void Bezier::recalc(QPointF p)
{
// http://www.flong.com/texts/code/shapers_bez/
// http://www.lemoda.net/maths/bezier-length/index.html,
// arbitrary but reasonable t-values for interior control points
double t0 = 0.3;
double t1 = 0.7;
if (m_drag_cp0) {
double x = (p.x() - m_endpoint0.x() * B0(t0) - m_cp1.x() * B2(t0) - m_endpoint1.x() * B3(t0)) / B1(t0);
double y = (p.y() - m_endpoint0.y() * B0(t0) - m_cp1.y() * B2(t0) - m_endpoint1.y() * B3(t0)) / B1(t0);
m_cp0 = QPointF(x, y);
}
else {
double x = (p.x() - m_endpoint0.x() * B0(t1) - m_cp0.x() * B1(t1) - m_endpoint1.x() * B3(t1)) / B2(t1);
double y = (p.y() - m_endpoint0.y() * B0(t1) - m_cp0.y() * B1(t1) - m_endpoint1.y() * B3(t1)) / B2(t1);
m_cp1 = QPointF(x, y);
}
/*
DebugDialog::debug(QString("ix:%1 p0x:%2,p0y:%3 p1x:%4,p1y:%5 px:%6,py:%7")
.arg(m_drag_cp0)
.arg(m_endpoint0.x())
.arg(m_endpoint0.y())
.arg(m_endpoint1.x())
.arg(m_endpoint1.y())
.arg(p.x())
.arg(p.y())
);
*/
m_isEmpty = false;
}
static int reallyroutespline (Pedge_t *edges, int edgen,
Ppoint_t *inps, int inpn, Ppoint_t ev0, Ppoint_t ev1) {
Ppoint_t p1, p2, cp1, cp2, p;
Pvector_t v1, v2, splitv, splitv1, splitv2;
double maxd, d, t;
int maxi, i, spliti;
static tna_t *tnas;
static int tnan;
if (tnan < inpn) {
if (!tnas) {
if (!(tnas = (tna_t *)malloc (sizeof (tna_t) * inpn)))
return -1;
} else {
if (!(tnas = (tna_t *)realloc (tnas, sizeof (tna_t) * inpn)))
return -1;
}
tnan = inpn;
}
tnas[0].t = 0;
for (i = 1; i < inpn; i++)
tnas[i].t = tnas[i - 1].t + dist (inps[i], inps[i - 1]);
for (i = 1; i < inpn; i++)
tnas[i].t /= tnas[inpn - 1].t;
for (i = 0; i < inpn; i++) {
tnas[i].a[0] = scale (ev0, B1 (tnas[i].t));
tnas[i].a[1] = scale (ev1, B2 (tnas[i].t));
}
if (mkspline (inps, inpn, tnas, ev0, ev1, &p1, &v1, &p2, &v2) == -1)
return -1;
if (splinefits (edges, edgen, p1, v1, p2, v2, (inpn == 2 ? 1 : 0)))
return 0;
cp1 = add (p1, scale (v1, 1 / 3.0));
cp2 = sub (p2, scale (v2, 1 / 3.0));
for (maxd = -1, maxi = -1, i = 1; i < inpn - 1; i++) {
t = tnas[i].t;
p.x = B0 (t) * p1.x + B1 (t) * cp1.x +
B2 (t) * cp2.x + B3 (t) * p2.x;
p.y = B0 (t) * p1.y + B1 (t) * cp1.y +
B2 (t) * cp2.y + B3 (t) * p2.y;
if ((d = dist (p, inps[i])) > maxd)
maxd = d, maxi = i;
}
spliti = maxi;
splitv1 = normv (sub (inps[spliti], inps[spliti - 1]));
splitv2 = normv (sub (inps[spliti + 1], inps[spliti]));
splitv = normv (add (splitv1, splitv2));
reallyroutespline (edges, edgen, inps, spliti + 1, ev0, splitv);
reallyroutespline (edges, edgen, &inps[spliti], inpn - spliti, splitv, ev1);
return 0;
}
// Shortest cubic spline through 4 on-curve points(chord approximation)
void Bezier::FitSpline(Vec3 p[])
{
// use chord length for shortest(best) cubic spline approximation
float c3 = (p[1] - p[0]).Magnitude();
float c2 = (p[2] - p[1]).Magnitude();
float c1 = (p[3] - p[2]).Magnitude();
// cases where p[1] is close to p[2] might lead to instabilities(need some heuristic)
if (50 * c2 < c1 + c3)
{
p[1] = p[0] + (p[1] - p[0]) * 0.98;
p[2] = p[3] + (p[2] - p[3]) * 0.98;
c3 = c3 * 0.98;
c2 = (p[2] - p[1]).Magnitude();
c1 = c1 * 0.98;
}
float t1 = c1 / (c1 + c2 + c3);
float t2 = (c1 + c2) / (c1 + c2 + c3);
// Solve M * x = y
float m00 = B1(t1);
float m01 = B2(t1);
float m10 = B1(t2);
float m11 = B2(t2);
float detM = m00 * m11 - m01 * m10;
if (fabs(detM) > 1E-3)
{
// y = p - p0 * B0(t) - p3 * B3(t)
Vec3 y1 = p[1] - p[0] * B0(t1) - p[3] * B3(t1);
Vec3 y2 = p[2] - p[0] * B0(t2) - p[3] * B3(t2);
// Minv
float s = 1 / detM;
float n00 = s * m11;
float n01 = -s * m01;
float n10 = -s * m10;
float n11 = s * m00;
// x = Minv * y
Vec3 x1 = y1 * n00 + y2 * n01;
Vec3 x2 = y1 * n10 + y2 * n11;
p[1] = x1;
p[2] = x2;
}
}
complex LoopToolsWrapper::PV_B0(const double mu2, const double p2,
const double m02, const double m12) const
{
setmudim(mu2);
std::complex<double> B0val = B0(p2, m02, m12);
return complex( B0val.real(), B0val.imag(), false );
}
int huddraw_help() {
fg_color = B0(COLOR_BLACK);
bg_color = B3(COLOR_WHITE);
int i,l;
for(i=0; HUD_HELP[i].title; i++) {
if (!strcmp(hud_help_page, HUD_HELP[i].title)) {
huddraw_help_text(HUD_HELP[i].text);
return 1;
}
}
char text[4096];
sprintf(text, "No help available for topic \n"
"%s\n"
"\n"
"Use #hud help to see the list of\n"
"available commands and topics \n"
"\n"
"Use #hud help <topic> for help \n"
"on specific topic \n",
hud_help_page);
huddraw_help_text(text);
return 1;
}
void huddraw_info_nav(int r) {
char text[256];
fg_color = B0(COLOR_BLACK);
bg_color = B3(COLOR_SAND_YELLOW);
draw_rect(0,r,128,35,1);
fg_color = B2(COLOR_SAND_YELLOW);
draw_rect(35,r+2,12,14,0);
draw_rect(76,r+2,12,14,0);
// coordinates
fg_color = B3(COLOR_REDSTONE_RED);
bg_color = COLOR_TRANSPARENT;
int32_t x = (int32_t)floor(gs.own.x);
int32_t z = (int32_t)floor(gs.own.z);
int32_t x_= (gs.world==&gs.nether) ? x*8 : x/8;
int32_t z_= (gs.world==&gs.nether) ? z*8 : z/8;
char * n_= (gs.world==&gs.nether) ? "Overworld" : "Nether";
draw_text(3, r+ 3, "X");
draw_text(3, r+10, "Z");
draw_text(3, r+19, "Y");
draw_text(3, r+26, "DIR");
sprintf(text, "%9d", x); draw_text(11,r+3,text);
sprintf(text, "%9d", x_); draw_text(53,r+3,text);
sprintf(text, "%9d", z); draw_text(11,r+10,text);
sprintf(text, "%9d", z_); draw_text(53,r+10,text);
sprintf(text, "%9d", (int32_t)floor(gs.own.y)); draw_text(11,r+19,text);
char * dir = "UNKNOWN";
switch(player_direction()) {
case DIR_NORTH : dir = "NORTH"; break;
case DIR_SOUTH : dir = "SOUTH"; break;
case DIR_EAST : dir = "EAST"; break;
case DIR_WEST : dir = "WEST"; break;
}
sprintf(text, "%9s", dir); draw_text(11,r+26,text);
int pos = (42-(strlen(n_)*4-1))/2+49;
draw_text(pos, r+26, n_);
// compass
huddraw_compass(108,r+17,B0(COLOR_BLACK), B3(COLOR_REDSTONE_RED));
}
/*
* The IXP425 expansion bus only allows 16-bit wide acceses
* when attached to a 16-bit wide device (such as the 28F128J3A),
* so we can't use a memcpy_fromio as it does byte acceses.
*/
static void ixp425_copy_from(struct map_info *map, void *to,
unsigned long from, ssize_t len)
{
int i;
u8 *dest = (u8*)to;
u16 *src = (u16 *)(map->map_priv_1 + from);
u16 data;
for(i = 0; i < (len / 2); i++) {
data = src[i];
dest[i * 2] = B0(data);
dest[i * 2 + 1] = B1(data);
}
if(len & 1)
dest[len - 1] = B0(src[i]);
}
void copy_from_flash(unsigned long from, void *to,ssize_t len)
{
int i;
u8 *dest = (u8*)to;
u16 data;
unsigned long remap = (unsigned long)ioremap(FLASH_BASE_ADDR,FLASH_SIZE);
u16 *src = (u16 *)(remap + from);
for(i = 0; i < (len / 2);i++){
data = src[i];
dest[i * 2] = B0(data);
dest[i * 2 + 1] = B1(data);
}
if(len & 1)
dest[len - 1] = B0(src[i]);
iounmap(remap);
}
void huddraw_info_health(int r) {
char text[256];
fg_color = B0(COLOR_BLACK);
bg_color = B3(COLOR_GRASS_GREEN);
draw_rect(0,r,128,12,1);
bg_color = COLOR_TRANSPARENT;
fg_color = B3(COLOR_REDSTONE_RED);
draw_blit(fonts, FONTS_ICON_HEART, 8, 8, 2, r+2);
fg_color = B0(COLOR_BLACK);
sprintf(text, "%.1f", gs.own.health);
draw_text(12, r+3, text);
fg_color = B2(COLOR_ORANGE);
draw_blit(fonts, FONTS_ICON_FOOD, 8, 8, 33, r+2);
fg_color = B0(COLOR_BLACK);
sprintf(text, "%d (%.1f)", gs.own.food, gs.own.saturation);
draw_text(43, r+3, text);
}
secure_vector<byte> CCM_Mode::format_b0(size_t sz)
{
secure_vector<byte> B0(BS);
const byte b_flags = (m_ad_buf.size() ? 64 : 0) + (((tag_size()/2)-1) << 3) + (L()-1);
B0[0] = b_flags;
copy_mem(&B0[1], m_nonce.data(), m_nonce.size());
encode_length(sz, &B0[m_nonce.size()+1]);
return B0;
}
secure_vector<uint8_t> CCM_Mode::format_b0(size_t sz)
{
secure_vector<uint8_t> B0(CCM_BS);
const uint8_t b_flags =
static_cast<uint8_t>((m_ad_buf.size() ? 64 : 0) + (((tag_size()/2)-1) << 3) + (L()-1));
B0[0] = b_flags;
copy_mem(&B0[1], m_nonce.data(), m_nonce.size());
encode_length(sz, &B0[m_nonce.size()+1]);
return B0;
}
void
CAST5decrypt(const PGPUInt8 *in, PGPUInt8 *out, const PGPUInt32 *xkey)
{
PGPUInt32 l, r, t;
r = (PGPUInt32) in[0]<<24 | (PGPUInt32) in[1]<<16 |
(PGPUInt32) in[2]<<8 | in[3];
l = (PGPUInt32) in[4]<<24 | (PGPUInt32) in[5]<<16 |
(PGPUInt32) in[6]<<8 | in[7];
t = F1(l, xkey, 15); r ^= G1(t);
t = F3(r, xkey, 14); l ^= G3(t);
t = F2(l, xkey, 13); r ^= G2(t);
t = F1(r, xkey, 12); l ^= G1(t);
// Start here if only doing 12 rounds
t = F3(l, xkey, 11); r ^= G3(t);
t = F2(r, xkey, 10); l ^= G2(t);
t = F1(l, xkey, 9); r ^= G1(t);
t = F3(r, xkey, 8); l ^= G3(t);
t = F2(l, xkey, 7); r ^= G2(t);
t = F1(r, xkey, 6); l ^= G1(t);
t = F3(l, xkey, 5); r ^= G3(t);
t = F2(r, xkey, 4); l ^= G2(t);
t = F1(l, xkey, 3); r ^= G1(t);
t = F3(r, xkey, 2); l ^= G3(t);
t = F2(l, xkey, 1); r ^= G2(t);
t = F1(r, xkey, 0); l ^= G1(t);
out[0] = (PGPUInt8) B0(l);
out[1] = (PGPUInt8) B1(l);
out[2] = (PGPUInt8) B2(l);
out[3] = (PGPUInt8) B3(l);
out[4] = (PGPUInt8) B0(r);
out[5] = (PGPUInt8) B1(r);
out[6] = (PGPUInt8) B2(r);
out[7] = (PGPUInt8) B3(r);
}
/*
* Encrypt the 8 bytes at *in into the 8 bytes at *out using the expanded
* key schedule from *xkey.
*/
static void
CAST5encrypt(PGPByte const *in, PGPByte *out, PGPUInt32 const *xkey)
{
PGPUInt32 l, r, t;
l = (PGPUInt32)
in[0]<<24 | (PGPUInt32)in[1]<<16 | (PGPUInt32)in[2]<<8 | in[3];
r = (PGPUInt32)
in[4]<<24 | (PGPUInt32)in[5]<<16 | (PGPUInt32)in[6]<<8 | in[7];
t = F1(r, xkey, 0); l ^= G1(t);
t = F2(l, xkey, 1); r ^= G2(t);
t = F3(r, xkey, 2); l ^= G3(t);
t = F1(l, xkey, 3); r ^= G1(t);
t = F2(r, xkey, 4); l ^= G2(t);
t = F3(l, xkey, 5); r ^= G3(t);
t = F1(r, xkey, 6); l ^= G1(t);
t = F2(l, xkey, 7); r ^= G2(t);
t = F3(r, xkey, 8); l ^= G3(t);
t = F1(l, xkey, 9); r ^= G1(t);
t = F2(r, xkey, 10); l ^= G2(t);
t = F3(l, xkey, 11); r ^= G3(t);
/* Stop here if only doing 12 rounds */
t = F1(r, xkey, 12); l ^= G1(t);
t = F2(l, xkey, 13); r ^= G2(t);
t = F3(r, xkey, 14); l ^= G3(t);
t = F1(l, xkey, 15); r ^= G1(t);
out[0] = B0(r);
out[1] = B1(r);
out[2] = B2(r);
out[3] = B3(r);
out[4] = B0(l);
out[5] = B1(l);
out[6] = B2(l);
out[7] = B3(l);
}
int huddraw_map() {
int shading[16] = { 3, 3, 3, 3, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2 };
if (!(hud_inv & HUDINVMASK_TUNNEL)) return 0;
bg_color = B0(COLOR_BLACK);
draw_clear();
int32_t x = (int32_t)floor(gs.own.x);
int32_t y = (int32_t)floor(gs.own.y);
int32_t z = (int32_t)floor(gs.own.z);
extent_t ex = { { x-80, y-12, z-80 }, { x+80, y+3, z+80 } };
cuboid_t cb = export_cuboid_extent(ex);
int r,c,i,j;
int32_t off = cb.boff + 16*cb.sa.x + 16;
for(r=0; r<128; r++) {
for(c=0; c<128; c++) {
int poff = off+r*cb.sa.x+c;
for(j=0; j<16; j++) {
if ( cb.data[j][poff].bid ) {
int8_t color = BLOCK_COLORMAP[cb.data[j][poff].bid][cb.data[j][poff].meta];
hud_image[r*128+c] = color*4 + shading[j];
}
}
}
}
for(i=0; i<256; i++) lh_free(cb.data[i]);
hud_image[64*128+64] = 126;
char text[256];
bg_color = B3(COLOR_WHITE);
fg_color = B3(COLOR_REDSTONE_RED);
sprintf(text, "%d,%d", x, z);
draw_text(2, 2, text);
sprintf(text, "Y:%d", y);
draw_text(2, 9, text);
bg_color = COLOR_TRANSPARENT;
huddraw_compass(111, 16, B3(COLOR_GOLD_YELLOW), B3(COLOR_WHITE));
return 1;
}
void huddraw_info_inv_item(int id, float damage, int r) {
int col = 4+31*(id%4);
int row = r+2+10*(id/4);
int fcol = FONTS_ICON_EQ_C+8*(id%4);
int frow = FONTS_ICON_EQ_R+8*(id/4);
bg_color = COLOR_TRANSPARENT;
fg_color = (damage<0) ? B0(COLOR_CLAY_GRAY) : B3(COLOR_DIAMOND_BLUE);
draw_blit(fonts, fcol, frow, 8, 8, col, row);
fg_color = B2(COLOR_CLAY_GRAY);
draw_rect(col+10, row+1, 16, 6, 1);
if (damage >= 0) {
fg_color = B3(COLOR_LAPIS_BLUE);
if (damage < 1.0) fg_color = B3(COLOR_EMERALD_GREEN);
if (damage < 0.5) fg_color = B3(COLOR_GOLD_YELLOW);
if (damage < 0.25) fg_color = B3(COLOR_ORANGE);
if (damage < 0.1) fg_color = B3(COLOR_REDSTONE_RED);
draw_rect(col+10, row+1, 16*damage, 6, 0);
}
}
int test()
{
int Error = 0;
int A0 = static_cast<int>(glm::log2(16.f));
glm::ivec1 B0(glm::log2(glm::vec1(16.f)));
glm::ivec2 C0(glm::log2(glm::vec2(16.f)));
glm::ivec3 D0(glm::log2(glm::vec3(16.f)));
glm::ivec4 E0(glm::log2(glm::vec4(16.f)));
int A1 = glm::log2(int(16));
glm::ivec1 B1 = glm::log2(glm::ivec1(16));
glm::ivec2 C1 = glm::log2(glm::ivec2(16));
glm::ivec3 D1 = glm::log2(glm::ivec3(16));
glm::ivec4 E1 = glm::log2(glm::ivec4(16));
Error += A0 == A1 ? 0 : 1;
Error += glm::all(glm::equal(B0, B1)) ? 0 : 1;
Error += glm::all(glm::equal(C0, C1)) ? 0 : 1;
Error += glm::all(glm::equal(D0, D1)) ? 0 : 1;
Error += glm::all(glm::equal(E0, E1)) ? 0 : 1;
glm::uint64 A2 = glm::log2(glm::uint64(16));
glm::u64vec1 B2 = glm::log2(glm::u64vec1(16));
glm::u64vec2 C2 = glm::log2(glm::u64vec2(16));
glm::u64vec3 D2 = glm::log2(glm::u64vec3(16));
glm::u64vec4 E2 = glm::log2(glm::u64vec4(16));
Error += A2 == glm::uint64(4) ? 0 : 1;
Error += glm::all(glm::equal(B2, glm::u64vec1(4))) ? 0 : 1;
Error += glm::all(glm::equal(C2, glm::u64vec2(4))) ? 0 : 1;
Error += glm::all(glm::equal(D2, glm::u64vec3(4))) ? 0 : 1;
Error += glm::all(glm::equal(E2, glm::u64vec4(4))) ? 0 : 1;
return Error;
}
void huddraw_info_inv(int r) {
bg_color = B3(COLOR_PINK);
fg_color = B0(COLOR_BLACK);
draw_rect(0,r,128,22,1);
int present[8] = { 0, 0, 0, 0, 0, 0, 0, 0 };
int i,j;
for(j=0; hii[j].pos>=0; j++) {
hudinvitem *h = hii+j;
for(i=h->ssid; i<=h->lsid; i++) {
slot_t *s = gs.inv.slots+i;
if (s->item == h->iid) {
float damage = ((float)h->dur-(float)s->damage)/(float)h->dur;
huddraw_info_inv_item(h->pos, damage, r);
present[h->pos] = 1;
}
}
}
for(i=0; i<8; i++)
if (!present[i])
huddraw_info_inv_item(i, -1, r);
}
int test()
{
int Error = 0;
int A0(glm::log2(10.f));
glm::ivec1 B0(glm::log2(glm::vec1(10.f)));
glm::ivec2 C0(glm::log2(glm::vec2(10.f)));
glm::ivec3 D0(glm::log2(glm::vec3(10.f)));
glm::ivec4 E0(glm::log2(glm::vec4(10.f)));
int A1 = glm::log2(int(10.f));
glm::ivec1 B1 = glm::log2(glm::ivec1(10.f));
glm::ivec2 C1 = glm::log2(glm::ivec2(10.f));
glm::ivec3 D1 = glm::log2(glm::ivec3(10.f));
glm::ivec4 E1 = glm::log2(glm::ivec4(10.f));
Error += A0 == A1 ? 0 : 1;
Error += glm::all(glm::equal(B0, B1)) ? 0 : 1;
Error += glm::all(glm::equal(C0, C1)) ? 0 : 1;
Error += glm::all(glm::equal(D0, D1)) ? 0 : 1;
Error += glm::all(glm::equal(E0, E1)) ? 0 : 1;
return Error;
}
/*
* GenerateBezier :
* Use least-squares method to find Bezier control points for region.
*
*/
static BezierCurve GenerateBezier(Point2 *d, int first, int last, double *uPrime,
Vector2 tHat1, Vector2 tHat2)
{
int i;
Vector2 A[MAXPOINTS][2]; /* Precomputed rhs for eqn */
int nPts; /* Number of pts in sub-curve */
double C[2][2]; /* Matrix C */
double X[2]; /* Matrix X */
double det_C0_C1, /* Determinants of matrices */
det_C0_X,
det_X_C1;
double alpha_l, /* Alpha values, left and right */
alpha_r;
Vector2 tmp; /* Utility variable */
BezierCurve bezCurve; /* RETURN bezier curve ctl pts */
bezCurve = (Point2 *)malloc(4 * sizeof(Point2));
nPts = last - first + 1;
/* Compute the A's */
for (i = 0; i < nPts; i++) {
Vector2 v1, v2;
v1 = tHat1;
v2 = tHat2;
V2Scale(&v1, B1(uPrime[i]));
V2Scale(&v2, B2(uPrime[i]));
A[i][0] = v1;
A[i][1] = v2;
}
/* Create the C and X matrices */
C[0][0] = 0.0;
C[0][1] = 0.0;
C[1][0] = 0.0;
C[1][1] = 0.0;
X[0] = 0.0;
X[1] = 0.0;
for (i = 0; i < nPts; i++) {
C[0][0] += V2Dot(&A[i][0], &A[i][0]);
C[0][1] += V2Dot(&A[i][0], &A[i][1]);
/* C[1][0] += V2Dot(&A[i][0], &A[i][1]);*/
C[1][0] = C[0][1];
C[1][1] += V2Dot(&A[i][1], &A[i][1]);
tmp = V2SubII(d[first + i],
V2AddII(
V2ScaleIII(d[first], B0(uPrime[i])),
V2AddII(
V2ScaleIII(d[first], B1(uPrime[i])),
V2AddII(
V2ScaleIII(d[last], B2(uPrime[i])),
V2ScaleIII(d[last], B3(uPrime[i]))))));
X[0] += V2Dot(&A[i][0], &tmp);
X[1] += V2Dot(&A[i][1], &tmp);
}
/* Compute the determinants of C and X */
det_C0_C1 = C[0][0] * C[1][1] - C[1][0] * C[0][1];
det_C0_X = C[0][0] * X[1] - C[1][0] * X[0];
det_X_C1 = X[0] * C[1][1] - X[1] * C[0][1];
/* Finally, derive alpha values */
alpha_l = (det_C0_C1 < ZERO_TOLERANCE) ? 0.0 : det_X_C1 / det_C0_C1;
alpha_r = (det_C0_C1 < ZERO_TOLERANCE) ? 0.0 : det_C0_X / det_C0_C1;
/* If alpha negative, use the Wu/Barsky heuristic (see text) */
/* (if alpha is 0, you get coincident control points that lead to
* divide by zero in any subsequent NewtonRaphsonRootFind() call. */
double segLength = V2DistanceBetween2Points(&d[last], &d[first]);
double epsilon = 1.0e-6 * segLength;
if (alpha_l < epsilon || alpha_r < epsilon)
{
/* fall back on standard (probably inaccurate) formula, and subdivide further if needed. */
double dist = segLength / 3.0;
bezCurve[0] = d[first];
bezCurve[3] = d[last];
V2Add(&bezCurve[0], V2Scale(&tHat1, dist), &bezCurve[1]);
V2Add(&bezCurve[3], V2Scale(&tHat2, dist), &bezCurve[2]);
return (bezCurve);
}
/* First and last control points of the Bezier curve are */
/* positioned exactly at the first and last data points */
/* Control points 1 and 2 are positioned an alpha distance out */
/* on the tangent vectors, left and right, respectively */
bezCurve[0] = d[first];
bezCurve[3] = d[last];
V2Add(&bezCurve[0], V2Scale(&tHat1, alpha_l), &bezCurve[1]);
V2Add(&bezCurve[3], V2Scale(&tHat2, alpha_r), &bezCurve[2]);
return (bezCurve);
}
void Trr2kNNNT
( UpperOrLower uplo,
Orientation orientationOfD,
T alpha, const DistMatrix<T>& A, const DistMatrix<T>& B,
const DistMatrix<T>& C, const DistMatrix<T>& D,
T beta, DistMatrix<T>& E )
{
#ifndef RELEASE
PushCallStack("internal::Trr2kNNNT");
if( E.Height() != E.Width() || A.Width() != C.Width() ||
A.Height() != E.Height() || C.Height() != E.Height() ||
B.Width() != E.Width() || D.Height() != E.Width() ||
A.Width() != B.Height() || C.Width() != D.Width() )
throw std::logic_error("Nonconformal Trr2kNNNT");
#endif
const Grid& g = E.Grid();
DistMatrix<T> AL(g), AR(g),
A0(g), A1(g), A2(g);
DistMatrix<T> BT(g), B0(g),
BB(g), B1(g),
B2(g);
DistMatrix<T> CL(g), CR(g),
C0(g), C1(g), C2(g);
DistMatrix<T> DL(g), DR(g),
D0(g), D1(g), D2(g);
DistMatrix<T,MC, STAR> A1_MC_STAR(g);
DistMatrix<T,MR, STAR> B1Trans_MR_STAR(g);
DistMatrix<T,MC, STAR> C1_MC_STAR(g);
DistMatrix<T,VR, STAR> D1_VR_STAR(g);
DistMatrix<T,STAR,MR > D1AdjOrTrans_STAR_MR(g);
A1_MC_STAR.AlignWith( E );
B1Trans_MR_STAR.AlignWith( E );
C1_MC_STAR.AlignWith( E );
D1_VR_STAR.AlignWith( E );
D1AdjOrTrans_STAR_MR.AlignWith( E );
LockedPartitionRight( A, AL, AR, 0 );
LockedPartitionDown
( B, BT,
BB, 0 );
LockedPartitionRight( C, CL, CR, 0 );
LockedPartitionRight( D, DL, DR, 0 );
while( AL.Width() < A.Width() )
{
LockedRepartitionRight
( AL, /**/ AR,
A0, /**/ A1, A2 );
LockedRepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
LockedRepartitionRight
( CL, /**/ CR,
C0, /**/ C1, C2 );
LockedRepartitionRight
( CL, /**/ CR,
C0, /**/ C1, C2 );
//--------------------------------------------------------------------//
A1_MC_STAR = A1;
C1_MC_STAR = C1;
B1Trans_MR_STAR.TransposeFrom( B1 );
D1_VR_STAR = D1;
if( orientationOfD == ADJOINT )
D1AdjOrTrans_STAR_MR.AdjointFrom( D1_VR_STAR );
else
D1AdjOrTrans_STAR_MR.TransposeFrom( D1_VR_STAR );
LocalTrr2k
( uplo, TRANSPOSE,
alpha, A1_MC_STAR, B1Trans_MR_STAR,
C1_MC_STAR, D1AdjOrTrans_STAR_MR,
beta, E );
//--------------------------------------------------------------------//
SlideLockedPartitionRight
( DL, /**/ DR,
D0, D1, /**/ D2 );
SlideLockedPartitionRight
( CL, /**/ CR,
C0, C1, /**/ C2 );
SlideLockedPartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
SlideLockedPartitionRight
( AL, /**/ AR,
A0, A1, /**/ A2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
/*
* GenerateBezier :
* Use least-squares method to find Bezier control points for region.
*
*/
QPointF* GenerateBezier(const QList<QPointF> &points, int first, int last, qreal *uPrime, FitVector tHat1, FitVector tHat2)
{
int i;
int nPts; /* Number of pts in sub-curve */
qreal C[2][2]; /* Matrix C */
qreal X[2]; /* Matrix X */
qreal det_C0_C1, /* Determinants of matrices */
det_C0_X,
det_X_C1;
qreal alpha_l, /* Alpha values, left and right */
alpha_r;
FitVector tmp; /* Utility variable */
QPointF *curve;
curve = new QPointF[4];
nPts = last - first + 1;
/* Precomputed rhs for eqn */
// FitVector A[nPts][2]
QVector< QVector<FitVector> > A(nPts, QVector<FitVector>(2));
/* Compute the A's */
for (i = 0; i < nPts; ++i) {
FitVector v1, v2;
v1 = tHat1;
v2 = tHat2;
v1.scale(B1(uPrime[i]));
v2.scale(B2(uPrime[i]));
A[i][0] = v1;
A[i][1] = v2;
}
/* Create the C and X matrices */
C[0][0] = 0.0;
C[0][1] = 0.0;
C[1][0] = 0.0;
C[1][1] = 0.0;
X[0] = 0.0;
X[1] = 0.0;
for (i = 0; i < nPts; ++i) {
C[0][0] += (A[i][0]).dot(A[i][0]);
C[0][1] += A[i][0].dot(A[i][1]);
/* C[1][0] += V2Dot(&A[i][0], &A[i][1]);*/
C[1][0] = C[0][1];
C[1][1] += A[i][1].dot(A[i][1]);
FitVector vfirstp1(points.at(first + i));
FitVector vfirst(points.at(first));
FitVector vlast(points.at(last));
tmp = VectorSub(vfirstp1,
VectorAdd(
VectorScale(vfirst, B0(uPrime[i])),
VectorAdd(
VectorScale(vfirst, B1(uPrime[i])),
VectorAdd(
VectorScale(vlast, B2(uPrime[i])),
VectorScale(vlast, B3(uPrime[i]))))));
X[0] += A[i][0].dot(tmp);
X[1] += A[i][1].dot(tmp);
}
/* Compute the determinants of C and X */
det_C0_C1 = C[0][0] * C[1][1] - C[1][0] * C[0][1];
det_C0_X = C[0][0] * X[1] - C[0][1] * X[0];
det_X_C1 = X[0] * C[1][1] - X[1] * C[0][1];
/* Finally, derive alpha values */
if (qFuzzyCompare(det_C0_C1, qreal(0.0))) {
det_C0_C1 = (C[0][0] * C[1][1]) * 10e-12;
if (qFuzzyCompare(det_C0_C1, qreal(0.0))) {
det_C0_C1 = Zero;
}
}
alpha_l = det_X_C1 / det_C0_C1;
alpha_r = det_C0_X / det_C0_C1;
/* If alpha negative, use the Wu/Barsky heuristic (see text) */
/* (if alpha is 0, you get coincident control points that lead to
* divide by zero in any subsequent NewtonRaphsonRootFind() call. */
if (alpha_l < 1.0e-6 || alpha_r < 1.0e-6) {
qreal dist = distance(points.at(last), points.at(first)) / 3.0;
curve[0] = points.at(first);
curve[3] = points.at(last);
tHat1.scale(dist);
tHat2.scale(dist);
curve[1] = tHat1 + curve[0];
curve[2] = tHat2 + curve[3];
return curve;
}
/* First and last control points of the Bezier curve are */
/* positioned exactly at the first and last data points */
/* Control points 1 and 2 are positioned an alpha distance out */
/* on the tangent vectors, left and right, respectively */
curve[0] = points.at(first);
curve[3] = points.at(last);
tHat1.scale(alpha_l);
tHat2.scale(alpha_r);
curve[1] = tHat1 + curve[0];
curve[2] = tHat2 + curve[3];
return (curve);
}
inline void
GemmTTA
( Orientation orientationOfA,
Orientation orientationOfB,
T alpha, const DistMatrix<T>& A,
const DistMatrix<T>& B,
T beta, DistMatrix<T>& C )
{
#ifndef RELEASE
PushCallStack("internal::GemmTTA");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
if( orientationOfA == NORMAL || orientationOfB == NORMAL )
throw std::logic_error
("GemmTTA expects A and B to be (Conjugate)Transposed");
if( A.Width() != C.Height() ||
B.Height() != C.Width() ||
A.Height() != B.Width() )
{
std::ostringstream msg;
msg << "Nonconformal GemmTTA: \n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str().c_str() );
}
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T> BT(g), B0(g),
BB(g), B1(g),
B2(g);
DistMatrix<T> CL(g), CR(g),
C0(g), C1(g), C2(g);
// Temporary distributions
DistMatrix<T,STAR,MC > B1_STAR_MC(g);
DistMatrix<T,MR, STAR> D1_MR_STAR(g);
DistMatrix<T,MR, MC > D1_MR_MC(g);
DistMatrix<T> D1(g);
B1_STAR_MC.AlignWith( A );
D1_MR_STAR.AlignWith( A );
// Start the algorithm
Scale( beta, C );
LockedPartitionDown
( B, BT,
BB, 0 );
PartitionRight( C, CL, CR, 0 );
while( BB.Height() > 0 )
{
LockedRepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
RepartitionRight
( CL, /**/ CR,
C0, /**/ C1, C2 );
D1.AlignWith( C1 );
Zeros( C1.Height(), C1.Width(), D1_MR_STAR );
//--------------------------------------------------------------------//
B1_STAR_MC = B1; // B1[*,MC] <- B1[MC,MR]
// D1[MR,*] := alpha (A[MC,MR])^T (B1[*,MC])^T
// = alpha (A^T)[MR,MC] (B1^T)[MC,*]
LocalGemm
( orientationOfA, orientationOfB,
alpha, A, B1_STAR_MC, T(0), D1_MR_STAR );
// C1[MC,MR] += scattered & transposed D1[MR,*] summed over grid cols
D1_MR_MC.SumScatterFrom( D1_MR_STAR );
D1 = D1_MR_MC;
Axpy( T(1), D1, C1 );
//--------------------------------------------------------------------//
D1.FreeAlignments();
SlideLockedPartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
SlidePartitionRight
( CL, /**/ CR,
C0, C1, /**/ C2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
GemmNNDot
( T alpha, const DistMatrix<T>& A,
const DistMatrix<T>& B,
T beta, DistMatrix<T>& C )
{
#ifndef RELEASE
PushCallStack("internal::GemmNNDot");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
if( A.Height() != C.Height() ||
B.Width() != C.Width() ||
A.Width() != B.Height() )
{
std::ostringstream msg;
msg << "Nonconformal GemmNNDot: \n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str().c_str() );
}
#endif
const Grid& g = A.Grid();
if( A.Height() > B.Width() )
{
// Matrix views
DistMatrix<T> AT(g), AB(g),
A0(g), A1(g), A2(g);
DistMatrix<T> BL(g), B0(g),
BR(g), B1(g),
B2(g);
DistMatrix<T> CT(g), C0(g), C1L(g), C1R(g),
CB(g), C1(g), C10(g), C11(g), C12(g),
C2(g);
// Temporary distributions
DistMatrix<T,STAR,VC> A1_STAR_VC(g);
DistMatrix<T,VC,STAR> B1_VC_STAR(g);
DistMatrix<T,STAR,STAR> C11_STAR_STAR(g);
// Star the algorithm
Scale( beta, C );
LockedPartitionDown
( A, AT,
AB, 0 );
PartitionDown
( C, CT,
CB, 0 );
while( AB.Height() > 0 )
{
LockedRepartitionDown
( AT, A0,
/**/ /**/
A1,
AB, A2 );
RepartitionDown
( CT, C0,
/**/ /**/
C1,
CB, C2 );
A1_STAR_VC = A1;
B1_VC_STAR.AlignWith( A1_STAR_VC );
LockedPartitionRight( B, BL, BR, 0 );
PartitionRight( C1, C1L, C1R, 0 );
while( BR.Width() > 0 )
{
LockedRepartitionRight
( BL, /**/ BR,
B0, /**/ B1, B2 );
RepartitionRight
( C1L, /**/ C1R,
C10, /**/ C11, C12 );
Zeros( C11.Height(), C11.Width(), C11_STAR_STAR );
//------------------------------------------------------------//
B1_VC_STAR = B1;
LocalGemm
( NORMAL, NORMAL,
alpha, A1_STAR_VC, B1_VC_STAR, T(0), C11_STAR_STAR );
C11.SumScatterUpdate( T(1), C11_STAR_STAR );
//------------------------------------------------------------//
SlideLockedPartitionRight
( BL, /**/ BR,
B0, B1, /**/ B2 );
SlidePartitionRight
( C1L, /**/ C1R,
C10, C11, /**/ C12 );
}
B1_VC_STAR.FreeAlignments();
SlideLockedPartitionDown
( AT, A0,
A1,
/**/ /**/
AB, A2 );
SlidePartitionDown
( CT, C0,
C1,
/**/ /**/
CB, C2 );
}
}
else
{
// Matrix views
DistMatrix<T> AT(g), AB(g),
A0(g), A1(g), A2(g);
DistMatrix<T> BL(g), B0(g),
BR(g), B1(g),
B2(g);
DistMatrix<T>
CL(g), CR(g), C1T(g), C01(g),
C0(g), C1(g), C2(g), C1B(g), C11(g),
C21(g);
// Temporary distributions
DistMatrix<T,STAR,VR> A1_STAR_VR(g);
DistMatrix<T,VR,STAR> B1_VR_STAR(g);
DistMatrix<T,STAR,STAR> C11_STAR_STAR(g);
// Star the algorithm
Scale( beta, C );
LockedPartitionRight( B, BL, BR, 0 );
PartitionRight( C, CL, CR, 0 );
while( BR.Width() > 0 )
{
LockedRepartitionRight
( BL, /**/ BR,
B0, /**/ B1, B2 );
RepartitionRight
( CL, /**/ CR,
C0, /**/ C1, C2 );
B1_VR_STAR = B1;
A1_STAR_VR.AlignWith( B1_VR_STAR );
LockedPartitionDown
( A, AT,
AB, 0 );
PartitionDown
( C1, C1T,
C1B, 0 );
while( AB.Height() > 0 )
{
LockedRepartitionDown
( AT, A0,
/**/ /**/
A1,
AB, A2 );
RepartitionDown
( C1T, C01,
/***/ /***/
C11,
C1B, C21 );
Zeros( C11.Height(), C11.Width(), C11_STAR_STAR );
//------------------------------------------------------------//
A1_STAR_VR = A1;
LocalGemm
( NORMAL, NORMAL,
alpha, A1_STAR_VR, B1_VR_STAR, T(0), C11_STAR_STAR );
C11.SumScatterUpdate( T(1), C11_STAR_STAR );
//------------------------------------------------------------//
SlideLockedPartitionDown
( AT, A0,
A1,
/**/ /**/
AB, A2 );
SlidePartitionDown
( C1T, C01,
C11,
/***/ /***/
C1B, C21 );
}
A1_STAR_VR.FreeAlignments();
SlideLockedPartitionRight
( BL, /**/ BR,
B0, B1, /**/ B2 );
SlidePartitionRight
( CL, /**/ CR,
C0, C1, /**/ C2 );
}
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
GemmNNC
( T alpha, const DistMatrix<T>& A,
const DistMatrix<T>& B,
T beta, DistMatrix<T>& C )
{
#ifndef RELEASE
PushCallStack("internal::GemmNNC");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
if( A.Height() != C.Height() ||
B.Width() != C.Width() ||
A.Width() != B.Height() )
{
std::ostringstream msg;
msg << "Nonconformal GemmNNC: \n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str().c_str() );
}
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T> AL(g), AR(g),
A0(g), A1(g), A2(g);
DistMatrix<T> BT(g), B0(g),
BB(g), B1(g),
B2(g);
// Temporary distributions
DistMatrix<T,MC,STAR> A1_MC_STAR(g);
DistMatrix<T,MR,STAR> B1Trans_MR_STAR(g);
A1_MC_STAR.AlignWith( C );
B1Trans_MR_STAR.AlignWith( C );
// Start the algorithm
Scale( beta, C );
LockedPartitionRight( A, AL, AR, 0 );
LockedPartitionDown
( B, BT,
BB, 0 );
while( AR.Width() > 0 )
{
LockedRepartitionRight( AL, /**/ AR,
A0, /**/ A1, A2 );
LockedRepartitionDown( BT, B0,
/**/ /**/
B1,
BB, B2 );
//--------------------------------------------------------------------//
A1_MC_STAR = A1;
B1Trans_MR_STAR.TransposeFrom( B1 );
// C[MC,MR] += alpha A1[MC,*] (B1^T[MR,*])^T
// = alpha A1[MC,*] B1[*,MR]
LocalGemm
( NORMAL, TRANSPOSE, alpha, A1_MC_STAR, B1Trans_MR_STAR, T(1), C );
//--------------------------------------------------------------------//
SlideLockedPartitionRight( AL, /**/ AR,
A0, A1, /**/ A2 );
SlideLockedPartitionDown( BT, B0,
B1,
/**/ /**/
BB, B2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
SymmLLA
( T alpha, const DistMatrix<T>& A, const DistMatrix<T>& B,
T beta, DistMatrix<T>& C )
{
#ifndef RELEASE
PushCallStack("internal::SymmLLA");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
#endif
const Grid& g = A.Grid();
DistMatrix<T>
BL(g), BR(g),
B0(g), B1(g), B2(g);
DistMatrix<T>
CL(g), CR(g),
C0(g), C1(g), C2(g);
DistMatrix<T,MC,STAR> B1_MC_STAR(g);
DistMatrix<T,VR,STAR> B1_VR_STAR(g);
DistMatrix<T,STAR,MR> B1Trans_STAR_MR(g);
DistMatrix<T> Z1(g);
DistMatrix<T,MC,STAR> Z1_MC_STAR(g);
DistMatrix<T,MR,STAR> Z1_MR_STAR(g);
DistMatrix<T,MR,MC > Z1_MR_MC(g);
B1_MC_STAR.AlignWith( A );
B1_VR_STAR.AlignWith( A );
B1Trans_STAR_MR.AlignWith( A );
Z1_MC_STAR.AlignWith( A );
Z1_MR_STAR.AlignWith( A );
Scale( beta, C );
LockedPartitionRight
( B, BL, BR, 0 );
PartitionRight
( C, CL, CR, 0 );
while( CL.Width() < C.Width() )
{
LockedRepartitionRight
( BL, /**/ BR,
B0, /**/ B1, B2 );
RepartitionRight
( CL, /**/ CR,
C0, /**/ C1, C2 );
Z1.AlignWith( C1 );
Zeros( C1.Height(), C1.Width(), Z1_MC_STAR );
Zeros( C1.Height(), C1.Width(), Z1_MR_STAR );
//--------------------------------------------------------------------//
B1_MC_STAR = B1;
B1_VR_STAR = B1_MC_STAR;
B1Trans_STAR_MR.TransposeFrom( B1_VR_STAR );
LocalSymmetricAccumulateLL
( TRANSPOSE,
alpha, A, B1_MC_STAR, B1Trans_STAR_MR, Z1_MC_STAR, Z1_MR_STAR );
Z1_MR_MC.SumScatterFrom( Z1_MR_STAR );
Z1 = Z1_MR_MC;
Z1.SumScatterUpdate( T(1), Z1_MC_STAR );
Axpy( T(1), Z1, C1 );
//--------------------------------------------------------------------//
Z1.FreeAlignments();
SlideLockedPartitionRight
( BL, /**/ BR,
B0, B1, /**/ B2 );
SlidePartitionRight
( CL, /**/ CR,
C0, C1, /**/ C2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
Her2kLN
( T alpha, const DistMatrix<T,MC,MR>& A,
const DistMatrix<T,MC,MR>& B,
T beta, DistMatrix<T,MC,MR>& C )
{
#ifndef RELEASE
PushCallStack("internal::Her2kLN");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
if( A.Height() != C.Height() || A.Height() != C.Width() ||
B.Height() != C.Height() || B.Height() != C.Width() ||
A.Width() != B.Width() )
{
std::ostringstream msg;
msg << "Nonconformal Her2kLN:\n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str() );
}
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T,MC,MR> AL(g), AR(g),
A0(g), A1(g), A2(g);
DistMatrix<T,MC,MR> BL(g), BR(g),
B0(g), B1(g), B2(g);
// Temporary distributions
DistMatrix<T,MC, STAR> A1_MC_STAR(g);
DistMatrix<T,MC, STAR> B1_MC_STAR(g);
DistMatrix<T,VR, STAR> A1_VR_STAR(g);
DistMatrix<T,VR, STAR> B1_VR_STAR(g);
DistMatrix<T,STAR,MR > A1Adj_STAR_MR(g);
DistMatrix<T,STAR,MR > B1Adj_STAR_MR(g);
A1_MC_STAR.AlignWith( C );
B1_MC_STAR.AlignWith( C );
A1_VR_STAR.AlignWith( C );
B1_VR_STAR.AlignWith( C );
A1Adj_STAR_MR.AlignWith( C );
B1Adj_STAR_MR.AlignWith( C );
// Start the algorithm
ScaleTrapezoid( beta, LEFT, LOWER, 0, C );
LockedPartitionRight( A, AL, AR, 0 );
LockedPartitionRight( B, BL, BR, 0 );
while( AR.Width() > 0 )
{
LockedRepartitionRight
( AL, /**/ AR,
A0, /**/ A1, A2 );
LockedRepartitionRight
( BL, /**/ BR,
B0, /**/ B1, B2 );
//--------------------------------------------------------------------//
A1_VR_STAR = A1_MC_STAR = A1;
A1Adj_STAR_MR.AdjointFrom( A1_VR_STAR );
B1_VR_STAR = B1_MC_STAR = B1;
B1Adj_STAR_MR.AdjointFrom( B1_VR_STAR );
LocalTrr2k
( LOWER,
alpha, A1_MC_STAR, B1Adj_STAR_MR,
B1_MC_STAR, A1Adj_STAR_MR,
T(1), C );
//--------------------------------------------------------------------//
SlideLockedPartitionRight
( AL, /**/ AR,
A0, A1, /**/ A2 );
SlideLockedPartitionRight
( BL, /**/ BR,
B0, B1, /**/ B2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
Syr2kUT
( T alpha, const DistMatrix<T>& A, const DistMatrix<T>& B,
T beta, DistMatrix<T>& C,
bool conjugate=false )
{
#ifndef RELEASE
CallStackEntry entry("internal::Syr2kUT");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
if( A.Width() != C.Height() ||
A.Width() != C.Width() ||
B.Width() != C.Height() ||
B.Width() != C.Width() ||
A.Height() != B.Height() )
{
std::ostringstream msg;
msg << "Nonconformal Syr2kUT:\n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str().c_str() );
}
#endif
const Grid& g = A.Grid();
const Orientation orientation = ( conjugate ? ADJOINT : TRANSPOSE );
// Matrix views
DistMatrix<T> AT(g), A0(g),
AB(g), A1(g),
A2(g);
DistMatrix<T> BT(g), B0(g),
BB(g), B1(g),
B2(g);
// Temporary distributions
DistMatrix<T,MR, STAR> A1Trans_MR_STAR(g);
DistMatrix<T,MR, STAR> B1Trans_MR_STAR(g);
DistMatrix<T,STAR,VR > A1_STAR_VR(g);
DistMatrix<T,STAR,VR > B1_STAR_VR(g);
DistMatrix<T,STAR,MC > A1_STAR_MC(g);
DistMatrix<T,STAR,MC > B1_STAR_MC(g);
A1Trans_MR_STAR.AlignWith( C );
B1Trans_MR_STAR.AlignWith( C );
A1_STAR_MC.AlignWith( C );
B1_STAR_MC.AlignWith( C );
// Start the algorithm
ScaleTrapezoid( beta, LEFT, UPPER, 0, C );
LockedPartitionDown
( A, AT,
AB, 0 );
LockedPartitionDown
( B, BT,
BB, 0 );
while( AB.Height() > 0 )
{
LockedRepartitionDown
( AT, A0,
/**/ /**/
A1,
AB, A2 );
LockedRepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
//--------------------------------------------------------------------//
A1Trans_MR_STAR.TransposeFrom( A1 );
A1_STAR_VR.TransposeFrom( A1Trans_MR_STAR );
A1_STAR_MC = A1_STAR_VR;
B1Trans_MR_STAR.TransposeFrom( B1 );
B1_STAR_VR.TransposeFrom( B1Trans_MR_STAR );
B1_STAR_MC = B1_STAR_VR;
LocalTrr2k
( UPPER, orientation, TRANSPOSE, orientation, TRANSPOSE,
alpha, A1_STAR_MC, B1Trans_MR_STAR,
B1_STAR_MC, A1Trans_MR_STAR,
T(1), C );
//--------------------------------------------------------------------//
SlideLockedPartitionDown
( AT, A0,
A1,
/**/ /**/
AB, A2 );
SlideLockedPartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
}
}
inline void
GemmTTC
( Orientation orientationOfA,
Orientation orientationOfB,
T alpha, const DistMatrix<T>& A,
const DistMatrix<T>& B,
T beta, DistMatrix<T>& C )
{
#ifndef RELEASE
PushCallStack("internal::GemmTTC");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
if( orientationOfA == NORMAL || orientationOfB == NORMAL )
throw std::logic_error
("GemmTTC expects A and B to be (Conjugate)Transposed");
if( A.Width() != C.Height() ||
B.Height() != C.Width() ||
A.Height() != B.Width() )
{
std::ostringstream msg;
msg << "Nonconformal GemmTTC: \n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str() );
}
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T> AT(g), A0(g),
AB(g), A1(g),
A2(g);
DistMatrix<T> BL(g), BR(g),
B0(g), B1(g), B2(g);
// Temporary distributions
DistMatrix<T,STAR,MC > A1_STAR_MC(g);
DistMatrix<T,VR, STAR> B1_VR_STAR(g);
DistMatrix<T,STAR,MR > B1AdjOrTrans_STAR_MR(g);
A1_STAR_MC.AlignWith( C );
B1_VR_STAR.AlignWith( C );
B1AdjOrTrans_STAR_MR.AlignWith( C );
// Start the algorithm
Scale( beta, C );
LockedPartitionDown
( A, AT,
AB, 0 );
LockedPartitionRight( B, BL, BR, 0 );
while( AB.Height() > 0 )
{
LockedRepartitionDown
( AT, A0,
/**/ /**/
A1,
AB, A2 );
LockedRepartitionRight
( BL, /**/ BR,
B0, /**/ B1, B2 );
//--------------------------------------------------------------------//
A1_STAR_MC = A1;
B1_VR_STAR = B1;
if( orientationOfB == ADJOINT )
B1AdjOrTrans_STAR_MR.AdjointFrom( B1_VR_STAR );
else
B1AdjOrTrans_STAR_MR.TransposeFrom( B1_VR_STAR );
// C[MC,MR] += alpha (A1[*,MC])^[T/H] (B1[MR,*])^[T/H]
// = alpha (A1^[T/H])[MC,*] (B1^[T/H])[*,MR]
LocalGemm
( orientationOfA, NORMAL,
alpha, A1_STAR_MC, B1AdjOrTrans_STAR_MR, T(1), C );
//--------------------------------------------------------------------//
SlideLockedPartitionDown
( AT, A0,
A1,
/**/ /**/
AB, A2 );
SlideLockedPartitionRight
( BL, /**/ BR,
B0, B1, /**/ B2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
SymmLLC
( T alpha, const DistMatrix<T>& A, const DistMatrix<T>& B,
T beta, DistMatrix<T>& C )
{
#ifndef RELEASE
PushCallStack("internal::SymmLLC");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T>
ATL(g), ATR(g), A00(g), A01(g), A02(g), AColPan(g),
ABL(g), ABR(g), A10(g), A11(g), A12(g), ARowPan(g),
A20(g), A21(g), A22(g);
DistMatrix<T>
BT(g), B0(g),
BB(g), B1(g),
B2(g);
DistMatrix<T>
CT(g), C0(g), CAbove(g),
CB(g), C1(g), CBelow(g),
C2(g);
// Temporary distributions
DistMatrix<T,MC, STAR> AColPan_MC_STAR(g);
DistMatrix<T,STAR,MC > ARowPan_STAR_MC(g);
DistMatrix<T,MR, STAR> B1Trans_MR_STAR(g);
B1Trans_MR_STAR.AlignWith( C );
// Start the algorithm
Scale( beta, C );
LockedPartitionDownDiagonal
( A, ATL, ATR,
ABL, ABR, 0 );
LockedPartitionDown
( B, BT,
BB, 0 );
PartitionDown
( C, CT,
CB, 0 );
while( CB.Height() > 0 )
{
LockedRepartitionDownDiagonal
( ATL, /**/ ATR, A00, /**/ A01, A02,
/*************/ /******************/
/**/ A10, /**/ A11, A12,
ABL, /**/ ABR, A20, /**/ A21, A22 );
LockedRepartitionDown
( BT, B0,
/**/ /**/
B1,
BB, B2 );
RepartitionDown
( CT, C0,
/**/ /**/
C1,
CB, C2 );
LockedView1x2( ARowPan, A10, A11 );
LockedView2x1
( AColPan, A11,
A21 );
View2x1
( CAbove, C0,
C1 );
View2x1
( CBelow, C1,
C2 );
AColPan_MC_STAR.AlignWith( CBelow );
ARowPan_STAR_MC.AlignWith( CAbove );
//--------------------------------------------------------------------//
AColPan_MC_STAR = AColPan;
ARowPan_STAR_MC = ARowPan;
MakeTrapezoidal( LEFT, LOWER, 0, AColPan_MC_STAR );
MakeTrapezoidal( RIGHT, LOWER, -1, ARowPan_STAR_MC );
B1Trans_MR_STAR.TransposeFrom( B1 );
LocalGemm
( NORMAL, TRANSPOSE,
alpha, AColPan_MC_STAR, B1Trans_MR_STAR, T(1), CBelow );
LocalGemm
( TRANSPOSE, TRANSPOSE,
alpha, ARowPan_STAR_MC, B1Trans_MR_STAR, T(1), CAbove );
//--------------------------------------------------------------------//
AColPan_MC_STAR.FreeAlignments();
ARowPan_STAR_MC.FreeAlignments();
SlideLockedPartitionDownDiagonal
( ATL, /**/ ATR, A00, A01, /**/ A02,
/**/ A10, A11, /**/ A12,
/*************/ /******************/
ABL, /**/ ABR, A20, A21, /**/ A22 );
SlideLockedPartitionDown
( BT, B0,
B1,
/**/ /**/
BB, B2 );
SlidePartitionDown
( CT, C0,
C1,
/**/ /**/
CB, C2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
inline void
GemmNNA
( T alpha, const DistMatrix<T>& A,
const DistMatrix<T>& B,
T beta, DistMatrix<T>& C )
{
#ifndef RELEASE
PushCallStack("internal::GemmNNA");
if( A.Grid() != B.Grid() || B.Grid() != C.Grid() )
throw std::logic_error
("{A,B,C} must be distributed over the same grid");
if( A.Height() != C.Height() ||
B.Width() != C.Width() ||
A.Width() != B.Height() )
{
std::ostringstream msg;
msg << "Nonconformal GemmNNA: \n"
<< " A ~ " << A.Height() << " x " << A.Width() << "\n"
<< " B ~ " << B.Height() << " x " << B.Width() << "\n"
<< " C ~ " << C.Height() << " x " << C.Width() << "\n";
throw std::logic_error( msg.str().c_str() );
}
#endif
const Grid& g = A.Grid();
// Matrix views
DistMatrix<T> BL(g), BR(g),
B0(g), B1(g), B2(g);
DistMatrix<T> CL(g), CR(g),
C0(g), C1(g), C2(g);
// Temporary distributions
DistMatrix<T,VR,STAR> B1_VR_STAR(g);
DistMatrix<T,STAR,MR> B1Trans_STAR_MR(g);
DistMatrix<T,MC,STAR> D1_MC_STAR(g);
B1_VR_STAR.AlignWith( A );
B1Trans_STAR_MR.AlignWith( A );
D1_MC_STAR.AlignWith( A );
// Start the algorithm
Scale( beta, C );
LockedPartitionRight( B, BL, BR, 0 );
PartitionRight( C, CL, CR, 0 );
while( BR.Width() > 0 )
{
LockedRepartitionRight
( BL, /**/ BR,
B0, /**/ B1, B2 );
RepartitionRight
( CL, /**/ CR,
C0, /**/ C1, C2 );
Zeros( C1.Height(), C1.Width(), D1_MC_STAR );
//--------------------------------------------------------------------//
B1_VR_STAR = B1;
B1Trans_STAR_MR.TransposeFrom( B1_VR_STAR );
// D1[MC,*] := alpha A[MC,MR] B1[MR,*]
LocalGemm
( NORMAL, TRANSPOSE, alpha, A, B1Trans_STAR_MR, T(0), D1_MC_STAR );
// C1[MC,MR] += scattered result of D1[MC,*] summed over grid rows
C1.SumScatterUpdate( T(1), D1_MC_STAR );
//--------------------------------------------------------------------//
SlideLockedPartitionRight
( BL, /**/ BR,
B0, B1, /**/ B2 );
SlidePartitionRight
( CL, /**/ CR,
C0, C1, /**/ C2 );
}
#ifndef RELEASE
PopCallStack();
#endif
}
