stream benchmark pthread version

Program: STREAM

Parallel Library: pthread

Orignial Algorithm by John D. McCalpin

Programmer: Jian Fang (TU Delft)

Copyright: Jian Fang([email protected])

License: You are free to use this program and modify this program
for your own use. If you publish results obtained from this program,
please cite both this program and the orignal virginia version.

• 1. There are three arrays which is indicate as a, b, c. You can
        define you own type of data for the array. For default, it is
     int64_t, which means each element takes up 8B.
• 2. The total data access amount consider the cache write strategy
        in most of the current system that if a write is not writing a
     whole cacheline, it needs to read the cacheline first before
     writing it back. So each write miss for L3 leads to an extra
     read. If the system you are testing is not using this strategy,
     please modify the calculation method or the 'amountFactor'.
• 3. The program works for both share and non-share memory
     channels. share channel means the read and write are using a
     same memory channel, and it can be use as only read or only
     write at a same time. non-share channels mean that the memory
     has both individual read channels and individual write channels,
     the read and write can happen in parallel.
• 4. This benchmark contain 9 operations: (n is a constant)

  OPERATION TYPE	CODE	DESCRIPTION	amountFactor
  READ	0	sum += c[j]	1
  WRITE	1	c[j] = n	2
  SELFINC	2	c[j] ++	2
  COPY	3	c[j] = a[j]	3
  MUL	4	c[j] = a[j]*n	3
  SELFADD	5	c[j] += a[j]	3
  ADD	6	c[j] = a[j]+b[j]	4
  MULADD	7	c[j] = a[j]*n+b[j]	4
  SELFDADD	8	c[j] += a[j]+b[j]	4

• 5. This program is written in C/C++ and using the pthread library.
        You can change the mapping[] array in the source code to
     change the core binding. For example you have a 2 nodes numa
     system. For node 1, it has core 0-9, and node 2 has core
     10-19. If you only want to test the bandwidth of node 1,
     just place 0-9 in the mapping[] array.
• 6. There is no Makefile yet. We provide a test.sh file example
        which contains instructoins and an example how to compile tihs
     program, as well as a way to run this test. You are welcome to
     help and make you own Makefile.
• 7. If you have any suggestions or comments, please feel free to
        contact the author:
     Jian Fang([email protected])