Skip to content

Faster and less stable #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
serge-sans-paille wants to merge 5 commits into
base: master
Choose a base branch
from
Open

Faster and less stable #2

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
6cc507b
Fix bug in C version
serge-sans-paille Feb 10, 2014
b7b9892
Fix a translation bug in stability
serge-sans-paille Feb 10, 2014
744f065
Add a computation intensive test case
serge-sans-paille Feb 10, 2014
9561966
Add a utility to dump the textual content of a .grid
serge-sans-paille Feb 11, 2014
e2d2548
Prevent denormalized number when SSE is activated
serge-sans-paille Feb 10, 2014
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
19 changes: 15 additions & 4 deletions ramsurflib.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -370,7 +370,7 @@ void deriv(int n, float sig, double alp,
dh1[0][1]=sig*0.5;
double exp1 = -0.5;
dh2[0][0]=alp;
dh2[0][1]=alp;
dh2[0][1]=0.;
double exp2=-1.0;
dh3[0][0]=-2.0*nu;
dh3[0][1]=0.;
Expand Down Expand Up @@ -830,7 +830,7 @@ void matrc(size_t mz, size_t mp, int const nz, int const np, int const iz, float
static
void updat( ramsurf_t const* rsurf, size_t *profl_index, size_t mr, size_t mz, size_t mp, int nz, int np, int *iz, int *ib,
float dr, float dz, float omega, float k0,
float r, float *rp, float rs,
float r, float *rp, float *rs,
float rb[mr], float zb[mr] , float cw[mz], float cb[mz], float rhob[mz],
float attn[mz], float alpw[mz], float alpb[mz],
fcomplex ksq[mz], fcomplex ksqw[mz], fcomplex ksqb[mz],
Expand Down Expand Up @@ -871,8 +871,9 @@ void updat( ramsurf_t const* rsurf, size_t *profl_index, size_t mr, size_t mz, s
//
// Turn off the stability constraints.
//
if(r>=rs) {
if(r>=*rs) {
int ns=0;
*rs=2.f*(rsurf->rmax);
epade(mp, np, ns, 1, k0, dr, (float (*)[2])pd1, (float (*)[2])pd2);
matrc(mz, mp, nz, np, *iz, dz, k0, rhob, alpw, alpb, (float (*)[2])ksq, (float (*)[2])ksqw, (float (*)[2])ksqb,
f1, f2, f3, r1, r2, r3, s1, s2, s3, (float (*)[2])pd1, (float (*)[2])pd2, *izsrf);
Expand Down Expand Up @@ -1118,6 +1119,13 @@ int ramsurf(ramsurf_t const* rsurf, int * lz, float *** ogrid, FILE *fdline)
fix_zmax(&lzmax, rsurf->rhob);
fix_zmax(&lzmax, rsurf->attn);

#ifdef __SSE3__
// this sets the "flush to zero" and "denormals are zero" bits (15,6) in the MXCSR register.
// otherwise we get very bad timings...
int flush_mode = _MM_GET_FLUSH_ZERO_MODE( );
_MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON);
#endif

size_t mr = rsurf->mr,
mz = lzmax / rsurf->dz + 2.5f,
mp = rsurf->np;
Expand Down Expand Up @@ -1177,7 +1185,7 @@ int ramsurf(ramsurf_t const* rsurf, int * lz, float *** ogrid, FILE *fdline)
//
while (r < rsurf->rmax) {
updat(rsurf, &profl_index, mr, mz, mp, nz, np, &iz, &ib, dr, dz, omega, k0, r,
&rp, rs, *rb, *zb, *cw, *cb, *rhob, *attn, *alpw, *alpb, *ksq, *ksqw, *ksqb, *f1, *f2, *f3,
&rp, &rs, *rb, *zb, *cw, *cb, *rhob, *attn, *alpw, *alpb, *ksq, *ksqw, *ksqb, *f1, *f2, *f3,
*r1, *r2, *r3, *s1, *s2, *s3, *pd1, *pd2, *rsrf, *zsrf, &izsrf, &isrf);
solve(mz, mp, nz, np, *u, *r1, *r3, *s1, *s2, *s3);
r=r+dr;
Expand All @@ -1191,6 +1199,9 @@ int ramsurf(ramsurf_t const* rsurf, int * lz, float *** ogrid, FILE *fdline)

// deallocation step
free(oscratch);
#ifdef __SSE3__
_MM_SET_FLUSH_ZERO_MODE(flush_mode);
#endif

return errorCode;
}
Expand Down
20 changes: 12 additions & 8 deletions tests/Makefile.am
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,7 @@
check_PROGRAMS=ramcmp
check_PROGRAMS=ramcmp ramdump

ramcmp_SOURCES=ramcmp.f
ramdump_SOURCES=ramdump.f

TESTS=decreasing_slop.sh\
flat.sh\
Expand All @@ -9,7 +10,8 @@ TESTS=decreasing_slop.sh\
flat_pade.sh\
flat_limit.sh\
increasing_slop.sh\
flat_2ssps.sh
flat_2ssps.sh\
intensive.sh

EXTRA_DIST=decreasing_slop.test\
flat.test\
Expand All @@ -18,18 +20,20 @@ EXTRA_DIST=decreasing_slop.test\
flat_pade.test\
flat_limit.test\
increasing_slop.test\
flat_2ssps.test
flat_2ssps.test\
intensive.test

SUFFIXES=.test .sh

.test.sh:$(srcdir)/Makefile.am
echo '#!/bin/sh' > $@
echo 'set -e' >> $@
echo 'chmod -R u+rwx `basename $<`.d || true' >> $@
echo 'rm -rf `basename $<`.d' >> $@
echo 'cp -r $(srcdir)/`basename $<` `basename $<`.d' >> $@
echo 'chmod -R u+rwx `basename $<`.d' >> $@
echo 'cd `basename $<`.d' >> $@
echo 'TARGET="`basename $<`"' >> $@
echo 'test ! -d $$TARGET.d || chmod -R u+rwx $$TARGET.d' >> $@
echo 'rm -rf $$TARGET' >> $@
echo 'cp -r $(srcdir)/$$TARGET $$TARGET.d' >> $@
echo 'chmod -R u+rwx $$TARGET.d' >> $@
echo 'cd $$TARGET.d' >> $@
echo '$(abs_top_builddir)/ramsurf1.5' >> $@
echo 'mv tl.grid ref.grid' >> $@
echo 'mv tl.line ref.line' >> $@
Expand Down
24 changes: 24 additions & 0 deletions tests/intensive.test/ram.in
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,24 @@
header line
2048.00 5 1000 1.0 frequency (Hz), source depth (m), tl.line depth (m)
400000. 10 40 domain range (m), computation step (m), mesure step (comp.step)
1000 0.2 10 1000 domain depth (m), computation step (m), mesure step (comp.step), last mesure depth (m)
1500.000000 4 1 0.000000 reference celerity, #PADE, #stability constraints, radius of constraint
0.0 0.0 surface depth (range, depth) (m)
-1 -1
0.0000 1000. bathymetry
400000.000 900
-1 -1
0.0000 1500
333.000 1500
334.000 1510
666.000 1510
667.000 1500
1000.000 1500
-1 -1
0.0000000 3000.0000000
-1 -1
0.0000000 2.
-1 -1
0.0000000 0.900
21 100.00
-1 -1
2 changes: 1 addition & 1 deletion tests/ramcmp.f
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
program ramdec
real tl(1000,1000)
real rtl(1000,1000)
threshold = 6.8
threshold = .5
open(unit=2,status='unknown',file='ref.grid', form='unformatted')
open(unit=3,status='unknown',file='tl.grid', form='unformatted')
read(2)nry
Expand Down
15 changes: 15 additions & 0 deletions tests/ramdump.f
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,15 @@
program ramdec
real tl(1000,1000)
CHARACTER(len=256) :: arg
threshold = .3
eps = 1e-10
CALL get_command_argument(1, arg)
open(unit=2,status='unknown',file=arg, form='unformatted')
read(2)nry
nx=1
5 read(2,end=6)(tl(j,nx),j=1,nry)
write(*,*)(tl(j,nx) ,j=1,nry)
nx=nx+1
go to 5
6 end