Add Support for Computations with 8-Bit Data.

[ballaib]

I wanted to use Kompute to do processing on 8-bit images. However, this does not appear to be possible without converting the image to another format beforehand. This made Kompute infeasible for my use case. It would be really helpful if Kompute supported computations with 8-bit data.

[axsaucedo]

You can add uint_8 data type quite easily by following

kompute/test/TestOpShadersFromStringAndFile.cpp

Lines 12 to 59 in a155585

	// Introducing custom struct that can be used for tensors
	struct TestStruct {
	float x;
	uint32_t y;
	int32_t z;
	// Creating an == operator overload for the comparison below
	bool operator==(const TestStruct rhs) const {
	return this->x == rhs.x && this->y == rhs.y && this->z == rhs.z;
	}
	};
	// Custom struct needs to be mapped the eCustom datatype
	template<>
	kp::Tensor::TensorDataTypes
	kp::TensorT<TestStruct>::dataType()
	{
	return Tensor::TensorDataTypes::eCustom;
	}
	TEST(TestShader, ShaderRawDataFromConstructorCustomDataType)
	{
	std::string shader(R"(
	#version 450
	layout (local_size_x = 1) in;
	layout(std140, binding = 0) buffer a {
	float ax;
	uint ay;
	int az;
	};
	layout(std140, binding = 1) buffer b {
	float bx;
	uint by;
	int bz;
	};
	void main() {
	bx = ax;
	by = ay;
	bz = az;
	}
	)");
	kp::Manager mgr;
	std::shared_ptr<kp::TensorT<TestStruct>> tensorA = mgr.tensorT<TestStruct>({ { 0.1, 2, 3} });
	std::shared_ptr<kp::TensorT<TestStruct>> tensorB = mgr.tensorT<TestStruct>({ { 0.0, 0, 0} });

We are looking to add this by default however.