Hi Gamedev! I've coded little shader code for my small game to handle lights on my moving rectangles. It work and I'm quite happy with the visual result. Unfortunately it's badly implemented and the number of lights I can have on screen at same time is not enough. Changing pixelshader to use ps_3_0 instead of ps_2_0 made it possible to use 18 lights instead of 3. I know this is a bad solution and what I've to do is lower the number of instructions that is made on the pixel shader.

I would love any help I could get to optimize my code, by moving some parts to the vertex shader or change instructions on pixel shader.

//-----------------------------------------------------------------------------
// Texture sampler
//-----------------------------------------------------------------------------
uniform const texture BasicTexture;

uniform const sampler TextureSampler : register(s0) = sampler_state
{
	Texture = (BasicTexture);
	MipFilter = Linear;
	MinFilter = Linear;
	MagFilter = Linear;
};

//-----------------------------------------------------------------------------
// Material settings
//-----------------------------------------------------------------------------

uniform const float3	DiffuseColor	: register(c0) = 1;
uniform const float	Alpha		: register(c1) = 1;

//-----------------------------------------------------------------------------
// Matrices
//-----------------------------------------------------------------------------

uniform const float4x4	World		: register(vs, c2);		
uniform const float4x4	View		: register(vs, c6);		
uniform const float4x4	Projection	: register(vs, c10);

//-----------------------------------------------------------------------------
// Structure definitions
//-----------------------------------------------------------------------------

struct CommonVSOutput
{
	float4	Pos_ws;
	float4	Pos_ps;
	float4	Diffuse;
};

//-----------------------------------------------------------------------------
// Vertex shader inputs
//-----------------------------------------------------------------------------
struct VSInputTx
{
	float4	Position	: POSITION;
	float2	TexCoord	: TEXCOORD0;
};

//-----------------------------------------------------------------------------
// Vertex shader outputs
//-----------------------------------------------------------------------------
struct VertexLightingVSOutputTx
{
	float4	PositionPS	: POSITION;		// Position in projection space
	float4	Diffuse		: COLOR0;
	float2	TexCoord	: TEXCOORD0;
};

//-----------------------------------------------------------------------------
// Pixel shader inputs
//-----------------------------------------------------------------------------

struct VertexLightingPSInputTx
{
	float4	Diffuse		: COLOR0;
	float2	TexCoord	: TEXCOORD0;
};

CommonVSOutput ComputeCommonVSOutput(float4 position)
{
	CommonVSOutput vout;
	
	float4 pos_ws = mul(position, World);
	float4 pos_vs = mul(pos_ws, View);
	float4 pos_ps = mul(pos_vs, Projection);
	vout.Pos_ws = pos_ws;
	vout.Pos_ps = pos_ps;
	
	vout.Diffuse	= float4(DiffuseColor, Alpha);
	
	return vout;
}

VertexLightingVSOutputTx VSBasicTx(VSInputTx vin)
{
	VertexLightingVSOutputTx vout;
	
	CommonVSOutput cout = ComputeCommonVSOutput(vin.Position);

	vout.PositionPS	= cout.Pos_ps;
	vout.Diffuse	= cout.Diffuse;
	vout.TexCoord	= vin.TexCoord;

	return vout;
}

//-----------------------------------------------------------------------------
// Pixel shaders
//-----------------------------------------------------------------------------

#define MAX_LIGHTS 18

float2 lightPos[MAX_LIGHTS];
float  lightRadius[MAX_LIGHTS];
float  lightIntensity[MAX_LIGHTS];
float4 lightColor[MAX_LIGHTS];

float objectWidth;
float objectHeight;
float2 objectPosition;
float grayScaleIntensity = 1.0f;

float4 PSBasicTx(VertexLightingPSInputTx pin) : COLOR
{
	float4 texCol = tex2D(TextureSampler, pin.TexCoord);
	float4 color = texCol * pin.Diffuse;
	//color.a = texCol.a;

	//Pixel position
        float2 pixelPos = pin.TexCoord.xy * float2(objectWidth, objectHeight) + objectPosition.xy;

	//Grayscale
	float4 grayScale = texCol;
	grayScale.rgb = dot(texCol.rgb, float3(0.3, 0.59, 0.11));

	for(int i = 0; i < MAX_LIGHTS; i++){
		float dist = distance(lightPos, pixelPos);
		float distanceFactor = saturate((lightRadius-dist) / lightRadius);

		grayScale.rgb = lerp(grayScale, texCol, distanceFactor * grayScaleIntensity);

		color += saturate(distanceFactor * lightIntensity) * (grayScale * lightColor);
	}

	//fix overlaping lights
	color.r = min(color.r, texCol.r);
	color.b = min(color.b, texCol.b);
	color.g = min(color.g, texCol.g);

	return color;
}

float4 PSBasicTxNoLight(VertexLightingPSInputTx pin) : COLOR
{
	float4 texCol = tex2D(TextureSampler, pin.TexCoord);
	float4 color  = texCol * pin.Diffuse;
	return color;
}

//-----------------------------------------------------------------------------
// Shader and technique definitions
//-----------------------------------------------------------------------------

int ShaderIndex = 0;


VertexShader VSArray[2] =
{
	compile vs_3_0 VSBasicTx(),
	compile vs_3_0 VSBasicTx(),
};


PixelShader PSArray[2] =
{
	compile ps_3_0 PSBasicTx(),
	compile ps_3_0 PSBasicTxNoLight(),
};


Technique BasicEffect
{
	Pass
	{
		VertexShader = (VSArray[ShaderIndex]);
		PixelShader  = (PSArray[ShaderIndex]);
	}
}

Another thing, and not entirely related to the pixel shader portion, is to merge your View, Projection and World matrices together on the engine side before sending them to the shader. You can then do a single MUL in the VS which should drop the instruction count. Depending on the number of objects this can be a good performance saver.

Another idea is to pass your light radius to the shader as a reciprocal, the result being 1.0/lightRadius. DIVs can be slower than MULs if not payed attention to an IIRC they're done per scalar. You can change your attenuation function to something like this (it's pseudocode, you might need to try it - I'm unfamiliar with the unroll for distance but this *might* give you less instructions, but don't take my word for it!):

float3 dist = (lightPos - pixelPos) * lightRadius; // gives you the distance (assuming pixelpos and lightpos are in the same space).
float distanceFactor = saturate(1.0 + dot(dist, dist)); // quadratic falloff

One more thing: consider merging your constants, that way you don't have to update as many, etc. For example:

float2 lightPos[MAX_LIGHTS];
float  lightRadius[MAX_LIGHTS];
float  lightIntensity[MAX_LIGHTS]; 

Can be (feel free to change the name):

float4 lightPos[MAX_LIGHTS]; // xy: pos, z: radius, w: intensity

The same can be applied to ObjectWidth and ObjectHeight, etc. It might help.

Of course YMMV with all of this but they're guidelines I always follow and they generally help.