MP_STACK_SIZE=10000000
4Mb pages used (sysconfigtab)
no processor sets used
Tuning pseudo-switches:
+BPD - Big page disable
+PFB - Post-link feedback for prefetches
See the flags description or the configuration file
for more details.
Base:
Fortran: f90 -arch ev7 -fast -omp -O5
C: cc -arch ev7 -fast -omp -O4
alternate source purdue1 for 330.art_m
Peak:
All use -arch host -omp ONESTEP
Individual benchmark tuning:
310.wupwise_m: f90 -arch host -omp -call_shared
-inline all -unroll 12 -align commons
+PFB +BPD
alternate source ompl.32
312.swim_m: f90 -arch host -omp -fast -O5
+PFB +BPD
PTHREAD_CONFIG=feature=def-scs,d4-scs
alternate source ompl.32
314.mgrid_m: f90 -arch host -omp -O5 -transform_loops
-tune ev7 -unroll 8
+PFB +BPD
PTHREAD_CONFIG=feature=def-scs,d4-scs
alternate source ompl.32
316.applu_m: f90 -arch host -omp -fast -O5 -unroll 14
+PFB
PTHREAD_CONFIG=feature=def-scs,d4-scs
alternate source ompl.32
318.galgel_m: f90 -arch host -omp -fast -unroll 5 -O5
+PFB
320.equake_m: cc -arch host -omp -fast -call_shared
-O4 -ldensemalloc -assume restricted_pointers
-inline speed -unroll 13
+PFB +BPD
PTHREAD_CONFIG=feature=def-scs,d4-scs
alternate source ompl.32
324.apsi_m: f90 -arch host -omp -O5 -transform_loops
-unroll 8
+PFB
PTHREAD_CONFIG=feature=def-scs,d4-scs
alternate source ompl.32
326.gafort_m: f90 -arch host -omp -fast -O5
-arch ev67 -tune ev67
alternate source ompl.32
328.fma3d: f90 -arch host -omp -fast -O4
-transform_loops
+PFB +BPD
PTHREAD_CONFIG=feature=def-scs,d4-scs
alternate source ompl.32
330.art_m: cc -arch host -omp -assume whole_program
-ldensemalloc -call_shared
-assume restricted_pointers -fast -O4 -unroll 16
-inline none
+PFB +BPD
alternate source purdue1
332.ammp_m: cc -arch host -omp -O4 -ifo
-assume nomath_errno
-assume trusted_short_alignment -fp_reorder
-readonly_strings -ldensemalloc
-assume restricted_pointers -unroll 9
PTHREAD_CONFIG=feature=def-scs,d4-scs
Information on UNIX V5.1B Patches can be found at
http://h30097.www3.hp.com/docs/patch/51B/bl25/ReleaseNotes.pdf
The densemalloc library provides an alternate set of malloc
routines that implement a "dense malloc" memory allocation
strategy. This can improve run time performance by reducing
bookkeeping overhead and rounding up padding attached to
allocated chunks.
File containing flag descriptions can be found at
http://www.spec.org/omp/results/flags/HP-ALPHA-TRU64-F90-CompaqC.txt
|
