vaDeriveImage returns VA_STATUS_ERROR_OPERATION_FAILED

[joeyes86]

It's a kind of weird situation. I wrote small sample code

#include <va/va.h>
#include <va/va_x11.h>
#include <stdio.h>

void main()
{   
    int major_ver;
    int minor_ver;

    Display* x11_display = XOpenDisplay(NULL);
    VADisplay va_dpy = vaGetDisplay(x11_display);

    VAStatus va_status = vaInitialize(va_dpy, &major_ver, &minor_ver);

    VASurfaceID surfaces[4];
    vaCreateSurfaces(va_dpy, VA_RT_FORMAT_YUV420, 1920, 1088, &surfaces[0], 4, NULL, 0);
    for (int i = 0; i < 4; i++) {
        VAImage surface_image;
        VAStatus va_status = vaDeriveImage(va_dpy, surfaces[i], &surface_image);
        printf("surface %d: vaDeriveImage returns %d\n", i, va_status);
    }
}

build command is gcc h264encode.c -lva -lva-x11 -lX11

Output:

libva info: VA-API version 1.14.0
libva info: Trying to open /usr/lib/x86_64-linux-gnu/dri/radeonsi_drv_video.so
libva info: Found init function __vaDriverInit_1_14
libva info: va_openDriver() returns 0
surface 0: vaDeriveImage returns 1
surface 1: vaDeriveImage returns 1
surface 2: vaDeriveImage returns 1
surface 3: vaDeriveImage returns 1

The same code at the very beginning of the https://github.com/intel/libva-utils/blob/master/encode/h264encode.c main fuction works fine and vaDeriveImage returns 0.
What I'm missing for?

[joeyes86]

I trying it on machine with AMD Ryzen 5 5300u

[vitaly-castLabs]

Apparently it's a commonly appearing issue AMD devs trying to fix for years, here are the latest reports (even though it was supposed to work since Mesa 20.3.0 released two years ago): https://gitlab.freedesktop.org/mesa/mesa/-/issues/1193