blob: 7bbf4e054a905e24e709c7d487633253d75ccbf9 [file] [log] [blame]
/*
* Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
/*
* lattice.c
*
* Contains the normalized lattice filter routines (MA and AR) for iSAC codec
*
*/
#include "modules/audio_coding/codecs/isac/fix/source/codec.h"
#include "modules/audio_coding/codecs/isac/fix/source/settings.h"
#include "rtc_base/sanitizer.h"
#define LATTICE_MUL_32_32_RSFT16(a32a, a32b, b32) \
((int32_t)(WEBRTC_SPL_MUL(a32a, b32) + (WEBRTC_SPL_MUL_16_32_RSFT16(a32b, b32))))
/* This macro is FORBIDDEN to use elsewhere than in a function in this file and
its corresponding neon version. It might give unpredictable results, since a
general int32_t*int32_t multiplication results in a 64 bit value.
The result is then shifted just 16 steps to the right, giving need for 48
bits, i.e. in the generel case, it will NOT fit in a int32_t. In the
cases used in here, the int32_t will be enough, since (for a good
reason) the involved multiplicands aren't big enough to overflow a
int32_t after shifting right 16 bits. I have compared the result of a
multiplication between t32 and tmp32, done in two ways:
1) Using (int32_t) (((float)(tmp32))*((float)(tmp32b))/65536.0);
2) Using LATTICE_MUL_32_32_RSFT16(t16a, t16b, tmp32b);
By running 25 files, I haven't found any bigger diff than 64 - this was in the
case when method 1) gave 650235648 and 2) gave 650235712.
*/
/* Function prototype: filtering ar_g_Q0[] and ar_f_Q0[] through an AR filter
with coefficients cth_Q15[] and sth_Q15[].
Implemented for both generic and ARMv7 platforms.
*/
void WebRtcIsacfix_FilterArLoop(int16_t* ar_g_Q0,
int16_t* ar_f_Q0,
int16_t* cth_Q15,
int16_t* sth_Q15,
size_t order_coef);
/* Inner loop used for function WebRtcIsacfix_NormLatticeFilterMa(). It does:
for 0 <= n < HALF_SUBFRAMELEN - 1:
*ptr2 = input2 * (*ptr2) + input0 * (*ptr0));
*ptr1 = input1 * (*ptr0) + input0 * (*ptr2);
Note, function WebRtcIsacfix_FilterMaLoopNeon and WebRtcIsacfix_FilterMaLoopC
are not bit-exact. The accuracy by the ARM Neon function is same or better.
*/
void WebRtcIsacfix_FilterMaLoopC(int16_t input0, // Filter coefficient
int16_t input1, // Filter coefficient
int32_t input2, // Inverse coeff. (1/input1)
int32_t* ptr0, // Sample buffer
int32_t* ptr1, // Sample buffer
int32_t* ptr2) { // Sample buffer
int n = 0;
// Separate the 32-bit variable input2 into two 16-bit integers (high 16 and
// low 16 bits), for using LATTICE_MUL_32_32_RSFT16 in the loop.
int16_t t16a = (int16_t)(input2 >> 16);
int16_t t16b = (int16_t)input2;
if (t16b < 0) t16a++;
// The loop filtering the samples *ptr0, *ptr1, *ptr2 with filter coefficients
// input0, input1, and input2.
for(n = 0; n < HALF_SUBFRAMELEN - 1; n++, ptr0++, ptr1++, ptr2++) {
int32_t tmp32a = 0;
int32_t tmp32b = 0;
// Calculate *ptr2 = input2 * (*ptr2 + input0 * (*ptr0));
tmp32a = WEBRTC_SPL_MUL_16_32_RSFT15(input0, *ptr0); // Q15 * Q15 >> 15 = Q15
tmp32b = *ptr2 + tmp32a; // Q15 + Q15 = Q15
*ptr2 = LATTICE_MUL_32_32_RSFT16(t16a, t16b, tmp32b);
// Calculate *ptr1 = input1 * (*ptr0) + input0 * (*ptr2);
tmp32a = WEBRTC_SPL_MUL_16_32_RSFT15(input1, *ptr0); // Q15*Q15>>15 = Q15
tmp32b = WEBRTC_SPL_MUL_16_32_RSFT15(input0, *ptr2); // Q15*Q15>>15 = Q15
*ptr1 = tmp32a + tmp32b; // Q15 + Q15 = Q15
}
}
/* filter the signal using normalized lattice filter */
/* MA filter */
void WebRtcIsacfix_NormLatticeFilterMa(size_t orderCoef,
int32_t *stateGQ15,
int16_t *lat_inQ0,
int16_t *filt_coefQ15,
int32_t *gain_lo_hiQ17,
int16_t lo_hi,
int16_t *lat_outQ9)
{
int16_t sthQ15[MAX_AR_MODEL_ORDER];
int16_t cthQ15[MAX_AR_MODEL_ORDER];
int u, n;
size_t i, k;
int16_t temp2,temp3;
size_t ord_1 = orderCoef+1;
int32_t inv_cthQ16[MAX_AR_MODEL_ORDER];
int32_t gain32, fQtmp;
int16_t gain16;
int16_t gain_sh;
int32_t tmp32, tmp32b;
int32_t fQ15vec[HALF_SUBFRAMELEN];
int32_t gQ15[MAX_AR_MODEL_ORDER+1][HALF_SUBFRAMELEN];
int16_t sh;
int16_t t16a;
int16_t t16b;
for (u=0;u<SUBFRAMES;u++)
{
int32_t temp1 = u * HALF_SUBFRAMELEN;
/* set the Direct Form coefficients */
temp2 = (int16_t)(u * orderCoef);
temp3 = (int16_t)(2 * u + lo_hi);
/* compute lattice filter coefficients */
memcpy(sthQ15, &filt_coefQ15[temp2], orderCoef * sizeof(int16_t));
WebRtcSpl_SqrtOfOneMinusXSquared(sthQ15, orderCoef, cthQ15);
/* compute the gain */
gain32 = gain_lo_hiQ17[temp3];
gain_sh = WebRtcSpl_NormW32(gain32);
gain32 <<= gain_sh; // Q(17+gain_sh)
for (k=0;k<orderCoef;k++)
{
gain32 = WEBRTC_SPL_MUL_16_32_RSFT15(cthQ15[k], gain32); //Q15*Q(17+gain_sh)>>15 = Q(17+gain_sh)
inv_cthQ16[k] = WebRtcSpl_DivW32W16((int32_t)2147483647, cthQ15[k]); // 1/cth[k] in Q31/Q15 = Q16
}
gain16 = (int16_t)(gain32 >> 16); // Q(1+gain_sh).
/* normalized lattice filter */
/*****************************/
/* initial conditions */
for (i=0;i<HALF_SUBFRAMELEN;i++)
{
fQ15vec[i] = lat_inQ0[i + temp1] << 15; // Q15
gQ15[0][i] = lat_inQ0[i + temp1] << 15; // Q15
}
fQtmp = fQ15vec[0];
/* get the state of f&g for the first input, for all orders */
for (i=1;i<ord_1;i++)
{
// Calculate f[i][0] = inv_cth[i-1]*(f[i-1][0] + sth[i-1]*stateG[i-1]);
tmp32 = WEBRTC_SPL_MUL_16_32_RSFT15(sthQ15[i-1], stateGQ15[i-1]);//Q15*Q15>>15 = Q15
tmp32b= fQtmp + tmp32; //Q15+Q15=Q15
tmp32 = inv_cthQ16[i-1]; //Q16
t16a = (int16_t)(tmp32 >> 16);
t16b = (int16_t)(tmp32 - (t16a << 16));
if (t16b<0) t16a++;
tmp32 = LATTICE_MUL_32_32_RSFT16(t16a, t16b, tmp32b);
fQtmp = tmp32; // Q15
// Calculate g[i][0] = cth[i-1]*stateG[i-1] + sth[i-1]* f[i][0];
tmp32 = WEBRTC_SPL_MUL_16_32_RSFT15(cthQ15[i-1], stateGQ15[i-1]); //Q15*Q15>>15 = Q15
tmp32b = WEBRTC_SPL_MUL_16_32_RSFT15(sthQ15[i-1], fQtmp); //Q15*Q15>>15 = Q15
tmp32 = tmp32 + tmp32b;//Q15+Q15 = Q15
gQ15[i][0] = tmp32; // Q15
}
/* filtering */
/* save the states */
for(k=0;k<orderCoef;k++)
{
// for 0 <= n < HALF_SUBFRAMELEN - 1:
// f[k+1][n+1] = inv_cth[k]*(f[k][n+1] + sth[k]*g[k][n]);
// g[k+1][n+1] = cth[k]*g[k][n] + sth[k]* f[k+1][n+1];
WebRtcIsacfix_FilterMaLoopFix(sthQ15[k], cthQ15[k], inv_cthQ16[k],
&gQ15[k][0], &gQ15[k+1][1], &fQ15vec[1]);
}
fQ15vec[0] = fQtmp;
for(n=0;n<HALF_SUBFRAMELEN;n++)
{
//gain32 >>= gain_sh; // Q(17+gain_sh) -> Q17
tmp32 = WEBRTC_SPL_MUL_16_32_RSFT16(gain16, fQ15vec[n]); //Q(1+gain_sh)*Q15>>16 = Q(gain_sh)
sh = 9-gain_sh; //number of needed shifts to reach Q9
t16a = (int16_t) WEBRTC_SPL_SHIFT_W32(tmp32, sh);
lat_outQ9[n + temp1] = t16a;
}
/* save the states */
for (i=0;i<ord_1;i++)
{
stateGQ15[i] = gQ15[i][HALF_SUBFRAMELEN-1];
}
//process next frame
}
return;
}
// Left shift of an int32_t that's allowed to overflow. (It's still undefined
// behavior, so not a good idea; this just makes UBSan ignore the violation, so
// that our old code can continue to do what it's always been doing.)
static inline int32_t RTC_NO_SANITIZE("shift")
OverflowingLShiftS32(int32_t x, int shift) {
return x << shift;
}
/* ----------------AR filter-------------------------*/
/* filter the signal using normalized lattice filter */
void WebRtcIsacfix_NormLatticeFilterAr(size_t orderCoef,
int16_t *stateGQ0,
int32_t *lat_inQ25,
int16_t *filt_coefQ15,
int32_t *gain_lo_hiQ17,
int16_t lo_hi,
int16_t *lat_outQ0)
{
size_t ii, k, i;
int n, u;
int16_t sthQ15[MAX_AR_MODEL_ORDER];
int16_t cthQ15[MAX_AR_MODEL_ORDER];
int32_t tmp32;
int16_t tmpAR;
int16_t ARfQ0vec[HALF_SUBFRAMELEN];
int16_t ARgQ0vec[MAX_AR_MODEL_ORDER+1];
int32_t inv_gain32;
int16_t inv_gain16;
int16_t den16;
int16_t sh;
int16_t temp2,temp3;
size_t ord_1 = orderCoef+1;
for (u=0;u<SUBFRAMES;u++)
{
int32_t temp1 = u * HALF_SUBFRAMELEN;
//set the denominator and numerator of the Direct Form
temp2 = (int16_t)(u * orderCoef);
temp3 = (int16_t)(2 * u + lo_hi);
for (ii=0; ii<orderCoef; ii++) {
sthQ15[ii] = filt_coefQ15[temp2+ii];
}
WebRtcSpl_SqrtOfOneMinusXSquared(sthQ15, orderCoef, cthQ15);
// Originally, this line was assumed to never overflow, since "[s]imulation
// of the 25 files shows that maximum value in the vector gain_lo_hiQ17[]
// is 441344, which means that it is log2((2^31)/441344) = 12.2 shifting
// bits from saturation. Therefore, it should be safe to use Q27 instead of
// Q17." However, a fuzzer test succeeded in provoking an overflow here,
// which we ignore on the theory that only "abnormal" inputs cause
// overflow.
tmp32 = OverflowingLShiftS32(gain_lo_hiQ17[temp3], 10); // Q27
for (k=0;k<orderCoef;k++) {
tmp32 = WEBRTC_SPL_MUL_16_32_RSFT15(cthQ15[k], tmp32); // Q15*Q27>>15 = Q27
}
sh = WebRtcSpl_NormW32(tmp32); // tmp32 is the gain
den16 = (int16_t) WEBRTC_SPL_SHIFT_W32(tmp32, sh-16); //Q(27+sh-16) = Q(sh+11) (all 16 bits are value bits)
inv_gain32 = WebRtcSpl_DivW32W16((int32_t)2147483647, den16); // 1/gain in Q31/Q(sh+11) = Q(20-sh)
//initial conditions
inv_gain16 = (int16_t)(inv_gain32 >> 2); // 1/gain in Q(20-sh-2) = Q(18-sh)
for (i=0;i<HALF_SUBFRAMELEN;i++)
{
tmp32 = OverflowingLShiftS32(lat_inQ25[i + temp1], 1); // Q25->Q26
tmp32 = WEBRTC_SPL_MUL_16_32_RSFT16(inv_gain16, tmp32); //lat_in[]*inv_gain in (Q(18-sh)*Q26)>>16 = Q(28-sh)
tmp32 = WEBRTC_SPL_SHIFT_W32(tmp32, -(28-sh)); // lat_in[]*inv_gain in Q0
ARfQ0vec[i] = (int16_t)WebRtcSpl_SatW32ToW16(tmp32); // Q0
}
// Get the state of f & g for the first input, for all orders.
for (i = orderCoef; i > 0; i--)
{
tmp32 = (cthQ15[i - 1] * ARfQ0vec[0] - sthQ15[i - 1] * stateGQ0[i - 1] +
16384) >> 15;
tmpAR = (int16_t)WebRtcSpl_SatW32ToW16(tmp32); // Q0
tmp32 = (sthQ15[i - 1] * ARfQ0vec[0] + cthQ15[i - 1] * stateGQ0[i - 1] +
16384) >> 15;
ARgQ0vec[i] = (int16_t)WebRtcSpl_SatW32ToW16(tmp32); // Q0
ARfQ0vec[0] = tmpAR;
}
ARgQ0vec[0] = ARfQ0vec[0];
// Filter ARgQ0vec[] and ARfQ0vec[] through coefficients cthQ15[] and sthQ15[].
WebRtcIsacfix_FilterArLoop(ARgQ0vec, ARfQ0vec, cthQ15, sthQ15, orderCoef);
for(n=0;n<HALF_SUBFRAMELEN;n++)
{
lat_outQ0[n + temp1] = ARfQ0vec[n];
}
/* cannot use memcpy in the following */
for (i=0;i<ord_1;i++)
{
stateGQ0[i] = ARgQ0vec[i];
}
}
return;
}