[Mesa-dev] [PATCH 3/3] i965/gen9+: Switch thread scratch space to non-coherent stateless access.

[Francisco Jerez]

Kristian Høgsberg <krh at bitplanet.net> writes:

> On Wed, Nov 25, 2015 at 11:37 AM, Francisco Jerez <currojerez at riseup.net> wrote:
>> The thread scratch space is thread-local so using the full IA-coherent
>> stateless surface index (255 since Gen8) is unnecessary and
>> potentially expensive.  On Gen8 and early steppings of Gen9 this is
>> not a functional change because the kernel already sets bit 4 of
>> HDC_CHICKEN0 which overrides all HDC memory access to be non-coherent
>> in order to workaround a hardware bug.
>>
>> This happens to fix a full system hang when running any spilling code
>> on a pre-production SKL GT4e machine I have on my desk (forcing all
>> HDC access to non-coherent from the kernel up to stepping F0 might be
>> a good idea though regardless of this patch), and improves performance
>> of the OglPSBump2 SynMark benchmark run with INTEL_DEBUG=spill_fs by
>> 33% (11 runs, 5% significance) on a production SKL GT2 (on which HDC
>> IA-coherency is apparently functional so it wouldn't make sense to
>> disable globally).
>
> Very nice, all looks good to me. All three patches:
>
> Reviewed-by: Kristian Høgsberg  <krh at bitplanet.net>
>
Thanks!

>> ---
>>  src/mesa/drivers/dri/i965/brw_eu.h               |  2 ++
>>  src/mesa/drivers/dri/i965/brw_eu_emit.c          | 17 +++++++++++++++--
>>  src/mesa/drivers/dri/i965/brw_vec4_generator.cpp |  4 ++--
>>  3 files changed, 19 insertions(+), 4 deletions(-)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_eu.h b/src/mesa/drivers/dri/i965/brw_eu.h
>> index 829e393..62891c7 100644
>> --- a/src/mesa/drivers/dri/i965/brw_eu.h
>> +++ b/src/mesa/drivers/dri/i965/brw_eu.h
>> @@ -317,6 +317,8 @@ void brw_oword_block_read(struct brw_codegen *p,
>>                           uint32_t offset,
>>                           uint32_t bind_table_index);
>>
>> +unsigned brw_scratch_surface_idx(const struct brw_codegen *p);
>> +
>>  void brw_oword_block_read_scratch(struct brw_codegen *p,
>>                                   struct brw_reg dest,
>>                                   struct brw_reg mrf,
>> diff --git a/src/mesa/drivers/dri/i965/brw_eu_emit.c b/src/mesa/drivers/dri/i965/brw_eu_emit.c
>> index da1ddfd..bb6f5dc 100644
>> --- a/src/mesa/drivers/dri/i965/brw_eu_emit.c
>> +++ b/src/mesa/drivers/dri/i965/brw_eu_emit.c
>> @@ -1997,6 +1997,19 @@ void gen6_math(struct brw_codegen *p,
>>     brw_set_src1(p, insn, src1);
>>  }
>>
>> +/**
>> + * Return the right surface index to access the thread scratch space using
>> + * stateless dataport messages.
>> + */
>> +unsigned
>> +brw_scratch_surface_idx(const struct brw_codegen *p)
>> +{
>> +   /* The scratch space is thread-local so IA coherency is unnecessary. */
>> +   if (p->devinfo->gen >= 8)
>> +      return GEN8_BTI_STATELESS_NON_COHERENT;
>> +   else
>> +      return BRW_BTI_STATELESS;
>> +}
>>
>>  /**
>>   * Write a block of OWORDs (half a GRF each) from the scratch buffer,
>> @@ -2097,7 +2110,7 @@ void brw_oword_block_write_scratch(struct brw_codegen *p,
>>
>>        brw_set_dp_write_message(p,
>>                                insn,
>> -                              255, /* binding table index (255=stateless) */
>> +                               brw_scratch_surface_idx(p),
>>                                msg_control,
>>                                msg_type,
>>                                mlen,
>> @@ -2183,7 +2196,7 @@ brw_oword_block_read_scratch(struct brw_codegen *p,
>>
>>        brw_set_dp_read_message(p,
>>                               insn,
>> -                             255, /* binding table index (255=stateless) */
>> +                              brw_scratch_surface_idx(p),
>>                               msg_control,
>>                               BRW_DATAPORT_READ_MESSAGE_OWORD_BLOCK_READ, /* msg_type */
>>                               BRW_DATAPORT_READ_TARGET_RENDER_CACHE,
>> diff --git a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
>> index 20107ac..65b4358 100644
>> --- a/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
>> +++ b/src/mesa/drivers/dri/i965/brw_vec4_generator.cpp
>> @@ -800,7 +800,7 @@ generate_scratch_read(struct brw_codegen *p,
>>     if (devinfo->gen < 6)
>>        brw_inst_set_cond_modifier(devinfo, send, inst->base_mrf);
>>     brw_set_dp_read_message(p, send,
>> -                          255, /* binding table index: stateless access */
>> +                           brw_scratch_surface_idx(p),
>>                            BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
>>                            msg_type,
>>                            BRW_DATAPORT_READ_TARGET_RENDER_CACHE,
>> @@ -873,7 +873,7 @@ generate_scratch_write(struct brw_codegen *p,
>>     if (devinfo->gen < 6)
>>        brw_inst_set_cond_modifier(p->devinfo, send, inst->base_mrf);
>>     brw_set_dp_write_message(p, send,
>> -                           255, /* binding table index: stateless access */
>> +                            brw_scratch_surface_idx(p),
>>                             BRW_DATAPORT_OWORD_DUAL_BLOCK_1OWORD,
>>                             msg_type,
>>                             3, /* mlen */
>> --
>> 2.5.1
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 212 bytes
Desc: not available
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20151126/e8f894c2/attachment.sig>