Skip to content

Conversation

@martin-frbg
Copy link
Collaborator

@martin-frbg martin-frbg commented Sep 17, 2022

(Hopefully temporary) fix for faulty DGEMM etc. with MIPS64_GENERIC, as without a KERNEL file the default selection of sources does not match the 2x2 MN defined in param.h. Ideally the MIPS64_GENERIC introduced with #3720 needs to be removed again, and the getarch code improved instead to properly distinguish between MIPS32 and MIPS64 on its own. (It has always been possible to build with TARGET=GENERIC, but autodetection of the "GENERIC" target never worked)

@martin-frbg martin-frbg added this to the 0.3.22 milestone Sep 17, 2022
@XiWeiGu
Copy link
Contributor

XiWeiGu commented Sep 19, 2022

There are many GENERIC. At the very top we have TARGET=GENERIC, for RISC-V 64 we have TARGET=RISCV64_GENERIC, for System Z we have TARGET=ZARCH_GENERIC and so on. This might be complicated.
GENERIC is meant to use the C kernel and turn off platform optimization. Can we just keep one TARGET=GENERIC and
remove the others?
If something is wrong with my understanding,please pardon me.

@martin-frbg
Copy link
Collaborator Author

Correct, this is just a quick fix to silence the CI jobs. But at least for mips this is complicated as autodetection (c_check, getarch etc) does not correctly determine 32 or 64bit ABI right now. i guess this is why FlyGoat added MIPS64_GENERIC to use in DYNAMIC_ARCH. The other NAMED_GENERIC may have additional parameters compared to, err, generic GENERIC - need to look into that.

@martin-frbg martin-frbg merged commit 667d0e0 into OpenMathLib:develop Sep 19, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants