starseeker (May 22 2026 at 00:12): starseeker (May 22 2026 at 00:12):bitv_t bits[2]; in include/bu/bitv.h is giving me grief with the -Werror=stringop-overflow= error again. Multiple AI sessions indicate the right fix is to use a C99 variable array here... @Sean is there any reason not to go ahead and make that shift at this point, since we require that standard anyway?
starseeker (May 22 2026 at 16:12):Never mind, I see C++ doesn't like it
Sean (May 23 2026 at 15:10):I was going to say, that trick is squarely in low-level C territory that doesn't have a migration equivalent. My gut reaction is to set up performance harness around it as it's a critical ray tracing book-keeping structure, but then work to replace it with a C++ construct under the hood as an opaque type. It'll be hard to get it right, i.e., without screwing performance either on reads or writes.
Sean (May 23 2026 at 17:22):I'm setting up some more test harnessing to benchmark it, and have a refactoring in progress.
Sean (May 27 2026 at 00:29):okay, I think I'm done -- short summary is bu_bitv is surprisingly about as fast as I could get it to go.
switching it to a 64bit type fixes the alignment and either that or the fact that it's no longer a 2-element array seems to avoid triggering the compiler warning. I added api to do callback iteration instead of macro hacking, and a few more missing funcs.
also ran through a slew of performance testing, both before and after as well as against STL vector<bool>. summary is we should probably continue to leave it alone... bu_bitv absolutely crushes it on boolean ops and set/testing on sparse bitsets, and is otherwise equivalent on bit set/test and clearing. room for improvement in deserialize/serialize calls.
Sean (May 27 2026 at 01:02):morrison@Miniagua brlcad.main % ./a.out
=============================================================
BRL-CAD bu_bitv vs STL std::vector<bool> benchmark
=============================================================
Vector size : 100000000 bits
Iterations : 10
Ops total : ~1000000000 per test
Operation bu_bitv (us) vector<bool> (us) ratio winner
-------------------------------- ------------ ----------------- ---------- -------------
ratio = stl_time / bitv_time: >1.0 => bu_bitv faster, <1.0 => STL faster
new/free 5953 2308 0.39x STL wins
BU_BITSET / v[i]=1 3460549 3468407 1.00x bu_bitv wins
BU_BITTEST / v[i] 643232 643769 1.00x bu_bitv wins
BU_BITCLR / v[i]=0 3463984 3463004 1.00x STL wins
bu_bitv_clear / assign(0) 1974 1576 0.80x STL wins
bu_bitv_or / stl_or 7959 4833792 607.34x bu_bitv wins
BU_BITS2WORDS / bits2words() 206027 379963 1.84x bu_bitv wins
bu_bitv_dup / copy-ctor 12086 3300 0.27x STL wins
bu_bitv_to_hex / stl_to_hex 155474 9750651 62.72x bu_bitv wins
bu_bitv_vls / stl_vls 18281 929630 50.85x bu_bitv wins
Aggregate totals across all tests:
bu_bitv total: 7975519 us
vector<bool> total: 23476400 us
overall ratio: 2.94x
Last updated: Jun 06 2026 at 02:08 UTC