Skip to content

Latest commit

 

History

History
45 lines (37 loc) · 3.44 KB

README.md

File metadata and controls

45 lines (37 loc) · 3.44 KB
Raw

How to compile RACE examples?

RACE provides examples to illustrate the usage and easiness of using the RACE library. To try it out:

  • cd race/example
  • mkdir build && cd build
  • CC=$(C\_COMPILER) CXX=$(CXX\_COMPILER) cmake .. -DRACE\_DIR=$(RACE\_LIB)
  • optionally set RACE_USE_SPMP for pre-permuting matrix with RCM permutation. Intel SpMP library is used for RCM permutation.
  • make

Note that as some examples depend on sparse BLAS library, linking with BLAS is necessary. The CMake tries to automatically detect BLAS library through environment flags, for example MKLROOT of Intel MKL. However, if RACE couldn't find a BLAS library please link it manually by specifying the include directory using -DCBLAS_INCLUDE_DIR=<path to BLAS include directory>.

Running coloring examples

  • To run benchmarks SpMTV, SymmSpMV, KACZ, GS with 10 threads on first 10 cores, and ε0= ε1=0.8 use the following command:

    RACE_EFFICIENCY=80,80 taskset -c 0-9 ./coloring  -m <matrix market file> -c 10 -t 1

  • Note pinning is carried out by RACE in this case

  • To get other options use: ./coloring -h

  • Relevant environment variables are:

    Name Values Default Recomendation Purpose Example
    RACE_EFFICIENCY [40, 100] 40 If you are unsure, set 80 for first few stages Controls recursion. It sets the load imbalance tolerance εs at each recursive stage s RACE_EFFICIENCY=50,80 implies ε0=0.5 and ε1=0.8.
    RACE_THREADS [1, c], where c is the total number of cores. auto If you are unsure, don't specify this variable. Assign n number of threads at a particular recusion stage. This overrides the internal heuristic to determine number of threads at each stage. RACE_THREADS=2,3 implies 2 threads in stage 0 and 3 threads in stage 1 of recursion.

Running temporal blocking examples

  • To run MPK kernel with power 3, i.e., A3x computation, with 10 threads on first 10 cores on an architecture with last level cache size of 40 MB, use the following command:

    RACE_CACHE_VIOLATION_CUTOFF=1,1 OMP_NUM_THREADS=10 OMP_PLACES=cores OMP_PROC_BIND=close  taskset -c 0-9 ./mtxPower -m <matrix market file> -s 40 -i 3

  • Note in this case pinning is left to user, therefore we use here OMP environment variables

  • To get other options use: ./mtxPower -h

  • Relevant environment variables are:

    Name Values Default Recomendation Purpose Example
    RACE_CACHE_VIOLATION_CUTOFF [1,inf] see next If you are unsure, set 1 for first few stages Controls recursion. It sets the safety factor of cache at different recursion stage. Higher the value less the probablity of going for further recursion. RACE_CACHE_VIOLATION_CUTOFF=1,2 implies safety factor of 1 at stage 0 and 2 at stage 1.
    RACE_CACHE_VIOLATION_CUTOFF_DEFAULT [1, inf] 50 If you are unsure, don't specify this variable. Sets the default safety factor value of all stages. RACE_CACHE_VIOLATION_CUTOFF_DEFAULT=2 implies safety factor is 2 by default on all stages.
    RACE_MAX_RECURSION_STAGES [1, inf] auto If you are unsure, don't specify this variable. Sets the maximum number of recursion stages. RACE_MAX_RECURSION_STAGES=1 implies only 1 recursion stage is allowed.