intel: Fix software copying of miptree faces for weird formats.

Now that we have W-tiled S8, we can't just region_map and poke at bits --
there has to be some swizzling.  Rely on intel_miptree_map to get that job
done.  This should also get the highest performance path we know of for the
mapping (interesting if I get around to finishing movntdqa some day).

v2: Fix stale name of the bit in a comment.

Reviewed-by: Chad Versace <[email protected]>

3 files changed, 77 insertions, 61 deletions

diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
index b36a1eb85ce..ee058a1029e 100644
--- a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
+++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
@@ -702,6 +702,68 @@ intel_miptree_get_tile_offsets(struct intel_mipmap_tree *mt,
 }
 
 static void
+intel_miptree_copy_slice_sw(struct intel_context *intel,
+                            struct intel_mipmap_tree *dst_mt,
+                            struct intel_mipmap_tree *src_mt,
+                            int level,
+                            int slice,
+                            int width,
+                            int height)
+{
+   void *src, *dst;
+   int src_stride, dst_stride;
+   int cpp = dst_mt->cpp;
+
+   intel_miptree_map(intel, src_mt,
+                     level, slice,
+                     0, 0,
+                     width, height,
+                     GL_MAP_READ_BIT | BRW_MAP_DIRECT_BIT,
+                     &src, &src_stride);
+
+   intel_miptree_map(intel, dst_mt,
+                     level, slice,
+                     0, 0,
+                     width, height,
+                     GL_MAP_WRITE_BIT | GL_MAP_INVALIDATE_RANGE_BIT |
+                     BRW_MAP_DIRECT_BIT,
+                     &dst, &dst_stride);
+
+   DBG("sw blit %s mt %p %p/%d -> %s mt %p %p/%d (%dx%d)\n",
+       _mesa_get_format_name(src_mt->format),
+       src_mt, src, src_stride,
+       _mesa_get_format_name(dst_mt->format),
+       dst_mt, dst, dst_stride,
+       width, height);
+
+   int row_size = cpp * width;
+   if (src_stride == row_size &&
+       dst_stride == row_size) {
+      memcpy(dst, src, row_size * height);
+   } else {
+      for (int i = 0; i < height; i++) {
+         memcpy(dst, src, row_size);
+         dst += dst_stride;
+         src += src_stride;
+      }
+   }
+
+   intel_miptree_unmap(intel, dst_mt, level, slice);
+   intel_miptree_unmap(intel, src_mt, level, slice);
+
+   /* Don't forget to copy the stencil data over, too.  We could have skipped
+    * passing BRW_MAP_DIRECT_BIT, but that would have meant intel_miptree_map
+    * shuffling the two data sources in/out of temporary storage instead of
+    * the direct mapping we get this way.
+    */
+   if (dst_mt->stencil_mt) {
+      assert(src_mt->stencil_mt);
+      intel_miptree_copy_slice_sw(intel, dst_mt->stencil_mt, src_mt->stencil_mt,
+                                  level, slice, width, height);
+   }
+}
+
+static void
 intel_miptree_copy_slice(struct intel_context *intel,
 			 struct intel_mipmap_tree *dst_mt,
 			 struct intel_mipmap_tree *src_mt,
@@ -721,12 +783,25 @@ intel_miptree_copy_slice(struct intel_context *intel,
       slice = depth;
 
    assert(depth < src_mt->level[level].depth);
+   assert(src_mt->format == dst_mt->format);
 
    if (dst_mt->compressed) {
       height = ALIGN(height, dst_mt->align_h) / dst_mt->align_h;
       width = ALIGN(width, dst_mt->align_w);
    }
 
+   /* If it's a packed depth/stencil buffer with separate stencil, the blit
+    * below won't apply since we can't do the depth's Y tiling or the
+    * stencil's W tiling in the blitter.
+    */
+   if (src_mt->stencil_mt) {
+      intel_miptree_copy_slice_sw(intel,
+                                  dst_mt, src_mt,
+                                  level, slice,
+                                  width, height);
+      return;
+   }
+
    uint32_t dst_x, dst_y, src_x, src_y;
    intel_miptree_get_image_offset(dst_mt, level, slice, &dst_x, &dst_y);
    intel_miptree_get_image_offset(src_mt, level, slice, &src_x, &src_y);
@@ -751,25 +826,9 @@ intel_miptree_copy_slice(struct intel_context *intel,
 
       fallback_debug("miptree validate blit for %s failed\n",
 		     _mesa_get_format_name(format));
-      void *dst = intel_region_map(intel, dst_mt->region, GL_MAP_WRITE_BIT);
-      void *src = intel_region_map(intel, src_mt->region, GL_MAP_READ_BIT);
-
-      _mesa_copy_rect(dst,
-		      dst_mt->cpp,
-		      dst_mt->region->pitch,
-		      dst_x, dst_y,
-		      width, height,
-		      src, src_mt->region->pitch,
-		      src_x, src_y);
-
-      intel_region_unmap(intel, dst_mt->region);
-      intel_region_unmap(intel, src_mt->region);
-   }
 
-   if (src_mt->stencil_mt) {
-      intel_miptree_copy_slice(intel,
-                               dst_mt->stencil_mt, src_mt->stencil_mt,
-                               level, face, depth);
+      intel_miptree_copy_slice_sw(intel, dst_mt, src_mt, level, slice,
+                                  width, height);
    }
 }
 
diff --git a/src/mesa/drivers/dri/intel/intel_regions.c b/src/mesa/drivers/dri/intel/intel_regions.c
index 89d91b0f8fe..a5e49d99e9f 100644
--- a/src/mesa/drivers/dri/intel/intel_regions.c
+++ b/src/mesa/drivers/dri/intel/intel_regions.c
@@ -328,39 +328,6 @@ intel_region_release(struct intel_region **region_handle)
    *region_handle = NULL;
 }
 
-/*
- * XXX Move this into core Mesa?
- */
-void
-_mesa_copy_rect(GLubyte * dst,
-                GLuint cpp,
-                GLuint dst_pitch,
-                GLuint dst_x,
-                GLuint dst_y,
-                GLuint width,
-                GLuint height,
-                const GLubyte * src,
-                GLuint src_pitch, GLuint src_x, GLuint src_y)
-{
-   GLuint i;
-
-   dst += dst_x * cpp;
-   src += src_x * cpp;
-   dst += dst_y * dst_pitch;
-   src += src_y * src_pitch;
-   width *= cpp;
-
-   if (width == dst_pitch && width == src_pitch)
-      memcpy(dst, src, height * width);
-   else {
-      for (i = 0; i < height; i++) {
-         memcpy(dst, src, width);
-         dst += dst_pitch;
-         src += src_pitch;
-      }
-   }
-}
-
 /* Copy rectangular sub-regions. Need better logic about when to
  * push buffers into AGP - will currently do so whenever possible.
  */
diff --git a/src/mesa/drivers/dri/intel/intel_regions.h b/src/mesa/drivers/dri/intel/intel_regions.h
index 28d581c800c..93181414c91 100644
--- a/src/mesa/drivers/dri/intel/intel_regions.h
+++ b/src/mesa/drivers/dri/intel/intel_regions.h
@@ -123,16 +123,6 @@ intel_region_copy(struct intel_context *intel,
 		  bool flip,
 		  GLenum logicop);
 
-void _mesa_copy_rect(GLubyte * dst,
-                GLuint cpp,
-                GLuint dst_pitch,
-                GLuint dst_x,
-                GLuint dst_y,
-                GLuint width,
-                GLuint height,
-                const GLubyte * src,
-                GLuint src_pitch, GLuint src_x, GLuint src_y);
-
 void
 intel_region_get_tile_masks(struct intel_region *region,
                             uint32_t *mask_x, uint32_t *mask_y,