Not a member of Pastebin yet?
Sign Up,
it unlocks many cool features!
- 10.26
- 11.3, .6, .7, .14, .15
- 12.6, .12
- 13.3, .4
- 14.3, .5
- 15.1, .4
- 16.3, .7
- 17.5, .9, .16
- Addressing Modes
- Pipeline Hazards
- RISC Designs
- 1. More registers
- 2. Speed up
- 3. Loops and Loads/Stores
- Instruction lvl Parallelism
- -how muh
- Cache coherence
- -understand
- -protocol for c.c
- -how works: directory, mesi
- Multicore computers
- -
- Betterz
- -Fixed of variable length instructions: fixed -- fetch, look at opcode, then operands... If fixed, opcode easy
- -Large cache vs large registers: registers -- access patterns; registers fast, cache slow; cache -- less memory look-ups
- -Short vs long instruction pipeline: short -- bad because need to wait for opcode to load; long -- bad because need more registers to compensate for instruction
- -RISC vs CISC RISC -- fast, CICS --
- -Scoreboard vs Tomasulo tomasulo -- register renaming, avoid dependencies Write after read, read after write; scoreboard -- depect dependency and stall
- -Hardwired vs microprogammed hardwire -- faster, difficult to debug; microprogrammed -- functionality
- -Directory vs MESI cache coherence directory -- NUMA (to directory through bus, then everywhere else); MESI -- common bus braodcast faster
- I1: r3b = r3a + r5a
- I2: r4a = r3b + 1
- I3: r3c = r5a +1
- I4: r7a = r3c + r4a
- I5: r5b = r4a + r7a
- I2: RAW depends on I1, r3
- I3: WAR depends on I2, r3
- I4: RAW depends on I3, r3
- I5: WAR depends on I4, r7
- I1: r3b = r3a + r5a
- I2: r4a = r3b + 1
- I3: r3c = r5a +1
- I4: r7a = r3c + r4a
- I5: r5b = r4a + r7a
- redesign using fetch, indirect
- MBR --> PC
- PC --> MAR
- MAR --> memory
- memory --> MBR
Add Comment
Please, Sign In to add comment