[PATCH 00/13] Add support for the T-Head vendor extensions

Christoph Müllner christoph.muellner@vrull.eu
Sun Sep 18 06:51:05 GMT 2022


On Fri, Sep 16, 2022 at 11:58 AM Palmer Dabbelt <palmer@dabbelt.com> wrote:

> On Fri, 16 Sep 2022 02:36:32 PDT (-0700), nelson@rivosinc.com wrote:
> > Hi Christoph,
> >
> >
> > On Tue, Sep 6, 2022 at 8:22 PM Christoph Muellner
> > <christoph.muellner@vrull.eu> wrote:
> >>
> >> From: Christoph Müllner <christoph.muellner@vrull.eu>
> >>
> >> This series introduces support for the T-Head vendor extensions,
> >> which are implemented e.g. in the XuanTie C906 and XuanTie C910
> >> processors:
> >> * XTheadBa
> >> * XTheadBb
> >> * XTheadBs
> >> * XTheadCmo
> >> * XTheadCondMov
> >> * XTheadFMemIdx
> >> * XTheadMac
> >> * XTheadMemIdx
> >> * XTheadMemPair
> >> * XTheadSync
> >>
> >> The xthead* extensions are documented here:
> >>
> https://github.com/T-head-Semi/thead-extension-spec/releases/download/2.0.0/xthead-2022-09-05-2.0.0.pdf
> >>
> >> The "th." instruction prefix prevents future conflicts with standard
> >> extensions and has been documentented in a PR for the RISC-V toolchain
> >> conventions:
> >>   https://github.com/riscv-non-isa/riscv-toolchain-conventions/pull/19
> >>
> >> The goal of this patchset is to provide access to these instruction
> >> so that compilers/users can optimize SW accordingly.
> >>
> >> Note, that the T-Head vendor extensions do not contain all
> >> vendor-specific
> >> functionality of the T-Head SoCs (e.g. no vendor-specific CSRs are
> >> included).
> >> Instead the extensions cover coherent functionality, that is exposed to
> >> S and U mode.
> >>
> >> To enable the extensions above, the following two methods are possible:
> >> * add the extension to the arch string (e.g.
> >> * -march=rv64gc_xtheadcmo_xtheadsync)
> >> * implicitly select the extensions via CPU selection (e.g.
> >> * -mcpu=thead-c910)
> >>
> >> The patchset attempts to minimize code changes in generic/infrastructure
> >> code. All patches in this series come with tests to avoid future
> regressions.
> >>
> >> Christoph Müllner (13):
> >>   RISC-V: Add generic support for vendor extensions
> >>   RISC-V: Add T-Head CMO vendor extension
> >>   RISC-V: Add T-Head SYNC vendor extension
> >>   RISC-V: Add support for arbitrary immediate encoding formats
> >>   RISC-V: Add T-Head Bitmanip vendor extension
> >>   RISC-V: Add T-Head CondMov vendor extension
> >>   RISC-V: Add T-Head MAC vendor extension
> >>   RISC-V: Add T-Head FMemIdx vendor extension
> >>   RISC-V: Add T-Head MemIdx vendor extension
> >>   RISC-V: Add support for literal instruction arguments
> >>   RISC-V: Add T-Head MemPair vendor extension
> >
> > There are three minor issues,
> >
> > 1. Probably need to add new extension support in
> > riscv_multi_subset_supports_ext, although it is just used for
> > reporting better error messages.
> > 2. The operand L seems only to be used for t-head for now, so it
> > should be still a vendor operand, named with the X prefix probably
> > better.
> > 3. I remember Lifang mentioned before that we should add these vendor
> > stuff based on the vendor infra in the integration branch, and that's
> > what I asked him to rebase his patches  to there before, since the
> > previous policy cannot accept vendor stuff on master branch.  But now
> > I think most of us agree to accept t-head extensions on mainline
> > directly, so I figured these things cannot be blocked by the vendor
> > infra and cannot be committed.  That doesn't mean we don't need the
> > vendor infra, that means we can support the vendor infra later if me
> > or someone have time to move them back to mainline.
> >
> > Therefore, these patches look good to me for now, thank you very much
> > for spending so much time to implement these.  But personally, I think
> > we should get the approval from the T-Head guys at least.  Btw, If
> > these implementations are referred to and based on the integration
> > branch, then I think it would be better to keep at least the
> > co-authors there in the commit message here; if these patches are
> > re-write without referring to the implementation from the integration
> > branch, then I need to apologize to them, because I asked them to
> > rewrite their patches over there, and cannot be committed to mainline
> > at that time...
> >
> >>   riscv: gas: Add command line option '-mcpu=' to specify the CPU
> >>   riscv: Add T-Head entries for the -mcpu= flag
> >
> > Not sure if it is necessary to add -mcpu in assembler, since the
> > compiler should recognize -mcpu, and will expand it to the correct
> > extensions to assembler.  If there are no objections for a period of
> > time, then please go ahead and commit these.
>
> I'd also learn towards keeping -mcpu out of the assembler, it's just
> going to result in subtly different behavior between GCC and GAS which
> will cause subtle headaches in the future.  All -mcpu does is set -march
> and -mtune, so if we just continue keeping it in GCC then we have a
> single source of truth.  GCC will already need to understand -march for
> everything that GAS needs to know anyway, so it's not like it's adding
> complexity.
>
> Is there something broken in GCC where -mcpu isn't setting the right
> -march for GAS and thus we're not getting the right extensions enabled?
> If that's the case I'd argue we're better off just fixing the bug over
> there.
>

No, nothing is broken.
The main use case was direct invocations of the assembler,
but I'm fine with dropping the patch.

Thanks,
Christoph


>
> >
> > Thank you very much
> > Nelson
> >
> >>  bfd/elfxx-riscv.c                             |  39 ++-
> >>  gas/config/tc-riscv.c                         | 127 +++++++
> >>  gas/doc/c-riscv.texi                          |  70 ++++
> >>  gas/testsuite/gas/riscv/mcpu-fail-unknown.d   |   3 +
> >>  gas/testsuite/gas/riscv/mcpu-fail-unknown.l   |   2 +
> >>  gas/testsuite/gas/riscv/mcpu-ok-c906.d        |   6 +
> >>  gas/testsuite/gas/riscv/mcpu-ok-c910.d        |   6 +
> >>  gas/testsuite/gas/riscv/x-thead-ba-fail.d     |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-ba-fail.l     |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-ba-fail.s     |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-ba.d          |  13 +
> >>  gas/testsuite/gas/riscv/x-thead-ba.s          |   6 +
> >>  gas/testsuite/gas/riscv/x-thead-bb-fail.d     |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-bb-fail.l     |   7 +
> >>  gas/testsuite/gas/riscv/x-thead-bb-fail.s     |   7 +
> >>  gas/testsuite/gas/riscv/x-thead-bb.d          |  30 ++
> >>  gas/testsuite/gas/riscv/x-thead-bb.s          |  22 ++
> >>  gas/testsuite/gas/riscv/x-thead-bs-fail.d     |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-bs-fail.l     |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-bs-fail.s     |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-bs.d          |  14 +
> >>  gas/testsuite/gas/riscv/x-thead-bs.s          |   6 +
> >>  gas/testsuite/gas/riscv/x-thead-cmo-fail.d    |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-cmo-fail.l    |  22 ++
> >>  gas/testsuite/gas/riscv/x-thead-cmo-fail.s    |  22 ++
> >>  gas/testsuite/gas/riscv/x-thead-cmo.d         |  30 ++
> >>  gas/testsuite/gas/riscv/x-thead-cmo.s         |  22 ++
> >>  gas/testsuite/gas/riscv/x-thead-condmov.d     |  11 +
> >>  gas/testsuite/gas/riscv/x-thead-condmov.s     |   3 +
> >>  .../gas/riscv/x-thead-fmemidx-fail.d          |   3 +
> >>  .../gas/riscv/x-thead-fmemidx-fail.l          |  18 +
> >>  .../gas/riscv/x-thead-fmemidx-fail.s          |  17 +
> >>  gas/testsuite/gas/riscv/x-thead-fmemidx.d     |  25 ++
> >>  gas/testsuite/gas/riscv/x-thead-fmemidx.s     |  17 +
> >>  gas/testsuite/gas/riscv/x-thead-mac.d         |  15 +
> >>  gas/testsuite/gas/riscv/x-thead-mac.s         |   7 +
> >>  gas/testsuite/gas/riscv/x-thead-memidx-fail.d |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-memidx-fail.l |  14 +
> >>  gas/testsuite/gas/riscv/x-thead-memidx-fail.s |  14 +
> >>  gas/testsuite/gas/riscv/x-thead-memidx.d      |  53 +++
> >>  gas/testsuite/gas/riscv/x-thead-memidx.s      |  48 +++
> >>  .../gas/riscv/x-thead-mempair-fail.d          |   3 +
> >>  .../gas/riscv/x-thead-mempair-fail.l          |  30 ++
> >>  .../gas/riscv/x-thead-mempair-fail.s          |  30 ++
> >>  gas/testsuite/gas/riscv/x-thead-mempair.d     |  14 +
> >>  gas/testsuite/gas/riscv/x-thead-mempair.s     |   6 +
> >>  gas/testsuite/gas/riscv/x-thead-sync-fail.d   |   3 +
> >>  gas/testsuite/gas/riscv/x-thead-sync-fail.l   |   6 +
> >>  gas/testsuite/gas/riscv/x-thead-sync-fail.s   |   6 +
> >>  gas/testsuite/gas/riscv/x-thead-sync.d        |  14 +
> >>  gas/testsuite/gas/riscv/x-thead-sync.s        |   6 +
> >>  include/opcode/riscv-opc.h                    | 326 ++++++++++++++++++
> >>  include/opcode/riscv.h                        |  27 ++
> >>  opcodes/riscv-dis.c                           |  44 +++
> >>  opcodes/riscv-opc.c                           | 155 +++++++++
> >>  55 files changed, 1394 insertions(+), 2 deletions(-)
> >>  create mode 100644 gas/testsuite/gas/riscv/mcpu-fail-unknown.d
> >>  create mode 100644 gas/testsuite/gas/riscv/mcpu-fail-unknown.l
> >>  create mode 100644 gas/testsuite/gas/riscv/mcpu-ok-c906.d
> >>  create mode 100644 gas/testsuite/gas/riscv/mcpu-ok-c910.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-ba-fail.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-ba-fail.l
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-ba-fail.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-ba.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-ba.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bb-fail.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bb-fail.l
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bb-fail.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bb.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bb.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bs-fail.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bs-fail.l
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bs-fail.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bs.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-bs.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-cmo-fail.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-cmo-fail.l
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-cmo-fail.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-cmo.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-cmo.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-condmov.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-condmov.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-fmemidx-fail.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-fmemidx-fail.l
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-fmemidx-fail.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-fmemidx.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-fmemidx.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-mac.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-mac.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-memidx-fail.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-memidx-fail.l
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-memidx-fail.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-memidx.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-memidx.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-mempair-fail.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-mempair-fail.l
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-mempair-fail.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-mempair.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-mempair.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-sync-fail.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-sync-fail.l
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-sync-fail.s
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-sync.d
> >>  create mode 100644 gas/testsuite/gas/riscv/x-thead-sync.s
> >>
> >> --
> >> 2.37.2
> >>
>


More information about the Binutils mailing list