uniform float4x4 ViewProj;
uniform texture2d image;
uniform int level;
uniform float2 imageTexel;
uniform float strength;

sampler_state pointSampler {
	Filter    = Point;
	AddressU  = Clamp;
	AddressV  = Clamp;
};

sampler_state linearSampler {
	Filter    = Linear;
	AddressU  = Clamp;
	AddressV  = Clamp;
};

struct VertDataIn {
	float4 pos : POSITION;
	float2 uv  : TEXCOORD0;
};

struct VertDataOut {
	float4 pos : POSITION;
	float2 uv  : TEXCOORD0;
};

VertDataOut VSDefault(VertDataIn v_in)
{
	VertDataOut vert_out;
	vert_out.pos = mul(float4(v_in.pos.xyz, 1.0), ViewProj);
	vert_out.uv  = v_in.uv;
	return vert_out;
}

float4 PSPoint(VertDataOut v_in) : TARGET
{
	return image.SampleLevel(pointSampler, v_in.uv, level);
}

float4 PSLinear(VertDataOut v_in) : TARGET
{
	return image.SampleLevel(linearSampler, v_in.uv, level);
}

float4 PSSharpen(VertDataOut v_in) : TARGET
{
	float2 ul, ur, dl, dr, u, d, l, r;
	ul = float2(-imageTexel.x, -imageTexel.y);
	ur = float2(imageTexel.x, -imageTexel.y);
	dl = -ur;
	dr = -ul;
	u = float2(0, -imageTexel.y);
	d = -u;
	l = float2(-imageTexel.x, 0);
	r = -l;
	
	float4 tl, tc, tr, cl, cc, cr, bl, bc, br;
	tl = image.SampleLevel(pointSampler, v_in.uv + ul, level);
	tc = image.SampleLevel(pointSampler, v_in.uv + u, level);
	tr = image.SampleLevel(pointSampler, v_in.uv + ur, level);
	cl = image.SampleLevel(pointSampler, v_in.uv + l, level);
	cc = image.SampleLevel(pointSampler, v_in.uv, level);
	cr = image.SampleLevel(pointSampler, v_in.uv + r, level);
	bl = image.SampleLevel(pointSampler, v_in.uv + dl, level);
	bc = image.SampleLevel(pointSampler, v_in.uv + d, level);
	br = image.SampleLevel(pointSampler, v_in.uv + dr, level);

	float kernel1, kernel2, kernel3;
	kernel1 = -0.25 * strength;
	kernel2 = -0.50 * strength;
	kernel3 = abs(kernel1 * 4) + abs(kernel2 * 4);

	return (tl * kernel1) + (tr * kernel1) + (bl * kernel1) + (br * kernel1) + (cl * kernel2) + (cr * kernel2) + (tc * kernel2) + (bc * kernel2) + (cc * kernel3);
}

float4 PSSmoothen(VertDataOut v_in) : TARGET
{
	// If we use linear sampling, we can get away with just 4 total sampler queries.
	// However this is not a cheap implementation, it's just meant to be accurate so we do each sampler query and rely on the compiler.

	float3 smoothKernel3 = float3(0.0574428, 0.0947072, 0.3914000);
	float2 ul, ur, dl, dr, u, d, l, r;
	float4 tl, tc, tr, cl, cc, cr, bl, bc, br;
	float limitstr = clamp(strength, 0.0, 1.0);

	ul = float2(-imageTexel.x, -imageTexel.y);
	ur = float2(imageTexel.x, -imageTexel.y);
	dl = -ur;
	dr = -ul;
	u = float2(0, -imageTexel.y);
	d = -u;
	l = float2(-imageTexel.x, 0);
	r = -l;
	
	tl = image.SampleLevel(pointSampler, v_in.uv + ul, level) * smoothKernel3[0];
	tc = image.SampleLevel(pointSampler, v_in.uv + u, level) * smoothKernel3[1];
	tr = image.SampleLevel(pointSampler, v_in.uv + ur, level) * smoothKernel3[0];
	cl = image.SampleLevel(pointSampler, v_in.uv + l, level) * smoothKernel3[1];
	cc = image.SampleLevel(pointSampler, v_in.uv, level) * smoothKernel3[2];
	cr = image.SampleLevel(pointSampler, v_in.uv + r, level) * smoothKernel3[1];
	bl = image.SampleLevel(pointSampler, v_in.uv + dl, level) * smoothKernel3[0];
	bc = image.SampleLevel(pointSampler, v_in.uv + d, level) * smoothKernel3[1];
	br = image.SampleLevel(pointSampler, v_in.uv + dr, level) * smoothKernel3[0];

	return tl + tc + tr + cl + cc + cr + bl + bc + br;
}

float4 PSBicubic(VertDataOut v_in) : TARGET
{
	return float4(1.0, 0.0, 1.0, 1.0);
}

float4 PSLanczos(VertDataOut v_in) : TARGET
{
	return float4(1.0, 0.0, 1.0, 1.0);
}

technique Point
{
	pass
	{
		vertex_shader = VSDefault(v_in);
		pixel_shader  = PSPoint(v_in);
	}
}

technique Linear
{
	pass
	{
		vertex_shader = VSDefault(v_in);
		pixel_shader  = PSLinear(v_in);
	}
}

technique Sharpen
{
	pass
	{
		vertex_shader = VSDefault(v_in);
		pixel_shader  = PSSharpen(v_in);
	}
}

technique Smoothen
{
	pass
	{
		vertex_shader = VSDefault(v_in);
		pixel_shader  = PSSmoothen(v_in);
	}
}

technique Bicubic
{
	pass
	{
		vertex_shader = VSDefault(v_in);
		pixel_shader  = PSBicubic(v_in);
	}
}

technique Lanczos
{
	pass
	{
		vertex_shader = VSDefault(v_in);
		pixel_shader  = PSLanczos(v_in);
	}
}