[Mesa-dev] [PATCH 2/7] i965: Adjust set_message_descriptor to handle non-sends

[Chris Forbes]

Thanks, landed.

On Fri, Aug 15, 2014 at 6:44 PM, Kenneth Graunke <kenneth at whitecape.org> wrote:
> On Sunday, August 03, 2014 12:36:31 PM Chris Forbes wrote:
>> We're about to be using this infrastructure to build descriptors in
>> src1 of non-send instructions, when preparing to do an indirect send.
>>
>> Don't accidentally clobber the conditionalmod field of those
>> instructions with SFID bits, which aren't part of the descriptor.
>>
>> Signed-off-by: Chris Forbes <chrisf at ijw.co.nz>
>> ---
>>  src/mesa/drivers/dri/i965/brw_eu_emit.c | 14 +++++++++++++-
>>  1 file changed, 13 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
>> index bc74de3..cdfc78e 100644
>> --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c
>> +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c
>> @@ -540,7 +540,19 @@ brw_set_message_descriptor(struct brw_compile *p,
>>     struct brw_context *brw = p->brw;
>>
>>     brw_set_src1(p, inst, brw_imm_d(0));
>> -   brw_inst_set_sfid(brw, inst, sfid);
>> +
>> +   /* For indirect sends, `inst` will not be the SEND/SENDC instruction
>> +    * itself; instead, it will be a MOV/OR into the address register.
>> +    *
>> +    * In this case, we avoid setting the extended message descriptor bits,
>> +    * since they go on the later SEND/SENDC instead and if set here would
>> +    * instead clobber the conditionalmod bits.
>> +    */
>> +   unsigned opcode = brw_inst_opcode(brw, inst);
>> +   if (opcode == BRW_OPCODE_SEND || opcode == BRW_OPCODE_SENDC) {
>> +      brw_inst_set_sfid(brw, inst, sfid);
>> +   }
>> +
>>     brw_inst_set_mlen(brw, inst, msg_length);
>>     brw_inst_set_rlen(brw, inst, response_length);
>>     brw_inst_set_eot(brw, inst, end_of_thread);
>>
>
> Hey Chris,
>
> I was originally confused by this - setting message descriptor bits on non-SEND instructions seemed bizarre to me.  But, now that I've looked a bit more closely, I see what you're doing...the message descriptor bits are already in src1, so it actually is a convenient way to set up the right constant to copy into a0.0.
>
> Clever - nice work :)
>
> This series is:
> Reviewed-by: Kenneth Graunke <kenneth at whitecape.org>