vo_gpu: stop hard-coding max compute group threads

We've been assuming that maximum number of compute group threads is
never less than the 1024 defined by the desktop GL spec. Given that we
haven't had working compute shaders for GLES and I guess the Vulkan
spec defines at least as high a value, we've gotten away with it so
far.

But we should really look the value up and respect it.

video/out/d3d11/ra_d3d11.c

@@ -2386,6 +2386,8 @@ struct ra *ra_d3d11_create(ID3D11Device *dev, struct mp_log *log,
     if (p->fl >= D3D_FEATURE_LEVEL_11_0) {
         ra->caps |= RA_CAP_COMPUTE | RA_CAP_BUF_RW;
         ra->max_shmem = 32 * 1024;
+        ra->max_compute_group_threads =
+            D3D11_CS_THREAD_GROUP_MAX_THREADS_PER_GROUP;
     }
 
     if (p->fl >= D3D_FEATURE_LEVEL_11_1) {

video/out/gpu/ra.h

@@ -26,6 +26,10 @@ struct ra {
     // time.
     size_t max_shmem;
 
+    // Maximum number of threads in a compute work group. Set by the RA backend
+    // at init time.
+    size_t max_compute_group_threads;
+
     // Maximum push constant size. Set by the RA backend at init time.
     size_t max_pushc_size;
 

video/out/gpu/video.c

@@ -2698,12 +2698,9 @@ static void pass_dither(struct gl_video *p)
 
             struct image img = image_wrap(p->error_diffusion_tex[0], PLANE_RGB, p->components);
 
-            // Ensure the block size doesn't exceed the minimum defined by the
-            // specification (1024 in desktop GL, 128 in GLES).
-            // TODO: Look up the actual maximum block size for the
-            // implementation using:
-            //     glGetIntegerv(MAX_COMPUTE_WORK_GROUP_INVOCATIONS, &value);
-            int block_size = MPMIN(p->ra->glsl_es ? 128 : 1024, o_h);
+            // Ensure the block size doesn't exceed the maximum of the
+            // implementation.
+            int block_size = MPMIN(p->ra->max_compute_group_threads, o_h);
 
             pass_describe(p, "dither=error-diffusion (kernel=%s, depth=%d)",
                              kernel->name, dst_depth);

video/out/opengl/gl_headers.h

@@ -138,6 +138,7 @@ typedef uint64_t GLuint64;
 
 #define GL_COMPUTE_SHADER                 0x91B9
 #define GL_MAX_COMPUTE_SHARED_MEMORY_SIZE 0x8262
+#define GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS 0x90EB
 
 // --- GL 4.3 or GL_ARB_shader_storage_buffer_object
 

video/out/opengl/ra_gl.c

@@ -224,6 +224,8 @@ static int ra_init_gl(struct ra *ra, GL *gl)
     if (ra->caps & RA_CAP_COMPUTE) {
         gl->GetIntegerv(GL_MAX_COMPUTE_SHARED_MEMORY_SIZE, &ival);
         ra->max_shmem = ival;
+        gl->GetIntegerv(GL_MAX_COMPUTE_WORK_GROUP_INVOCATIONS, &ival);
+        ra->max_compute_group_threads = ival;
     }
 
     gl->Disable(GL_DITHER);

video/out/placebo/ra_pl.c

@@ -67,6 +67,7 @@ struct ra *ra_create_pl(const struct pl_gpu *gpu, struct mp_log *log)
     ra->max_texture_wh = gpu->limits.max_tex_2d_dim;
     ra->max_shmem = gpu->limits.max_shmem_size;
     ra->max_pushc_size = gpu->limits.max_pushc_size;
+    ra->max_compute_group_threads = gpu->limits.max_group_threads;
 
     // Set up format wrappers
     for (int i = 0; i < gpu->num_formats; i++) {