Skip to content

MDEV-35566 Ensure compatibility with ARMv9 by updating .arch directive #3675

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
cythe wants to merge 1 commit into from
Closed

MDEV-35566 Ensure compatibility with ARMv9 by updating .arch directive #3675

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion storage/innobase/sync/cache.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,7 +82,7 @@ static void pmem_cvap(const void* buf, size_t size)
for (uintptr_t u= uintptr_t(buf) & ~(CPU_LEVEL1_DCACHE_LINESIZE),
end= uintptr_t(buf) + size;
u < end; u+= CPU_LEVEL1_DCACHE_LINESIZE)
__asm__ __volatile__(".arch armv8.2-a\n dc cvap, %0" :: "r"(u) : "memory");
__asm__ __volatile__(".arch armv9.4-a\n dc cvap, %0" :: "r"(u) : "memory");

@dr-m dr-m Dec 4, 2024

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thank you for this fix! Because I am not very familiar with this architecture and toolchains, I played with https://godbolt.org using a modified version of their default example:

#include <cstdint>
uintptr_t square(uintptr_t num) {
      __asm__ __volatile__(".arch armv9.4-a\n dc cvap, %0" :: "r"(num) : "memory");
    return num * num;
}

GCC would just blindly emit the __asm__ snippet to an external assembler, even something nonsensical like dc cvap, eax on x86. I have to enable "compile to binary" to actually invoke an assembler and show the disassembled output. And that is where it seems to be failing for me, even with GCC 14.2.0:

/tmp/ccie1qI1.s: Assembler messages:
/tmp/ccie1qI1.s:21: Error: unknown architecture `armv9.4-a'

I don’t know which assembler version they are using, though. We seem to have a similar problem on our CI as well, for example in https://buildbot.mariadb.org/#/builders/830/builds/458/steps/5/logs/stdio:

/tmp/ccHlQ7lL.s: Assembler messages:
/tmp/ccHlQ7lL.s:79: Error: unknown architecture `armv9.4-a'
/tmp/ccHlQ7lL.s:80: Error: selected processor does not support system register name 'cvap'
make[2]: *** [storage/innobase/CMakeFiles/innobase.dir/build.make:1616: storage/innobase/CMakeFiles/innobase.dir/sync/cache.cc.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:5340: storage/innobase/CMakeFiles/innobase.dir/all] Error 2
make[1]: *** Waiting for unfinished jobs....

The oldest version of clang that recognizes .arch armv9.4-a is clang-16. At least back to clang-9, the sample program with .arch armv8.2-a would compile.

I think that we must support reasonably old compiler versions than that. For proper clang support, I suppose that we could check __clang_major__. But I see that there could be an even better trick:

#include <cstdint>
uintptr_t square(uintptr_t num) {
#if defined __ARM_ARCH && __ARM_ARCH == 9
      __asm__ __volatile__(".arch armv9.4-a\n dc cvap, %0" :: "r"(num) : "memory");
#else
      __asm__ __volatile__(".arch armv8.2-a\n dc cvap, %0" :: "r"(num) : "memory");
#endif
    return num * num;
}

Can you please revise this accordingly? Please rebase the fix on the 10.11 branch and edit the target branch of the pull request to that. That is the oldest version where this fix is applicable.

@dr-m dr-m Dec 4, 2024

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note: Not all our CI builders are reporting their status to GitHub. I found the above build failure in the grid view of this pull request.

@cythe cythe Dec 4, 2024

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for your prompt reply.

This part of the code is compiled for the arm64 structure, so it is pointless to compile this code under x86 and arm32. Here is a simplified part of the code cache.cc file:

#include <cstdint>

#if defined __x86_64__ || defined __aarch64__ || defined __powerpc64__

/* x86_64 code */

#elif defined __aarch64__

/* ... */
__asm__ __volatile__(".arch armv9.4-a\n dc cvap, %0" :: "r"(num) : "memory");
/* ... */

#elif defined __powerpc64__

/* powerpc64 code */

#endif

I see, we want to compile on older versions of the compiler, right? I will resend a pull request later.

@dr-m dr-m Jan 7, 2025

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, we are targeting whatever compiler versions the supported operating systems originally shipped with. Thanks to Red Hat Enterprise Linux 7 and CentOS 7 finally reaching their end-of-life last year, the minimum was bumped from GCC 4.8.5 to GCC 7. For clang, I do not know what a reasonable minimum might be.

@dr-m dr-m Mar 13, 2025

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Does my suggestion to check for

#if defined __ARM_ARCH && __ARM_ARCH == 9

work in your environment?

__asm__ __volatile__("dmb ishst" ::: "memory");
}

Expand Down