// Copyright (C) 2009-present, Panagiotis Christopoulos Charitos and contributors.
// All rights reserved.
// Code licensed under the BSD License.
// http://www.anki3d.org/LICENSE

#pragma anki mutator REDUCTION_TYPE 0 1 // 0: min 1: max
#pragma anki mutator MIN_MAX_SAMPLER 0 1

#pragma anki technique comp

#include <AnKi/Shaders/Common.hlsl>

struct Constants
{
	Vec2 m_invSrcTexSize;
	U32 m_threadGroupCount;
	U32 m_mipmapCount;
};
ANKI_FAST_CONSTANTS(Constants, g_consts)

globallycoherent RWStructuredBuffer<U32> g_spdCounter : register(u0);
globallycoherent RWTexture2D<Vec4> g_dstStorageTextures[kMaxMipsSinglePassDownsamplerCanProduce] : register(u1);
Texture2D<Vec4> g_srcTex : register(t0);

#if MIN_MAX_SAMPLER
SamplerState g_minMaxAnyClampSampler : register(s0);
#else
SamplerState g_linearAnyClampSampler : register(s0);
#endif

// Include SPD
#define A_GPU 1
#define A_HLSL 1
#include <ThirdParty/FidelityFX/ffx_a.h>

groupshared AU1 s_spdCounter;
groupshared AF1 s_spdIntermediateR[16][16];

AF4 SpdLoadSourceImage(AU2 p, AU1 slice)
{
	ANKI_MAYBE_UNUSED(slice);
	const AF2 uv = p * g_consts.m_invSrcTexSize + g_consts.m_invSrcTexSize;

#if MIN_MAX_SAMPLER
	const F32 f = g_srcTex.SampleLevel(g_minMaxAnyClampSampler, uv, 0.0).r;
#else
	Vec4 samples = g_srcTex.GatherRed(g_linearAnyClampSampler, uv);

#	if REDUCTION_TYPE == 0
	const F32 f = min4(samples);
#	else
	const F32 f = max4(samples);
#	endif
#endif

	return AF4(f, 0.0, 0.0, 0.0);
}

AF4 SpdLoad(AU2 p, AU1 slice)
{
	ANKI_MAYBE_UNUSED(slice);
	return AF4(g_dstStorageTextures[5][p].r, 0.0, 0.0, 0.0);
}

void SpdStore(AU2 p, AF4 value, AU1 mip, AU1 slice)
{
	ANKI_MAYBE_UNUSED(slice);
	g_dstStorageTextures[mip][p] = Vec4(value.x, 0.0, 0.0, 0.0);
}

void SpdIncreaseAtomicCounter(AU1 slice)
{
	ANKI_MAYBE_UNUSED(slice);
	InterlockedAdd(g_spdCounter[0], 1u, s_spdCounter);
}

AU1 SpdGetAtomicCounter()
{
	return s_spdCounter;
}

void SpdResetAtomicCounter(AU1 slice)
{
	ANKI_MAYBE_UNUSED(slice);
	g_spdCounter[0] = 0u;
}

AF4 SpdLoadIntermediate(AU1 x, AU1 y)
{
	return AF4(s_spdIntermediateR[x][y], 0.0, 0.0, 0.0);
}

void SpdStoreIntermediate(AU1 x, AU1 y, AF4 value)
{
	s_spdIntermediateR[x][y] = value.x;
}

AF4 SpdReduce4(AF4 v0, AF4 v1, AF4 v2, AF4 v3)
{
#if REDUCTION_TYPE == 0
	const F32 value = min4(v0.x, v1.x, v2.x, v3.x);
#else
	const F32 value = max4(v0.x, v1.x, v2.x, v3.x);
#endif
	return AF4(value, 0.0, 0.0, 0.0);
}

#define SPD_LINEAR_SAMPLER 1
#include <ThirdParty/FidelityFX/ffx_spd.h>

[numthreads(256, 1, 1)] void main(UVec3 svGroupId : SV_GROUPID, U32 svGroupIndex : SV_GROUPINDEX)
{
	const U32 slice = 0u;
	const UVec2 offset = UVec2(0, 0);
	SpdDownsample(AU2(svGroupId.xy), AU1(svGroupIndex), AU1(g_consts.m_mipmapCount), AU1(g_consts.m_threadGroupCount), slice, offset);
}