reveal

[jg piccinali]

• man reveal

step1: Loop work estimate

./reframe.py --exec-policy async -p PrgEnv-cray --system dom:gpu --keep-stage-files 
--prefix=$SCRATCH/reframe -r 
-c checks/tools/profiling_and_debugging/reveal1.py 
-n reveal_jacobi_C

cp -a jacobi+*t $HOME

CrayPat/X:  Version 6.5.1 Revision ae17549  07/24/17 13:15:12
Experiment:                  lite  lite/loop_profile
Number of PEs (MPI ranks):     12
Numbers of PEs per Node:       12
Numbers of Threads per PE:      1
Number of Cores per Socket:    12
Execution start time:  Tue Mar 20 15:34:04 2018
System name and speed:  nid00033  2601 MHz (approx)
Intel haswell CPU  Family:  6  Model: 63  Stepping:  2


Avg Process Time:  0.86 secs        
High Memory:      589.5 MBytes 49.1 MBytes per PE

Table 1:  Inclusive and Exclusive Time in Loops (from -hprofile_generate)

 Loop Incl |     Loop |     Time |   Loop |    Loop |  Loop |  Loop | Function=/.LOOP[.]
     Time% |     Incl |    (Loop |    Hit |   Trips | Trips | Trips |  PE=HIDE
           |     Time |    Adj.) |        |     Avg |   Min |   Max | 
|-----------------------------------------------------------------------------
| 34,779.8% | 0.809934 | 0.363888 |      1 |   200.0 |   200 |   200 | Jacobi.LOOP.0001.li.59
| 19,153.9% | 0.446046 | 0.000724 |    200 |   166.5 |   166 |   167 | Jacobi.LOOP.0002.li.69
| 19,122.8% | 0.445322 | 0.445322 | 33,300 | 1,998.0 | 1,998 | 1,998 | Jacobi.LOOP.0003.li.71
| 13,418.7% | 0.312487 | 0.000967 |    200 |   166.5 |   166 |   167 | ExchangeJacobiMpiData.LOOP.0001.li.155
| 13,377.2% | 0.311520 | 0.311520 | 33,300 | 2,000.0 | 2,000 | 2,000 | ExchangeJacobiMpiData.LOOP.0002.li.157
|     93.6% | 0.002180 | 0.000002 |      1 |   168.5 |   168 |   169 | InitializeMatrix.LOOP.0001.li.210
|     93.5% | 0.002178 | 0.002178 | 168.50 | 2,000.0 | 2,000 | 2,000 | InitializeMatrix.LOOP.0002.li.212
|     33.3% | 0.000775 | 0.000002 |      1 |   168.5 |   168 |   169 | CheckError.LOOP.0001.li.237
|     33.2% | 0.000773 | 0.000773 | 166.67 | 2,000.0 | 2,000 | 2,000 | CheckError.LOOP.0002.li.245

step2: Program Library

./reframe.py --exec-policy async -p PrgEnv-cray --system dom:gpu --keep-stage-files 
--prefix=$SCRATCH/reframe -r 
-c checks/tools/profiling_and_debugging/reveal2.py 
-n reveal_jacobi_C

cp -a 0.pl $HOME

step3: Report (needs X11)

cd $HOME
reveal ./0.pl   ./jacobi+*t

scope loop

• add loop to the scoping list by right clicking on "Scope Loops":

• scope the loop(s):

• view openMP directives: