1. EECS 583 – Class 22 Research Topic 4: Automatic SIMDization - Superword Level Parallelism University of Michigan December 10, 2012

2. - 2 - Announcements v Last class today! »No more reading v Dec 12-18 – Project presentations »Each group sign up for 30-minute slot »See me after class if you have not signed up v Course evaluations reminder »Please fill one out, it will only take 5 minutes »I do read them »Improve the experience for future 583 students

3. - 3 - Notes on Project Demos v Demo format »Each group gets 30 minutes  Strict deadlines enforced because many back to back groups  Don’t be late!  Figure out your room number ahead of time (see schedule on my door) »Plan for 20 mins of presentation (no more!), 10 mins questions  Some slides are helpful, try to have all group members say something  Talk about what you did (basic idea, previous work), how you did it (approach + implementation), and results  Demo or real code examples are good v Report »5 pg double spaced including figures – what you did + why, implementation, and results »Due either when you do your demo or Dec 18 at 6pm

4. - 4 - SIMD Processors: Larrabee (now called Knights Corner) Block Diagram

5. - 5 - Vector Unit Block Diagram

6. - 6 - Processor Core Block Diagram

7. - 7 - Larrabee vs Conventional GPUs v Each Larrabee core is a complete Intel processor »Context switching & pre-emptive multi-tasking »Virtual memory and page swapping, even in texture logic »Fully coherent caches at all levels of the hierarchy v Efficient inter-block communication »Ring bus for full inter-processor communication »Low latency high bandwidth L1 and L2 caches »Fast synchronization between cores and caches v Larrabee: the programmability of IA with the parallelism of graphics processors

8. Exploiting Superword Level Parallelism with Multimedia Instruction Sets

9. Multimedia Extensions Additions to all major ISAs SIMD operations

10. Using Multimedia Extensions Library calls and inline assembly –Difficult to program –Not portable Different extensions to the same ISA –MMX and SSE –SSE vs. 3DNow! Need automatic compilation

11. Vector Compilation Pros: –Successful for vector computers –Large body of research Cons: –Involved transformations –Targets loop nests

12. Superword Level Parallelism (SLP) Small amount of parallelism –Typically 2 to 8-way Exists within basic blocks Uncovered with a simple analysis Independent isomorphic operations –New paradigm

13. 1. Independent ALU Ops R = R + XR * 1.08327 G = G + XG * 1.89234 B = B + XB * 1.29835 R R XR 1.08327 G = G + XG * 1.89234 B B XB 1.29835

14. 2. Adjacent Memory References R = R + X[i+0] G = G + X[i+1] B = B + X[i+2] R G = G + X[i:i+2] B

15. for (i=0; i<100; i+=1) A[i+0] = A[i+0] + B[i+0] 3. Vectorizable Loops

16. for (i=0; i<100; i+=4) A[i:i+3] = B[i:i+3] + C[i:i+3] for (i=0; i<100; i+=4) A[i+0] = A[i+0] + B[i+0] A[i+1] = A[i+1] + B[i+1] A[i+2] = A[i+2] + B[i+2] A[i+3] = A[i+3] + B[i+3]

17. 4. Partially Vectorizable Loops for (i=0; i<16; i+=1) L = A[i+0] – B[i+0] D = D + abs(L)

18. 4. Partially Vectorizable Loops for (i=0; i<16; i+=2) L0 L1 = A[i:i+1] – B[i:i+1] D = D + abs(L0) D = D + abs(L1) for (i=0; i<16; i+=2) L = A[i+0] – B[i+0] D = D + abs(L) L = A[i+1] – B[i+1] D = D + abs(L)

19. Exploiting SLP with SIMD Execution Benefit: –Multiple ALU ops  One SIMD op –Multiple ld/st ops  One wide mem op Cost: –Packing and unpacking –Reshuffling within a register

20. Packing/Unpacking Costs C = A + 2 D = B + 3 C A 2 D B 3 = +

21. Packing/Unpacking Costs Packing source operands A B A = f() B = g() C = A + 2 D = B + 3 C A 2 D B 3 = +

22. Packing/Unpacking Costs Packing source operands Unpacking destination operands C D A = f() B = g() C = A + 2 D = B + 3 E = C / 5 F = D * 7 A B C A 2 D B 3 = +

23. Optimizing Program Performance To achieve the best speedup: –Maximize parallelization –Minimize packing/unpacking Many packing possibilities –Worst case: n ops  n! configurations –Different cost/benefit for each choice

24. A = B + C D = E + F Observation 1: Packing Costs can be Amortized Use packed result operands G = A - H I = D - J

25. Observation 1: Packing Costs can be Amortized Use packed result operands Share packed source operands A = B + C D = E + F G = B + H I = E + J A = B + C D = E + F G = A - H I = D - J

26. Observation 2: Adjacent Memory is Key Large potential performance gains –Eliminate ld/str instructions –Reduce memory bandwidth Few packing possibilities –Only one ordering exploits pre-packing

27. SLP Extraction Algorithm Identify adjacent memory references A = X[i+0] C = E * 3 B = X[i+1] H = C – A D = F * 5 J = D - B

28. SLP Extraction Algorithm Identify adjacent memory references A = X[i+0] C = E * 3 B = X[i+1] H = C – A D = F * 5 J = D - B ABAB = X[i:i+1]

29. SLP Extraction Algorithm Follow def-use chains A = X[i+0] C = E * 3 B = X[i+1] H = C – A D = F * 5 J = D - B ABAB = X[i:i+1]

30. SLP Extraction Algorithm Follow def-use chains A = X[i+0] C = E * 3 B = X[i+1] H = C – A D = F * 5 J = D - B ABAB = X[i:i+1] HJHJ CDCD ABAB = -

31. SLP Extraction Algorithm Follow use-def chains A = X[i+0] C = E * 3 B = X[i+1] H = C – A D = F * 5 J = D - B ABAB = X[i:i+1] HJHJ CDCD ABAB = -

32. SLP Extraction Algorithm Follow use-def chains A = X[i+0] C = E * 3 B = X[i+1] H = C – A D = F * 5 J = D - B ABAB = X[i:i+1] CDCD EFEF 3535 = * HJHJ CDCD ABAB = -

33. SLP Extraction Algorithm Follow use-def chains A = X[i+0] C = E * 3 B = X[i+1] H = C – A D = F * 5 J = D - B ABAB = X[i:i+1] CDCD EFEF 3535 = * HJHJ CDCD ABAB = -

34. SLP Availability

35. SLP vs. Vector Parallelism

36. Conclusions Multimedia architectures abundant –Need automatic compilation SLP is the right paradigm –20% non-vectorizable in SPEC95fp SLP extraction successful –Simple, local analysis –Provides speedups from 1.24 – 6.70 Found SLP in general-purpose codes