[Mesa-dev] [PATCH 3/3] Revert "i965: Disable regular fast-clears (CCS_D) on gen9+"

Jason Ekstrand jason at jlekstrand.net
Sat Dec 16 23:07:01 UTC 2017


On December 16, 2017 14:35:29 Nanley Chery <nanleychery at gmail.com> wrote:

> On Wed, Dec 13, 2017 at 05:52:03PM -0800, Jason Ekstrand wrote:
>> This reverts commit ee57b15ec764736e2d5360beaef9fb2045ed0f68.
>>
>> Cc: "17.3" <mesa-stable at lists.freedesktop.org>
>> ---
>>  src/mesa/drivers/dri/i965/brw_meta_util.c     | 10 -----
>>  src/mesa/drivers/dri/i965/intel_mipmap_tree.c | 57 ++++++++++++---------------
>>  2 files changed, 25 insertions(+), 42 deletions(-)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_meta_util.c 
>> b/src/mesa/drivers/dri/i965/brw_meta_util.c
>> index 54dc6a5..b311815 100644
>> --- a/src/mesa/drivers/dri/i965/brw_meta_util.c
>> +++ b/src/mesa/drivers/dri/i965/brw_meta_util.c
>> @@ -293,17 +293,7 @@ brw_is_color_fast_clear_compatible(struct brw_context 
>> *brw,
>>         brw->mesa_to_isl_render_format[mt->format])
>>        return false;
>>
>> -   /* Gen9 doesn't support fast clear on single-sampled SRGB buffers. When
>> -    * GL_FRAMEBUFFER_SRGB is enabled any color renderbuffers will be
>> -    * resolved in intel_update_state. In that case it's pointless to do a
>> -    * fast clear because it's very likely to be immediately resolved.
>> -    */
>>     const bool srgb_rb = _mesa_get_srgb_format_linear(mt->format) != mt->format;
>> -   if (devinfo->gen >= 9 &&
>> -       mt->surf.samples == 1 &&
>> -       ctx->Color.sRGBEnabled && srgb_rb)
>> -      return false;
>> -
>>    /* Gen10 doesn't automatically decode the clear color of sRGB buffers. Since
>>     * we currently don't perform this decode in software, avoid a fast-clear
>>     * altogether. TODO: Do this in software.
>> diff --git a/src/mesa/drivers/dri/i965/intel_mipmap_tree.c 
>> b/src/mesa/drivers/dri/i965/intel_mipmap_tree.c
>> index c1a4ce1..b87d356 100644
>> --- a/src/mesa/drivers/dri/i965/intel_mipmap_tree.c
>> +++ b/src/mesa/drivers/dri/i965/intel_mipmap_tree.c
>> @@ -207,13 +207,7 @@ intel_miptree_supports_ccs(struct brw_context *brw,
>>     if (!brw->mesa_format_supports_render[mt->format])
>>        return false;
>>
>> -   if (devinfo->gen >= 9) {
>> -      mesa_format linear_format = _mesa_get_srgb_format_linear(mt->format);
>> -      const enum isl_format isl_format =
>> -         brw_isl_format_for_mesa_format(linear_format);
>> -      return isl_format_supports_ccs_e(&brw->screen->devinfo, isl_format);
>> -   } else
>> -      return true;
>> +   return true;
>>  }
>>
>>  static bool
>> @@ -256,7 +250,7 @@ intel_miptree_supports_hiz(const struct brw_context *brw,
>>   * our HW tends to support more linear formats than sRGB ones, we use this
>>   * format variant for check for CCS_E compatibility.
>>   */
>> -MAYBE_UNUSED static bool
>> +static bool
>>  format_ccs_e_compat_with_miptree(const struct gen_device_info *devinfo,
>>                                   const struct intel_mipmap_tree *mt,
>>                                   enum isl_format access_format)
>> @@ -290,12 +284,13 @@ intel_miptree_supports_ccs_e(struct brw_context *brw,
>>     if (!intel_miptree_supports_ccs(brw, mt))
>>        return false;
>>
>> -   /* Fast clear can be also used to clear srgb surfaces by using equivalent
>> -    * linear format. This trick, however, can't be extended to be used with
>> -    * lossless compression and therefore a check is needed to see if the 
>> format
>> -    * really is linear.
>> +   /* Many window system buffers are sRGB even if they are never rendered as
>> +    * sRGB.  For those, we want CCS_E for when sRGBEncode is false.  When the
>> +    * surface is used as sRGB, we fall back to CCS_D.
>>      */
>> -   return _mesa_get_srgb_format_linear(mt->format) == mt->format;
>> +   mesa_format linear_format = _mesa_get_srgb_format_linear(mt->format);
>> +   enum isl_format isl_format = brw_isl_format_for_mesa_format(linear_format);
>> +   return isl_format_supports_ccs_e(&brw->screen->devinfo, isl_format);
>>  }
>>
>>  /**
>> @@ -2686,29 +2681,27 @@ intel_miptree_render_aux_usage(struct brw_context *brw,
>>        return ISL_AUX_USAGE_MCS;
>>
>>     case ISL_AUX_USAGE_CCS_D:
>> -      /* If FRAMEBUFFER_SRGB is used on Gen9+ then we need to resolve any of
>> -       * the single-sampled color renderbuffers because the CCS buffer isn't
>> -       * supported for SRGB formats. This only matters if FRAMEBUFFER_SRGB is
>> -       * enabled because otherwise the surface state will be programmed with
>> -       * the linear equivalent format anyway.
>> -       */
>> -      if (isl_format_is_srgb(render_format) &&
>> -          _mesa_get_srgb_format_linear(mt->format) != mt->format) {
>> -         return ISL_AUX_USAGE_NONE;
>> -      } else if (!mt->mcs_buf) {
>> -         return ISL_AUX_USAGE_NONE;
>> -      } else {
>> -         return ISL_AUX_USAGE_CCS_D;
>> -      }
>> +      return mt->mcs_buf ? ISL_AUX_USAGE_CCS_D : ISL_AUX_USAGE_NONE;
>>
>>     case ISL_AUX_USAGE_CCS_E: {
>> -      /* Lossless compression is not supported for SRGB formats, it
>> -       * should be impossible to get here with such surfaces.
>> +      /* If the format supports CCS_E and is compatible with the miptree,
>> +       * then we can use it.
>>         */
>> -      assert(!isl_format_is_srgb(render_format) ||
>> -             _mesa_get_srgb_format_linear(mt->format) == mt->format);
>> +      if (format_ccs_e_compat_with_miptree(&brw->screen->devinfo,
>> +                                           mt, render_format))
>> +         return ISL_AUX_USAGE_CCS_E;
>> +
>> +      /* Otherwise, we have to fall back to CCS_D */
>> +
>> +      /* gen9 hardware technically supports non-0/1 clear colors with sRGB
>> +       * formats.  However, there are issues with blending where it doesn't
>> +       * properly apply the sRGB curve to the clear color when blending.
>> +       */
>> +      if (blend_enabled && isl_format_is_srgb(render_format) &&
>> +          !isl_color_value_is_zero_one(mt->fast_clear_color, render_format))
>> +         return ISL_AUX_USAGE_NONE;
>>
>
> We also need the blend workaround for the CCS_D case. Otherwise, we'll
> start failing piglit's arb_framebuffer_srgb-fast-clear-blend when
> DEBUG_NO_RBC is set (MD5-287).

Initially, I waz very confused as to why that's in the CCS_E case at all.  
However, as of this patch, mt->aux_usage is always set to CCS_D for any 
8888 surfaces.  That said, it would probably be better to have it in both 
cases.

> -Nanley
>
>> -      return ISL_AUX_USAGE_CCS_E;
>> +      return ISL_AUX_USAGE_CCS_D;
>>     }
>>
>>     default:
>> --
>> 2.5.0.400.gff86faf
>>
>> _______________________________________________
>> mesa-dev mailing list
>> mesa-dev at lists.freedesktop.org
>> https://lists.freedesktop.org/mailman/listinfo/mesa-dev




More information about the mesa-dev mailing list