Kyrylo Tkachov <ktkac...@nvidia.com> writes:
> Hi Spencer,
>
>> On 28 Jul 2025, at 16:25, Spencer Abson <spencer.ab...@arm.com> wrote:
>>
>> Streaming-compatible functions can be compiled without SME enabled, but need
>> to use "SMSTART SM" and "SMSTOP SM" to temporarily switch into the streaming
>> mode of a callee. These switches are conditional on the current mode being
>> opposite to the target mode, so no SME instructions are executed if SME is
>> not
>> available.
>>
>> However, in GAS, "SMSTART SM" and "SMSTOP SM" always require +sme. A call
>> from a streaming-compatible function, compiled without SME enabled, to a non
>> -streaming function will be rejected as:
>
> I guess this is an argument for allowing SMTART/SMSTOP in gas unconditionally?
> Though it would need to be consistent with LLVM and we’d still need a
> workaround for older binutils, so okay...
I'm not sure it would be appropriate to allow SMSTART and SMSTOP without SME.
Although the instructions are used in streaming-compatible functions,
their use is guarded by a runtime test that streaming mode is active
(and thus SME is available). In that sense, they are really being used
in blocks of SME-specific code that are embedded in a function that is
PSTATE.SM-agnostic overall.
If we needed to do other SME things besides SMSTART and SMSTOP then
it might make sense to add SME to the assembler architecture extensions
(aarch64_set_asm_isa_flags). But that would also remove some sanity
checking of inline asms, and I suppose of GCC's own code. .inst
seems more robust.
>> Error: selected processor does not support `smstop sm'..
>>
>> To work around this, we make use of the .inst directive to insert the literal
>> encodings of "SMSTART SM" and "SMSTOP SM".
>>
>> gcc/ChangeLog:
>> PR target/121028
>> * config/aarch64/aarch64-sme.md (aarch64_smstart_sm): Use the .inst
>> directive if !TARGET_SME.
>> (aarch64_smstop_sm): Likewise.
>>
>> gcc/testsuite/ChangeLog:
>> PR target/121028
>> * gcc.target/aarch64/sme/call_sm_switch_1.c: Tell check-function
>> -bodies not to ignore .inst directives, and replace the test for
>> "smstart sm" with one for it's encoding.
>> * gcc.target/aarch64/sme/call_sm_switch_11.c: Likewise.
>> * gcc.target/aarch64/sme/pr121028.c: New test.
>> ---
>> gcc/config/aarch64/aarch64-sme.md | 12 ++++-
>> .../gcc.target/aarch64/sme/call_sm_switch_1.c | 4 +-
>> .../aarch64/sme/call_sm_switch_11.c | 5 +-
>> .../gcc.target/aarch64/sme/pr121028.c | 46 +++++++++++++++++++
>> 4 files changed, 61 insertions(+), 6 deletions(-)
>> create mode 100644 gcc/testsuite/gcc.target/aarch64/sme/pr121028.c
>>
>> diff --git a/gcc/config/aarch64/aarch64-sme.md
>> b/gcc/config/aarch64/aarch64-sme.md
>> index 6b3f4390943..5264bf75c0b 100644
>> --- a/gcc/config/aarch64/aarch64-sme.md
>> +++ b/gcc/config/aarch64/aarch64-sme.md
>> @@ -62,6 +62,10 @@
>> ;; (b) they are sometimes used conditionally, particularly in streaming-
>> ;; compatible code.
>> ;;
>> +;; To prevent the latter from upsetting the assembler, we emit the literal
>> +;; encodings of "SMSTART SM" and "SMSTOP SM" when compiling without
>> +;; TARGET_SME.
>> +;;
>> ;; =========================================================================
>>
>> ;; -------------------------------------------------------------------------
>> @@ -161,7 +165,9 @@
>> (clobber (reg:VNx16BI P14_REGNUM))
>> (clobber (reg:VNx16BI P15_REGNUM))]
>> ""
>> - "smstart\tsm"
>> + {
>> + return TARGET_SME ? "smstart\tsm" : ".inst 0xd503437f";
>> + }
>
> Usually when we emit such .inst directives we also emit the mnemonic as an
> assembly comment to aid human assembly debugging.
Yeah, agreed that we should have a comment ("// smstart sm", etc.).
> Ok by me with these changes, but maybe Richard is best-placed to comment on
> the SME parts.
LGTM too with that change.
Thanks,
Richard
