[PATCH] RISC-V: Check the Zca extension when disabling the C extension using .option norvc.

[Nelson Chu]

Yeah thanks for the clarification.  The .option norvc intents to make all
instructions are 32 bits, although removing extensions is not the original
intention, but for the correctness probably also needs to remove all 16-bit
extensions, including the new zcmop/zcmt/zcmp?

However, I prefer to keep the original intention of ".option rvc/norvc",
which means allow/dis-allow 16-bit instructions, so that probably not same
as "allow/dis-allow the standard C extension" since we will/already ratify
more 16-bit ZC extensions, which aren't included in standard C.  That
should also works for RVC flags since it means no 16 bits instructions if
the flag isn't raised.  Therefore,

.option rvc = +c,+zcmop,+zcmt,+zcmp, ...
.option norvc = -c,-zcmop,-zcmt,-zcmt, ...

cc more people, Jim, Palmer and Kito ;)

Nelson


On Tue, Oct 8, 2024 at 2:41 PM Andrew Waterman <andrew@sifive.com> wrote:

> On Mon, Oct 7, 2024 at 11:32 PM Nelson Chu <nelson@rivosinc.com> wrote:
> >
> > Hi,
> >
> > According to the newest ISA spec,
> https://github.com/riscv/riscv-isa-manual/blob/main/src/zc.adoc, the C is
> actually same as Zca, Zcf and Zcd, so this change looks conflict.  The
> original code should be correct, the RVC flag means enable the C extension,
> so NORVC means disable C, and it makes sense to disable the RVC flag.  But
> this patch also reminds us that the current toolchain needs to be updated
> to the newest spec.
> >
> > 1. C needs imply Zca, Zcf and Zcd, so the implicit table needs to be
> updated.
> >
> > 2. Removing extensions using ".option arch" is more complicated than the
> first implemented.  For this case, C is the set of Zca, Zcf and Zcd, so we
> need to imply those extensions when adding C, and should also need to
> remove those extensions when removing C.  That is - riscv_update_subset
> (&riscv_rps_as, "-c,-zca,-zcf,-zcd") for .option norvc.  But unfortunately
> it's not always work to directly remove the extension from the implicit
> table in reverse.  For example, zcd implies d and zca, but remove zcd
> doesn't mean remove d since zcd isn't a super set of d and zca.
> >
> > cc Andrew in case I am wrong ;)
>
> Nelson's right about the ISA implications.
>
> ".option rvc" is of course not defined by the ISA, and the ASM manual
> doesn't do a great job of defining it, either.  But the original
> intent was to suppress automatic instruction compression (regardless
> of exactly which extension it was). IOW, the intent was to make it so
> that all instructions that are notionally 32 bits long remain 32 bits
> long.  Enabling or disabling extensions was not the intent.
>
> >
> > Thanks
> > Nelson
> >
> >
> > On Fri, Oct 4, 2024 at 3:12 PM Ethan Yu-Cheng Liang(梁淯程) <
> ycl669@andestech.com> wrote:
> >>
> >> .option norvc disables only the C extension, not all 16-bit
> instructions.
> >>
> >> Signed-off-by: Ethan Yu-Cheng Liang <ycl669@andestech.com>
> >> ---
> >>  gas/config/tc-riscv.c | 2 +-
> >>  1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/gas/config/tc-riscv.c b/gas/config/tc-riscv.c
> >> index ef455e449b9..4ef7b8e689e 100644
> >> --- a/gas/config/tc-riscv.c
> >> +++ b/gas/config/tc-riscv.c
> >> @@ -4917,7 +4917,7 @@ s_riscv_option (int x ATTRIBUTE_UNUSED)
> >>      {
> >>        riscv_update_subset (&riscv_rps_as, "-c");
> >>        riscv_set_arch_str (&riscv_rps_as.subset_list->arch_str);
> >> -      riscv_set_rvc (false);
> >> +      riscv_set_rvc (riscv_subset_supports (&riscv_rps_as, "zca"));
> >>      }
> >>    else if (strcmp (name, "pic") == 0)
> >>      riscv_opts.pic = true;
> >> --
> >> 2.45.2
> >>
> >> CONFIDENTIALITY NOTICE:
> >>
> >> This e-mail (and its attachments) may contain confidential and legally
> privileged information or information protected from disclosure. If you are
> not the intended recipient, you are hereby notified that any disclosure,
> copying, distribution, or use of the information contained herein is
> strictly prohibited. In this case, please immediately notify the sender by
> return e-mail, delete the message (and any accompanying documents) and
> destroy all printed hard copies. Thank you for your cooperation.
> >>
> >> Copyright ANDES TECHNOLOGY CORPORATION - All Rights Reserved.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://sourceware.org/pipermail/binutils/attachments/20241008/9b2ed57c/attachment.htm>