126         .warps      = 1,
131         .warps      = 1,
227   hsg_config.warp.skpw_bs = bs_keys / hsg_merge[0].warps;  in hsg_config_init_shared()
240     uint32_t const bs_threads   = merge->warps << hsg_config.warp.lanes_log2;  in hsg_merge_levels_init_shared()
269     uint32_t const bc_warps_min  = MAX_MACRO(merge->warps,hsg_config.block.warps_min);  in hsg_merge_levels_init_shared()
292 hsg_merge_levels_init_1(struct hsg_merge * const merge, uint32_t const warps, uint32_t const level,…  in hsg_merge_levels_init_1()  argument
294   uint32_t const even_odd = warps & 1;  in hsg_merge_levels_init_1()
297   merge->levels[level].networks[even_odd] = warps;  in hsg_merge_levels_init_1()
299   if (warps == 1)  in hsg_merge_levels_init_1()
302   merge->levels[level].active.b64 |= BITS_TO_MASK_AT_64(warps,offset);  in hsg_merge_levels_init_1()
333   merge->networks[index] = warps;  in hsg_merge_levels_init_1()
336   uint32_t const l = (warps+1)/2; // lower/larger  on left  in hsg_merge_levels_init_1()
337   uint32_t const r = (warps+0)/2; // higher/smaller on right  in hsg_merge_levels_init_1()
757   if (active < merge->warps)  in hsg_bc_half_merge_level()
806   uint32_t const warps    = MAX_MACRO(merge->warps,hsg_config.block.warps_min);  in hsg_bc_half_merge()  local
837         ops = hsg_op(ops,BC_REG_SHARED_LOAD_V(warps,r_lo+c,c));  in hsg_bc_half_merge()
892       uint32_t active = merge->warps;  in hsg_bs_flip_merge_level()
895       if (merge->warps > s_rows) {  in hsg_bs_flip_merge_level()
946           base += active * merge->warps;  in hsg_bs_flip_merge_level()
1009       if (merge->levels[level].active.b64 != BITS_TO_MASK_64(merge->warps))  in hsg_bs_flip_merge()
1075   if (merge->warps > 1)  in hsg_bs_sort()
1101       if (m->warps == 0)  in hsg_bs_sort_all()
1128   if (merge->warps == 1)  in hsg_bc_clean()
1167       if (m->warps == 0)  in hsg_bc_clean_all()
1174       if (!is_pow2_u32(m->warps))  in hsg_bc_clean_all()
1272 hsg_fm_merge_all(struct hsg_op * ops, uint32_t const scale_log2, uint32_t const warps)  in hsg_fm_merge_all()  argument
1274   uint32_t const span_left    = (warps << scale_log2) / 2;  in hsg_fm_merge_all()
1345   uint32_t const warps      = hsg_merge[0].warps;  in hsg_xm_merge_all()  local
1346   uint32_t const warps_pow2 = pow2_rd_u32(warps);  in hsg_xm_merge_all()
1352     ops = hsg_fm_merge_all(ops,scale_log2,warps);  in hsg_xm_merge_all()
1468             uint32_t const warps = atoi(optarg);  in main()  local
1471             if ((warps & 1) != 0)  in main()
1478             hsg_merge[0].warps = warps;  in main()
1482               hsg_config.block.warps_max = pow2_ru_u32(warps);  in main()
1540           hsg_config.merge.flip.warps = atoi(optarg);  in main()
1544           hsg_config.merge.half.warps = atoi(optarg);  in main()
1560   uint32_t const warps_ru_pow2 = pow2_ru_u32(hsg_merge[0].warps);  in main()
1565       hsg_merge[ii].warps = warps_ru_pow2 >> ii;  in main()
1601       if (merge->warps == 0)  in main()
1609               merge->warps);  in main()
1613       hsg_merge_levels_init_1(merge,merge->warps,0,0);  in main()