[Mesa-dev] [PATCH 51/53] intel/fs: Remove support push constants in repclear shaders

[Jason Ekstrand]

On Thu, May 31, 2018 at 11:28 AM, Jason Ekstrand <jason at jlekstrand.net>
wrote:

> On Thu, May 31, 2018 at 10:43 AM, Matt Turner <mattst88 at gmail.com> wrote:
>
>> On Thu, May 24, 2018 at 2:56 PM, Jason Ekstrand <jason at jlekstrand.net>
>> wrote:
>> > ---
>> >  src/intel/compiler/brw_fs.cpp | 25 ++++++-------------------
>> >  1 file changed, 6 insertions(+), 19 deletions(-)
>> >
>> > diff --git a/src/intel/compiler/brw_fs.cpp
>> b/src/intel/compiler/brw_fs.cpp
>> > index 1f87f06..7e532af 100644
>> > --- a/src/intel/compiler/brw_fs.cpp
>> > +++ b/src/intel/compiler/brw_fs.cpp
>> > @@ -3220,21 +3220,14 @@ fs_visitor::emit_repclear_shader()
>> >     brw_wm_prog_key *key = (brw_wm_prog_key*) this->key;
>> >     int base_mrf = 0;
>> >     int color_mrf = base_mrf + 2;
>> > -   fs_inst *mov;
>> >
>> > -   if (uniforms > 0) {
>> > -      mov = bld.exec_all().group(4, 0)
>> > -               .MOV(brw_message_reg(color_mrf),
>> > -                    fs_reg(UNIFORM, 0, BRW_REGISTER_TYPE_F));
>> > -   } else {
>> > -      struct brw_reg reg =
>> > -         brw_reg(BRW_GENERAL_REGISTER_FILE, 2, 3, 0, 0,
>> BRW_REGISTER_TYPE_F,
>> > -                 BRW_VERTICAL_STRIDE_8, BRW_WIDTH_2,
>> BRW_HORIZONTAL_STRIDE_4,
>> > -                 BRW_SWIZZLE_XYZW, WRITEMASK_XYZW);
>> > +   const struct brw_reg color_in_reg =
>> > +      brw_reg(BRW_GENERAL_REGISTER_FILE, 2, 3, 0, 0,
>> BRW_REGISTER_TYPE_F,
>> > +              BRW_VERTICAL_STRIDE_8, BRW_WIDTH_2,
>> BRW_HORIZONTAL_STRIDE_4,
>> > +              BRW_SWIZZLE_XYZW, WRITEMASK_XYZW);
>> >
>> > -      mov = bld.exec_all().group(4, 0)
>> > -               .MOV(vec4(brw_message_reg(color_mrf)), fs_reg(reg));
>> > -   }
>> > +   bld.exec_all().group(4, 0)
>> > +      .MOV(vec4(brw_message_reg(color_mrf)), fs_reg(color_in_reg));
>> >
>> >     fs_inst *write = NULL;
>> >     if (key->nr_color_regions == 1) {
>> > @@ -3273,12 +3266,6 @@ fs_visitor::emit_repclear_shader()
>> >
>> >     assign_constant_locations();
>> >     assign_curb_setup();
>> > -
>> > -   /* Now that we have the uniform assigned, go ahead and force it to
>> a vec4. */
>> > -   if (uniforms > 0) {
>> > -      assert(mov->src[0].file == FIXED_GRF);
>> > -      mov->src[0] = brw_vec4_grf(mov->src[0].nr, 0);
>> > -   }
>> >  }
>> >
>> >  /**
>> > --
>>
>> Some explanation required: why are we making this change, how is it
>> safe (i.e., did something earlier in the series make this possible?).
>>
>
> It hasn't been used since
>
> commit 46e1132b80f01f4670dac4e393d36b0e9266b02f
> Author: Topi Pohjolainen <topi.pohjolainen at intel.com>
> Date:   Sun May 15 07:43:39 2016 +0300
>
>     i965/blorp: Use flat inputs instead of uniforms
>
>     v2 (Jason): Use LOAD_INPUT() macro
>
>     Signed-off-by: Topi Pohjolainen <topi.pohjolainen at intel.com>
>     Reviewed-by: Jason Ekstrand <jason at jlekstrand.net>
>
>
That said, SIMD32 with repclear has shown itself to be a bit more annoying
than first thought so I think it's probably best to drop 51 and 52.  They
were added somewhat last-minute.

--Jason
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20180531/61ec1716/attachment.html>