pan/crc: Simplify CRC buffer initialization

Transaction Elimination on a RT is disabled until there's a full frame
render with all tiles forcefully written back. This is currently done
by letting the Gallium driver track states and fix up FB preload by
disabling clean_fragment_write on the pre-frame DCD and by setting the
pre-frame mode to "always" (instead of "intersect").

This commit forces the write-back of all the tiles by setting
clean_tile_write_enable on the FBD instead. This simplifies the code
and removes most of the CRC state tracking from the Gallium driver.

Signed-off-by: Loïc Molinari <loic.molinari@collabora.com>

src/gallium/drivers/panfrost/pan_fb_preload.c

@@ -1018,8 +1018,7 @@ pan_preload_emit_viewport(struct pan_pool *pool, uint16_t minx, uint16_t miny,
 static void
 pan_preload_emit_dcd(struct pan_fb_preload_cache *cache, struct pan_pool *pool,
                      struct pan_fb_info *fb, bool zs, uint64_t coordinates,
-                     uint64_t tsd, struct mali_draw_packed *out,
-                     bool always_write)
+                     uint64_t tsd, struct mali_draw_packed *out)
 {
    unsigned tex_count = 0;
    uint64_t textures = pan_preload_emit_textures(pool, fb, zs, &tex_count);
@@ -1031,7 +1030,7 @@ pan_preload_emit_dcd(struct pan_fb_preload_cache *cache, struct pan_pool *pool,
    /* Tiles updated by preload shaders are still considered clean (separate
     * for colour and Z/S), allowing us to suppress unnecessary writeback
     */
-   UNUSED bool clean_fragment_write = !always_write;
+   UNUSED bool clean_fragment_write = true;
 
    /* Image view used when patching stencil formats for combined
     * depth/stencil preloads.
@@ -1186,25 +1185,8 @@ pan_preload_emit_pre_frame_dcd(struct pan_fb_preload_cache *cache,
 
    void *dcd = fb->bifrost.pre_post.dcds.cpu + (dcd_idx * pan_size(DRAW));
 
-   /* We only use crc_rt to determine whether to force writes for updating
-    * the CRCs, so use a conservative tile size (16x16).
-    */
-   int crc_rt = GENX(pan_select_crc_rt)(fb, 16 * 16);
+   pan_preload_emit_dcd(cache, desc_pool, fb, zs, coords, tsd, dcd);
 
-   bool always_write = false;
-
-   /* If CRC data is currently invalid and this batch will make it valid,
-    * write even clean tiles to make sure CRC data is updated. */
-   if (crc_rt >= 0) {
-      bool *valid = fb->rts[crc_rt].crc_valid;
-      bool full = pan_fb_info_is_fully_covered(fb);
-
-      if (full && !(*valid))
-         always_write = true;
-   }
-
-   pan_preload_emit_dcd(cache, desc_pool, fb, zs, coords, tsd, dcd,
-                        always_write);
    if (zs) {
       enum pipe_format fmt = fb->zs.view.zs
                                 ? fb->zs.view.zs->planes[0].image->props.format
@@ -1264,8 +1246,7 @@ pan_preload_emit_pre_frame_dcd(struct pan_fb_preload_cache *cache,
 #endif
    } else {
       fb->bifrost.pre_post.modes[dcd_idx] =
-         always_write ? MALI_PRE_POST_FRAME_SHADER_MODE_ALWAYS
-                      : MALI_PRE_POST_FRAME_SHADER_MODE_INTERSECT;
+         MALI_PRE_POST_FRAME_SHADER_MODE_INTERSECT;
    }
 }
 #else
@@ -1280,7 +1261,7 @@ pan_preload_emit_tiler_job(struct pan_fb_preload_cache *cache,
       return (struct pan_ptr){0};
 
    pan_preload_emit_dcd(cache, desc_pool, fb, zs, coords, tsd,
-                        pan_section_ptr(job.cpu, TILER_JOB, DRAW), false);
+                        pan_section_ptr(job.cpu, TILER_JOB, DRAW));
 
    pan_section_pack(job.cpu, TILER_JOB, PRIMITIVE, cfg) {
       cfg.draw_mode = MALI_DRAW_MODE_TRIANGLE_STRIP;

src/panfrost/lib/pan_desc.c

@@ -1045,8 +1045,8 @@ pan_fb_color_attachment_should_crc(const struct pan_fb_color_attachment *rt,
    return true;
 }
 
-int
-GENX(pan_select_crc_rt)(const struct pan_fb_info *fb, unsigned tile_size)
+static int
+pan_select_crc_rt(const struct pan_fb_info *fb)
 {
    int best_rt = -1;
 
@@ -1059,7 +1059,7 @@ GENX(pan_select_crc_rt)(const struct pan_fb_info *fb, unsigned tile_size)
     * CRCs are more expensive at smaller tile sizes, reducing the benefit.
     * Restricting CRC to 16x16 should work in practice.
     */
-   if (tile_size < 16 * 16)
+   if (fb->tile_size < 16 * 16)
       return best_rt;
 
 #if PAN_ARCH <= 6
@@ -1069,7 +1069,7 @@ GENX(pan_select_crc_rt)(const struct pan_fb_info *fb, unsigned tile_size)
 
    for (unsigned i = 0; i < fb->rt_count; i++) {
       /* Skip unusable RTs. */
-      if (!pan_fb_color_attachment_should_crc(&fb->rts[i], tile_size))
+      if (!pan_fb_color_attachment_should_crc(&fb->rts[i], fb->tile_size))
          continue;
 
       /* Select the first RT with a valid CRC buffer. */
@@ -1118,8 +1118,18 @@ GENX(pan_force_clean_write_on)(const struct pan_image *image,
 #endif
 }
 
+static bool
+pan_force_clean_write_crc(const struct pan_fb_info *fb, int index, int crc_rt)
+{
+   bool match = index == crc_rt;
+   bool valid = *(fb->rts[index].crc_valid);
+   bool full = pan_fb_info_is_fully_covered(fb);
+
+   return match && !valid && full;
+}
+
 static struct pan_clean_tile
-pan_get_clean_tile_info(const struct pan_fb_info *fb)
+pan_get_clean_tile_info(const struct pan_fb_info *fb, int crc_rt)
 {
    struct pan_clean_tile clean_tile = { 0, };
    const struct pan_image *img;
@@ -1134,7 +1144,8 @@ pan_get_clean_tile_info(const struct pan_fb_info *fb)
       img = fb->rts[i].view ?
          pan_image_view_get_color_plane(fb->rts[i].view).image : NULL;
       if (fb->rts[i].clear ||
-          GENX(pan_force_clean_write_on)(img, fb->tile_size))
+          GENX(pan_force_clean_write_on)(img, fb->tile_size) ||
+          pan_force_clean_write_crc(fb, i, crc_rt))
          clean_tile.write_rt_mask |= 1 << i;
    }
 
@@ -1195,9 +1206,9 @@ GENX(pan_emit_fbd)(const struct pan_fb_info *fb, unsigned layer_idx,
    GENX(pan_emit_tls)(tls, pan_section_ptr(fbd, FRAMEBUFFER, LOCAL_STORAGE));
 #endif
 
-   int crc_rt = GENX(pan_select_crc_rt)(fb, fb->tile_size);
+   int crc_rt = pan_select_crc_rt(fb);
    bool has_zs_crc_ext = (fb->zs.view.zs || fb->zs.view.s || crc_rt >= 0);
-   struct pan_clean_tile clean_tile = pan_get_clean_tile_info(fb);
+   struct pan_clean_tile clean_tile = pan_get_clean_tile_info(fb, crc_rt);
 
    pan_section_pack(fbd, FRAMEBUFFER, PARAMETERS, cfg) {
 #if PAN_ARCH >= 6

src/panfrost/lib/pan_desc.h

@@ -303,8 +303,6 @@ bool GENX(pan_force_clean_write_on)(const struct pan_image *image,
 void GENX(pan_emit_tls)(const struct pan_tls_info *info,
                         struct mali_local_storage_packed *out);
 
-int GENX(pan_select_crc_rt)(const struct pan_fb_info *fb, unsigned tile_size);
-
 struct pan_attachment_info {
    const struct pan_image_view *iview;
    unsigned layer_or_z_slice;