Some Concluding Pipeline thoughts.
Notes
- For these notes ARM8 = ARM Cortex-A72
- VLIW - very large instruction word/ Multiple Issue
- I fetch two instructions at the same time (if possible)
- A load/store and a ALU type
- They can execute at the same time if I add another adder (See page 374)
- I probably need the compiler to fix my program for me.
- But I might get a CPI of .5 if EVERYTHING is wonderful.
- Apparently the ARM8 chip has the capacity to issue three instructions at the same time.
- Hence the CPI of better than 1 in the early experiments.
- Out of order execution
- Can be done in hardware
- See the picture p 381
- As long as the program does what the programmer wants.
- ARM 8 does this
- Speculative Execution
- Start doing both sides of a conditional branch instruction
- Flush the parts you don't need
- The ARM 8 does this.
- Deep pipelines
- I believe that the ARM 8 has a 15 state pipeline.
- Different version, but a deep pipeline picture is on page 356
- Intel comparison chart on 354
- The book has a nice section on intel and arm hardware and how it differs from the leg 8.
- If you are moderately interested, this is well worth a read.
- Intel has implemented a "micro architecture"
- The crunchy outside translates the intel instructions into a RISK like language
- The soft caramel inside is a "wide, out-of-order, super-scalar micro architecture" (gemini)
- Branch prediction and speculative execution
- Out of order execution
- 14-19 stages in the pipeline