[Mesa-dev] [PATCH 3/3] i965/screen: Allow modifiers on sRGB formats

[Eero Tamminen]

Hi,

On 29.08.2018 01:22, Jason Ekstrand wrote:
> This effectively reverts a26693493570a9d0f0fba1be617e01ee7bfff4db which
> was a misguided attempt at protecting intel_query_dma_buf_modifiers from
> invalid formats.  Unfortunately, in some internal EGL cases, we can get
> an SRGB format validly in this function.

While in SynMark EGL tests E2E RBC got completely lost, there was some 
perf impact (at most half of E2E RBC impact) also in GLX using onscreen 
versions of GfxBench ALU2, Tessellation and Manhattan 3.0 tests.

This patch series (with the "continue" fix you mentioned in later 
comment) fixes both SynMark & GfxBench regressions.

Tested-By: Eero Tamminen <eero.t.tamminen at intel.com>


	- Eero

> Rejecting such formats caused
> us to not allow CCS in some cases where we should have been allowing it.
> 
> There's some question of whether or not we really should be using SRGB
> "fourcc" formats that aren't actually in drm_foucc.h but there's not
> much harm in allowing them through here.
> 
> Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107223
> Fixes: a26693493570 "i965/screen: Return false for unsupported..."
> ---
>   src/mesa/drivers/dri/i965/intel_screen.c | 14 +++++++++++---
>   1 file changed, 11 insertions(+), 3 deletions(-)
> 
> diff --git a/src/mesa/drivers/dri/i965/intel_screen.c b/src/mesa/drivers/dri/i965/intel_screen.c
> index eaf5a3b9feb..ac1938f6204 100644
> --- a/src/mesa/drivers/dri/i965/intel_screen.c
> +++ b/src/mesa/drivers/dri/i965/intel_screen.c
> @@ -1274,9 +1274,9 @@ static bool
>   intel_image_format_is_supported(const struct gen_device_info *devinfo,
>                                   const struct intel_image_format *fmt)
>   {
> -   if (fmt->fourcc == __DRI_IMAGE_FOURCC_SARGB8888 ||
> -       fmt->fourcc == __DRI_IMAGE_FOURCC_SABGR8888)
> -      return false;
> +   /* Currently, all formats with an intel_image_format are available on all
> +    * platforms so there's really nothing to check there.
> +    */
>   
>   #ifndef NDEBUG
>      if (fmt->nplanes == 1) {
> @@ -1302,6 +1302,14 @@ intel_query_dma_buf_formats(__DRIscreen *_screen, int max,
>      int num_formats = 0, i;
>   
>      for (i = 0; i < ARRAY_SIZE(intel_image_formats); i++) {
> +      /* These two formats are valid DRI formats but do not exist in
> +       * drm_fourcc.h in the Linux kernel.  We don't want to accidentally
> +       * advertise them through the EGL layer.
> +       */
> +      if (intel_image_formats[i].fourcc == __DRI_IMAGE_FOURCC_SARGB8888 ||
> +          intel_image_formats[i].fourcc == __DRI_IMAGE_FOURCC_SABGR8888)
> +         return false;
> +
>         if (!intel_image_format_is_supported(&screen->devinfo,
>                                              &intel_image_formats[i]))
>            continue;
>