[Mesa-dev] [Mesa-users] Problem with ARB_copy_buffer on Mesa 9.2.4

[Ian Romanick]

Carl (the stable branch maintainer) doesn't have plans to do another
10.0.x release.  That said, if the fix is found (by bisecting), it might
still get picked to the branch.  If Fedora is shipping 9.2 or 10.0, they
might get the update.  Dunno.

On 05/06/2014 11:01 AM, Benjamin Bellec wrote:
> Hello Jonas,
> 
> I tested your program and on my system (Fedora 19) on an Evergreen
> (Radeon HD 5850) and I have the same issue indeed.
> Here is my result :
> 
> Mesa 9.2.4 (from F19 repo) => Data does NOT match up!
> Mesa 9.2.5 => Data does NOT match up!
> Mesa 10.0.5 => Data does NOT match up!
> Mesa 10.1.0 => Data matches.
> 
> So this is fixed in newer version.
> That said, Mesa 9.2 is not supported anymore and I really don't know if
> there will be a new Mesa 10.0.x release given the imminence of Mesa 10.2.
> If yes, I can bisect and you can open a bug.
> 
> Mesa-dev, any new 10.0.x release planned ?
> 
> Regards.
> 
> Benjamin
> 
> Le 06/05/2014 13:40, Jonas Kulla a écrit :
>> Hello list,
>>
>> after about 3 days of debugging, I was able to isolate a rather weird
>> behavior in Mesa GL.
>> The gist of it is the following: When I create a buffer object and
>> allocate uninitilaized
>> memory for it (glBufferData() with nullptr as 'data'),
>> then glCopyBufferSubData() data into
>> it from another buffer object, then subsequently fill a part of it
>> with glBufferSubData(),
>> this new data isn't visible to the buffer object. In fact, it seems
>> that the SubData'ed bytes
>> are completely lost. Any further data uploads however work as
>> expected. I will attach
>> a small C test case below that demonstrates this behavior.
>>
>> I realize that I am working with an old Mesa release (on Fedora 19),
>> but I'm afraid of
>> upgrading my system to the newest distro release as I might break my
>> working environment.
>> That's why I would like to kindly ask if someone could verify that
>> this problem still persists
>> on the newest Mesa code, in which case I would go ahead and file a bug
>> report. At the
>> same time, maybe someone could spot a critical mistake in my code that
>> would explain
>> this strange behavior I'm seeing. I think the code paths I'm hitting
>> here in the driver are
>> sufficiently obscure though.
>>
>> I should probably mention that my card is a Mobility Radeon HD 3650
>> (ie. r600).
>>
>> Here's the code sample (you can replace the GL setup code with your own):
>>
>> #include <stdlib.h>
>> #include <string.h>
>>  
>> #include <GL/glew.h>
>> #include <SDL2/SDL.h>
>>  
>> static SDL_Window *win;
>> static SDL_GLContext *ctx;
>>  
>> void setupGL()
>> {
>>         SDL_Init(SDL_INIT_VIDEO);
>>         win = SDL_CreateWindow("CopyBufferBug",
>> SDL_WINDOWPOS_UNDEFINED, SDL_WINDOWPOS_UNDEFINED, 64, 64,
>> SDL_WINDOW_OPENGL);
>>         ctx = SDL_GL_CreateContext(win);
>>         glewInit();
>> }
>>  
>> static void teardownGL()
>> {
>>         SDL_GL_DeleteContext(ctx);
>>         SDL_DestroyWindow(win);
>>  
>>         SDL_Quit();
>> }
>>  
>> int main(int argc, char *argv[])
>> {
>>         setupGL();
>>  
>>         /* These don't matter I think */
>>         #define BLOCK_SIZE 128
>>         #define BUFFER1_SIZE BLOCK_SIZE
>>         #define BUFFER2_SIZE BLOCK_SIZE
>>         #define BUFFER1_TARGET GL_COPY_READ_BUFFER
>>         #define BUFFER2_TARGET GL_COPY_WRITE_BUFFER
>>         #define BUFFER1_USAGE GL_DYNAMIC_DRAW
>>         #define BUFFER2_USAGE GL_DYNAMIC_DRAW
>>  
>>         GLuint buffers[2];
>>         glGenBuffers(2, buffers);
>>  
>>         /* We allocate both buffers with undefined memory */
>>         glBindBuffer(BUFFER1_TARGET, buffers[0]);
>>         glBufferData(BUFFER1_TARGET, BUFFER1_SIZE, 0, BUFFER1_USAGE);
>>  
>>         glBindBuffer(BUFFER2_TARGET, buffers[1]);
>>         glBufferData(BUFFER2_TARGET, BUFFER2_SIZE, 0, BUFFER2_USAGE);
>>  
>>         /* Then copy (undefined) bytes from the first into the second
>> buffer */
>>         /* Note: If I comment this line out, everything works */
>>         glCopyBufferSubData(BUFFER1_TARGET, BUFFER2_TARGET, 0, 0,
>> BUFFER1_SIZE);
>>  
>>         /* Generate random string */
>>         FILE *rand = fopen("/dev/urandom", "r");
>>         char data[BLOCK_SIZE];
>>         fread(data, 1, sizeof(data), rand);
>>         fclose(rand);
>>  
>>         /* We fill the second buffer with defined data */
>>         /* Note: If I execute this call twice (just copy paste the
>> line), everything works */
>>         glBufferSubData(BUFFER2_TARGET, 0, sizeof(data), data);
>>  
>>         /* Then download it again to compare its contents against our
>> test string */
>>         char data2[BLOCK_SIZE];
>>         glGetBufferSubData(BUFFER2_TARGET, 0, sizeof(data2), data2);
>>  
>>         if (memcmp(data, data2, sizeof(data)))
>>                 printf("Data does NOT match up!\n");
>>         else
>>                 printf("Data matches.\n");
>>  
>>         glDeleteBuffers(2, buffers);
>>  
>>         teardownGL();
>>  
>>         return 0;
>> }
>>
>> Thank you very much for your time.
>> Jonas
>>
>>
>> _______________________________________________
>> mesa-users mailing list
>> mesa-users at lists.freedesktop.org
>> http://lists.freedesktop.org/mailman/listinfo/mesa-users
> 
> 
> 
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
>