dolphin/Source/Core/VideoBackends/Vulkan/StreamBuffer.cpp
Stenzek 37550501cc Vulkan: Fix incorrect handling of buffer wrap-around in StreamBuffer
This was happening when a fence wait happened mid-frame. The data written
between the fence being queued and the allocation occuring was incorrectly
assumed to be consumed by the GPU.
2016-11-30 22:44:52 +10:00

347 lines
12 KiB
C++

// Copyright 2016 Dolphin Emulator Project
// Licensed under GPLv2+
// Refer to the license.txt file included.
#include "VideoBackends/Vulkan/StreamBuffer.h"
#include <algorithm>
#include <cstdint>
#include <functional>
#include "Common/Assert.h"
#include "Common/MsgHandler.h"
#include "VideoBackends/Vulkan/CommandBufferManager.h"
#include "VideoBackends/Vulkan/Util.h"
#include "VideoBackends/Vulkan/VulkanContext.h"
namespace Vulkan
{
StreamBuffer::StreamBuffer(VkBufferUsageFlags usage, size_t max_size)
: m_usage(usage), m_maximum_size(max_size)
{
// Add a callback that fires on fence point creation and signal
g_command_buffer_mgr->AddFencePointCallback(
this, std::bind(&StreamBuffer::OnCommandBufferQueued, this, std::placeholders::_1,
std::placeholders::_2),
std::bind(&StreamBuffer::OnCommandBufferExecuted, this, std::placeholders::_1));
}
StreamBuffer::~StreamBuffer()
{
g_command_buffer_mgr->RemoveFencePointCallback(this);
if (m_host_pointer)
vkUnmapMemory(g_vulkan_context->GetDevice(), m_memory);
if (m_buffer != VK_NULL_HANDLE)
g_command_buffer_mgr->DeferBufferDestruction(m_buffer);
if (m_memory != VK_NULL_HANDLE)
g_command_buffer_mgr->DeferDeviceMemoryDestruction(m_memory);
}
std::unique_ptr<StreamBuffer> StreamBuffer::Create(VkBufferUsageFlags usage, size_t initial_size,
size_t max_size)
{
std::unique_ptr<StreamBuffer> buffer = std::make_unique<StreamBuffer>(usage, max_size);
if (!buffer->ResizeBuffer(initial_size))
return nullptr;
return buffer;
}
bool StreamBuffer::ResizeBuffer(size_t size)
{
// Create the buffer descriptor
VkBufferCreateInfo buffer_create_info = {
VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO, // VkStructureType sType
nullptr, // const void* pNext
0, // VkBufferCreateFlags flags
static_cast<VkDeviceSize>(size), // VkDeviceSize size
m_usage, // VkBufferUsageFlags usage
VK_SHARING_MODE_EXCLUSIVE, // VkSharingMode sharingMode
0, // uint32_t queueFamilyIndexCount
nullptr // const uint32_t* pQueueFamilyIndices
};
VkBuffer buffer = VK_NULL_HANDLE;
VkResult res =
vkCreateBuffer(g_vulkan_context->GetDevice(), &buffer_create_info, nullptr, &buffer);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkCreateBuffer failed: ");
return false;
}
// Get memory requirements (types etc) for this buffer
VkMemoryRequirements memory_requirements;
vkGetBufferMemoryRequirements(g_vulkan_context->GetDevice(), buffer, &memory_requirements);
// Aim for a coherent mapping if possible.
u32 memory_type_index = g_vulkan_context->GetUploadMemoryType(memory_requirements.memoryTypeBits,
&m_coherent_mapping);
// Allocate memory for backing this buffer
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 = VK_NULL_HANDLE;
res = vkAllocateMemory(g_vulkan_context->GetDevice(), &memory_allocate_info, nullptr, &memory);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkAllocateMemory failed: ");
vkDestroyBuffer(g_vulkan_context->GetDevice(), buffer, nullptr);
return false;
}
// Bind memory to buffer
res = vkBindBufferMemory(g_vulkan_context->GetDevice(), buffer, memory, 0);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkBindBufferMemory failed: ");
vkDestroyBuffer(g_vulkan_context->GetDevice(), buffer, nullptr);
vkFreeMemory(g_vulkan_context->GetDevice(), memory, nullptr);
return false;
}
// Map this buffer into user-space
void* mapped_ptr = nullptr;
res = vkMapMemory(g_vulkan_context->GetDevice(), memory, 0, size, 0, &mapped_ptr);
if (res != VK_SUCCESS)
{
LOG_VULKAN_ERROR(res, "vkMapMemory failed: ");
vkDestroyBuffer(g_vulkan_context->GetDevice(), buffer, nullptr);
vkFreeMemory(g_vulkan_context->GetDevice(), memory, nullptr);
return false;
}
// Unmap current host pointer (if there was a previous buffer)
if (m_host_pointer)
vkUnmapMemory(g_vulkan_context->GetDevice(), m_memory);
// Destroy the backings for the buffer after the command buffer executes
if (m_buffer != VK_NULL_HANDLE)
g_command_buffer_mgr->DeferBufferDestruction(m_buffer);
if (m_memory != VK_NULL_HANDLE)
g_command_buffer_mgr->DeferDeviceMemoryDestruction(m_memory);
// Replace with the new buffer
m_buffer = buffer;
m_memory = memory;
m_host_pointer = reinterpret_cast<u8*>(mapped_ptr);
m_current_size = size;
m_current_offset = 0;
m_current_gpu_position = 0;
m_tracked_fences.clear();
return true;
}
bool StreamBuffer::ReserveMemory(size_t num_bytes, size_t alignment, bool allow_reuse /* = true */,
bool allow_growth /* = true */,
bool reallocate_if_full /* = false */)
{
size_t required_bytes = num_bytes + alignment;
// Check for sane allocations
if (required_bytes > m_maximum_size)
{
PanicAlert("Attempting to allocate %u bytes from a %u byte stream buffer",
static_cast<uint32_t>(num_bytes), static_cast<uint32_t>(m_maximum_size));
return false;
}
// Is the GPU behind or up to date with our current offset?
if (m_current_offset >= m_current_gpu_position)
{
size_t remaining_bytes = m_current_size - m_current_offset;
if (required_bytes <= remaining_bytes)
{
// Place at the current position, after the GPU position.
m_current_offset = Util::AlignBufferOffset(m_current_offset, alignment);
m_last_allocation_size = num_bytes;
return true;
}
// Check for space at the start of the buffer
// We use < here because we don't want to have the case of m_current_offset ==
// m_current_gpu_position. That would mean the code above would assume the
// GPU has caught up to us, which it hasn't.
if (allow_reuse && required_bytes < m_current_gpu_position)
{
// Reset offset to zero, since we're allocating behind the gpu now
m_current_offset = 0;
m_last_allocation_size = num_bytes;
return true;
}
}
// Is the GPU ahead of our current offset?
if (m_current_offset < m_current_gpu_position)
{
// We have from m_current_offset..m_current_gpu_position space to use.
size_t remaining_bytes = m_current_gpu_position - m_current_offset;
if (required_bytes < remaining_bytes)
{
// Place at the current position, since this is still behind the GPU.
m_current_offset = Util::AlignBufferOffset(m_current_offset, alignment);
m_last_allocation_size = num_bytes;
return true;
}
}
// Try to grow the buffer up to the maximum size before waiting.
// Double each time until the maximum size is reached.
if (allow_growth && m_current_size < m_maximum_size)
{
size_t new_size = std::min(std::max(num_bytes, m_current_size * 2), m_maximum_size);
if (ResizeBuffer(new_size))
{
// Allocating from the start of the buffer.
m_last_allocation_size = new_size;
return true;
}
}
// Can we find a fence to wait on that will give us enough memory?
if (allow_reuse && WaitForClearSpace(required_bytes))
{
_assert_(m_current_offset == m_current_gpu_position ||
(m_current_offset + required_bytes) < m_current_gpu_position);
m_current_offset = Util::AlignBufferOffset(m_current_offset, alignment);
m_last_allocation_size = num_bytes;
return true;
}
// If we are not allowed to execute in our current state (e.g. in the middle of a render pass),
// as a last resort, reallocate the buffer. This will incur a performance hit and is not
// encouraged.
if (reallocate_if_full && ResizeBuffer(m_current_size))
{
m_last_allocation_size = num_bytes;
return true;
}
// We tried everything we could, and still couldn't get anything. If we're not at a point
// where the state is known and can be resumed, this is probably a fatal error.
return false;
}
void StreamBuffer::CommitMemory(size_t final_num_bytes)
{
_assert_((m_current_offset + final_num_bytes) <= m_current_size);
_assert_(final_num_bytes <= m_last_allocation_size);
// For non-coherent mappings, flush the memory range
if (!m_coherent_mapping)
{
VkMappedMemoryRange range = {VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE, nullptr, m_memory,
m_current_offset, final_num_bytes};
vkFlushMappedMemoryRanges(g_vulkan_context->GetDevice(), 1, &range);
}
m_current_offset += final_num_bytes;
}
void StreamBuffer::OnCommandBufferQueued(VkCommandBuffer command_buffer, VkFence fence)
{
// Don't create a tracking entry if the GPU is caught up with the buffer.
if (m_current_offset == m_current_gpu_position)
return;
// Has the offset changed since the last fence?
if (!m_tracked_fences.empty() && m_tracked_fences.back().second == m_current_offset)
{
// No need to track the new fence, the old one is sufficient.
return;
}
m_tracked_fences.emplace_back(fence, m_current_offset);
}
void StreamBuffer::OnCommandBufferExecuted(VkFence fence)
{
// Locate the entry for this fence (if any, we may have been forced to wait already)
auto iter = std::find_if(m_tracked_fences.begin(), m_tracked_fences.end(),
[fence](const auto& it) { return it.first == fence; });
if (iter != m_tracked_fences.end())
{
// Update the GPU position, and remove any fences before this fence (since
// it is implied that they have been signaled as well, though the callback
// should have removed them already).
m_current_gpu_position = iter->second;
m_tracked_fences.erase(m_tracked_fences.begin(), ++iter);
}
}
bool StreamBuffer::WaitForClearSpace(size_t num_bytes)
{
size_t new_offset = 0;
size_t new_gpu_position = 0;
auto iter = m_tracked_fences.begin();
for (; iter != m_tracked_fences.end(); iter++)
{
// Would this fence bring us in line with the GPU?
// This is the "last resort" case, where a command buffer execution has been forced
// after no additional data has been written to it, so we can assume that after the
// fence has been signaled the entire buffer is now consumed.
size_t gpu_position = iter->second;
if (m_current_offset == gpu_position)
{
// Start at the start of the buffer again.
new_offset = 0;
new_gpu_position = 0;
break;
}
// Assuming that we wait for this fence, are we allocating in front of the GPU?
if (m_current_offset > gpu_position)
{
// We can wrap around to the start, behind the GPU, if there is enough space.
// We use > here because otherwise we'd end up lining up with the GPU, and then the
// allocator would assume that the GPU has consumed what we just wrote.
if (gpu_position > num_bytes)
{
new_offset = 0;
new_gpu_position = gpu_position;
break;
}
}
else
{
// We're currently allocating behind the GPU. This would give us between the current
// offset and the GPU position worth of space to work with. Again, > because we can't
// align the GPU position with the buffer offset.
size_t available_space_inbetween = m_current_offset - gpu_position;
if (available_space_inbetween > num_bytes)
{
// Leave the offset as-is, but update the GPU position.
new_offset = m_current_offset;
new_gpu_position = gpu_position;
break;
}
}
}
// Did any fences satisfy this condition?
if (iter == m_tracked_fences.end())
return false;
// Wait until this fence is signaled.
VkResult res =
vkWaitForFences(g_vulkan_context->GetDevice(), 1, &iter->first, VK_TRUE, UINT64_MAX);
if (res != VK_SUCCESS)
LOG_VULKAN_ERROR(res, "vkWaitForFences failed: ");
// Update GPU position, and remove all fences up to (and including) this fence.
m_current_offset = new_offset;
m_current_gpu_position = new_gpu_position;
m_tracked_fences.erase(m_tracked_fences.begin(), ++iter);
return true;
}
} // namespace Vulkan