--- Log opened Thu Jun 05 00:00:04 2025
01:16 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has quit [Quit: My iMac has gone to sleep. ZZZzzz…]
02:28 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has joined #secp256k1
04:01 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has quit [Quit: My iMac has gone to sleep. ZZZzzz…]
05:34 < hebasto> on which systems does enabling x86_64 assembly support noticeably improve performance? I’m comparing builds with `-DSECP256K1_ASM=x86_64` and `-DSECP256K1_ASM=OFF`, but I don’t observe any significant difference on my Intel-based machines. Also see https://github.com/bitcoin-core/secp256k1/pull/1681#issuecomment-2944005145.
05:49 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has joined #secp256k1
05:49 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has quit [Client Quit]
05:51 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has joined #secp256k1
05:54 < sipa> hmm, i don't see a noticeable difference either on Ryzen 5950X
06:03 < sipa> oh, right, we removed the field x86_64 asm code after it was shown to not be a win anymore compared to what compilers on pure C code
06:04 < sipa> there is still asm logic for the scalar arithmetic, but that's a much smaller overall part of the runtime
06:05 < hebasto> is it necessary to maintain two implementations then?
06:11 < sipa> that's a good question
06:12 < sipa> i see some 10% speedup with asm for scalar multiplications, but those only constitute a small portion of the overall runtime of most API calls
06:12 < sipa> except multiplicative private key tweaking functions, i guess, which have no benchmarks
06:13 < sipa> $ ./build_noasm/bin/bench_internal mul
06:13 < sipa> scalar_mul                    ,     0.0261    ,     0.0274    ,     0.0294 
06:13 < sipa> ./build/bin/bench_internal mul
06:13 < sipa> scalar_mul                    ,     0.0237    ,     0.0246    ,     0.0280 
06:15 < hebasto> I have a similar difference for the `scalar_mul` benchmark
06:17 < sipa> it's not a particularly large burden to maintain, though - just an #ifdef in two functions
06:17 < sipa> the 32-bit arm asm is a lot worse in that regard, as it's a separate .s file etc
06:28 < hebasto> observing the opposite performance change on Windows
06:28 < hebasto> > .\build_clangcl_noasm\bin\Release\bench_internal.exe
06:28 < hebasto> scalar_mul                    ,     0.0275    ,     0.0278    ,     0.0303
06:28 < hebasto> > .\build_clangcl_asm\bin\Release\bench_internal.exe
06:28 < hebasto> scalar_mul                    ,     0.0328    ,     0.0329    ,     0.0331
06:31 < sipa> strange!
06:35 < hebasto> the same, reverse effect, is observed on linux when building with clang 19.1.7
06:37 < hebasto> $ ./build_noasm/bin/bench_internal mul
06:37 < hebasto> scalar_mul                    ,     0.0270    ,     0.0270    ,     0.0272 
06:37 < hebasto> $ ./build_asm/bin/bench_internal mul
06:37 < hebasto> scalar_mul                    ,     0.0330    ,     0.0330    ,     0.0332
08:24 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has quit [Quit: My iMac has gone to sleep. ZZZzzz…]
08:45 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has joined #secp256k1
08:56 < hebasto> were github actions repo settings modified recently? asking because of ci error https://github.com/bitcoin-core/secp256k1/actions/runs/15470455100
09:05 < sipa> hebasto: yeah, i touched some things to enable it for minisketch
09:14 < sipa> hebasto: reverted
09:15 < hebasto> thanks! it works
09:52 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has quit [Quit: My iMac has gone to sleep. ZZZzzz…]
09:56 -!- jon_atack [~jonatack@user/jonatack] has joined #secp256k1
09:59 -!- jonatack [~jonatack@user/jonatack] has quit [Ping timeout: 252 seconds]
10:00 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has joined #secp256k1
10:55 -!- jon_atack [~jonatack@user/jonatack] has quit [Ping timeout: 252 seconds]
11:26 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has quit [Quit: My iMac has gone to sleep. ZZZzzz…]
13:01 -!- jonatack [~jonatack@user/jonatack] has joined #secp256k1
13:31 < bitcoin-git> [secp256k1] real-or-random pushed 2 commits to master: https://github.com/bitcoin-core/secp256k1/compare/201b2b8f06eb...f24b838bedff
13:31 < bitcoin-git> secp256k1/master 3f31ac4 Hennadii Stepanov: doc: Promote "Building with CMake" to standard procedure
13:31 < bitcoin-git> secp256k1/master f24b838 merge-script: Merge bitcoin-core/secp256k1#1680: doc: Promote "Building with CMake" to s...
13:31 < bitcoin-git> [secp256k1] real-or-random merged pull request #1680: doc: Promote "Building with CMake" to standard procedure (master...250604-cmake) https://github.com/bitcoin-core/secp256k1/pull/1680
13:39 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has joined #secp256k1
14:01 -!- ghost43_ [~ghost43@gateway/tor-sasl/ghost43] has joined #secp256k1
14:02 -!- ghost43 [~ghost43@gateway/tor-sasl/ghost43] has quit [Remote host closed the connection]
14:37 -!- jonatack [~jonatack@user/jonatack] has quit [Quit: WeeChat 4.6.3]
14:45 -!- jonatack [~jonatack@user/jonatack] has joined #secp256k1
14:59 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has quit [Quit: My iMac has gone to sleep. ZZZzzz…]
17:45 -!- jon_atack [~jonatack@user/jonatack] has joined #secp256k1
17:47 -!- jonatack [~jonatack@user/jonatack] has quit [Ping timeout: 248 seconds]
23:58 -!- tromp [~textual@2001:1c00:3487:1b00:fdb0:a3b6:295:d798] has joined #secp256k1
--- Log closed Fri Jun 06 00:00:05 2025