mirrors/dolphin
1
0
Fork 0
mirror of https://github.com/dolphin-emu/dolphin.git synced 2025-02-09 00:12:14 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

Vulkan: Implement StagingTexture2D on top of StagingBuffer

Browse Source 
Greatly simplifies things, and we weren't using the linear texture
implementation anyway.
This commit is contained in:
Stenzek 2016-11-19 22:22:04 +10:00
parent 4bc0e14995
commit e241ec6666
5 changed files with 91 additions and 510 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
Source/Core/VideoBackends/Vulkan/FramebufferManager.cpp
View File

@ -1007,10 +1007,8 @@ bool FramebufferManager::CreateReadbackTextures()
VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT |
VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT); VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT);
// We can't copy to/from color<->depth formats, so using a linear texture is not an option here. m_depth_readback_texture = StagingTexture2D::Create(STAGING_BUFFER_TYPE_READBACK, EFB_WIDTH,
// TODO: Investigate if vkCmdBlitImage can be used. The documentation isn't that clear. EFB_HEIGHT, EFB_DEPTH_TEXTURE_FORMAT);
m_depth_readback_texture = StagingTexture2DBuffer::Create(STAGING_BUFFER_TYPE_READBACK, EFB_WIDTH,
EFB_HEIGHT, EFB_DEPTH_TEXTURE_FORMAT);
if (!m_depth_copy_texture || !m_depth_readback_texture) if (!m_depth_copy_texture || !m_depth_readback_texture)
{ {
ERROR_LOG(VIDEO, "Failed to create EFB depth readback texture"); ERROR_LOG(VIDEO, "Failed to create EFB depth readback texture");

55
Source/Core/VideoBackends/Vulkan/StagingBuffer.cpp
View File

@ -80,6 +80,9 @@ void StagingBuffer::InvalidateGPUCache(VkCommandBuffer command_buffer,
VkPipelineStageFlagBits dest_pipeline_stage, VkPipelineStageFlagBits dest_pipeline_stage,
VkDeviceSize offset, VkDeviceSize size) VkDeviceSize offset, VkDeviceSize size)
{ {
if (m_coherent)
return;
_assert_((offset + size) <= m_size || (offset < m_size && size == VK_WHOLE_SIZE)); _assert_((offset + size) <= m_size || (offset < m_size && size == VK_WHOLE_SIZE));
Util::BufferMemoryBarrier(command_buffer, m_buffer, VK_ACCESS_HOST_WRITE_BIT, dest_access_flags, Util::BufferMemoryBarrier(command_buffer, m_buffer, VK_ACCESS_HOST_WRITE_BIT, dest_access_flags,
offset, size, VK_PIPELINE_STAGE_HOST_BIT, dest_pipeline_stage); offset, size, VK_PIPELINE_STAGE_HOST_BIT, dest_pipeline_stage);
@ -90,6 +93,9 @@ void StagingBuffer::PrepareForGPUWrite(VkCommandBuffer command_buffer,
VkPipelineStageFlagBits dst_pipeline_stage, VkPipelineStageFlagBits dst_pipeline_stage,
VkDeviceSize offset, VkDeviceSize size) VkDeviceSize offset, VkDeviceSize size)
{ {
if (m_coherent)
return;
_assert_((offset + size) <= m_size || (offset < m_size && size == VK_WHOLE_SIZE)); _assert_((offset + size) <= m_size || (offset < m_size && size == VK_WHOLE_SIZE));
Util::BufferMemoryBarrier(command_buffer, m_buffer, 0, dst_access_flags, offset, size, Util::BufferMemoryBarrier(command_buffer, m_buffer, 0, dst_access_flags, offset, size,
VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, dst_pipeline_stage); VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT, dst_pipeline_stage);
@ -99,6 +105,9 @@ void StagingBuffer::FlushGPUCache(VkCommandBuffer command_buffer, VkAccessFlagBi
VkPipelineStageFlagBits src_pipeline_stage, VkDeviceSize offset, VkPipelineStageFlagBits src_pipeline_stage, VkDeviceSize offset,
VkDeviceSize size) VkDeviceSize size)
{ {
if (m_coherent)
return;
_assert_((offset + size) <= m_size || (offset < m_size && size == VK_WHOLE_SIZE)); _assert_((offset + size) <= m_size || (offset < m_size && size == VK_WHOLE_SIZE));
Util::BufferMemoryBarrier(command_buffer, m_buffer, src_access_flags, VK_ACCESS_HOST_READ_BIT, Util::BufferMemoryBarrier(command_buffer, m_buffer, src_access_flags, VK_ACCESS_HOST_READ_BIT,
offset, size, src_pipeline_stage, VK_PIPELINE_STAGE_HOST_BIT); offset, size, src_pipeline_stage, VK_PIPELINE_STAGE_HOST_BIT);
@ -136,8 +145,9 @@ void StagingBuffer::Write(VkDeviceSize offset, const void* data, size_t size,
FlushCPUCache(offset, size); FlushCPUCache(offset, size);
} }
std::unique_ptr<Vulkan::StagingBuffer> bool StagingBuffer::AllocateBuffer(STAGING_BUFFER_TYPE type, VkDeviceSize size,
StagingBuffer::Create(STAGING_BUFFER_TYPE type, VkDeviceSize size, VkBufferUsageFlags usage) VkBufferUsageFlags usage, VkBuffer* out_buffer,
VkDeviceMemory* out_memory, bool* out_coherent)
{ {
VkBufferCreateInfo buffer_create_info = { VkBufferCreateInfo buffer_create_info = {
VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType
@ -149,24 +159,22 @@ StagingBuffer::Create(STAGING_BUFFER_TYPE type, VkDeviceSize size, VkBufferUsage
0, // uint32_t queueFamilyIndexCount 0, // uint32_t queueFamilyIndexCount
nullptr // const uint32_t* pQueueFamilyIndices nullptr // const uint32_t* pQueueFamilyIndices
}; };
VkBuffer buffer;
VkResult res = VkResult res =
vkCreateBuffer(g_vulkan_context->GetDevice(), &buffer_create_info, nullptr, &buffer); vkCreateBuffer(g_vulkan_context->GetDevice(), &buffer_create_info, nullptr, out_buffer);
if (res != VK_SUCCESS) if (res != VK_SUCCESS)
{ {
LOG_VULKAN_ERROR(res, "vkCreateBuffer failed: "); LOG_VULKAN_ERROR(res, "vkCreateBuffer failed: ");
return nullptr; return false;
} }
VkMemoryRequirements requirements; VkMemoryRequirements requirements;
vkGetBufferMemoryRequirements(g_vulkan_context->GetDevice(), buffer, &requirements); vkGetBufferMemoryRequirements(g_vulkan_context->GetDevice(), *out_buffer, &requirements);
bool is_coherent;
u32 type_index; u32 type_index;
if (type == STAGING_BUFFER_TYPE_UPLOAD) if (type == STAGING_BUFFER_TYPE_UPLOAD)
type_index = g_vulkan_context->GetUploadMemoryType(requirements.memoryTypeBits, &is_coherent); type_index = g_vulkan_context->GetUploadMemoryType(requirements.memoryTypeBits, out_coherent);
else else
type_index = g_vulkan_context->GetReadbackMemoryType(requirements.memoryTypeBits, &is_coherent); type_index = g_vulkan_context->GetReadbackMemoryType(requirements.memoryTypeBits, out_coherent);
VkMemoryAllocateInfo memory_allocate_info = { VkMemoryAllocateInfo memory_allocate_info = {
VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, // VkStructureType sType VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, // VkStructureType sType
@ -174,25 +182,36 @@ StagingBuffer::Create(STAGING_BUFFER_TYPE type, VkDeviceSize size, VkBufferUsage
requirements.size, // VkDeviceSize allocationSize requirements.size, // VkDeviceSize allocationSize
type_index // uint32_t memoryTypeIndex type_index // uint32_t memoryTypeIndex
}; };
VkDeviceMemory memory; res = vkAllocateMemory(g_vulkan_context->GetDevice(), &memory_allocate_info, nullptr, out_memory);
res = vkAllocateMemory(g_vulkan_context->GetDevice(), &memory_allocate_info, nullptr, &memory);
if (res != VK_SUCCESS) if (res != VK_SUCCESS)
{ {
LOG_VULKAN_ERROR(res, "vkAllocateMemory failed: "); LOG_VULKAN_ERROR(res, "vkAllocateMemory failed: ");
vkDestroyBuffer(g_vulkan_context->GetDevice(), buffer, nullptr); vkDestroyBuffer(g_vulkan_context->GetDevice(), *out_buffer, nullptr);
return nullptr; return false;
} }
res = vkBindBufferMemory(g_vulkan_context->GetDevice(), buffer, memory, 0); res = vkBindBufferMemory(g_vulkan_context->GetDevice(), *out_buffer, *out_memory, 0);
if (res != VK_SUCCESS) if (res != VK_SUCCESS)
{ {
LOG_VULKAN_ERROR(res, "vkBindBufferMemory failed: "); LOG_VULKAN_ERROR(res, "vkBindBufferMemory failed: ");
vkDestroyBuffer(g_vulkan_context->GetDevice(), buffer, nullptr); vkDestroyBuffer(g_vulkan_context->GetDevice(), *out_buffer, nullptr);
vkFreeMemory(g_vulkan_context->GetDevice(), memory, nullptr); vkFreeMemory(g_vulkan_context->GetDevice(), *out_memory, nullptr);
return nullptr; return false;
} }
return std::make_unique<Vulkan::StagingBuffer>(type, buffer, memory, size, is_coherent); return true;
}
std::unique_ptr<StagingBuffer> StagingBuffer::Create(STAGING_BUFFER_TYPE type, VkDeviceSize size,
VkBufferUsageFlags usage)
{
VkBuffer buffer;
VkDeviceMemory memory;
bool coherent;
if (!AllocateBuffer(type, size, usage, &buffer, &memory, &coherent))
return nullptr;
return std::make_unique<StagingBuffer>(type, buffer, memory, size, coherent);
} }
} // namespace Vulkan } // namespace Vulkan

7
Source/Core/VideoBackends/Vulkan/StagingBuffer.h
View File

@ -16,7 +16,7 @@ class StagingBuffer
public: public:
StagingBuffer(STAGING_BUFFER_TYPE type, VkBuffer buffer, VkDeviceMemory memory, VkDeviceSize size, StagingBuffer(STAGING_BUFFER_TYPE type, VkBuffer buffer, VkDeviceMemory memory, VkDeviceSize size,
bool coherent); bool coherent);
~StagingBuffer(); virtual ~StagingBuffer();
STAGING_BUFFER_TYPE GetType() const { return m_type; } STAGING_BUFFER_TYPE GetType() const { return m_type; }
VkDeviceSize GetSize() const { return m_size; } VkDeviceSize GetSize() const { return m_size; }
@ -33,6 +33,7 @@ public:
void FlushCPUCache(VkDeviceSize offset = 0, VkDeviceSize size = VK_WHOLE_SIZE); void FlushCPUCache(VkDeviceSize offset = 0, VkDeviceSize size = VK_WHOLE_SIZE);
// Upload part 2: Prepare for device read from the GPU side // Upload part 2: Prepare for device read from the GPU side
// Implicit when submitting the command buffer, so rarely needed.
void InvalidateGPUCache(VkCommandBuffer command_buffer, VkAccessFlagBits dst_access_flags, void InvalidateGPUCache(VkCommandBuffer command_buffer, VkAccessFlagBits dst_access_flags,
VkPipelineStageFlagBits dst_pipeline_stage, VkDeviceSize offset = 0, VkPipelineStageFlagBits dst_pipeline_stage, VkDeviceSize offset = 0,
VkDeviceSize size = VK_WHOLE_SIZE); VkDeviceSize size = VK_WHOLE_SIZE);
@ -59,6 +60,10 @@ public:
VkBufferUsageFlags usage); VkBufferUsageFlags usage);
protected: protected:
// Allocates the resources needed to create a staging buffer.
static bool AllocateBuffer(STAGING_BUFFER_TYPE type, VkDeviceSize size, VkBufferUsageFlags usage,
VkBuffer* out_buffer, VkDeviceMemory* out_memory, bool* out_coherent);
STAGING_BUFFER_TYPE m_type; STAGING_BUFFER_TYPE m_type;
VkBuffer m_buffer; VkBuffer m_buffer;
VkDeviceMemory m_memory; VkDeviceMemory m_memory;

442
Source/Core/VideoBackends/Vulkan/StagingTexture2D.cpp
View File

@ -14,16 +14,16 @@
namespace Vulkan namespace Vulkan
{ {
StagingTexture2D::StagingTexture2D(STAGING_BUFFER_TYPE type, u32 width, u32 height, VkFormat format, StagingTexture2D::StagingTexture2D(STAGING_BUFFER_TYPE type, VkBuffer buffer, VkDeviceMemory memory,
u32 stride) VkDeviceSize size, bool coherent, u32 width, u32 height,
: m_type(type), m_width(width), m_height(height), m_format(format), VkFormat format, u32 stride)
m_texel_size(Util::GetTexelSize(format)), m_row_stride(stride) : StagingBuffer(type, buffer, memory, size, coherent), m_width(width), m_height(height),
m_format(format), m_texel_size(Util::GetTexelSize(format)), m_row_stride(stride)
{ {
} }
StagingTexture2D::~StagingTexture2D() StagingTexture2D::~StagingTexture2D()
{ {
_assert_(!m_map_pointer);
} }
void StagingTexture2D::ReadTexel(u32 x, u32 y, void* data, size_t data_size) const void StagingTexture2D::ReadTexel(u32 x, u32 y, void* data, size_t data_size) const
@ -96,283 +96,13 @@ void StagingTexture2D::WriteTexels(u32 x, u32 y, u32 width, u32 height, const vo
} }
} }
std::unique_ptr<StagingTexture2D> StagingTexture2D::Create(STAGING_BUFFER_TYPE type, u32 width, void StagingTexture2D::CopyFromImage(VkCommandBuffer command_buffer, VkImage image,
u32 height, VkFormat format) VkImageAspectFlags src_aspect, u32 x, u32 y, u32 width,
{ u32 height, u32 level, u32 layer)
// TODO: Using a buffer here as opposed to a linear texture is faster on AMD.
// NVIDIA also seems faster with buffers over textures.
#if 0
// Check for support for this format as a linear texture.
// Some drivers don't support this (e.g. adreno).
VkImageFormatProperties properties;
VkResult res = vkGetPhysicalDeviceImageFormatProperties(
g_object_cache->GetPhysicalDevice(), format, VK_IMAGE_TYPE_2D, VK_IMAGE_TILING_LINEAR,
VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 0, &properties);
if (res == VK_SUCCESS && width <= properties.maxExtent.width &&
height <= properties.maxExtent.height)
{
return StagingTexture2DLinear::Create(type, width, height, format);
}
#endif
// Fall back to a buffer copy.
return StagingTexture2DBuffer::Create(type, width, height, format);
}
StagingTexture2DLinear::StagingTexture2DLinear(STAGING_BUFFER_TYPE type, u32 width, u32 height,
VkFormat format, u32 stride, VkImage image,
VkDeviceMemory memory, VkDeviceSize size,
bool coherent)
: StagingTexture2D(type, width, height, format, stride), m_image(image), m_memory(memory),
m_size(size), m_layout(VK_IMAGE_LAYOUT_PREINITIALIZED), m_coherent(coherent)
{
}
StagingTexture2DLinear::~StagingTexture2DLinear()
{
if (m_map_pointer)
Unmap();
g_command_buffer_mgr->DeferDeviceMemoryDestruction(m_memory);
g_command_buffer_mgr->DeferImageDestruction(m_image);
}
void StagingTexture2DLinear::CopyFromImage(VkCommandBuffer command_buffer, VkImage image,
VkImageAspectFlags src_aspect, u32 x, u32 y, u32 width,
u32 height, u32 level, u32 layer)
{
// Prepare the buffer for copying.
// We don't care about the existing contents, so set to UNDEFINED.
VkImageMemoryBarrier before_transfer_barrier = {
VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType
nullptr, // const void* pNext
0, // VkAccessFlags srcAccessMask
VK_ACCESS_TRANSFER_WRITE_BIT, // VkAccessFlags dstAccessMask
VK_IMAGE_LAYOUT_UNDEFINED, // VkImageLayout oldLayout
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, // VkImageLayout newLayout
VK_QUEUE_FAMILY_IGNORED, // uint32_t srcQueueFamilyIndex
VK_QUEUE_FAMILY_IGNORED, // uint32_t dstQueueFamilyIndex
m_image, // VkImage image
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1} // VkImageSubresourceRange subresourceRange
};
vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT, 0, 0, nullptr, 0, nullptr, 1,
&before_transfer_barrier);
// Issue the image copy, gpu -> host.
VkImageCopy copy_region = {
{src_aspect, level, layer, 1}, // VkImageSubresourceLayers srcSubresource
{static_cast<s32>(x), static_cast<s32>(y), 0}, // VkOffset3D srcOffset
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1}, // VkImageSubresourceLayers dstSubresource
{0, 0, 0}, // VkOffset3D dstOffset
{width, height, 1} // VkExtent3D extent
};
vkCmdCopyImage(command_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, m_image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &copy_region);
// Ensure writes are visible to the host.
VkImageMemoryBarrier visible_barrier = {
VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType
nullptr, // const void* pNext
VK_ACCESS_TRANSFER_WRITE_BIT, // VkAccessFlags srcAccessMask
VK_ACCESS_HOST_READ_BIT, // VkAccessFlags dstAccessMask
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, // VkImageLayout oldLayout
VK_IMAGE_LAYOUT_GENERAL, // VkImageLayout newLayout
VK_QUEUE_FAMILY_IGNORED, // uint32_t srcQueueFamilyIndex
VK_QUEUE_FAMILY_IGNORED, // uint32_t dstQueueFamilyIndex
m_image, // VkImage image
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1} // VkImageSubresourceRange subresourceRange
};
vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_TRANSFER_BIT, VK_PIPELINE_STAGE_HOST_BIT,
0, 0, nullptr, 0, nullptr, 1, &visible_barrier);
m_layout = VK_IMAGE_LAYOUT_GENERAL;
// Invalidate memory range if currently mapped.
if (m_map_pointer && !m_coherent)
{
VkMappedMemoryRange range = {VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, nullptr, m_memory,
m_map_offset, m_map_size};
vkInvalidateMappedMemoryRanges(g_vulkan_context->GetDevice(), 1, &range);
}
}
void StagingTexture2DLinear::CopyToImage(VkCommandBuffer command_buffer, VkImage image,
VkImageAspectFlags dst_aspect, u32 x, u32 y, u32 width,
u32 height, u32 level, u32 layer)
{
// Flush memory range if currently mapped.
if (m_map_pointer && !m_coherent)
{
VkMappedMemoryRange range = {VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, nullptr, m_memory,
m_map_offset, m_map_size};
vkFlushMappedMemoryRanges(g_vulkan_context->GetDevice(), 1, &range);
}
// Ensure any writes to the image are visible to the GPU.
VkImageMemoryBarrier barrier = {
VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER, // VkStructureType sType
nullptr, // const void* pNext
VK_ACCESS_HOST_WRITE_BIT, // VkAccessFlags srcAccessMask
VK_ACCESS_TRANSFER_READ_BIT, // VkAccessFlags dstAccessMask
m_layout, // VkImageLayout oldLayout
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, // VkImageLayout newLayout
VK_QUEUE_FAMILY_IGNORED, // uint32_t srcQueueFamilyIndex
VK_QUEUE_FAMILY_IGNORED, // uint32_t dstQueueFamilyIndex
m_image, // VkImage image
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 1, 0, 1} // VkImageSubresourceRange subresourceRange
};
vkCmdPipelineBarrier(command_buffer, VK_PIPELINE_STAGE_HOST_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
0, 0, nullptr, 0, nullptr, 1, &barrier);
m_layout = VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL;
// Issue the image copy, host -> gpu.
VkImageCopy copy_region = {
{VK_IMAGE_ASPECT_COLOR_BIT, 0, 0, 1}, // VkImageSubresourceLayers srcSubresource
{0, 0, 0}, // VkOffset3D srcOffset
{dst_aspect, level, layer, 1}, // VkImageSubresourceLayers dstSubresource
{static_cast<s32>(x), static_cast<s32>(y), 0}, // VkOffset3D dstOffset
{width, height, 1} // VkExtent3D extent
};
vkCmdCopyImage(command_buffer, m_image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, image,
VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &copy_region);
}
bool StagingTexture2DLinear::Map(VkDeviceSize offset /* = 0 */,
VkDeviceSize size /* = VK_WHOLE_SIZE */)
{
m_map_offset = offset;
if (size == VK_WHOLE_SIZE)
m_map_size = m_size - offset;
else
m_map_size = size;
_assert_(!m_map_pointer);
_assert_(m_map_offset + m_map_size <= m_size);
void* map_pointer;
VkResult res = vkMapMemory(g_vulkan_context->GetDevice(), m_memory, m_map_offset, m_map_size, 0,
&map_pointer);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkMapMemory failed: ");
return false;
}
m_map_pointer = reinterpret_cast<char*>(map_pointer);
return true;
}
void StagingTexture2DLinear::Unmap()
{
_assert_(m_map_pointer);
vkUnmapMemory(g_vulkan_context->GetDevice(), m_memory);
m_map_pointer = nullptr;
m_map_offset = 0;
m_map_size = 0;
}
std::unique_ptr<StagingTexture2D>
StagingTexture2DLinear::Create(STAGING_BUFFER_TYPE type, u32 width, u32 height, VkFormat format)
{
VkImageUsageFlags usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT;
VkImageCreateInfo create_info = {
VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO, // VkStructureType sType
nullptr, // const void* pNext
0, // VkImageCreateFlags flags
VK_IMAGE_TYPE_2D, // VkImageType imageType
format, // VkFormat format
{width, height, 1}, // VkExtent3D extent
1, // uint32_t mipLevels
1, // uint32_t arrayLayers
VK_SAMPLE_COUNT_1_BIT, // VkSampleCountFlagBits samples
VK_IMAGE_TILING_LINEAR, // VkImageTiling tiling
usage, // VkImageUsageFlags usage
VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode
0, // uint32_t queueFamilyIndexCount
nullptr, // const uint32_t* pQueueFamilyIndices
VK_IMAGE_LAYOUT_PREINITIALIZED // VkImageLayout initialLayout
};
VkImage image;
VkResult res = vkCreateImage(g_vulkan_context->GetDevice(), &create_info, nullptr, &image);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkCreateImage failed: ");
return nullptr;
}
VkMemoryRequirements memory_requirements;
vkGetImageMemoryRequirements(g_vulkan_context->GetDevice(), image, &memory_requirements);
bool is_coherent;
u32 memory_type_index;
if (type == STAGING_BUFFER_TYPE_READBACK)
{
memory_type_index =
g_vulkan_context->GetReadbackMemoryType(memory_requirements.memoryTypeBits, &is_coherent);
}
else
{
memory_type_index =
g_vulkan_context->GetUploadMemoryType(memory_requirements.memoryTypeBits, &is_coherent);
}
VkMemoryAllocateInfo memory_allocate_info = {
VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, // VkStructureType sType
nullptr, // const void* pNext
memory_requirements.size, // VkDeviceSize allocationSize
memory_type_index // uint32_t memoryTypeIndex
};
VkDeviceMemory memory;
res = vkAllocateMemory(g_vulkan_context->GetDevice(), &memory_allocate_info, nullptr, &memory);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkAllocateMemory failed: ");
vkDestroyImage(g_vulkan_context->GetDevice(), image, nullptr);
return nullptr;
}
res = vkBindImageMemory(g_vulkan_context->GetDevice(), image, memory, 0);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkBindImageMemory failed: ");
vkDestroyImage(g_vulkan_context->GetDevice(), image, nullptr);
vkFreeMemory(g_vulkan_context->GetDevice(), memory, nullptr);
return nullptr;
}
// Assume tight packing. Is this correct?
u32 stride = width * Util::GetTexelSize(format);
return std::make_unique<StagingTexture2DLinear>(type, width, height, format, stride, image,
memory, memory_requirements.size, is_coherent);
}
StagingTexture2DBuffer::StagingTexture2DBuffer(STAGING_BUFFER_TYPE type, u32 width, u32 height,
VkFormat format, u32 stride, VkBuffer buffer,
VkDeviceMemory memory, VkDeviceSize size,
bool coherent)
: StagingTexture2D(type, width, height, format, stride), m_buffer(buffer), m_memory(memory),
m_size(size), m_coherent(coherent)
{
}
StagingTexture2DBuffer::~StagingTexture2DBuffer()
{
if (m_map_pointer)
Unmap();
g_command_buffer_mgr->DeferDeviceMemoryDestruction(m_memory);
g_command_buffer_mgr->DeferBufferDestruction(m_buffer);
}
void StagingTexture2DBuffer::CopyFromImage(VkCommandBuffer command_buffer, VkImage image,
VkImageAspectFlags src_aspect, u32 x, u32 y, u32 width,
u32 height, u32 level, u32 layer)
{ {
// Issue the image->buffer copy. // Issue the image->buffer copy.
VkBufferImageCopy image_copy = { VkBufferImageCopy image_copy = {
0, // VkDeviceSize bufferOffset y * m_row_stride + x * m_texel_size, // VkDeviceSize bufferOffset
m_width, // uint32_t bufferRowLength m_width, // uint32_t bufferRowLength
0, // uint32_t bufferImageHeight 0, // uint32_t bufferImageHeight
{src_aspect, level, layer, 1}, // VkImageSubresourceLayers imageSubresource {src_aspect, level, layer, 1}, // VkImageSubresourceLayers imageSubresource
@ -382,42 +112,28 @@ void StagingTexture2DBuffer::CopyFromImage(VkCommandBuffer command_buffer, VkIma
vkCmdCopyImageToBuffer(command_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, m_buffer, 1, vkCmdCopyImageToBuffer(command_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, m_buffer, 1,
&image_copy); &image_copy);
// Ensure the write has completed. // Flush CPU and GPU caches if not coherent mapping.
VkDeviceSize copy_size = m_row_stride * height; VkDeviceSize buffer_flush_offset = y * m_row_stride;
Util::BufferMemoryBarrier(command_buffer, m_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VkDeviceSize buffer_flush_size = height * m_row_stride;
VK_ACCESS_HOST_READ_BIT, 0, copy_size, VK_PIPELINE_STAGE_TRANSFER_BIT, FlushGPUCache(command_buffer, VK_ACCESS_TRANSFER_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_HOST_BIT); buffer_flush_offset, buffer_flush_size);
InvalidateCPUCache(buffer_flush_offset, buffer_flush_size);
// If we're still mapped, invalidate the mapped range
if (m_map_pointer && !m_coherent)
{
VkMappedMemoryRange range = {VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, nullptr, m_memory,
m_map_offset, m_map_size};
vkInvalidateMappedMemoryRanges(g_vulkan_context->GetDevice(), 1, &range);
}
} }
void StagingTexture2DBuffer::CopyToImage(VkCommandBuffer command_buffer, VkImage image, void StagingTexture2D::CopyToImage(VkCommandBuffer command_buffer, VkImage image,
VkImageAspectFlags dst_aspect, u32 x, u32 y, u32 width, VkImageAspectFlags dst_aspect, u32 x, u32 y, u32 width,
u32 height, u32 level, u32 layer) u32 height, u32 level, u32 layer)
{ {
// If we're still mapped, flush the mapped range // Flush CPU and GPU caches if not coherent mapping.
if (m_map_pointer && !m_coherent) VkDeviceSize buffer_flush_offset = y * m_row_stride;
{ VkDeviceSize buffer_flush_size = height * m_row_stride;
VkMappedMemoryRange range = {VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, nullptr, m_memory, FlushCPUCache(buffer_flush_offset, buffer_flush_size);
m_map_offset, m_map_size}; InvalidateGPUCache(command_buffer, VK_ACCESS_HOST_WRITE_BIT, VK_PIPELINE_STAGE_TRANSFER_BIT,
vkFlushMappedMemoryRanges(g_vulkan_context->GetDevice(), 1, &range); buffer_flush_offset, buffer_flush_size);
}
// Ensure writes are visible to GPU. // Issue the buffer->image copy.
VkDeviceSize copy_size = m_row_stride * height;
Util::BufferMemoryBarrier(command_buffer, m_buffer, VK_ACCESS_HOST_WRITE_BIT,
VK_ACCESS_TRANSFER_READ_BIT, 0, copy_size, VK_PIPELINE_STAGE_HOST_BIT,
VK_PIPELINE_STAGE_TRANSFER_BIT);
// Issue the buffer->image copy
VkBufferImageCopy image_copy = { VkBufferImageCopy image_copy = {
0, // VkDeviceSize bufferOffset y * m_row_stride + x * m_texel_size, // VkDeviceSize bufferOffset
m_width, // uint32_t bufferRowLength m_width, // uint32_t bufferRowLength
0, // uint32_t bufferImageHeight 0, // uint32_t bufferImageHeight
{dst_aspect, level, layer, 1}, // VkImageSubresourceLayers imageSubresource {dst_aspect, level, layer, 1}, // VkImageSubresourceLayers imageSubresource
@ -428,109 +144,21 @@ void StagingTexture2DBuffer::CopyToImage(VkCommandBuffer command_buffer, VkImage
&image_copy); &image_copy);
} }
bool StagingTexture2DBuffer::Map(VkDeviceSize offset /* = 0 */, std::unique_ptr<StagingTexture2D> StagingTexture2D::Create(STAGING_BUFFER_TYPE type, u32 width,
VkDeviceSize size /* = VK_WHOLE_SIZE */) u32 height, VkFormat format)
{
m_map_offset = offset;
if (size == VK_WHOLE_SIZE)
m_map_size = m_size - offset;
else
m_map_size = size;
_assert_(!m_map_pointer);
_assert_(m_map_offset + m_map_size <= m_size);
void* map_pointer;
VkResult res = vkMapMemory(g_vulkan_context->GetDevice(), m_memory, m_map_offset, m_map_size, 0,
&map_pointer);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkMapMemory failed: ");
return false;
}
m_map_pointer = reinterpret_cast<char*>(map_pointer);
return true;
}
void StagingTexture2DBuffer::Unmap()
{
_assert_(m_map_pointer);
vkUnmapMemory(g_vulkan_context->GetDevice(), m_memory);
m_map_pointer = nullptr;
m_map_offset = 0;
m_map_size = 0;
}
std::unique_ptr<StagingTexture2D>
StagingTexture2DBuffer::Create(STAGING_BUFFER_TYPE type, u32 width, u32 height, VkFormat format)
{ {
// Assume tight packing. // Assume tight packing.
u32 row_stride = Util::GetTexelSize(format) * width; u32 stride = Util::GetTexelSize(format) * width;
u32 buffer_size = row_stride * height; u32 size = stride * height;
VkImageUsageFlags usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT; VkImageUsageFlags usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_TRANSFER_DST_BIT;
VkBufferCreateInfo buffer_create_info = {
VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType
nullptr, // const void* pNext
0, // VkBufferCreateFlags flags
buffer_size, // VkDeviceSize size
usage, // VkBufferUsageFlags usage
VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode
0, // uint32_t queueFamilyIndexCount
nullptr // const uint32_t* pQueueFamilyIndices
};
VkBuffer buffer; VkBuffer buffer;
VkResult res =
vkCreateBuffer(g_vulkan_context->GetDevice(), &buffer_create_info, nullptr, &buffer);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkCreateBuffer failed: ");
return nullptr;
}
VkMemoryRequirements memory_requirements;
vkGetBufferMemoryRequirements(g_vulkan_context->GetDevice(), buffer, &memory_requirements);
bool is_coherent;
u32 memory_type_index;
if (type == STAGING_BUFFER_TYPE_READBACK)
{
memory_type_index =
g_vulkan_context->GetReadbackMemoryType(memory_requirements.memoryTypeBits, &is_coherent);
}
else
{
memory_type_index =
g_vulkan_context->GetUploadMemoryType(memory_requirements.memoryTypeBits, &is_coherent);
}
VkMemoryAllocateInfo memory_allocate_info = {
VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO, // VkStructureType sType
nullptr, // const void* pNext
memory_requirements.size, // VkDeviceSize allocationSize
memory_type_index // uint32_t memoryTypeIndex
};
VkDeviceMemory memory; VkDeviceMemory memory;
res = vkAllocateMemory(g_vulkan_context->GetDevice(), &memory_allocate_info, nullptr, &memory); bool coherent;
if (res != VK_SUCCESS) if (!AllocateBuffer(type, size, usage, &buffer, &memory, &coherent))
{
LOG_VULKAN_ERROR(res, "vkAllocateMemory failed: ");
vkDestroyBuffer(g_vulkan_context->GetDevice(), buffer, nullptr);
return nullptr; return nullptr;
}
res = vkBindBufferMemory(g_vulkan_context->GetDevice(), buffer, memory, 0); return std::make_unique<StagingTexture2D>(type, buffer, memory, size, coherent, width, height,
if (res != VK_SUCCESS) format, stride);
{
LOG_VULKAN_ERROR(res, "vkBindBufferMemory failed: ");
vkDestroyBuffer(g_vulkan_context->GetDevice(), buffer, nullptr);
vkFreeMemory(g_vulkan_context->GetDevice(), memory, nullptr);
return nullptr;
}
return std::make_unique<StagingTexture2DBuffer>(type, width, height, format, row_stride, buffer,
memory, buffer_size, is_coherent);
} }
} // namespace Vulkan } // namespace Vulkan

91
Source/Core/VideoBackends/Vulkan/StagingTexture2D.h
View File

@ -9,29 +9,26 @@
#include "Common/CommonTypes.h" #include "Common/CommonTypes.h"
#include "VideoBackends/Vulkan/Constants.h" #include "VideoBackends/Vulkan/Constants.h"
#include "VideoBackends/Vulkan/StagingBuffer.h"
namespace Vulkan namespace Vulkan
{ {
class StagingTexture2D class StagingTexture2D final : public StagingBuffer
{ {
public: public:
StagingTexture2D(STAGING_BUFFER_TYPE type, u32 width, u32 height, VkFormat format, u32 stride); StagingTexture2D(STAGING_BUFFER_TYPE type, VkBuffer buffer, VkDeviceMemory memory,
virtual ~StagingTexture2D(); VkDeviceSize size, bool coherent, u32 width, u32 height, VkFormat format,
u32 stride);
~StagingTexture2D();
STAGING_BUFFER_TYPE GetType() const { return m_type; }
u32 GetWidth() const { return m_width; } u32 GetWidth() const { return m_width; }
u32 GetHeight() const { return m_height; } u32 GetHeight() const { return m_height; }
VkFormat GetFormat() const { return m_format; } VkFormat GetFormat() const { return m_format; }
u32 GetRowStride() const { return m_row_stride; } u32 GetRowStride() const { return m_row_stride; }
u32 GetTexelSize() const { return m_texel_size; } u32 GetTexelSize() const { return m_texel_size; }
bool IsMapped() const { return m_map_pointer != nullptr; } // Requires Map() to be called first.
const char* GetMapPointer() const { return m_map_pointer; }
char* GetMapPointer() { return m_map_pointer; }
VkDeviceSize GetMapOffset() const { return m_map_offset; }
VkDeviceSize GetMapSize() const { return m_map_size; }
const char* GetRowPointer(u32 row) const { return m_map_pointer + row * m_row_stride; } const char* GetRowPointer(u32 row) const { return m_map_pointer + row * m_row_stride; }
char* GetRowPointer(u32 row) { return m_map_pointer + row * m_row_stride; } char* GetRowPointer(u32 row) { return m_map_pointer + row * m_row_stride; }
// Requires Map() to be called first.
void ReadTexel(u32 x, u32 y, void* data, size_t data_size) const; void ReadTexel(u32 x, u32 y, void* data, size_t data_size) const;
void WriteTexel(u32 x, u32 y, const void* data, size_t data_size); void WriteTexel(u32 x, u32 y, const void* data, size_t data_size);
void ReadTexels(u32 x, u32 y, u32 width, u32 height, void* data, u32 data_stride) const; void ReadTexels(u32 x, u32 y, u32 width, u32 height, void* data, u32 data_stride) const;
@ -39,89 +36,23 @@ public:
// Assumes that image is in TRANSFER_SRC layout. // Assumes that image is in TRANSFER_SRC layout.
// Results are not ready until command_buffer has been executed. // Results are not ready until command_buffer has been executed.
virtual void CopyFromImage(VkCommandBuffer command_buffer, VkImage image, void CopyFromImage(VkCommandBuffer command_buffer, VkImage image, VkImageAspectFlags src_aspect,
VkImageAspectFlags src_aspect, u32 x, u32 y, u32 width, u32 height, u32 x, u32 y, u32 width, u32 height, u32 level, u32 layer);
u32 level, u32 layer) = 0;
// Assumes that image is in TRANSFER_DST layout. // Assumes that image is in TRANSFER_DST layout.
// Buffer is not safe for re-use until after command_buffer has been executed. // Buffer is not safe for re-use until after command_buffer has been executed.
virtual void CopyToImage(VkCommandBuffer command_buffer, VkImage image, void CopyToImage(VkCommandBuffer command_buffer, VkImage image, VkImageAspectFlags dst_aspect,
VkImageAspectFlags dst_aspect, u32 x, u32 y, u32 width, u32 height, u32 x, u32 y, u32 width, u32 height, u32 level, u32 layer);
u32 level, u32 layer) = 0;
virtual bool Map(VkDeviceSize offset = 0, VkDeviceSize size = VK_WHOLE_SIZE) = 0;
virtual void Unmap() = 0;
// Creates the optimal format of image copy. // Creates the optimal format of image copy.
static std::unique_ptr<StagingTexture2D> Create(STAGING_BUFFER_TYPE type, u32 width, u32 height, static std::unique_ptr<StagingTexture2D> Create(STAGING_BUFFER_TYPE type, u32 width, u32 height,
VkFormat format); VkFormat format);
protected: protected:
STAGING_BUFFER_TYPE m_type;
u32 m_width; u32 m_width;
u32 m_height; u32 m_height;
VkFormat m_format; VkFormat m_format;
u32 m_texel_size; u32 m_texel_size;
u32 m_row_stride; u32 m_row_stride;
char* m_map_pointer = nullptr;
VkDeviceSize m_map_offset = 0;
VkDeviceSize m_map_size = 0;
};
class StagingTexture2DLinear : public StagingTexture2D
{
public:
StagingTexture2DLinear(STAGING_BUFFER_TYPE type, u32 width, u32 height, VkFormat format,
u32 stride, VkImage image, VkDeviceMemory memory, VkDeviceSize size,
bool coherent);
~StagingTexture2DLinear();
void CopyFromImage(VkCommandBuffer command_buffer, VkImage image, VkImageAspectFlags src_aspect,
u32 x, u32 y, u32 width, u32 height, u32 level, u32 layer) override;
void CopyToImage(VkCommandBuffer command_buffer, VkImage image, VkImageAspectFlags dst_aspect,
u32 x, u32 y, u32 width, u32 height, u32 level, u32 layer) override;
bool Map(VkDeviceSize offset = 0, VkDeviceSize size = VK_WHOLE_SIZE) override;
void Unmap() override;
static std::unique_ptr<StagingTexture2D> Create(STAGING_BUFFER_TYPE type, u32 width, u32 height,
VkFormat format);
private:
VkImage m_image;
VkDeviceMemory m_memory;
VkDeviceSize m_size;
VkImageLayout m_layout;
bool m_coherent;
};
class StagingTexture2DBuffer : public StagingTexture2D
{
public:
StagingTexture2DBuffer(STAGING_BUFFER_TYPE type, u32 width, u32 height, VkFormat format,
u32 stride, VkBuffer buffer, VkDeviceMemory memory, VkDeviceSize size,
bool coherent);
~StagingTexture2DBuffer();
void CopyFromImage(VkCommandBuffer command_buffer, VkImage image, VkImageAspectFlags src_aspect,
u32 x, u32 y, u32 width, u32 height, u32 level, u32 layer) override;
void CopyToImage(VkCommandBuffer command_buffer, VkImage image, VkImageAspectFlags dst_aspect,
u32 x, u32 y, u32 width, u32 height, u32 level, u32 layer) override;
bool Map(VkDeviceSize offset = 0, VkDeviceSize size = VK_WHOLE_SIZE) override;
void Unmap() override;
static std::unique_ptr<StagingTexture2D> Create(STAGING_BUFFER_TYPE type, u32 width, u32 height,
VkFormat format);
private:
VkBuffer m_buffer;
VkDeviceMemory m_memory;
VkDeviceSize m_size;
bool m_coherent;
}; };
} }