Bathymetry 2d boxes

[jmsexton03]

Add FillBoundary to fix ghost zones in W for 3d rhs and in DUon, DVom for 2d rhs

[asalmgren]

That box is definitely not 2D -- when I ran Upwelling just now it prints as ((-2,-2,0) (42,81,15) (0,0,0))

…

On Mon, Sep 18, 2023 at 8:59 PM Weiqun Zhang ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In Source/TimeIntegration/ROMSX_advance_2d.cpp <#30 (comment)>: > + }); + amrex::ParallelFor(tbxp2, + [=] AMREX_GPU_DEVICE (int i, int j, int ) + { + + const auto prob_lo = geomdata.ProbLo(); + const auto dx = geomdata.CellSize(); + + pm(i,j,0)=dxi[0]; + pn(i,j,0)=dxi[1]; + //defined UPWELLING + Real f0=-8.26e-5; + Real beta=0.0; + Real Esize=1000*(Mm); + Real y = prob_lo[1] + (j + 0.5) * dx[1]; + Real f=fomn(i,j,0)=f0+beta*(y-.5*Esize); This pattern appears in a number of places. I will submit a PR tomorrow. I did not change it last time because if the Box is 2D, this is harmless even though it wastes a write. But if it is not 2d, there are race conditions. — Reply to this email directly, view it on GitHub <#30 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACRE6YXCYUCHXJAKS67AYR3X3EKCNANCNFSM6AAAAAA45CPWTE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

-- Ann Almgren Senior Scientist; Dept. Head, Applied Mathematics Pronouns: she/her/hers

[asalmgren]

what does this line even do? Real f=fomn(i,j,0)=f0+beta*(y-.5*Esize); Is Real a = b = c the same as Real a = c Real b = c ??

…

On Tue, Sep 19, 2023 at 9:31 AM Weiqun Zhang ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In Source/TimeIntegration/ROMSX_advance_2d.cpp <#30 (comment)>: > + + const auto prob_lo = geomdata.ProbLo(); + const auto dx = geomdata.CellSize(); + + pm(i,j,0)=dxi[0]; + pn(i,j,0)=dxi[1]; + //defined UPWELLING + Real f0=-8.26e-5; + Real beta=0.0; + Real Esize=1000*(Mm); + Real y = prob_lo[1] + (j + 0.5) * dx[1]; + Real f=fomn(i,j,0)=f0+beta*(y-.5*Esize); + fomn(i,j,0)=f*(1.0/(pm(i,j,0)*pn(i,j,0))); + }); + + amrex::ParallelFor(tbxp2, ⬇️ Suggested change - amrex::ParallelFor(tbxp2, + amrex::ParallelFor(tbxp2D, — Reply to this email directly, view it on GitHub <#30 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACRE6YUETFLUZBCHJ4T53GTX3HCFFANCNFSM6AAAAAA45CPWTE> . You are receiving this because you commented.Message ID: ***@***.***>

-- Ann Almgren Senior Scientist; Dept. Head, Applied Mathematics Pronouns: she/her/hers

[WeiqunZhang]

Right. The original code is like

Real a = b = c;
b = a * x;

This is equivalent to

b = c;
Real a = b;
b = a * x;

For CPU, this is fine even though unnecessary. But for GPU, Ssince b is shared by all gpu threads, there is a race condition.

[WeiqunZhang]

After merging #35 into this, the diff between CPU and GPU is

$ fcompare.gnu.ex plt00100 cpu/plt00100/

            variable name            absolute error            relative error
                                        (||A - B||)         (||A - B||/||A||)
 ----------------------------------------------------------------------------
 level = 0
 temp                               5.329070518e-14           2.464088545e-15
 omega                                            0                         0
 salt                               7.105427358e-15           2.030122102e-16
 x_velocity                         8.326672685e-17           1.132812766e-15
 y_velocity                         6.245004514e-17           1.085894626e-15
 z_velocity                                       0                         0