-1

I have got a list of 2 types of hazards:

1a. EX/MEM.RegisterRd = ID/EX.RegisterRs
1b. EX/MEM.RegisterRd = ID/EX.RegisterRt

2a. MEM/WB.RegisterRd = ID/EX.RegisterRs
2b. MEM/WB.RegisterRd = ID/EX.RegisterRt

I am not able to understand the intuition behind these 2 rules which can help me know the technical terms and also explains me the concept? Any explanations are welcome :)

Progman
  • 16,827
  • 6
  • 33
  • 48
brainstorm12
  • 1
  • In most 3-operand ISAs (e.g. MIPS documentation like http://www.mrc.uidaho.edu/mrc/people/jff/digital/MIPSir.html uses this convention), `rd` is the destination register, and Rs, Rt are source registers. e.g. `add rd, rs, rt`. (rs and rt might be second and third, or source and third, IDK). – Peter Cordes Jun 25 '20 at 21:26

1 Answers1

0

In most 3-operand ISAs (e.g. MIPS documentation like http://www.mrc.uidaho.edu/mrc/people/jff/digital/MIPSir.html uses this convention), rd is the destination register, and Rs, Rt are source registers. e.g. add rd, rs, rt. (rs and rt might be second and third, or source and third, IDK).

If you're reading a register (in ID) which was recently written (the instruction writing it hasn't reached the write-back stage), that's a RAW read-after-write true dependency.

Out-of-order exec also introduces the possibility of write-after-write and write-after-read anti-dependency hazards. https://en.wikipedia.org/wiki/Hazard_(computer_architecture)#Data_hazards. But in a scalar in-order pipeline, only true dependencies are a concern, I think. At least if all instructions have fixed 1-cycle latency, so excluding funky stuff like MIPS mult or div that write the hi:lo pair.

Peter Cordes
  • 328,167
  • 45
  • 605
  • 847