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

Revert "Rewrite &= -> = and simplify the initial blocking phase."

This commit is contained in:
Martin Kroeker 2018-07-03 21:42:28 +02:00 committed by GitHub
parent a83f01e0ee
commit 5f2a3c05cd
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 14 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
driver/level3/level3_thread.c
View File

@ -344,6 +344,12 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
div_n = (n_to - n_from + DIVIDE_RATE - 1) / DIVIDE_RATE; div_n = (n_to - n_from + DIVIDE_RATE - 1) / DIVIDE_RATE;
for (js = n_from, bufferside = 0; js < n_to; js += div_n, bufferside ++) { for (js = n_from, bufferside = 0; js < n_to; js += div_n, bufferside ++) {
/* Make sure if no one is using workspace */
START_RPCC();
for (i = 0; i < args -> nthreads; i++)
while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;MB;};
STOP_RPCC(waiting1);
#if defined(FUSED_GEMM) && !defined(TIMING) #if defined(FUSED_GEMM) && !defined(TIMING)
/* Fused operation to copy region of B into workspace and apply kernel */ /* Fused operation to copy region of B into workspace and apply kernel */
@ -381,16 +387,11 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
} }
#endif #endif
for (i = mypos_n * nthreads_m; i < (mypos_n + 1) * nthreads_m; i++) {
/* Make sure if no one is using workspace */
START_RPCC();
while (job[mypos].working[i][CACHE_LINE_SIZE * bufferside]) {YIELDING;MB;};
STOP_RPCC(waiting1);
/* Set flag so other threads can access local region of B */ /* Set flag so other threads can access local region of B */
for (i = mypos_n * nthreads_m; i < (mypos_n + 1) * nthreads_m; i++)
job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside]; job[mypos].working[i][CACHE_LINE_SIZE * bufferside] = (BLASLONG)buffer[bufferside];
WMB; WMB;
} }
}
/* Get regions of B from other threads and apply kernel */ /* Get regions of B from other threads and apply kernel */
current = mypos; current = mypos;
@ -425,7 +426,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
/* Clear synchronization flag if this thread is done with other region of B */ /* Clear synchronization flag if this thread is done with other region of B */
if (m_to - m_from == min_i) { if (m_to - m_from == min_i) {
job[current].working[mypos][CACHE_LINE_SIZE * bufferside] = 0; job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
WMB; WMB;
} }
} }
@ -468,7 +469,7 @@ static int inner_thread(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n,
/* Clear synchronization flag if this thread is done with region of B */ /* Clear synchronization flag if this thread is done with region of B */
if (is + min_i >= m_to) { if (is + min_i >= m_to) {
job[current].working[mypos][CACHE_LINE_SIZE * bufferside] = 0; job[current].working[mypos][CACHE_LINE_SIZE * bufferside] &= 0;
WMB; WMB;
} }
} }