bool ktx_texture::consistency_check() const
{
if (!check_header())
return false;
uint32_t block_dim = 0, bytes_per_block = 0;
if ((!m_header.m_glType) || (!m_header.m_glFormat))
{
if ((m_header.m_glType) || (m_header.m_glFormat))
return false;
if (!ktx_get_ogl_fmt_desc(m_header.m_glInternalFormat, m_header.m_glType, block_dim, bytes_per_block))
return false;
if (block_dim == 1)
return false;
//if ((get_width() % block_dim) || (get_height() % block_dim))
// return false;
}
else
{
if (!ktx_get_ogl_fmt_desc(m_header.m_glFormat, m_header.m_glType, block_dim, bytes_per_block))
return false;
if (block_dim > 1)
return false;
}
if ((m_block_dim != block_dim) || (m_bytes_per_block != bytes_per_block))
return false;
uint32_t total_expected_images = get_total_images();
if (m_image_data.size() != total_expected_images)
return false;
for (uint32_t mip_level = 0; mip_level < get_num_mips(); mip_level++)
{
uint32_t mip_width, mip_height, mip_depth;
get_mip_dim(mip_level, mip_width, mip_height, mip_depth);
const uint32_t mip_row_blocks = (mip_width + m_block_dim - 1) / m_block_dim;
const uint32_t mip_col_blocks = (mip_height + m_block_dim - 1) / m_block_dim;
if ((!mip_row_blocks) || (!mip_col_blocks))
return false;
for (uint32_t array_element = 0; array_element < get_array_size(); array_element++)
{
for (uint32_t face = 0; face < get_num_faces(); face++)
{
for (uint32_t zslice = 0; zslice < mip_depth; zslice++)
{
const uint8_vec &image_data = get_image_data(get_image_index(mip_level, array_element, face, zslice));
uint32_t expected_image_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
if (image_data.size() != expected_image_size)
return false;
}
}
}
}
return true;
}
bool ktx_texture::check_header() const
{
if (((get_num_faces() != 1) && (get_num_faces() != 6)) || (!m_header.m_pixelWidth))
return false;
if ((!m_header.m_pixelHeight) && (m_header.m_pixelDepth))
return false;
if ((get_num_faces() == 6) && ((m_header.m_pixelDepth) || (!m_header.m_pixelHeight)))
return false;
#if 0
if (m_header.m_numberOfMipmapLevels)
{
const uint32_t max_mipmap_dimension = 1U << (m_header.m_numberOfMipmapLevels - 1U);
if (max_mipmap_dimension > (VOGL_MAX(VOGL_MAX(m_header.m_pixelWidth, m_header.m_pixelHeight), m_header.m_pixelDepth)))
return false;
}
#endif
return true;
}
uint32_t ktx_texture::get_total_images() const
{
if (!is_valid() || !get_num_mips())
return 0;
// bogus:
//return get_num_mips() * (get_depth() * get_num_faces() * get_array_size());
// Naive algorithm, could just compute based off the # of mips
uint32_t max_index = 0;
for (uint32_t mip_level = 0; mip_level < get_num_mips(); mip_level++)
{
uint32_t total_zslices = math::maximum<uint32_t>(get_depth() >> mip_level, 1U);
uint32_t index = get_image_index(mip_level, get_array_size() - 1, get_num_faces() - 1, total_zslices - 1);
max_index = math::maximum<uint32_t>(max_index, index);
}
return max_index + 1;
}
void
build_elem_elem(Exo_DB *exo)
{
int ce;
int count;
int e;
int ebi;
int elem;
int ename;
int face;
//int his_dim, her_dim;
int i, j;
int index;
int len_curr;
int len_prev;
int len_intr;
int length, length_new, num_faces;
int iel;
int ioffset;
int n;
int neighbor_name = -1;
int node;
int num_elem_sides;
int num_nodes;
int snl[MAX_NODES_PER_SIDE]; /* Side Node List - NOT Saturday Night Live! */
char err_msg[MAX_CHAR_ERR_MSG];
/*
* Integer arrays used to find intersection sets of node->element lists.
*/
int prev_set[MAX_EPN]; /* list of elements attached to previous node*/
int curr_set[MAX_EPN]; /* list of elements attached to "this" node */
int interset[MAX_EPN]; /* values of hits between */
int ip[MAX_EPN]; /* indeces of hits for prev_set[] */
int ic[MAX_EPN]; /* indeces of hits for curr_set[] */
/*
* If the element->node and node->element connectivities have not been
* built, then we won't be able to do this task.
*/
if ( ! exo->elem_node_conn_exists ||
! exo->node_elem_conn_exists )
{
EH(-1, "Build elem->node before node->elem.");
return;
}
/*
* The number of elements connected via conventional faces may be deduced
* from the number of elements and their type.
*/
exo->elem_elem_pntr = (int *) smalloc((exo->num_elems+1)*sizeof(int));
length = 0;
for ( i=0; i<exo->num_elem_blocks; i++)
{
length += exo->eb_num_elems[i] * get_num_faces(exo->eb_elem_type[i]);
}
exo->elem_elem_list = (int *) smalloc(length*sizeof(int));
/*
* Initialize...
*/
for ( i=0; i<length; i++)
{
exo->elem_elem_list[i] = UNASSIGNED_YET;
}
/*
elem = 0;
for ( ebi=0; ebi<exo->num_elem_blocks; ebi++)
{
num_elem_sides = get_num_faces(exo->eb_elem_type[ebi]);
for ( e=0; e<exo->eb_num_elems[ebi]; e++)
{
exo->elem_elem_pntr[elem] = count;
elem++;
count += num_elem_sides;
}
}
*/
/*
* Walk through the elements, block by block.
*/
count = 0;
elem = 0;
for ( ebi=0; ebi<exo->num_elem_blocks; ebi++)
{
num_elem_sides = get_num_faces(exo->eb_elem_type[ebi]);
for ( e=0; e<exo->eb_num_elems[ebi]; e++,elem++)
{
exo->elem_elem_pntr[elem] = count;
count += num_elem_sides;
/*
* Look at each side of the element, collecting a unique
* list of integers corresponding to the minimum number of nodes
* needed to identify an entire side.
*
* Typically, the same number of nodes as space dimensions are
* needed, with exceptions being the various "sides" of shells,
* beams and trusses...
*/
for ( face=0; face<num_elem_sides; face++)
{
/*
* Given the element and the face construct the
* list of node numbers that determine that side.
*/
/*
* Later, we might not need *all* the nodes on a side,
* particularly for high order elements. It may suffice
* to check only as many nodes as space dimensions that
* the element lives in...
*/
num_nodes = build_side_node_list(elem, face, ebi, exo, snl);
#ifdef DEBUG
fprintf(stderr, "Elem %d, face %d has %d nodes: ", elem, face,
num_nodes);
for ( i=0; i<num_nodes; i++)
{
fprintf(stderr, " %d", snl[i]);
}
fprintf(stderr, "\n");
#endif /* DEBUG */
/*
* Cross check: for each node in the side there is a list
* of elements connected to it. Beginning with all the
* elements connected to the first node (except for this given
* element), cross check with all the elements connected with
* the 2nd node to build an intersection set of one element.
*/
for ( i=0; i<MAX_EPN; i++)
{
prev_set[i] = -1;
curr_set[i] = -1;
interset[i] = -1;
}
len_prev = 0;
len_curr = 0;
len_intr = 0;
for ( n=0; n<num_nodes; n++)
{
/*
* Copy this node's element list into a clean "curr_set" array
* that will be intersected with any previously gathered
* lists of elements that qualify as promiscuously in
* contact with nodes...
*/
node = snl[n];
for ( i=0; i<MAX_EPN; i++)
{
curr_set[i] = -1;
}
len_curr = 0;
#ifdef DEBUG
fprintf(stderr, "Traversing n->e connectivity of node %d\n",
node);
#endif /* DEBUG */
for ( ce=exo->node_elem_pntr[node];
ce<exo->node_elem_pntr[node+1]; ce++)
{
ename = exo->node_elem_list[ce];
#ifdef DEBUG
fprintf(stderr, "\telem %d\n", ename);
#endif /* DEBUG */
/*
* Go ahead and accumulate the self element name
* just as a consistency check....
*/
/*
if ( ename != e )
{
}
*/
/*
* PKN: The current Goma use of ->elem_elem...
* is such that this connectivity should list
* connections like QUAD-BAR or HEX-SHELL.
* So, I'll add this dimension matching conditional
*/
/* PRS (Summer 2012): Need to change this for shell stacks which
have the same dim*/
/* We need however to consider a special case (as of 8/30/2012
* this is a SHELL-on-SHELL stack. Viz. two materials, each a shell material
* which share not a side but a face. Since faces of shells are sides
* in patran speak, we need some special logic. We need to avoid adding
* the friend shell element (neighboring material) to the current shell element
* even though each material has the same number of sides.
* Here goes (BTW, I cannot find max-nodes-per-element anywhere!!!!)
*/
int shell_on_shell = 0; int flippy_flop = 0;
int nbr_ebid; int nbr_num_elem_sides;
nbr_ebid = fence_post(ename, exo->eb_ptr,
exo->num_elem_blocks+1);
EH(nbr_ebid, "Bad element block ID!");
nbr_num_elem_sides = get_num_faces(exo->eb_elem_type[nbr_ebid]);
shell_on_shell = 0;
flippy_flop = 0;
if (exo->eb_id[ebi] < 100 && exo->eb_id[nbr_ebid] >= 100) flippy_flop=1;
if (exo->eb_id[ebi] >= 100 && exo->eb_id[nbr_ebid] < 100) flippy_flop=1;
if ((nbr_ebid != ebi) &&
(strstr(exo->eb_elem_type[nbr_ebid], "SHELL")) &&
(strstr(exo->eb_elem_type[ebi], "SHELL")) &&
flippy_flop) shell_on_shell = 1;
// his_dim = elem_info(NDIM, exo->eb_elem_itype[ebi]);
// her_dim = elem_info(NDIM, exo->eb_elem_itype[exo->elem_eb[ename]]);
// if( his_dim == her_dim )
if (nbr_num_elem_sides == num_elem_sides && !shell_on_shell)
{
curr_set[len_curr] = ename;
len_curr++;
}
}
/*
* The first node is special - we'll just compare
* it with itself by making the "previous set" just the
* same as the current set...
*/
if ( n == 0 )
{
for ( i=0; i<MAX_EPN; i++)
{
prev_set[i] = curr_set[i];
}
len_prev = len_curr;
}
#ifdef DEBUG
fprintf(stderr, "\ncurr_set: ");
for ( i=0; i<len_curr; i++)
{
fprintf(stderr, "%d ", curr_set[i]);
}
fprintf(stderr, "\nprev_set: ");
for ( i=0; i<len_prev; i++)
{
fprintf(stderr, "%d ", prev_set[i]);
}
#endif /* DEBUG */
/*
* First, clean the intersection list and the list of
* hit indeces in the previous and current lists.
*
* Then find the intersection of the previous and current
* sets of elements attached to the previous and current
* nodes...
*/
for ( i=0; i<MAX_EPN; i++)
{
interset[i] = -1;
ip[i] = -1;
ic[i] = -1;
}
len_intr = 0;
len_intr = int_intersect(prev_set, curr_set, len_prev,
len_curr, ip, ic);
#ifdef DEBUG
fprintf(stderr, "num_hits = %d\n", len_intr);
#endif /* DEBUG */
/*
* Now, let's make the intersection set the next previous
* set of elements, a standard for comparison. We should
* eventually boil down to either one or zero elements
* that qualify...
*/
for ( i=0; i<MAX_EPN; i++)
{
prev_set[i] = -1;
}
for ( i=0; i<len_intr; i++)
{
prev_set[i] = curr_set[ic[i]];
}
len_prev = len_intr;
}
#ifdef DEBUG
fprintf(stderr, "Element [%d], face [%d], local_node [%d]\n",
elem, face, n);
fprintf(stderr, "Intersection set length = %d\n", len_intr);
#endif /* DEBUG */
/*
* Now consider the different cases.
*/
if ( len_intr == 2 )
{
/*
* The boiled list contains self and one other element.
*/
if ( prev_set[0] == elem )
{
neighbor_name = prev_set[1];
}
else
{
neighbor_name = prev_set[0];
if ( prev_set[1] != elem )
{
sr = sprintf(err_msg,
"2 elems ( %d %d ) 1 should be %d!",
prev_set[0], prev_set[1], elem);
EH(-1, err_msg);
}
}
}
else if ( len_intr == 1 && prev_set[0] == elem )
{
/*
* The boiled list has one member, this element.
*
* The face must connect either to outer space or to
* another processor.
*/
if ( Num_Proc == 1 )
{
neighbor_name = -1;
}
else
{
neighbor_name = -1;
/*
* I am going to punt for now. Later, revisit this
* condition and insert code to check for neighbor
* processors containing all the same face nodes.
*
* EH(-1, "Not done yet...");
*
*/
/*
* Check if ALL the nodes on this face belong
* to another processors list of nodes. I.e., the
* node must all be in the external node list of
* and belong to the same external processor.
*/
}
}
/*
* Pathological cases that normally should not occur....
*/
else if ( len_intr == 0 )
{
sr = sprintf(err_msg, "Elem %d, face %d should self contain!",
elem, face);
EH(-1, err_msg);
}
else if ( len_intr == 1 && prev_set[0] != elem )
{
sr = sprintf(err_msg,
"Elem %d, face %d only connects with elem %d ?",
elem, face, prev_set[0]);
EH(-1, err_msg);
}
else
{
sr = sprintf(err_msg,
"Unknown elem-elem connection elem %d, face %d, len_intr=%d",
elem, face, len_intr);
WH(-1, err_msg);
}
/*
* Now we know how to assign the neighbor name for this face
* of the element.
*/
index = exo->elem_elem_pntr[elem] + face;
exo->elem_elem_list[index] = neighbor_name;
} /* end face loop this elem */
} /* end elem loop this elemblock */
} /* end elem block loop */
exo->elem_elem_pntr[exo->num_elems] = count; /* last fencepost */
exo->elem_elem_conn_exists = TRUE;
if (Linear_Solver == FRONT)
{
/*
* Now that we have elem_elem_pntr and elem_elem_list for our parallel
* world, we are going to use them also for optimal element bandwidth
* reduction ordering. We will use METIS, but METIS requires the CSR
* format, which is compressed, viz. we need to remove the -1s. Here
* we go
*/
/* First check for the assumption that all blocks have same number of
element faces. Stop if they don't and issue an error to the next
aspiring developer */
for ( i=0; i<exo->num_elem_blocks; i++)
{
if(get_num_faces(exo->eb_elem_type[0]) !=
get_num_faces(exo->eb_elem_type[i]) )
{
EH(-1,"Stop! We cannot reorder these elements with METIS with elemement type changes");
}
}
/* Now begin */
exo->elem_elem_xadj = (int *) smalloc((exo->num_elems+1)*sizeof(int));
/*initialize */
for(e=0; e<exo->num_elems+1 ; e++)
{
exo->elem_elem_xadj[e] = exo->elem_elem_pntr[e];
}
/* Recompute length of adjacency list by removing external edges */
length_new = 0;
for (i = 0; i < length; i++) {
if(exo->elem_elem_list[i] != -1) length_new++;
}
exo->elem_elem_adjncy = alloc_int_1(length_new, -1);
/* Now convert */
ioffset=0;
for(iel = 0; iel < exo->num_elems; iel++)
{
/* Big assumption here that all blocks have the same */
/* element type. Can be furbished later since this is */
/* just for the frontal solver */
num_faces = get_num_faces(exo->eb_elem_type[0]);
for (i= iel*num_faces; i < (iel+1)*num_faces; i++)
{
j = i - ioffset;
if(exo->elem_elem_list[i] == -1)
{
ioffset++;
for(e=iel +1; e <exo->num_elems+1; e++)exo->elem_elem_xadj[e]--;
}
else
{
exo->elem_elem_adjncy[j] = exo->elem_elem_list[i];
}
}
}
/* convert to Fortran style */
for(e=0; e<exo->num_elems+1 ; e++) exo->elem_elem_xadj[e]++;
for ( i=0; i<length_new; i++) exo->elem_elem_adjncy[i]++;
} /* End FRONTAL_SOLVER if */
/*
* Verification that every element/face has assigned something besides
* the initial default value of "unassigned".
*
* For your convenience - FORTRAN 1-based numbering.
*/
#ifdef DEBUG
for ( e=0; e<exo->num_elems; e++)
{
fprintf(stdout, "Elem %3d:", e+1);
for ( ce=exo->elem_elem_pntr[e]; ce<exo->elem_elem_pntr[e+1]; ce++)
{
if ( exo->elem_elem_list[ce] == -1 )
{
fprintf(stdout, " spc");
}
else if ( exo->elem_elem_list[ce] < -1 )
{
fprintf(stdout, " prc");
}
else
{
fprintf(stdout, " %3d", exo->elem_elem_list[ce] + 1);
}
if ( exo->elem_elem_list[ce] == UNASSIGNED_YET )
{
sr = sprintf(err_msg,
"You need to plug a leak at elem (%d) face (%d)",
exo->elem_elem_list[ce] + 1,
ce - exo->elem_elem_pntr[e] + 1);
EH(-1, err_msg);
}
}
fprintf(stdout, "\n");
}
#endif /* DEBUG */
#if FALSE
demo_elem_elem_conn(exo);
#endif
return;
}
void do_physics_align_object( object * obj )
{
vms_vector desired_upvec;
fixang delta_ang,roll_ang;
//vms_vector forvec = {0,0,f1_0};
vms_matrix temp_matrix;
fix d,largest_d=-f1_0;
int i,best_side;
best_side=0;
// bank player according to segment orientation
//find side of segment that player is most alligned with
for (i=0;i<6;i++) {
#ifdef COMPACT_SEGS
vms_vector _tv1;
get_side_normal( &Segments[obj->segnum], i, 0, &_tv1 );
d = vm_vec_dot(&_tv1,&obj->orient.uvec);
#else
d = vm_vec_dot(&Segments[obj->segnum].sides[i].normals[0],&obj->orient.uvec);
#endif
if (d > largest_d) {largest_d = d; best_side=i;}
}
if (floor_levelling) {
// old way: used floor's normal as upvec
#ifdef COMPACT_SEGS
get_side_normal(&Segments[obj->segnum], 3, 0, &desired_upvec );
#else
desired_upvec = Segments[obj->segnum].sides[3].normals[0];
#endif
}
else // new player leveling code: use normal of side closest to our up vec
if (get_num_faces(&Segments[obj->segnum].sides[best_side])==2) {
#ifdef COMPACT_SEGS
vms_vector normals[2];
get_side_normals(&Segments[obj->segnum], best_side, &normals[0], &normals[1] );
desired_upvec.x = (normals[0].x + normals[1].x) / 2;
desired_upvec.y = (normals[0].y + normals[1].y) / 2;
desired_upvec.z = (normals[0].z + normals[1].z) / 2;
vm_vec_normalize(&desired_upvec);
#else
side *s = &Segments[obj->segnum].sides[best_side];
desired_upvec.x = (s->normals[0].x + s->normals[1].x) / 2;
desired_upvec.y = (s->normals[0].y + s->normals[1].y) / 2;
desired_upvec.z = (s->normals[0].z + s->normals[1].z) / 2;
vm_vec_normalize(&desired_upvec);
#endif
}
else
#ifdef COMPACT_SEGS
get_side_normal(&Segments[obj->segnum], best_side, 0, &desired_upvec );
#else
desired_upvec = Segments[obj->segnum].sides[best_side].normals[0];
#endif
if (labs(vm_vec_dot(&desired_upvec,&obj->orient.fvec)) < f1_0/2) {
vms_angvec tangles;
vm_vector_2_matrix(&temp_matrix,&obj->orient.fvec,&desired_upvec,NULL);
delta_ang = vm_vec_delta_ang(&obj->orient.uvec,&temp_matrix.uvec,&obj->orient.fvec);
delta_ang += obj->mtype.phys_info.turnroll;
if (abs(delta_ang) > DAMP_ANG) {
vms_matrix rotmat, new_pm;
roll_ang = fixmul(FrameTime,ROLL_RATE);
if (abs(delta_ang) < roll_ang) roll_ang = delta_ang;
else if (delta_ang<0) roll_ang = -roll_ang;
tangles.p = tangles.h = 0; tangles.b = roll_ang;
vm_angles_2_matrix(&rotmat,&tangles);
vm_matrix_x_matrix(&new_pm,&obj->orient,&rotmat);
obj->orient = new_pm;
}
else floor_levelling=0;
}
}
bool ktx_texture::write_to_stream(data_stream_serializer &serializer, bool no_keyvalue_data) const
{
if (!consistency_check())
{
VOGL_ASSERT_ALWAYS;
return false;
}
memcpy(m_header.m_identifier, s_ktx_file_id, sizeof(m_header.m_identifier));
m_header.m_endianness = m_opposite_endianness ? KTX_OPPOSITE_ENDIAN : KTX_ENDIAN;
if (m_block_dim == 1)
{
m_header.m_glTypeSize = ktx_get_ogl_type_size(m_header.m_glType);
m_header.m_glBaseInternalFormat = m_header.m_glFormat;
}
else
{
m_header.m_glBaseInternalFormat = ktx_get_ogl_compressed_base_internal_fmt(m_header.m_glInternalFormat);
}
m_header.m_bytesOfKeyValueData = 0;
if (!no_keyvalue_data)
{
for (uint32_t i = 0; i < m_key_values.size(); i++)
m_header.m_bytesOfKeyValueData += sizeof(uint32_t) + ((m_key_values[i].size() + 3) & ~3);
}
if (m_opposite_endianness)
m_header.endian_swap();
bool success = (serializer.write(&m_header, sizeof(m_header), 1) == 1);
if (m_opposite_endianness)
m_header.endian_swap();
if (!success)
return success;
uint32_t total_key_value_bytes = 0;
const uint8_t padding[3] = { 0, 0, 0 };
if (!no_keyvalue_data)
{
for (uint32_t i = 0; i < m_key_values.size(); i++)
{
uint32_t key_value_size = m_key_values[i].size();
if (m_opposite_endianness)
key_value_size = utils::swap32(key_value_size);
success = (serializer.write(&key_value_size, sizeof(key_value_size), 1) == 1);
total_key_value_bytes += sizeof(key_value_size);
if (m_opposite_endianness)
key_value_size = utils::swap32(key_value_size);
if (!success)
return false;
if (key_value_size)
{
if (serializer.write(&m_key_values[i][0], key_value_size, 1) != 1)
return false;
total_key_value_bytes += key_value_size;
uint32_t num_padding = 3 - ((key_value_size + 3) % 4);
if ((num_padding) && (serializer.write(padding, num_padding, 1) != 1))
return false;
total_key_value_bytes += num_padding;
}
}
(void)total_key_value_bytes;
}
VOGL_ASSERT(total_key_value_bytes == m_header.m_bytesOfKeyValueData);
for (uint32_t mip_level = 0; mip_level < get_num_mips(); mip_level++)
{
uint32_t mip_width, mip_height, mip_depth;
get_mip_dim(mip_level, mip_width, mip_height, mip_depth);
const uint32_t mip_row_blocks = (mip_width + m_block_dim - 1) / m_block_dim;
const uint32_t mip_col_blocks = (mip_height + m_block_dim - 1) / m_block_dim;
if ((!mip_row_blocks) || (!mip_col_blocks))
return false;
uint32_t image_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
if ((m_header.m_numberOfArrayElements) || (get_num_faces() == 1))
image_size *= (get_array_size() * get_num_faces() * mip_depth);
if (!image_size)
{
VOGL_ASSERT_ALWAYS;
return false;
}
if (m_opposite_endianness)
image_size = utils::swap32(image_size);
success = (serializer.write(&image_size, sizeof(image_size), 1) == 1);
if (m_opposite_endianness)
image_size = utils::swap32(image_size);
if (!success)
return false;
uint32_t total_mip_size = 0;
uint32_t total_image_data_size = 0;
if ((!m_header.m_numberOfArrayElements) && (get_num_faces() == 6))
{
// plain non-array cubemap
for (uint32_t face = 0; face < get_num_faces(); face++)
{
const uint8_vec &image_data = get_image_data(get_image_index(mip_level, 0, face, 0));
if ((!image_data.size()) || (image_data.size() != image_size))
return false;
if (m_opposite_endianness)
{
uint8_vec tmp_image_data(image_data);
utils::endian_swap_mem(&tmp_image_data[0], tmp_image_data.size(), m_header.m_glTypeSize);
if (serializer.write(&tmp_image_data[0], tmp_image_data.size(), 1) != 1)
return false;
}
else if (serializer.write(&image_data[0], image_data.size(), 1) != 1)
return false;
// Not +=, but =, because of the silly image_size plain cubemap exception in the KTX file format
total_image_data_size = image_data.size();
uint32_t num_cube_pad_bytes = 3 - ((image_data.size() + 3) % 4);
if ((num_cube_pad_bytes) && (serializer.write(padding, num_cube_pad_bytes, 1) != 1))
return false;
total_mip_size += image_size + num_cube_pad_bytes;
}
}
else
{
// 1D, 2D, 3D (normal or array texture), or array cubemap
for (uint32_t array_element = 0; array_element < get_array_size(); array_element++)
{
for (uint32_t face = 0; face < get_num_faces(); face++)
{
for (uint32_t zslice = 0; zslice < mip_depth; zslice++)
{
const uint8_vec &image_data = get_image_data(get_image_index(mip_level, array_element, face, zslice));
if (!image_data.size())
return false;
if (m_opposite_endianness)
{
uint8_vec tmp_image_data(image_data);
utils::endian_swap_mem(&tmp_image_data[0], tmp_image_data.size(), m_header.m_glTypeSize);
if (serializer.write(&tmp_image_data[0], tmp_image_data.size(), 1) != 1)
return false;
}
else if (serializer.write(&image_data[0], image_data.size(), 1) != 1)
return false;
total_image_data_size += image_data.size();
total_mip_size += image_data.size();
}
}
}
uint32_t num_mip_pad_bytes = 3 - ((total_mip_size + 3) % 4);
if ((num_mip_pad_bytes) && (serializer.write(padding, num_mip_pad_bytes, 1) != 1))
return false;
total_mip_size += num_mip_pad_bytes;
}
VOGL_ASSERT((total_mip_size & 3) == 0);
VOGL_ASSERT(total_image_data_size == image_size);
}
return true;
}
bool ktx_texture::read_from_stream(data_stream_serializer &serializer)
{
clear();
// Read header
if (serializer.read(&m_header, 1, sizeof(m_header)) != sizeof(ktx_header))
return false;
// Check header
if (memcmp(s_ktx_file_id, m_header.m_identifier, sizeof(m_header.m_identifier)))
return false;
if ((m_header.m_endianness != KTX_OPPOSITE_ENDIAN) && (m_header.m_endianness != KTX_ENDIAN))
return false;
m_opposite_endianness = (m_header.m_endianness == KTX_OPPOSITE_ENDIAN);
if (m_opposite_endianness)
{
m_header.endian_swap();
if ((m_header.m_glTypeSize != sizeof(uint8_t)) && (m_header.m_glTypeSize != sizeof(uint16_t)) && (m_header.m_glTypeSize != sizeof(uint32_t)))
return false;
}
if (!check_header())
return false;
if (!compute_pixel_info())
{
#if VOGL_KTX_PVRTEX_WORKAROUNDS
// rg [9/10/13] - moved this check into here, instead of in compute_pixel_info(), but need to retest it.
if ((!m_header.m_glInternalFormat) && (!m_header.m_glType) && (!m_header.m_glTypeSize) && (!m_header.m_glBaseInternalFormat))
{
// PVRTexTool writes bogus headers when outputting ETC1.
console::warning("ktx_texture::compute_pixel_info: Header doesn't specify any format, assuming ETC1 and hoping for the best\n");
m_header.m_glBaseInternalFormat = KTX_RGB;
m_header.m_glInternalFormat = KTX_ETC1_RGB8_OES;
m_header.m_glTypeSize = 1;
m_block_dim = 4;
m_bytes_per_block = 8;
}
else
#endif
return false;
}
uint8_t pad_bytes[3];
// Read the key value entries
uint32_t num_key_value_bytes_remaining = m_header.m_bytesOfKeyValueData;
while (num_key_value_bytes_remaining)
{
if (num_key_value_bytes_remaining < sizeof(uint32_t))
return false;
uint32_t key_value_byte_size;
if (serializer.read(&key_value_byte_size, 1, sizeof(uint32_t)) != sizeof(uint32_t))
return false;
num_key_value_bytes_remaining -= sizeof(uint32_t);
if (m_opposite_endianness)
key_value_byte_size = utils::swap32(key_value_byte_size);
if (key_value_byte_size > num_key_value_bytes_remaining)
return false;
uint8_vec key_value_data;
if (key_value_byte_size)
{
key_value_data.resize(key_value_byte_size);
if (serializer.read(&key_value_data[0], 1, key_value_byte_size) != key_value_byte_size)
return false;
}
m_key_values.push_back(key_value_data);
uint32_t padding = 3 - ((key_value_byte_size + 3) % 4);
if (padding)
{
if (serializer.read(pad_bytes, 1, padding) != padding)
return false;
}
num_key_value_bytes_remaining -= key_value_byte_size;
if (num_key_value_bytes_remaining < padding)
return false;
num_key_value_bytes_remaining -= padding;
}
// Now read the mip levels
uint32_t total_faces = get_num_mips() * get_array_size() * get_num_faces() * get_depth();
if ((!total_faces) || (total_faces > 65535))
return false;
// See Section 2.8 of KTX file format: No rounding to block sizes should be applied for block compressed textures.
// OK, I'm going to break that rule otherwise KTX can only store a subset of textures that DDS can handle for no good reason.
#if 0
const uint32_t mip0_row_blocks = m_header.m_pixelWidth / m_block_dim;
const uint32_t mip0_col_blocks = VOGL_MAX(1, m_header.m_pixelHeight) / m_block_dim;
#else
const uint32_t mip0_row_blocks = (m_header.m_pixelWidth + m_block_dim - 1) / m_block_dim;
const uint32_t mip0_col_blocks = (VOGL_MAX(1, m_header.m_pixelHeight) + m_block_dim - 1) / m_block_dim;
#endif
if ((!mip0_row_blocks) || (!mip0_col_blocks))
return false;
const uint32_t mip0_depth = VOGL_MAX(1, m_header.m_pixelDepth);
VOGL_NOTE_UNUSED(mip0_depth);
bool has_valid_image_size_fields = true;
bool disable_mip_and_cubemap_padding = false;
#if VOGL_KTX_PVRTEX_WORKAROUNDS
{
// PVRTexTool has a bogus KTX writer that doesn't write any imageSize fields. Nice.
size_t expected_bytes_remaining = 0;
for (uint32_t mip_level = 0; mip_level < get_num_mips(); mip_level++)
{
uint32_t mip_width, mip_height, mip_depth;
get_mip_dim(mip_level, mip_width, mip_height, mip_depth);
const uint32_t mip_row_blocks = (mip_width + m_block_dim - 1) / m_block_dim;
const uint32_t mip_col_blocks = (mip_height + m_block_dim - 1) / m_block_dim;
if ((!mip_row_blocks) || (!mip_col_blocks))
return false;
expected_bytes_remaining += sizeof(uint32_t);
if ((!m_header.m_numberOfArrayElements) && (get_num_faces() == 6))
{
for (uint32_t face = 0; face < get_num_faces(); face++)
{
uint32_t slice_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
expected_bytes_remaining += slice_size;
uint32_t num_cube_pad_bytes = 3 - ((slice_size + 3) % 4);
expected_bytes_remaining += num_cube_pad_bytes;
}
}
else
{
uint32_t total_mip_size = 0;
for (uint32_t array_element = 0; array_element < get_array_size(); array_element++)
{
for (uint32_t face = 0; face < get_num_faces(); face++)
{
for (uint32_t zslice = 0; zslice < mip_depth; zslice++)
{
uint32_t slice_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
total_mip_size += slice_size;
}
}
}
expected_bytes_remaining += total_mip_size;
uint32_t num_mip_pad_bytes = 3 - ((total_mip_size + 3) % 4);
expected_bytes_remaining += num_mip_pad_bytes;
}
}
if (serializer.get_stream()->get_remaining() < expected_bytes_remaining)
{
has_valid_image_size_fields = false;
disable_mip_and_cubemap_padding = true;
console::warning("ktx_texture::read_from_stream: KTX file size is smaller than expected - trying to read anyway without imageSize fields\n");
}
}
#endif
for (uint32_t mip_level = 0; mip_level < get_num_mips(); mip_level++)
{
uint32_t mip_width, mip_height, mip_depth;
get_mip_dim(mip_level, mip_width, mip_height, mip_depth);
const uint32_t mip_row_blocks = (mip_width + m_block_dim - 1) / m_block_dim;
const uint32_t mip_col_blocks = (mip_height + m_block_dim - 1) / m_block_dim;
if ((!mip_row_blocks) || (!mip_col_blocks))
return false;
uint32_t image_size = 0;
if (!has_valid_image_size_fields)
{
if ((!m_header.m_numberOfArrayElements) && (get_num_faces() == 6))
{
// The KTX file format has an exception for plain cubemap textures, argh.
image_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
}
else
{
image_size = mip_depth * mip_row_blocks * mip_col_blocks * m_bytes_per_block * get_array_size() * get_num_faces();
}
}
else
{
if (serializer.read(&image_size, 1, sizeof(image_size)) != sizeof(image_size))
return false;
if (m_opposite_endianness)
image_size = utils::swap32(image_size);
}
if (!image_size)
return false;
uint32_t total_mip_size = 0;
// The KTX file format has an exception for plain cubemap textures, argh.
if ((!m_header.m_numberOfArrayElements) && (get_num_faces() == 6))
{
// plain non-array cubemap
for (uint32_t face = 0; face < get_num_faces(); face++)
{
VOGL_ASSERT(m_image_data.size() == get_image_index(mip_level, 0, face, 0));
m_image_data.push_back(uint8_vec());
uint8_vec &image_data = m_image_data.back();
image_data.resize(image_size);
if (serializer.read(&image_data[0], 1, image_size) != image_size)
return false;
if (m_opposite_endianness)
utils::endian_swap_mem(&image_data[0], image_size, m_header.m_glTypeSize);
uint32_t num_cube_pad_bytes = disable_mip_and_cubemap_padding ? 0 : (3 - ((image_size + 3) % 4));
if (serializer.read(pad_bytes, 1, num_cube_pad_bytes) != num_cube_pad_bytes)
return false;
total_mip_size += image_size + num_cube_pad_bytes;
}
}
else
{
uint32_t num_image_bytes_remaining = image_size;
// 1D, 2D, 3D (normal or array texture), or array cubemap
for (uint32_t array_element = 0; array_element < get_array_size(); array_element++)
{
for (uint32_t face = 0; face < get_num_faces(); face++)
{
for (uint32_t zslice = 0; zslice < mip_depth; zslice++)
{
uint32_t slice_size = mip_row_blocks * mip_col_blocks * m_bytes_per_block;
if ((!slice_size) || (slice_size > num_image_bytes_remaining))
return false;
uint32_t image_index = get_image_index(mip_level, array_element, face, zslice);
m_image_data.ensure_element_is_valid(image_index);
uint8_vec &image_data = m_image_data[image_index];
image_data.resize(slice_size);
if (serializer.read(&image_data[0], 1, slice_size) != slice_size)
return false;
if (m_opposite_endianness)
utils::endian_swap_mem(&image_data[0], slice_size, m_header.m_glTypeSize);
num_image_bytes_remaining -= slice_size;
total_mip_size += slice_size;
}
}
}
if (num_image_bytes_remaining)
{
VOGL_ASSERT_ALWAYS;
return false;
}
}
uint32_t num_mip_pad_bytes = disable_mip_and_cubemap_padding ? 0 : (3 - ((total_mip_size + 3) % 4));
if (serializer.read(pad_bytes, 1, num_mip_pad_bytes) != num_mip_pad_bytes)
return false;
}
return true;
}
bool ktx_texture::operator==(const ktx_texture &rhs) const
{
if (this == &rhs)
return true;
// This is not super deep because I want to avoid poking around into internal state (such as the header)
#define CMP(x) \
if (x != rhs.x) \
return false;
CMP(get_ogl_internal_fmt());
CMP(get_width());
CMP(get_height());
CMP(get_depth());
CMP(get_num_mips());
CMP(get_array_size());
CMP(get_num_faces());
CMP(is_compressed());
CMP(get_block_dim());
// The image fmt/type shouldn't matter with compressed textures.
if (!is_compressed())
{
CMP(get_ogl_fmt());
CMP(get_ogl_type());
}
CMP(get_total_images());
CMP(get_opposite_endianness());
// Do an order insensitive key/value comparison.
dynamic_string_array lhs_keys;
get_keys(lhs_keys);
dynamic_string_array rhs_keys;
rhs.get_keys(rhs_keys);
if (lhs_keys.size() != rhs_keys.size())
return false;
lhs_keys.sort(dynamic_string_less_than_case_sensitive());
rhs_keys.sort(dynamic_string_less_than_case_sensitive());
for (uint32_t i = 0; i < lhs_keys.size(); i++)
if (lhs_keys[i].compare(rhs_keys[i], true) != 0)
return false;
for (uint32_t i = 0; i < lhs_keys.size(); i++)
{
uint8_vec lhs_data, rhs_data;
if (!get_key_value_data(lhs_keys[i].get_ptr(), lhs_data))
return false;
if (!get_key_value_data(lhs_keys[i].get_ptr(), rhs_data))
return false;
if (lhs_data != rhs_data)
return false;
}
// Compare images.
for (uint32_t l = 0; l < get_num_mips(); l++)
{
for (uint32_t a = 0; a < get_array_size(); a++)
{
for (uint32_t f = 0; f < get_num_faces(); f++)
{
for (uint32_t z = 0; z < get_depth(); z++)
{
const uint8_vec &lhs_img = get_image_data(l, a, f, z);
const uint8_vec &rhs_img = rhs.get_image_data(l, a, f, z);
if (lhs_img != rhs_img)
return false;
}
}
}
}
#undef CMP
return true;
}
