[i*N+j]; |
---|
| 223 | tmp+=cur; |
---|
| 224 | if (tmp>clipval) { |
---|
| 225 | ret+=10; |
---|
| 226 | tmp=0; |
---|
| 227 | } else { |
---|
| 228 | ret += (cur>prev) ? 1 : 0; |
---|
| 229 | } |
---|
| 230 | prev=cur; |
---|
| 231 | } |
---|
| 232 | } |
---|
| 233 | return ret; |
---|
| 234 | } |
---|
| 235 | |
---|
| 236 | /* Function: matrix_mul_const |
---|
| 237 | Multiply a matrix by a constant. |
---|
| 238 | This could be used as a scaler for instance. |
---|
| 239 | */ |
---|
| 240 | void matrix_mul_const(ee_u32 N, MATRES *C, MATDAT *A, MATDAT val) { |
---|
| 241 | ee_u32 i,j; |
---|
| 242 | for (i=0; i<N; i++) { |
---|
| 243 | for (j=0; j<N; j++) { |
---|
| 244 | C[i*N+j]=(MATRES)A[i*N+j] * (MATRES)val; |
---|
| 245 | } |
---|
| 246 | } |
---|
| 247 | } |
---|
| 248 | |
---|
| 249 | /* Function: matrix_add_const |
---|
| 250 | Add a constant value to all elements of a matrix. |
---|
| 251 | */ |
---|
| 252 | void matrix_add_const(ee_u32 N, MATDAT *A, MATDAT val) { |
---|
| 253 | ee_u32 i,j; |
---|
| 254 | for (i=0; i<N; i++) { |
---|
| 255 | for (j=0; j<N; j++) { |
---|
| 256 | A[i*N+j] += val; |
---|
| 257 | } |
---|
| 258 | } |
---|
| 259 | } |
---|
| 260 | |
---|
| 261 | /* Function: matrix_mul_vect |
---|
| 262 | Multiply a matrix by a vector. |
---|
| 263 | This is common in many simple filters (e.g. fir where a vector of coefficients is applied to the matrix.) |
---|
| 264 | */ |
---|
| 265 | void matrix_mul_vect(ee_u32 N, MATRES *C, MATDAT *A, MATDAT *B) { |
---|
| 266 | ee_u32 i,j; |
---|
| 267 | for (i=0; i<N; i++) { |
---|
| 268 | C[i]=0; |
---|
| 269 | for (j=0; j<N; j++) { |
---|
| 270 | C[i]+=(MATRES)A[i*N+j] * (MATRES)B[j]; |
---|
| 271 | } |
---|
| 272 | } |
---|
| 273 | } |
---|
| 274 | |
---|
| 275 | /* Function: matrix_mul_matrix |
---|
| 276 | Multiply a matrix by a matrix. |
---|
| 277 | Basic code is used in many algorithms, mostly with minor changes such as scaling. |
---|
| 278 | */ |
---|
| 279 | void matrix_mul_matrix(ee_u32 N, MATRES *C, MATDAT *A, MATDAT *B) { |
---|
| 280 | ee_u32 i,j,k; |
---|
| 281 | for (i=0; i<N; i++) { |
---|
| 282 | for (j=0; j<N; j++) { |
---|
| 283 | C[i*N+j]=0; |
---|
| 284 | for(k=0;k<N;k++) |
---|
| 285 | { |
---|
| 286 | C[i*N+j]+=(MATRES)A[i*N+k] * (MATRES)B[k*N+j]; |
---|
| 287 | } |
---|
| 288 | } |
---|
| 289 | } |
---|
| 290 | } |
---|
| 291 | |
---|
| 292 | /* Function: matrix_mul_matrix_bitextract |
---|
| 293 | Multiply a matrix by a matrix, and extract some bits from the result. |
---|
| 294 | Basic code is used in many algorithms, mostly with minor changes such as scaling. |
---|
| 295 | */ |
---|
| 296 | void matrix_mul_matrix_bitextract(ee_u32 N, MATRES *C, MATDAT *A, MATDAT *B) { |
---|
| 297 | ee_u32 i,j,k; |
---|
| 298 | for (i=0; i<N; i++) { |
---|
| 299 | for (j=0; j<N; j++) { |
---|
| 300 | C[i*N+j]=0; |
---|
| 301 | for(k=0;k<N;k++) |
---|
| 302 | { |
---|
| 303 | MATRES tmp=(MATRES)A[i*N+k] * (MATRES)B[k*N+j]; |
---|
| 304 | C[i*N+j]+=bit_extract(tmp,2,4)*bit_extract(tmp,5,7); |
---|
| 305 | } |
---|
| 306 | } |
---|
| 307 | } |
---|
| 308 | } |
---|