From 280f334f81c439d391d9934f9cf1791f074c3773 Mon Sep 17 00:00:00 2001 From: reduz Date: Mon, 25 Jan 2021 21:52:58 -0300 Subject: Reorganize RenderingDevice barriers -Removed sync to draw, now everything syncs to draw by default. -Fixed many validation layer errors. -Added support for VkImageViewUsageCreateInfo to fix validation layer warnings. -Texture, buffer, raster and compute functions now all allow spcifying which barriers will be used. --- drivers/vulkan/rendering_device_vulkan.cpp | 381 +++++++++++++++++++++++------ drivers/vulkan/rendering_device_vulkan.h | 21 +- drivers/vulkan/vulkan_context.cpp | 19 ++ 3 files changed, 331 insertions(+), 90 deletions(-) (limited to 'drivers/vulkan') diff --git a/drivers/vulkan/rendering_device_vulkan.cpp b/drivers/vulkan/rendering_device_vulkan.cpp index 0689b3f2dd..ef331ec4b6 100644 --- a/drivers/vulkan/rendering_device_vulkan.cpp +++ b/drivers/vulkan/rendering_device_vulkan.cpp @@ -41,28 +41,60 @@ //#define FORCE_FULL_BARRIER // Get the Vulkan object information and possible stage access types (bitwise OR'd with incoming values) -RenderingDeviceVulkan::Buffer *RenderingDeviceVulkan::_get_buffer_from_owner(RID p_buffer, VkPipelineStageFlags &stage_mask, VkAccessFlags &access_mask) { +RenderingDeviceVulkan::Buffer *RenderingDeviceVulkan::_get_buffer_from_owner(RID p_buffer, VkPipelineStageFlags &r_stage_mask, VkAccessFlags &r_access_mask, uint32_t p_post_barrier) { Buffer *buffer = nullptr; if (vertex_buffer_owner.owns(p_buffer)) { - stage_mask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT; - access_mask |= VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT; buffer = vertex_buffer_owner.getornull(p_buffer); + + r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT; + r_access_mask |= VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT; + if (buffer->usage & VK_BUFFER_USAGE_STORAGE_BUFFER_BIT) { + if (p_post_barrier & BARRIER_MASK_RASTER) { + r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + } + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + } + } } else if (index_buffer_owner.owns(p_buffer)) { - stage_mask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT; - access_mask |= VK_ACCESS_INDEX_READ_BIT; + r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT; + r_access_mask |= VK_ACCESS_INDEX_READ_BIT; buffer = index_buffer_owner.getornull(p_buffer); } else if (uniform_buffer_owner.owns(p_buffer)) { - stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; - access_mask |= VK_ACCESS_UNIFORM_READ_BIT; + if (p_post_barrier & BARRIER_MASK_RASTER) { + r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + } + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + } + r_access_mask |= VK_ACCESS_UNIFORM_READ_BIT; buffer = uniform_buffer_owner.getornull(p_buffer); } else if (texture_buffer_owner.owns(p_buffer)) { - stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; - access_mask |= VK_ACCESS_SHADER_READ_BIT; + if (p_post_barrier & BARRIER_MASK_RASTER) { + r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + } + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + } + r_access_mask |= VK_ACCESS_SHADER_READ_BIT; buffer = &texture_buffer_owner.getornull(p_buffer)->buffer; } else if (storage_buffer_owner.owns(p_buffer)) { - stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; - access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; buffer = storage_buffer_owner.getornull(p_buffer); + if (p_post_barrier & BARRIER_MASK_RASTER) { + r_stage_mask |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + } + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + r_stage_mask |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + r_access_mask |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + } + + if (buffer->usage & VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT) { + r_stage_mask |= VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT; + r_access_mask |= VK_ACCESS_INDIRECT_COMMAND_READ_BIT; + } } return buffer; } @@ -2067,6 +2099,48 @@ RID RenderingDeviceVulkan::texture_create_shared(const TextureView &p_view, RID image_view_create_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT; } + VkImageViewUsageCreateInfo usage_info; + usage_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO; + usage_info.pNext = nullptr; + if (p_view.format_override != DATA_FORMAT_MAX) { + //need to validate usage with vulkan + + usage_info.usage = 0; + + if (texture.usage_flags & TEXTURE_USAGE_SAMPLING_BIT) { + usage_info.usage |= VK_IMAGE_USAGE_SAMPLED_BIT; + } + + if (texture.usage_flags & TEXTURE_USAGE_STORAGE_BIT) { + if (texture_is_format_supported_for_usage(p_view.format_override, TEXTURE_USAGE_STORAGE_BIT)) { + usage_info.usage |= VK_IMAGE_USAGE_STORAGE_BIT; + } + } + + if (texture.usage_flags & TEXTURE_USAGE_COLOR_ATTACHMENT_BIT) { + if (texture_is_format_supported_for_usage(p_view.format_override, TEXTURE_USAGE_COLOR_ATTACHMENT_BIT)) { + usage_info.usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; + } + } + + if (texture.usage_flags & TEXTURE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT) { + usage_info.usage |= VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT; + } + + if (texture.usage_flags & TEXTURE_USAGE_CAN_UPDATE_BIT) { + usage_info.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT; + } + if (texture.usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT) { + usage_info.usage |= VK_IMAGE_USAGE_TRANSFER_SRC_BIT; + } + + if (texture.usage_flags & TEXTURE_USAGE_CAN_COPY_TO_BIT) { + usage_info.usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT; + } + + image_view_create_info.pNext = &usage_info; + } + VkResult err = vkCreateImageView(device, &image_view_create_info, nullptr, &texture.view); ERR_FAIL_COND_V_MSG(err, RID(), "vkCreateImageView failed with error " + itos(err) + "."); @@ -2196,11 +2270,11 @@ RID RenderingDeviceVulkan::texture_create_shared_from_slice(const TextureView &p return id; } -Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, const Vector &p_data, bool p_sync_with_draw) { +Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, const Vector &p_data, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ - ERR_FAIL_COND_V_MSG(draw_list && p_sync_with_draw, ERR_INVALID_PARAMETER, - "Updating textures in 'sync to draw' mode is forbidden during creation of a draw list"); + ERR_FAIL_COND_V_MSG(draw_list || compute_list, ERR_INVALID_PARAMETER, + "Updating textures in is forbidden during creation of a draw or compute list"); Texture *texture = texture_owner.getornull(p_texture); ERR_FAIL_COND_V(!texture, ERR_INVALID_PARAMETER); @@ -2241,7 +2315,7 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con const uint8_t *r = p_data.ptr(); - VkCommandBuffer command_buffer = p_sync_with_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer; + VkCommandBuffer command_buffer = p_post_barrier ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer; //barrier to transfer { @@ -2266,6 +2340,10 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con } uint32_t mipmap_offset = 0; + + uint32_t logic_width = texture->width; + uint32_t logic_height = texture->height; + for (uint32_t mm_i = 0; mm_i < texture->mipmaps; mm_i++) { uint32_t depth; uint32_t image_total = get_image_format_required_size(texture->format, texture->width, texture->height, texture->depth, mm_i + 1, &width, &height, &depth); @@ -2282,12 +2360,15 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con uint32_t region_w = MIN(region_size, width - x); uint32_t region_h = MIN(region_size, height - y); + uint32_t region_logic_w = MIN(region_size, logic_width - x); + uint32_t region_logic_h = MIN(region_size, logic_height - y); + uint32_t pixel_size = get_image_format_pixel_size(texture->format); uint32_t to_allocate = region_w * region_h * pixel_size; to_allocate >>= get_compressed_image_format_pixel_rshift(texture->format); uint32_t alloc_offset, alloc_size; - Error err = _staging_buffer_allocate(to_allocate, required_align, alloc_offset, alloc_size, false, p_sync_with_draw); + Error err = _staging_buffer_allocate(to_allocate, required_align, alloc_offset, alloc_size, false, p_post_barrier); ERR_FAIL_COND_V(err, ERR_CANT_CREATE); uint8_t *write_ptr; @@ -2363,8 +2444,8 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con buffer_image_copy.imageOffset.y = y; buffer_image_copy.imageOffset.z = z; - buffer_image_copy.imageExtent.width = region_w; - buffer_image_copy.imageExtent.height = region_h; + buffer_image_copy.imageExtent.width = region_logic_w; + buffer_image_copy.imageExtent.height = region_logic_h; buffer_image_copy.imageExtent.depth = 1; vkCmdCopyBufferToImage(command_buffer, staging_buffer_blocks[staging_buffer_current].buffer, texture->image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &buffer_image_copy); @@ -2375,15 +2456,32 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con } mipmap_offset = image_total; + logic_width = MAX(1, logic_width >> 1); + logic_height = MAX(1, logic_height >> 1); } //barrier to restore layout { + uint32_t barrier_flags = 0; + uint32_t access_flags = 0; + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT; + } + if (p_post_barrier & BARRIER_MASK_RASTER) { + barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT; + } + if (p_post_barrier & BARRIER_MASK_TRANSFER) { + barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; + access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; + } + VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; - image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; image_memory_barrier.newLayout = texture->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; @@ -2395,7 +2493,7 @@ Error RenderingDeviceVulkan::texture_update(RID p_texture, uint32_t p_layer, con image_memory_barrier.subresourceRange.baseArrayLayer = p_layer; image_memory_barrier.subresourceRange.layerCount = 1; - vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); + vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } return OK; @@ -2608,13 +2706,13 @@ bool RenderingDeviceVulkan::texture_is_valid(RID p_texture) { return texture_owner.owns(p_texture); } -Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, const Vector3 &p_from, const Vector3 &p_to, const Vector3 &p_size, uint32_t p_src_mipmap, uint32_t p_dst_mipmap, uint32_t p_src_layer, uint32_t p_dst_layer, bool p_sync_with_draw) { +Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, const Vector3 &p_from, const Vector3 &p_to, const Vector3 &p_size, uint32_t p_src_mipmap, uint32_t p_dst_mipmap, uint32_t p_src_layer, uint32_t p_dst_layer, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ Texture *src_tex = texture_owner.getornull(p_from_texture); ERR_FAIL_COND_V(!src_tex, ERR_INVALID_PARAMETER); - ERR_FAIL_COND_V_MSG(p_sync_with_draw && src_tex->bound, ERR_INVALID_PARAMETER, + ERR_FAIL_COND_V_MSG(src_tex->bound, ERR_INVALID_PARAMETER, "Source texture can't be copied while a render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(src_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT), ERR_INVALID_PARAMETER, "Source texture requires the TEXTURE_USAGE_CAN_COPY_FROM_BIT in order to be retrieved."); @@ -2635,7 +2733,7 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, Texture *dst_tex = texture_owner.getornull(p_to_texture); ERR_FAIL_COND_V(!dst_tex, ERR_INVALID_PARAMETER); - ERR_FAIL_COND_V_MSG(p_sync_with_draw && dst_tex->bound, ERR_INVALID_PARAMETER, + ERR_FAIL_COND_V_MSG(dst_tex->bound, ERR_INVALID_PARAMETER, "Destination texture can't be copied while a render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(dst_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_TO_BIT), ERR_INVALID_PARAMETER, "Destination texture requires the TEXTURE_USAGE_CAN_COPY_TO_BIT in order to be retrieved."); @@ -2656,7 +2754,7 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, ERR_FAIL_COND_V_MSG(src_tex->read_aspect_mask != dst_tex->read_aspect_mask, ERR_INVALID_PARAMETER, "Source and destination texture must be of the same type (color or depth)."); - VkCommandBuffer command_buffer = p_sync_with_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer; + VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; { //PRE Copy the image @@ -2731,12 +2829,27 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, // RESTORE LAYOUT for SRC and DST + uint32_t barrier_flags = 0; + uint32_t access_flags = 0; + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + } + if (p_post_barrier & BARRIER_MASK_RASTER) { + barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + } + if (p_post_barrier & BARRIER_MASK_TRANSFER) { + barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; + access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; + } + { //restore src VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT; - image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL; image_memory_barrier.newLayout = src_tex->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; @@ -2748,7 +2861,7 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, image_memory_barrier.subresourceRange.baseArrayLayer = p_src_layer; image_memory_barrier.subresourceRange.layerCount = 1; - vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); + vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } { //make dst readable @@ -2757,7 +2870,7 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; - image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; image_memory_barrier.newLayout = dst_tex->layout; @@ -2770,20 +2883,20 @@ Error RenderingDeviceVulkan::texture_copy(RID p_from_texture, RID p_to_texture, image_memory_barrier.subresourceRange.baseArrayLayer = p_src_layer; image_memory_barrier.subresourceRange.layerCount = 1; - vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); + vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } } return OK; } -Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID p_to_texture, bool p_sync_with_draw) { +Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID p_to_texture, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ Texture *src_tex = texture_owner.getornull(p_from_texture); ERR_FAIL_COND_V(!src_tex, ERR_INVALID_PARAMETER); - ERR_FAIL_COND_V_MSG(p_sync_with_draw && src_tex->bound, ERR_INVALID_PARAMETER, + ERR_FAIL_COND_V_MSG(src_tex->bound, ERR_INVALID_PARAMETER, "Source texture can't be copied while a render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(src_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_FROM_BIT), ERR_INVALID_PARAMETER, "Source texture requires the TEXTURE_USAGE_CAN_COPY_FROM_BIT in order to be retrieved."); @@ -2794,7 +2907,7 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID Texture *dst_tex = texture_owner.getornull(p_to_texture); ERR_FAIL_COND_V(!dst_tex, ERR_INVALID_PARAMETER); - ERR_FAIL_COND_V_MSG(p_sync_with_draw && dst_tex->bound, ERR_INVALID_PARAMETER, + ERR_FAIL_COND_V_MSG(dst_tex->bound, ERR_INVALID_PARAMETER, "Destination texture can't be copied while a render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V_MSG(!(dst_tex->usage_flags & TEXTURE_USAGE_CAN_COPY_TO_BIT), ERR_INVALID_PARAMETER, "Destination texture requires the TEXTURE_USAGE_CAN_COPY_TO_BIT in order to be retrieved."); @@ -2808,7 +2921,7 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID ERR_FAIL_COND_V_MSG(src_tex->read_aspect_mask != dst_tex->read_aspect_mask, ERR_INVALID_PARAMETER, "Source and destination texture must be of the same type (color or depth)."); - VkCommandBuffer command_buffer = p_sync_with_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer; + VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; { //PRE Copy the image @@ -2883,12 +2996,27 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID // RESTORE LAYOUT for SRC and DST + uint32_t barrier_flags = 0; + uint32_t access_flags = 0; + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + } + if (p_post_barrier & BARRIER_MASK_RASTER) { + barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + } + if (p_post_barrier & BARRIER_MASK_TRANSFER) { + barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; + access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; + } + { //restore src VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_READ_BIT; - image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL; image_memory_barrier.newLayout = src_tex->layout; image_memory_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; @@ -2900,7 +3028,7 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer; image_memory_barrier.subresourceRange.layerCount = 1; - vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); + vkCmdPipelineBarrier(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } { //make dst readable @@ -2909,7 +3037,7 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; - image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; image_memory_barrier.newLayout = dst_tex->layout; @@ -2922,20 +3050,20 @@ Error RenderingDeviceVulkan::texture_resolve_multisample(RID p_from_texture, RID image_memory_barrier.subresourceRange.baseArrayLayer = dst_tex->base_layer; image_memory_barrier.subresourceRange.layerCount = 1; - vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); + vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } } return OK; } -Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color, uint32_t p_base_mipmap, uint32_t p_mipmaps, uint32_t p_base_layer, uint32_t p_layers, bool p_sync_with_draw) { +Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color, uint32_t p_base_mipmap, uint32_t p_mipmaps, uint32_t p_base_layer, uint32_t p_layers, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ Texture *src_tex = texture_owner.getornull(p_texture); ERR_FAIL_COND_V(!src_tex, ERR_INVALID_PARAMETER); - ERR_FAIL_COND_V_MSG(p_sync_with_draw && src_tex->bound, ERR_INVALID_PARAMETER, + ERR_FAIL_COND_V_MSG(src_tex->bound, ERR_INVALID_PARAMETER, "Source texture can't be cleared while a render pass that uses it is being created. Ensure render pass is finalized (and that it was created with RENDER_PASS_CONTENTS_FINISH) to unbind this texture."); ERR_FAIL_COND_V(p_layers == 0, ERR_INVALID_PARAMETER); @@ -2952,7 +3080,7 @@ Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color, ERR_FAIL_COND_V(p_base_mipmap + p_mipmaps > src_tex->mipmaps, ERR_INVALID_PARAMETER); ERR_FAIL_COND_V(p_base_layer + p_layers > src_layer_count, ERR_INVALID_PARAMETER); - VkCommandBuffer command_buffer = p_sync_with_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer; + VkCommandBuffer command_buffer = frames[frame].draw_command_buffer; VkImageLayout clear_layout = (src_tex->layout == VK_IMAGE_LAYOUT_GENERAL) ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL; @@ -2999,11 +3127,27 @@ Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color, vkCmdClearColorImage(command_buffer, src_tex->image, clear_layout, &clear_color, 1, &range); { // Barrier to post clear accesses (changing back the layout if needed) + + uint32_t barrier_flags = 0; + uint32_t access_flags = 0; + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + } + if (p_post_barrier & BARRIER_MASK_RASTER) { + barrier_flags |= VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + } + if (p_post_barrier & BARRIER_MASK_TRANSFER) { + barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; + access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; + } + VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_TRANSFER_WRITE_BIT; - image_memory_barrier.dstAccessMask = valid_texture_access; + image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = clear_layout; image_memory_barrier.newLayout = src_tex->layout; @@ -3016,7 +3160,7 @@ Error RenderingDeviceVulkan::texture_clear(RID p_texture, const Color &p_color, image_memory_barrier.subresourceRange.baseArrayLayer = src_tex->base_layer + p_base_layer; image_memory_barrier.subresourceRange.layerCount = p_layers; - vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, valid_texture_stages, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); + vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); } return OK; @@ -5038,19 +5182,22 @@ bool RenderingDeviceVulkan::uniform_set_is_valid(RID p_uniform_set) { return uniform_set_owner.owns(p_uniform_set); } -Error RenderingDeviceVulkan::buffer_update(RID p_buffer, uint32_t p_offset, uint32_t p_size, const void *p_data, bool p_sync_with_draw) { +Error RenderingDeviceVulkan::buffer_update(RID p_buffer, uint32_t p_offset, uint32_t p_size, const void *p_data, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ - ERR_FAIL_COND_V_MSG(draw_list && p_sync_with_draw, ERR_INVALID_PARAMETER, - "Updating buffers in 'sync to draw' mode is forbidden during creation of a draw list"); - ERR_FAIL_COND_V_MSG(compute_list && p_sync_with_draw, ERR_INVALID_PARAMETER, - "Updating buffers in 'sync to draw' mode is forbidden during creation of a compute list"); + ERR_FAIL_COND_V_MSG(draw_list, ERR_INVALID_PARAMETER, + "Updating buffers is forbidden during creation of a draw list"); + ERR_FAIL_COND_V_MSG(compute_list, ERR_INVALID_PARAMETER, + "Updating buffers is forbidden during creation of a compute list"); - // Protect subsequent updates... - VkPipelineStageFlags dst_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT; - VkAccessFlags dst_access = VK_ACCESS_TRANSFER_WRITE_BIT; - - Buffer *buffer = _get_buffer_from_owner(p_buffer, dst_stage_mask, dst_access); + VkPipelineStageFlags dst_stage_mask = 0; + VkAccessFlags dst_access = 0; + if (p_post_barrier & BARRIER_MASK_TRANSFER) { + // Protect subsequent updates... + dst_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT; + dst_access = VK_ACCESS_TRANSFER_WRITE_BIT; + } + Buffer *buffer = _get_buffer_from_owner(p_buffer, dst_stage_mask, dst_access, p_post_barrier); if (!buffer) { ERR_FAIL_V_MSG(ERR_INVALID_PARAMETER, "Buffer argument is not a valid buffer of any type."); } @@ -5058,35 +5205,41 @@ Error RenderingDeviceVulkan::buffer_update(RID p_buffer, uint32_t p_offset, uint ERR_FAIL_COND_V_MSG(p_offset + p_size > buffer->size, ERR_INVALID_PARAMETER, "Attempted to write buffer (" + itos((p_offset + p_size) - buffer->size) + " bytes) past the end."); - _buffer_memory_barrier(buffer->buffer, p_offset, p_size, dst_stage_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_access, VK_ACCESS_TRANSFER_WRITE_BIT, p_sync_with_draw); - Error err = _buffer_update(buffer, p_offset, (uint8_t *)p_data, p_size, p_sync_with_draw); + // no barrier should be needed here + // _buffer_memory_barrier(buffer->buffer, p_offset, p_size, dst_stage_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_access, VK_ACCESS_TRANSFER_WRITE_BIT, true); + + Error err = _buffer_update(buffer, p_offset, (uint8_t *)p_data, p_size, p_post_barrier); if (err) { return err; } #ifdef FORCE_FULL_BARRIER - _full_barrier(p_sync_with_draw); + _full_barrier(true); #else - _buffer_memory_barrier(buffer->buffer, p_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, p_sync_with_draw); + _buffer_memory_barrier(buffer->buffer, p_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, true); #endif return err; } -Error RenderingDeviceVulkan::buffer_clear(RID p_buffer, uint32_t p_offset, uint32_t p_size, bool p_sync_with_draw) { +Error RenderingDeviceVulkan::buffer_clear(RID p_buffer, uint32_t p_offset, uint32_t p_size, uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_V_MSG((p_size % 4) != 0, ERR_INVALID_PARAMETER, "Size must be a multiple of four"); - ERR_FAIL_COND_V_MSG(draw_list && p_sync_with_draw, ERR_INVALID_PARAMETER, - "Updating buffers in 'sync to draw' mode is forbidden during creation of a draw list"); - ERR_FAIL_COND_V_MSG(compute_list && p_sync_with_draw, ERR_INVALID_PARAMETER, - "Updating buffers in 'sync to draw' mode is forbidden during creation of a compute list"); + ERR_FAIL_COND_V_MSG(draw_list, ERR_INVALID_PARAMETER, + "Updating buffers in is forbidden during creation of a draw list"); + ERR_FAIL_COND_V_MSG(compute_list, ERR_INVALID_PARAMETER, + "Updating buffers is forbidden during creation of a compute list"); - // Protect subsequent updates... - VkPipelineStageFlags dst_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT; - VkAccessFlags dst_access = VK_ACCESS_TRANSFER_WRITE_BIT; + VkPipelineStageFlags dst_stage_mask = 0; + VkAccessFlags dst_access = 0; + if (p_post_barrier & BARRIER_MASK_TRANSFER) { + // Protect subsequent updates... + dst_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT; + dst_access = VK_ACCESS_TRANSFER_WRITE_BIT; + } - Buffer *buffer = _get_buffer_from_owner(p_buffer, dst_stage_mask, dst_access); + Buffer *buffer = _get_buffer_from_owner(p_buffer, dst_stage_mask, dst_access, p_post_barrier); if (!buffer) { ERR_FAIL_V_MSG(ERR_INVALID_PARAMETER, "Buffer argument is not a valid buffer of any type."); } @@ -5094,14 +5247,15 @@ Error RenderingDeviceVulkan::buffer_clear(RID p_buffer, uint32_t p_offset, uint3 ERR_FAIL_COND_V_MSG(p_offset + p_size > buffer->size, ERR_INVALID_PARAMETER, "Attempted to write buffer (" + itos((p_offset + p_size) - buffer->size) + " bytes) past the end."); - _buffer_memory_barrier(buffer->buffer, p_offset, p_size, dst_stage_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_access, VK_ACCESS_TRANSFER_WRITE_BIT, p_sync_with_draw); + // should not be needed + // _buffer_memory_barrier(buffer->buffer, p_offset, p_size, dst_stage_mask, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_access, VK_ACCESS_TRANSFER_WRITE_BIT, p_post_barrier); - vkCmdFillBuffer(p_sync_with_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer, buffer->buffer, p_offset, p_size, 0); + vkCmdFillBuffer(frames[frame].draw_command_buffer, buffer->buffer, p_offset, p_size, 0); #ifdef FORCE_FULL_BARRIER - _full_barrier(p_sync_with_draw); + _full_barrier(true); #else - _buffer_memory_barrier(buffer->buffer, p_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, p_sync_with_draw); + _buffer_memory_barrier(buffer->buffer, p_offset, p_size, VK_PIPELINE_STAGE_TRANSFER_BIT, dst_stage_mask, VK_ACCESS_TRANSFER_WRITE_BIT, dst_access, p_post_barrier); #endif return OK; } @@ -5113,7 +5267,7 @@ Vector RenderingDeviceVulkan::buffer_get_data(RID p_buffer) { VkPipelineShaderStageCreateFlags src_stage_mask = VK_PIPELINE_STAGE_TRANSFER_BIT; VkAccessFlags src_access_mask = VK_ACCESS_TRANSFER_WRITE_BIT; // Get the vulkan buffer and the potential stage/access possible - Buffer *buffer = _get_buffer_from_owner(p_buffer, src_stage_mask, src_access_mask); + Buffer *buffer = _get_buffer_from_owner(p_buffer, src_stage_mask, src_access_mask, BARRIER_MASK_ALL); if (!buffer) { ERR_FAIL_V_MSG(Vector(), "Buffer is either invalid or this type of buffer can't be retrieved. Only Index and Vertex buffers allow retrieving."); } @@ -6476,7 +6630,7 @@ void RenderingDeviceVulkan::draw_list_disable_scissor(DrawListID p_list) { vkCmdSetScissor(dl->command_buffer, 0, 1, &scissor); } -void RenderingDeviceVulkan::draw_list_end() { +void RenderingDeviceVulkan::draw_list_end(uint32_t p_post_barrier) { _THREAD_SAFE_METHOD_ ERR_FAIL_COND_MSG(!draw_list, "Immediate draw list is already inactive."); @@ -6512,6 +6666,21 @@ void RenderingDeviceVulkan::draw_list_end() { } } + uint32_t barrier_flags = 0; + uint32_t access_flags = 0; + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; + } + if (p_post_barrier & BARRIER_MASK_RASTER) { + barrier_flags |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; + } + if (p_post_barrier & BARRIER_MASK_TRANSFER) { + barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; + access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT; + } + draw_list_bound_textures.clear(); for (int i = 0; i < draw_list_storage_textures.size(); i++) { @@ -6521,7 +6690,7 @@ void RenderingDeviceVulkan::draw_list_end() { image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; - image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT; + image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = texture->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; @@ -6534,7 +6703,7 @@ void RenderingDeviceVulkan::draw_list_end() { image_memory_barrier.subresourceRange.baseArrayLayer = texture->base_layer; image_memory_barrier.subresourceRange.layerCount = texture->layers; - vkCmdPipelineBarrier(frames[frame].draw_command_buffer, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); + vkCmdPipelineBarrier(frames[frame].draw_command_buffer, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); texture->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; } @@ -6548,7 +6717,7 @@ void RenderingDeviceVulkan::draw_list_end() { #ifdef FORCE_FULL_BARRIER _full_barrier(true); #else - _memory_barrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT, VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT, VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, true); + _memory_barrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT, barrier_flags, VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT, access_flags, true); #endif } @@ -6871,14 +7040,30 @@ void RenderingDeviceVulkan::compute_list_add_barrier(ComputeListID p_list) { #endif } -void RenderingDeviceVulkan::compute_list_end() { +void RenderingDeviceVulkan::compute_list_end(uint32_t p_post_barrier) { ERR_FAIL_COND(!compute_list); + + uint32_t barrier_flags = 0; + uint32_t access_flags = 0; + if (p_post_barrier & BARRIER_MASK_COMPUTE) { + barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; + } + if (p_post_barrier & BARRIER_MASK_RASTER) { + barrier_flags |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT; + access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; + } + if (p_post_barrier & BARRIER_MASK_TRANSFER) { + barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; + access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT; + } + for (Set::Element *E = compute_list->state.textures_to_sampled_layout.front(); E; E = E->next()) { VkImageMemoryBarrier image_memory_barrier; image_memory_barrier.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER; image_memory_barrier.pNext = nullptr; image_memory_barrier.srcAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT; - image_memory_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT; + image_memory_barrier.dstAccessMask = access_flags; image_memory_barrier.oldLayout = E->get()->layout; image_memory_barrier.newLayout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; @@ -6892,7 +7077,7 @@ void RenderingDeviceVulkan::compute_list_end() { image_memory_barrier.subresourceRange.layerCount = E->get()->layers; // TODO: Look at the usages in the compute list and determine tighter dst stage and access masks based on some "final" usage equivalent - vkCmdPipelineBarrier(compute_list->command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); + vkCmdPipelineBarrier(compute_list->command_buffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, barrier_flags, 0, 0, nullptr, 0, nullptr, 1, &image_memory_barrier); E->get()->layout = VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL; } @@ -6902,10 +7087,44 @@ void RenderingDeviceVulkan::compute_list_end() { #ifdef FORCE_FULL_BARRIER _full_barrier(true); #else - _memory_barrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT | VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_TRANSFER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT, VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT, true); + _memory_barrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, barrier_flags, VK_ACCESS_SHADER_WRITE_BIT, VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT, true); #endif } +void RenderingDeviceVulkan::barrier(uint32_t p_from, uint32_t p_to) { + uint32_t src_barrier_flags = 0; + uint32_t src_access_flags = 0; + if (p_from & BARRIER_MASK_COMPUTE) { + src_barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + src_access_flags |= VK_ACCESS_SHADER_WRITE_BIT; + } + if (p_from & BARRIER_MASK_RASTER) { + src_barrier_flags |= VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT | VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT; + src_access_flags |= VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT | VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT; + } + if (p_from & BARRIER_MASK_TRANSFER) { + src_barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; + src_access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT; + } + + uint32_t dst_barrier_flags = 0; + uint32_t dst_access_flags = 0; + if (p_to & BARRIER_MASK_COMPUTE) { + dst_barrier_flags |= VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT; + dst_access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; + } + if (p_to & BARRIER_MASK_RASTER) { + dst_barrier_flags |= VK_PIPELINE_STAGE_VERTEX_INPUT_BIT | VK_PIPELINE_STAGE_VERTEX_SHADER_BIT | VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT | VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT; + dst_access_flags |= VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT | VK_ACCESS_INDEX_READ_BIT | VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT | VK_ACCESS_INDIRECT_COMMAND_READ_BIT; + } + if (p_to & BARRIER_MASK_TRANSFER) { + dst_barrier_flags |= VK_PIPELINE_STAGE_TRANSFER_BIT; + dst_access_flags |= VK_ACCESS_TRANSFER_WRITE_BIT | VK_ACCESS_TRANSFER_READ_BIT; + } + + _memory_barrier(src_barrier_flags, dst_barrier_flags, src_access_flags, dst_access_flags, true); +} + void RenderingDeviceVulkan::full_barrier() { #ifndef DEBUG_ENABLED ERR_PRINT("Full barrier is debug-only, should not be used in production"); @@ -7524,9 +7743,10 @@ void RenderingDeviceVulkan::_free_rids(T &p_owner, const char *p_type) { } } -void RenderingDeviceVulkan::capture_timestamp(const String &p_name, bool p_sync_to_draw) { +void RenderingDeviceVulkan::capture_timestamp(const String &p_name) { ERR_FAIL_COND(frames[frame].timestamp_count >= max_timestamp_query_elements); + //this should be optional for profiling, else it will slow things down { VkMemoryBarrier memoryBarrier; @@ -7563,9 +7783,10 @@ void RenderingDeviceVulkan::capture_timestamp(const String &p_name, bool p_sync_ VK_ACCESS_HOST_READ_BIT | VK_ACCESS_HOST_WRITE_BIT; - vkCmdPipelineBarrier(p_sync_to_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1, &memoryBarrier, 0, nullptr, 0, nullptr); + vkCmdPipelineBarrier(frames[frame].draw_command_buffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0, 1, &memoryBarrier, 0, nullptr, 0, nullptr); } - vkCmdWriteTimestamp(p_sync_to_draw ? frames[frame].draw_command_buffer : frames[frame].setup_command_buffer, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, frames[frame].timestamp_pool, frames[frame].timestamp_count); + + vkCmdWriteTimestamp(frames[frame].draw_command_buffer, VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT, frames[frame].timestamp_pool, frames[frame].timestamp_count); frames[frame].timestamp_names[frames[frame].timestamp_count] = p_name; frames[frame].timestamp_cpu_values[frames[frame].timestamp_count] = OS::get_singleton()->get_ticks_usec(); frames[frame].timestamp_count++; diff --git a/drivers/vulkan/rendering_device_vulkan.h b/drivers/vulkan/rendering_device_vulkan.h index ba3e1b243c..4bea17e4a1 100644 --- a/drivers/vulkan/rendering_device_vulkan.h +++ b/drivers/vulkan/rendering_device_vulkan.h @@ -785,7 +785,7 @@ class RenderingDeviceVulkan : public RenderingDevice { Error _draw_list_setup_framebuffer(Framebuffer *p_framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, VkFramebuffer *r_framebuffer, VkRenderPass *r_render_pass); Error _draw_list_render_pass_begin(Framebuffer *framebuffer, InitialAction p_initial_color_action, FinalAction p_final_color_action, InitialAction p_initial_depth_action, FinalAction p_final_depth_action, const Vector &p_clear_colors, float p_clear_depth, uint32_t p_clear_stencil, Point2i viewport_offset, Point2i viewport_size, VkFramebuffer vkframebuffer, VkRenderPass render_pass, VkCommandBuffer command_buffer, VkSubpassContents subpass_contents, const Vector &p_storage_textures); _FORCE_INLINE_ DrawList *_get_draw_list_ptr(DrawListID p_id); - Buffer *_get_buffer_from_owner(RID p_buffer, VkPipelineStageFlags &dst_stage_mask, VkAccessFlags &dst_access); + Buffer *_get_buffer_from_owner(RID p_buffer, VkPipelineStageFlags &dst_stage_mask, VkAccessFlags &dst_access, uint32_t p_post_barrier); /**********************/ /**** COMPUTE LIST ****/ @@ -913,16 +913,16 @@ public: virtual RID texture_create_shared(const TextureView &p_view, RID p_with_texture); virtual RID texture_create_shared_from_slice(const TextureView &p_view, RID p_with_texture, uint32_t p_layer, uint32_t p_mipmap, TextureSliceType p_slice_type = TEXTURE_SLICE_2D); - virtual Error texture_update(RID p_texture, uint32_t p_layer, const Vector &p_data, bool p_sync_with_draw = false); + virtual Error texture_update(RID p_texture, uint32_t p_layer, const Vector &p_data, uint32_t p_post_barrier = BARRIER_MASK_ALL); virtual Vector texture_get_data(RID p_texture, uint32_t p_layer); virtual bool texture_is_format_supported_for_usage(DataFormat p_format, uint32_t p_usage) const; virtual bool texture_is_shared(RID p_texture); virtual bool texture_is_valid(RID p_texture); - virtual Error texture_copy(RID p_from_texture, RID p_to_texture, const Vector3 &p_from, const Vector3 &p_to, const Vector3 &p_size, uint32_t p_src_mipmap, uint32_t p_dst_mipmap, uint32_t p_src_layer, uint32_t p_dst_layer, bool p_sync_with_draw = false); - virtual Error texture_clear(RID p_texture, const Color &p_color, uint32_t p_base_mipmap, uint32_t p_mipmaps, uint32_t p_base_layer, uint32_t p_layers, bool p_sync_with_draw = false); - virtual Error texture_resolve_multisample(RID p_from_texture, RID p_to_texture, bool p_sync_with_draw = false); + virtual Error texture_copy(RID p_from_texture, RID p_to_texture, const Vector3 &p_from, const Vector3 &p_to, const Vector3 &p_size, uint32_t p_src_mipmap, uint32_t p_dst_mipmap, uint32_t p_src_layer, uint32_t p_dst_layer, uint32_t p_post_barrier = BARRIER_MASK_ALL); + virtual Error texture_clear(RID p_texture, const Color &p_color, uint32_t p_base_mipmap, uint32_t p_mipmaps, uint32_t p_base_layer, uint32_t p_layers, uint32_t p_post_barrier = BARRIER_MASK_ALL); + virtual Error texture_resolve_multisample(RID p_from_texture, RID p_to_texture, uint32_t p_post_barrier = BARRIER_MASK_ALL); /*********************/ /**** FRAMEBUFFER ****/ @@ -975,8 +975,8 @@ public: virtual RID uniform_set_create(const Vector &p_uniforms, RID p_shader, uint32_t p_shader_set); virtual bool uniform_set_is_valid(RID p_uniform_set); - virtual Error buffer_update(RID p_buffer, uint32_t p_offset, uint32_t p_size, const void *p_data, bool p_sync_with_draw = false); //works for any buffer - virtual Error buffer_clear(RID p_buffer, uint32_t p_offset, uint32_t p_size, bool p_sync_with_draw = false); + virtual Error buffer_update(RID p_buffer, uint32_t p_offset, uint32_t p_size, const void *p_data, uint32_t p_post_barrier = BARRIER_MASK_ALL); //works for any buffer + virtual Error buffer_clear(RID p_buffer, uint32_t p_offset, uint32_t p_size, uint32_t p_post_barrier = BARRIER_MASK_ALL); virtual Vector buffer_get_data(RID p_buffer); /*************************/ @@ -1022,7 +1022,7 @@ public: virtual void draw_list_enable_scissor(DrawListID p_list, const Rect2 &p_rect); virtual void draw_list_disable_scissor(DrawListID p_list); - virtual void draw_list_end(); + virtual void draw_list_end(uint32_t p_post_barrier = BARRIER_MASK_ALL); /***********************/ /**** COMPUTE LISTS ****/ @@ -1036,8 +1036,9 @@ public: virtual void compute_list_dispatch(ComputeListID p_list, uint32_t p_x_groups, uint32_t p_y_groups, uint32_t p_z_groups); virtual void compute_list_dispatch_indirect(ComputeListID p_list, RID p_buffer, uint32_t p_offset); - virtual void compute_list_end(); + virtual void compute_list_end(uint32_t p_post_barrier = BARRIER_MASK_ALL); + virtual void barrier(uint32_t p_from = BARRIER_MASK_ALL, uint32_t p_to = BARRIER_MASK_ALL); virtual void full_barrier(); /**************/ @@ -1050,7 +1051,7 @@ public: /**** Timing ****/ /****************/ - virtual void capture_timestamp(const String &p_name, bool p_sync_to_draw); + virtual void capture_timestamp(const String &p_name); virtual uint32_t get_captured_timestamps_count() const; virtual uint64_t get_captured_timestamps_frame() const; virtual uint64_t get_captured_timestamp_gpu_time(uint32_t p_index) const; diff --git a/drivers/vulkan/vulkan_context.cpp b/drivers/vulkan/vulkan_context.cpp index c644395b54..7ac860087e 100644 --- a/drivers/vulkan/vulkan_context.cpp +++ b/drivers/vulkan/vulkan_context.cpp @@ -54,11 +54,30 @@ VKAPI_ATTR VkBool32 VKAPI_CALL VulkanContext::_debug_messenger_callback( strstr(pCallbackData->pMessage, "can result in undefined behavior if this memory is used by the device") != nullptr) { return VK_FALSE; } + // This needs to be ignored because Validator is wrong here. + if (strstr(pCallbackData->pMessage, "Invalid SPIR-V binary version 1.3") != nullptr) { + return VK_FALSE; + } + // This needs to be ignored because Validator is wrong here. + if (strstr(pCallbackData->pMessage, "Shader requires flag") != nullptr) { + return VK_FALSE; + } + // This needs to be ignored because Validator is wrong here. if (strstr(pCallbackData->pMessage, "SPIR-V module not valid: Pointer operand") != nullptr && strstr(pCallbackData->pMessage, "must be a memory object") != nullptr) { return VK_FALSE; } + /* + // This is a valid warning because its illegal in Vulkan, but in practice it should work according to VK_KHR_maintenance2 + if (strstr(pCallbackData->pMessage, "VK_FORMAT_E5B9G9R9_UFLOAT_PACK32 with tiling VK_IMAGE_TILING_OPTIMAL does not support usage that includes VK_IMAGE_USAGE_STORAGE_BIT") != nullptr) { + return VK_FALSE; + } + + if (strstr(pCallbackData->pMessage, "VK_FORMAT_R4G4B4A4_UNORM_PACK16 with tiling VK_IMAGE_TILING_OPTIMAL does not support usage that includes VK_IMAGE_USAGE_STORAGE_BIT") != nullptr) { + return VK_FALSE; + } +*/ // Workaround for Vulkan-Loader usability bug: https://github.com/KhronosGroup/Vulkan-Loader/issues/262. if (strstr(pCallbackData->pMessage, "wrong ELF class: ELFCLASS32") != nullptr) { return VK_FALSE; -- cgit v1.2.3