static int calc_one_scale(int32_t peak_cb, int abits, softfloat *quant)
{
int32_t peak;
int our_nscale, try_remove;
softfloat our_quant;
av_assert0(peak_cb <= 0);
av_assert0(peak_cb >= -2047);
our_nscale = 127;
peak = cb_to_level[-peak_cb];
for (try_remove = 64; try_remove > 0; try_remove >>= 1) {
if (scalefactor_inv[our_nscale - try_remove].e + stepsize_inv[abits].e <= 17)
continue;
our_quant.m = mul32(scalefactor_inv[our_nscale - try_remove].m, stepsize_inv[abits].m);
our_quant.e = scalefactor_inv[our_nscale - try_remove].e + stepsize_inv[abits].e - 17;
if ((quant_levels[abits] - 1) / 2 < quantize_value(peak, our_quant))
continue;
our_nscale -= try_remove;
}
if (our_nscale >= 125)
our_nscale = 124;
quant->m = mul32(scalefactor_inv[our_nscale].m, stepsize_inv[abits].m);
quant->e = scalefactor_inv[our_nscale].e + stepsize_inv[abits].e - 17;
av_assert0((quant_levels[abits] - 1) / 2 >= quantize_value(peak, *quant));
return our_nscale;
}
static void lfe_downsample(DCAContext *c, const int32_t *input)
{
/* FIXME: make 128x LFE downsampling possible */
int i, j, lfes;
int32_t hist[512];
int32_t accum;
int hist_start = 0;
for (i = 0; i < 512; i++)
hist[i] = c->history[i][c->channels - 1];
for (lfes = 0; lfes < DCA_LFE_SAMPLES; lfes++) {
/* Calculate the convolution */
accum = 0;
for (i = hist_start, j = 0; i < 512; i++, j++)
accum += mul32(hist[i], lfe_fir_64i[j]);
for (i = 0; i < hist_start; i++, j++)
accum += mul32(hist[i], lfe_fir_64i[j]);
c->downsampled_lfe[lfes] = accum;
/* Copy in 64 new samples from input */
for (i = 0; i < 64; i++)
hist[i + hist_start] = input[(lfes * 64 + i) * c->channels + c->channels - 1];
hist_start = (hist_start + 64) & 511;
}
}
static void qmf_decompose(DCAContext *c, int32_t in[32], int32_t out[32],
int channel)
{
int band, i, j, k;
int32_t resp;
int32_t accum[DCA_SUBBANDS_32] = {0};
add_new_samples(c, in, DCA_SUBBANDS_32, channel);
/* Calculate the dot product of the signal with the (possibly inverted)
reference decoder's response to this vector:
(0.0, 0.0, ..., 0.0, -1.0, 1.0, 0.0, ..., 0.0)
so that -1.0 cancels 1.0 from the previous step */
for (k = 48, j = 0, i = c->start[channel]; i < 512; k++, j++, i++)
accum[(k & 32) ? (31 - (k & 31)) : (k & 31)] += mul32(c->history[channel][i], UnQMF[j]);
for (i = 0; i < c->start[channel]; k++, j++, i++)
accum[(k & 32) ? (31 - (k & 31)) : (k & 31)] += mul32(c->history[channel][i], UnQMF[j]);
resp = 0;
/* TODO: implement FFT instead of this naive calculation */
for (band = 0; band < DCA_SUBBANDS_32; band++) {
for (j = 0; j < 32; j++)
resp += mul32(accum[j], band_delta_factor(band, j));
out[band] = (band & 2) ? (-resp) : resp;
}
}
h_generic_calc_Mul32x4 ( /*OUT*/V128* res,
V128* argL, V128* argR )
{
res->w32[0] = mul32(argL->w32[0], argR->w32[0]);
res->w32[1] = mul32(argL->w32[1], argR->w32[1]);
res->w32[2] = mul32(argL->w32[2], argR->w32[2]);
res->w32[3] = mul32(argL->w32[3], argR->w32[3]);
}
static int lfe_downsample(DCAContext *c, int32_t in[LFE_INTERPOLATION])
{
int i, j;
int channel = c->prim_channels;
int32_t accum = 0;
add_new_samples(c, in, LFE_INTERPOLATION, channel);
for (i = c->start[channel], j = 0; i < 512; i++, j++)
accum += mul32(c->history[channel][i], lfe_fir_64i[j]);
for (i = 0; i < c->start[channel]; i++, j++)
accum += mul32(c->history[channel][i], lfe_fir_64i[j]);
return accum;
}
static void subband_transform(DCAEncContext *c, const int32_t *input)
{
int ch, subs, i, k, j;
for (ch = 0; ch < c->fullband_channels; ch++) {
/* History is copied because it is also needed for PSY */
int32_t hist[512];
int hist_start = 0;
const int chi = c->channel_order_tab[ch];
for (i = 0; i < 512; i++)
hist[i] = c->history[i][ch];
for (subs = 0; subs < SUBBAND_SAMPLES; subs++) {
int32_t accum[64];
int32_t resp;
int band;
/* Calculate the convolutions at once */
for (i = 0; i < 64; i++)
accum[i] = 0;
for (k = 0, i = hist_start, j = 0;
i < 512; k = (k + 1) & 63, i++, j++)
accum[k] += mul32(hist[i], c->band_interpolation[j]);
for (i = 0; i < hist_start; k = (k + 1) & 63, i++, j++)
accum[k] += mul32(hist[i], c->band_interpolation[j]);
for (k = 16; k < 32; k++)
accum[k] = accum[k] - accum[31 - k];
for (k = 32; k < 48; k++)
accum[k] = accum[k] + accum[95 - k];
for (band = 0; band < 32; band++) {
resp = 0;
for (i = 16; i < 48; i++) {
int s = (2 * band + 1) * (2 * (i + 16) + 1);
resp += mul32(accum[i], cos_t(s << 3)) >> 3;
}
c->subband[subs][band][ch] = ((band + 1) & 2) ? -resp : resp;
}
/* Copy in 32 new samples from input */
for (i = 0; i < 32; i++)
hist[i + hist_start] = input[(subs * 32 + i) * c->channels + chi];
hist_start = (hist_start + 32) & 511;
}
}
}
void JIT::compileOpCallVarargs(Instruction* instruction)
{
int callee = instruction[1].u.operand;
int argCountRegister = instruction[2].u.operand;
int registerOffset = instruction[3].u.operand;
emitGetVirtualRegister(argCountRegister, regT1);
emitFastArithImmToInt(regT1);
emitGetVirtualRegister(callee, regT0);
addPtr(Imm32(registerOffset), regT1, regT2);
// Check for JSFunctions.
emitJumpSlowCaseIfNotJSCell(regT0);
addSlowCase(branchPtr(NotEqual, Address(regT0), TrustedImmPtr(m_globalData->jsFunctionVPtr)));
// Speculatively roll the callframe, assuming argCount will match the arity.
mul32(TrustedImm32(sizeof(Register)), regT2, regT2);
intptr_t offset = (intptr_t)sizeof(Register) * (intptr_t)RegisterFile::CallerFrame;
addPtr(Imm32((int32_t)offset), regT2, regT3);
addPtr(callFrameRegister, regT3);
storePtr(callFrameRegister, regT3);
addPtr(regT2, callFrameRegister);
emitNakedCall(m_globalData->jitStubs->ctiVirtualCall());
sampleCodeBlock(m_codeBlock);
}
void JIT::compileOpCallVarargs(Instruction* instruction)
{
int callee = instruction[1].u.operand;
int argCountRegister = instruction[2].u.operand;
int registerOffset = instruction[3].u.operand;
emitLoad(callee, regT1, regT0);
emitLoadPayload(argCountRegister, regT2); // argCount
addPtr(Imm32(registerOffset), regT2, regT3); // registerOffset
emitJumpSlowCaseIfNotJSCell(callee, regT1);
addSlowCase(branchPtr(NotEqual, Address(regT0), TrustedImmPtr(m_globalData->jsFunctionVPtr)));
// Speculatively roll the callframe, assuming argCount will match the arity.
mul32(TrustedImm32(sizeof(Register)), regT3, regT3);
addPtr(callFrameRegister, regT3);
store32(TrustedImm32(JSValue::CellTag), tagFor(RegisterFile::CallerFrame, regT3));
storePtr(callFrameRegister, payloadFor(RegisterFile::CallerFrame, regT3));
move(regT3, callFrameRegister);
move(regT2, regT1); // argCount
emitNakedCall(m_globalData->jitStubs->ctiVirtualCall());
sampleCodeBlock(m_codeBlock);
}
static int32_t quantize_value(int32_t value, softfloat quant)
{
int32_t offset = 1 << (quant.e - 1);
value = mul32(value, quant.m) + offset;
value = value >> quant.e;
return value;
}
static void qmf_init(void)
{
int i;
int32_t c[17], s[17];
s[0] = 0; /* sin(index * PI / 64) * 0x7fffffff */
c[0] = 0x7fffffff; /* cos(index * PI / 64) * 0x7fffffff */
for (i = 1; i <= 16; i++) {
s[i] = 2 * (mul32(c[i - 1], 105372028) + mul32(s[i - 1], 2144896908));
c[i] = 2 * (mul32(c[i - 1], 2144896908) - mul32(s[i - 1], 105372028));
}
for (i = 0; i < 16; i++) {
cos_table[i ] = c[i] >> 3; /* avoid output overflow */
cos_table[i + 16] = s[16 - i] >> 3;
cos_table[i + 32] = -s[i] >> 3;
cos_table[i + 48] = -c[16 - i] >> 3;
cos_table[i + 64] = -c[i] >> 3;
cos_table[i + 80] = -s[16 - i] >> 3;
cos_table[i + 96] = s[i] >> 3;
cos_table[i + 112] = c[16 - i] >> 3;
}
}
void main() {
struct long32 first, second, working;
char s[20];
while (TRUE) {
printf ("\r\n\r\nEnter the first number: ");
get_string (s,20);
atol32 (s,&first);
printf ("\r\nEnter the second number: ");
get_string (s,20);
atol32 (s,&second);
printf ("\r\n\r\nA: ");
print_long32 (&first);
printf ("\r\nB: ");
print_long32 (&second);
working.hi = first.hi;
working.lo = first.lo;
add32 (&working, &second);
printf ("\r\na + b = ");
print_long32 (&working);
working.hi = first.hi;
working.lo = first.lo;
sub32 (&working, &second);
printf ("\r\na - b = ");
print_long32 (&working);
working.hi = first.hi;
working.lo = first.lo;
mul32 (&working, &second);
printf ("\r\na * b = ");
print_long32 (&working);
working.hi = first.hi;
working.lo = first.lo;
div32 (&working, &second);
printf ("\r\na / b = ");
print_long32 (&working);
rem32 (&first, &second, &working);
printf ("\r\na modulus b = ");
print_long32 (&working);
}
}
static int init_quantization_noise(DCAContext *c, int noise)
{
int ch, band, ret = 0;
c->consumed_bits = 132 + 493 * c->fullband_channels;
if (c->lfe_channel)
c->consumed_bits += 72;
/* attempt to guess the bit distribution based on the prevoius frame */
for (ch = 0; ch < c->fullband_channels; ch++) {
for (band = 0; band < 32; band++) {
int snr_cb = c->peak_cb[band][ch] - c->band_masking_cb[band] - noise;
if (snr_cb >= 1312) {
c->abits[band][ch] = 26;
ret |= USED_26ABITS;
} else if (snr_cb >= 222) {
c->abits[band][ch] = 8 + mul32(snr_cb - 222, 69000000);
ret |= USED_NABITS;
} else if (snr_cb >= 0) {
c->abits[band][ch] = 2 + mul32(snr_cb, 106000000);
ret |= USED_NABITS;
} else {
c->abits[band][ch] = 1;
ret |= USED_1ABITS;
}
}
}
for (band = 0; band < 32; band++)
for (ch = 0; ch < c->fullband_channels; ch++) {
c->consumed_bits += bit_consumption[c->abits[band][ch]];
}
return ret;
}
void print_long32 (struct long32 *input) {
byte i;
struct long32 divisor, digit, temp, value;
divisor.hi = 0x3B9A;
divisor.lo = 0xCA00;
value.hi = input->hi;
value.lo = input->lo;
for(i=0;i<10;++i) {
digit = value;
div32 (&digit,&divisor);
temp = digit;
mul32 (&temp,&divisor);
sub32 (&value, &temp);
putc(digit.lo+'0');
temp.hi = 0;
temp.lo = 0x000A;
div32 (&divisor, &temp);
}
}
static VOID UpdateTimeStamp( USHORT new8253 )
{
USHORT delta;
ULONG nanos;
USHORT mills;
if( Last8253 >= new8253 ) {
delta = Last8253-new8253;
} else { // wrapped
delta = 0xFFFF - new8253 + Last8253;
}
nanos = mul32( delta, NANOS_IN_TIC );
Last8253 = new8253;
nanos += ReadDataBuf.nanosecs;
if( nanos >= AMILL ) { // overflow to millsecs
mills = 1;
nanos -= AMILL; // the most we need to do this is 5 times anyways
while( nanos >= AMILL ) { // overflow into millisecs
++mills; // try and avoid a runtime divide
nanos -= AMILL;
}
ReadDataBuf.millisecs += mills;
}
ReadDataBuf.nanosecs = nanos;
}
static void fft(const int32_t in[2 * 256], cplx32 out[256])
{
cplx32 buf[256], rin[256], rout[256];
int i, j, k, l;
/* do two transforms in parallel */
for (i = 0; i < 256; i++) {
/* Apply the Hann window */
rin[i].re = mul32(in[2 * i], 0x3fffffff - (cos_t(8 * i + 2) >> 1));
rin[i].im = mul32(in[2 * i + 1], 0x3fffffff - (cos_t(8 * i + 6) >> 1));
}
/* pre-rotation */
for (i = 0; i < 256; i++) {
buf[i].re = mul32(cos_t(4 * i + 2), rin[i].re)
- mul32(sin_t(4 * i + 2), rin[i].im);
buf[i].im = mul32(cos_t(4 * i + 2), rin[i].im)
+ mul32(sin_t(4 * i + 2), rin[i].re);
}
for (j = 256, l = 1; j != 1; j >>= 1, l <<= 1) {
for (k = 0; k < 256; k += j) {
for (i = k; i < k + j / 2; i++) {
cplx32 sum, diff;
int t = 8 * l * i;
sum.re = buf[i].re + buf[i + j / 2].re;
sum.im = buf[i].im + buf[i + j / 2].im;
diff.re = buf[i].re - buf[i + j / 2].re;
diff.im = buf[i].im - buf[i + j / 2].im;
buf[i].re = half32(sum.re);
buf[i].im = half32(sum.im);
buf[i + j / 2].re = mul32(diff.re, cos_t(t))
- mul32(diff.im, sin_t(t));
buf[i + j / 2].im = mul32(diff.im, cos_t(t))
+ mul32(diff.re, sin_t(t));
}
}
}
/* post-rotation */
for (i = 0; i < 256; i++) {
int b = ff_reverse[i];
rout[i].re = mul32(buf[b].re, cos_t(4 * i))
- mul32(buf[b].im, sin_t(4 * i));
rout[i].im = mul32(buf[b].im, cos_t(4 * i))
+ mul32(buf[b].re, sin_t(4 * i));
}
for (i = 0; i < 256; i++) {
/* separate the results of the two transforms */
cplx32 o1, o2;
o1.re = rout[i].re - rout[255 - i].re;
o1.im = rout[i].im + rout[255 - i].im;
o2.re = rout[i].im - rout[255 - i].im;
o2.im = -rout[i].re - rout[255 - i].re;
/* combine them into one long transform */
out[i].re = mul32( o1.re + o2.re, cos_t(2 * i + 1))
+ mul32( o1.im - o2.im, sin_t(2 * i + 1));
out[i].im = mul32( o1.im + o2.im, cos_t(2 * i + 1))
+ mul32(-o1.re + o2.re, sin_t(2 * i + 1));
}
}
static int init_quantization_noise(DCAEncContext *c, int noise)
{
int ch, band, ret = 0;
uint32_t huff_bit_count_accum[MAX_CHANNELS][DCA_CODE_BOOKS][7];
uint32_t clc_bit_count_accum[MAX_CHANNELS][DCA_CODE_BOOKS];
uint32_t bits_counter = 0;
c->consumed_bits = 132 + 333 * c->fullband_channels;
if (c->lfe_channel)
c->consumed_bits += 72;
/* attempt to guess the bit distribution based on the prevoius frame */
for (ch = 0; ch < c->fullband_channels; ch++) {
for (band = 0; band < 32; band++) {
int snr_cb = c->peak_cb[ch][band] - c->band_masking_cb[band] - noise;
if (snr_cb >= 1312) {
c->abits[ch][band] = 26;
ret |= USED_26ABITS;
} else if (snr_cb >= 222) {
c->abits[ch][band] = 8 + mul32(snr_cb - 222, 69000000);
ret |= USED_NABITS;
} else if (snr_cb >= 0) {
c->abits[ch][band] = 2 + mul32(snr_cb, 106000000);
ret |= USED_NABITS;
} else {
c->abits[ch][band] = 1;
ret |= USED_1ABITS;
}
}
c->consumed_bits += set_best_abits_code(c->abits[ch], 32, &c->bit_allocation_sel[ch]);
}
/* Recalc scale_factor each time to get bits consumption in case of Huffman coding.
It is suboptimal solution */
/* TODO: May be cache scaled values */
for (ch = 0; ch < c->fullband_channels; ch++) {
for (band = 0; band < 32; band++) {
c->scale_factor[ch][band] = calc_one_scale(c->peak_cb[ch][band],
c->abits[ch][band],
&c->quant[ch][band]);
}
}
quantize_all(c);
memset(huff_bit_count_accum, 0, MAX_CHANNELS * DCA_CODE_BOOKS * 7 * sizeof(uint32_t));
memset(clc_bit_count_accum, 0, MAX_CHANNELS * DCA_CODE_BOOKS * sizeof(uint32_t));
for (ch = 0; ch < c->fullband_channels; ch++) {
for (band = 0; band < 32; band++) {
if (c->abits[ch][band] && c->abits[ch][band] <= DCA_CODE_BOOKS) {
accumulate_huff_bit_consumption(c->abits[ch][band], c->quantized[ch][band], huff_bit_count_accum[ch][c->abits[ch][band] - 1]);
clc_bit_count_accum[ch][c->abits[ch][band] - 1] += bit_consumption[c->abits[ch][band]];
} else {
bits_counter += bit_consumption[c->abits[ch][band]];
}
}
}
for (ch = 0; ch < c->fullband_channels; ch++) {
bits_counter += set_best_code(huff_bit_count_accum[ch], clc_bit_count_accum[ch], c->quant_index_sel[ch]);
}
c->consumed_bits += bits_counter;
return ret;
}
