Thijs Koerselman

Hi. I'm stuck again... Finally I'm talking to a matrix in my patch.

x->pos_mx is a valid pointer to a t_jit_matrix. The atom array is filled

with valid arguments. The setall() function works fine, the setcell()

I'm using the windows sdk, while reading the new mac sdk. Is this maybe the

problem? I noticed that jit_object_method_typed is also not available for

What the hell am I missing here? I don't get any errors from max or my

compiler. If setcell() is not available what methods can I use for fast

matrix access? In this case I want to write xyz values to a 3 plane matrix.

void jit_csvs_stream(t_jit_csvs *x, long offset)

m = jit_csvs_get_marker(x, mrk_id, offset);

----------------------------------------------

jit_atom_setfloat(&tmpatom, (float)(offset/128.f));

jit_object_method(x->pos_mx, gensym("setall"),0L,1,&tmpatom);

jit_object_method(x->pos_mx, gensym("setcell"),0L,argc,argv);

Hi. I'm stuck again... Finally I'm talking to a matrix in my patch.
x->pos_mx is a valid pointer to a t_jit_matrix. The atom array is filled
with valid arguments. The setall() function works fine, the setcell()
function does nothing.

I'm using the windows sdk, while reading the new mac sdk. Is this maybe the
problem? I noticed that jit_object_method_typed is also not available for
windows....(?).

What the hell am I missing here? I don't get any errors from max or my
compiler. If setcell() is not available what methods can I use for fast
matrix access? In this case I want to write xyz values to a 3 plane matrix.

void jit_csvs_stream(t_jit_csvs *x, long offset)
{
    long mrk_id = 5, tmplong;
    t_mrk *m = NULL;
    long argc = 5;
    t_atom argv[5];
    t_atom tmpatom;

    if(flags.pos_mx_isset)
    {
        m = jit_csvs_get_marker(x, mrk_id, offset);
        if(m)
        {
            jit_atom_setlong(argv, offset);
            jit_atom_setlong(argv+1, mrk_id);
            jit_atom_setfloat(argv+2,(m->x)*0.001f);
            jit_atom_setfloat(argv+3,(m->y)*0.001f);
            jit_atom_setfloat(argv+4,(m->z)*0.001f);

         // ----------this one works fine
----------------------------------------------
            jit_atom_setfloat(&tmpatom,  (float)(offset/128.f));
            jit_object_method(x->pos_mx, gensym("setall"),0L,1,&tmpatom);

         // -------- this seems to do nothing
------------------------------------
            jit_object_method(x->pos_mx, gensym("setcell"),0L,argc,argv);

        }else{
            post("marker pointer null");
        }
    }
}

setcell-jit_object_method

After some further investigation I drew the conclusion that you probably

have to access the matrix data directly. Although the scalebias example is

still a bit confusing to me, its in there, so I guess I will figure it out.

I think it would still be very convenient to have an interface like

setcell(). So if its available please let me know.

> Hi. I'm stuck again... Finally I'm talking to a matrix in my patch.

> x->pos_mx is a valid pointer to a t_jit_matrix. The atom array is filled

> with valid arguments. The setall() function works fine, the setcell()

> I'm using the windows sdk, while reading the new mac sdk. Is this maybe

> the problem? I noticed that jit_object_method_typed is also not available

> What the hell am I missing here? I don't get any errors from max or my

> compiler. If setcell() is not available what methods can I use for fast

> matrix access? In this case I want to write xyz values to a 3 plane matrix.

> void jit_csvs_stream(t_jit_csvs *x, long offset)

> m = jit_csvs_get_marker(x, mrk_id, offset);

> jit_atom_setfloat(argv+2,(m->x)*0.001f);

> jit_atom_setfloat(argv+3,(m->y)*0.001f);

> jit_atom_setfloat(argv+4,(m->z)*0.001f);

> ----------------------------------------------

> jit_atom_setfloat(&tmpatom, (float)(offset/128.f));

> jit_object_method(x->pos_mx, gensym("setall"),0L,1,&tmpatom);

> jit_object_method(x->pos_mx, gensym("setcell"),0L,argc,argv);

After some further investigation I drew the conclusion that you probably
have to access the matrix data directly. Although the scalebias example is
still a bit confusing to me, its in there, so I guess I will figure it out.
I think it would still be very convenient to have an interface like
setcell(). So if its available please let me know.

On 10/30/06, Thijs Koerselman  wrote:
>
> Hi. I'm stuck again... Finally I'm talking to a matrix in my patch.
> x->pos_mx is a valid pointer to a t_jit_matrix. The atom array is filled
> with valid arguments. The setall() function works fine, the setcell()
> function does nothing.
>
> I'm using the windows sdk, while reading the new mac sdk. Is this maybe
> the problem? I noticed that jit_object_method_typed is also not available
> for windows....(?).
>
> What the hell am I missing here? I don't get any errors from max or my
> compiler. If setcell() is not available what methods can I use for fast
> matrix access? In this case I want to write xyz values to a 3 plane matrix.
>
> cheers, -thijs
>
>
>
> void jit_csvs_stream(t_jit_csvs *x, long offset)
> {
>     long mrk_id = 5, tmplong;
>     t_mrk *m = NULL;
>     long argc = 5;
>     t_atom argv[5];
>     t_atom tmpatom;
>
>     if(flags.pos_mx_isset)
>     {
>         m = jit_csvs_get_marker(x, mrk_id, offset);
>         if(m)
>         {
>             jit_atom_setlong(argv, offset);
>             jit_atom_setlong(argv+1, mrk_id);
>             jit_atom_setfloat(argv+2,(m->x)*0.001f);
>             jit_atom_setfloat(argv+3,(m->y)*0.001f);
>             jit_atom_setfloat(argv+4,(m->z)*0.001f);
>
>          // ----------this one works fine
> ----------------------------------------------
>             jit_atom_setfloat(&tmpatom,  (float)(offset/128.f));
>             jit_object_method(x->pos_mx, gensym("setall"),0L,1,&tmpatom);
>
>          // -------- this seems to do nothing
> ------------------------------------
>             jit_object_method(x->pos_mx, gensym("setcell"),0L,argc,argv);
>
>         }else{
>             post("marker pointer null");
>         }
>     }
> }
>

On Oct 30, 2006, at 12:35 PM, Thijs Koerselman wrote:

> After some further investigation I drew the conclusion that you 

> probably have to access the matrix data directly. Although the 

> scalebias example is still a bit confusing to me, its in there, so 

> I guess I will figure it out. I think it would still be very 

> convenient to have an interface like setcell(). So if its available 

From your code sample, it appears that you're forgetting to use the 

"val" delimiter required by the "setcell" method to identify where 

the cell position ends and the value begins. If you're only using a 

1/2/3D matrix, you can use the setcell1d, setcell2d, setcell3d 

utility methods which don't require the val delimiter.

However, if you're going to set lots of values in your matrix, it's 

definitely faster to access the matrix data directly. You could to 

make a macro for matrix cell random access given t_jit_matrix_info 

and a data pointer. Something like the following email client code 

(might be typos) should get you going in the right direction

#define CELL_PTR_1D (info,data,x) (((uchar *)(data))+(info)->dimstride 

#define CELL_PTR_2D (info,data,x,y) (CELL_PTR_1D(info,data,x)+(info)- 

#define CELL_PTR_3D (info,data,x,y,z) (CELL_PTR_2D(info,data,x,y)+ 

f = (float *) CELL_PTR_2D(info,data,x,y);

> After some further investigation I drew the conclusion that you  
> probably have to access the matrix data directly. Although the  
> scalebias example is still a bit confusing to me, its in there, so  
> I guess I will figure it out. I think it would still be very  
> convenient to have an interface like setcell(). So if its available  
> please let me know.

 From your code sample, it appears that you're forgetting to use the  
"val" delimiter required by the "setcell" method to identify where  
the cell position ends and the value begins. If you're only using a  
1/2/3D matrix, you can use the setcell1d, setcell2d, setcell3d  
utility methods which don't require the val delimiter.

However, if you're going to set lots of values in your matrix, it's  
definitely faster to access the matrix data directly. You could to  
make a macro for matrix cell random access given t_jit_matrix_info  
and a data pointer. Something like the following email client code  
(might be typos) should get you going in the right direction

#define CELL_PTR_1D (info,data,x) (((uchar *)(data))+(info)->dimstride 
[0]*(x))
#define CELL_PTR_2D (info,data,x,y) (CELL_PTR_1D(info,data,x)+(info)- 
 >dimstride[1]*(y))
#define CELL_PTR_3D (info,data,x,y,z) (CELL_PTR_2D(info,data,x,y)+ 
(info)->dimstride[2]*(z))

f = (float *) CELL_PTR_2D(info,data,x,y);
f[0] = 0.f;
f[1] = 1/2.f;
f[2] = 666.f;

Thanks Joshua, didn't know about the delimiters. I guess I'm used to js too

much. I don't have much experience with macros, but I'll look into it.

It turned out to be quite easy to access my matrix directly. As it is always

3plane float32 2d matrix. I finally got it working, and its FAST. I replaced

a max patch using 100 % cpu with an external that uses only 3%... or maybe

> On Oct 30, 2006, at 12:35 PM, Thijs Koerselman wrote:

> > After some further investigation I drew the conclusion that you

> > probably have to access the matrix data directly. Although the

> > scalebias example is still a bit confusing to me, its in there, so

> > I guess I will figure it out. I think it would still be very

> > convenient to have an interface like setcell(). So if its available

> From your code sample, it appears that you're forgetting to use the

> "val" delimiter required by the "setcell" method to identify where

> the cell position ends and the value begins. If you're only using a

> 1/2/3D matrix, you can use the setcell1d, setcell2d, setcell3d

> utility methods which don't require the val delimiter.

> However, if you're going to set lots of values in your matrix, it's

> definitely faster to access the matrix data directly. You could to

> make a macro for matrix cell random access given t_jit_matrix_info

> and a data pointer. Something like the following email client code

> (might be typos) should get you going in the right direction

> #define CELL_PTR_1D (info,data,x) (((uchar *)(data))+(info)->dimstride

> #define CELL_PTR_2D (info,data,x,y) (CELL_PTR_1D(info,data,x)+(info)-

> #define CELL_PTR_3D (info,data,x,y,z) (CELL_PTR_2D(info,data,x,y)+

> f = (float *) CELL_PTR_2D(info,data,x,y);

Thanks Joshua, didn't know about the delimiters. I guess I'm used to js too
much. I don't have much experience with macros, but I'll look into it.

It turned out to be quite easy to access my matrix directly. As it is always
3plane float32 2d matrix. I finally got it working, and its FAST. I replaced
a max patch using 100 % cpu with an external that uses only 3%... or maybe
I'm really bad at patching;-)

On 10/30/06, Joshua Kit Clayton  wrote:
>
>
> On Oct 30, 2006, at 12:35 PM, Thijs Koerselman wrote:
>
> > After some further investigation I drew the conclusion that you
> > probably have to access the matrix data directly. Although the
> > scalebias example is still a bit confusing to me, its in there, so
> > I guess I will figure it out. I think it would still be very
> > convenient to have an interface like setcell(). So if its available
> > please let me know.
>
> From your code sample, it appears that you're forgetting to use the
> "val" delimiter required by the "setcell" method to identify where
> the cell position ends and the value begins. If you're only using a
> 1/2/3D matrix, you can use the setcell1d, setcell2d, setcell3d
> utility methods which don't require the val delimiter.
>
> However, if you're going to set lots of values in your matrix, it's
> definitely faster to access the matrix data directly. You could to
> make a macro for matrix cell random access given t_jit_matrix_info
> and a data pointer. Something like the following email client code
> (might be typos) should get you going in the right direction
>
> #define CELL_PTR_1D (info,data,x) (((uchar *)(data))+(info)->dimstride
> [0]*(x))
> #define CELL_PTR_2D (info,data,x,y) (CELL_PTR_1D(info,data,x)+(info)-
> >dimstride[1]*(y))
> #define CELL_PTR_3D (info,data,x,y,z) (CELL_PTR_2D(info,data,x,y)+
> (info)->dimstride[2]*(z))
>
> float *f;
>
> f = (float *) CELL_PTR_2D(info,data,x,y);
> f[0] = 0.f;
> f[1] = 1/2.f;
> f[2] = 666.f;
>
> -Joshua
>

setcell jit_object_method