Solar Designer

> This is without explicit caching/reuse - letting the compiler possibly do it. Looks like explicit caching is sometimes needed. Experimenting with `yescrypt/sha256.c`, I get 1944 lines of `objdump -d...

> MD4's `G()` too is that same function. Thanks, @magnumripper! I've just tried patching my generic `md4.c` accordingly. On i386 (does anyone still care?), this results in code size increase...

> Committed SHA-256 changes into yescrypt (not yet updated in JtR) It is now. We should also make similar changes to `sha2.c`, which we use in `--without-openssl` builds, or switch...

> Do we use self hosted github runners? No, we currently run CI jobs in GitHub and CircleCI infrastructure. I see no need for self-hosted for this task - testing...

OTOH, in all of those places we use `SECP256K1_CONTEXT_NONE`, which may skip much of the initialization, and maybe our uses are such that they don't require nor trigger further initialization....

Upon a closer look, that code in`get_compute_capability()` may be dead anyway? We check for `if (!major) {` before reaching the "Apple, VCL and some other environments" fallback code that checks...

> Upon a closer look, that code in`get_compute_capability()` may be dead anyway? We check for `if (!major) {` before reaching the "Apple, VCL and some other environments" fallback code that...

I think we don't need to "fix" this issue because it was only in legacy code, which is unlikely to be needed on systems recent enough to have RTX 4000...

> it doesn't want to work with the Керlr hash We have no OpenCL (and thus no GPU) support for Keplr. It uses scrypt at 128 MiB per hash, which...

Done for OpenCL via #5401, yet to do for CPU.