Index: openmp/trunk/runtime/src/kmp_affinity.h =================================================================== --- openmp/trunk/runtime/src/kmp_affinity.h +++ openmp/trunk/runtime/src/kmp_affinity.h @@ -237,53 +237,56 @@ void resize(kmp_uint32 nproc) { kmp_int8 bool_result = KMP_COMPARE_AND_STORE_ACQ8(&resizing, 0, 1); - if (bool_result == 0) { // Someone else is resizing - while (TCR_1(resizing) != 0) KMP_CPU_PAUSE(); - return; + while (bool_result == 0) { // someone else is trying to resize + KMP_CPU_PAUSE(); + if (nproc <= base_num_threads) // happy with other thread's resize + return; + else // try to resize + bool_result = KMP_COMPARE_AND_STORE_ACQ8(&resizing, 0, 1); } KMP_DEBUG_ASSERT(bool_result!=0); - KMP_DEBUG_ASSERT(nproc > base_num_threads); + if (nproc <= base_num_threads) return; // happy with other thread's resize // Calculate new maxLevels kmp_uint32 old_sz = skipPerLevel[depth-1]; kmp_uint32 incs = 0, old_maxLevels = maxLevels; - // First see if old maxLevels is enough to contain new size + // First see if old maxLevels is enough to contain new size for (kmp_uint32 i=depth; iold_sz; ++i) { skipPerLevel[i] = 2*skipPerLevel[i-1]; + numPerLevel[i-1] *= 2; old_sz *= 2; depth++; } - if (nproc <= old_sz) // enough space already - return; - // Not enough space, need to expand hierarchy - while (nproc > old_sz) { - old_sz *=2; - incs++; - depth++; - } - maxLevels += incs; - - // Resize arrays - kmp_uint32 *old_numPerLevel = numPerLevel; - kmp_uint32 *old_skipPerLevel = skipPerLevel; - numPerLevel = skipPerLevel = NULL; - numPerLevel = (kmp_uint32 *)__kmp_allocate(maxLevels*2*sizeof(kmp_uint32)); - skipPerLevel = &(numPerLevel[maxLevels]); + if (nproc > old_sz) { // Not enough space, need to expand hierarchy + while (nproc > old_sz) { + old_sz *=2; + incs++; + depth++; + } + maxLevels += incs; - // Copy old elements from old arrays - for (kmp_uint32 i=0; i 0); // Adjust the hierarchy in case num threads exceeds original - if (nproc > machine_hierarchy.skipPerLevel[depth-1]) + if (nproc > machine_hierarchy.base_num_threads) machine_hierarchy.resize(nproc); + depth = machine_hierarchy.depth; + KMP_DEBUG_ASSERT(depth > 0); + thr_bar->depth = depth; thr_bar->base_leaf_kids = (kmp_uint8)machine_hierarchy.numPerLevel[0]-1; thr_bar->skip_per_level = machine_hierarchy.skipPerLevel;