void xfer_dobj_flag(Data_Obj *dpto, Data_Obj *dpfr, uint32_t flagbit)
{
if( OBJ_FLAGS(dpfr) & flagbit ){
SET_OBJ_FLAG_BITS(dpto, flagbit);
} else {
CLEAR_OBJ_FLAG_BITS(dpto, flagbit);
}
}
static void _update_pf_viewer(QSP_ARG_DECL Platform_Viewer *pvp, Data_Obj *dp)
{
#ifdef HAVE_OPENGL
int t;
//cudaError_t e;
// unmap buffer before using w/ GL
if( BUF_IS_MAPPED(dp) ){
if( (*PF_UNMAPBUF_FN(PFDEV_PLATFORM(OBJ_PFDEV(dp))))
(QSP_ARG dp) < 0 ) {
warn("update_pf_viewer: buffer unmap error!?");
}
CLEAR_OBJ_FLAG_BITS(dp, DT_BUF_MAPPED);
// propagate change to children and parents
propagate_flag(dp,DT_BUF_MAPPED);
}
glClear(GL_COLOR_BUFFER_BIT);
glBindTexture(GL_TEXTURE_2D, OBJ_TEX_ID(dp));
// is glBindBuffer REALLY part of libGLEW???
//#ifdef HAVE_LIBGLEW
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, OBJ_BUF_ID(dp));
//#endif // HAVE_LIBGLEW
t=gl_pixel_type(dp);
glTexSubImage2D(GL_TEXTURE_2D, 0, // target, level
0, 0, // x0, y0
OBJ_COLS(dp), OBJ_ROWS(dp), // dx, dy
t,
GL_UNSIGNED_BYTE, // type
OFFSET(0)); // offset into PIXEL_UNPACK_BUFFER
//#ifdef HAVE_LIBGLEW
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
//#endif // HAVE_LIBGLEW
glBegin(GL_QUADS);
glTexCoord2f(0, 1); glVertex2f(-1.0, -1.0);
glTexCoord2f(0, 0); glVertex2f(-1.0, 1.0);
glTexCoord2f(1, 0); glVertex2f(1.0, 1.0);
glTexCoord2f(1, 1); glVertex2f(1.0, -1.0);
glEnd();
glBindTexture(GL_TEXTURE_2D, 0);
if( (*PF_MAPBUF_FN(PFDEV_PLATFORM(OBJ_PFDEV(dp))))(QSP_ARG dp) < 0 ){
warn("update_pf_viewer: Error mapping buffer!?");
}
SET_OBJ_FLAG_BITS(dp, DT_BUF_MAPPED);
// propagate change to children and parents
propagate_flag(dp,DT_BUF_MAPPED);
#else // ! HAVE_OPENGL
NO_OGL_MSG
#endif // ! HAVE_OPENGL
}
int enlarge(QSP_ARG_DECL Data_Obj *big_dp,Data_Obj *lil_dp) /* reduce into lil_dp */
{
if( change_size(QSP_ARG big_dp,lil_dp) < 0 )
return(-1);
SET_OBJ_FLAG_BITS(big_dp, DT_ASSIGNED);
return(0);
}
static COMMAND_FUNC( do_protect )
{
Data_Obj *dp;
dp=pick_obj("");
if( dp == NULL ) return;
if( IS_STATIC(dp) ){
sprintf(ERROR_STRING,"do_protect: Object %s is already static!?",OBJ_NAME(dp));
warn(ERROR_STRING);
return;
}
SET_OBJ_FLAG_BITS(dp,DT_STATIC);
}
static void cuda_display_finish(QSP_ARG_DECL Data_Obj *dp)
{
cudaError_t e;
// re-map so we can use again with CUDA
// BUG? Is it safe to do this before the call to swap_buffers???
//cutilSafeCall(cudaGLMapBufferObject( &OBJ_DATA_PTR(dp), OBJ_BUF_ID(dp) ));
e = cudaGLMapBufferObject( &OBJ_DATA_PTR(dp), OBJ_BUF_ID(dp) );
if( e != cudaSuccess ){
WARN("Error mapping buffer object!?");
// should we return now, with possibly other cleanup???
}
SET_OBJ_FLAG_BITS(dp, DT_BUF_MAPPED);
// propagate change to children and parents
propagate_flag(dp,DT_BUF_MAPPED);
}
static void _make_zombie(QSP_ARG_DECL Data_Obj *dp)
{
static int n_zombie=1;
char zname[LLEN];
/* this function removes it from the hash table and
* active item list, doesn't add the structure to the
* item free list. This is why we release the node
* and later free the object...
*
* I think the idea here is that there might be a dangling reference
* to an image - for instance, if we display an image, then delete it,
* and then later want to refresh the window...
*/
zombie_item(dobj_itp,(Item *)dp);
sprintf(zname,"Z.%s.%d",OBJ_NAME(dp),n_zombie++);
fprintf(stderr,"make_zombie, changing object %s to %s\n",OBJ_NAME(dp),zname);
rls_str( (char *) OBJ_NAME(dp) ); /* unsave old name, make_zombie */
SET_OBJ_NAME(dp,savestr(zname));
SET_OBJ_FLAG_BITS(dp,DT_ZOMBIE);
} // make_zombie
// This is the normal display path
static void update_pf_viewer(QSP_ARG_DECL Platform_Viewer *pvp, Data_Obj *dp)
{
#ifdef HAVE_OPENGL
int t;
//cudaError_t e;
// unmap buffer before using w/ GL
if( BUF_IS_MAPPED(dp) ){
if( (*PF_UNMAPBUF_FN(PFDEV_PLATFORM(OBJ_PFDEV(dp))))
(QSP_ARG dp) < 0 ) {
WARN("update_pf_viewer: buffer unmap error!?");
}
#ifdef FOOBAR
e = cudaGLUnmapBufferObject( OBJ_BUF_ID(dp) );
if( e != cudaSuccess ){
describe_cuda_driver_error2("update_pf_viewer",
"cudaGLUnmapBufferObject",e);
NERROR1("failed to unmap buffer object");
}
#endif // FOOBAR
CLEAR_OBJ_FLAG_BITS(dp, DT_BUF_MAPPED);
// propagate change to children and parents
propagate_flag(dp,DT_BUF_MAPPED);
}
//
//bind_texture(OBJ_DATA_PTR(dp));
glClear(GL_COLOR_BUFFER_BIT);
/*
sprintf(ERROR_STRING,"update_pf_viewer: tex_id = %d, buf_id = %d",
OBJ_TEX_ID(dp),OBJ_BUF_ID(dp));
advise(ERROR_STRING);
*/
glBindTexture(GL_TEXTURE_2D, OBJ_TEX_ID(dp));
// is glBindBuffer REALLY part of libGLEW???
//#ifdef HAVE_LIBGLEW
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, OBJ_BUF_ID(dp));
//#endif // HAVE_LIBGLEW
#ifdef FOOBAR
switch(OBJ_COMPS(dp)){
/* what used to be here??? */
}
#endif /* FOOBAR */
t=gl_pixel_type(dp);
glTexSubImage2D(GL_TEXTURE_2D, 0, // target, level
0, 0, // x0, y0
OBJ_COLS(dp), OBJ_ROWS(dp), // dx, dy
t,
GL_UNSIGNED_BYTE, // type
OFFSET(0)); // offset into PIXEL_UNPACK_BUFFER
//#ifdef HAVE_LIBGLEW
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
//#endif // HAVE_LIBGLEW
glBegin(GL_QUADS);
glTexCoord2f(0, 1); glVertex2f(-1.0, -1.0);
glTexCoord2f(0, 0); glVertex2f(-1.0, 1.0);
glTexCoord2f(1, 0); glVertex2f(1.0, 1.0);
glTexCoord2f(1, 1); glVertex2f(1.0, -1.0);
glEnd();
glBindTexture(GL_TEXTURE_2D, 0);
#ifdef FOOBAR
e = cudaGLMapBufferObject( &OBJ_DATA_PTR(dp), OBJ_BUF_ID(dp) );
if( e != cudaSuccess ){
WARN("Error mapping buffer object!?");
// should we return now, with possibly other cleanup???
}
#endif // FOOBAR
if( (*PF_MAPBUF_FN(PFDEV_PLATFORM(OBJ_PFDEV(dp))))(QSP_ARG dp) < 0 ){
WARN("update_pf_viewer: Error mapping buffer!?");
}
SET_OBJ_FLAG_BITS(dp, DT_BUF_MAPPED);
// propagate change to children and parents
propagate_flag(dp,DT_BUF_MAPPED);
#else // ! HAVE_OPENGL
NO_OGL_MSG
#endif // ! HAVE_OPENGL
}
static COMMAND_FUNC( do_new_gl_buffer )
{
const char *s;
Data_Obj *dp;
Platform_Device *pdp;
Compute_Platform *cdp;
dimension_t d,w,h;
#ifdef HAVE_OPENGL
Dimension_Set ds;
int t;
#endif // HAVE_OPENGL
s = NAMEOF("name for GL buffer object");
cdp = pick_platform("platform");
if( cdp != NULL )
push_pfdev_context(QSP_ARG PF_CONTEXT(cdp) );
pdp = pick_pfdev("device");
if( cdp != NULL )
pop_pfdev_context(SINGLE_QSP_ARG);
w = (int)HOW_MANY("width");
h = (int)HOW_MANY("height");
d = (int)HOW_MANY("depth");
/* what should the depth be??? default to 1 for now... */
if( pdp == NULL ) return;
/* Make sure this name isn't already in use... */
dp = dobj_of(s);
if( dp != NULL ){
sprintf(ERROR_STRING,"Data object name '%s' is already in use, can't use for GL buffer object.",s);
warn(ERROR_STRING);
return;
}
#ifdef HAVE_OPENGL
// BUG need to be able to set the cuda device.
// Note, however, that we don't need GL buffers on the Tesla...
//set_data_area(cuda_data_area[0][0]);
set_data_area( PFDEV_AREA(pdp,PFDEV_GLOBAL_AREA_INDEX) );
ds.ds_dimension[0]=d;
ds.ds_dimension[1]=w;
ds.ds_dimension[2]=h;
ds.ds_dimension[3]=1;
ds.ds_dimension[4]=1;
dp = _make_dp(QSP_ARG s,&ds,PREC_FOR_CODE(PREC_UBY));
if( dp == NULL ){
sprintf(ERROR_STRING,
"Error creating data_obj header for %s",s);
error1(ERROR_STRING);
}
SET_OBJ_FLAG_BITS(dp, DT_NO_DATA); /* can't free this data */
SET_OBJ_FLAG_BITS(dp, DT_GL_BUF); /* indicate obj is a GL buffer */
SET_OBJ_DATA_PTR(dp, NULL);
//fprintf(stderr,"do_new_gl_buffer: allocating gl_info for %s\n",OBJ_NAME(dp));
SET_OBJ_GL_INFO(dp, (GL_Info *) getbuf( sizeof(GL_Info) ) );
//fprintf(stderr,"do_new_gl_buffer: DONE allocating gl_info for %s\n",OBJ_NAME(dp));
glew_check(SINGLE_QSP_ARG); /* without this, we get a segmentation
* violation on glGenBuffers???
*/
// We need an extra field in which to store the GL identifier...
// AND another extra field in which to store the associated texid.
// Why is this ifdef here? These don't seem to depend
// on libglew???
// Answer: We need libglew to bring in openGL extensions like glBindBuffer...
//advise("calling glGenBuffers");
//fprintf(stderr,"OBJ_GL_INFO(%s) = 0x%lx\n",OBJ_NAME(dp),(long)OBJ_GL_INFO(dp));
//fprintf(stderr,"OBJ_BUF_ID_P(%s) = 0x%lx\n",OBJ_NAME(dp),(long)OBJ_BUF_ID_P(dp));
// BUG glGenBuffers seems to require v1.5???
glGenBuffers(1, OBJ_BUF_ID_P(dp) ); // first arg is # buffers to generate?
//sprintf(ERROR_STRING,"glGenBuffers gave us buf_id = %d",OBJ_BUF_ID(dp));
//advise(ERROR_STRING);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, OBJ_BUF_ID(dp) );
// glBufferData will allocate the memory for the buffer,
// but won't copy unless the pointer is non-null
// How do we get the gpu memory space address?
// That must be with map
glBufferData(GL_PIXEL_UNPACK_BUFFER,
OBJ_COMPS(dp) * OBJ_COLS(dp) * OBJ_ROWS(dp), NULL, GL_STREAM_DRAW);
/* buffer arg set to 0 unbinds any previously bound buffers...
* and restores client memory usage.
*/
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
//#endif // HAVE_LIBGLEW
glGenTextures(1, OBJ_TEX_ID_P(dp) ); // makes a texture name
fprintf(stderr,"new_gl_buffer: new texture name is 0x%x\n",OBJ_TEX_ID(dp));
glBindTexture(GL_TEXTURE_2D, OBJ_TEX_ID(dp) );
t = gl_pixel_type(dp);
glTexImage2D( GL_TEXTURE_2D,
0, // level-of-detail - is this the same as miplevel???
OBJ_COMPS(dp), // internal format, can also be symbolic constant such as
// GL_RGBA etc
OBJ_COLS(dp), // width - must be 2^n+2 (border) for some n???
OBJ_ROWS(dp), // height - must be 2^m+2 (border) for some m???
0, // border - must be 0 or 1
t, // format of pixel data
GL_UNSIGNED_BYTE, // type of pixel data
NULL // pixel data - null pointer means
// allocate but do not copy?
// - offset into PIXEL_UNPACK_BUFFER??
);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
// Why was this here? It would seem to un-bind the target???
glBindTexture(GL_TEXTURE_2D, 0);
//glFinish(); // necessary or not?
//advise("calling platform-specific buffer registration function");
if( (*PF_REGBUF_FN(PFDEV_PLATFORM(pdp)))( QSP_ARG dp ) < 0 ){
WARN("do_new_gl_buffer: Error in platform-specific buffer registration!?");
// BUG? - should clean up here!
}
// Leave the buffer mapped by default
//cutilSafeCall(cudaGLMapBufferObject( &OBJ_DATA_PTR(dp), OBJ_BUF_ID(dp) ));
//advise("calling platform-specific buffer mapping function");
if( (*PF_MAPBUF_FN(PFDEV_PLATFORM(pdp)))( QSP_ARG dp ) < 0 ){
WARN("do_new_gl_buffer: Error in platform-specific buffer mapping!?");
// BUG? - should clean up here!
}
SET_OBJ_FLAG_BITS(dp, DT_BUF_MAPPED);
// propagate change to children and parents
propagate_flag(dp,DT_BUF_MAPPED);
#else // ! HAVE_OPENGL
NO_OGL_MSG
#endif // ! HAVE_OPENGL
} /* end do_new_gl_buffer */
static int change_size(QSP_ARG_DECL Data_Obj *dst_dp,Data_Obj *src_dp )
{
Dimension_Set ef, *enlargement_factor=&ef;
Dimension_Set rf, *reduction_factor=&rf;
Vec_Obj_Args oa1, *oap=&oa1;
Dimension_Set size_ds, n_ds;
Dimension_Set *size_dsp=(&size_ds), *n_dsp=(&n_ds);
Data_Obj *src_ss_dp, *dst_ss_dp;
dimension_t i,j,k,l,m;
index_t offsets[N_DIMENSIONS]={0,0,0,0,0};
incr_t dst_incrs[N_DIMENSIONS], src_incrs[N_DIMENSIONS];
index_t dst_indices[N_DIMENSIONS]={0,0,0,0,0}, src_indices[N_DIMENSIONS]={0,0,0,0,0};
/* For simplicity, we don't allow size changes to be combined with conversions */
if( !dp_same_prec(QSP_ARG dst_dp,src_dp,"change_size") )
return(-1);
for(i=0;i<N_DIMENSIONS;i++){
if( OBJ_TYPE_DIM(dst_dp,i) > OBJ_TYPE_DIM(src_dp,i) ){
/* enlargement - subsample the destination */
SET_DIMENSION(enlargement_factor,i,
floor( OBJ_TYPE_DIM(dst_dp,i) / OBJ_TYPE_DIM(src_dp,i) ) );
SET_DIMENSION(reduction_factor,i, 0);
SET_DIMENSION(size_dsp,i, OBJ_TYPE_DIM(src_dp,i) );
SET_DIMENSION(n_dsp,i, DIMENSION(enlargement_factor,i) );
dst_incrs[i] = DIMENSION(n_dsp,i);
src_incrs[i] = 1;
} else {
/* reduction - subsample the source */
SET_DIMENSION(reduction_factor,i,
ceil( OBJ_TYPE_DIM(src_dp,i) / OBJ_TYPE_DIM(dst_dp,i) ) );
SET_DIMENSION(enlargement_factor,i, 0 );
SET_DIMENSION(size_dsp,i, floor( OBJ_TYPE_DIM(src_dp,i) /
DIMENSION(reduction_factor,i) ) );
/* We don't need to do this multiple times, just pick one and do it */
/*SET_DIMENSION(n_dsp,i, DIMENSION(reduction_factor,i) ); */
SET_DIMENSION(n_dsp,i, 1);
src_incrs[i] = DIMENSION(reduction_factor,i);
dst_incrs[i] = 1;
}
}
/* make the subsamples.
* the column increment is expressed in columns, etc.
*/
dst_ss_dp=make_subsamp(QSP_ARG "chngsize_dst_obj",dst_dp,size_dsp,offsets,dst_incrs);
src_ss_dp=make_subsamp(QSP_ARG "chngsize_src_obj",src_dp,size_dsp,offsets,src_incrs);
clear_obj_args(oap);
SET_OA_DEST(oap,dst_ss_dp);
SET_OA_SRC_OBJ(oap,0, src_ss_dp);
SET_OA_ARGSTYPE(oap, REAL_ARGS);
SET_OA_PFDEV(oap,OBJ_PFDEV(dst_dp));
for(i=0;i<DIMENSION(n_dsp,4);i++){ /* foreach sequence to copy */
if( dst_incrs[4] > 1 )
dst_indices[4]=i;
else src_indices[4]=i;
for(j=0;j<DIMENSION(n_dsp,3);j++){ /* foreach frame to copy */
if( dst_incrs[3] > 1 )
dst_indices[3]=j;
else src_indices[3]=j;
for(k=0;k<DIMENSION(n_dsp,2);k++){ /* foreach row */
if( dst_incrs[2] > 1 )
dst_indices[2]=k;
else src_indices[2]=k;
for(l=0;l<DIMENSION(n_dsp,1);l++){ /* foreach col */
if( dst_incrs[1] > 1 )
dst_indices[1]=l;
else src_indices[1]=l;
for(m=0;m<DIMENSION(n_dsp,0);m++){ /* foreach comp */
if( dst_incrs[0] > 1 )
dst_indices[0]=m;
else src_indices[0]=m;
/* relocate the appropriate subsample */
SET_OBJ_DATA_PTR(dst_ss_dp, multiply_indexed_data(dst_dp,dst_indices) );
SET_OBJ_DATA_PTR(src_ss_dp, multiply_indexed_data(src_dp,src_indices) );
// This doesn't check for cuda obj...
//vmov(oap);
perf_vfunc(QSP_ARG FVMOV, oap );
}
}
}
}
}
delvec(QSP_ARG dst_ss_dp);
delvec(QSP_ARG src_ss_dp);
SET_OBJ_FLAG_BITS(dst_dp, DT_ASSIGNED);
return(0);
}
static Data_Obj *insure_ram_obj(QSP_ARG_DECL Data_Obj *dp)
{
Data_Obj *tmp_dp;
char *tname;
Data_Area *save_ap;
Data_Obj *c_dp=NULL;
if( OBJ_IS_RAM(dp) ) return dp;
// This object lives on a different platform.
// We create a copy in RAM, and download the data
// using the platform download function.
save_ap = curr_ap;
curr_ap = ram_area_p;
tname = getbuf( strlen(OBJ_NAME(dp)) + strlen(DNAME_PREFIX) + 1 );
sprintf(tname,"%s%s",DNAME_PREFIX,OBJ_NAME(dp));
tmp_dp = dup_obj(QSP_ARG dp, tname);
givbuf(tname);
if( tmp_dp == NO_OBJ ){
// This can happen if the object is subscripted,
// as the bracket characters are illegal in names
return NO_OBJ;
}
curr_ap = save_ap;
// We can't download if the source data is not contiguous...
//
// We have a problem with bit precision, because the bits can
// be non-contiguous when the long words are - any time the number of columns
// is not evenly divided by the bits-per-word
if( (! IS_CONTIGUOUS(dp)) && ! HAS_CONTIGUOUS_DATA(dp) ){
Vec_Obj_Args oa1, *oap=&oa1;
advise("object is not contiguous, and does not have contiguous data...");
longlist(QSP_ARG dp);
save_ap = curr_ap;
curr_ap = OBJ_AREA( dp );
tname = getbuf( strlen(OBJ_NAME(dp)) + strlen(CNAME_PREFIX) + 1 );
sprintf(tname,"%s%s",CNAME_PREFIX,OBJ_NAME(dp));
c_dp = dup_obj(QSP_ARG dp, tname );
givbuf(tname);
curr_ap = save_ap;
// Now do the move...
setvarg2(oap,c_dp,dp);
if( IS_BITMAP(dp) ){
SET_OA_SBM(oap,dp);
SET_OA_SRC1(oap,NO_OBJ);
}
if( IS_REAL(dp) ) /* BUG case for QUAT too? */
OA_ARGSTYPE(oap) = REAL_ARGS;
else if( IS_COMPLEX(dp) ) /* BUG case for QUAT too? */
OA_ARGSTYPE(oap) = COMPLEX_ARGS;
else if( IS_QUAT(dp) ) /* BUG case for QUAT too? */
OA_ARGSTYPE(oap) = QUATERNION_ARGS;
else
//ERROR1("CAUTIOUS: insure_ram_obj: bad argset type!?");
assert( AERROR("insure_ram_obj: bad argset type!?") );
//fprintf(stderr,"insure_ram_obj: moving remote data to a contiguous object\n");
call_vfunc( QSP_ARG FIND_VEC_FUNC(FVMOV), oap );
//fprintf(stderr,"insure_ram_obj: DONE moving remote data to a contiguous object\n");
dp = c_dp;
}
gen_obj_dnload(QSP_ARG tmp_dp, dp);
if( c_dp != NO_OBJ )
delvec(QSP_ARG c_dp);
// BUG - when to delete?
// We try using the VOLATILE flag. This will work as long as
// the input object is not VOLATILE!?
SET_OBJ_FLAG_BITS(tmp_dp, DT_VOLATILE ) ;
return tmp_dp;
}
