Call back to intel_miptree_map when mapping the separate depth miptree in map_depthstencil. This brings us back to the mapping method decision tree in miptree_map where we hopefully find the most performant mapping method for depth. --- Here's an idea for a replacement of patch 5. Squirreling the depthstencil map away from the miptree when doing the recursive map is kinda yuck, but it lets us go back to the main miptree_map function for the depth map where we can get cpu detiling or whatever else.
src/mesa/drivers/dri/i965/intel_mipmap_tree.c | 54 +++++++++++++++------------ 1 file changed, 31 insertions(+), 23 deletions(-) diff --git a/src/mesa/drivers/dri/i965/intel_mipmap_tree.c b/src/mesa/drivers/dri/i965/intel_mipmap_tree.c index 98f71471bda..652074eb9fc 100644 --- a/src/mesa/drivers/dri/i965/intel_mipmap_tree.c +++ b/src/mesa/drivers/dri/i965/intel_mipmap_tree.c @@ -3525,27 +3525,30 @@ intel_miptree_map_depthstencil(struct brw_context *brw, if (!(map->mode & GL_MAP_INVALIDATE_RANGE_BIT)) { uint32_t *packed_map = map->ptr; uint8_t *s_map = intel_miptree_map_raw(brw, s_mt, GL_MAP_READ_BIT); - uint32_t *z_map = intel_miptree_map_raw(brw, z_mt, GL_MAP_READ_BIT); unsigned int s_image_x, s_image_y; - unsigned int z_image_x, z_image_y; + + uint32_t *z_map = NULL; + ptrdiff_t z_stride = 0; + + z_mt->level[level].slice[slice].map = NULL; + intel_miptree_map(brw, z_mt, level, slice, map->x, map->y, map->w, map->h, + map->mode | BRW_MAP_DIRECT_BIT, (void **)&z_map, + &z_stride); + assert(z_map && z_stride); intel_miptree_get_image_offset(s_mt, level, slice, &s_image_x, &s_image_y); - intel_miptree_get_image_offset(z_mt, level, slice, - &z_image_x, &z_image_y); for (uint32_t y = 0; y < map->h; y++) { + uint32_t *z_line = (uint32_t *)((uint8_t *)z_map + z_stride * y); for (uint32_t x = 0; x < map->w; x++) { int map_x = map->x + x, map_y = map->y + y; ptrdiff_t s_offset = intel_offset_S8(s_mt->surf.row_pitch, map_x + s_image_x, map_y + s_image_y, brw->has_swizzling); - ptrdiff_t z_offset = ((map_y + z_image_y) * - (z_mt->surf.row_pitch / 4) + - (map_x + z_image_x)); uint8_t s = s_map[s_offset]; - uint32_t z = z_map[z_offset]; + uint32_t z = z_line[x]; if (map_z32f_x24s8) { packed_map[(y * map->w + x) * 2 + 0] = z; @@ -3557,12 +3560,13 @@ intel_miptree_map_depthstencil(struct brw_context *brw, } intel_miptree_unmap_raw(s_mt); - intel_miptree_unmap_raw(z_mt); + intel_miptree_unmap(brw, z_mt, level, slice); + z_mt->level[level].slice[slice].map = map; - DBG("%s: %d,%d %dx%d from z mt %p %d,%d, s mt %p %d,%d = %p/%d\n", + DBG("%s: %d,%d %dx%d from z mt %p (%d,%d) @ (level:%d, slice:%d), s mt %p %d,%d = %p/%d\n", __func__, map->x, map->y, map->w, map->h, - z_mt, map->x + z_image_x, map->y + z_image_y, + z_mt, map->x, map->y, level, slice, s_mt, map->x + s_image_x, map->y + s_image_y, map->ptr, map->stride); } else { @@ -3586,44 +3590,48 @@ intel_miptree_unmap_depthstencil(struct brw_context *brw, if (map->mode & GL_MAP_WRITE_BIT) { uint32_t *packed_map = map->ptr; uint8_t *s_map = intel_miptree_map_raw(brw, s_mt, GL_MAP_WRITE_BIT); - uint32_t *z_map = intel_miptree_map_raw(brw, z_mt, GL_MAP_WRITE_BIT); unsigned int s_image_x, s_image_y; - unsigned int z_image_x, z_image_y; + + uint32_t *z_map = NULL; + ptrdiff_t z_stride = 0; + + z_mt->level[level].slice[slice].map = NULL; + intel_miptree_map(brw, z_mt, level, slice, map->x, map->y, map->w, map->h, + map->mode | GL_MAP_INVALIDATE_RANGE_BIT | BRW_MAP_DIRECT_BIT, + (void **)&z_map, &z_stride); + assert(z_map && z_stride); intel_miptree_get_image_offset(s_mt, level, slice, &s_image_x, &s_image_y); - intel_miptree_get_image_offset(z_mt, level, slice, - &z_image_x, &z_image_y); for (uint32_t y = 0; y < map->h; y++) { + uint32_t *z_line = (uint32_t *)((uint8_t *)z_map + z_stride * y); for (uint32_t x = 0; x < map->w; x++) { ptrdiff_t s_offset = intel_offset_S8(s_mt->surf.row_pitch, x + s_image_x + map->x, y + s_image_y + map->y, brw->has_swizzling); - ptrdiff_t z_offset = ((y + z_image_y + map->y) * - (z_mt->surf.row_pitch / 4) + - (x + z_image_x + map->x)); if (map_z32f_x24s8) { - z_map[z_offset] = packed_map[(y * map->w + x) * 2 + 0]; + z_line[x] = packed_map[(y * map->w + x) * 2 + 0]; s_map[s_offset] = packed_map[(y * map->w + x) * 2 + 1]; } else { uint32_t packed = packed_map[y * map->w + x]; s_map[s_offset] = packed >> 24; - z_map[z_offset] = packed; + z_line[x] = packed; } } } intel_miptree_unmap_raw(s_mt); - intel_miptree_unmap_raw(z_mt); + intel_miptree_unmap(brw, z_mt, level, slice); + z_mt->level[level].slice[slice].map = map; - DBG("%s: %d,%d %dx%d from z mt %p (%s) %d,%d, s mt %p %d,%d = %p/%d\n", + DBG("%s: %d,%d %dx%d from z mt %p (%s) (%d,%d) @ (level:%d, slice:%d), s mt %p %d,%d = %p/%d\n", __func__, map->x, map->y, map->w, map->h, z_mt, _mesa_get_format_name(z_mt->format), - map->x + z_image_x, map->y + z_image_y, + map->x, map->y, level, slice, s_mt, map->x + s_image_x, map->y + s_image_y, map->ptr, map->stride); } -- 2.14.3 _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev