etnaviv: hook-up etc2 patching

Changes v1 -> v2:
 - Avoid the GPU sampling from the resource that gets mutated by the the
   transfer map by setting DRM_ETNA_PREP_WRITE.

Changes v2 -> v3:
 - make use of likely(..)
 - drop minor optimization regarding rsc->layout == ETNA_LAYOUT_LINEAR
 - better documentation why DRM_ETNA_PREP_WRITE is needed

Signed-off-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>

src/gallium/drivers/etnaviv/etnaviv_resource.c

@@ -478,6 +478,9 @@ etna_resource_destroy(struct pipe_screen *pscreen, struct pipe_resource *prsc)
    pipe_resource_reference(&rsc->texture, NULL);
    pipe_resource_reference(&rsc->external, NULL);
 
+   for (unsigned i = 0; i < ETNA_NUM_LOD; i++)
+      FREE(rsc->levels[i].patch_offsets);
+
    FREE(rsc);
 }
 

src/gallium/drivers/etnaviv/etnaviv_resource.h

@@ -33,6 +33,7 @@
 #include "util/list.h"
 
 struct pipe_screen;
+struct util_dynarray;
 
 struct etna_resource_level {
    unsigned width, padded_width; /* in pixels */
@@ -47,6 +48,10 @@ struct etna_resource_level {
    uint32_t ts_size;
    uint32_t clear_value; /* clear value of resource level (mainly for TS) */
    bool ts_valid;
+
+   /* keep track if we have done some per block patching */
+   bool patched;
+   struct util_dynarray *patch_offsets;
 };
 
 enum etna_resource_addressing_mode {

src/gallium/drivers/etnaviv/etnaviv_transfer.c

@@ -28,6 +28,7 @@
 #include "etnaviv_clear_blit.h"
 #include "etnaviv_context.h"
 #include "etnaviv_debug.h"
+#include "etnaviv_etc2.h"
 #include "etnaviv_screen.h"
 
 #include "pipe/p_defines.h"
@@ -57,6 +58,46 @@ etna_compute_offset(enum pipe_format format, const struct pipe_box *box,
              util_format_get_blocksize(format);
 }
 
+static void etna_patch_data(void *buffer, const struct pipe_transfer *ptrans)
+{
+   struct pipe_resource *prsc = ptrans->resource;
+   struct etna_resource *rsc = etna_resource(prsc);
+   struct etna_resource_level *level = &rsc->levels[ptrans->level];
+
+   if (likely(!etna_etc2_needs_patching(prsc)))
+      return;
+
+   if (level->patched)
+      return;
+
+   /* do have the offsets of blocks to patch? */
+   if (!level->patch_offsets) {
+      level->patch_offsets = CALLOC_STRUCT(util_dynarray);
+
+      etna_etc2_calculate_blocks(buffer, ptrans->stride,
+                                         ptrans->box.width, ptrans->box.height,
+                                         prsc->format, level->patch_offsets);
+   }
+
+   etna_etc2_patch(buffer, level->patch_offsets);
+
+   level->patched = true;
+}
+
+static void etna_unpatch_data(void *buffer, const struct pipe_transfer *ptrans)
+{
+   struct pipe_resource *prsc = ptrans->resource;
+   struct etna_resource *rsc = etna_resource(prsc);
+   struct etna_resource_level *level = &rsc->levels[ptrans->level];
+
+   if (!level->patched)
+      return;
+
+   etna_etc2_patch(buffer, level->patch_offsets);
+
+   level->patched = false;
+}
+
 static void
 etna_transfer_unmap(struct pipe_context *pctx, struct pipe_transfer *ptrans)
 {
@@ -119,6 +160,9 @@ etna_transfer_unmap(struct pipe_context *pctx, struct pipe_transfer *ptrans)
       }
    }
 
+   /* We need to have the patched data ready for the GPU. */
+   etna_patch_data(trans->mapped, ptrans);
+
    /*
     * Transfers without a temporary are only pulled into the CPU domain if they
     * are not mapped unsynchronized. If they are, must push them back into GPU
@@ -321,6 +365,14 @@ etna_transfer_map(struct pipe_context *pctx, struct pipe_resource *prsc,
       if (usage & PIPE_TRANSFER_WRITE)
          prep_flags |= DRM_ETNA_PREP_WRITE;
 
+      /*
+       * The ETC2 patching operates in-place on the resource, so the resource will
+       * get written even on read-only transfers. This blocks the GPU to sample
+       * from this resource.
+       */
+      if ((usage & PIPE_TRANSFER_READ) && etna_etc2_needs_patching(prsc))
+         prep_flags |= DRM_ETNA_PREP_WRITE;
+
       if (etna_bo_cpu_prep(rsc->bo, prep_flags))
          goto fail_prep;
    }
@@ -340,6 +392,10 @@ etna_transfer_map(struct pipe_context *pctx, struct pipe_resource *prsc,
              etna_compute_offset(prsc->format, box, res_level->stride,
                                  res_level->layer_stride);
 
+      /* We need to have the unpatched data ready for the gfx stack. */
+      if (usage & PIPE_TRANSFER_READ)
+         etna_unpatch_data(trans->mapped, ptrans);
+
       return trans->mapped;
    } else {
       unsigned divSizeX = util_format_get_blockwidth(format);