v3d: flush jobs reading from transform feedback output buffers

[mesa.git] / src / gallium / drivers / v3d / v3dx_draw.c

diff --git a/src/gallium/drivers/v3d/v3dx_draw.c b/src/gallium/drivers/v3d/v3dx_draw.c
index 14e95c71204c7c31ad6c411af33ccf81f6f7b40d..98f630766ca9d62a1d7064804f2b5be4660fa2e2 100644 (file)
--- a/src/gallium/drivers/v3d/v3dx_draw.c
+++ b/src/gallium/drivers/v3d/v3dx_draw.c
@@ -157,7 +157,8 @@ v3d_predraw_check_stage_inputs(struct pipe_context *pctx,
                         continue;
                 struct v3d_sampler_view *view = v3d_sampler_view(pview);
 
-                if (view->texture != view->base.texture)
+                if (view->texture != view->base.texture &&
+                    view->base.format != PIPE_FORMAT_X32_S8X24_UINT)
                         v3d_update_shadow_texture(pctx, &view->base);
 
                 v3d_flush_jobs_writing_resource(v3d, view->texture);
@@ -178,6 +179,26 @@ v3d_predraw_check_stage_inputs(struct pipe_context *pctx,
         }
 }
 
+static void
+v3d_predraw_check_outputs(struct pipe_context *pctx)
+{
+        struct v3d_context *v3d = v3d_context(pctx);
+
+        /* Flush jobs reading from TF buffers that we are about to write. */
+        if (v3d_transform_feedback_enabled(v3d)) {
+                struct v3d_streamout_stateobj *so = &v3d->streamout;
+
+                for (int i = 0; i < so->num_targets; i++) {
+                        if (!so->targets[i])
+                                continue;
+
+                        const struct pipe_stream_output_target *target =
+                                so->targets[i];
+                        v3d_flush_jobs_reading_resource(v3d, target->buffer);
+                }
+        }
+}
+
 static void
 v3d_emit_gl_shader_state(struct v3d_context *v3d,
                          const struct pipe_draw_info *info)
@@ -235,6 +256,11 @@ v3d_emit_gl_shader_state(struct v3d_context *v3d,
                 shader.fragment_shader_uses_real_pixel_centre_w_in_addition_to_centroid_w2 =
                         v3d->prog.fs->prog_data.fs->uses_center_w;
 
+#if V3D_VERSION >= 40
+               shader.disable_implicit_point_line_varyings =
+                        !v3d->prog.fs->prog_data.fs->uses_implicit_point_line_varyings;
+#endif
+
                 shader.number_of_varyings_in_fragment_shader =
                         v3d->prog.fs->prog_data.fs->num_inputs;
 
@@ -261,9 +287,11 @@ v3d_emit_gl_shader_state(struct v3d_context *v3d,
                         v3d->prog.vs->prog_data.vs->separate_segments;
 
                 shader.coordinate_shader_input_vpm_segment_size =
-                        v3d->prog.cs->prog_data.vs->vpm_input_size;
+                        v3d->prog.cs->prog_data.vs->separate_segments ?
+                        v3d->prog.cs->prog_data.vs->vpm_input_size : 1;
                 shader.vertex_shader_input_vpm_segment_size =
-                        v3d->prog.vs->prog_data.vs->vpm_input_size;
+                        v3d->prog.vs->prog_data.vs->separate_segments ?
+                        v3d->prog.vs->prog_data.vs->vpm_input_size : 1;
 
                 shader.coordinate_shader_output_vpm_segment_size =
                         v3d->prog.cs->prog_data.vs->vpm_output_size;
@@ -320,6 +348,7 @@ v3d_emit_gl_shader_state(struct v3d_context *v3d,
                                    vtx->defaults_offset);
         }
 
+        bool cs_loaded_any = false;
         for (int i = 0; i < vtx->num_elements; i++) {
                 struct pipe_vertex_element *elem = &vtx->pipe[i];
                 struct pipe_vertex_buffer *vb =
@@ -339,6 +368,20 @@ v3d_emit_gl_shader_state(struct v3d_context *v3d,
                                 v3d->prog.cs->prog_data.vs->vattr_sizes[i];
                         attr.number_of_values_read_by_vertex_shader =
                                 v3d->prog.vs->prog_data.vs->vattr_sizes[i];
+
+                        /* GFXH-930: At least one attribute must be enabled
+                         * and read by CS and VS.  If we have attributes being
+                         * consumed by the VS but not the CS, then set up a
+                         * dummy load of the last attribute into the CS's VPM
+                         * inputs.  (Since CS is just dead-code-elimination
+                         * compared to VS, we can't have CS loading but not
+                         * VS).
+                         */
+                        if (v3d->prog.cs->prog_data.vs->vattr_sizes[i])
+                                cs_loaded_any = true;
+                        if (i == vtx->num_elements - 1 && !cs_loaded_any) {
+                                attr.number_of_values_read_by_coordinate_shader = 1;
+                        }
 #if V3D_VERSION >= 41
                         attr.maximum_index = 0xffffff;
 #endif
@@ -486,8 +529,8 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
                 return;
         }
 
-        /* Before setting up the draw, flush anything writing to the textures
-         * that we read from.
+        /* Before setting up the draw, flush anything writing to the resources
+         * that we read from or reading from resources we write to.
          */
         for (int s = 0; s < PIPE_SHADER_COMPUTE; s++)
                 v3d_predraw_check_stage_inputs(pctx, s);
@@ -495,6 +538,8 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
         if (info->indirect)
                 v3d_flush_jobs_writing_resource(v3d, info->indirect->buffer);
 
+        v3d_predraw_check_outputs(pctx);
+
         struct v3d_job *job = v3d_get_job_for_fbo(v3d);
 
         /* If vertex texturing depends on the output of rendering, we need to