Matrox Helios Series Installation And Hardware Reference page 92

92 Chapter 4: Matrox Helios hardware reference
Lookup tables
For each pixel, up to five values (MAC results or point sources) can be transferred
to the register file of the ALU, into the first five of the seven 40-bit registers.
Registers that are not needed to hold MAC results or point sources can be used
to accumulate results during a processing pass, and any of them can be used to
hold intermediate values temporarily during a single sequence of instructions.
The ALU also has sixteen 40-bit constant registers, which are pre-loaded at the
beginning of a processing pass with values needed by the instruction sequence.
On each clock cycle, the ALU can execute an instruction that operates on one or
two operands. The operands can be:
• A register from the register file.
• A constant register.
• The result of an ALU operation (except for multiply) from one or two instructions
before.
• The result of a multiply operation from two or three instructions before.
The ALU can perform one of the following instructions: absolute value, clip, shift,
add, subtract, multiply, minimum, maximum, logical AND, logical OR, logical
XOR, logical NOT, or LUT map. It can also perform other more specialized
operations. Note that all instructions can work on 40 bits, except multiply
instructions which work on 16 bits (producing a 32-bit result).
The hardware provides direct support in the ALU for LUTs with a size of 256 bits.
This is enough to implement an 8-in:1-out, 7-in:2-out, 6-in:4-out, or 5-in:8-out
LUT in a single instruction. Using multiple 8-in:1-out LUT mapping instructions
(which can shift and merge mapping results with the result of the previous LUT
instruction), the ALU can perform a very fast 256 entry 8-bit LUT mapping. Even
larger LUT mappings can be constructed out of a combination of basic LUT
instructions; however, the larger the LUT, the slower the operation.