v3dv: implement vkCmdCopyImageToBuffer
For now we only support the TLB path, which limits us to copying
regions that start at offset (0,0). In the future, we will need to add
a fallback path that uses blitting to copy regions with an offset.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6766>
diff --git a/src/broadcom/vulkan/v3dv_meta_copy.c b/src/broadcom/vulkan/v3dv_meta_copy.c
index 98e6f0b..a7728a2 100644
--- a/src/broadcom/vulkan/v3dv_meta_copy.c
+++ b/src/broadcom/vulkan/v3dv_meta_copy.c
@@ -23,6 +23,379 @@
#include "v3dv_private.h"
+#include "broadcom/cle/v3dx_pack.h"
+#include "vk_format_info.h"
+
+static void
+emit_image_loads(struct v3dv_cmd_buffer *cmd_buffer,
+ struct v3dv_cl *cl,
+ struct v3dv_image *image,
+ uint32_t layer,
+ uint32_t mip_level)
+{
+ uint32_t layer_offset = v3dv_layer_offset(image, mip_level, layer);
+
+ const struct util_format_description *desc =
+ vk_format_description(image->vk_format);
+ bool swap_rb = desc->swizzle[0] == PIPE_SWIZZLE_Z &&
+ image->vk_format != VK_FORMAT_B5G6R5_UNORM_PACK16;
+
+ cl_emit(cl, LOAD_TILE_BUFFER_GENERAL, load) {
+ load.buffer_to_load = RENDER_TARGET_0;
+ load.address = v3dv_cl_address(image->mem->bo, layer_offset);
+
+ load.input_image_format = image->format->rt_type;
+ load.r_b_swap = swap_rb;
+ load.memory_format = image->slices[0].tiling;
+
+ const struct v3d_resource_slice *slice = &image->slices[0];
+ if (slice->tiling == VC5_TILING_UIF_NO_XOR ||
+ slice->tiling == VC5_TILING_UIF_XOR) {
+ load.height_in_ub_or_stride =
+ slice->padded_height_of_output_image_in_uif_blocks;
+ } else if (slice->tiling == VC5_TILING_RASTER) {
+ load.height_in_ub_or_stride = slice->stride;
+ }
+
+ if (image->samples > VK_SAMPLE_COUNT_1_BIT)
+ load.decimate_mode = V3D_DECIMATE_MODE_ALL_SAMPLES;
+ else
+ load.decimate_mode = V3D_DECIMATE_MODE_SAMPLE_0;
+ }
+
+ cl_emit(cl, END_OF_LOADS, end);
+}
+
+static void
+emit_buffer_stores(struct v3dv_cmd_buffer *cmd_buffer,
+ struct v3dv_cl *cl,
+ struct v3dv_buffer *buffer,
+ struct v3dv_image *image,
+ uint32_t buffer_offset,
+ uint32_t buffer_stride)
+{
+ cl_emit(cl, STORE_TILE_BUFFER_GENERAL, store) {
+ store.buffer_to_store = RENDER_TARGET_0;
+ store.address = v3dv_cl_address(buffer->mem->bo, buffer_offset);
+ store.clear_buffer_being_stored = false;
+
+ store.output_image_format = image->format->rt_type;
+ store.r_b_swap = false;
+ store.memory_format = VC5_TILING_RASTER;
+ store.height_in_ub_or_stride = buffer_stride;
+
+ if (image->samples > VK_SAMPLE_COUNT_1_BIT)
+ store.decimate_mode = V3D_DECIMATE_MODE_ALL_SAMPLES;
+ else
+ store.decimate_mode = V3D_DECIMATE_MODE_SAMPLE_0;
+ }
+}
+
+static void
+emit_copy_layer_to_buffer_per_tile_list(struct v3dv_cmd_buffer *cmd_buffer,
+ struct v3dv_buffer *buffer,
+ struct v3dv_image *image,
+ uint32_t layer,
+ const VkBufferImageCopy *region)
+{
+ struct v3dv_cl *cl = &cmd_buffer->indirect;
+ v3dv_cl_ensure_space(cl, 200, 1);
+ struct v3dv_cl_reloc tile_list_start = v3dv_cl_get_address(cl);
+
+ cl_emit(cl, TILE_COORDINATES_IMPLICIT, coords);
+
+ const VkImageSubresourceLayers *imgrsc = ®ion->imageSubresource;
+ assert(layer < imgrsc->layerCount);
+
+ /* Load image to TLB */
+ emit_image_loads(cmd_buffer, cl, image,
+ imgrsc->baseArrayLayer + layer, imgrsc->mipLevel);
+
+ cl_emit(cl, PRIM_LIST_FORMAT, fmt) {
+ fmt.primitive_type = LIST_TRIANGLES;
+ }
+
+ cl_emit(cl, BRANCH_TO_IMPLICIT_TILE_LIST, branch);
+
+ /* Store TLB to buffer */
+ uint32_t width, height;
+ if (region->bufferRowLength == 0 || region->bufferImageHeight == 0) {
+ width = image->extent.width;
+ height = image->extent.height;
+ } else {
+ width = region->bufferRowLength;
+ height = region->bufferImageHeight;
+ }
+
+ uint32_t buffer_stride = width * image->cpp;
+ uint32_t buffer_offset =
+ region->bufferOffset + height * buffer_stride * layer;
+ emit_buffer_stores(cmd_buffer, cl, buffer, image,
+ buffer_offset, buffer_stride);
+
+ cl_emit(cl, END_OF_TILE_MARKER, end);
+
+ cl_emit(cl, RETURN_FROM_SUB_LIST, ret);
+
+ cl_emit(&cmd_buffer->rcl, START_ADDRESS_OF_GENERIC_TILE_LIST, branch) {
+ branch.start = tile_list_start;
+ branch.end = v3dv_cl_get_address(cl);
+ }
+}
+
+static void
+emit_copy_layer_to_buffer(struct v3dv_cmd_buffer *cmd_buffer,
+ uint32_t min_x_supertile,
+ uint32_t min_y_supertile,
+ uint32_t max_x_supertile,
+ uint32_t max_y_supertile,
+ struct v3dv_buffer *buffer,
+ struct v3dv_image *image,
+ struct v3dv_framebuffer *framebuffer,
+ uint32_t layer,
+ const VkBufferImageCopy *region)
+{
+ struct v3dv_cl *rcl = &cmd_buffer->rcl;
+
+ const uint32_t tile_alloc_offset =
+ 64 * layer * framebuffer->draw_tiles_x * framebuffer->draw_tiles_y;
+ cl_emit(rcl, MULTICORE_RENDERING_TILE_LIST_SET_BASE, list) {
+ list.address = v3dv_cl_address(cmd_buffer->tile_alloc, tile_alloc_offset);
+ }
+
+ cl_emit(rcl, MULTICORE_RENDERING_SUPERTILE_CFG, config) {
+ config.number_of_bin_tile_lists = 1;
+ config.total_frame_width_in_tiles = framebuffer->draw_tiles_x;
+ config.total_frame_height_in_tiles = framebuffer->draw_tiles_y;
+
+ config.supertile_width_in_tiles = framebuffer->supertile_width;
+ config.supertile_height_in_tiles = framebuffer->supertile_height;
+
+ config.total_frame_width_in_supertiles =
+ framebuffer->frame_width_in_supertiles;
+ config.total_frame_height_in_supertiles =
+ framebuffer->frame_height_in_supertiles;
+ }
+
+ /* GFXH-1742 workaround */
+ for (int i = 0; i < 2; i++) {
+ cl_emit(rcl, TILE_COORDINATES, coords);
+ cl_emit(rcl, END_OF_LOADS, end);
+ cl_emit(rcl, STORE_TILE_BUFFER_GENERAL, store) {
+ store.buffer_to_store = NONE;
+ }
+ cl_emit(rcl, END_OF_TILE_MARKER, end);
+ }
+
+ cl_emit(rcl, FLUSH_VCD_CACHE, flush);
+
+ emit_copy_layer_to_buffer_per_tile_list(cmd_buffer, buffer, image,
+ layer, region);
+
+ for (int y = min_y_supertile; y <= max_y_supertile; y++) {
+ for (int x = min_x_supertile; x <= max_x_supertile; x++) {
+ cl_emit(rcl, SUPERTILE_COORDINATES, coords) {
+ coords.column_number_in_supertiles = x;
+ coords.row_number_in_supertiles = y;
+ }
+ }
+ }
+}
+
+static void
+emit_copy_image_to_buffer_rcl(struct v3dv_cmd_buffer *cmd_buffer,
+ struct v3dv_buffer *buffer,
+ struct v3dv_image *image,
+ struct v3dv_framebuffer *framebuffer,
+ uint32_t internal_type,
+ const VkBufferImageCopy *region)
+{
+ const VkImageSubresourceLayers *imgrsc = ®ion->imageSubresource;
+
+ struct v3dv_cl *rcl = &cmd_buffer->rcl;
+ v3dv_cl_ensure_space_with_branch(rcl, 200 +
+ imgrsc->layerCount * 256 *
+ cl_packet_length(SUPERTILE_COORDINATES));
+
+ uint32_t level_width = u_minify(image->extent.width, imgrsc->mipLevel);
+ uint32_t level_height = u_minify(image->extent.height, imgrsc->mipLevel);
+ cl_emit(rcl, TILE_RENDERING_MODE_CFG_COMMON, config) {
+ config.early_z_disable = true;
+ config.image_width_pixels = level_width;
+ config.image_height_pixels = level_height;
+ config.number_of_render_targets = 1;
+ config.multisample_mode_4x = false; /* FIXME */
+ config.maximum_bpp_of_all_render_targets = framebuffer->internal_bpp;
+ }
+
+ cl_emit(rcl, TILE_RENDERING_MODE_CFG_COLOR, rt) {
+ rt.render_target_0_internal_bpp = framebuffer->internal_bpp;
+ rt.render_target_0_internal_type = internal_type;
+ rt.render_target_0_clamp = V3D_RENDER_TARGET_CLAMP_NONE;
+ }
+
+ /* We always need to emit this, since it signals the end of the RCL config */
+ cl_emit(rcl, TILE_RENDERING_MODE_CFG_ZS_CLEAR_VALUES, clear) {
+ clear.z_clear_value = 0;
+ clear.stencil_clear_value = 0;
+ };
+
+ cl_emit(rcl, TILE_LIST_INITIAL_BLOCK_SIZE, init) {
+ init.use_auto_chained_tile_lists = true;
+ init.size_of_first_block_in_chained_tile_lists =
+ TILE_ALLOCATION_BLOCK_SIZE_64B;
+ }
+
+ uint32_t supertile_w_in_pixels =
+ framebuffer->tile_width * framebuffer->supertile_width;
+ uint32_t supertile_h_in_pixels =
+ framebuffer->tile_height * framebuffer->supertile_height;
+ const uint32_t min_x_supertile =
+ region->imageOffset.x / supertile_w_in_pixels;
+ const uint32_t min_y_supertile =
+ region->imageOffset.y / supertile_h_in_pixels;
+
+ const uint32_t max_render_x =
+ region->imageOffset.x + region->imageExtent.width - 1;
+ const uint32_t max_render_y =
+ region->imageOffset.y + region->imageExtent.height - 1;
+ const uint32_t max_x_supertile = max_render_x / supertile_w_in_pixels;
+ const uint32_t max_y_supertile = max_render_y / supertile_h_in_pixels;
+
+ for (int layer = 0; layer < imgrsc->layerCount; layer++) {
+ emit_copy_layer_to_buffer(cmd_buffer,
+ min_x_supertile, min_y_supertile,
+ max_x_supertile, max_y_supertile,
+ buffer, image, framebuffer,
+ layer,
+ region);
+ }
+
+ cl_emit(rcl, END_OF_RENDERING, end);
+}
+
+static void
+emit_copy_image_to_buffer_bcl(struct v3dv_cmd_buffer *cmd_buffer,
+ struct v3dv_framebuffer *framebuffer,
+ const VkBufferImageCopy *region)
+{
+ v3dv_cl_ensure_space_with_branch(&cmd_buffer->bcl, 256);
+
+ cl_emit(&cmd_buffer->bcl, NUMBER_OF_LAYERS, config) {
+ config.number_of_layers = framebuffer->layers;
+ }
+
+ cl_emit(&cmd_buffer->bcl, TILE_BINNING_MODE_CFG, config) {
+ config.width_in_pixels = framebuffer->width;
+ config.height_in_pixels = framebuffer->height;
+ config.number_of_render_targets = 1;
+ config.multisample_mode_4x = false; /* FIXME */
+ config.maximum_bpp_of_all_render_targets = framebuffer->internal_bpp;
+ }
+
+ cl_emit(&cmd_buffer->bcl, FLUSH_VCD_CACHE, bin);
+
+ cl_emit(&cmd_buffer->bcl, OCCLUSION_QUERY_COUNTER, counter);
+
+ cl_emit(&cmd_buffer->bcl, START_TILE_BINNING, bin);
+
+ cl_emit(&cmd_buffer->bcl, CLIP_WINDOW, clip) {
+ clip.clip_window_left_pixel_coordinate = region->imageOffset.x;
+ clip.clip_window_bottom_pixel_coordinate = region->imageOffset.y;
+ clip.clip_window_width_in_pixels = region->imageExtent.width;
+ clip.clip_window_height_in_pixels = region->imageExtent.height;
+ }
+
+ cl_emit(&cmd_buffer->bcl, FLUSH, flush);
+}
+
+/* Sets framebuffer dimensions and computes tile size parameters based on the
+ * maximum internal bpp provided.
+ */
+static void
+setup_framebuffer_params(struct v3dv_framebuffer *fb,
+ struct v3dv_image *image,
+ uint32_t layer_count,
+ uint32_t internal_bpp)
+{
+ fb->width = image->extent.width;
+ fb->height = image->extent.height;
+ fb->layers = layer_count;
+ fb->internal_bpp = MAX2(RENDER_TARGET_MAXIMUM_32BPP, internal_bpp);
+
+ /* We are only interested in the framebufer description required to compute
+ * the tiling setup parameters below, so we don't need real attachments,
+ * only the framebuffer size and the internal bpp.
+ */
+ fb->attachment_count = 0;
+
+ v3dv_framebuffer_compute_tiling_params(fb);
+}
+
+static inline bool
+can_use_tlb_copy_for_image_region(const VkBufferImageCopy *region)
+{
+ return region->imageOffset.x == 0 && region->imageOffset.y == 0;
+}
+
+/* Implements a copy using the TLB.
+ *
+ * This only works if we are copying from offset (0,0), since a TLB store for
+ * tile (x,y) will be written at the same tile offset into the destination.
+ * When this requirement is not met, we need to use a blit instead.
+ */
+static void
+copy_image_to_buffer_tlb(struct v3dv_cmd_buffer *cmd_buffer,
+ struct v3dv_buffer *buffer,
+ struct v3dv_image *image,
+ const VkBufferImageCopy *region)
+{
+ assert(can_use_tlb_copy_for_image_region(region));
+
+ /* FIXME: pre-compute this at image creation time? */
+ uint32_t internal_type;
+ uint32_t internal_bpp;
+ v3dv_get_internal_type_bpp_for_output_format(image->format->rt_type,
+ &internal_type,
+ &internal_bpp);
+
+ uint32_t num_layers = region->imageSubresource.layerCount;
+ assert(num_layers > 0);
+
+ struct v3dv_framebuffer framebuffer;
+ setup_framebuffer_params(&framebuffer, image, num_layers, internal_bpp);
+
+ /* FIXME: here we assume that we have a valid tile alloc/state setup,
+ * which is usually the case for copy after render scenarios. The
+ * code below simply checks and asserts this requirement,
+ * however, a proper implementation should allocate new tile
+ * alloc/state if we don't have one (for example if we haven't
+ * recorded a render pass yet) or the one we have isn't large
+ * enough. We still need to figure out how we want to handle
+ * varying tile alloc/state requirements in a command buffer.
+ */
+ uint32_t tile_alloc_size = 64 * num_layers *
+ framebuffer.draw_tiles_x *
+ framebuffer.draw_tiles_y;
+ tile_alloc_size = align(tile_alloc_size, 4096);
+ tile_alloc_size += 8192;
+ tile_alloc_size += 512 * 1024;
+ assert(cmd_buffer->tile_alloc &&
+ cmd_buffer->tile_alloc->size >= tile_alloc_size);
+
+ const uint32_t tsda_per_tile_size = 256;
+ const uint32_t tile_state_size = num_layers *
+ framebuffer.draw_tiles_x *
+ framebuffer.draw_tiles_y *
+ tsda_per_tile_size;
+ assert(cmd_buffer->tile_state &&
+ cmd_buffer->tile_state->size >= tile_state_size);
+
+ emit_copy_image_to_buffer_bcl(cmd_buffer, &framebuffer, region);
+ emit_copy_image_to_buffer_rcl(cmd_buffer, buffer, image,
+ &framebuffer, internal_type, region);
+}
+
void
v3dv_CmdCopyImageToBuffer(VkCommandBuffer commandBuffer,
VkImage srcImage,
@@ -31,5 +404,13 @@
uint32_t regionCount,
const VkBufferImageCopy *pRegions)
{
+ V3DV_FROM_HANDLE(v3dv_cmd_buffer, cmd_buffer, commandBuffer);
+ V3DV_FROM_HANDLE(v3dv_image, image, srcImage);
+ V3DV_FROM_HANDLE(v3dv_buffer, buffer, destBuffer);
+
+ for (uint32_t i = 0; i < regionCount; i++) {
+ if (can_use_tlb_copy_for_image_region(&pRegions[i]))
+ copy_image_to_buffer_tlb(cmd_buffer, buffer, image, &pRegions[i]);
+ }
}
