DoLoops performance in Fortran - Revision history

Guillaume: /* SetMatrix3D_Constant */

2010-10-29T15:48:15Z

‎SetMatrix3D_Constant

2010-10-29T12:07:48Z

‎Test-cases

2010-10-29T12:07:27Z

‎Test-cases

2010-10-29T12:06:31Z

‎Test-cases

2010-10-29T11:47:45Z

‎Test-cases

2010-10-29T11:37:30Z

‎Test-cases

2010-10-29T10:20:01Z

‎Test-cases

2010-10-29T10:17:39Z

‎Test-cases

2010-10-28T20:02:32Z

‎Test-cases

2010-10-28T20:00:57Z

‎Conclusions

@@ Line 247: / Line 247: @@
 === SetMatrix3D_Constant===
 ====Description====
-This subroutine is in the ModuleFunctions of MohidWater. In the context of MohidWater, parallelizing this subroutine yields a worse performance. However, in the little test program, the same OMP directives yield quite good results.
+This subroutine is in the ModuleFunctions of MohidWater. In the context of MohidWater, parallelizing this subroutine yields up to 15% increase in performance. However, in the little test program, the same OMP directives yield quite good results (under 1/3 of the simulation time or a 200% increase in performance).
 ====Hardware====
 *Core i7-870

@@ Line 335: / Line 335: @@
 <webimage>http://i.imgur.com/0bDqn.png</webimage>
-=== Test-cases===
+=== PCOMS test-case===
 *Here's the present situation with the codeplex build (20101029). The chart below depicts the PCOMS test-case performance with the growing number of threads (up to 8). Maximum performance gains are roughly 15% for the 4 threads. Since the i7-870 is a 4 core machine, it makes sense that 4 threads perform better than 5 or more, or than 3 or less. Also note that a single-threaded openmp code is slower by 9% than a no-openmp code.
 <webimage>http://i.imgur.com/dhSg5.png</webimage>

@@ Line 336: / Line 336: @@
 === Test-cases===
-*Here's the present situation with the codeplex build (20101029). The chart below depicts the PCOMS test-case performance with the growing number of threads (up to 8). Maximum performance gains are roughly 15% for the 4 threads. Since the i7-870 is a 4 core machine, it makes sense that 4 threads perform better than 5 or more, or than 3 or less.
+*Here's the present situation with the codeplex build (20101029). The chart below depicts the PCOMS test-case performance with the growing number of threads (up to 8). Maximum performance gains are roughly 15% for the 4 threads. Since the i7-870 is a 4 core machine, it makes sense that 4 threads perform better than 5 or more, or than 3 or less. Also note that a single-threaded openmp code is slower by 9% than a no-openmp code.
 <webimage>http://i.imgur.com/dhSg5.png</webimage>

@@ Line 336: / Line 336: @@
 === Test-cases===
 ====Without openmp compiler option====
 A 3 hour run of the PCOMS is made, which takes around '''870s''' without parallelization.

@@ Line 337: / Line 337: @@
 === Test-cases===
 ====Without openmp compiler option====
-A 1 hour run of the PCOMS is made, which takes around '''870s''' without parallelization.
+A 3 hour run of the PCOMS is made, which takes around '''870s''' without parallelization.
                            Main ModifyMohidWater                          863.33
                 ModuleFunctions THOMASZ                                    46.06
@@ Line 343: / Line 345: @@
                 ModuleFunctions SetMatrixValues3D_R8_FromMatri             16.69
                 ModuleFunctions InterpolateLinearyMatrix3D                  8.42
 ====With openmp compiler option, with current code from Codeplex (codename: Angela)====

@@ Line 372: / Line 372: @@
 *5 Threads only (set OMP_NUM_THREADS=5) take more than '''370s''':
 <webimage>http://i.imgur.com/6dOxk.png</webimage>
 ==== With openmp, but without any openmp directives ====

@@ Line 358: / Line 358: @@
                 ModuleFunctions InterpolateLinearyMatrix2D                  0.08
-*1 Thread only (set OMP_NUM_THREADS=2) takes more than '''460s''':
+*1 Thread only (set OMP_NUM_THREADS=1) takes more than '''460s''':
 <webimage>http://i.imgur.com/zY3K3.png</webimage>

@@ Line 358: / Line 358: @@
                 ModuleFunctions InterpolateLinearyMatrix2D                  0.08
 *2 Threads only (set OMP_NUM_THREADS=2) take less than '''380s''':
@@ Line 366: / Line 368: @@
 *4 Threads only (set OMP_NUM_THREADS=3) take less than '''360s''':
-<webimage>http://i.imgur.com/VNFuQ.png</webimage>
+<webimage>http://i.imgur.com/VNFuQ.png</webimage>~
 ==== With openmp, but without any openmp directives ====

@@ Line 364: / Line 364: @@
 *3 Threads only (set OMP_NUM_THREADS=3) take less than '''370s''':
 <webimage>http://i.imgur.com/TKBrC.png</webimage>
 ==== With openmp, but without any openmp directives ====

@@ Line 241: / Line 241: @@
 * dynamic small chunks, or no chunk at all yield 10% increased performance over large dynamic chunks. Probably better off with no-chunk.
 * More test-cases representing different scenarios of do-loops may yield different choices of CHUNK/scheduling.
 ----