Bitmanip and Scalar Crypto Intrinsics in GCC

[pz9115]

Technical Group

Crypto TG

ratification-pkg

Scalar Crypto & Bitmanip extensions

Technical Liaison

[email protected]

Task Category

Compilers/Toolchain

Task Sub Category

• gcc
• binutils
• gdb
• intrinsics
• Java
• KVM
• ld
• llvm
• Linux kernel
• QEMU
• Spike

Ratification Target

2021

Statement of Work (SOW)

Component names: gcc

Requirements:
riscv-non-isa/riscv-c-api-doc#44 merged

Deliverables:

• gcc patches

Acceptance Criteria:

• gcc support Bitmanip and Scalar Crypto extension intrinsic format in GCC upstream

Projected timeframe:

• Supports Bitmanip intrinsics in gcc within Nov 2023
• Supports Scalar Crypto intrinsics in gcc within Dec 2023

SOW Signoffs:

• Task group liaison sign-off
• Development partner sign-off
• Toolchains & Runtimes sign-off (if gcc, LLVM, binutils, gdb work)

Waiver

• Freeze
• Ratification

Pull Request Details

riscv-non-isa/riscv-c-api-doc#44 finished review in the last week,
Now we are preparing to support this new feature in gcc upstream.

[cmuellner]

@pz9115, thank you very much for picking this up!

[jjscheel]

IYes, @pz9115, thank you!!!

A couple questions to consider as you plan this work:

1. What size is this (S, M, ...)? See the field values for month estimates.
2. By what date do target having PR(s) accepted upstream?

I'm sure I'll have more questions, but these would be good to understand first.

[pz9115]

Still WIP, plan to send patches within this month.

[pz9115]

Sent the Scalar Crypto intrinsic implement to GCC upstream, Bitmanip part is still debugging, plan to send patch within this week.

https://gcc.gnu.org/pipermail/gcc-patches/2023-November/638238.html

[cmuellner]

I reviewed the initial patch from Nov 28 and we identified that we need to change some of the way the intrinsics are implemented.

The first complete patchset of scalar crypto and bitmanip intrinsics has been published on Dec 5: https://gcc.gnu.org/pipermail/gcc-patches/2023-December/639371.html

I reviewed and identified a few things that need to be improved (e.g. extend test coverage to also test with -O0 and implement intrinsics with immediates such that the compiler will always "see" this value).

A second patchset has been published yesterday, on Dec 7: https://gcc.gnu.org/pipermail/gcc-patches/2023-December/639725.html

I also reviewed this patch set and identified that the tests were not passing.
So, another revision is needed.

[pz9115]

WIP, Shihua is updating the patch following Christoph's comments.

[pz9115]

Shihua resends his third version patches and waits for a new cycle review.

https://gcc.gnu.org/pipermail/gcc-patches/2023-December/641432.html

[pz9115]

Merged into upstream:

https://gcc.gnu.org/git?p=gcc.git;a=commit;h=c7ad6bf5b9e478d87294230318db675d7c8f3ce4

[jjscheel]

Thanks, Jiawei. Does this mean that all work for this SOW is complete?

[pz9115]

Yes, it does. We can close this issue :)

[jjscheel]

Great news, @pz9115. I've updated the State correctly and consider.

@cmuellner, would you mind confirming that we can sign-off of this as our TG rep for me?

[cmuellner]

Yes, I am happy to confirm that bitmanip and scalar crypto intrinsics are specified and available in LLVM and GCC.
To all involved people: thank you very much (especially Craig and @pz9115).

[jjscheel]

Complete. Nice work, @pz9115!!! Thanks for your help @cmuellner!