--- /dev/null
+#include "stdlib.h"
+
+#include "util.h"
+
+#include "dataset.h"\r
+void __attribute__((noinline)) matmul(const int coreid, const int ncores, const int lda, const data_t A[], const data_t B[], data_t C[] )\r
+{\r
+ \r
+ // ***************************** //\r
+ // **** ADD YOUR CODE HERE ***** //\r
+ // ***************************** //\r
+ //\r
+ // feel free to make a separate function for MI and MSI versions.\r
+ \r
+ int m, i, j, k, iB0, iB1;\r
+ data_t tempC0, tempC1, tempC2, tempC3, tempC4, tempC5, tempC6, tempC7;\r
+ data_t tempA0, tempA1;\r
+ \r
+ if(coreid > 1) return;
+ if (coreid == 0){\r
+ for (m = 0; m < 2; m++){\r
+ for (j = 0; j < lda/2; j++){\r
+ for (i = 0; i < lda; i+=8){\r
+ tempC0 = C[i + j*lda];\r
+ tempC1 = C[i + j*lda+1];\r
+ tempC2 = C[i + j*lda+2];\r
+ tempC3 = C[i + j*lda+3];\r
+ tempC4 = C[i + j*lda+4];\r
+ tempC5 = C[i + j*lda+5];\r
+ tempC6 = C[i + j*lda+6];\r
+ tempC7 = C[i + j*lda+7];\r
+ iB0 = m*lda*lda/2+i;\r
+ iB1 = iB0+lda;\r
+ for (k = m*lda/2; k < (m+1)*lda/2; k+=2){\r
+ tempA0 = A[j*lda+k];\r
+ tempA1 = A[j*lda+k+1];\r
+ tempC0 += tempA0*B[iB0]+tempA1*B[iB1];\r
+ tempC1 += tempA0*B[iB0+1]+tempA1*B[iB1+1];\r
+ tempC2 += tempA0*B[iB0+2]+tempA1*B[iB1+2];\r
+ tempC3 += tempA0*B[iB0+3]+tempA1*B[iB1+3];\r
+ tempC4 += tempA0*B[iB0+4]+tempA1*B[iB1+4];\r
+ tempC5 += tempA0*B[iB0+5]+tempA1*B[iB1+5];\r
+ tempC6 += tempA0*B[iB0+6]+tempA1*B[iB1+6];\r
+ tempC7 += tempA0*B[iB0+7]+tempA1*B[iB1+7];\r
+ iB0 += 2*lda;\r
+ iB1 += 2*lda;\r
+ \r
+ }\r
+ C[i + j*lda] = tempC0;\r
+ C[i + j*lda + 1] = tempC1;\r
+ C[i + j*lda + 2] = tempC2;\r
+ C[i + j*lda + 3] = tempC3;\r
+ C[i + j*lda + 4] = tempC4;\r
+ C[i + j*lda + 5] = tempC5;\r
+ C[i + j*lda + 6] = tempC6;\r
+ C[i + j*lda + 7] = tempC7;\r
+ }\r
+ }\r
+ }\r
+ }
+ if(coreid == 1 || ncores == 1) {\r
+ for (m = 2; m > 0; m--){\r
+ for (j = lda-1; j >= lda/2; j--){\r
+ for (i = lda-1; i >= 0; i-=8){\r
+ tempC0 = C[i + j*lda];\r
+ tempC1 = C[i + j*lda - 1];\r
+ tempC2 = C[i + j*lda - 2];\r
+ tempC3 = C[i + j*lda - 3];\r
+ tempC4 = C[i + j*lda - 4];\r
+ tempC5 = C[i + j*lda - 5];\r
+ tempC6 = C[i + j*lda - 6];\r
+ tempC7 = C[i + j*lda - 7];\r
+ for (k = m*lda/2-1; k >= (m-1)*lda/2; k-=2){\r
+ tempA0 = A[j*lda+k];\r
+ tempA1 = A[j*lda+k-1];\r
+ tempC0 += tempA0*B[k*lda+i]+tempA1*B[(k-1)*lda+i];\r
+ tempC1 += tempA0*B[k*lda+i-1]+tempA1*B[(k-1)*lda+i-1];\r
+ tempC2 += tempA0*B[k*lda+i-2]+tempA1*B[(k-1)*lda+i-2];\r
+ tempC3 += tempA0*B[k*lda+i-3]+tempA1*B[(k-1)*lda+i-3];\r
+ tempC4 += tempA0*B[k*lda+i-4]+tempA1*B[(k-1)*lda+i-4];\r
+ tempC5 += tempA0*B[k*lda+i-5]+tempA1*B[(k-1)*lda+i-5];\r
+ tempC6 += tempA0*B[k*lda+i-6]+tempA1*B[(k-1)*lda+i-6];\r
+ tempC7 += tempA0*B[k*lda+i-7]+tempA1*B[(k-1)*lda+i-7];\r
+ }\r
+ C[i + j*lda] = tempC0;\r
+ C[i + j*lda - 1] = tempC1;\r
+ C[i + j*lda - 2] = tempC2;\r
+ C[i + j*lda - 3] = tempC3;\r
+ C[i + j*lda - 4] = tempC4;\r
+ C[i + j*lda - 5] = tempC5;\r
+ C[i + j*lda - 6] = tempC6;\r
+ C[i + j*lda - 7] = tempC7;\r
+ }\r
+ }\r
+ }\r
+ }\r
+}\r