/*
===========================================================================
Copyright (C) 2023-2024 Gian 'myT' Schellenbaum

This file is part of Challenge Quake 3 (CNQ3).

Challenge Quake 3 is free software; you can redistribute it
and/or modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version 2 of the License,
or (at your option) any later version.

Challenge Quake 3 is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
GNU General Public License for more details.

You should have received a copy of the GNU General Public License
along with Challenge Quake 3. If not, see <https://www.gnu.org/licenses/>.
===========================================================================
*/
// add fragments of transparent surfaces to per-pixel linked lists


#include "common.hlsli"
#include "alpha_test.h.hlsli"
#include "oit.h.hlsli"
#include "scene_view.h.hlsli"


cbuffer RootConstants
{
	// geometry
	matrix modelViewMatrix;

	// general
	uint textureIndex;
	uint samplerIndex;
	uint alphaTest;
	uint counterBuffer;
	uint indexTexture;
	uint fragmentBuffer;
	float greyscale;
	uint stateBits;
	uint shaderTrace;
	uint depthFadeDistOffset; // offset: fp16 - distance: fp16
	uint depthFadeScaleBiasPO; // polygon offset: 1 - enable: 1 - color bias: 4 - color scale: 4
};

struct VIn
{
	float3 position : POSITION;
	float3 normal : NORMAL;
	float2 texCoords : TEXCOORD0;
	float4 color : COLOR0;
};

struct VOut
{
	float4 position : SV_Position;
	float3 normal : NORMAL;
	float2 texCoords : TEXCOORD0;
	float4 color : COLOR0;
	float clipDist : SV_ClipDistance0;
	float2 proj2232 : PROJ;
	float depthVS : DEPTHVS;
};

VOut vs(VIn input)
{
	SceneView scene = GetSceneView();
	float4 positionVS = mul(modelViewMatrix, float4(input.position.xyz, 1));
	matrix projectionMatrix = scene.projectionMatrix;

	VOut output;
	output.position = mul(projectionMatrix, positionVS);
	output.normal = input.normal;
	output.texCoords = input.texCoords;
	output.color = input.color;
	output.clipDist = dot(positionVS, scene.clipPlane);
	output.proj2232 = float2(-projectionMatrix[2][2], projectionMatrix[2][3]);
	output.depthVS = -positionVS.z;

	return output;
}

bool IsFragmentUseless(uint blendBits, float4 color)
{
	const float epsilon = 1.0 / 1024.0;

	if(blendBits == GLS_BLEND_ADDITIVE &&
		all(color.rgb < epsilon.xxx))
	{
		return true;
	}
	if((blendBits == GLS_BLEND_STD_ALPHA || blendBits == GLS_BLEND_PMUL_ALPHA) &&
		color.a < epsilon)
	{
		return true;
	}
	if((blendBits == GLS_BLEND_FILTER || blendBits == GLS_BLEND_FILTER_V2) &&
		all(color.rgb > (1.0 - epsilon).xxx) && all(color.rgb < (1.0 + epsilon).xxx))
	{
		return true;
	}

	return false;
}

[earlydepthstencil]
void ps(VOut input)
{
	Texture2D texture0 = ResourceDescriptorHeap[textureIndex];
	SamplerState sampler0 = SamplerDescriptorHeap[samplerIndex];
	float4 dst = texture0.Sample(sampler0, input.texCoords) * input.color;
	if(FailsAlphaTest(dst.a, alphaTest))
	{
		return;
	}

	dst = MakeGreyscale(dst, greyscale);
	uint blendBits = stateBits & GLS_BLEND_BITS;
	if(IsFragmentUseless(blendBits, dst))
	{
		return;
	}

	RWStructuredBuffer<OIT_Counter> counter = ResourceDescriptorHeap[counterBuffer];
	uint fragmentIndex;
	InterlockedAdd(counter[0].fragmentCount, 1, fragmentIndex);
	if(fragmentIndex < counter[0].maxFragmentCount)
	{
		RWTexture2D<uint> indexTex = ResourceDescriptorHeap[indexTexture];
		RWStructuredBuffer<OIT_Fragment> fragments = ResourceDescriptorHeap[fragmentBuffer];
		uint prevFragmentIndex;
		InterlockedExchange(indexTex[int2(input.position.xy)], fragmentIndex, prevFragmentIndex);
		OIT_Fragment fragment;
		fragment.color = PackColor(dst);
		fragment.depth = input.depthVS;
		fragment.stateBits = stateBits;
		fragment.next = prevFragmentIndex;
		fragment.shaderTrace = shaderTrace;
		fragment.depthFadeDistOffset = depthFadeDistOffset;
		fragment.depthFadeScaleBiasPO = depthFadeScaleBiasPO;
		fragments[fragmentIndex] = fragment;
	}
	else
	{
		uint garbage;
		InterlockedAdd(counter[0].overflowCount, 1, garbage);
		InterlockedAdd(counter[0].fragmentCount, -1, garbage);
	}
}