layout(set=0, binding=0) writeonly buffer destBuffer{
    FLOAT4 data[];
}uOutput;

layout(set=0, binding=1) readonly buffer inputBuffer {
    FLOAT4 data[];
} uInput;

layout(set=0, binding=2) readonly buffer KernelBuffer{
    FLOAT4 data[];
} uSlope;

layout(set = 0, binding = 3) uniform reluBuffer{
    ivec4 imgSize; // plane, 1, channelC4, plane * channelC4
}uPreluParam;

layout(local_size_x = 256, local_size_y = 1, local_size_z = 1) in;

void main()
{
	int index = int(gl_GlobalInvocationID.x);
	int oz = index / uPreluParam.imgSize.x;
    if(index < uPreluParam.imgSize.w)
	{
		FLOAT4 slope = uSlope.data[oz];
		FLOAT4 dataIn = uInput.data[index];
		FLOAT4 dataTemp = dataIn * slope;
		bvec4 lessZero = bvec4(lessThan(dataIn, FLOAT4(0.0)));
		uOutput.data[index] = mix(dataIn, dataTemp, lessZero);
	}

}