v3dv: defer tile_alloc creation in meta TLB ops

Remove the allocate_tile_state_now parameter from v3dv_job_start_frame().
So v3dv_job_allocate_tile_state() is explicitly called after
job_emit_binning_flush() as we know the value of job->draw_count instead
of using always 0.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40554>

src/broadcom/vulkan/v3dv_cmd_buffer.c

@@ -490,7 +490,6 @@ v3dv_job_start_frame(struct v3dv_job *job,
                      uint32_t height,
                      uint32_t layers,
                      bool allocate_tile_state_for_all_layers,
-                     bool allocate_tile_state_now,
                      uint32_t render_target_count,
                      uint8_t max_internal_bpp,
                      uint8_t total_color_bpp,
@@ -511,14 +510,6 @@ v3dv_job_start_frame(struct v3dv_job *job,
 
    job->allocate_tile_state_for_all_layers = allocate_tile_state_for_all_layers;
 
-   /* For subpass jobs we postpone tile state allocation until we are finishing
-    * the job and have made a decision about double-buffer.
-    */
-   if (allocate_tile_state_now) {
-      if (!v3dv_job_allocate_tile_state(job))
-         return;
-   }
-
    v3d_X((&job->device->devinfo), job_emit_binning_prolog)(job, tiling,
       allocate_tile_state_for_all_layers ? tiling->layers : 1);
 
@@ -1785,7 +1776,7 @@ cmd_buffer_subpass_create_job(struct v3dv_cmd_buffer *cmd_buffer,
                            width,
                            height,
                            layers,
-                           true, false,
+                           true,
                            subpass->color_count,
                            max_internal_bpp,
                            total_color_bpp,
@@ -2776,7 +2767,7 @@ cmd_buffer_restart_job_for_msaa_if_needed(struct v3dv_cmd_buffer *cmd_buffer)
                         old_job->frame_tiling.width,
                         old_job->frame_tiling.height,
                         old_job->frame_tiling.layers,
-                        true, false,
+                        true,
                         old_job->frame_tiling.render_target_count,
                         old_job->frame_tiling.internal_bpp,
                         old_job->frame_tiling.total_color_bpp,

src/broadcom/vulkan/v3dv_cmd_buffer.h

@@ -306,7 +306,6 @@ void v3dv_job_start_frame(struct v3dv_job *job,
                           uint32_t height,
                           uint32_t layers,
                           bool allocate_tile_state_for_all_layers,
-                          bool allocate_tile_state_now,
                           uint32_t render_target_count,
                           uint8_t max_internal_bpp,
                           uint8_t total_color_bpp,

src/broadcom/vulkan/v3dv_meta_clear.c

@@ -138,9 +138,8 @@ clear_image_tlb(struct v3dv_cmd_buffer *cmd_buffer,
       if (!job)
          return true;
 
-      v3dv_job_start_frame(job, width, height, max_layer,
-                           false, true, 1, internal_bpp,
-                           4 * v3d_internal_bpp_words(internal_bpp),
+      v3dv_job_start_frame(job, width, height, max_layer, false, 1,
+                           internal_bpp, 4 * v3d_internal_bpp_words(internal_bpp),
                            image->vk.samples > VK_SAMPLE_COUNT_1_BIT);
 
       struct v3dv_meta_framebuffer framebuffer;
@@ -149,6 +148,8 @@ clear_image_tlb(struct v3dv_cmd_buffer *cmd_buffer,
                                                  &job->frame_tiling);
 
       v3d_X((&job->device->devinfo), job_emit_binning_flush)(job);
+      if (!v3dv_job_allocate_tile_state(job))
+         return true;
 
       /* If this triggers it is an application bug: the spec requires
        * that any aspects to clear are present in the image.

src/broadcom/vulkan/v3dv_meta_copy.c

@@ -485,15 +485,17 @@ copy_image_to_buffer_tlb(struct v3dv_cmd_buffer *cmd_buffer,
    const uint32_t width = DIV_ROUND_UP(region->imageExtent.width, block_w);
    const uint32_t height = DIV_ROUND_UP(region->imageExtent.height, block_h);
 
-   v3dv_job_start_frame(job, width, height, num_layers, false, true, 1,
-                        internal_bpp, 4 * v3d_internal_bpp_words(internal_bpp),
-                        false);
+   v3dv_job_start_frame(job, width, height, num_layers, false, 1, internal_bpp,
+                        4 * v3d_internal_bpp_words(internal_bpp), false);
 
    struct v3dv_meta_framebuffer framebuffer;
    v3d_X((&job->device->devinfo), meta_framebuffer_init)(&framebuffer, fb_format,
                                               internal_type, &job->frame_tiling);
 
    v3d_X((&job->device->devinfo), job_emit_binning_flush)(job);
+   if (!v3dv_job_allocate_tile_state(job))
+      return true;
+
    v3d_X((&job->device->devinfo), meta_emit_copy_image_to_buffer_rcl)
       (job, buffer, image, &framebuffer, region);
 
@@ -1377,7 +1379,7 @@ copy_image_tlb(struct v3dv_cmd_buffer *cmd_buffer,
    const uint32_t width = DIV_ROUND_UP(region->extent.width, block_w);
    const uint32_t height = DIV_ROUND_UP(region->extent.height, block_h);
 
-   v3dv_job_start_frame(job, width, height, num_layers, false, true, 1,
+   v3dv_job_start_frame(job, width, height, num_layers, false, 1,
                         internal_bpp, 4 * v3d_internal_bpp_words(internal_bpp),
                         src->vk.samples > VK_SAMPLE_COUNT_1_BIT);
 
@@ -1386,6 +1388,9 @@ copy_image_tlb(struct v3dv_cmd_buffer *cmd_buffer,
                                               internal_type, &job->frame_tiling);
 
    v3d_X((&job->device->devinfo), job_emit_binning_flush)(job);
+   if (!v3dv_job_allocate_tile_state(job))
+      return true;
+
    v3d_X((&job->device->devinfo), meta_emit_copy_image_rcl)(job, dst, src, &framebuffer, region);
 
    v3dv_cmd_buffer_finish_job(cmd_buffer);
@@ -2061,7 +2066,7 @@ copy_buffer_to_image_tlb(struct v3dv_cmd_buffer *cmd_buffer,
    const uint32_t width = DIV_ROUND_UP(region->imageExtent.width, block_w);
    const uint32_t height = DIV_ROUND_UP(region->imageExtent.height, block_h);
 
-   v3dv_job_start_frame(job, width, height, num_layers, false, true, 1,
+   v3dv_job_start_frame(job, width, height, num_layers, false, 1,
                         internal_bpp, 4 * v3d_internal_bpp_words(internal_bpp),
                         false);
 
@@ -2070,6 +2075,9 @@ copy_buffer_to_image_tlb(struct v3dv_cmd_buffer *cmd_buffer,
                                               internal_type, &job->frame_tiling);
 
    v3d_X((&job->device->devinfo), job_emit_binning_flush)(job);
+   if (!v3dv_job_allocate_tile_state(job))
+      return true;
+
    v3d_X((&job->device->devinfo), meta_emit_copy_buffer_to_image_rcl)
       (job, image, buffer, &framebuffer, region);
 
@@ -4923,15 +4931,17 @@ resolve_image_tlb(struct v3dv_cmd_buffer *cmd_buffer,
       (resolve_format, region->srcSubresource.aspectMask,
        &internal_type, &internal_bpp);
 
-   v3dv_job_start_frame(job, width, height, num_layers, false, true, 1,
-                        internal_bpp, 4 * v3d_internal_bpp_words(internal_bpp),
-                        true);
+   v3dv_job_start_frame(job, width, height, num_layers, false, 1, internal_bpp,
+                        4 * v3d_internal_bpp_words(internal_bpp), true);
 
    struct v3dv_meta_framebuffer framebuffer;
    v3d_X((&job->device->devinfo), meta_framebuffer_init)(&framebuffer, resolve_format,
                                               internal_type, &job->frame_tiling);
 
    v3d_X((&job->device->devinfo), job_emit_binning_flush)(job);
+   if (!v3dv_job_allocate_tile_state(job))
+      return true;
+
    v3d_X((&job->device->devinfo), meta_emit_resolve_image_rcl)(job, dst, src,
                                                     &framebuffer, region);
 

src/broadcom/vulkan/v3dvx_meta_common.c

@@ -1410,15 +1410,16 @@ v3dX(meta_copy_buffer)(struct v3dv_cmd_buffer *cmd_buffer,
       uint32_t width, height;
       framebuffer_size_for_pixel_count(num_items, &width, &height);
 
-      v3dv_job_start_frame(job, width, height, 1, true, true, 1,
-                           internal_bpp, 4 * v3d_internal_bpp_words(internal_bpp),
-                           false);
+      v3dv_job_start_frame(job, width, height, 1, true, 1, internal_bpp,
+                           4 * v3d_internal_bpp_words(internal_bpp), false);
 
       struct v3dv_meta_framebuffer framebuffer;
       v3dX(meta_framebuffer_init)(&framebuffer, vk_format, internal_type,
                                   &job->frame_tiling);
 
       v3dX(job_emit_binning_flush)(job);
+      if (!v3dv_job_allocate_tile_state(job))
+         return NULL;
 
       v3dX(meta_emit_copy_buffer_rcl)(job, dst, src, dst_offset, src_offset,
                                       &framebuffer, format, item_size);
@@ -1458,15 +1459,16 @@ v3dX(meta_fill_buffer)(struct v3dv_cmd_buffer *cmd_buffer,
       uint32_t width, height;
       framebuffer_size_for_pixel_count(num_items, &width, &height);
 
-      v3dv_job_start_frame(job, width, height, 1, true, true, 1,
-                           internal_bpp, 4 * v3d_internal_bpp_words(internal_bpp),
-                           false);
+      v3dv_job_start_frame(job, width, height, 1, true, 1, internal_bpp,
+                           4 * v3d_internal_bpp_words(internal_bpp), false);
 
       struct v3dv_meta_framebuffer framebuffer;
       v3dX(meta_framebuffer_init)(&framebuffer, VK_FORMAT_R8G8B8A8_UINT,
                                   internal_type, &job->frame_tiling);
 
       v3dX(job_emit_binning_flush)(job);
+      if (!v3dv_job_allocate_tile_state(job))
+         return;
 
       v3dX(meta_emit_fill_buffer_rcl)(job, bo, offset, &framebuffer, data);
 

src/broadcom/vulkan/v3dvx_queue.c

@@ -30,9 +30,10 @@
 void
 v3dX(job_emit_noop)(struct v3dv_job *job)
 {
-   v3dv_job_start_frame(job, 1, 1, 1, true, true, 1,
-                        V3D_INTERNAL_BPP_32, 4, false);
+   v3dv_job_start_frame(job, 1, 1, 1, true, 1, V3D_INTERNAL_BPP_32, 4, false);
    v3dX(job_emit_binning_flush)(job);
+   if (!v3dv_job_allocate_tile_state(job))
+      return;
 
    struct v3dv_cl *rcl = &job->rcl;
    v3dv_cl_ensure_space_with_branch(rcl, 200 + 1 * 256 *