Tool/software: TI C/C++ Compiler
Hello.
I`m myungjin Lee.
I refered following documents : (Ref. TMS320C66x DSP Literature Number: SPRUGH7 : CPU and Instruction Set)
I`m going to implement matrix multiplier.
I already implemented 16 bit-multiplier using 4 way instruction.
And now, I want to implement 8 bit multiplier.
It should be faster than 16 bit multiplier.
For the implementation, I think I should use MPY or DOT instruction.
But I found out that c66x can do 4 way 16bit signed operation(multiply and dot product), but don`t have 4 way 8 bit signed operation.
Is it right? It just have signed-unsigned and unsigned-unsigned operation.
This document describes that c66x can execute up to 32 (16x16-bit) multiplies per cycle.
But I couldn`t find out instructions that can execute up to 32 multiplies.
Would you let me know, what is the best instruction about mac-performance?
Best regards,
Myungjin.
