Update libsecp256k1 subtree to current master #29169
Conversation
|
The following sections might be updated with supplementary metadata relevant to reviewers and maintainers. Code CoverageFor detailed information about the code coverage, see the test coverage report. ReviewsSee the guideline for information on the review process.
If your review is incorrectly listed, please react with 👎 to this comment and the bot will ignore it on the next update. |
|
Obvious ACK c13a17c is obvious. |
|
Shouldn't the subtree be synced to the |
Why? It makes 0 difference. |
|
I would also prefer not to introduce some policy about only using tags, or similar, because we want to retain the ability to update the subtree to any commit we'd like, at any point. |
There was a problem hiding this comment.
ACK c13a17c, I've updated the secp256k1 to its current master branch (efe85c70a2e357e3605a8901a9662295bae1001f) locally and got zero diff with this PR.
However, the PR title and description are a bit misleading as they mention "0.4.1 release", which implies (at least for me) syncing to the v0.4.1 tag.
There was a problem hiding this comment.
Concept ACK c13a17c I haven't checked that the subtree is correct but no objections to update to this commit
However, the PR title and description are a bit misleading as they mention "0.4.1 release", which implies (at least for me) syncing to the
v0.4.1tag.
True, this updates to the merge commit of the "cleanup" PR after the release. But yeah, it really doesn't matter in the end, and I think there's no need to invalidate the ACKs here.
What do you mean by/where are we reporting the version number? |
The commit title, "Update secp256k1 subtree to upstream release 0.4.1", makes it sound like it's updating to the exact 0.4.1 tag. I agree it's fine to update to master, and that we shouldn't need a policy of only using tagged releases, but it'd be better if the commit title said "Update secp256k1 subtree to current master", or "to commit after 0.4.1". |
Plus, |
efe85c70a2 Merge bitcoin-core/secp256k1#1466: release cleanup: bump version after 0.4.1 4b2e06f460 release cleanup: bump version after 0.4.1 1ad5185 Merge bitcoin-core/secp256k1#1465: release: prepare for 0.4.1 672053d release: prepare for 0.4.1 1a81df8 Merge bitcoin-core/secp256k1#1380: Add ABI checking tool for release process 74a4d97 doc: Add ABI checking with `check-abi.sh` to the Release Process e7f830e Add `tools/check-abi.sh` 77af1da Merge bitcoin-core/secp256k1#1455: doc: improve secp256k1_fe_set_b32_mod doc 3928b7c doc: improve secp256k1_fe_set_b32_mod doc 5e9a4d7 Merge bitcoin-core/secp256k1#990: Add comment on length checks when parsing ECDSA sigs 4197d66 Merge bitcoin-core/secp256k1#1431: Add CONTRIBUTING.md 0e5ea62 CONTRIBUTING: add some coding and style conventions e2c9888 Merge bitcoin-core/secp256k1#1451: changelog: add entry for "field: Remove x86_64 asm" d2e36a2 changelog: add entry for "field: Remove x86_64 asm" 1a432cb README: update first sentence 0922a04 docs: move coverage report instructions to CONTRIBUTING 76880e4 Add CONTRIBUTING.md including scope and guidelines for new code d3e29db Merge bitcoin-core/secp256k1#1450: Add group.h ge/gej equality functions 04af0ba Replace ge_equals_ge[,j] calls with group.h equality calls 60525f6 Add unit tests for group.h equality functions a47cd97 Add group.h ge/gej equality functions 10e6d29 Merge bitcoin-core/secp256k1#1446: field: Remove x86_64 asm 07687e8 Merge bitcoin-core/secp256k1#1393: Implement new policy for VERIFY_CHECK and #ifdef VERIFY (issue bitcoin#1381) bb46723 remove VERIFY_SETUP define a3a3e11 remove unneeded VERIFY_SETUP uses in ECMULT_CONST_TABLE_GET_GE macro a0fb68a introduce and use SECP256K1_SCALAR_VERIFY macro cf25c86 introduce and use SECP256K1_{FE,GE,GEJ}_VERIFY macros 5d89bc0 remove superfluous `#ifdef VERIFY`/`#endif` preprocessor conditions c2688f8 redefine VERIFY_CHECK to empty in production (non-VERIFY) mode 5814d84 Merge bitcoin-core/secp256k1#1438: correct assertion for secp256k1_fe_mul_inner c1b4966 Merge bitcoin-core/secp256k1#1445: bench: add --help option to bench_internal f07cead build: Don't call assembly an optimization 2f0762f field: Remove x86_64 asm 1ddd76a bench: add --help option to bench_internal e721039 Merge bitcoin-core/secp256k1#1441: asm: add .note.GNU-stack section for non-exec stack ea47c82 Merge bitcoin-core/secp256k1#1442: Return temporaries to being unsigned in secp256k1_fe_sqr_inner dcdda31 Tighten secp256k1_fe_mul_inner's VERIFY_BITS checks 1027135 Return temporaries to being unsigned in secp256k1_fe_sqr_inner 33dc7e4 asm: add .note.GNU-stack section for non-exec stack c891c5c Merge bitcoin-core/secp256k1#1437: ci: Ignore internal errors of snapshot compilers 8185e72 ci: Ignore internal errors in snapshot compilers 40f50d0 Merge bitcoin-core/secp256k1#1184: Signed-digit based ecmult_const algorithm 8e2a5fe correct assertion for secp256k1_fe_mul_inner 355bbdf Add changelog entry for signed-digit ecmult_const algorithm 21f49d9 Remove unused secp256k1_scalar_shr_int 115fdc7 Remove unused secp256k1_wnaf_const aa9f3a3 ecmult_const: add/improve tests 4d16e90 Signed-digit based ecmult_const algorithm ba523be make SECP256K1_SCALAR_CONST reduce modulo exhaustive group order 2140da9 Add secp256k1_scalar_half for halving scalars (+ tests/benchmarks). 1f1bb78 Merge bitcoin-core/secp256k1#1430: README: remove CI badge 5dab0ba README: remove CI badge b314cf2 Merge bitcoin-core/secp256k1#1426: ci/cirrus: Add native ARM64 jobs fa4d6c7 ci/cirrus: Add native ARM64 persistent workers ee7aaf2 Merge bitcoin-core/secp256k1#1395: tests: simplify `random_fe_non_zero` (remove loop limit and unneeded normalize) ba9cb6f Merge bitcoin-core/secp256k1#1424: ci: Bump major versions for docker actions d9d80fd ci: Bump major versions for docker actions 4fd00f4 Merge bitcoin-core/secp256k1#1422: cmake: Install `libsecp256k1.pc` file 421d848 ci: Align Autotools/CMake `CI_INSTALL` directory names 9f005c6 cmake: Install `libsecp256k1.pc` file 2262d0e ci/cirrus: Bring back skeleton .cirrus.yml without jobs b10ddd2 Merge bitcoin-core/secp256k1#1416: doc: Align documented scripts with CI ones 49be5be Merge bitcoin-core/secp256k1#1390: tests: Replace counting_illegal_callbacks with CHECK_ILLEGAL_VOID cbf3053 Merge bitcoin-core/secp256k1#1417: release cleanup: bump version after 0.4.0 9b118bc release cleanup: bump version after 0.4.0 7030364 tests: add CHECK_ERROR_VOID and use it in scratch tests f8d7ea6 tests: Replace counting_illegal_callbacks with CHECK_ILLEGAL_VOID b0f7bfe doc: Do not mention soname in CHANGELOG.md "ABI Compatibility" section bd9d98d doc: Align documented scripts with CI ones a1d52e3 tests: remove unnecessary test in run_ec_pubkey_parse_test 875b0ad tests: remove unnecessary set_illegal_callback c45b7c4 refactor: introduce testutil.h (deduplicate `random_fe_`, `ge_equals_` helpers) dc55141 tests: simplify `random_fe_non_zero` (remove loop limit and unneeded normalize) e02f313 Add comment on length checks when parsing ECDSA sigs git-subtree-dir: src/secp256k1 git-subtree-split: efe85c70a2e357e3605a8901a9662295bae1001f
fanquake
changed the title
fanquake
force-pushed
the
libsecp256k1_0_4_1
branch
from
c13a17c
to
e2cdeb5
Compare
|
I've updated to commit message to make it clear that this isn't exactly the 0.4.1 tag.
I was thinking more like output from bitcoind or similar. |
There was a problem hiding this comment.
Concept ACK c13a17c I haven't checked that the subtree is correct but no objections to update to this commit
|
Oddly enough, this change benched out as being slightly slower (~4.8%) than the commit before it in master. Not a huge regression, but not what I expected. Compiled with
#29169 vs. f8ca135 (absolute)
#29169 vs. f8ca135 (relative)
|
|
Hm, this is really not expected.
You could see if gcc 10.5.0 makes a difference. AFAIU this is used in the official builds currently (until #27897 lands), and this was one of our considerations. If that doesn't help, please benchmark libsecp256k1 with (bitcoin-core/secp256k1@10e6d29) and without bitcoin-core/secp256k1#1446 (bitcoin-core/secp256k1@07687e8), see the PR for instructions and also consider setting |
|
Quick update: I did confirm (on another machine) that more contemporary versions of gcc ( #29169 vs. f8ca135 (relative)
I also confirmed that the gcc 10 result was not a fluke; a rerun again showed 5% regression. I'm rerunning with gcc 12 on that same machine and will post the results when they're up. I guess this is all expected since the libsecp change was advertised to be contingent on improvements in gcc. |
|
Oddly enough, even with Will run the secp benches in isolation. |
These benchmarks are indeed slower on the first host when compiling with |
|
Wow, bizarre, but not impossible. It seems that the handwritten asm code we had was faster on your CPU, but is slower on all CPUs we tested on. Yours seems to be a fairly old architecture (2nd gen Intel Core, first released in 2011), so it's not impossible there are differences. |
This includes changes from the 0.4.1 release: https://github.com/bitcoin-core/secp256k1/releases/tag/v0.4.1.