YCbCr Sampler in Vulkan - c++

I've been trying to sample a YCbCr image in Vulkan but I keep getting incorrect results, and I was hoping someone might be able to spot my mistake.
I have a NV12 YCbCr image which I want to render onto two triangles forming a quad. If i understand correctly, the VkFormat that corresponds to NV12 is VK_FORMAT_G8_B8R8_2PLANE_420_UNORM. Below is the code that I would expect to work, but I'll try to explain what I'm trying to do as well:
Create a VkSampler with a VkSamplerYcbcrConversion (with the correct format) in pNext
Read NV12 data into staging buffer
Create VkImage with the correct format and specify that the planes are disjoint
Get memory requirements (and offset for plane 1) for each plane (0 and 1)
Allocate device local memory for the image data
Bind each plane to the correct location in memory
Copy staging buffer to image memory
Create VkImageView with the same format as the VkImage and the same VkSamplerYcbcrConversionInfo as the VkSampler in pNext.
Code:
VkSamplerYcbcrConversion ycbcr_sampler_conversion;
VkSamplerYcbcrConversionInfo ycbcr_info;
VkSampler ycbcr_sampler;
VkImage image;
VkDeviceMemory image_memory;
VkDeviceSize memory_offset_plane0, memory_offset_plane1;
VkImageView image_view;
enum YCbCrStorageFormat
{
NV12
};
unsigned char* ReadYCbCrFile(const std::string& filename, YCbCrStorageFormat storage_format, VkFormat vulkan_format, uint32_t* buffer_size, uint32_t* buffer_offset_plane1, uint32_t* buffer_offset_plane2)
{
std::ifstream file;
file.open(filename.c_str(), std::ios::in | std::ios::binary | std::ios::ate);
if (!file.is_open()) { ELOG("Failed to open YCbCr image"); }
*buffer_size = file.tellg();
file.seekg(0);
unsigned char* data;
switch (storage_format)
{
case NV12:
{
if (vulkan_format != VK_FORMAT_G8_B8R8_2PLANE_420_UNORM)
{
ILOG("A 1:1 relationship doesn't exist between NV12 and 420, exiting");
exit(1);
}
*buffer_offset_plane1 = (*buffer_size / 3) * 2;
*buffer_offset_plane2 = 0; //Not used
data = new unsigned char[*buffer_size];
file.read((char*)(data), *buffer_size);
break;
}
default:
ELOG("A YCbCr storage format is required");
break;
}
file.close();
return data;
}
VkFormatProperties format_properties;
vkGetPhysicalDeviceFormatProperties(physical_device, VK_FORMAT_G8_B8R8_2PLANE_420_UNORM, &format_properties);
bool cosited = false, midpoint = false;
if (format_properties.optimalTilingFeatures & VK_FORMAT_FEATURE_COSITED_CHROMA_SAMPLES_BIT)
{
cosited = true;
}
else if (format_properties.optimalTilingFeatures & VK_FORMAT_FEATURE_MIDPOINT_CHROMA_SAMPLES_BIT)
{
midpoint = true;
}
if (!cosited && !midpoint)
{
ELOG("Nither VK_FORMAT_FEATURE_COSITED_CHROMA_SAMPLES_BIT nor VK_FORMAT_FEATURE_MIDPOINT_CHROMA_SAMPLES_BIT is supported for VK_FORMAT_G8_B8R8_2PLANE_420_UNORM");
}
VkSamplerYcbcrConversionCreateInfo conversion_info = {};
conversion_info.sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_CREATE_INFO;
conversion_info.pNext = NULL;
conversion_info.format = VK_FORMAT_G8_B8R8_2PLANE_420_UNORM;
conversion_info.ycbcrModel = VK_SAMPLER_YCBCR_MODEL_CONVERSION_YCBCR_709;
conversion_info.ycbcrRange = VK_SAMPLER_YCBCR_RANGE_ITU_FULL;
conversion_info.components.r = VK_COMPONENT_SWIZZLE_IDENTITY;
conversion_info.components.g = VK_COMPONENT_SWIZZLE_IDENTITY;
conversion_info.components.b = VK_COMPONENT_SWIZZLE_IDENTITY;
conversion_info.components.a = VK_COMPONENT_SWIZZLE_IDENTITY;
if (cosited)
{
conversion_info.xChromaOffset = VK_CHROMA_LOCATION_COSITED_EVEN;
conversion_info.yChromaOffset = VK_CHROMA_LOCATION_COSITED_EVEN;
}
else
{
conversion_info.xChromaOffset = VK_CHROMA_LOCATION_MIDPOINT;
conversion_info.yChromaOffset = VK_CHROMA_LOCATION_MIDPOINT;
}
conversion_info.chromaFilter = VK_FILTER_LINEAR;
conversion_info.forceExplicitReconstruction = VK_FALSE;
VkResult res = vkCreateSamplerYcbcrConversion(logical_device, &conversion_info, NULL, &ycbcr_sampler_conversion);
CHECK_VK_RESULT(res, "Failed to create YCbCr conversion sampler");
ILOG("Successfully created YCbCr conversion");
ycbcr_info.sType = VK_STRUCTURE_TYPE_SAMPLER_YCBCR_CONVERSION_INFO;
ycbcr_info.pNext = NULL;
ycbcr_info.conversion = ycbcr_sampler_conversion;
VkSamplerCreateInfo sampler_info = {};
sampler_info.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO;
sampler_info.pNext = &ycbcr_info;
sampler_info.flags = 0;
sampler_info.magFilter = VK_FILTER_LINEAR;
sampler_info.minFilter = VK_FILTER_LINEAR;
sampler_info.mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR;
sampler_info.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
sampler_info.addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
sampler_info.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE;
sampler_info.mipLodBias = 0.0f;
sampler_info.anisotropyEnable = VK_FALSE;
//sampler_info.maxAnisotropy IGNORED
sampler_info.compareEnable = VK_FALSE;
//sampler_info.compareOp = IGNORED
sampler_info.minLod = 0.0f;
sampler_info.maxLod = 1.0f;
sampler_info.borderColor = VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK;
sampler_info.unnormalizedCoordinates = VK_FALSE;
res = vkCreateSampler(logical_device, &sampler_info, NULL, &ycbcr_sampler);
CHECK_VK_RESULT(res, "Failed to create YUV sampler");
ILOG("Successfully created sampler with YCbCr in pNext");
std::string filename = "tree_nv12_1920x1080.yuv";
uint32_t width = 1920, height = 1080;
VkFormat format = VK_FORMAT_G8_B8R8_2PLANE_420_UNORM;
uint32_t buffer_size, buffer_offset_plane1, buffer_offset_plane2;
unsigned char* ycbcr_data = ReadYCbCrFile(filename, NV12, VK_FORMAT_G8_B8R8_2PLANE_420_UNORM, &buffer_size, &buffer_offset_plane1, &buffer_offset_plane2);
//Load image into staging buffer
VkDeviceMemory stage_buffer_memory;
VkBuffer stage_buffer = create_vk_buffer(buffer_size, VK_BUFFER_USAGE_TRANSFER_SRC_BIT, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT, stage_buffer_memory);
void* stage_memory_ptr;
vkMapMemory(logical_device, stage_buffer_memory, 0, buffer_size, 0, &stage_memory_ptr);
memcpy(stage_memory_ptr, ycbcr_data, buffer_size);
vkUnmapMemory(logical_device, stage_buffer_memory);
delete[] ycbcr_data;
//Create image
VkImageCreateInfo img_info = {};
img_info.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
img_info.flags = VK_IMAGE_CREATE_DISJOINT_BIT;
img_info.imageType = VK_IMAGE_TYPE_2D;
img_info.extent.width = width;
img_info.extent.height = height;
img_info.extent.depth = 1;
img_info.mipLevels = 1;
img_info.arrayLayers = 1;
img_info.format = format;
img_info.tiling = VK_IMAGE_TILING_LINEAR;//VK_IMAGE_TILING_OPTIMAL;
img_info.initialLayout = VK_IMAGE_LAYOUT_PREINITIALIZED;
img_info.usage = VK_BUFFER_USAGE_TRANSFER_DST_BIT | VK_IMAGE_USAGE_SAMPLED_BIT;
img_info.samples = VK_SAMPLE_COUNT_1_BIT;
img_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
VkResult result = vkCreateImage(logical_device, &img_info, NULL, &image);
CHECK_VK_RESULT(result, "vkCreateImage failed to create image handle");
ILOG("Image created!");
//Get memory requirements for each plane and combine
//Plane 0
VkImagePlaneMemoryRequirementsInfo image_plane_info = {};
image_plane_info.sType = VK_STRUCTURE_TYPE_IMAGE_PLANE_MEMORY_REQUIREMENTS_INFO;
image_plane_info.pNext = NULL;
image_plane_info.planeAspect = VK_IMAGE_ASPECT_PLANE_0_BIT;
VkImageMemoryRequirementsInfo2 image_info2 = {};
image_info2.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_REQUIREMENTS_INFO_2;
image_info2.pNext = &image_plane_info;
image_info2.image = image;
VkImagePlaneMemoryRequirementsInfo memory_plane_requirements = {};
memory_plane_requirements.sType = VK_STRUCTURE_TYPE_IMAGE_PLANE_MEMORY_REQUIREMENTS_INFO;
memory_plane_requirements.pNext = NULL;
memory_plane_requirements.planeAspect = VK_IMAGE_ASPECT_PLANE_0_BIT;
VkMemoryRequirements2 memory_requirements2 = {};
memory_requirements2.sType = VK_STRUCTURE_TYPE_MEMORY_REQUIREMENTS_2;
memory_requirements2.pNext = &memory_plane_requirements;
vkGetImageMemoryRequirements2(logical_device, &image_info2, &memory_requirements2);
VkDeviceSize image_size = memory_requirements2.memoryRequirements.size;
uint32_t image_bits = memory_requirements2.memoryRequirements.memoryTypeBits;
//Set offsets
memory_offset_plane0 = 0;
memory_offset_plane1 = image_size;
//Plane 1
image_plane_info.planeAspect = VK_IMAGE_ASPECT_PLANE_1_BIT;
memory_plane_requirements.planeAspect = VK_IMAGE_ASPECT_PLANE_1_BIT;
vkGetImageMemoryRequirements2(logical_device, &image_info2, &memory_requirements2);
image_size += memory_requirements2.memoryRequirements.size;
image_bits = image_bits | memory_requirements2.memoryRequirements.memoryTypeBits;
//Allocate image memory
VkMemoryAllocateInfo allocate_info = {};
allocate_info.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO;
allocate_info.allocationSize = image_size;
allocate_info.memoryTypeIndex = get_device_memory_type(image_bits, VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT);
result = vkAllocateMemory(logical_device, &allocate_info, NULL, &image_memory);
CHECK_VK_RESULT(result, "vkAllocateMemory failed to allocate image memory");
//Bind each image plane to memory
std::vector<VkBindImageMemoryInfo> bind_image_memory_infos(2);
//Plane 0
VkBindImagePlaneMemoryInfo bind_image_plane0_info = {};
bind_image_plane0_info.sType = VK_STRUCTURE_TYPE_BIND_IMAGE_PLANE_MEMORY_INFO;
bind_image_plane0_info.pNext = NULL;
bind_image_plane0_info.planeAspect = VK_IMAGE_ASPECT_PLANE_0_BIT;
VkBindImageMemoryInfo& bind_image_memory_plane0_info = bind_image_memory_infos[0];
bind_image_memory_plane0_info.sType = VK_STRUCTURE_TYPE_BIND_IMAGE_MEMORY_INFO;
bind_image_memory_plane0_info.pNext = &bind_image_plane0_info;
bind_image_memory_plane0_info.image = image;
bind_image_memory_plane0_info.memory = image_memory;
bind_image_memory_plane0_info.memoryOffset = memory_offset_plane0;
//Plane 1
VkBindImagePlaneMemoryInfo bind_image_plane1_info = {};
bind_image_plane1_info.sType = VK_STRUCTURE_TYPE_BIND_IMAGE_PLANE_MEMORY_INFO;
bind_image_plane1_info.pNext = NULL;
bind_image_plane1_info.planeAspect = VK_IMAGE_ASPECT_PLANE_1_BIT;
VkBindImageMemoryInfo& bind_image_memory_plane1_info = bind_image_memory_infos[1];
bind_image_memory_plane1_info.sType = VK_STRUCTURE_TYPE_BIND_IMAGE_MEMORY_INFO;
bind_image_memory_plane1_info.pNext = &bind_image_plane1_info;
bind_image_memory_plane1_info.image = image;
bind_image_memory_plane1_info.memory = image_memory;
bind_image_memory_plane1_info.memoryOffset = memory_offset_plane1;
vkBindImageMemory2(logical_device, bind_image_memory_infos.size(), bind_image_memory_infos.data());
context.transition_vk_image_layout(image, format, VK_IMAGE_ASPECT_COLOR_BIT, VK_IMAGE_LAYOUT_PREINITIALIZED, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
//Copy staging buffer to device local buffer
VkCommandBuffer tmp_cmd_buffer = begin_tmp_vk_cmd_buffer();
std::vector<VkBufferImageCopy> plane_regions(2);
plane_regions[0].bufferOffset = 0;
plane_regions[0].bufferRowLength = 0;
plane_regions[0].bufferImageHeight = 0;
plane_regions[0].imageSubresource.aspectMask = VK_IMAGE_ASPECT_PLANE_0_BIT;
plane_regions[0].imageSubresource.mipLevel = 0;
plane_regions[0].imageSubresource.baseArrayLayer = 0;
plane_regions[0].imageSubresource.layerCount = 1;
plane_regions[0].imageOffset = { 0, 0, 0 };
plane_regions[0].imageExtent = { width, height, 1 };
plane_regions[1].bufferOffset = buffer_offset_plane1;
plane_regions[1].bufferRowLength = 0;
plane_regions[1].bufferImageHeight = 0;
plane_regions[1].imageSubresource.aspectMask = VK_IMAGE_ASPECT_PLANE_1_BIT;
plane_regions[1].imageSubresource.mipLevel = 0;
plane_regions[1].imageSubresource.baseArrayLayer = 0;
plane_regions[1].imageSubresource.layerCount = 1;
plane_regions[1].imageOffset = { 0, 0, 0 };
plane_regions[1].imageExtent = { width / 2, height / 2, 1 };
vkCmdCopyBufferToImage(tmp_cmd_buffer, stage_buffer, image, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, plane_regions.size(), plane_regions.data());
end_tmp_vk_cmd_buffer(tmp_cmd_buffer); //Submit and waits
vkFreeMemory(logical_device, stage_buffer_memory, NULL);
vkDestroyBuffer(logical_device, stage_buffer, NULL);
transition_vk_image_layout(image, format, VK_IMAGE_ASPECT_COLOR_BIT, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL);
VkImageViewCreateInfo image_view_info = {};
image_view_info.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
image_view_info.pNext = &ycbcr_info;
image_view_info.flags = 0;
image_view_info.image = image;
image_view_info.viewType = VK_IMAGE_VIEW_TYPE_2D;
image_view_info.format = format;
image_view_info.components.r = VK_COMPONENT_SWIZZLE_IDENTITY;
image_view_info.components.b = VK_COMPONENT_SWIZZLE_IDENTITY;
image_view_info.components.g = VK_COMPONENT_SWIZZLE_IDENTITY;
image_view_info.components.a = VK_COMPONENT_SWIZZLE_IDENTITY;
image_view_info.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
image_view_info.subresourceRange.baseMipLevel = 0;
image_view_info.subresourceRange.levelCount = 1;
image_view_info.subresourceRange.baseArrayLayer = 0;
image_view_info.subresourceRange.layerCount = 1;
VkResult res = vkCreateImageView(logical_device, &image_view_info, NULL, &.image_view);
CHECK_VK_RESULT(res, "Failed to create image view");
ILOG("Successfully created image, allocated image memory and created image view");
I receive one validation error: vkCmdCopyBufferToImage() parameter, VkImageAspect pRegions->imageSubresource.aspectMask, is an unrecognized enumerator, but from inspecting the validation code, it seems that it's just a bit outdated and this shouldn't be an issue.
The rest of the code just sets up regular descriptor layouts/pools and allocated and updates accordingly (I've verified with a regular RGB texture).
The fragment shader is as follows:
vec2 uv = vec2(gl_FragCoord.x / 1024.0, 1.0 - (gl_FragCoord.y / 1024.0));
out_color = vec4(texture(ycbcr_image, uv).rgb, 1.0f);
When I run my program I only get a red components (the image is essentially a greyscale image). from a little testing, it seems that the VkSamplerYcbcrconversion setup as removing it from both the VkSamplerCreateInfo.pNext and VkImageViewCreateInfo.pNext doesn't change anything.
I've also looked here, Khronos YCbCr tests, but I can't find any real mistake.

Solution: according to the spec, sec. 12.1, Conversion must be fixed at pipeline creation time, through use of a combined image sampler with an immutable sampler in VkDescriptorSetLayoutBinding.
By adding the ycbcr_sampler to pImmutableSamplers when setting up the descriptor set layout binding it now works:
VkDescriptorSetLayoutBinding image_binding = {};
image_binding.binding = 0;
image_binding.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
image_binding.descriptorCount = 1;
image_binding.stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT;
image_binding.pImmutableSamplers = &ycbcr_sampler;

Related

I want to use a UAV in a pixel shader to read the data in the buffer with the CPU

I would like to have information on the number of vertices that have been increased by doing Tessellation.
To do this, we send the vertex information from the Domain Shader to the Pixel Shader and use the RWStructureBuffer in the Pixel Shader as shown below.
struct Data
{
float3 position;
};
RWStructuredBuffer<Data> rwBuffer0 : register(u1);
・・・
Data data;
data.position = input.position;
rwBuffer0[id] = data;
・・・
}
On the CPU side, we are trying to receive the following.
struct ReternUAV
{
DirectX::XMFLOA3 position;
};
HRESULT hr = S_OK;
Microsoft::WRL::ComPtr<ID3D11Buffer> outputBuffer;
D3D11_BUFFER_DESC outputDesc;
outputDesc.Usage = D3D11_USAGE_DEFAULT;
outputDesc.ByteWidth = sizeof(ReternUAV) * 10000;
outputDesc.BindFlags = D3D11_BIND_UNORDERED_ACCESS;
outputDesc.CPUAccessFlags = 0;
outputDesc.StructureByteStride = sizeof(ReternUAV);
outputDesc.MiscFlags = 0;
device->CreateBuffer(&outputDesc, nullptr, outputBuffer.GetAddressOf());
Microsoft::WRL::ComPtr<ID3D11Buffer> outputResultBuffer;
outputDesc.Usage = D3D11_USAGE_STAGING;
outputDesc.BindFlags = 0;
outputDesc.CPUAccessFlags = D3D11_CPU_ACCESS_READ;
device->CreateBuffer(&outputDesc, nullptr, outputResultBuffer.GetAddressOf());
D3D11_UNORDERED_ACCESS_VIEW_DESC uavDesc;
uavDesc.Buffer.FirstElement = 0;
uavDesc.Buffer.Flags = 0;
uavDesc.Buffer.NumElements = 10000;
uavDesc.Format = DXGI_FORMAT_R32G32B32_FLOAT;
uavDesc.ViewDimension = D3D11_UAV_DIMENSION_BUFFER;
Microsoft::WRL::ComPtr<ID3D11UnorderedAccessVie>unorderedAccessView;
hr = device->CreateUnorderedAccessView(outputBuffer.Get(), &uavDesc, unorderedAccessView.GetAddressOf());
if (FAILED(hr))
{
assert(!"CreateUnorderedAccessView"); // <ーFailed to create
}
ID3D11RenderTargetView* renderTarget = GameScene::GetRenderTargetView();
ID3D11DepthStencilView* deStencilView = GameScene::GetDepthStencilView();
context>OMSetRenderTargetsAndUnorderedAccessViews(1, &renderTarget, deStencilView,1, 1, unorderedAccessView.GetAddressOf(),NULL);
context->DrawIndexed(subset.indexCount, subset.indexStart, 0);
Microsoft::WRL::ComPtr<ID3D11UnorderedAccessView> unCom = nullptr;
context->OMSetRenderTargetsAndUnorderedAccessViews(1, &renderTarget, deStencilView,1, 1, unCom.GetAddressOf(),NULL);
context->CopyResource(outputResultBuffer.Get(), outputBuffer.Get());
D3D11_MAPPED_SUBRESOURCE mappedBuffer;
D3D11_MAP map = D3D11_MAP_READ;
hr = context->Map(outputResultBuffer.Get(), 0, map, 0, &mappedBuffer);
ReternUAV* copy = reinterpret_cast<ReternUAV*>(mappedBuffer.pData);
UINT num = sizeof(copy);
for (int i = 0; i < num; i++)
{
ReternUAV a = copy[i];
a = a;
}
context->Unmap(outputResultBuffer.Get(), 0);
It may be that the CreateUnorderedAccessView is failing to create it, but I couldn't figure out what was causing it.
If I ignore this and run,
The data in "copy" that I mapped and read is all 0,0,0 and there are only 8 elements.
I would like to ask you where I am going wrong.
If there is a better way to achieve the goal, I would like to hear about it.
Eventually, I would like to tessellation and handle the newly obtained data with the CPU.
Thank you very much for your help.
uavDesc.Format must be DXGI_FORMAT_UNKNOWN when creating a View of a Structured Buffer. Also "UINT num = sizeof(copy);" will not return the number of written vertices. :)
I recommend to create a device using D3D11_CREATE_DEVICE_DEBUG flag and then you will get an explanation why it failed to create the UAV. Just pass the flag to the D3D11CreateDevice().
The best way is to use D3D11_QUERY if you need only the number of vertices.
https://learn.microsoft.com/en-us/windows/win32/api/d3d11/ne-d3d11-d3d11_query
https://learn.microsoft.com/en-us/windows/win32/api/d3d11/ns-d3d11-d3d11_query_data_pipeline_statistics
D3D11_QUERY_DESC qdesc = {D3D11_QUERY_PIPELINE_STATISTICS};
ID3D11Query* query = 0;
device->CreateQuery(&qdesc, &query);
context->Begin(query);
context->DrawIndexed(index_count, 0, 0);
context->End(query);
D3D11_QUERY_DATA_PIPELINE_STATISTICS stats = {};
while (S_FALSE == context->GetData(query, &stats, sizeof(stats), 0))
;
query->Release();

Clip Raster with Polygon with GDAL C++

I am trying to clip a raster using a polygon an GDAL. At the moment i get an error that there is a read access violation when initializing the WarpOperation. I can access my Shapefile and check the num of features so the access is fine i think. Also i can access my Raster Data (GetProjectionRef).. All files are in the same CRS. Is there a way to use GdalWarp with Cutline?
const char* inputPath = "input.tif";
const char* outputPath = "output.tif";
//clipper Polygon
auto w_read_filenamePoly = "Polygon.shp";
char* read_filenamePoly = new char[w_read_filenamePoly.length() + 1];
wcstombs(read_filenamePoly, w_read_filenamePoly.c_str(), w_read_filenamePoly.length() + 1);
GDALDataset* hSrcDS;
GDALDataset* hDstDS;
GDALAllRegister();
hSrcDS =(GDALDataset *) GDALOpen(inputPath, GA_Update);
hDstDS = (GDALDataset*)GDALOpen(outputPath, GA_Update);
const char* proj = hSrcDS->GetProjectionRef();
const char* proj2 = hDstDS->GetProjectionRef();
//clipper Layer
GDALDataset* poDSClipper;
poDSClipper = (GDALDataset*)GDALOpenEx(read_filenamePoly, GDAL_OF_UPDATE, NULL, NULL, NULL);
Assert::IsNotNull(poDSClipper);
delete[]read_filenamePoly;
OGRLayer* poLayerClipper;
poLayerClipper = poDSClipper->GetLayerByName("Polygon");
int numClip = poLayerClipper->GetFeatureCount();
//setup warp options
GDALWarpOptions* psWarpOptions = GDALCreateWarpOptions();
psWarpOptions->hSrcDS = hSrcDS;
psWarpOptions->hDstDS = hDstDS;
psWarpOptions->nBandCount = 1;
psWarpOptions->panSrcBands = (int *) CPLMalloc(sizeof(int) * psWarpOptions->nBandCount);
psWarpOptions->panSrcBands[0] = 1;
psWarpOptions->panDstBands = (int*)CPLMalloc(sizeof(int) * psWarpOptions->nBandCount);
psWarpOptions->panDstBands[0] = 1;
psWarpOptions->pfnProgress = GDALTermProgress;
psWarpOptions->hCutline = poLayerClipper;
// Establish reprojection transformer.
psWarpOptions->pTransformerArg = GDALCreateGenImgProjTransformer(hSrcDS,proj, hDstDS, proj2, FALSE, 0.0, 1);
psWarpOptions->pfnTransformer = GDALGenImgProjTransform;
GDALWarpOperation oOperation;
oOperation.Initialize(psWarpOptions);
oOperation.ChunkAndWarpImage(0, 0, GDALGetRasterXSize(hDstDS), GDALGetRasterYSize(hDstDS));
GDALDestroyGenImgProjTransformer(psWarpOptions->pTransformerArg);
GDALDestroyWarpOptions(psWarpOptions);
GDALClose(hDstDS);
GDALClose(hSrcDS);
Your psWarpOptions->hCutline should be a polygon, not a layer.
Also the cutline should be in source pixel/line coordinates.
Check TransformCutlineToSource from gdalwarp_lib.cpp, you can probably simply get the code from there.
This particular GDAL operation, when called from C++, is so full of pitfalls - and there are so many open questions about it here - that I am reproducing a full working example:
Warping (reprojecting) a raster image with a polygon mask (cutline):
#include <gdal/gdal.h>
#include <gdal/gdal_priv.h>
#include <gdal/gdalwarper.h>
#include <gdal/ogrsf_frmts.h>
int main() {
const char *inputPath = "input.tif";
const char *outputPath = "output.tif";
// clipper Polygon
// THIS FILE MUST BE IN PIXEL/LINE COORDINATES or otherwise one should
// copy the function gdalwarp_lib.cpp:TransformCutlineToSource()
// from GDAL's sources
// It is expected that it contains a single polygon feature
const char *read_filenamePoly = "cutline.json";
GDALDataset *hSrcDS;
GDALDataset *hDstDS;
GDALAllRegister();
auto poDriver = GetGDALDriverManager()->GetDriverByName("GTiff");
hSrcDS = (GDALDataset *)GDALOpen(inputPath, GA_ReadOnly);
hDstDS = (GDALDataset *)poDriver->CreateCopy(
outputPath, hSrcDS, 0, nullptr, nullptr, nullptr);
// Without this step the cutline is useless - because the background
// will be carried over from the original image
CPLErr e = hDstDS->GetRasterBand(1)->Fill(0);
const char *src_srs = hSrcDS->GetProjectionRef();
const char *dst_srs = hDstDS->GetProjectionRef();
// clipper Layer
GDALDataset *poDSClipper;
poDSClipper = (GDALDataset *)GDALOpenEx(
read_filenamePoly, GDAL_OF_UPDATE, NULL, NULL, NULL);
auto poLayerClipper = poDSClipper->GetLayer(0);
auto geom = poLayerClipper->GetNextFeature()->GetGeometryRef();
// setup warp options
GDALWarpOptions *psWarpOptions = GDALCreateWarpOptions();
psWarpOptions->hSrcDS = hSrcDS;
psWarpOptions->hDstDS = hDstDS;
psWarpOptions->nBandCount = 1;
psWarpOptions->panSrcBands =
(int *)CPLMalloc(sizeof(int) * psWarpOptions->nBandCount);
psWarpOptions->panSrcBands[0] = 1;
psWarpOptions->panDstBands =
(int *)CPLMalloc(sizeof(int) * psWarpOptions->nBandCount);
psWarpOptions->panDstBands[0] = 1;
psWarpOptions->pfnProgress = GDALTermProgress;
psWarpOptions->hCutline = geom;
// Establish reprojection transformer.
psWarpOptions->pTransformerArg = GDALCreateGenImgProjTransformer(
hSrcDS, src_srs, hDstDS, dst_srs, TRUE, 1000, 1);
psWarpOptions->pfnTransformer = GDALGenImgProjTransform;
GDALWarpOperation oOperation;
oOperation.Initialize(psWarpOptions);
oOperation.ChunkAndWarpImage(
0, 0, GDALGetRasterXSize(hDstDS), GDALGetRasterYSize(hDstDS));
GDALDestroyGenImgProjTransformer(psWarpOptions->pTransformerArg);
GDALDestroyWarpOptions(psWarpOptions);
GDALClose(hDstDS);
GDALClose(hSrcDS);
}

How to make OpenVR multiview rendering to work?

I'm optimizing my already working VR rendering by implementing multiview rendering into my custom C++ engine, but can't get the other eye to render. Here's my vertex shader:
layout(set=0, binding=2) Buffer<float3> positions : register(b2);
VSOutput unlitVS( uint vertexId : SV_VertexID, uint viewId : SV_ViewID )
{
VSOutput vsOut;
vsOut.uv = uvs[ vertexId ];
vsOut.pos = mul( data.localToClip[ viewId ], float4( positions[ vertexId ], 1 ) );
return vsOut;
}
If I use viewId to select a transformation matrix, I'm getting an error:
ERROR: Vertex shader consumes input at location 0 but not provided
If I don't use viewId my other eye doesn't render at all. Here's how I setup my framebuffer, it has 2 layers and after rendering the left eye, I copy the second layer to the right eye.
VkImageCreateInfo imageCreateInfo = {};
imageCreateInfo.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
imageCreateInfo.pNext = nullptr;
imageCreateInfo.imageType = VK_IMAGE_TYPE_2D;
imageCreateInfo.extent.width = width;
imageCreateInfo.extent.height = height;
imageCreateInfo.extent.depth = 1;
imageCreateInfo.mipLevels = 1;
imageCreateInfo.arrayLayers = 2;
imageCreateInfo.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED;
imageCreateInfo.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
imageCreateInfo.format = VK_FORMAT_R8G8B8A8_SRGB;
imageCreateInfo.tiling = VK_IMAGE_TILING_OPTIMAL;
imageCreateInfo.samples = sampleCount == 4 ? VK_SAMPLE_COUNT_4_BIT : VK_SAMPLE_COUNT_1_BIT;
imageCreateInfo.usage = (VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT | VK_IMAGE_USAGE_SAMPLED_BIT | VK_IMAGE_USAGE_TRANSFER_SRC_BIT);
imageCreateInfo.flags = 0;
VkResult result = vkCreateImage( dev, &imageCreateInfo, nullptr, &outFramebufferDesc.image );
...
// Bit mask that specifies which view rendering is broadcast to.
// 0011 = Broadcast to first and second view (layer)
const uint32_t viewMask = 0b00000011;
// Bit mask that specifices correlation between views
// An implementation may use this for optimizations (concurrent render)
const uint32_t correlationMask = 0b00000011;
VkRenderPassMultiviewCreateInfo renderPassMultiviewCI{};
renderPassMultiviewCI.sType = VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO;
renderPassMultiviewCI.subpassCount = 1;
renderPassMultiviewCI.pViewMasks = &viewMask;
renderPassMultiviewCI.correlationMaskCount = 1;
renderPassMultiviewCI.pCorrelationMasks = &correlationMask;
VkRenderPassCreateInfo renderPassCreateInfo = {};
renderPassCreateInfo.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
renderPassCreateInfo.flags = 0;
renderPassCreateInfo.pNext = &renderPassMultiviewCI;
renderPassCreateInfo.attachmentCount = 2;
renderPassCreateInfo.pAttachments = &attachmentDescs[ 0 ];
renderPassCreateInfo.subpassCount = 1;
renderPassCreateInfo.pSubpasses = &subPassCreateInfo;
renderPassCreateInfo.dependencyCount = 0;
renderPassCreateInfo.pDependencies = nullptr;
result = vkCreateRenderPass( dev, &renderPassCreateInfo, nullptr, &outFramebufferDesc.renderPass );
assert( result == VK_SUCCESS );
VkImageView attachments[ 2 ] = { outFramebufferDesc.imageView, outFramebufferDesc.depthStencilImageView };
VkFramebufferCreateInfo framebufferCreateInfo = {};
framebufferCreateInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
framebufferCreateInfo.pNext = nullptr;
framebufferCreateInfo.renderPass = outFramebufferDesc.renderPass;
framebufferCreateInfo.attachmentCount = 2;
framebufferCreateInfo.pAttachments = &attachments[ 0 ];
framebufferCreateInfo.width = width;
framebufferCreateInfo.height = height;
framebufferCreateInfo.layers = 1;
result = vkCreateFramebuffer( dev, &framebufferCreateInfo, nullptr, &outFramebufferDesc.framebuffer );
...
// After rendering the left eye:
VkImageCopy region = {};
region.extent.width = device.width;
region.extent.height = device.height;
region.extent.depth = 1;
region.srcSubresource.baseArrayLayer = 1;
region.srcSubresource.layerCount = 1;
region.srcSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
region.dstSubresource.layerCount = 1;
region.dstSubresource.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
vkCmdCopyImage( gCurrentDrawCommandBuffer, device.fbDesc.image, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, device.fbDesc.imageCopy, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, 1, &region );
Here's my VkPipelineVertexInputStateCreateInfo, I'm using programmable vertex pulling:
VkPipelineVertexInputStateCreateInfo inputState = {};
inputState.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO;
How can I fix the error and get right eye working? I'm using HTC Vive, OpenVR 1.7.15, AMD R9 Nano, Vulkan 1.1 on Vulkan SDK 1.1.126.0, Windows 10. I have enabled device extension VK_KHR_MULTIVIEW_EXTENSION_NAME.

Swapchain and Render Pass created successfully, but validation layers give an error when using them

I am trying to render a triangle and I'm following a tutorial from https://vulkan-tutorial.org
I am using GLFW, GLSL and Visual Studio 2017.
When render pass is created, it's value is always 0xc, and when swapchain is created, it's value is always 0x2. Creation functions always return VK_SUCCESS, and there is no validation layer output.
When I try to create a frame buffer, using vkCreateFramebuffer() I get prompted that my render pass object handle is invalid, and that my ImageView object handles are invalid. Their values are always 0x6, 0x7, and 0x8.
Also, when I try to invoke vkAcquireNextImageKHR(), I get prompted that my swapchain object handle is invalid.
My swapchain creation code is below. Please, ignore the comments, they are for learning purposes only.
void Swapchain::initSwapchain() {
VkPresentModeKHR presentMode = getAvaiablePresentMode();
QueueFamilyIndices indices = *mainWindow->getRenderer()->getQueueIndices();
swapchainCreateInfo.sType = VK_STRUCTURE_TYPE_SWAPCHAIN_CREATE_INFO_KHR;
swapchainCreateInfo.surface = this->mainWindow->getSurface();
swapchainCreateInfo.minImageCount = swapchainImageCount; //Bufferovanje slika display buffera, koliko slika odjednom moze biti u redu
swapchainCreateInfo.imageFormat = this->mainWindow->getSurfaceFormat().format;
swapchainCreateInfo.imageColorSpace = this->mainWindow->getSurfaceFormat().colorSpace;
swapchainCreateInfo.imageExtent = this->swapExtent;
swapchainCreateInfo.imageArrayLayers = 1; //Koliko slojeva ima slika (1 je obicno renderovanje, 2 je stetoskopsko)
swapchainCreateInfo.imageUsage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT; //Za koju vrstu operacija koristimo slike? Renderujemo ih, sto znaci da su oni COLOR ATTACHMENTS
if (indices.getGraphicsFamilyIndex() != indices.getPresentationFamilyIndex()) {
uint32_t queueIndices[] = { indices.getGraphicsFamilyIndex(), indices.getPresentationFamilyIndex() };
swapchainCreateInfo.imageSharingMode = VK_SHARING_MODE_CONCURRENT; //Slika moze da se koristi paralelno, bez transfera vlasnistva nad slikom.
swapchainCreateInfo.queueFamilyIndexCount = 2;
swapchainCreateInfo.pQueueFamilyIndices = queueIndices;
}
else {
swapchainCreateInfo.imageSharingMode = VK_SHARING_MODE_EXCLUSIVE; //Slika je u vlasnistvu jednog reda u jedno vreme, i vlasnistvo mora biti prebaceno na drugi da bi taj drugi mogao da ga koristi.
swapchainCreateInfo.queueFamilyIndexCount = 0; //Za exclusive je uvek 0
swapchainCreateInfo.pQueueFamilyIndices = nullptr; //Ignorisemo za Exclusive
}
swapchainCreateInfo.preTransform = mainWindow->getSurfaceCapatibilities().currentTransform; //mainWindow->getCapabilities().currentTransform ako necemo transformaciju. VK_SURFACE_TRANSFORM_IDENTITY_BIT_KHR;
swapchainCreateInfo.compositeAlpha = VK_COMPOSITE_ALPHA_OPAQUE_BIT_KHR; //Alfa kanal SURFACE-a, da li je ona transparentna
swapchainCreateInfo.presentMode = presentMode; //Vertical Sync
swapchainCreateInfo.clipped = VK_TRUE; //Ukljucujemo clipping, jako bitno za telefone
swapchainCreateInfo.oldSwapchain = VK_NULL_HANDLE; //Ako rekonstruisemo swapchain, pokazivac na stari
util->ErrorCheck(vkCreateSwapchainKHR(renderer->getDevice(), &swapchainCreateInfo, nullptr, &swapchain));
util->ErrorCheck(vkGetSwapchainImagesKHR(renderer->getDevice(), swapchain, &swapchainImageCount, nullptr));
}
void Swapchain::initSwapchainImgs()
{
images.resize(swapchainImageCount);
imageViews.resize(swapchainImageCount);
util->ErrorCheck(vkGetSwapchainImagesKHR(renderer->getDevice(), swapchain, &swapchainImageCount, images.data()));
for (uint32_t i = 0; i < swapchainImageCount; i++) {
VkImageViewCreateInfo imgCreateInfo = {};
imgCreateInfo.components.r = VK_COMPONENT_SWIZZLE_R;
imgCreateInfo.components.g = VK_COMPONENT_SWIZZLE_G;
imgCreateInfo.components.b = VK_COMPONENT_SWIZZLE_B;
imgCreateInfo.components.a = VK_COMPONENT_SWIZZLE_A;
imgCreateInfo.subresourceRange.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT;
imgCreateInfo.subresourceRange.baseMipLevel = 0;
imgCreateInfo.subresourceRange.levelCount = 1;
imgCreateInfo.subresourceRange.baseArrayLayer = 0;
imgCreateInfo.subresourceRange.layerCount = 1;
imgCreateInfo.format = this->imagesFormat;
imgCreateInfo.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO;
imgCreateInfo.image = images[i];
imgCreateInfo.viewType = VK_IMAGE_VIEW_TYPE_2D;
util->ErrorCheck(vkCreateImageView(renderer->getDevice(), &imgCreateInfo, nullptr, &imageViews[i]));
}
}
Render pass creation code:
void RenderPass::createColor() {
VkAttachmentDescription attachment = {};
VkAttachmentReference reference = {};
VkSubpassDescription subpass = {};
VkRenderPassCreateInfo info = {};
attachment.format = surfaceFormat.format; //Mora da se poklapa sa formatom slika iz swapchaina
attachment.samples = VK_SAMPLE_COUNT_1_BIT; //Odnosi se na multisampling
attachment.loadOp = VK_ATTACHMENT_LOAD_OP_CLEAR; //Operacija koju render pass attachment treba da obavi pri ucitavanju
attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE; //Operacija koju treba odraditi posle rendera
attachment.stencilLoadOp = VK_ATTACHMENT_LOAD_OP_DONT_CARE;
attachment.stencilStoreOp = VK_ATTACHMENT_STORE_OP_DONT_CARE;
attachment.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED; //Nije nam bitno kog je formata bila prosla slika, to ovo znaci.
attachment.finalLayout = VK_IMAGE_LAYOUT_PRESENT_SRC_KHR; //Slike koje treba da budu predstavljene u swapchainu
reference.layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL;//Layout slike u ovom subpassu
reference.attachment = 0; //Index attachmenta koji referenciramo ovim subpassom
subpass.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS;
subpass.colorAttachmentCount = 1;
subpass.pColorAttachments = &reference; //REFERENCIRAN JE IZ FRAGMENT SHADERA
info.attachmentCount = 1;
info.dependencyCount = 0;
info.pAttachments = &attachment;
info.pDependencies = nullptr;
info.subpassCount = 1;
info.pSubpasses = &subpass;
info.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO;
util->ErrorCheck(vkCreateRenderPass(this->renderer->getDevice(), &info, nullptr, &this->renderPass));
}
And finally, Framebuffer creation code:
void FrameBuffer::initFrameBuffer(
uint32_t swapchainImageCount,
std::vector<VkImageView> imageViews,
VkRenderPass renderPass,
VkExtent2D surfaceSize,
std::vector<VkImageView> attachments
)
{
frameBuffers.resize(swapchainImageCount);
for (uint32_t i = 0; i < swapchainImageCount; ++i) {
VkFramebufferCreateInfo frameBufferCreateInfo{};
std::vector<VkImageView> allAttachments = { imageViews[i] };
allAttachments.insert(allAttachments.begin(), attachments.begin(), attachments.end());
frameBufferCreateInfo.sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO;
frameBufferCreateInfo.renderPass = renderPass;
frameBufferCreateInfo.width = surfaceSize.width;
frameBufferCreateInfo.height = surfaceSize.height;
frameBufferCreateInfo.layers = 1;
frameBufferCreateInfo.pAttachments = allAttachments.data();
frameBufferCreateInfo.attachmentCount = allAttachments.size();
util->ErrorCheck(vkCreateFramebuffer(renderer->getDevice(), &frameBufferCreateInfo, nullptr, &frameBuffers[i]));
}
}
After running framebuffer creation code, I get these outputs:
Render Pass error on framebuffer creation:
ImageView error on framebuffer creation:
This error I get when I invoke vkAcquireNextImageKHR():
This can't be a problem with a device or instance because I have enabled VK_surface_khr and VK_KHR_swapchain, and I have checked if my device and window support these extensions. Also, I have rebooted my PC and this errors still occur. I am using Vulkan SDK 1.1.85.0.
Also, there are no missing values in creation info structures except for pNext, which I think shouldn't even be used for now.

RGB to x264 : Strange color render

i'm trying to make a video from an OpenGl context.
I'm Using glReadPixel, to be sure RGB buffer data is Ok i save it into a bmp file, wich i can read correctly.
My .h264 video is encoded but there are some artefact and i don't understand why.
I tried a lot of different parameters for the x264_param_t but anything better !
Bitmap saved (OpenGL real data) : Bitmap from OpenGl (1mo)
Raw h264 with error : Raw h264 video (1mo)
OpenGl ReadPixel :
int nSize = ClientHeight * ClientWidth * 3;
GLubyte *inBuff = new GLubyte[nSize];
glReadBuffer(GL_FRONT);
glReadPixels(0, 0, ldwidth, ldheight, GL_BGR, GL_UNSIGNED_BYTE, inBuff);
The params define :
x264_param_default(&mX264_param_t);
x264_param_default_preset(&mX264_param_t, "placebo", "film");
mX264_param_t.i_csp = X264_CSP_BGR;
mX264_param_t.i_threads = 6;
mX264_param_t.i_width = mWidth;
mX264_param_t.i_height = mHeight;
mX264_param_t.i_fps_num = mFps;
mX264_param_t.i_fps_den = 1;
// Intra refres:
mX264_param_t.i_keyint_max = mFps;
mX264_param_t.b_intra_refresh = 1;
//Rate control:
mX264_param_t.rc.i_rc_method = X264_RC_CRF;
mX264_param_t.rc.f_rf_constant = 25;
mX264_param_t.rc.f_rf_constant_max = 35;
int bps = 5000;
mX264_param_t.rc.i_bitrate = bps;
mX264_param_t.rc.i_vbv_max_bitrate = bps;
mX264_param_t.i_bframe = 2;
mX264_param_t.i_keyint_min = mFps / 4;
//For streaming:
mX264_param_t.b_repeat_headers = 1;
mX264_param_t.b_annexb = 1;
mX264_param_t.i_log_level = X264_LOG_DEBUG;
x264_param_apply_profile(&mX264_param_t, "baseline");
mpEncoder = x264_encoder_open(&mX264_param_t);
x264_encoder_parameters(mpEncoder, &mX264_param_t);
mpPictureOut = new x264_picture_t();
mpPictureIn = new x264_picture_t();
x264_picture_alloc(mpPictureIn, X264_CSP_BGR | X264_CSP_VFLIP, mWidth, mHeight);
Then the encoding loop :
mpPictureIn->img.i_csp = X264_CSP_BGR;
mpPictureIn->img.i_plane = 1;
mpPictureIn->img.i_stride[0] = 3 * mWidth;
mpPictureIn->img.plane[0] = rgbframe;
mpPictureIn->i_pts = mFrameCount;
mpPictureIn->i_type = X264_TYPE_AUTO;
mpPictureOut->i_pts = mFrameCount;
int i_nals;
x264_nal_t* nals;
int frame_size = x264_encoder_encode(mpEncoder, &nals, &i_nals, mpPictureIn, mpPictureOut);
if(frame_size > 0)
{
mpFileOut->write_frame(nals[0].p_payload, frame_size, mpPictureOut);
mFrameCount++;
}
The write frame :
int TVideoFileWriter::write_frame(uint8_t *p_nalu, int i_size, x264_picture_t *p_picture)
{
if(fwrite(p_nalu, i_size, 1, mFileHandle))
return i_size;
return -1;
}
You opened your output file in text mode (and not binary mode) and so all 0x0A bytes where replaced with 0x0D 0x0A bytes.
Here is your output with this replace reverted: out_fixed.h264
And it plays fine.