[Mesa-dev] [PATCH 4/5] egl/wayland: Allow client->server format conversion for PRIME offload.

Mario Kleiner mario.kleiner.de at gmail.com
Wed May 23 04:51:12 UTC 2018


On Mon, May 21, 2018 at 4:42 PM, Eric Engestrom
<eric.engestrom at intel.com> wrote:
> On Saturday, 2018-05-19 05:32:41 +0200, Mario Kleiner wrote:
>> Support PRIME render offload between a Wayland server gpu and a Wayland
>> client gpu with different channel ordering for their color formats,
>> e.g., between Intel drivers which currently only support ARGB2101010
>> and XRGB2101010 import/display and nouveau which only supports ABGR2101010
>> rendering and display on nv-50 and later.
>>
>> In the wl_visuals table, we also store for each format an alternate
>> sibling format which stores colors at the same precision, but with
>> different channel ordering, e.g., ARGB2101010 <-> ABGR2101010.
>>
>> If a given client-gpu renderable format is not supported by the server
>> for import, but the alternate format is supported by the server, expose
>> the client-gpu renderable format as a valid EGLConfig to the client. At
>> eglSwapBuffers time, during the blitImage() detiling blit from the client
>> backbuffer to the linear buffer, the client format is converted to the
>> server supported format. As we have to do a copy for PRIME anyway,
>> this channel swizzling conversion comes essentially for free.
>>
>> Note that even if a server gpu in principle does support sampling
>> from the clients native format, this conversion will be a performance
>> advantage if it allows to convert to the servers preferred format
>> for direct scanout, as the Wayland compositor may then be able to
>> directly page-flip a fullscreen client wl_buffer onto the primary
>> plane, or onto a hardware overlay plane, avoiding an extra data copy
>> for desktop composition.
>>
>> Tested so far under Weston with: nouveau single-gpu, Intel single-gpu,
>> AMD single-gpu, "Optimus" Intel server iGPU for display + NVidia
>> client dGPU for rendering.
>>
>> Signed-off-by: Mario Kleiner <mario.kleiner.de at gmail.com>
>> Cc: Daniel Stone <daniels at collabora.com>
>> ---
>>  src/egl/drivers/dri2/platform_wayland.c | 67 ++++++++++++++++++++++++++++-----
>>  1 file changed, 58 insertions(+), 9 deletions(-)
>>
>> diff --git a/src/egl/drivers/dri2/platform_wayland.c b/src/egl/drivers/dri2/platform_wayland.c
>> index 89a7f90118..fb364a6233 100644
>> --- a/src/egl/drivers/dri2/platform_wayland.c
>> +++ b/src/egl/drivers/dri2/platform_wayland.c
>> @@ -68,49 +68,50 @@ static const struct dri2_wl_visual {
>>     uint32_t wl_drm_format;
>>     uint32_t wl_shm_format;
>>     int dri_image_format;
>> +   int alt_dri_image_format;
>

Thanks for the review.

> A comment here wouldn't hurt, to explain what this "alt" is to someone
> who sees the code after it landed :)
>

Will do, although i usually feel a bit guilty that my patches usually
suffer from over-commenting and essay length commit messages :)

>>     int bpp;
>>     unsigned int rgba_masks[4];
>>  } dri2_wl_visuals[] = {
>>     {
>>       "XRGB2101010",
>>       WL_DRM_FORMAT_XRGB2101010, WL_SHM_FORMAT_XRGB2101010,
>> -     __DRI_IMAGE_FORMAT_XRGB2101010, 32,
>> +     __DRI_IMAGE_FORMAT_XRGB2101010, __DRI_IMAGE_FORMAT_XBGR2101010, 32,
>>       { 0x3ff00000, 0x000ffc00, 0x000003ff, 0x00000000 }
>>     },
>>     {
>>       "ARGB2101010",
>>       WL_DRM_FORMAT_ARGB2101010, WL_SHM_FORMAT_ARGB2101010,
>> -     __DRI_IMAGE_FORMAT_ARGB2101010, 32,
>> +     __DRI_IMAGE_FORMAT_ARGB2101010, __DRI_IMAGE_FORMAT_ABGR2101010, 32,
>>       { 0x3ff00000, 0x000ffc00, 0x000003ff, 0xc0000000 }
>>     },
>>     {
>>       "XBGR2101010",
>>       WL_DRM_FORMAT_XBGR2101010, WL_SHM_FORMAT_XBGR2101010,
>> -     __DRI_IMAGE_FORMAT_XBGR2101010, 32,
>> +     __DRI_IMAGE_FORMAT_XBGR2101010, __DRI_IMAGE_FORMAT_XRGB2101010, 32,
>>       { 0x000003ff, 0x000ffc00, 0x3ff00000, 0x00000000 }
>>     },
>>     {
>>       "ABGR2101010",
>>       WL_DRM_FORMAT_ABGR2101010, WL_SHM_FORMAT_ABGR2101010,
>> -     __DRI_IMAGE_FORMAT_ABGR2101010, 32,
>> +     __DRI_IMAGE_FORMAT_ABGR2101010, __DRI_IMAGE_FORMAT_ARGB2101010, 32,
>>       { 0x000003ff, 0x000ffc00, 0x3ff00000, 0xc0000000 }
>>     },
>>     {
>>       "XRGB8888",
>>       WL_DRM_FORMAT_XRGB8888, WL_SHM_FORMAT_XRGB8888,
>> -     __DRI_IMAGE_FORMAT_XRGB8888, 32,
>> +     __DRI_IMAGE_FORMAT_XRGB8888, __DRI_IMAGE_FORMAT_NONE, 32,
>>       { 0x00ff0000, 0x0000ff00, 0x000000ff, 0x00000000 }
>>     },
>>     {
>>       "ARGB8888",
>>       WL_DRM_FORMAT_ARGB8888, WL_SHM_FORMAT_ARGB8888,
>> -     __DRI_IMAGE_FORMAT_ARGB8888, 32,
>> +     __DRI_IMAGE_FORMAT_ARGB8888, __DRI_IMAGE_FORMAT_NONE, 32,
>>       { 0x00ff0000, 0x0000ff00, 0x000000ff, 0xff000000 }
>>     },
>>     {
>>       "RGB565",
>>       WL_DRM_FORMAT_RGB565, WL_SHM_FORMAT_RGB565,
>> -     __DRI_IMAGE_FORMAT_RGB565, 16,
>> +     __DRI_IMAGE_FORMAT_RGB565, __DRI_IMAGE_FORMAT_NONE, 16,
>>       { 0xf800, 0x07e0, 0x001f, 0x0000 }
>>     },
>>  };
>> @@ -450,15 +451,23 @@ get_back_bo(struct dri2_egl_surface *dri2_surf)
>>     int use_flags;
>>     int visual_idx;
>>     unsigned int dri_image_format;
>> +   unsigned int linear_dri_image_format;
>>     uint64_t *modifiers;
>>     int num_modifiers;
>>
>>     visual_idx = dri2_wl_visual_idx_from_fourcc(dri2_surf->format);
>>     assert(visual_idx != -1);
>>     dri_image_format = dri2_wl_visuals[visual_idx].dri_image_format;
>> +   linear_dri_image_format = dri_image_format;
>>     modifiers = u_vector_tail(&dri2_dpy->wl_modifiers[visual_idx]);
>>     num_modifiers = u_vector_length(&dri2_dpy->wl_modifiers[visual_idx]);
>>
>> +   /* Substitute dri image format if server does not support original format */
>> +   if (!(dri2_dpy->formats & (1 << visual_idx)))
>> +      linear_dri_image_format = dri2_wl_visuals[visual_idx].alt_dri_image_format;
>
> Could we test that the substitution would actually work better before
> doing it?
>

We test that, but already in dri2_wl_add_configs_for_visuals() below.
The only way to get to the if statement (ie. visual_idx != -1) and
then have it trigger the linear_dri_image_format substitution is if
the new hunk of code in dri2_wl_add_configs_for_visuals() found out
that the substitution is necessary, possible and valid. A EGLconfig
can support the client format with the given visual_idx if either the
wayland-server supports it (so the above if statement will skip the
substitution), or if the server doesn't support it, but does support
sampling/displaying the substituted format, according to the checks
done in dri2_wl_add_configs_for_visuals before adding the EGLconfig to
the list of client configs.

I could add a test to document this, which would be redundant though.
Or maybe a code comment? Or maybe the test in an assert() to document
that that test should always succeed in the absence of bugs in the
code?

>> +
>> +   assert(linear_dri_image_format != __DRI_IMAGE_FORMAT_NONE);
>> +
>>     /* There might be a buffer release already queued that wasn't processed */
>>     wl_display_dispatch_queue_pending(dri2_dpy->wl_dpy, dri2_surf->wl_queue);
>>
>> @@ -505,7 +514,7 @@ get_back_bo(struct dri2_egl_surface *dri2_surf)
>>              dri2_dpy->image->createImageWithModifiers(dri2_dpy->dri_screen,
>>                                                        dri2_surf->base.Width,
>>                                                        dri2_surf->base.Height,
>> -                                                      dri_image_format,
>> +                                                      linear_dri_image_format,
>>                                                        &linear_mod,
>>                                                        1,
>>                                                        NULL);
>> @@ -514,7 +523,7 @@ get_back_bo(struct dri2_egl_surface *dri2_surf)
>>              dri2_dpy->image->createImage(dri2_dpy->dri_screen,
>>                                           dri2_surf->base.Width,
>>                                           dri2_surf->base.Height,
>> -                                         dri_image_format,
>> +                                         linear_dri_image_format,
>>                                           use_flags |
>>                                           __DRI_IMAGE_USE_LINEAR,
>>                                           NULL);
>> @@ -1278,8 +1287,11 @@ dri2_wl_add_configs_for_visuals(_EGLDriver *drv, _EGLDisplay *disp)
>>     struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
>>     unsigned int format_count[ARRAY_SIZE(dri2_wl_visuals)] = { 0 };
>>     unsigned int count = 0;
>> +   bool assigned;
>>
>>     for (unsigned i = 0; dri2_dpy->driver_configs[i]; i++) {
>> +      assigned = false;
>> +
>>        for (unsigned j = 0; j < ARRAY_SIZE(dri2_wl_visuals); j++) {
>>           struct dri2_egl_config *dri2_conf;
>>
>> @@ -1292,6 +1304,43 @@ dri2_wl_add_configs_for_visuals(_EGLDriver *drv, _EGLDisplay *disp)
>>              if (dri2_conf->base.ConfigID == count + 1)
>>                 count++;
>>              format_count[j]++;
>> +            assigned = true;
>> +         }
>> +      }
>> +
>> +      if (!assigned && dri2_dpy->is_different_gpu) {
>> +         struct dri2_egl_config *dri2_conf;
>> +         int alt_dri_image_format, j, k;
>
> Let's avoid reusing the same names as the loop variables above?

Okay, c for client visual index and s for substitute visual index are
still free as short names.

thanks,
-mario

>
>> +
>> +         /* No match for config. Try if we can blitImage convert to a visual */
>> +         j = dri2_wl_visual_idx_from_config(dri2_dpy,
>> +                                            dri2_dpy->driver_configs[i]);
>> +
>> +         if (j == -1)
>> +            continue;
>> +
>> +         /* Find optimal target visual for blitImage conversion, if any. */
>> +         alt_dri_image_format = dri2_wl_visuals[j].alt_dri_image_format;
>> +         k = dri2_wl_visual_idx_from_dri_image_format(alt_dri_image_format);
>> +
>> +         if (k == -1 || !(dri2_dpy->formats & (1 << k)))
>> +            continue;
>> +
>> +         /* Visual k works for the Wayland server, and j can be converted into k
>> +          * by our client gpu during PRIME blitImage conversion to a linear
>> +          * wl_buffer, so add visual j as supported by the client renderer.
>> +          */
>> +         dri2_conf = dri2_add_config(disp, dri2_dpy->driver_configs[i],
>> +                                     count + 1, EGL_WINDOW_BIT, NULL,
>> +                                     dri2_wl_visuals[j].rgba_masks);
>> +         if (dri2_conf) {
>> +            if (dri2_conf->base.ConfigID == count + 1)
>> +               count++;
>> +            format_count[j]++;
>> +            if (format_count[j] == 1)
>> +               _eglLog(_EGL_DEBUG, "Client format %s to server format %s via "
>> +                       "PRIME blitImage.", dri2_wl_visuals[j].format_name,
>> +                       dri2_wl_visuals[k].format_name);
>>           }
>>        }
>>     }
>> --
>> 2.13.0.rc1.294.g07d810a77f
>>
>> _______________________________________________
>> 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