}
-static VkFormat pipeline_formats[] = {
- VK_FORMAT_R8G8B8A8_UNORM,
- VK_FORMAT_R8G8B8A8_UINT,
- VK_FORMAT_R8G8B8A8_SINT,
- VK_FORMAT_A2R10G10B10_UINT_PACK32,
- VK_FORMAT_A2R10G10B10_SINT_PACK32,
- VK_FORMAT_R16G16B16A16_UNORM,
- VK_FORMAT_R16G16B16A16_SNORM,
- VK_FORMAT_R16G16B16A16_UINT,
- VK_FORMAT_R16G16B16A16_SINT,
- VK_FORMAT_R32_SFLOAT,
- VK_FORMAT_R32G32_SFLOAT,
- VK_FORMAT_R32G32B32A32_SFLOAT
-};
-
VkResult
-radv_device_init_meta_resolve_state(struct radv_device *device)
+radv_device_init_meta_resolve_state(struct radv_device *device, bool on_demand)
{
+ if (on_demand)
+ return VK_SUCCESS;
+
VkResult res = VK_SUCCESS;
struct radv_meta_state *state = &device->meta_state;
struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
goto fail;
}
- for (uint32_t i = 0; i < ARRAY_SIZE(pipeline_formats); ++i) {
- VkFormat format = pipeline_formats[i];
+ for (uint32_t i = 0; i < NUM_META_FS_KEYS; ++i) {
+ VkFormat format = radv_fs_key_format_exemplars[i];
unsigned fs_key = radv_format_meta_fs_key(format);
res = create_pass(device, format, &state->resolve.pass[fs_key]);
if (res != VK_SUCCESS)
RESOLVE_FRAGMENT,
};
-static void radv_pick_resolve_method_images(struct radv_image *src_image,
+static void radv_pick_resolve_method_images(struct radv_device *device,
+ struct radv_image *src_image,
+ VkFormat src_format,
struct radv_image *dest_image,
VkImageLayout dest_image_layout,
+ bool dest_render_loop,
struct radv_cmd_buffer *cmd_buffer,
enum radv_resolve_method *method)
cmd_buffer->queue_family_index,
cmd_buffer->queue_family_index);
- if (vk_format_is_int(src_image->vk_format))
- *method = RESOLVE_COMPUTE;
+ if (vk_format_is_color(src_format)) {
+ if (src_format == VK_FORMAT_R16G16_UNORM ||
+ src_format == VK_FORMAT_R16G16_SNORM)
+ *method = RESOLVE_COMPUTE;
+ else if (vk_format_is_int(src_format))
+ *method = RESOLVE_COMPUTE;
+ else if (src_image->info.array_size > 1 ||
+ dest_image->info.array_size > 1)
+ *method = RESOLVE_COMPUTE;
- if (radv_layout_dcc_compressed(dest_image, dest_image_layout, queue_mask)) {
- *method = RESOLVE_FRAGMENT;
- } else if (dest_image->surface.micro_tile_mode != src_image->surface.micro_tile_mode) {
- *method = RESOLVE_COMPUTE;
+ if (radv_layout_dcc_compressed(device, dest_image, dest_image_layout,
+ dest_render_loop, queue_mask)) {
+ *method = RESOLVE_FRAGMENT;
+ } else if (dest_image->planes[0].surface.micro_tile_mode !=
+ src_image->planes[0].surface.micro_tile_mode) {
+ *method = RESOLVE_COMPUTE;
+ }
+ } else {
+ if (src_image->info.array_size > 1 ||
+ dest_image->info.array_size > 1)
+ *method = RESOLVE_COMPUTE;
+ else
+ *method = RESOLVE_FRAGMENT;
+ }
+}
+
+static VkResult
+build_resolve_pipeline(struct radv_device *device,
+ unsigned fs_key)
+{
+ VkResult result = VK_SUCCESS;
+
+ if (device->meta_state.resolve.pipeline[fs_key])
+ return result;
+
+ mtx_lock(&device->meta_state.mtx);
+ if (device->meta_state.resolve.pipeline[fs_key]) {
+ mtx_unlock(&device->meta_state.mtx);
+ return result;
}
+
+ struct radv_shader_module vs_module = { .nir = radv_meta_build_nir_vs_generate_vertices() };
+
+ result = create_pass(device, radv_fs_key_format_exemplars[fs_key], &device->meta_state.resolve.pass[fs_key]);
+ if (result != VK_SUCCESS)
+ goto fail;
+
+ VkShaderModule vs_module_h = radv_shader_module_to_handle(&vs_module);
+ result = create_pipeline(device, vs_module_h, &device->meta_state.resolve.pipeline[fs_key], device->meta_state.resolve.pass[fs_key]);
+
+fail:
+ ralloc_free(vs_module.nir);
+ mtx_unlock(&device->meta_state.mtx);
+ return result;
}
void radv_CmdResolveImage(
} else
resolve_method = RESOLVE_COMPUTE;
- radv_pick_resolve_method_images(src_image, dest_image,
- dest_image_layout, cmd_buffer,
+ radv_pick_resolve_method_images(cmd_buffer->device, src_image,
+ src_image->vk_format, dest_image,
+ dest_image_layout, false, cmd_buffer,
&resolve_method);
if (resolve_method == RESOLVE_FRAGMENT) {
if (resolve_method == RESOLVE_COMPUTE) {
radv_meta_resolve_compute_image(cmd_buffer,
src_image,
+ src_image->vk_format,
src_image_layout,
dest_image,
+ dest_image->vk_format,
dest_image_layout,
region_count, regions);
return;
}
assert(dest_image->info.samples == 1);
- if (src_image->info.samples >= 16) {
- /* See commit aa3f9aaf31e9056a255f9e0472ebdfdaa60abe54 for the
- * glBlitFramebuffer workaround for samples >= 16.
- */
- radv_finishme("vkCmdResolveImage: need interpolation workaround when "
- "samples >= 16");
- }
-
if (src_image->info.array_size > 1)
radv_finishme("vkCmdResolveImage: multisample array images");
- if (dest_image->surface.dcc_size) {
- radv_initialize_dcc(cmd_buffer, dest_image, 0xffffffff);
- }
unsigned fs_key = radv_format_meta_fs_key(dest_image->vk_format);
for (uint32_t r = 0; r < region_count; ++r) {
const VkImageResolve *region = ®ions[r];
const struct VkOffset3D dstOffset =
radv_sanitize_image_offset(dest_image->type, region->dstOffset);
+ if (radv_dcc_enabled(dest_image, region->dstSubresource.mipLevel)) {
+ VkImageSubresourceRange range = {
+ .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
+ .baseMipLevel = region->dstSubresource.mipLevel,
+ .levelCount = 1,
+ .baseArrayLayer = dest_base_layer,
+ .layerCount = region->dstSubresource.layerCount,
+ };
+
+ radv_initialize_dcc(cmd_buffer, dest_image, &range, 0xffffffff);
+ }
for (uint32_t layer = 0; layer < region->srcSubresource.layerCount;
++layer) {
+ VkResult ret = build_resolve_pipeline(device, fs_key);
+ if (ret != VK_SUCCESS) {
+ cmd_buffer->record_result = ret;
+ break;
+ }
+
struct radv_image_view src_iview;
radv_image_view_init(&src_iview, cmd_buffer->device,
&(VkImageViewCreateInfo) {
.baseArrayLayer = src_base_layer + layer,
.layerCount = 1,
},
- });
+ }, NULL);
struct radv_image_view dest_iview;
radv_image_view_init(&dest_iview, cmd_buffer->device,
.baseArrayLayer = dest_base_layer + layer,
.layerCount = 1,
},
- });
+ }, NULL);
VkFramebuffer fb_h;
radv_CreateFramebuffer(device_h,
struct radv_meta_saved_state saved_state;
enum radv_resolve_method resolve_method = RESOLVE_HW;
- /* FINISHME(perf): Skip clears for resolve attachments.
- *
- * From the Vulkan 1.0 spec:
- *
- * If the first use of an attachment in a render pass is as a resolve
- * attachment, then the loadOp is effectively ignored as the resolve is
- * guaranteed to overwrite all pixels in the render area.
- */
+ if (subpass->ds_resolve_attachment) {
+ struct radv_subpass_attachment src_att = *subpass->depth_stencil_attachment;
+ struct radv_subpass_attachment dst_att = *subpass->ds_resolve_attachment;
+ struct radv_image_view *src_iview =
+ cmd_buffer->state.attachments[src_att.attachment].iview;
+ struct radv_image_view *dst_iview =
+ cmd_buffer->state.attachments[dst_att.attachment].iview;
+
+ radv_pick_resolve_method_images(cmd_buffer->device,
+ src_iview->image,
+ src_iview->vk_format,
+ dst_iview->image,
+ dst_att.layout,
+ dst_att.in_render_loop,
+ cmd_buffer,
+ &resolve_method);
+
+ if ((src_iview->aspect_mask & VK_IMAGE_ASPECT_DEPTH_BIT) &&
+ subpass->depth_resolve_mode != VK_RESOLVE_MODE_NONE_KHR) {
+ if (resolve_method == RESOLVE_FRAGMENT) {
+ radv_depth_stencil_resolve_subpass_fs(cmd_buffer,
+ VK_IMAGE_ASPECT_DEPTH_BIT,
+ subpass->depth_resolve_mode);
+ } else {
+ assert(resolve_method == RESOLVE_COMPUTE);
+ radv_depth_stencil_resolve_subpass_cs(cmd_buffer,
+ VK_IMAGE_ASPECT_DEPTH_BIT,
+ subpass->depth_resolve_mode);
+ }
+ }
+
+ if ((src_iview->aspect_mask & VK_IMAGE_ASPECT_STENCIL_BIT) &&
+ subpass->stencil_resolve_mode != VK_RESOLVE_MODE_NONE_KHR) {
+ if (resolve_method == RESOLVE_FRAGMENT) {
+ radv_depth_stencil_resolve_subpass_fs(cmd_buffer,
+ VK_IMAGE_ASPECT_STENCIL_BIT,
+ subpass->stencil_resolve_mode);
+ } else {
+ assert(resolve_method == RESOLVE_COMPUTE);
+ radv_depth_stencil_resolve_subpass_cs(cmd_buffer,
+ VK_IMAGE_ASPECT_STENCIL_BIT,
+ subpass->stencil_resolve_mode);
+ }
+ }
+ }
- if (!subpass->has_resolve)
+ if (!subpass->has_color_resolve)
return;
for (uint32_t i = 0; i < subpass->color_count; ++i) {
- VkAttachmentReference src_att = subpass->color_attachments[i];
- VkAttachmentReference dest_att = subpass->resolve_attachments[i];
+ struct radv_subpass_attachment src_att = subpass->color_attachments[i];
+ struct radv_subpass_attachment dest_att = subpass->resolve_attachments[i];
- if (src_att.attachment == VK_ATTACHMENT_UNUSED ||
- dest_att.attachment == VK_ATTACHMENT_UNUSED)
+ if (dest_att.attachment == VK_ATTACHMENT_UNUSED)
continue;
- struct radv_image *dst_img = cmd_buffer->state.framebuffer->attachments[dest_att.attachment].attachment->image;
- struct radv_image *src_img = cmd_buffer->state.framebuffer->attachments[src_att.attachment].attachment->image;
+ /* Make sure to not clear color attachments after resolves. */
+ cmd_buffer->state.attachments[dest_att.attachment].pending_clear_aspects = 0;
+
+ struct radv_image *dst_img = cmd_buffer->state.attachments[dest_att.attachment].iview->image;
+ struct radv_image_view *src_iview= cmd_buffer->state.attachments[src_att.attachment].iview;
+ struct radv_image *src_img = src_iview->image;
+
+ radv_pick_resolve_method_images(cmd_buffer->device, src_img,
+ src_iview->vk_format, dst_img,
+ dest_att.layout,
+ dest_att.in_render_loop,
+ cmd_buffer, &resolve_method);
- radv_pick_resolve_method_images(dst_img, src_img, dest_att.layout, cmd_buffer, &resolve_method);
if (resolve_method == RESOLVE_FRAGMENT) {
break;
}
RADV_META_SAVE_GRAPHICS_PIPELINE);
for (uint32_t i = 0; i < subpass->color_count; ++i) {
- VkAttachmentReference src_att = subpass->color_attachments[i];
- VkAttachmentReference dest_att = subpass->resolve_attachments[i];
+ struct radv_subpass_attachment src_att = subpass->color_attachments[i];
+ struct radv_subpass_attachment dest_att = subpass->resolve_attachments[i];
- if (src_att.attachment == VK_ATTACHMENT_UNUSED ||
- dest_att.attachment == VK_ATTACHMENT_UNUSED)
+ if (dest_att.attachment == VK_ATTACHMENT_UNUSED)
continue;
- struct radv_image *dst_img = cmd_buffer->state.framebuffer->attachments[dest_att.attachment].attachment->image;
+ struct radv_image_view *dest_iview = cmd_buffer->state.attachments[dest_att.attachment].iview;
+ struct radv_image *dst_img = dest_iview->image;
+
+ if (radv_dcc_enabled(dst_img, dest_iview->base_mip)) {
+ VkImageSubresourceRange range = {
+ .aspectMask = dest_iview->aspect_mask,
+ .baseMipLevel = dest_iview->base_mip,
+ .levelCount = dest_iview->level_count,
+ .baseArrayLayer = dest_iview->base_layer,
+ .layerCount = dest_iview->layer_count,
+ };
- if (dst_img->surface.dcc_size) {
- radv_initialize_dcc(cmd_buffer, dst_img, 0xffffffff);
+ radv_initialize_dcc(cmd_buffer, dst_img, &range, 0xffffffff);
cmd_buffer->state.attachments[dest_att.attachment].current_layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;
}
struct radv_subpass resolve_subpass = {
.color_count = 2,
- .color_attachments = (VkAttachmentReference[]) { src_att, dest_att },
- .depth_stencil_attachment = { .attachment = VK_ATTACHMENT_UNUSED },
+ .color_attachments = (struct radv_subpass_attachment[]) { src_att, dest_att },
+ .depth_stencil_attachment = NULL,
};
- radv_cmd_buffer_set_subpass(cmd_buffer, &resolve_subpass, false);
+ radv_cmd_buffer_set_subpass(cmd_buffer, &resolve_subpass);
+
+ VkResult ret = build_resolve_pipeline(cmd_buffer->device, radv_format_meta_fs_key(dest_iview->vk_format));
+ if (ret != VK_SUCCESS) {
+ cmd_buffer->record_result = ret;
+ continue;
+ }
emit_resolve(cmd_buffer,
- dst_img->vk_format,
+ dest_iview->vk_format,
&(VkOffset2D) { 0, 0 },
&(VkExtent2D) { fb->width, fb->height });
}
- cmd_buffer->state.subpass = subpass;
+ radv_cmd_buffer_set_subpass(cmd_buffer, subpass);
+
radv_meta_restore(&saved_state, cmd_buffer);
}
+
+/**
+ * Decompress CMask/FMask before resolving a multisampled source image inside a
+ * subpass.
+ */
+void
+radv_decompress_resolve_subpass_src(struct radv_cmd_buffer *cmd_buffer)
+{
+ const struct radv_subpass *subpass = cmd_buffer->state.subpass;
+ struct radv_framebuffer *fb = cmd_buffer->state.framebuffer;
+ uint32_t layer_count = fb->layers;
+
+ if (subpass->view_mask)
+ layer_count = util_last_bit(subpass->view_mask);
+
+ for (uint32_t i = 0; i < subpass->color_count; ++i) {
+ struct radv_subpass_attachment src_att = subpass->color_attachments[i];
+ struct radv_subpass_attachment dest_att = subpass->resolve_attachments[i];
+
+ if (dest_att.attachment == VK_ATTACHMENT_UNUSED)
+ continue;
+
+ struct radv_image_view *src_iview = cmd_buffer->state.attachments[src_att.attachment].iview;
+ struct radv_image *src_image = src_iview->image;
+
+ VkImageResolve region = {};
+ region.srcSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
+ region.srcSubresource.mipLevel = 0;
+ region.srcSubresource.baseArrayLayer = src_iview->base_layer;
+ region.srcSubresource.layerCount = layer_count;
+
+ radv_decompress_resolve_src(cmd_buffer, src_image,
+ src_att.layout, 1, ®ion);
+ }
+
+ if (subpass->ds_resolve_attachment) {
+ struct radv_subpass_attachment src_att = *subpass->depth_stencil_attachment;
+ struct radv_image_view *src_iview = fb->attachments[src_att.attachment];
+ struct radv_image *src_image = src_iview->image;
+
+ VkImageResolve region = {};
+ region.srcSubresource.aspectMask = VK_IMAGE_ASPECT_DEPTH_BIT;
+ region.srcSubresource.mipLevel = 0;
+ region.srcSubresource.baseArrayLayer = src_iview->base_layer;
+ region.srcSubresource.layerCount = layer_count;
+
+ radv_decompress_resolve_src(cmd_buffer, src_image,
+ src_att.layout, 1, ®ion);
+ }
+}
+
+static struct radv_sample_locations_state *
+radv_get_resolve_sample_locations(struct radv_cmd_buffer *cmd_buffer)
+{
+ struct radv_cmd_state *state = &cmd_buffer->state;
+ uint32_t subpass_id = radv_get_subpass_id(cmd_buffer);
+
+ for (uint32_t i = 0; i < state->num_subpass_sample_locs; i++) {
+ if (state->subpass_sample_locs[i].subpass_idx == subpass_id)
+ return &state->subpass_sample_locs[i].sample_location;
+ }
+
+ return NULL;
+}
+
+/**
+ * Decompress CMask/FMask before resolving a multisampled source image.
+ */
+void
+radv_decompress_resolve_src(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *src_image,
+ VkImageLayout src_image_layout,
+ uint32_t region_count,
+ const VkImageResolve *regions)
+{
+ for (uint32_t r = 0; r < region_count; ++r) {
+ const VkImageResolve *region = ®ions[r];
+ const uint32_t src_base_layer =
+ radv_meta_get_iview_layer(src_image, ®ion->srcSubresource,
+ ®ion->srcOffset);
+
+ VkImageMemoryBarrier barrier = {};
+ barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT;
+ barrier.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT;
+ barrier.oldLayout = src_image_layout;
+ barrier.newLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
+ barrier.image = radv_image_to_handle(src_image);
+ barrier.subresourceRange = (VkImageSubresourceRange) {
+ .aspectMask = region->srcSubresource.aspectMask,
+ .baseMipLevel = region->srcSubresource.mipLevel,
+ .levelCount = 1,
+ .baseArrayLayer = src_base_layer,
+ .layerCount = region->srcSubresource.layerCount,
+ };
+
+ if (src_image->flags & VK_IMAGE_CREATE_SAMPLE_LOCATIONS_COMPATIBLE_DEPTH_BIT_EXT) {
+ /* If the depth/stencil image uses different sample
+ * locations, we need them during HTILE decompressions.
+ */
+ struct radv_sample_locations_state *sample_locs =
+ radv_get_resolve_sample_locations(cmd_buffer);
+
+ barrier.pNext = &(VkSampleLocationsInfoEXT) {
+ .sType = VK_STRUCTURE_TYPE_SAMPLE_LOCATIONS_INFO_EXT,
+ .sampleLocationsPerPixel = sample_locs->per_pixel,
+ .sampleLocationGridSize = sample_locs->grid_size,
+ .sampleLocationsCount = sample_locs->count,
+ .pSampleLocations = sample_locs->locations,
+ };
+ }
+
+ radv_CmdPipelineBarrier(radv_cmd_buffer_to_handle(cmd_buffer),
+ VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
+ VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT,
+ false, 0, NULL, 0, NULL, 1, &barrier);
+ }
+}
projects / mesa.git / blobdiff