Compute shader: nvidia: render full image (incorrect dispatch size)

1 files changed, 6 insertions, 7 deletions

diff --git a/src/color_conversion.c b/src/color_conversion.c
index 58c9661..a54407b 100644
--- a/src/color_conversion.c
+++ b/src/color_conversion.c
@@ -251,7 +251,6 @@ int gsr_color_conversion_init(gsr_color_conversion *self, const gsr_color_conver
     int max_compute_work_group_invocations = 256;
     self->params.egl->glGetIntegerv(GL_MAX_COMPUTE_FIXED_GROUP_INVOCATIONS, &max_compute_work_group_invocations);
     self->max_local_size_dim = sqrt(max_compute_work_group_invocations);
-    fprintf(stderr, "max local size: %d, max_local_size_dim: %d\n", max_compute_work_group_invocations, self->max_local_size_dim);
 
     switch(params->destination_color) {
         case GSR_DESTINATION_COLOR_NV12:
@@ -435,8 +434,8 @@ void gsr_color_conversion_draw(gsr_color_conversion *self, unsigned int texture_
                 self->params.egl->glUniform2i(uniform->target_position, destination_pos.x, destination_pos.y);
                 self->params.egl->glUniform2f(uniform->scale, scale.x, scale.y);
                 self->params.egl->glBindImageTexture(0, self->params.destination_textures[0], 0, GL_FALSE, 0, GL_WRITE_ONLY, use_16bit_colors ? GL_R16 : GL_R8);
-                const double num_groups_x = (double)texture_size.x/(double)self->max_local_size_dim + 0.5;
-                const double num_groups_y = (double)texture_size.y/(double)self->max_local_size_dim + 0.5;
+                const double num_groups_x = ceil((double)texture_size.x/(double)self->max_local_size_dim);
+                const double num_groups_y = ceil((double)texture_size.y/(double)self->max_local_size_dim);
                 self->params.egl->glDispatchCompute(max_int(1, num_groups_x), max_int(1, num_groups_y), 1);
             }
 
@@ -453,8 +452,8 @@ void gsr_color_conversion_draw(gsr_color_conversion *self, unsigned int texture_
                 self->params.egl->glUniform2i(uniform->target_position, destination_pos.x, destination_pos.y);
                 self->params.egl->glUniform2f(uniform->scale, scale.x, scale.y);
                 self->params.egl->glBindImageTexture(0, self->params.destination_textures[1], 0, GL_FALSE, 0, GL_WRITE_ONLY, use_16bit_colors ? GL_RG16 : GL_RG8);
-                const double num_groups_x = (double)texture_size.x*0.5/(double)self->max_local_size_dim + 0.5;
-                const double num_groups_y = (double)texture_size.y*0.5/(double)self->max_local_size_dim + 0.5;
+                const double num_groups_x = ceil((double)texture_size.x*0.5/(double)self->max_local_size_dim);
+                const double num_groups_y = ceil((double)texture_size.y*0.5/(double)self->max_local_size_dim);
                 self->params.egl->glDispatchCompute(max_int(1, num_groups_x), max_int(1, num_groups_y), 1);
             }
             break;
@@ -473,8 +472,8 @@ void gsr_color_conversion_draw(gsr_color_conversion *self, unsigned int texture_
             self->params.egl->glUniform2i(uniform->target_position, destination_pos.x, destination_pos.y);
             self->params.egl->glUniform2f(uniform->scale, scale.x, scale.y);
             self->params.egl->glBindImageTexture(0, self->params.destination_textures[0], 0, GL_FALSE, 0, GL_READ_WRITE, GL_RGBA8);
-            const double num_groups_x = (double)texture_size.x/(double)self->max_local_size_dim + 0.5;
-            const double num_groups_y = (double)texture_size.y/(double)self->max_local_size_dim + 0.5;
+            const double num_groups_x = ceil((double)texture_size.x/(double)self->max_local_size_dim);
+            const double num_groups_y = ceil((double)texture_size.y/(double)self->max_local_size_dim);
             self->params.egl->glDispatchCompute(max_int(1, num_groups_x), max_int(1, num_groups_y), 1);
             break;
         }