floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Calculation of range limits was ignoring num_cpu 

bug introduced by me in #1262
This commit is contained in:
Martin Kroeker 2017-12-09 18:47:24 +01:00 committed by GitHub
parent e90a147493
commit 7777580501
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
driver/level2/sbmv_thread.c
View File

@ -246,7 +246,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width; range_m[MAX_CPU_NUMBER - num_cpu - 1] = range_m[MAX_CPU_NUMBER - num_cpu] - width;
range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16); range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
if (range_n[num_cpu] > n) range_n[num_cpu] = n; if (range_n[num_cpu] > n * num_cpu) range_n[num_cpu] = n * num_cpu;
queue[num_cpu].mode = mode; queue[num_cpu].mode = mode;
queue[num_cpu].routine = sbmv_kernel; queue[num_cpu].routine = sbmv_kernel;
@ -286,7 +286,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
range_m[num_cpu + 1] = range_m[num_cpu] + width; range_m[num_cpu + 1] = range_m[num_cpu] + width;
range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16); range_n[num_cpu] = num_cpu * (((n + 15) & ~15) + 16);
if (range_n[num_cpu] > n) range_n[num_cpu] = n; if (range_n[num_cpu] > n * num_cpu) range_n[num_cpu] = n * num_cpu;
queue[num_cpu].mode = mode; queue[num_cpu].mode = mode;
queue[num_cpu].routine = sbmv_kernel; queue[num_cpu].routine = sbmv_kernel;
@ -318,7 +318,7 @@ int CNAME(BLASLONG n, BLASLONG k, FLOAT *alpha, FLOAT *a, BLASLONG lda, FLOAT *x
range_m[num_cpu + 1] = range_m[num_cpu] + width; range_m[num_cpu + 1] = range_m[num_cpu] + width;
range_n[num_cpu] = num_cpu * ((n + 15) & ~15); range_n[num_cpu] = num_cpu * ((n + 15) & ~15);
if (range_n[num_cpu] > n) range_n[num_cpu] = n; if (range_n[num_cpu] > n * num_cpu) range_n[num_cpu] = n * num_cpu;
queue[num_cpu].mode = mode; queue[num_cpu].mode = mode;
queue[num_cpu].routine = sbmv_kernel; queue[num_cpu].routine = sbmv_kernel;