uint64_t t1ha2_atonce(const void *data, size_t length, uint64_t seed) {
t1ha_state256_t state;
init_ab(&state, seed, length);
#if T1HA_SYS_UNALIGNED_ACCESS == T1HA_UNALIGNED_ACCESS__EFFICIENT
if (unlikely(length > 32)) {
init_cd(&state, seed, length);
T1HA2_LOOP(le, unaligned, &state, data, length);
squash(&state);
length &= 31;
}
T1HA2_TAIL_AB(le, unaligned, &state, data, length);
#else
const bool misaligned = (((uintptr_t)data) & (ALIGNMENT_64 - 1)) != 0;
if (misaligned) {
if (unlikely(length > 32)) {
init_cd(&state, seed, length);
T1HA2_LOOP(le, unaligned, &state, data, length);
squash(&state);
length &= 31;
}
T1HA2_TAIL_AB(le, unaligned, &state, data, length);
} else {
if (unlikely(length > 32)) {
init_cd(&state, seed, length);
T1HA2_LOOP(le, aligned, &state, data, length);
squash(&state);
length &= 31;
}
T1HA2_TAIL_AB(le, aligned, &state, data, length);
}
#endif
}
void ConcatPeephole::append(const Bytecode& op,
const State& state,
const std::vector<Op>& srcStack) {
FTRACE(1, "ConcatPeephole::append {}\n", show(op));
assert(state.stack.size() == srcStack.size());
int nstack = state.stack.size();
// Size of the stack at the previous Concat.
int prevsz = m_working.empty() ? -1 : m_working.back().stacksz;
// Squash the innermost concat stream if we consumed its concat result.
if (nstack < prevsz - 1 || (nstack == prevsz - 1 &&
srcStack[nstack - 1] != Op::Concat)) {
squash();
}
if (op.op == Op::Concat) {
auto ind1 = nstack - 1;
auto ind2 = nstack - 2;
// Non-string concat; just append, squashing if this terminates a stream.
if (!state.stack[ind1].subtypeOf(TStr) ||
!state.stack[ind2].subtypeOf(TStr)) {
if (nstack == prevsz) {
squash();
}
return push_back(op);
}
// If the first concat operand is from the previous concat in the stream,
// continue the current stream.
if (srcStack[ind2] == Op::Concat && nstack == prevsz) {
return push_back(op, true);
}
// Correction for cases where we might have bizarre opcode sequences like
// [stk: 2] Concat, [stk: 1] CGetL2, [stk: 2] Concat, where it's unsafe
// to reorder.
if (nstack == prevsz) {
squash();
}
// Start a new stream.
m_working.push_back({{}, nstack, 0});
return push_back(op, true);
}
// Just push by default.
push_back(op);
}
int crook(d00d *master) {
d00d *copy;
b8 *mc, *cc;
b8 pos, val=0, dummy;
if(!(copy=(d00d *)malloc(sizeof(d00d)))) {
fputs("crook: RAM denied, 0", stderr);
exit(0);
}
if(first(master, &pos)) {
mc=&master->i_v[pos];
cc=©->i_v[pos];
} else return 0;
while(trial(mc, &val)) {
*copy=*master;
*cc=(((b8)val)|open|may_b[val]);
if(squash(copy)){
rm(mc,val);
/* should i check? :\ */
if(idea(master, &pos, &dummy)&&(dummy>val)) {
free(copy);
return 1;
}
} else val++;
}
free(copy);
return 0;
}
int DominatorTree::eval(int v)
{
if (ANCESTOR(v) < 0)
return v;
squash(v);
return LABEL(v);
}
static void bal()
{
assert(stage_context[1].decoded==NULL);
squash();
regs[8] = stage_context[2].pc+1;//save return address
reg_valid_bits[8]=1;
PC=stage_context[2].decoded->literal + stage_context[2].decoded->src1;
}
void apply(database &db, const operation &o, const options_type &opts) {
auto undo_session = db.start_undo_session(!(opts &
skip_undo_operation));
db.pre_apply_operation(o);
o.visit(apply_operation_visitor(db, opts));
db.post_apply_operation(o);
undo_session.squash();
}
static void bz()
{
if(ZF)
{//branch is taken
//squash the instructions in fetch stage and D/RF stage by turning them into nop
assert( stage_context[1].decoded==NULL);
squash();
PC = stage_context[2].pc+stage_context[2].decoded->literal;
}
}
Stretch::Stretch() {
int pi=0;
for (int x=-2047; x<=2047; ++x) { // invert squash()
int i=squash(x);
for (int j=pi; j<=i; ++j)
t[j]=x;
pi=i+1;
}
t[4095]=2047;
}
void DominatorTree::squash(int v)
{
if (ANCESTOR(ANCESTOR(v)) >= 0) {
squash(ANCESTOR(v));
if (SEMI(LABEL(ANCESTOR(v))) < SEMI(LABEL(v)))
LABEL(v) = LABEL(ANCESTOR(v));
ANCESTOR(v) = ANCESTOR(ANCESTOR(v));
}
}
/* solves input dudoku by hook or crook coroutines
returns 1 if input is incorrect
*/
int main(int argc, char **argv){
d00d *master;
char *in;
if(!(master = (d00d *)malloc(sizeof(d00d)))) {
fputs("main: RAM denied, 0", stderr);
return 0;
}
in=argc>1?argv[1]:def;
D00D(in, master);
if(squash(master)) fputs("invalid constraints\n",stderr);
free(master);
return 0;
}
void StaticLstmLayer::gradient(Parameters& w, Parameters& grad, FwdState&, BwdState& d, Matrix&, Matrix& x, Matrix& ) {
//! \f$\frac{dE}{dW_ZX} += \frac{dE}{da_Z} * x(t)\f$
//! \f$\frac{dE}{dW_FX} += \frac{dE}{da_F} * x(t)\f$
//! \f$\frac{dE}{dW_IX} += \frac{dE}{da_I} * x(t)\f$
//! \f$\frac{dE}{dW_OX} += \frac{dE}{da_O} * x(t)\f$
grad.IX.set_all_elements_to(0.0);
grad.ZX.set_all_elements_to(0.0);
grad.FX.set_all_elements_to(0.0);
grad.OX.set_all_elements_to(0.0);
for (size_t t(1); t < x.n_slices; ++t) {
mult_add(d.Za.slice(t), x.row_slice(w.I_bias.n_rows, x.n_rows).slice(t).T(), grad.ZX); // 1.0 / 1.0); //(double) n_time);
mult_add(d.Fa.slice(t), x.row_slice(w.I_bias.n_rows, x.n_rows).slice(t).T(), grad.FX); // 1.0 / 1.0); //(double) n_time);
mult_add(d.Ia.slice(t), x.row_slice(w.I_bias.n_rows, x.n_rows).slice(t).T(), grad.IX); //1.0 / 1.0); //(double) n_time);
mult_add(d.Oa.slice(t), x.row_slice(w.I_bias.n_rows, x.n_rows).slice(t).T(), grad.OX); // 1.0 / 1.0); //(double) n_time);
}
squash(d.Ia, grad.I_bias); //, 1.0 / (double) n_time);
squash(d.Fa, grad.F_bias); //, 1.0 / (double) n_time);
squash(d.Za, grad.Z_bias); //, 1.0 / (double) n_time);
squash(d.Oa, grad.O_bias); //, 1.0 / (double)n_time); //, 1.0 / (double) n_time);
}
void BBFind::clipSquash(Map3 &map, int numPasses, float initialMax)
{
// This gently reigns in the higher values while minimizing clipping.
// Brings values into the range 0 - 1
float inc = (initialMax - 1.0f) / (numPasses * 2);
for(int p = 0; p < numPasses; p++)
{
clip(map, 0.0f, initialMax);
initialMax -= inc;
squash(map, 0.0f, initialMax);
initialMax -= inc;
}
}
const char *pick_move(MazeMap *mm, int distsq)
{
Point dst;
find_distance(mm, dist, mm->loc.r, mm->loc.c);
if (mm_count_squares(mm) < WIDTH*HEIGHT)
dst = explore(mm);
else /* mm_count_squares(mm) == WIDTH*HEIGHT */
dst = squash(mm, distsq);
if (dst.r == mm->loc.r && dst.c == mm->loc.c)
return "T";
return construct_turn(mm, dist, mm->loc.r, mm->loc.c, mm->dir, dst.r, dst.c,
NULL, NULL);
}
void BBFind::accumulateIntoPrev(Map3 &prevMap, const Map3 ¤tMap, float accumulateAmt, float frameMemory, float scaleMin, float scaleMax)
{
// This is the entire update step for the previous confidence values
// buffer. We add currentMap * accumulateAmt to the prevMap buffer,
// then decay it by e ^ (-1 / frameMemory). Lastly, we clip any values
// that have grown too large and squash the values back to an acceptable
// range. This ends up functioning as a form of competition, as large
// new values cause older values to be squashed.
int numCategories = currentMap.size();
int mapWidth = currentMap[0][0].size();
int mapHeight = currentMap[0].size();
float fadeFactor = exp(-1.0f / frameMemory);
if(prevMap.empty())
{
prevMap = currentMap;
for(int c = 0; c < numCategories; c++)
{
for(int y = 0; y < mapHeight; y++)
{
for(int x = 0; x < mapWidth; x++)
{
prevMap[c][y][x] = scaleMin;
}
}
}
}
#ifdef PV_USE_OPENMP_THREADS
#pragma omp parallel for
#endif
for(int c = 0; c < numCategories; c++)
{
for(int y = 0; y < mapHeight; y++)
{
for(int x = 0; x < mapWidth; x++)
{
float val = (prevMap[c][y][x] + currentMap[c][y][x] * accumulateAmt) * fadeFactor;
if(val > scaleMax) val = scaleMax + (val - scaleMax) * accumulateAmt; //hard knee
prevMap[c][y][x] = val;
}
}
}
clip(prevMap, scaleMin, scaleMax * 1.5f);
squash(prevMap, scaleMin, scaleMax);
}
void
down (BOARD *b)
{
int i;
int s[4];
for (i = 0; i < 4; i++)
{
s[0] = b->cells[3][i];
s[1] = b->cells[2][i];
s[2] = b->cells[1][i];
s[3] = b->cells[0][i];
b->score += squash (s);
b->cells[3][i] = s[0];
b->cells[2][i] = s[1];
b->cells[1][i] = s[2];
b->cells[0][i] = s[3];
}
}
void
right (BOARD *b)
{
int i;
int s[4];
for (i = 0; i < 4; i++)
{
s[0] = b->cells[i][3];
s[1] = b->cells[i][2];
s[2] = b->cells[i][1];
s[3] = b->cells[i][0];
b->score += squash (s);
b->cells[i][3] = s[0];
b->cells[i][2] = s[1];
b->cells[i][1] = s[2];
b->cells[i][0] = s[3];
}
}
void apply(database &db, const signed_transaction &t, const options_type &opts) {
auto undo_session = db.start_undo_session(!(opts &
skip_undo_transaction));
db.pre_apply_transaction(t);
db.create<transaction_object>([&](transaction_object &trx) {
trx.trx_id = t.id();
trx.block_num = db.head_block().block_num;
auto pack_size = fc::raw::pack_size(t);
trx.packed_transaction.resize(pack_size);
fc::datastream<char *> ds(trx.packed_transaction.data(), pack_size);
fc::raw::pack(ds, t);
});
for (const auto &op : t.operations) {
apply(db, op, opts);
}
db.post_apply_transaction(t);
undo_session.squash();
}
void bpnn_layerforward(float *l1, float *l2, float **conn, int n1, int n2)
{
float sum;
int j, k;
/*** Set up thresholding unit ***/
l1[0] = 1.0;
#ifdef OPEN
omp_set_num_threads(NUM_THREAD);
#pragma omp parallel for shared(conn, n1, n2, l1) private(k, j) reduction(+: sum) schedule(static)
#endif
/*** For each unit in second layer ***/
for (j = 1; j <= n2; j++) {
/*** Compute weighted sum of its inputs ***/
sum = 0.0;
for (k = 0; k <= n1; k++) {
sum += conn[k][j] * l1[k];
}
l2[j] = squash(sum);
}
}
// show compression component statistics
int Predictor::stat(int id) {
printf("Memory utilization:\n");
int cp=7;
for (int i=0; i<z.header[6]; ++i) {
assert(cp<z.header.isize());
int type=z.header[cp];
assert(compsize[type]>0);
printf("%2d %s", i, compname[type]);
for (int j=1; j<compsize[type]; ++j)
printf(" %d", z.header[cp+j]);
Component& cr=comp[i];
if (type==MATCH) {
size_t count=0;
for (size_t j=0; j<cr.cm.size(); ++j)
if (cr.cm[j]) ++count;
printf(": buffer=%1.0f/%1.0f index=%1.0f/%1.0f (%1.2f%%)",
cr.limit/8.0, double(cr.ht.size()), double(count), double(cr.cm.size()),
count*100.0/cr.cm.size());
}
else if (type==SSE) {
size_t count=0;
for (size_t j=0; j<cr.cm.size(); ++j) {
if (int(cr.cm[j])!=(squash((j&31)*64-992)<<17|z.header[cp+3]))
++count;
}
printf(": %1.0f/%1.0f (%1.2f%%)", double(count),
double(cr.cm.size()), count*100.0/cr.cm.size());
}
else if (type==CM) {
size_t count=0;
for (size_t j=0; j<cr.cm.size(); ++j)
if (cr.cm[j]!=0x80000000) ++count;
printf(": %1.0f/%1.0f (%1.2f%%)", double(count),
double(cr.cm.size()), count*100.0/cr.cm.size());
}
else if (type==MIX) {
size_t count=0;
int m=z.header[cp+3];
assert(m>0);
for (size_t j=0; j<cr.cm.size(); ++j)
if (int(cr.cm[j])!=65536/m) ++count;
printf(": %1.0f/%1.0f (%1.2f%%)", double(count),
double(cr.cm.size()), count*100.0/cr.cm.size());
}
else if (type==MIX2) {
size_t count=0;
for (size_t j=0; j<cr.a16.size(); ++j)
if (int(cr.a16[j])!=32768) ++count;
printf(": %1.0f/%1.0f (%1.2f%%)", double(count),
double(cr.a16.size()), count*100.0/cr.a16.size());
}
else if (cr.ht.size()>0) {
double hcount=0;
for (size_t j=0; j<cr.ht.size(); ++j)
if (cr.ht[j]>0) ++hcount;
printf(": %1.0f/%1.0f (%1.2f%%)",
double(hcount), double(cr.ht.size()), hcount*100.0/cr.ht.size());
}
cp+=compsize[type];
printf("\n");
}
printf("\n");
return 0;
}
void LLOcclusionCullingGroup::doOcclusion(LLCamera* camera, const LLVector4a* shift)
{
LLGLDisable stencil(GL_STENCIL_TEST);
if (mSpatialPartition->isOcclusionEnabled() && LLPipeline::sUseOcclusion > 1)
{
//move mBounds to the agent space if necessary
LLVector4a bounds[2];
bounds[0] = mBounds[0];
bounds[1] = mBounds[1];
if(shift != NULL)
{
bounds[0].add(*shift);
}
// Don't cull hole/edge water, unless we have the GL_ARB_depth_clamp extension
if (earlyFail(camera, bounds))
{
LLFastTimer t(FTM_OCCLUSION_EARLY_FAIL);
setOcclusionState(LLOcclusionCullingGroup::DISCARD_QUERY);
assert_states_valid(this);
clearOcclusionState(LLOcclusionCullingGroup::OCCLUDED, LLOcclusionCullingGroup::STATE_MODE_DIFF);
assert_states_valid(this);
}
else
{
if (!isOcclusionState(QUERY_PENDING) || isOcclusionState(DISCARD_QUERY))
{
{ //no query pending, or previous query to be discarded
LLFastTimer t(FTM_RENDER_OCCLUSION);
if (!mOcclusionQuery[LLViewerCamera::sCurCameraID])
{
LLFastTimer t(FTM_OCCLUSION_ALLOCATE);
mOcclusionQuery[LLViewerCamera::sCurCameraID] = getNewOcclusionQueryObjectName();
}
// Depth clamp all water to avoid it being culled as a result of being
// behind the far clip plane, and in the case of edge water to avoid
// it being culled while still visible.
bool const use_depth_clamp = gGLManager.mHasDepthClamp &&
(mSpatialPartition->mDrawableType == LLDrawPool::POOL_WATER ||
mSpatialPartition->mDrawableType == LLDrawPool::POOL_VOIDWATER);
LLGLEnable clamp(use_depth_clamp ? GL_DEPTH_CLAMP : 0);
#if !LL_DARWIN
U32 mode = gGLManager.mHasOcclusionQuery2 ? GL_ANY_SAMPLES_PASSED : GL_SAMPLES_PASSED_ARB;
#else
U32 mode = GL_SAMPLES_PASSED_ARB;
#endif
#if LL_TRACK_PENDING_OCCLUSION_QUERIES
sPendingQueries.insert(mOcclusionQuery[LLViewerCamera::sCurCameraID]);
#endif
{
LLFastTimer t(FTM_PUSH_OCCLUSION_VERTS);
//store which frame this query was issued on
mOcclusionIssued[LLViewerCamera::sCurCameraID] = gFrameCount;
{
LLFastTimer t(FTM_OCCLUSION_BEGIN_QUERY);
glBeginQueryARB(mode, mOcclusionQuery[LLViewerCamera::sCurCameraID]);
}
LLGLSLShader* shader = LLGLSLShader::sCurBoundShaderPtr;
llassert(shader);
shader->uniform3fv(LLShaderMgr::BOX_CENTER, 1, bounds[0].getF32ptr());
//static LLVector4a fudge(SG_OCCLUSION_FUDGE);
static LLCachedControl<F32> vel("SHOcclusionFudge",SG_OCCLUSION_FUDGE);
LLVector4a fudge(SG_OCCLUSION_FUDGE);
static LLVector4a fudged_bounds;
fudged_bounds.setAdd(fudge, bounds[1]);
shader->uniform3fv(LLShaderMgr::BOX_SIZE, 1, fudged_bounds.getF32ptr());
if (!use_depth_clamp && mSpatialPartition->mDrawableType == LLDrawPool::POOL_VOIDWATER)
{
LLFastTimer t(FTM_OCCLUSION_DRAW_WATER);
LLGLSquashToFarClip squash(glh_get_current_projection(), 1);
if (camera->getOrigin().isExactlyZero())
{ //origin is invalid, draw entire box
gPipeline.mCubeVB->drawRange(LLRender::TRIANGLE_FAN, 0, 7, 8, 0);
gPipeline.mCubeVB->drawRange(LLRender::TRIANGLE_FAN, 0, 7, 8, b111*8);
}
else
{
gPipeline.mCubeVB->drawRange(LLRender::TRIANGLE_FAN, 0, 7, 8, get_box_fan_indices(camera, bounds[0]));
}
}
else
{
LLFastTimer t(FTM_OCCLUSION_DRAW);
if (camera->getOrigin().isExactlyZero())
{ //origin is invalid, draw entire box
gPipeline.mCubeVB->drawRange(LLRender::TRIANGLE_FAN, 0, 7, 8, 0);
gPipeline.mCubeVB->drawRange(LLRender::TRIANGLE_FAN, 0, 7, 8, b111*8);
}
else
{
gPipeline.mCubeVB->drawRange(LLRender::TRIANGLE_FAN, 0, 7, 8, get_box_fan_indices(camera, bounds[0]));
}
}
{
LLFastTimer t(FTM_OCCLUSION_END_QUERY);
glEndQueryARB(mode);
}
}
}
{
LLFastTimer t(FTM_SET_OCCLUSION_STATE);
setOcclusionState(LLOcclusionCullingGroup::QUERY_PENDING);
clearOcclusionState(LLOcclusionCullingGroup::DISCARD_QUERY);
}
}
}
}
}
static void jump()
{
assert(stage_context[1].decoded==NULL);
squash();
PC=stage_context[2].decoded->literal + stage_context[2].decoded->src1;
}
void ConcatPeephole::finalize() {
while (!m_working.empty()) {
squash();
}
m_next.finalize();
}
HIDDEN void
do_Char(int c, int xpos, int ypos, int odd)
{
register int i, j;
int base;
int totwid = font.width;
int down;
static float resbuf[FONTBUFSZ];
static RGBpixel fbline[FONTBUFSZ];
#if DEBUG_STRINGS
fb_log( "do_Char: c='%c' xpos=%d ypos=%d odd=%d\n",
c, xpos, ypos, odd );
#endif
/* read in character bit map, with two blank lines on each end */
for (i = 0; i < 2; i++)
clear_buf (totwid, filterbuf[i]);
for (i = font.height + 1; i >= 2; i--)
fill_buf (font.width, filterbuf[i]);
for (i = font.height + 2; i < font.height + 4; i++)
clear_buf (totwid, filterbuf[i]);
(void)SignedChar( font.dir[c].up );
down = SignedChar( font.dir[c].down );
/* Initial base line for filtering depends on odd flag. */
base = (odd ? 1 : 2);
/* Produce a RGBpixel buffer from a description of the character and
the read back data from the frame buffer for anti-aliasing.
*/
for (i = font.height + base; i >= base; i--)
{
squash( filterbuf[i - 1], /* filter info */
filterbuf[i],
filterbuf[i + 1],
resbuf,
totwid + 4
);
fb_read( fbp, xpos, ypos - down + i, (unsigned char *)fbline, totwid+3);
for (j = 0; j < (totwid + 3) - 1; j++)
{
register int tmp;
/* EDITOR'S NOTE : do not rearrange this code,
the SUN compiler can't handle more
complex expressions. */
tmp = fbline[j][RED] & 0377;
fbline[j][RED] =
(int)(paint[RED]*resbuf[j]+(1-resbuf[j])*tmp);
fbline[j][RED] &= 0377;
tmp = fbline[j][GRN] & 0377;
fbline[j][GRN] =
(int)(paint[GRN]*resbuf[j]+(1-resbuf[j])*tmp);
fbline[j][GRN] &= 0377;
tmp = fbline[j][BLU] & 0377;
fbline[j][BLU] =
(int)(paint[BLU]*resbuf[j]+(1-resbuf[j])*tmp);
fbline[j][BLU] &= 0377;
}
fb_write( fbp, xpos, ypos - down + i, (unsigned char *)fbline, totwid+3 );
}
return;
}
void SMPCache::doReceiveFromBelow(SMPMemRequest *sreq)
{
MemOperation memOp = sreq->getMemOperation();
//printf("Symbolic Name: %s\n",symbolicName);
// request from other caches, need to process them accordingly
PAddr addr= sreq->getPAddr();
Line *l = cache->findLine(addr);
//READ HIT/MISS------------
if(memOp == MemRead) {
#ifdef TLS
//check if cache has line marked with epoch
//Add the chache flags to the request
if(sreq->isDataReq() && l)
{
if (sreq->needsData()) //READ MISS was broadcast
{
MemRequest *oreq=sreq->getOriginalRequest();
//If Line epoch is < request epoch set Others exposed
long int tmpclk=((oreq->getEpoch())->getClock());
if ((l->getEpoch())->getClock()<tmpclk)
{
l->getCacheFlags()->setOE();
}
//READ MISS--- then Add line cache flags to request
if (sreq->needsData())
{
//HACK: some bug causes this value already set to disappear
(l->getCacheFlags())->setEpoch(l->getEpoch());
sreq->setCacheFlags(l->getCacheFlags());
}
}
else //READ HIT was broadcast
{
MemRequest *oreq=sreq->getOriginalRequest();
//If Line epoch is < request epoch set Others exposed
long int tmpclk=((oreq->getEpoch())->getClock());
if ((l->getEpoch())->getClock()<tmpclk)
{
l->getCacheFlags()->setOE();
}
}
}
#endif
protocol->readMissHandler(sreq);
return;
}
//WRITE HIT/MISS----------
if(memOp == MemReadW) {
//L1 Write Miss
if(sreq->needsData())
{
#ifdef TLS
//Write miss -check if cache has line marked with epoch
if (sreq->isDataReq() && l)
{
MemRequest *oreq=sreq->getOriginalRequest();
//SQUASH EPOCH!!!
if (((l->getEpoch())->getClock()>((oreq->getEpoch())->getClock())) && l->getCacheFlags()->getWordER(getOffsetInLine(addr)))
{
//realInvalidateTLS(addr,cache->getLineSize());
int sqhCount=squash(l->getEpoch());
printf("Backend Squash: Invalidated lines- %ld\n",sqhCount);
squashes.inc();
}
//WRITE MISS--- then Add line cache flags to request
else if (sreq->needsData())
{
//HACK: some bug causes this value already set to disappear
(l->getCacheFlags())->setEpoch(l->getEpoch());
sreq->setCacheFlags(l->getCacheFlags());
}
}
#endif
protocol->writeMissHandler(sreq);
return;
}
else{
//WRITE HIT
#ifdef TLS
//Write Hit -check if cache has line marked with epoch
if (sreq->isDataReq() && l)
{
MemRequest *oreq=sreq->getOriginalRequest();
//SQUASH EPOCH!!!
if (((l->getEpoch())->getClock()>((oreq->getEpoch())->getClock())) && l->getCacheFlags()->getWordER(getOffsetInLine(addr)))
{
//realInvalidateTLS(addr,cache->getLineSize());
int sqhCount=squash(l->getEpoch());
printf("Backend Squash: Invalidated lines- %ld\n",sqhCount);
squashes.inc();
}
//WRITE Hit--- then Add line cache flags to request
else if (sreq->needsData())
{
//HACK: some bug causes this value already set to disappear
(l->getCacheFlags())->setEpoch(l->getEpoch());
sreq->setCacheFlags(l->getCacheFlags());
}
}
protocol->writeMissHandler(sreq);
return;
#else
protocol->invalidateHandler(sreq);
#endif
}
return;
}
#ifdef TLS
//L1 HIT,L2 HIT
//INVALIDATION MESSAGE FROM ANOTHER CACHE AS IT WANTS IT IN EXCLUSIVE STATE
if(memOp == MemWrite) {
protocol->invalidateHandler(sreq);
return;
}
#else
if(memOp == MemWrite) {
I(!sreq->needsData());
protocol->invalidateHandler(sreq);
return;
}
#endif
#ifdef TLS
if (memOp==MemPush)
{
if (l)
{
MemRequest *oreq=sreq->getOriginalRequest();
//All epochs preceeding the one being written back MUST be commited
if ((l->getEpoch())->getClock()<((oreq->getEpoch())->getClock()))
{
//HACK: some bug causes this value already set to disappear
(l->getCacheFlags())->setEpoch(l->getEpoch());
sreq->setCacheFlags(l->getCacheFlags());
realInvalidateTLS(addr,cache->getLineSize());
}
}
sreq->goDown(0, lowerLevel[0]);
return;
}
#else
I(0); // this routine should not be called for any other memory request
#endif
}
void
do_char(struct vfont *vfp, struct vfont_dispatch *vdp, int x, int y)
{
int i, j;
int base;
int totwid = width;
int ln;
static float resbuf[FONTBUFSZ];
static RGBpixel fbline[FONTBUFSZ];
int bytes_wide; /* # bytes/row in bitmap */
bytes_wide = (width+7)>>3;
/* Read in the character bit map, with two blank lines on each end. */
for (i = 0; i < 2; i++)
memset((char *)&filterbuf[i][0], 0, (totwid+4)*sizeof(int));
for (ln=0, i = height + 1; i >= 2; i--, ln++)
fill_buf (width, &filterbuf[i][0],
&vfp->vf_bits[vdp->vd_addr + bytes_wide*ln]);
for (i = height + 2; i < height + 4; i++)
memset((char *)&filterbuf[i][0], 0, (totwid+4)*sizeof(int));
/* Initial base line for filtering depends on odd flag. */
if (vdp->vd_down % 2)
base = 1;
else
base = 2;
/* Produce a RGBpixel buffer from a description of the character
* and the read back data from the frame buffer for anti-aliasing.
*/
for (i = height + base; i >= base; i--) {
squash(filterbuf[i - 1], /* filter info */
filterbuf[i],
filterbuf[i + 1],
resbuf,
totwid + 4
);
fb_read(fbp, x, y - vdp->vd_down + i, (unsigned char *)fbline, totwid+3);
for (j = 0; j < (totwid + 3) - 1; j++) {
int tmp;
/* EDITOR'S NOTE : do not rearrange this code, the SUN
* compiler can't handle more complex expressions.
*/
tmp = fbline[j][RED] & 0377;
fbline[j][RED] =
(int)(pixcolor[RED]*resbuf[j]+(1-resbuf[j])*tmp);
fbline[j][RED] &= 0377;
tmp = fbline[j][GRN] & 0377;
fbline[j][GRN] =
(int)(pixcolor[GRN]*resbuf[j]+(1-resbuf[j])*tmp);
fbline[j][GRN] &= 0377;
tmp = fbline[j][BLU] & 0377;
fbline[j][BLU] =
(int)(pixcolor[BLU]*resbuf[j]+(1-resbuf[j])*tmp);
fbline[j][BLU] &= 0377;
}
if (fb_write(fbp, x, y-vdp->vd_down+i, (unsigned char *)fbline, totwid+3) < totwid+3) {
fprintf(stderr, "fblabel: pixel write error\n");
bu_exit(1, NULL);
}
}
}