gpuopen-librariesandsdks/cauldron

A simple framework for rapid prototyping on Vulkan or Direct3D 12

cauldron's Issues

Linux support

Linux support would be nice.

Tonemappers 'abs' are useless

https://github.com/GPUOpen-LibrariesAndSDKs/Cauldron/blob/master/src/DX12/shaders/Tonemappers.hlsl#L75
https://github.com/GPUOpen-LibrariesAndSDKs/Cauldron/blob/master/src/DX12/shaders/Tonemappers.hlsl#L77

There are 2 abs, which I believe are useless since ratio should always be in 0..1 as long as colors are >=0

SwapChain::WaitForSwapChain wrong "CPU wait for fence" setup

I catch next one, here:

Cauldron/src/DX12/base/SwapChain.cpp

Line 163 in b3a4f62

m_swapChainFence.CpuWaitForFence(m_BackBufferCount - 1);

For now, this fires only once at the very first round.
It should wait for previous backBuffferIndex - it works now only due to engine happy-scenario ...

Perhaps: m_swapChainFence.CpuWaitForFence(backBuffferIndex == 0 ? m_BackBufferCount : backBuffferIndex - 1);

Fix copyright issue link

Change https://github.com/KhronosGroup/glTF-WebGL-PBR to https://github.com/KhronosGroup/glTF-Sample-Viewer/

friendly headsup note

the generated sln's name is modified from Framework_{GFX_API} to Cauldron_{GFX_API} in May 5, 2020

GLTFCommon::GetInverseBindMatricesBufferSizeByID should use sizeof(Matrix2)

(4 * 4 * sizeof(float)) less than (sizeof(Matrix2))

CommandListRing::GetPool always return frame #0's pool.

On Vk impl, class CommandListRing::GetPool() just return m_pCommandBuffers, i guess it should return m_pCurrentFrame's pool.

[Question]: D3D12_RESOURCE_STATE_INDEX_BUFFER

Why does the static buffer pool always transition to a D3D12_RESOURCE_STATE_VERTEX_AND_CONSTANT_BUFFER state even when used for indices? Shouldn't D3D12_RESOURCE_STATE_INDEX_BUFFER be used for that purpose as already mentioned in the comments?

GenerateSolutions.bat creating solution folders,but doesn't create visual studio solution files.

Changing a shader include files does not trigger a recompilation

Since the shader cache hash is only based on the input source file and its defines and arguments, editing the contents of an include file will never trigger a re-compilation.

Please support building on MSYS2 MINGW64

MSYS2 MINGW64 doesn't have DX12 but it has working Vulkan. The current CMakeLists.txt assumed MSVC and refused to configure on MSYS2 MINGW64.

Depth test is incorrectly inverted in `GltfDepthPass` for both platforms

Which leads to shadow maps missing in glTFSample - GPUOpen-LibrariesAndSDKs/glTFSample#6

RingWithTabs::OnBeginFrame wrong m_backBufferIndex calculations

@rys next one:

Cauldron/src/common/Misc/Ring.h

Line 137 in b3a4f62

m_backBufferIndex = (m_backBufferIndex + 1) % m_numberOfBackBuffers;

This one should release (through m_mem.Free(...)) stuff from previous back_buffer, not the upcoming one.
Something like: m_backBufferIndex = m_backBufferIndex == 0 ? m_numberOfBackBuffers : m_backBufferIndex - 1;

Block compressed dds

Is loading of BC formats supported? I have tried with a few dds files and it always shows garbage data in both VK and D3D12
Attached a BC3 version of Default_albedo.png from the Damaged Helmet GLTF in Cauldron Media repo and screenshot from the app.

Default_albedo.zip

Thank you

How to tell actual RAM usage?

Given that Cauldron allocates large swaths of memory to use as a buffer, how can one tell the how much actual memory is being used?

Can the pre-allocation system be disabled? Or is there any chance that the system metrics window will contain memory usage information too in the future?

Thank you

DivideRoundingUp compile error

template<typename T> inline T DivideRoundingUp(T a, T b)
{
    return (x + y - (T)1) / y;
}

You probably wanted to use x and y so the definition should be
template<typename T> inline T DivideRoundingUp(T x, T y)

Trace does not actually lock

While debugging the SSSR sample I stumbled upon a lock allocated on stack:
https://github.com/GPUOpen-LibrariesAndSDKs/Cauldron/blob/master/src/common/Misc/Misc.cpp#L83

Camera position is not correctly updated in Camera::UpdateCameraPolar()

Hello.

I found that I was not able to update camera position when using Camera::UpdateCameraPolar().
I think eyePos should be used instead of m_eyePos in this line, is this correct?
https://github.com/GPUOpen-LibrariesAndSDKs/Cauldron/blob/master/src/common/Misc/Camera.cpp#L122

UploadHeap::Suballocate assert(uSize...) will not fire at heap over size allocation...

I catch this one, here:

Cauldron/src/DX12/base/UploadHeap.cpp

Line 94 in b3a4f62

assert(uSize < (size_t)(m_pDataBegin - m_pDataEnd));

It should be reversed, i.e.: assert(uSize < (size_t)(m_pDataEnd - m_pDataBegin));
Perhaps, additional - similar assert for AlignUp() case also...

Camera::UpdatePreviousMatrices should always be called

Right now it is not called when loading or in benchmark mode, resulting in corrupted motion vectors.

MacOS support

MacOS support would be nice.