void
genFillTileWithNAN(struct KgenContext *ctx, const Tile *tile)
{
char tmp[1024];
Kstring elem;
unsigned int incRows, incCols;
unsigned int i, j, v;
if (!tile->trans) {
incRows = 1;
v = incCols = umin(tile->vecLen, tile->nrCols);
}
else {
v = incRows = umin(tile->vecLen, tile->nrRows);
incCols = 1;
}
for (i = 0; i < tile->nrRows; i += incRows) {
for (j = 0; j < tile->nrCols; j += incCols) {
sprintfTileElement(&elem, tile, i, j, v);
sprintf(tmp, "%s = NAN;\n", elem.buf);
kgenAddStmt(ctx, tmp);
}
}
kgenAddBlankLine(ctx);
}
////////////////////////////////////////////////////////////////////////////////
// Merge step 1: find sample ranks in each segment
////////////////////////////////////////////////////////////////////////////////
static void generateSampleRanks(
uint *ranksA,
uint *ranksB,
uint *srcKey,
uint stride,
uint N,
uint sortDir
){
uint lastSegmentElements = N % (2 * stride);
uint sampleCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE) : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
for(uint pos = 0; pos < sampleCount; pos++){
const uint i = pos & ( (stride / SAMPLE_STRIDE) - 1 );
const uint segmentBase = (pos - i) * (2 * SAMPLE_STRIDE);
const uint lenA = stride;
const uint lenB = umin(stride, N - segmentBase - stride);
const uint nA = stride / SAMPLE_STRIDE;
const uint nB = getSampleCount(lenB);
if(i < nA){
ranksA[(segmentBase + 0) / SAMPLE_STRIDE + i] = i * SAMPLE_STRIDE;
ranksB[(segmentBase + 0) / SAMPLE_STRIDE + i] = binarySearchExclusive(srcKey[segmentBase + i * SAMPLE_STRIDE], srcKey + segmentBase + stride, lenB, sortDir);
}
if(i < nB){
ranksB[(segmentBase + stride) / SAMPLE_STRIDE + i] = i * SAMPLE_STRIDE;
ranksA[(segmentBase + stride) / SAMPLE_STRIDE + i] = binarySearchInclusive(srcKey[segmentBase + stride + i * SAMPLE_STRIDE], srcKey + segmentBase, lenA, sortDir);
}
}
}
bool BitmapFloat_copy_linear_over_srgb(Context * context, BitmapFloat * src, const uint32_t from_row, BitmapBgra * dest, const uint32_t dest_row, const uint32_t row_count, const uint32_t from_col, const uint32_t col_count, const bool transpose)
{
const uint32_t dest_bytes_pp = BitmapPixelFormat_bytes_per_pixel (dest->fmt);
const uint32_t dest_row_stride = transpose ? dest_bytes_pp : dest->stride;
const uint32_t dest_pixel_stride = transpose ? dest->stride : dest_bytes_pp;
const uint32_t srcitems = umin(from_col + col_count, src->w) *src->channels;
const uint32_t ch = src->channels;
const bool copy_alpha = dest->fmt == Bgra32 && src->channels == 4 && src->alpha_meaningful;
const bool clean_alpha = !copy_alpha && dest->fmt == Bgra32;
for (uint32_t row = 0; row < row_count; row++) {
float * src_row = src->pixels + (row + from_row) * src->float_stride;
uint8_t * dest_row_bytes = dest->pixels + (dest_row + row) * dest_row_stride + (from_col * dest_pixel_stride);
for (uint32_t ix = from_col * ch; ix < srcitems; ix += ch) {
dest_row_bytes[0] = Context_floatspace_to_srgb (context, src_row[ix]);
dest_row_bytes[1] = Context_floatspace_to_srgb (context, src_row[ix + 1]);
dest_row_bytes[2] = Context_floatspace_to_srgb (context, src_row[ix + 2]);
if (copy_alpha) {
dest_row_bytes[3] = uchar_clamp_ff (src_row[ix + 3] * 255.0f);
}
if (clean_alpha) {
dest_row_bytes[3] = 0xff;
}
dest_row_bytes += dest_pixel_stride;
}
}
return true;
}
////////////////////////////////////////////////////////////////////////////////
// Merge step 2: merge ranks and indices to derive elementary intervals
////////////////////////////////////////////////////////////////////////////////
static void mergeRanksAndIndices(
uint *limits,
uint *ranks,
uint stride,
uint N
){
uint lastSegmentElements = N % (2 * stride);
uint sampleCount = (lastSegmentElements > stride) ? (N + 2 * stride - lastSegmentElements) / (2 * SAMPLE_STRIDE) : (N - lastSegmentElements) / (2 * SAMPLE_STRIDE);
for(uint pos = 0; pos < sampleCount; pos++){
const uint i = pos & ( (stride / SAMPLE_STRIDE) - 1 );
const uint segmentBase = (pos - i) * (2 * SAMPLE_STRIDE);
const uint lenA = stride;
const uint lenB = umin(stride, N - segmentBase - stride);
const uint nA = stride / SAMPLE_STRIDE;
const uint nB = getSampleCount(lenB);
if(i < nA){
uint dstPosA = binarySearchExclusive(ranks[(segmentBase + 0) / SAMPLE_STRIDE + i], ranks + (segmentBase + stride) / SAMPLE_STRIDE, nB, 1) + i;
assert( dstPosA < nA + nB );
limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] = ranks[(segmentBase + 0) / SAMPLE_STRIDE + i];
}
if(i < nB){
uint dstPosA = binarySearchInclusive(ranks[(segmentBase + stride) / SAMPLE_STRIDE + i], ranks + (segmentBase + 0) / SAMPLE_STRIDE, nA, 1) + i;
assert( dstPosA < nA + nB );
limits[(segmentBase / SAMPLE_STRIDE) + dstPosA] = ranks[(segmentBase + stride) / SAMPLE_STRIDE + i];
}
}
}
unsigned int
tileLineSegmentLen(const Tile *tile)
{
unsigned int pitch;
unsigned int len;
pitch = tilePitch(tile);
len = umin(pitch, tile->vecLen);
if (tile->trans) {
len = umin(len, tile->nrRows);
}
else {
len = umin(len, tile->nrCols);
}
return len;
}
/*
* Common line segment length of 2 tiles being arguments in tile multiplication
*/
static unsigned int
commonTileSegmentLen(const Tile *tile1, const Tile *tile2)
{
unsigned int u1, u2;
u1 = tileLineSegmentLen(tile1);
u2 = tileLineSegmentLen(tile2);
return umin(u1, u2);
}
static NTSTATUS DDKAPI finaread(PDEVICE_OBJECT d, PIRP i) {
NTSTATUS status = STATUS_SUCCESS;
PIO_STACK_LOCATION sl = IoGetCurrentIrpStackLocation(i);
unsigned sz = umin(sl->Parameters.Read.Length, g_outcurr);
RtlCopyMemory(i->AssociatedIrp.SystemBuffer, g_outbuf, sz);
g_outcurr = 0;
i->IoStatus.Status = status;
i->IoStatus.Information = sz;
IoCompleteRequest(i, IO_NO_INCREMENT);
return status;
}
static size_t r_stats( struct uufile *f, void *dest, size_t bytes)
{
struct kernel_stats out;
errno_t rc = get_stats_record( f->pos++, &out );
if( rc )
return 0;
int nc = umin( bytes, sizeof(out) );
memcpy( dest, &out, nc );
return nc;
}
// TODO are we sure we can give it all out?
static size_t r_threads( struct uufile *f, void *dest, size_t bytes)
{
phantom_thread_t out;
tid_t t = get_next_tid( f->pos, &out);
f->pos = t;
if( t < 0 )
return 0;
int nc = umin( bytes, sizeof(out) );
memcpy( dest, &out, nc );
return nc;
}
static uint binarySearchExclusive(uint val, uint *data, uint L, uint sortDir){
if(L == 0)
return 0;
uint pos = 0;
for(uint stride = nextPowerOfTwo(L); stride > 0; stride >>= 1){
uint newPos = umin(pos + stride, L);
if( (sortDir && (data[newPos - 1] < val)) || (!sortDir && (data[newPos - 1] > val)) )
pos = newPos;
}
return pos;
}
static bool BitmapFloat_compose_linear_over_srgb(Context * context, BitmapFloat * src, const uint32_t from_row, BitmapBgra * dest, const uint32_t dest_row, const uint32_t row_count, const uint32_t from_col, const uint32_t col_count, const bool transpose)
{
const uint32_t dest_bytes_pp = BitmapPixelFormat_bytes_per_pixel (dest->fmt);
const uint32_t dest_row_stride = transpose ? dest_bytes_pp : dest->stride;
const uint32_t dest_pixel_stride = transpose ? dest->stride : dest_bytes_pp;
const uint32_t srcitems = umin(from_col + col_count, src->w) *src->channels;
const uint32_t ch = src->channels;
const bool dest_alpha = dest->fmt == Bgra32 && dest->alpha_meaningful;
const uint8_t dest_alpha_index = dest_alpha ? 3 : 0;
const float dest_alpha_to_float_coeff = dest_alpha ? 1.0f / 255.0f : 0.0f;
const float dest_alpha_to_float_offset = dest_alpha ? 0.0f : 1.0f;
for (uint32_t row = 0; row < row_count; row++) {
//const float * const __restrict src_row = src->pixels + (row + from_row) * src->float_stride;
float * src_row = src->pixels + (row + from_row) * src->float_stride;
uint8_t * dest_row_bytes = dest->pixels + (dest_row + row) * dest_row_stride + (from_col * dest_pixel_stride);
for (uint32_t ix = from_col * ch; ix < srcitems; ix += ch) {
const uint8_t dest_b = dest_row_bytes[0];
const uint8_t dest_g = dest_row_bytes[1];
const uint8_t dest_r = dest_row_bytes[2];
const uint8_t dest_a = dest_row_bytes[dest_alpha_index];
const float src_b = src_row[ix + 0];
const float src_g = src_row[ix + 1];
const float src_r = src_row[ix + 2];
const float src_a = src_row[ix + 3];
const float a = (1.0f - src_a) * (dest_alpha_to_float_coeff * dest_a + dest_alpha_to_float_offset);
const float b = Context_srgb_to_floatspace (context, dest_b) * a + src_b;
const float g = Context_srgb_to_floatspace (context, dest_g) * a + src_g;
const float r = Context_srgb_to_floatspace (context, dest_r) * a + src_r;
const float final_alpha = src_a + a;
dest_row_bytes[0] = Context_floatspace_to_srgb (context, b / final_alpha);
dest_row_bytes[1] = Context_floatspace_to_srgb (context,g / final_alpha);
dest_row_bytes[2] = Context_floatspace_to_srgb (context,r / final_alpha);
if (dest_alpha) {
dest_row_bytes[3] = uchar_clamp_ff(final_alpha * 255);
}
dest_row_bytes += dest_pixel_stride;
}
}
return true;
}
//==============================================================================
bool LRBSpline2D::overlaps(Element2D *el) const
//==============================================================================
{
// Does it make sense to include equality?
if (el->umin() >= umax())
return false;
if (el->umax() <= umin())
return false;
if (el->vmin() >= vmax())
return false;
if (el->vmax() <= vmin())
return false;
return true;
}
//==============================================================================
bool LRBSpline2D::overlaps(double domain[]) const
//==============================================================================
{
// Does it make sense to include equality?
if (domain[0] >= umax())
return false;
if (domain[1] <= umin())
return false;
if (domain[2] >= vmax())
return false;
if (domain[3] <= vmin())
return false;
return true;
}
static void mergeElementaryIntervals(
uint *dstKey,
uint *dstVal,
uint *srcKey,
uint *srcVal,
uint *limitsA,
uint *limitsB,
uint stride,
uint N,
uint sortDir
){
uint lastSegmentElements = N % (2 * stride);
uint mergePairs = (lastSegmentElements > stride) ? getSampleCount(N) : (N - lastSegmentElements) / SAMPLE_STRIDE;
for(uint pos = 0; pos < mergePairs; pos++){
uint i = pos & ( (2 * stride) / SAMPLE_STRIDE - 1 );
uint segmentBase = (pos - i) * SAMPLE_STRIDE;
const uint lenA = stride;
const uint lenB = umin(stride, N - segmentBase - stride);
const uint nA = stride / SAMPLE_STRIDE;
const uint nB = getSampleCount(lenB);
const uint n = nA + nB;
const uint startPosA = limitsA[pos];
const uint endPosA = (i + 1 < n) ? limitsA[pos + 1] : lenA;
const uint startPosB = limitsB[pos];
const uint endPosB = (i + 1 < n) ? limitsB[pos + 1] : lenB;
const uint startPosDst = startPosA + startPosB;
assert( startPosA <= endPosA && endPosA <= lenA);
assert( startPosB <= endPosB && endPosB <= lenB);
assert( (endPosA - startPosA) <= SAMPLE_STRIDE);
assert( (endPosB - startPosB) <= SAMPLE_STRIDE);
merge(
dstKey + segmentBase + startPosDst,
dstVal + segmentBase + startPosDst,
(srcKey + segmentBase + 0) + startPosA,
(srcVal + segmentBase + 0) + startPosA,
(srcKey + segmentBase + stride) + startPosB,
(srcVal + segmentBase + stride) + startPosB,
endPosA - startPosA,
endPosB - startPosB,
sortDir
);
}
}
bool BitmapBgra_convert_srgb_to_linear(Context * context, BitmapBgra * src, uint32_t from_row, BitmapFloat * dest, uint32_t dest_row, uint32_t row_count)
{
if (src->w != dest->w || BitmapPixelFormat_bytes_per_pixel(src->fmt) < dest->channels) {
CONTEXT_error(context, Invalid_internal_state);
return false;
}
if (!(from_row + row_count <= src->h && dest_row + row_count <= dest->h)) {
CONTEXT_error(context, Invalid_internal_state);
return false;
}
const uint32_t w = src->w;
const uint32_t units = w * BitmapPixelFormat_bytes_per_pixel(src->fmt);
const uint32_t from_step = BitmapPixelFormat_bytes_per_pixel(src->fmt);
const uint32_t to_step = dest->channels;
const uint32_t copy_step = umin(from_step, to_step);
for (uint32_t row = 0; row < row_count; row++) {
uint8_t* src_start = src->pixels + (from_row + row)*src->stride;
float* buf = dest->pixels + (dest->float_stride * (row + dest_row));
if (copy_step == 3) {
for (uint32_t to_x = 0, bix = 0; bix < units; to_x += to_step, bix += from_step) {
buf[to_x] = Context_srgb_to_floatspace(context, src_start[bix]);
buf[to_x + 1] = Context_srgb_to_floatspace (context, src_start[bix + 1]);
buf[to_x + 2] = Context_srgb_to_floatspace (context, src_start[bix + 2]);
}
//We're only working on a portion... dest->alpha_premultiplied = false;
} else if (copy_step == 4) {
for (uint32_t to_x = 0, bix = 0; bix < units; to_x += to_step, bix += from_step) {
{
const float alpha = ((float)src_start[bix + 3]) / 255.0f;
buf[to_x] = alpha * Context_srgb_to_floatspace (context, src_start[bix]);
buf[to_x + 1] = alpha * Context_srgb_to_floatspace (context, src_start[bix + 1]);
buf[to_x + 2] = alpha * Context_srgb_to_floatspace (context, src_start[bix + 2]);
buf[to_x + 3] = alpha;
}
}
//We're only working on a portion... dest->alpha_premultiplied = true;
} else {
CONTEXT_error (context, Unsupported_pixel_format);
return false;
}
}
return true;
}
int _GrViDrvVESAsetVirtualSize(GrVideoMode *md,int w,int h,GrVideoMode *result)
{
Int86Regs r;
sttzero(&r);
IREG_AX(r) = VESA_FUNC + VESA_SCAN_LNLEN;
IREG_BX(r) = 0;
IREG_CX(r) = w;
int10(&r);
if(IREG_AX(r) == VESA_SUCCESS) {
result->lineoffset = IREG_BX(r);
result->width = IREG_CX(r);
result->height = umin(IREG_DX(r),h);
return(TRUE);
}
return(FALSE);
}
bool BitmapFloat_scale_rows(Context * context, BitmapFloat * from, uint32_t from_row, BitmapFloat * to, uint32_t to_row, uint32_t row_count, PixelContributions * weights)
{
const uint32_t from_step = from->channels;
const uint32_t to_step = to->channels;
const uint32_t dest_buffer_count = to->w;
const uint32_t min_channels = umin(from_step, to_step);
uint32_t ndx;
if (min_channels > 4) {
CONTEXT_error(context, Invalid_internal_state);
return false;
}
float avg[4];
// if both have alpha, process it
if (from_step == 4 && to_step == 4) {
for (uint32_t row = 0; row < row_count; row++) {
const float* __restrict source_buffer = from->pixels + ((from_row + row) * from->float_stride);
float* __restrict dest_buffer = to->pixels + ((to_row + row) * to->float_stride);
for (ndx = 0; ndx < dest_buffer_count; ndx++) {
float r = 0, g = 0, b = 0, a = 0;
const int left = weights[ndx].Left;
const int right = weights[ndx].Right;
const float* __restrict weightArray = weights[ndx].Weights;
int i;
/* Accumulate each channel */
for (i = left; i <= right; i++) {
const float weight = weightArray[i - left];
b += weight * source_buffer[i * from_step];
g += weight * source_buffer[i * from_step + 1];
r += weight * source_buffer[i * from_step + 2];
a += weight * source_buffer[i * from_step + 3];
}
dest_buffer[ndx * to_step] = b;
dest_buffer[ndx * to_step + 1] = g;
dest_buffer[ndx * to_step + 2] = r;
dest_buffer[ndx * to_step + 3] = a;
}
}
} else if (from_step == 3 && to_step == 3) {
void check_branch_relocation( void* fromArg, void* toArg, int32 countArg) {
inst_t *from = (inst_t*)fromArg, *to = (inst_t*)toArg;
int32 count = countArg / sizeof(inst_t);
if ( umax((uint32)from, (uint32)to) < umin((uint32)from + count, (uint32) to + count))
return; // overlapping, too hard to check
for ( int32 i = 0; i < count; ++i) {
inst_t* f = (inst_t*)get_target_of_branch_instruction(&from[i]);
inst_t* t = (inst_t*)get_target_of_branch_instruction(& to[i]);
if (f == NULL) ;
else if (from <= f && f < from+count) f += to - from;
if (f != t)
fatal2("check_branch_relocation: branch at 0x%x moved to 0x%x but is wrong",
&from[i], &to[i]);
}
}
bool BitmapBgra_flip_vertical(Context * context, BitmapBgra * b)
{
void* swap = CONTEXT_malloc(context,b->stride);
if (swap == NULL) {
CONTEXT_error(context, Out_of_memory);
return false;
}
//Dont' copy the full stride (padding), it could be windowed!
uint32_t row_length = umin (b->stride, b->w * BitmapPixelFormat_bytes_per_pixel (b->fmt));
for (uint32_t i = 0; i < b->h / 2; i++) {
void* top = b->pixels + (i * b->stride);
void* bottom = b->pixels + ((b->h - 1 - i) * b->stride);
memcpy (swap, top, row_length);
memcpy(top, bottom, row_length);
memcpy (bottom, swap, row_length);
}
CONTEXT_free(context,swap);
return true;
}
int
forEachTile(Kstring *kstr, unsigned int row, unsigned int col,
unsigned int num, Tile *first, ...)
{
unsigned int minVecLen = first->vecLen;
unsigned int valRow = first->nrRows;
unsigned int valCol = first->nrCols;
va_list argptr;
unsigned int i;
va_start(argptr, first);
for (i = 1; i < num; i++) {
Tile * cur = va_arg( argptr, Tile * );
minVecLen = umin(minVecLen, cur->vecLen);
}
va_end(argptr);
if (first->trans) {
valRow /= minVecLen;
}
else {
valCol /= minVecLen;
}
if (row >= valRow || col >= valCol /*|| row < 0 || col < 0*/) { //would be signed
return 0;
}
if (kstr) {
va_start(argptr, first);
for (i = 0; i < num; i++) {
Tile * cur = i ? va_arg( argptr, Tile * ) : first;
if (cur->baseName) {
unsigned int vRow = (cur->trans ? row * minVecLen : row);
unsigned int vCol = (cur->trans ? col : col * minVecLen);
sprintfTileElement(&kstr[i], cur, vRow, vCol, minVecLen);
}
}
va_end(argptr);
}
return first->trans ? valRow : valCol;
}
void
initTile(
Tile *tile,
const char *baseName,
unsigned int nrRows,
unsigned int nrCols,
unsigned int vecLen,
DataType dtype,
PrivateStorageType storType,
bool trans,
bool packed)
{
assert(baseName == NULL || strlen(baseName) <= MAX_TILE_BASE_NAMELEN);
tile->baseName = baseName;
tile->nrRows = nrRows;
tile->nrCols = nrCols;
tile->vecLen = umin(MAX_TILE_VECLEN, vecLen);
tile->dtype = dtype;
tile->storType = storType;
tile->trans = trans;
tile->packed = packed;
}
void
genTileCopy(
struct KgenContext *ctx,
const Tile *dst,
const Tile *src,
TileCopyOps op)
{
char tmp[1024];
Kstring el1, el2;
unsigned int nrRows, nrCols;
unsigned int incRows, incCols;
unsigned int vlen;
unsigned int i, j;
nrRows = umin(dst->nrRows, src->nrRows);
nrCols = umin(dst->nrCols, src->nrCols);
if (dst->trans != src->trans) {
vlen = 1;
incRows = incCols = 1;
}
else {
vlen = umin(dst->vecLen, src->vecLen);
if (!dst->trans) {
incRows = 1;
incCols = umin(dst->nrCols, src->nrCols);
incCols = umin(incCols, vlen);
}
else {
incRows = umin(dst->nrRows, src->nrRows);
incRows = umin(incRows, vlen);
incCols = 1;
}
}
for (i = 0; i < nrRows; i += incRows) {
for (j = 0; j < nrCols; j += incCols) {
sprintfTileElement(&el1, dst, i, j, vlen);
sprintfTileElement(&el2, src, i, j, vlen);
switch( op )
{
case TILECOPY_ASSIGN:
sprintf(tmp, "%s = %s;\n", el1.buf, el2.buf);
break;
case TILECOPY_ADD_ASSIGN:
sprintf(tmp, "%s += %s;\n", el1.buf, el2.buf);
break;
case TILECOPY_SUB_ASSIGN:
sprintf(tmp, "%s -= %s;\n", el1.buf, el2.buf);
break;
case TILECOPY_MUL_ASSIGN:
sprintf(tmp, "%s *= %s;\n", el1.buf, el2.buf);
break;
case TILECOPY_DIV_ASSIGN:
sprintf(tmp, "%s /= %s;\n", el1.buf, el2.buf);
break;
case TILECOPY_MOD_ASSIGN:
sprintf(tmp, "%s %%= %s;\n", el1.buf, el2.buf);
break;
default:
break;
}
kgenAddStmt(ctx, tmp);
}
}
kgenAddBlankLine(ctx);
}
static void
initTiles(
BlasGenSettings* gset,
TileSet* tileSet,
const struct SubproblemDim *subdims,
KernelExtraFlags kflags,
DataType dtype,
PrivateStorageType storType)
{
unsigned int rowsA;
unsigned int rowsB;
unsigned int rowsC;
unsigned int colsA;
unsigned int colsB;
unsigned int colsC;
bool transA;
bool transB;
unsigned int vecLenA;
unsigned int vecLenB;
unsigned int vecLenC;
rowsA = (unsigned int)subdims[1].y;
colsA = (unsigned int)szmax(subdims[1].y, subdims[1].bwidth);
rowsB = (unsigned int)szmax(subdims[1].y, subdims[1].bwidth);
colsB = (unsigned int)szmax(subdims[1].x, subdims[1].y);
rowsC = (unsigned int)subdims[1].y;
colsC = (unsigned int)subdims[1].x;
transA = isMatrixAccessColMaj(CLBLAS_TRSM, kflags, MATRIX_A);
transB = isMatrixAccessColMaj(CLBLAS_TRSM, kflags, MATRIX_B);
vecLenA = (unsigned int)((transA) ? subdims[1].y : subdims[1].bwidth);
vecLenA = umin(vecLenA, MAX_TILE_VECLEN);
vecLenB = (unsigned int)((transB) ? subdims[1].x : subdims[1].bwidth);
vecLenB = umin(vecLenB, MAX_TILE_VECLEN);
vecLenC = (transB) ? vecLenB : vecLenA;
initTile(&tileSet->rectA, "a", (unsigned int)subdims[1].y,
(unsigned int)subdims[1].bwidth, vecLenA, dtype,
storType, transA, false);
initTile(&tileSet->squareA, "a", (unsigned int)subdims[1].y,
(unsigned int)subdims[1].y, vecLenA, dtype, storType,
transA, false);
initTile(&tileSet->origB, "b", (unsigned int)subdims[1].bwidth,
(unsigned int)subdims[1].x, vecLenB, dtype, storType,
!transB, false);
initTile(&tileSet->bStage2, "b", (unsigned int)subdims[1].y,
(unsigned int)subdims[1].x, vecLenB, dtype, storType,
!transB, false);
initTile(&tileSet->bAsSqA, "b", (unsigned int)subdims[1].y,
(unsigned int)subdims[1].y, vecLenB, dtype, storType,
transA, false);
initTile(&tileSet->bAsC, "b", (unsigned int)subdims[1].y,
(unsigned int)subdims[1].x, vecLenB, dtype, storType,
gset->tileCY.trans, false);
initTile(&gset->tileA, "a", rowsA, colsA,
vecLenA, dtype, storType, transA, false);
initTile(&gset->tileBX, "b", rowsB, colsB,
vecLenB, dtype, storType, !transB, false);
initTile(&gset->tileCY, "c", rowsC, colsC,
vecLenC, dtype, storType, !transB, false);
tileSet->A = gset->tileA;
tileSet->B = gset->tileBX;
}
int
DjVuPalette::compute_palette(int maxcolors, int minboxsize)
{
if (!hist)
G_THROW( ERR_MSG("DjVuPalette.no_color") );
if (maxcolors<1 || maxcolors>MAXPALETTESIZE)
G_THROW( ERR_MSG("DjVuPalette.many_colors") );
// Paul Heckbert: "Color Image Quantization for Frame Buffer Display",
// SIGGRAPH '82 Proceedings, page 297. (also in ppmquant)
// Collect histogram colors
int sum = 0;
int ncolors = 0;
GTArray<PData> pdata;
{ // extra nesting for windows
for (GPosition p = *hist; p; ++p)
{
pdata.touch(ncolors);
PData &data = pdata[ncolors++];
int k = hist->key(p);
data.p[0] = (k>>16) & 0xff;
data.p[1] = (k>>8) & 0xff;
data.p[2] = (k) & 0xff;
data.w = (*hist)[p];
sum += data.w;
}
}
// Create first box
GList<PBox> boxes;
PBox newbox;
newbox.data = pdata;
newbox.colors = ncolors;
newbox.boxsize = 256;
newbox.sum = sum;
boxes.append(newbox);
// Repeat spliting boxes
while (boxes.size() < maxcolors)
{
// Find suitable box
GPosition p;
for (p=boxes; p; ++p)
if (boxes[p].colors>=2 && boxes[p].boxsize>minboxsize)
break;
if (! p)
break;
// Find box boundaries
PBox &splitbox = boxes[p];
unsigned char pmax[3];
unsigned char pmin[3];
pmax[0] = pmin[0] = splitbox.data->p[0];
pmax[1] = pmin[1] = splitbox.data->p[1];
pmax[2] = pmin[2] = splitbox.data->p[2];
{ // extra nesting for windows
for (int j=1; j<splitbox.colors; j++)
{
pmax[0] = umax(pmax[0], splitbox.data[j].p[0]);
pmax[1] = umax(pmax[1], splitbox.data[j].p[1]);
pmax[2] = umax(pmax[2], splitbox.data[j].p[2]);
pmin[0] = umin(pmin[0], splitbox.data[j].p[0]);
pmin[1] = umin(pmin[1], splitbox.data[j].p[1]);
pmin[2] = umin(pmin[2], splitbox.data[j].p[2]);
}
}
// Determine split direction and sort
int bl = pmax[0]-pmin[0];
int gl = pmax[1]-pmin[1];
int rl = pmax[2]-pmin[2];
splitbox.boxsize = (bl>gl ? (rl>bl ? rl : bl) : (rl>gl ? rl : gl));
if (splitbox.boxsize <= minboxsize)
continue;
if (gl == splitbox.boxsize)
qsort(splitbox.data, splitbox.colors, sizeof(PData), gcomp);
else if (rl == splitbox.boxsize)
qsort(splitbox.data, splitbox.colors, sizeof(PData), rcomp);
else
qsort(splitbox.data, splitbox.colors, sizeof(PData), bcomp);
// Find median
int lowercolors = 0;
int lowersum = 0;
while (lowercolors<splitbox.colors-1 && lowersum+lowersum<splitbox.sum)
lowersum += splitbox.data[lowercolors++].w;
// Compute new boxes
newbox.data = splitbox.data + lowercolors;
newbox.colors = splitbox.colors - lowercolors;
newbox.sum = splitbox.sum - lowersum;
splitbox.colors = lowercolors;
splitbox.sum = lowersum;
// Insert boxes at proper location
GPosition q;
for (q=p; q; ++q)
if (boxes[q].sum < newbox.sum)
break;
boxes.insert_before(q, newbox);
for (q=p; q; ++q)
if (boxes[q].sum < splitbox.sum)
break;
boxes.insert_before(q, boxes, p);
}
// Fill palette array
ncolors = 0;
palette.empty();
palette.resize(0,boxes.size()-1);
{ // extra nesting for windows
for (GPosition p=boxes; p; ++p)
{
PBox &box = boxes[p];
// Compute box representative color
float bsum = 0;
float gsum = 0;
float rsum = 0;
for (int j=0; j<box.colors; j++)
{
float w = (float)box.data[j].w;
bsum += box.data[j].p[0] * w;
gsum += box.data[j].p[1] * w;
rsum += box.data[j].p[2] * w;
}
PColor &color = palette[ncolors++];
color.p[0] = (unsigned char) fmin(255, bsum/box.sum);
color.p[1] = (unsigned char) fmin(255, gsum/box.sum);
color.p[2] = (unsigned char) fmin(255, rsum/box.sum);
color.p[3] = ( color.p[0]*BMUL + color.p[1]*GMUL + color.p[2]*RMUL) / SMUL;
}
}
// Save dominant color
PColor dcolor = palette[0];
// Sort palette colors in luminance order
qsort((PColor*)palette, ncolors, sizeof(PColor), lcomp);
// Clear invalid data
colordata.empty();
delete pmap;
pmap = 0;
// Return dominant color
return color_to_index_slow(dcolor.p);
}
static bool Render1D(Context * context,
const Renderer * r,
BitmapBgra * pSrc,
BitmapBgra * pDst,
const RenderDetails * details,
bool transpose,
int call_number)
{
bool success= true;
//How many rows to buffer and process at a time.
uint32_t buffer_row_count = 4; //using buffer=5 seems about 6% better than most other non-zero values.
//How many bytes per pixel are we scaling?
BitmapPixelFormat scaling_format = (pSrc->fmt == Bgra32 && !pSrc->alpha_meaningful) ? Bgr24 : pSrc->fmt;
BitmapFloat * buf = BitmapFloat_create(context,pSrc->w, buffer_row_count, scaling_format, false);
if (buf == NULL) {
return false;
}
buf->alpha_meaningful = pSrc->alpha_meaningful;
buf->alpha_premultiplied = buf->channels == 4;
/* Scale each set of lines */
for (uint32_t source_start_row = 0; source_start_row < pSrc->h; source_start_row += buffer_row_count) {
const uint32_t row_count = umin(pSrc->h - source_start_row, buffer_row_count);
if (!BitmapBgra_convert_srgb_to_linear(context, pSrc, source_start_row, buf, 0, row_count)) {
CONTEXT_add_to_callstack (context);
success=false;
goto cleanup;
}
if (!ApplyConvolutionsFloat1D(context, r, buf, 0, row_count, 0)) {
CONTEXT_add_to_callstack (context);
success=false;
goto cleanup;
}
if (details->apply_color_matrix && call_number == 2) {
if (!ApplyColorMatrix(context, r, buf, row_count)) {
CONTEXT_add_to_callstack (context);
success=false;
goto cleanup;
}
}
if (!BitmapFloat_pivoting_composite_linear_over_srgb(context, buf, 0, pDst, source_start_row, row_count, transpose)) {
CONTEXT_add_to_callstack (context);
success=false;
goto cleanup;
}
}
//sRGB sharpening
//Color matrix
cleanup:
BitmapFloat_destroy(context,buf);
return success;
}
static bool ScaleAndRender1D(Context * context, const Renderer * r,
BitmapBgra * pSrc,
BitmapBgra * pDst,
const RenderDetails * details,
bool transpose,
int call_number)
{
LineContributions * contrib = NULL;
BitmapFloat * source_buf = NULL;
BitmapFloat * dest_buf = NULL;
uint32_t from_count = pSrc->w;
uint32_t to_count = transpose ? pDst->h : pDst->w;
bool success = true;
if (details->interpolation->window == 0) {
CONTEXT_error(context, Invalid_argument);
return false;
}
//How many rows to buffer and process at a time.
const uint32_t buffer_row_count = 4; //using buffer=5 seems about 6% better than most other non-zero values.
//How many bytes per pixel are we scaling?
BitmapPixelFormat scaling_format = (pSrc->fmt == Bgra32 && !pSrc->alpha_meaningful) ? Bgr24 : pSrc->fmt;
prof_start(context,"contributions_calc", false);
contrib = LineContributions_create(context, to_count, from_count, details->interpolation);
if (contrib == NULL) {
CONTEXT_add_to_callstack (context);
success = false;
goto cleanup;
}
prof_stop(context,"contributions_calc", true, false);
prof_start(context,"create_bitmap_float (buffers)", false);
source_buf = BitmapFloat_create(context, from_count, buffer_row_count, scaling_format, false);
if (source_buf == NULL) {
CONTEXT_add_to_callstack (context);
success = false;
goto cleanup;
}
dest_buf = BitmapFloat_create(context, to_count, buffer_row_count, scaling_format, false);
if (dest_buf == NULL) {
CONTEXT_add_to_callstack (context);
success = false;
goto cleanup;
}
source_buf->alpha_meaningful = pSrc->alpha_meaningful;
dest_buf->alpha_meaningful = source_buf->alpha_meaningful;
source_buf->alpha_premultiplied = source_buf->channels == 4;
dest_buf->alpha_premultiplied = source_buf->alpha_premultiplied;
prof_stop(context,"create_bitmap_float (buffers)", true, false);
/* Scale each set of lines */
for (uint32_t source_start_row = 0; source_start_row < pSrc->h; source_start_row += buffer_row_count) {
const uint32_t row_count = umin(pSrc->h - source_start_row, buffer_row_count);
prof_start(context,"convert_srgb_to_linear", false);
if (!BitmapBgra_convert_srgb_to_linear(context,pSrc, source_start_row, source_buf, 0, row_count)) {
CONTEXT_add_to_callstack (context);
success=false;
goto cleanup;
}
prof_stop(context,"convert_srgb_to_linear", true, false);
prof_start(context,"ScaleBgraFloatRows", false);
if (!BitmapFloat_scale_rows(context, source_buf, 0, dest_buf, 0, row_count, contrib->ContribRow)) {
CONTEXT_add_to_callstack (context);
success=false;
goto cleanup;
}
prof_stop(context,"ScaleBgraFloatRows", true, false);
if (!ApplyConvolutionsFloat1D(context, r, dest_buf, 0, row_count, contrib->percent_negative)) {
CONTEXT_add_to_callstack (context);
success=false;
goto cleanup;
}
if (details->apply_color_matrix && call_number == 2) {
if (!ApplyColorMatrix(context, r, dest_buf, row_count)) {
CONTEXT_add_to_callstack (context);
success=false;
goto cleanup;
}
}
prof_start(context,"pivoting_composite_linear_over_srgb", false);
if (!BitmapFloat_pivoting_composite_linear_over_srgb(context, dest_buf, 0, pDst, source_start_row, row_count, transpose)) {
CONTEXT_add_to_callstack (context);
success=false;
goto cleanup;
}
prof_stop(context,"pivoting_composite_linear_over_srgb", true, false);
}
//sRGB sharpening
//Color matrix
cleanup:
//p->Start("Free Contributions,FloatBuffers", false);
if (contrib != NULL) LineContributions_destroy(context, contrib);
if (source_buf != NULL) BitmapFloat_destroy(context, source_buf);
if (dest_buf != NULL) BitmapFloat_destroy(context, dest_buf);
///p->Stop("Free Contributions,FloatBuffers", true, false);
return success;
}
void GrMouseSetAccel(int thresh, int accel)
{
MOUINFO->thresh = umin(64, umax(1, thresh));
MOUINFO->accel = umin(16, umax(1, accel));
}
void GrMouseSetSpeed(int spmult, int spdiv)
{
MOUINFO->spmult = umin(16, umax(1, spmult));
MOUINFO->spdiv = umin(16, umax(1, spdiv));
}
afs_int32
UDP_GetTicket(int ksoc, struct packet *pkt, afs_int32 kvno,
char *authDomain, char *ticket, int ticketLen, char *auth,
int authLen)
{
afs_int32 code;
struct ktc_encryptionKey tgskey;
char name[MAXKTCNAMELEN];
char inst[MAXKTCNAMELEN];
char cell[MAXKTCREALMLEN];
struct ktc_encryptionKey authSessionKey;
afs_int32 host;
Date start;
Date authEnd;
Date now = time(0);
int celllen;
int import;
char *packet;
int slen;
int byteOrder = pkt->byteOrder;
char sname[MAXKTCNAMELEN];
char sinst[MAXKTCNAMELEN];
afs_int32 time_ws;
unsigned char life;
struct ubik_trans *tt;
afs_int32 to;
struct kaentry caller;
struct kaentry server;
Date reqEnd;
struct ktc_encryptionKey sessionKey;
int newTicketLen;
char newTicket[MAXKTCTICKETLEN];
char cipher[2 * MAXKTCTICKETLEN]; /* put encrypted part of answer here */
int cipherLen;
struct packet ans;
COUNT_REQ(UGetTicket);
if ((code = InitAuthServ(&tt, LOCKREAD, this_op)))
goto fail;
code =
ka_LookupKvno(tt, KA_TGS_NAME,
((strlen(authDomain) > 0) ? authDomain : lrealm), kvno,
&tgskey);
if (code)
goto abort;
code =
tkt_DecodeTicket(ticket, ticketLen, &tgskey, name, inst, cell,
&authSessionKey, &host, &start, &authEnd);
pkt->name = name;
pkt->inst = inst;
pkt->realm = cell;
if (code) {
code = KERB_ERR_AUTH_EXP; /* was KANOAUTH */
goto abort;
}
save_principal(udptgsPrincipal, name, inst, cell);
code = tkt_CheckTimes(start, authEnd, now);
if (code <= 0) {
if (code == -1) {
code = KERB_ERR_SERVICE_EXP; /* was RXKADEXPIRED */
goto abort;
}
code = KERB_ERR_AUTH_EXP; /* was KANOAUTH */
goto abort;
}
celllen = strlen(cell);
import = 0;
if ((strlen(authDomain) > 0) && (strcmp(authDomain, lrealm) != 0))
import = 1;
if (import && (celllen == 0)) {
code = KERB_ERR_PKT_VER; /* was KABADTICKET */
goto abort;
}
if (celllen == 0) {
strncpy(cell, lrealm, MAXKTCREALMLEN - 1);
cell[MAXKTCREALMLEN - 1] = 0;
};
if (!krb4_cross && strcmp(lrealm, cell) != 0) {
code = KERB_ERR_PRINCIPAL_UNKNOWN;
goto abort;
}
if (krb_udp_debug) {
printf("UGetTicket: got ticket from '%s'.'%s'@'%s'\n", name, inst,
cell);
}
code = check_auth(pkt, auth, authLen, &authSessionKey, name, inst, cell);
if (code)
goto abort;
/* authenticator and all is OK so read actual request */
packet = pkt->rest;
getint(time_ws);
life = *(unsigned char *)packet++;
getstr(sname);
getstr(sinst);
start = now;
reqEnd = life_to_time(start, life);
if (krb_udp_debug) {
printf("UGetTicket: request for server '%s'.'%s'\n", sname, sinst);
}
save_principal(udptgsServerPrincipal, sname, sinst, 0);
if (import) {
strcpy(caller.userID.name, name);
strcpy(caller.userID.instance, inst);
caller.max_ticket_lifetime = htonl(MAXKTCTICKETLIFETIME);
} else {
code = FindBlock(tt, name, inst, &to, &caller);
if (code)
goto abort;
if (to == 0) {
ka_PrintUserID("GetTicket: User ", name, inst, " unknown.\n");
code = KERB_ERR_PRINCIPAL_UNKNOWN; /* KANOENT */
goto abort;
}
if (ntohl(caller.flags) & KAFNOTGS) {
code = KERB_ERR_AUTH_EXP; /* was KABADUSER */
goto abort;
}
}
code = FindBlock(tt, sname, sinst, &to, &server); /* get server's entry */
if (code)
goto abort;
if (to == 0) { /* entry not found */
ka_PrintUserID("GetTicket: Server ", sname, sinst, " unknown.\n");
code = KERB_ERR_PRINCIPAL_UNKNOWN; /* KANOENT */
goto abort;
}
code = ubik_EndTrans(tt);
if (code)
goto fail;
if (ntohl(server.flags) & KAFNOSEAL)
return KABADSERVER;
code = DES_new_random_key(ktc_to_cblock(&sessionKey));
if (code) {
code = KERB_ERR_NULL_KEY; /* was KANOKEYS */
goto fail;
}
reqEnd =
umin(umin(reqEnd, authEnd),
umin(start + ntohl(caller.max_ticket_lifetime),
start + ntohl(server.max_ticket_lifetime)));
code =
tkt_MakeTicket(newTicket, &newTicketLen, &server.key,
caller.userID.name, caller.userID.instance, cell,
start, reqEnd, &sessionKey,
htonl(pkt->from.sin_addr.s_addr), server.userID.name,
server.userID.instance);
if (code)
goto fail;
cipherLen = sizeof(cipher);
code =
create_cipher(cipher, &cipherLen, &sessionKey, sname, sinst, start,
reqEnd, ntohl(server.key_version), newTicket,
newTicketLen, &authSessionKey);
if (code)
goto fail;
code =
create_reply(&ans, name, inst, start, reqEnd, 0, cipher, cipherLen);
if (code)
goto fail;
code =
sendto(ksoc, ans.data, ans.len, 0, (struct sockaddr *)&pkt->from,
sizeof(pkt->from));
if (code != ans.len) {
perror("calling sendto");
code = -1;
goto fail;
}
if (cipherLen != 0) {
KALOG(name, inst, sname, sinst, NULL, host, LOG_GETTICKET);
}
osi_audit(UDPGetTicketEvent, 0, AUD_STR, name, AUD_STR, inst, AUD_STR,
cell, AUD_STR, sname, AUD_STR, sinst, AUD_END);
return 0;
abort:
ubik_AbortTrans(tt);
fail:
osi_audit(UDPGetTicketEvent, code, AUD_STR, name, AUD_STR, inst, AUD_STR,
NULL, AUD_STR, NULL, AUD_STR, NULL, AUD_END);
return code;
}
afs_int32
UDP_Authenticate(int ksoc, struct sockaddr_in *client, char *name,
char *inst, Date startTime, Date endTime, char *sname,
char *sinst)
{
struct ubik_trans *tt;
afs_int32 to; /* offset of block */
struct kaentry tentry;
afs_int32 tgskvno; /* key version of service key */
struct ktc_encryptionKey tgskey; /* service key for encrypting ticket */
int tgt;
Date now = time(0);
afs_int32 code;
char ticket[MAXKTCTICKETLEN]; /* our copy of the ticket */
int ticketLen;
struct ktc_encryptionKey sessionKey; /* we have to invent a session key */
char cipher[2 * MAXKTCTICKETLEN]; /* put encrypted part of answer here */
int cipherLen;
struct packet ans;
COUNT_REQ(UAuthenticate);
if (!name_instance_legal(name, inst))
return KERB_ERR_NAME_EXP; /* KABADNAME */
if ((code = InitAuthServ(&tt, LOCKREAD, this_op)))
return code;
code = FindBlock(tt, name, inst, &to, &tentry);
if (code)
goto abort;
if (to) { /* if user exists check other stuff */
afs_int32 sto;
struct kaentry sentry;
save_principal(udpAuthPrincipal, name, inst, 0);
tgt = ((strcmp(sname, KA_TGS_NAME) == 0)
&& (strcmp(sinst, lrealm) == 0));
if ((ntohl(tentry.user_expiration) < now)
|| (tgt && (ntohl(tentry.flags) & KAFNOTGS))) {
code = KERB_ERR_NAME_EXP; /* KABADUSER */
goto abort;
}
code = FindBlock(tt, KA_TGS_NAME, lrealm, &sto, &sentry);
if (code)
goto abort;
if (sto == 0) {
code = KANOENT;
goto abort;
}
if ((ntohl(sentry.user_expiration) < now)) {
code = KERB_ERR_NAME_EXP; /* XXX Could use another error code XXX */
goto abort;
}
if (abs(startTime - now) > KTC_TIME_UNCERTAINTY) {
code = KERB_ERR_SERVICE_EXP; /* was KABADREQUEST */
goto abort;
}
if (tentry.misc_auth_bytes) {
unsigned char misc_auth_bytes[4];
afs_uint32 temp; /* unsigned for safety */
afs_uint32 pwexpires;
memcpy(&temp, tentry.misc_auth_bytes, sizeof(afs_uint32));
temp = ntohl(temp);
unpack_long(temp, misc_auth_bytes);
pwexpires = misc_auth_bytes[0];
if (pwexpires) {
pwexpires =
ntohl(tentry.change_password_time) +
24 * 60 * 60 * pwexpires;
if (pwexpires < now) {
code = KERB_ERR_AUTH_EXP; /* was KAPWEXPIRED */
goto abort;
}
}
}
/* make the ticket */
code = DES_new_random_key(ktc_to_cblock(&sessionKey));
if (code) {
code = KERB_ERR_NULL_KEY; /* was KANOKEYS */
goto abort;
}
endTime =
umin(endTime, startTime + ntohl(tentry.max_ticket_lifetime));
if ((code = ka_LookupKey(tt, sname, sinst, &tgskvno, &tgskey))
|| (code =
tkt_MakeTicket(ticket, &ticketLen, &tgskey, name, inst,
lrealm, startTime, endTime, &sessionKey,
htonl(client->sin_addr.s_addr), sname, sinst)))
goto abort;
cipherLen = sizeof(cipher);
code =
create_cipher(cipher, &cipherLen, &sessionKey, sname, sinst,
startTime, endTime, tgskvno, ticket, ticketLen,
&tentry.key);
if (code)
goto abort;
} else { /* no such user */
cipherLen = 0;
tentry.key_version = 0;
}
code = ubik_EndTrans(tt);
if (code)
goto fail;
code =
create_reply(&ans, name, inst, startTime, endTime,
ntohl(tentry.key_version), cipher, cipherLen);
if (code)
goto fail;
if (krb_udp_debug) {
printf("Sending %d bytes ending in: ", ans.len);
ka_PrintBytes(ans.data + ans.len - 8, 8);
printf("\n");
}
code =
sendto(ksoc, ans.data, ans.len, 0, (struct sockaddr *)client,
sizeof(*client));
if (code != ans.len) {
perror("calling sendto");
code = -1;
goto fail;
}
KALOG(name, inst, sname, sinst, NULL, client->sin_addr.s_addr,
LOG_AUTHENTICATE);
if (cipherLen != 0) {
KALOG(name, inst, sname, sinst, NULL, client->sin_addr.s_addr,
LOG_TGTREQUEST);
}
osi_audit(UDPAuthenticateEvent, 0, AUD_STR, name, AUD_STR, inst, AUD_END);
return 0;
abort:
COUNT_ABO;
ubik_AbortTrans(tt);
fail:
osi_audit(UDPAuthenticateEvent, code, AUD_STR, name, AUD_STR, inst,
AUD_END);
return code;
}