MIPS optimizations for ISAC (patch #2)
Implemented functions: - WebRtcIsacfix_CalculateResidualEnergy - WebRtcIsacfix_Spec2Time - WebRtcIsacfix_Time2Spec - WebRtcIsacfix_HighpassFilterFixDec32 - WebRtcIsacfix_PCorr2Q32 Gain achieved: aprox. further 5% on top of patch#1 on ISAC encoding path. The optimizations are bit-exact to the C code, with the excception of the MIPS DSPr2 variant of the WebRtcIsacfix_Time2Spec function (the accuracy of the WebRtcIsacfix_Time2Spec MIPS DSPr2 variant is same or better than C variant). Code verification and improvement achieved have been determined using the iSACFixtest application. R=andrew@webrtc.org, tina.legrand@webrtc.org Review URL: https://webrtc-codereview.appspot.com/19749004 Patch from Ljubomir Papuga <lpapuga@mips.com>. git-svn-id: http://webrtc.googlecode.com/svn/trunk@6749 4adac7df-926f-26a2-2b94-8c16560cd09d
This commit is contained in:
andrew@webrtc.org
parent
908f57ed94
commit
ceafa8cce9
@ -101,6 +101,16 @@ void WebRtcIsacfix_Spec2TimeNeon(int16_t* inreQ7,
|
||||
int32_t* outre2Q16);
|
||||
#endif
|
||||
|
||||
#if defined(MIPS32_LE)
|
||||
void WebRtcIsacfix_Time2SpecMIPS(int16_t* inre1Q9,
|
||||
int16_t* inre2Q9,
|
||||
int16_t* outre,
|
||||
int16_t* outim);
|
||||
void WebRtcIsacfix_Spec2TimeMIPS(int16_t* inreQ7,
|
||||
int16_t* inimQ7,
|
||||
int32_t* outre1Q16,
|
||||
int32_t* outre2Q16);
|
||||
#endif
|
||||
|
||||
/* filterbank functions */
|
||||
|
||||
|
||||
@ -23,10 +23,23 @@ extern "C" {
|
||||
* coefficient: Input.
|
||||
* state: Input/output, filter state, in Q4.
|
||||
*/
|
||||
void WebRtcIsacfix_HighpassFilterFixDec32(int16_t *io,
|
||||
int16_t len,
|
||||
const int16_t *coefficient,
|
||||
int32_t *state);
|
||||
typedef void (*HighpassFilterFixDec32)(int16_t* io,
|
||||
int16_t len,
|
||||
const int16_t* coefficient,
|
||||
int32_t* state);
|
||||
extern HighpassFilterFixDec32 WebRtcIsacfix_HighpassFilterFixDec32;
|
||||
|
||||
void WebRtcIsacfix_HighpassFilterFixDec32C(int16_t* io,
|
||||
int16_t len,
|
||||
const int16_t* coefficient,
|
||||
int32_t* state);
|
||||
|
||||
#if defined(MIPS_DSP_R1_LE)
|
||||
void WebRtcIsacfix_HighpassFilterFixDec32MIPS(int16_t* io,
|
||||
int16_t len,
|
||||
const int16_t* coefficient,
|
||||
int32_t* state);
|
||||
#endif
|
||||
|
||||
typedef void (*AllpassFilter2FixDec16)(
|
||||
int16_t *data_ch1, // Input and output in channel 1, in Q0
|
||||
|
||||
@ -86,10 +86,13 @@ void WebRtcIsacfix_AllpassFilter2FixDec16C(
|
||||
filter_state_ch2[1] = state1_ch2;
|
||||
}
|
||||
|
||||
void WebRtcIsacfix_HighpassFilterFixDec32(int16_t *io,
|
||||
int16_t len,
|
||||
const int16_t *coefficient,
|
||||
int32_t *state)
|
||||
// Declare a function pointer.
|
||||
HighpassFilterFixDec32 WebRtcIsacfix_HighpassFilterFixDec32;
|
||||
|
||||
void WebRtcIsacfix_HighpassFilterFixDec32C(int16_t *io,
|
||||
int16_t len,
|
||||
const int16_t *coefficient,
|
||||
int32_t *state)
|
||||
{
|
||||
int k;
|
||||
int32_t a1 = 0, b1 = 0, c = 0, in = 0;
|
||||
|
||||
@ -10,26 +10,26 @@
|
||||
|
||||
#include "webrtc/modules/audio_coding/codecs/isac/fix/source/filterbank_internal.h"
|
||||
|
||||
// WebRtcIsacfix_AllpassFilter2FixDec16 function optimized for MIPSDSP platform
|
||||
// Bit-exact with WebRtcIsacfix_AllpassFilter2FixDec16C from filterbanks.c
|
||||
// WebRtcIsacfix_AllpassFilter2FixDec16 function optimized for MIPSDSP platform.
|
||||
// Bit-exact with WebRtcIsacfix_AllpassFilter2FixDec16C from filterbanks.c.
|
||||
void WebRtcIsacfix_AllpassFilter2FixDec16MIPS(
|
||||
int16_t *data_ch1, // Input and output in channel 1, in Q0
|
||||
int16_t *data_ch2, // Input and output in channel 2, in Q0
|
||||
const int16_t *factor_ch1, // Scaling factor for channel 1, in Q15
|
||||
const int16_t *factor_ch2, // Scaling factor for channel 2, in Q15
|
||||
const int length, // Length of the data buffers
|
||||
int32_t *filter_state_ch1, // Filter state for channel 1, in Q16
|
||||
int32_t *filter_state_ch2) { // Filter state for channel 2, in Q16
|
||||
int16_t* data_ch1, // Input and output in channel 1, in Q0.
|
||||
int16_t* data_ch2, // Input and output in channel 2, in Q0.
|
||||
const int16_t* factor_ch1, // Scaling factor for channel 1, in Q15.
|
||||
const int16_t* factor_ch2, // Scaling factor for channel 2, in Q15.
|
||||
const int length, // Length of the data buffers.
|
||||
int32_t* filter_state_ch1, // Filter state for channel 1, in Q16.
|
||||
int32_t* filter_state_ch2) { // Filter state for channel 2, in Q16.
|
||||
|
||||
int32_t st0_ch1, st1_ch1; // channel1 state variables
|
||||
int32_t st0_ch2, st1_ch2; // channel2 state variables
|
||||
int32_t f_ch10, f_ch11, f_ch20, f_ch21; // factor variables
|
||||
int32_t r0, r1, r2, r3, r4, r5; // temporary ragister variables
|
||||
int32_t st0_ch1, st1_ch1; // channel1 state variables.
|
||||
int32_t st0_ch2, st1_ch2; // channel2 state variables.
|
||||
int32_t f_ch10, f_ch11, f_ch20, f_ch21; // factor variables.
|
||||
int32_t r0, r1, r2, r3, r4, r5; // temporary register variables.
|
||||
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
// Load all the state and factor variables
|
||||
// Load all the state and factor variables.
|
||||
"lh %[f_ch10], 0(%[factor_ch1]) \n\t"
|
||||
"lh %[f_ch20], 0(%[factor_ch2]) \n\t"
|
||||
"lh %[f_ch11], 2(%[factor_ch1]) \n\t"
|
||||
@ -38,7 +38,7 @@ void WebRtcIsacfix_AllpassFilter2FixDec16MIPS(
|
||||
"lw %[st1_ch1], 4(%[filter_state_ch1]) \n\t"
|
||||
"lw %[st0_ch2], 0(%[filter_state_ch2]) \n\t"
|
||||
"lw %[st1_ch2], 4(%[filter_state_ch2]) \n\t"
|
||||
// Allpass filtering loop
|
||||
// Allpass filtering loop.
|
||||
"1: \n\t"
|
||||
"lh %[r0], 0(%[data_ch1]) \n\t"
|
||||
"lh %[r1], 0(%[data_ch2]) \n\t"
|
||||
@ -80,7 +80,7 @@ void WebRtcIsacfix_AllpassFilter2FixDec16MIPS(
|
||||
"subq_s.w %[st1_ch2], %[r3], %[r1] \n\t"
|
||||
"bgtz %[length], 1b \n\t"
|
||||
" addiu %[data_ch2], %[data_ch2], 2 \n\t"
|
||||
// Store channel states
|
||||
// Store channel states.
|
||||
"sw %[st0_ch1], 0(%[filter_state_ch1]) \n\t"
|
||||
"sw %[st1_ch1], 4(%[filter_state_ch1]) \n\t"
|
||||
"sw %[st0_ch2], 0(%[filter_state_ch2]) \n\t"
|
||||
@ -100,3 +100,143 @@ void WebRtcIsacfix_AllpassFilter2FixDec16MIPS(
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
}
|
||||
|
||||
// WebRtcIsacfix_HighpassFilterFixDec32 function optimized for MIPSDSP platform.
|
||||
// Bit-exact with WebRtcIsacfix_HighpassFilterFixDec32C from filterbanks.c.
|
||||
void WebRtcIsacfix_HighpassFilterFixDec32MIPS(int16_t* io,
|
||||
int16_t len,
|
||||
const int16_t* coefficient,
|
||||
int32_t* state) {
|
||||
int k;
|
||||
int32_t a1, a2, b1, b2, in;
|
||||
int32_t state0 = state[0];
|
||||
int32_t state1 = state[1];
|
||||
|
||||
int32_t c0, c1, c2, c3;
|
||||
int32_t c4, c5, c6, c7;
|
||||
int32_t state0_lo, state0_hi;
|
||||
int32_t state1_lo, state1_hi;
|
||||
int32_t t0, t1, t2, t3, t4, t5;
|
||||
|
||||
__asm __volatile (
|
||||
"lh %[c0], 0(%[coeff_ptr]) \n\t"
|
||||
"lh %[c1], 2(%[coeff_ptr]) \n\t"
|
||||
"lh %[c2], 4(%[coeff_ptr]) \n\t"
|
||||
"lh %[c3], 6(%[coeff_ptr]) \n\t"
|
||||
"sra %[state0_hi], %[state0], 16 \n\t"
|
||||
"sra %[state1_hi], %[state1], 16 \n\t"
|
||||
"andi %[state0_lo], %[state0], 0xFFFF \n\t"
|
||||
"andi %[state1_lo], %[state1], 0xFFFF \n\t"
|
||||
"lh %[c4], 8(%[coeff_ptr]) \n\t"
|
||||
"lh %[c5], 10(%[coeff_ptr]) \n\t"
|
||||
"lh %[c6], 12(%[coeff_ptr]) \n\t"
|
||||
"lh %[c7], 14(%[coeff_ptr]) \n\t"
|
||||
"sra %[state0_lo], %[state0_lo], 1 \n\t"
|
||||
"sra %[state1_lo], %[state1_lo], 1 \n\t"
|
||||
: [c0] "=&r" (c0), [c1] "=&r" (c1), [c2] "=&r" (c2), [c3] "=&r" (c3),
|
||||
[c4] "=&r" (c4), [c5] "=&r" (c5), [c6] "=&r" (c6), [c7] "=&r" (c7),
|
||||
[state0_hi] "=&r" (state0_hi), [state0_lo] "=&r" (state0_lo),
|
||||
[state1_hi] "=&r" (state1_hi), [state1_lo] "=&r" (state1_lo)
|
||||
: [coeff_ptr] "r" (coefficient), [state0] "r" (state0),
|
||||
[state1] "r" (state1)
|
||||
: "memory"
|
||||
);
|
||||
|
||||
for (k = 0; k < len; k++) {
|
||||
in = (int32_t)io[k];
|
||||
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"mul %[t2], %[c4], %[state0_lo] \n\t"
|
||||
"mul %[t0], %[c5], %[state0_lo] \n\t"
|
||||
"mul %[t1], %[c4], %[state0_hi] \n\t"
|
||||
"mul %[a1], %[c5], %[state0_hi] \n\t"
|
||||
"mul %[t5], %[c6], %[state1_lo] \n\t"
|
||||
"mul %[t3], %[c7], %[state1_lo] \n\t"
|
||||
"mul %[t4], %[c6], %[state1_hi] \n\t"
|
||||
"mul %[b1], %[c7], %[state1_hi] \n\t"
|
||||
"shra_r.w %[t2], %[t2], 15 \n\t"
|
||||
"shra_r.w %[t0], %[t0], 15 \n\t"
|
||||
"addu %[t1], %[t1], %[t2] \n\t"
|
||||
"addu %[a1], %[a1], %[t0] \n\t"
|
||||
"sra %[t1], %[t1], 16 \n\t"
|
||||
"addu %[a1], %[a1], %[t1] \n\t"
|
||||
"shra_r.w %[t5], %[t5], 15 \n\t"
|
||||
"shra_r.w %[t3], %[t3], 15 \n\t"
|
||||
"addu %[t4], %[t4], %[t5] \n\t"
|
||||
"addu %[b1], %[b1], %[t3] \n\t"
|
||||
"sra %[t4], %[t4], 16 \n\t"
|
||||
"addu %[b1], %[b1], %[t4] \n\t"
|
||||
"mul %[t2], %[c0], %[state0_lo] \n\t"
|
||||
"mul %[t0], %[c1], %[state0_lo] \n\t"
|
||||
"mul %[t1], %[c0], %[state0_hi] \n\t"
|
||||
"mul %[a2], %[c1], %[state0_hi] \n\t"
|
||||
"mul %[t5], %[c2], %[state1_lo] \n\t"
|
||||
"mul %[t3], %[c3], %[state1_lo] \n\t"
|
||||
"mul %[t4], %[c2], %[state1_hi] \n\t"
|
||||
"mul %[b2], %[c3], %[state1_hi] \n\t"
|
||||
"shra_r.w %[t2], %[t2], 15 \n\t"
|
||||
"shra_r.w %[t0], %[t0], 15 \n\t"
|
||||
"addu %[t1], %[t1], %[t2] \n\t"
|
||||
"addu %[a2], %[a2], %[t0] \n\t"
|
||||
"sra %[t1], %[t1], 16 \n\t"
|
||||
"addu %[a2], %[a2], %[t1] \n\t"
|
||||
"shra_r.w %[t5], %[t5], 15 \n\t"
|
||||
"shra_r.w %[t3], %[t3], 15 \n\t"
|
||||
"addu %[t4], %[t4], %[t5] \n\t"
|
||||
"addu %[b2], %[b2], %[t3] \n\t"
|
||||
"sra %[t4], %[t4], 16 \n\t"
|
||||
"addu %[b2], %[b2], %[t4] \n\t"
|
||||
"addu %[a1], %[a1], %[b1] \n\t"
|
||||
"sra %[a1], %[a1], 7 \n\t"
|
||||
"addu %[a1], %[a1], %[in] \n\t"
|
||||
"sll %[t0], %[in], 2 \n\t"
|
||||
"addu %[a2], %[a2], %[b2] \n\t"
|
||||
"subu %[t0], %[t0], %[a2] \n\t"
|
||||
"shll_s.w %[a1], %[a1], 16 \n\t"
|
||||
"shll_s.w %[t0], %[t0], 2 \n\t"
|
||||
"sra %[a1], %[a1], 16 \n\t"
|
||||
"addu %[state1_hi], %[state0_hi], $0 \n\t"
|
||||
"addu %[state1_lo], %[state0_lo], $0 \n\t"
|
||||
"sra %[state0_hi], %[t0], 16 \n\t"
|
||||
"andi %[state0_lo], %[t0], 0xFFFF \n\t"
|
||||
"sra %[state0_lo], %[state0_lo], 1 \n\t"
|
||||
".set pop \n\t"
|
||||
: [a1] "=&r" (a1), [b1] "=&r" (b1), [a2] "=&r" (a2), [b2] "=&r" (b2),
|
||||
[state0_hi] "+r" (state0_hi), [state0_lo] "+r" (state0_lo),
|
||||
[state1_hi] "+r" (state1_hi), [state1_lo] "+r" (state1_lo),
|
||||
[t0] "=&r" (t0), [t1] "=&r" (t1), [t2] "=&r" (t2),
|
||||
[t3] "=&r" (t3), [t4] "=&r" (t4), [t5] "=&r" (t5)
|
||||
: [c0] "r" (c0), [c1] "r" (c1), [c2] "r" (c2), [c3] "r" (c3),
|
||||
[c4] "r" (c4), [c5] "r" (c5), [c6] "r" (c6), [c7] "r" (c7),
|
||||
[in] "r" (in)
|
||||
: "hi", "lo"
|
||||
);
|
||||
io[k] = (int16_t)a1;
|
||||
}
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
#if !defined(MIPS_DSP_R2_LE)
|
||||
"sll %[state0_hi], %[state0_hi], 16 \n\t"
|
||||
"sll %[state0_lo], %[state0_lo], 1 \n\t"
|
||||
"sll %[state1_hi], %[state1_hi], 16 \n\t"
|
||||
"sll %[state1_lo], %[state1_lo], 1 \n\t"
|
||||
"or %[state0_hi], %[state0_hi], %[state0_lo] \n\t"
|
||||
"or %[state1_hi], %[state1_hi], %[state1_lo] \n\t"
|
||||
#else
|
||||
"sll %[state0_lo], %[state0_lo], 1 \n\t"
|
||||
"sll %[state1_lo], %[state1_lo], 1 \n\t"
|
||||
"precr_sra.ph.w %[state0_hi], %[state0_lo], 0 \n\t"
|
||||
"precr_sra.ph.w %[state1_hi], %[state1_lo], 0 \n\t"
|
||||
#endif
|
||||
"sw %[state0_hi], 0(%[state]) \n\t"
|
||||
"sw %[state1_hi], 4(%[state]) \n\t"
|
||||
".set pop \n\t"
|
||||
: [state0_hi] "+r" (state0_hi), [state0_lo] "+r" (state0_lo),
|
||||
[state1_hi] "+r" (state1_hi), [state1_lo] "+r" (state1_lo)
|
||||
: [state] "r" (state)
|
||||
: "memory"
|
||||
);
|
||||
}
|
||||
|
||||
@ -86,6 +86,13 @@ TEST_F(FilterBanksTest, HighpassFilterFixDec32Test) {
|
||||
-1280, -8554, -14496, -7561, -23541, -27263, -30560, -32768, -3441, -32768,
|
||||
25203, -27550, 22419};
|
||||
#endif
|
||||
HighpassFilterFixDec32 WebRtcIsacfix_HighpassFilterFixDec32;
|
||||
#if defined(MIPS_DSP_R1_LE)
|
||||
WebRtcIsacfix_HighpassFilterFixDec32 =
|
||||
WebRtcIsacfix_HighpassFilterFixDec32MIPS;
|
||||
#else
|
||||
WebRtcIsacfix_HighpassFilterFixDec32 = WebRtcIsacfix_HighpassFilterFixDec32C;
|
||||
#endif
|
||||
|
||||
for (int i = 0; i < kSamples; i++) {
|
||||
in[i] = WEBRTC_SPL_WORD32_MAX / (i + 1);
|
||||
|
||||
@ -209,9 +209,17 @@ static void WebRtcIsacfix_InitNeon(void) {
|
||||
static void WebRtcIsacfix_InitMIPS(void) {
|
||||
WebRtcIsacfix_AutocorrFix = WebRtcIsacfix_AutocorrMIPS;
|
||||
WebRtcIsacfix_FilterMaLoopFix = WebRtcIsacfix_FilterMaLoopMIPS;
|
||||
WebRtcIsacfix_Spec2Time = WebRtcIsacfix_Spec2TimeMIPS;
|
||||
WebRtcIsacfix_Time2Spec = WebRtcIsacfix_Time2SpecMIPS;
|
||||
#if defined(MIPS_DSP_R1_LE)
|
||||
WebRtcIsacfix_AllpassFilter2FixDec16 =
|
||||
WebRtcIsacfix_AllpassFilter2FixDec16MIPS;
|
||||
WebRtcIsacfix_HighpassFilterFixDec32 =
|
||||
WebRtcIsacfix_HighpassFilterFixDec32MIPS;
|
||||
#endif
|
||||
#if defined(MIPS_DSP_R2_LE)
|
||||
WebRtcIsacfix_CalculateResidualEnergy =
|
||||
WebRtcIsacfix_CalculateResidualEnergyMIPS;
|
||||
#endif
|
||||
}
|
||||
#endif
|
||||
@ -300,10 +308,11 @@ int16_t WebRtcIsacfix_EncoderInit(ISACFIX_MainStruct *ISAC_main_inst,
|
||||
WebRtcIsacfix_CalculateResidualEnergy =
|
||||
WebRtcIsacfix_CalculateResidualEnergyC;
|
||||
WebRtcIsacfix_AllpassFilter2FixDec16 = WebRtcIsacfix_AllpassFilter2FixDec16C;
|
||||
WebRtcIsacfix_HighpassFilterFixDec32 = WebRtcIsacfix_HighpassFilterFixDec32C;
|
||||
WebRtcIsacfix_Time2Spec = WebRtcIsacfix_Time2SpecC;
|
||||
WebRtcIsacfix_Spec2Time = WebRtcIsacfix_Spec2TimeC;
|
||||
WebRtcIsacfix_MatrixProduct1 = WebRtcIsacfix_MatrixProduct1C;
|
||||
WebRtcIsacfix_MatrixProduct2 = WebRtcIsacfix_MatrixProduct2C ;
|
||||
WebRtcIsacfix_MatrixProduct2 = WebRtcIsacfix_MatrixProduct2C;
|
||||
|
||||
#ifdef WEBRTC_DETECT_ARM_NEON
|
||||
if ((WebRtc_GetCPUFeaturesARM() & kCPUFeatureNEON) != 0) {
|
||||
|
||||
@ -47,12 +47,14 @@
|
||||
'lpc_masking_model.c',
|
||||
'lpc_tables.c',
|
||||
'pitch_estimator.c',
|
||||
'pitch_estimator_c.c',
|
||||
'pitch_filter.c',
|
||||
'pitch_filter_c.c',
|
||||
'pitch_gain_tables.c',
|
||||
'pitch_lag_tables.c',
|
||||
'spectrum_ar_model_tables.c',
|
||||
'transform.c',
|
||||
'transform_tables.c',
|
||||
'arith_routins.h',
|
||||
'bandwidth_estimator.h',
|
||||
'codec.h',
|
||||
@ -89,9 +91,12 @@
|
||||
'sources': [
|
||||
'filters_mips.c',
|
||||
'lattice_mips.c',
|
||||
'pitch_estimator_mips.c',
|
||||
'transform_mips.c',
|
||||
],
|
||||
'sources!': [
|
||||
'lattice_c.c',
|
||||
'pitch_estimator_c.c',
|
||||
],
|
||||
'conditions': [
|
||||
['mips_dsp_rev>0', {
|
||||
@ -101,6 +106,7 @@
|
||||
}],
|
||||
['mips_dsp_rev>1', {
|
||||
'sources': [
|
||||
'lpc_masking_model_mips.c',
|
||||
'pitch_filter_mips.c',
|
||||
],
|
||||
'sources!': [
|
||||
|
||||
@ -62,6 +62,15 @@ int32_t WebRtcIsacfix_CalculateResidualEnergyNeon(int lpc_order,
|
||||
int* q_val_residual_energy);
|
||||
#endif
|
||||
|
||||
#if defined(MIPS_DSP_R2_LE)
|
||||
int32_t WebRtcIsacfix_CalculateResidualEnergyMIPS(int lpc_order,
|
||||
int32_t q_val_corr,
|
||||
int q_val_polynomial,
|
||||
int16_t* a_polynomial,
|
||||
int32_t* corr_coeffs,
|
||||
int* q_val_residual_energy);
|
||||
#endif
|
||||
|
||||
#ifdef __cplusplus
|
||||
} /* extern "C" */
|
||||
#endif
|
||||
|
||||
@ -0,0 +1,237 @@
|
||||
/*
|
||||
* Copyright (c) 2014 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.
|
||||
*/
|
||||
|
||||
#include "webrtc/modules/audio_coding/codecs/isac/fix/source/lpc_masking_model.h"
|
||||
|
||||
// MIPS DSPR2 optimization for function WebRtcIsacfix_CalculateResidualEnergy
|
||||
// Bit-exact with WebRtcIsacfix_CalculateResidualEnergyC from file
|
||||
// lpc_masking_model.c
|
||||
int32_t WebRtcIsacfix_CalculateResidualEnergyMIPS(int lpc_order,
|
||||
int32_t q_val_corr,
|
||||
int q_val_polynomial,
|
||||
int16_t* a_polynomial,
|
||||
int32_t* corr_coeffs,
|
||||
int* q_val_residual_energy) {
|
||||
|
||||
int i = 0, j = 0;
|
||||
int shift_internal = 0, shift_norm = 0;
|
||||
int32_t tmp32 = 0, word32_high = 0, word32_low = 0, residual_energy = 0;
|
||||
int32_t tmp_corr_c = corr_coeffs[0];
|
||||
int16_t* tmp_a_poly = &a_polynomial[0];
|
||||
int32_t sum64_hi = 0;
|
||||
int32_t sum64_lo = 0;
|
||||
|
||||
for (j = 0; j <= lpc_order; j++) {
|
||||
// For the case of i == 0:
|
||||
// residual_energy +=
|
||||
// a_polynomial[j] * corr_coeffs[i] * a_polynomial[j - i];
|
||||
|
||||
int32_t tmp2, tmp3;
|
||||
int16_t sign_1;
|
||||
int16_t sign_2;
|
||||
int16_t sign_3;
|
||||
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"lh %[tmp2], 0(%[tmp_a_poly]) \n\t"
|
||||
"mul %[tmp32], %[tmp2], %[tmp2] \n\t"
|
||||
"addiu %[tmp_a_poly], %[tmp_a_poly], 2 \n\t"
|
||||
"sra %[sign_2], %[sum64_hi], 31 \n\t"
|
||||
"mult $ac0, %[tmp32], %[tmp_corr_c] \n\t"
|
||||
"shilov $ac0, %[shift_internal] \n\t"
|
||||
"mfhi %[tmp2], $ac0 \n\t"
|
||||
"mflo %[tmp3], $ac0 \n\t"
|
||||
"sra %[sign_1], %[tmp2], 31 \n\t"
|
||||
"xor %[sign_3], %[sign_1], %[sign_2] \n\t"
|
||||
".set pop \n\t"
|
||||
: [tmp2] "=&r" (tmp2), [tmp3] "=&r" (tmp3), [tmp32] "=&r" (tmp32),
|
||||
[tmp_a_poly] "+r" (tmp_a_poly), [sign_1] "=&r" (sign_1),
|
||||
[sign_3] "=&r" (sign_3), [sign_2] "=&r" (sign_2),
|
||||
[sum64_hi] "+r" (sum64_hi), [sum64_lo] "+r" (sum64_lo)
|
||||
: [tmp_corr_c] "r" (tmp_corr_c), [shift_internal] "r" (shift_internal)
|
||||
: "hi", "lo", "memory"
|
||||
);
|
||||
|
||||
if (sign_3 != 0) {
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"addsc %[sum64_lo], %[sum64_lo], %[tmp3] \n\t"
|
||||
"addwc %[sum64_hi], %[sum64_hi], %[tmp2] \n\t"
|
||||
".set pop \n\t"
|
||||
: [sum64_hi] "+r" (sum64_hi), [sum64_lo] "+r" (sum64_lo)
|
||||
: [tmp2] "r" (tmp2), [tmp3] "r" (tmp3)
|
||||
: "hi", "lo", "memory"
|
||||
);
|
||||
} else {
|
||||
if (((!(sign_1 || sign_2)) && (0x7FFFFFFF - sum64_hi < tmp2)) ||
|
||||
((sign_1 && sign_2) && (sum64_hi + tmp2 > 0))) {
|
||||
// Shift right for overflow.
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"addiu %[shift_internal], %[shift_internal], 1 \n\t"
|
||||
"prepend %[sum64_lo], %[sum64_hi], 1 \n\t"
|
||||
"sra %[sum64_hi], %[sum64_hi], 1 \n\t"
|
||||
"prepend %[tmp3], %[tmp2], 1 \n\t"
|
||||
"sra %[tmp2], %[tmp2], 1 \n\t"
|
||||
"addsc %[sum64_lo], %[sum64_lo], %[tmp3] \n\t"
|
||||
"addwc %[sum64_hi], %[sum64_hi], %[tmp2] \n\t"
|
||||
".set pop \n\t"
|
||||
: [tmp2] "+r" (tmp2), [tmp3] "+r" (tmp3),
|
||||
[shift_internal] "+r" (shift_internal),
|
||||
[sum64_hi] "+r" (sum64_hi), [sum64_lo] "+r" (sum64_lo)
|
||||
:
|
||||
: "hi", "lo", "memory"
|
||||
);
|
||||
} else {
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"addsc %[sum64_lo], %[sum64_lo], %[tmp3] \n\t"
|
||||
"addwc %[sum64_hi], %[sum64_hi], %[tmp2] \n\t"
|
||||
".set pop \n\t"
|
||||
: [sum64_hi] "+r" (sum64_hi), [sum64_lo] "+r" (sum64_lo)
|
||||
: [tmp2] "r" (tmp2), [tmp3] "r" (tmp3)
|
||||
: "hi", "lo", "memory"
|
||||
);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
for (i = 1; i <= lpc_order; i++) {
|
||||
tmp_corr_c = corr_coeffs[i];
|
||||
int16_t* tmp_a_poly_j = &a_polynomial[i];
|
||||
int16_t* tmp_a_poly_j_i = &a_polynomial[0];
|
||||
for (j = i; j <= lpc_order; j++) {
|
||||
// For the case of i = 1 .. lpc_order:
|
||||
// residual_energy +=
|
||||
// a_polynomial[j] * corr_coeffs[i] * a_polynomial[j - i] * 2;
|
||||
|
||||
int32_t tmp2, tmp3;
|
||||
int16_t sign_1;
|
||||
int16_t sign_2;
|
||||
int16_t sign_3;
|
||||
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"lh %[tmp3], 0(%[tmp_a_poly_j]) \n\t"
|
||||
"lh %[tmp2], 0(%[tmp_a_poly_j_i]) \n\t"
|
||||
"addiu %[tmp_a_poly_j], %[tmp_a_poly_j], 2 \n\t"
|
||||
"addiu %[tmp_a_poly_j_i], %[tmp_a_poly_j_i], 2 \n\t"
|
||||
"mul %[tmp32], %[tmp3], %[tmp2] \n\t"
|
||||
"sll %[tmp32], %[tmp32], 1 \n\t"
|
||||
"mult $ac0, %[tmp32], %[tmp_corr_c] \n\t"
|
||||
"shilov $ac0, %[shift_internal] \n\t"
|
||||
"mfhi %[tmp2], $ac0 \n\t"
|
||||
"mflo %[tmp3], $ac0 \n\t"
|
||||
"sra %[sign_1], %[tmp2], 31 \n\t"
|
||||
"sra %[sign_2], %[sum64_hi], 31 \n\t"
|
||||
"xor %[sign_3], %[sign_1], %[sign_2] \n\t"
|
||||
".set pop \n\t"
|
||||
: [tmp2] "=&r" (tmp2), [tmp3] "=&r" (tmp3), [tmp32] "=&r" (tmp32),
|
||||
[tmp_a_poly_j] "+r" (tmp_a_poly_j), [sign_1] "=&r" (sign_1),
|
||||
[tmp_a_poly_j_i] "+r" (tmp_a_poly_j_i), [sign_2] "=&r" (sign_2),
|
||||
[sign_3] "=&r" (sign_3), [sum64_hi] "+r" (sum64_hi),
|
||||
[sum64_lo] "+r" (sum64_lo)
|
||||
: [tmp_corr_c] "r" (tmp_corr_c), [shift_internal] "r" (shift_internal)
|
||||
: "hi", "lo", "memory"
|
||||
);
|
||||
if (sign_3 != 0) {
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"addsc %[sum64_lo], %[sum64_lo], %[tmp3] \n\t"
|
||||
"addwc %[sum64_hi], %[sum64_hi], %[tmp2] \n\t"
|
||||
".set pop \n\t"
|
||||
: [tmp2] "+r" (tmp2), [tmp3] "+r" (tmp3), [sum64_hi] "+r" (sum64_hi),
|
||||
[sum64_lo] "+r" (sum64_lo)
|
||||
:
|
||||
:"memory"
|
||||
);
|
||||
} else {
|
||||
// Test overflow and sum the result.
|
||||
if (((!(sign_1 || sign_2)) && (0x7FFFFFFF - sum64_hi < tmp2)) ||
|
||||
((sign_1 && sign_2) && (sum64_hi + tmp2 > 0))) {
|
||||
// Shift right for overflow.
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"addiu %[shift_internal], %[shift_internal], 1 \n\t"
|
||||
"prepend %[sum64_lo], %[sum64_hi], 1 \n\t"
|
||||
"sra %[sum64_hi], %[sum64_hi], 1 \n\t"
|
||||
"prepend %[tmp3], %[tmp2], 1 \n\t"
|
||||
"sra %[tmp2], %[tmp2], 1 \n\t"
|
||||
"addsc %[sum64_lo], %[sum64_lo], %[tmp3] \n\t"
|
||||
"addwc %[sum64_hi], %[sum64_hi], %[tmp2] \n\t"
|
||||
".set pop \n\t"
|
||||
: [tmp2] "+r" (tmp2), [tmp3] "+r" (tmp3),
|
||||
[shift_internal] "+r" (shift_internal),
|
||||
[sum64_hi] "+r" (sum64_hi), [sum64_lo] "+r" (sum64_lo)
|
||||
:
|
||||
: "hi", "lo", "memory"
|
||||
);
|
||||
} else {
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"addsc %[sum64_lo], %[sum64_lo], %[tmp3] \n\t"
|
||||
"addwc %[sum64_hi], %[sum64_hi], %[tmp2] \n\t"
|
||||
".set pop \n\t"
|
||||
: [tmp2] "+r" (tmp2), [tmp3] "+r" (tmp3),
|
||||
[sum64_hi] "+r" (sum64_hi), [sum64_lo] "+r" (sum64_lo)
|
||||
:
|
||||
: "hi", "lo", "memory"
|
||||
);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
word32_high = sum64_hi;
|
||||
word32_low = sum64_lo;
|
||||
|
||||
// Calculate the value of shifting (shift_norm) for the 64-bit sum.
|
||||
if (word32_high != 0) {
|
||||
shift_norm = 32 - WebRtcSpl_NormW32(word32_high);
|
||||
int tmp1;
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"srl %[residual_energy], %[sum64_lo], %[shift_norm] \n\t"
|
||||
"li %[tmp1], 32 \n\t"
|
||||
"subu %[tmp1], %[tmp1], %[shift_norm] \n\t"
|
||||
"sll %[tmp1], %[sum64_hi], %[tmp1] \n\t"
|
||||
"or %[residual_energy], %[residual_energy], %[tmp1] \n\t"
|
||||
".set pop \n\t"
|
||||
: [residual_energy] "=&r" (residual_energy), [tmp1]"=&r"(tmp1),
|
||||
[sum64_hi] "+r" (sum64_hi), [sum64_lo] "+r" (sum64_lo)
|
||||
: [shift_norm] "r" (shift_norm)
|
||||
: "memory"
|
||||
);
|
||||
} else {
|
||||
if ((word32_low & 0x80000000) != 0) {
|
||||
shift_norm = 1;
|
||||
residual_energy = (uint32_t)word32_low >> 1;
|
||||
} else {
|
||||
shift_norm = WebRtcSpl_NormW32(word32_low);
|
||||
residual_energy = word32_low << shift_norm;
|
||||
shift_norm = -shift_norm;
|
||||
}
|
||||
}
|
||||
|
||||
// Q(q_val_polynomial * 2) * Q(q_val_corr) >> shift_internal >> shift_norm
|
||||
// = Q(q_val_corr - shift_internal - shift_norm + q_val_polynomial * 2)
|
||||
*q_val_residual_energy =
|
||||
q_val_corr - shift_internal - shift_norm + q_val_polynomial * 2;
|
||||
|
||||
return residual_energy;
|
||||
}
|
||||
@ -29,7 +29,7 @@ static const int16_t kACoefQ12[3] = {
|
||||
|
||||
|
||||
|
||||
static __inline int32_t Log2Q8( uint32_t x ) {
|
||||
__inline int32_t WebRtcIsacfix_Log2Q8( uint32_t x ) {
|
||||
|
||||
int32_t zeros, lg2;
|
||||
int16_t frac;
|
||||
@ -153,109 +153,7 @@ static void FindFour32(int32_t *in, int16_t length, int16_t *bestind)
|
||||
|
||||
|
||||
|
||||
static void PCorr2Q32(const int16_t *in, int32_t *logcorQ8)
|
||||
{
|
||||
int16_t scaling,n,k;
|
||||
int32_t ysum32,csum32, lys, lcs;
|
||||
int32_t oneQ8;
|
||||
|
||||
|
||||
const int16_t *x, *inptr;
|
||||
|
||||
oneQ8 = WEBRTC_SPL_LSHIFT_W32((int32_t)1, 8); // 1.00 in Q8
|
||||
|
||||
x = in + PITCH_MAX_LAG/2 + 2;
|
||||
scaling = WebRtcSpl_GetScalingSquare ((int16_t *) in, PITCH_CORR_LEN2, PITCH_CORR_LEN2);
|
||||
ysum32 = 1;
|
||||
csum32 = 0;
|
||||
x = in + PITCH_MAX_LAG/2 + 2;
|
||||
for (n = 0; n < PITCH_CORR_LEN2; n++) {
|
||||
ysum32 += WEBRTC_SPL_MUL_16_16_RSFT( (int16_t) in[n],(int16_t) in[n], scaling); // Q0
|
||||
csum32 += WEBRTC_SPL_MUL_16_16_RSFT((int16_t) x[n],(int16_t) in[n], scaling); // Q0
|
||||
}
|
||||
|
||||
logcorQ8 += PITCH_LAG_SPAN2 - 1;
|
||||
|
||||
lys=Log2Q8((uint32_t) ysum32); // Q8
|
||||
lys=WEBRTC_SPL_RSHIFT_W32(lys, 1); //sqrt(ysum);
|
||||
|
||||
if (csum32>0) {
|
||||
|
||||
lcs=Log2Q8((uint32_t) csum32); // 2log(csum) in Q8
|
||||
|
||||
if (lcs>(lys + oneQ8) ){ // csum/sqrt(ysum) > 2 in Q8
|
||||
*logcorQ8 = lcs - lys; // log2(csum/sqrt(ysum))
|
||||
} else {
|
||||
*logcorQ8 = oneQ8; // 1.00
|
||||
}
|
||||
|
||||
} else {
|
||||
*logcorQ8 = 0;
|
||||
}
|
||||
|
||||
|
||||
for (k = 1; k < PITCH_LAG_SPAN2; k++) {
|
||||
inptr = &in[k];
|
||||
ysum32 -= WEBRTC_SPL_MUL_16_16_RSFT( (int16_t) in[k-1],(int16_t) in[k-1], scaling);
|
||||
ysum32 += WEBRTC_SPL_MUL_16_16_RSFT( (int16_t) in[PITCH_CORR_LEN2 + k - 1],(int16_t) in[PITCH_CORR_LEN2 + k - 1], scaling);
|
||||
|
||||
#ifdef WEBRTC_ARCH_ARM_NEON
|
||||
{
|
||||
int32_t vbuff[4];
|
||||
int32x4_t int_32x4_sum = vmovq_n_s32(0);
|
||||
// Can't shift a Neon register to right with a non-constant shift value.
|
||||
int32x4_t int_32x4_scale = vdupq_n_s32(-scaling);
|
||||
// Assert a codition used in loop unrolling at compile-time.
|
||||
COMPILE_ASSERT(PITCH_CORR_LEN2 %4 == 0);
|
||||
|
||||
for (n = 0; n < PITCH_CORR_LEN2; n += 4) {
|
||||
int16x4_t int_16x4_x = vld1_s16(&x[n]);
|
||||
int16x4_t int_16x4_in = vld1_s16(&inptr[n]);
|
||||
int32x4_t int_32x4 = vmull_s16(int_16x4_x, int_16x4_in);
|
||||
int_32x4 = vshlq_s32(int_32x4, int_32x4_scale);
|
||||
int_32x4_sum = vaddq_s32(int_32x4_sum, int_32x4);
|
||||
}
|
||||
|
||||
// Use vector store to avoid long stall from data trasferring
|
||||
// from vector to general register.
|
||||
vst1q_s32(vbuff, int_32x4_sum);
|
||||
csum32 = vbuff[0] + vbuff[1];
|
||||
csum32 += vbuff[2];
|
||||
csum32 += vbuff[3];
|
||||
}
|
||||
#else
|
||||
csum32 = 0;
|
||||
if(scaling == 0) {
|
||||
for (n = 0; n < PITCH_CORR_LEN2; n++) {
|
||||
csum32 += x[n] * inptr[n];
|
||||
}
|
||||
} else {
|
||||
for (n = 0; n < PITCH_CORR_LEN2; n++) {
|
||||
csum32 += (x[n] * inptr[n]) >> scaling;
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
logcorQ8--;
|
||||
|
||||
lys=Log2Q8((uint32_t)ysum32); // Q8
|
||||
lys=WEBRTC_SPL_RSHIFT_W32(lys, 1); //sqrt(ysum);
|
||||
|
||||
if (csum32>0) {
|
||||
|
||||
lcs=Log2Q8((uint32_t) csum32); // 2log(csum) in Q8
|
||||
|
||||
if (lcs>(lys + oneQ8) ){ // csum/sqrt(ysum) > 2
|
||||
*logcorQ8 = lcs - lys; // log2(csum/sqrt(ysum))
|
||||
} else {
|
||||
*logcorQ8 = oneQ8; // 1.00
|
||||
}
|
||||
|
||||
} else {
|
||||
*logcorQ8 = 0;
|
||||
}
|
||||
}
|
||||
}
|
||||
extern void WebRtcIsacfix_PCorr2Q32(const int16_t *in, int32_t *logcorQ8);
|
||||
|
||||
|
||||
|
||||
@ -311,12 +209,13 @@ void WebRtcIsacfix_InitialPitch(const int16_t *in, /* Q0 */
|
||||
|
||||
|
||||
/* compute correlation for first and second half of the frame */
|
||||
PCorr2Q32(buf_dec16, crrvecQ8_1);
|
||||
PCorr2Q32(buf_dec16 + PITCH_CORR_STEP2, crrvecQ8_2);
|
||||
WebRtcIsacfix_PCorr2Q32(buf_dec16, crrvecQ8_1);
|
||||
WebRtcIsacfix_PCorr2Q32(buf_dec16 + PITCH_CORR_STEP2, crrvecQ8_2);
|
||||
|
||||
|
||||
/* bias towards pitch lag of previous frame */
|
||||
tmp32a = Log2Q8((uint32_t) old_lagQ8) - 2304; // log2(0.5*oldlag) in Q8
|
||||
tmp32a = WebRtcIsacfix_Log2Q8((uint32_t) old_lagQ8) - 2304;
|
||||
// log2(0.5*oldlag) in Q8
|
||||
tmp32b = WEBRTC_SPL_MUL_16_16_RSFT(oldgQ12,oldgQ12, 10); //Q12 & * 4.0;
|
||||
gain_bias16 = (int16_t) tmp32b; //Q12
|
||||
if (gain_bias16 > 3276) gain_bias16 = 3276; // 0.8 in Q12
|
||||
@ -325,7 +224,7 @@ void WebRtcIsacfix_InitialPitch(const int16_t *in, /* Q0 */
|
||||
for (k = 0; k < PITCH_LAG_SPAN2; k++)
|
||||
{
|
||||
if (crrvecQ8_1[k]>0) {
|
||||
tmp32b = Log2Q8((uint32_t) (k + (PITCH_MIN_LAG/2-2)));
|
||||
tmp32b = WebRtcIsacfix_Log2Q8((uint32_t) (k + (PITCH_MIN_LAG/2-2)));
|
||||
tmp16a = (int16_t) (tmp32b - tmp32a); // Q8 & fabs(ratio)<4
|
||||
tmp32c = WEBRTC_SPL_MUL_16_16_RSFT(tmp16a,tmp16a, 6); //Q10
|
||||
tmp16b = (int16_t) tmp32c; // Q10 & <8
|
||||
@ -334,7 +233,8 @@ void WebRtcIsacfix_InitialPitch(const int16_t *in, /* Q0 */
|
||||
tmp16d = Exp2Q10((int16_t) -tmp16c); //Q10
|
||||
tmp32c = WEBRTC_SPL_MUL_16_16_RSFT(gain_bias16,tmp16d,13); // Q10 & * 0.5
|
||||
bias16 = (int16_t) (1024 + tmp32c); // Q10
|
||||
tmp32b = Log2Q8((uint32_t) bias16) - 2560; // Q10 in -> Q8 out with 10*2^8 offset
|
||||
tmp32b = WebRtcIsacfix_Log2Q8((uint32_t)bias16) - 2560;
|
||||
// Q10 in -> Q8 out with 10*2^8 offset
|
||||
crrvecQ8_1[k] += tmp32b ; // -10*2^8 offset
|
||||
}
|
||||
}
|
||||
@ -407,7 +307,7 @@ void WebRtcIsacfix_InitialPitch(const int16_t *in, /* Q0 */
|
||||
xq[0] = WEBRTC_SPL_LSHIFT_W32(xq[0], 8);
|
||||
Intrp1DQ8(xq, fxq, yq, fyq);
|
||||
|
||||
tmp32a= Log2Q8((uint32_t) *yq) - 2048; // offset 8*2^8
|
||||
tmp32a= WebRtcIsacfix_Log2Q8((uint32_t) *yq) - 2048; // offset 8*2^8
|
||||
/* Bias towards short lags */
|
||||
/* log(pow(0.8, log(2.0 * *y )))/log(2.0) */
|
||||
tmp32b= WEBRTC_SPL_MUL_16_16_RSFT((int16_t) tmp32a, -42, 8);
|
||||
@ -437,10 +337,13 @@ void WebRtcIsacfix_InitialPitch(const int16_t *in, /* Q0 */
|
||||
tmp32b = (int32_t) (WEBRTC_SPL_LSHIFT_W32(tmp32a, 1)) - ratq; // Q8
|
||||
tmp32c = WEBRTC_SPL_MUL_16_16_RSFT((int16_t) tmp32b, (int16_t) tmp32b, 8); // Q8
|
||||
|
||||
tmp32b = (int32_t) tmp32c + (int32_t) WEBRTC_SPL_RSHIFT_W32(ratq, 1); // (k-r)^2 + 0.5 * r Q8
|
||||
tmp32c = Log2Q8((uint32_t) tmp32a) - 2048; // offset 8*2^8 , log2(0.5*k) Q8
|
||||
tmp32d = Log2Q8((uint32_t) tmp32b) - 2048; // offset 8*2^8 , log2(0.5*k) Q8
|
||||
tmp32e = tmp32c -tmp32d;
|
||||
tmp32b = (int32_t)tmp32c + (int32_t)WEBRTC_SPL_RSHIFT_W32(ratq, 1);
|
||||
// (k-r)^2 + 0.5 * r Q8
|
||||
tmp32c = WebRtcIsacfix_Log2Q8((uint32_t)tmp32a) - 2048;
|
||||
// offset 8*2^8 , log2(0.5*k) Q8
|
||||
tmp32d = WebRtcIsacfix_Log2Q8((uint32_t)tmp32b) - 2048;
|
||||
// offset 8*2^8 , log2(0.5*k) Q8
|
||||
tmp32e = tmp32c - tmp32d;
|
||||
|
||||
cv2q[k] += WEBRTC_SPL_RSHIFT_W32(tmp32e, 1);
|
||||
|
||||
@ -481,7 +384,7 @@ void WebRtcIsacfix_InitialPitch(const int16_t *in, /* Q0 */
|
||||
|
||||
/* Bias towards short lags */
|
||||
/* log(pow(0.8, log(2.0f * *y )))/log(2.0f) */
|
||||
tmp32a= Log2Q8((uint32_t) *yq) - 2048; // offset 8*2^8
|
||||
tmp32a= WebRtcIsacfix_Log2Q8((uint32_t) *yq) - 2048; // offset 8*2^8
|
||||
tmp32b= WEBRTC_SPL_MUL_16_16_RSFT((int16_t) tmp32a, -82, 8);
|
||||
tmp32c= tmp32b + 256;
|
||||
*fyq += tmp32c;
|
||||
|
||||
@ -58,4 +58,8 @@ void WebRtcIsacfix_DecimateAllpass32(const int16_t *in,
|
||||
int16_t N, /* number of input samples */
|
||||
int16_t *out); /* array of size N/2 */
|
||||
|
||||
int32_t WebRtcIsacfix_Log2Q8( uint32_t x );
|
||||
|
||||
void WebRtcIsacfix_PCorr2Q32(const int16_t* in, int32_t* logcorQ8);
|
||||
|
||||
#endif /* WEBRTC_MODULES_AUDIO_CODING_CODECS_ISAC_FIX_SOURCE_PITCH_ESTIMATOR_H_ */
|
||||
|
||||
@ -0,0 +1,122 @@
|
||||
/*
|
||||
* Copyright (c) 2014 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.
|
||||
*/
|
||||
|
||||
#include "webrtc/modules/audio_coding/codecs/isac/fix/source/pitch_estimator.h"
|
||||
|
||||
#ifdef WEBRTC_ARCH_ARM_NEON
|
||||
#include <arm_neon.h>
|
||||
#endif
|
||||
|
||||
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
|
||||
#include "webrtc/system_wrappers/interface/compile_assert_c.h"
|
||||
|
||||
extern int32_t WebRtcIsacfix_Log2Q8(uint32_t x);
|
||||
|
||||
void WebRtcIsacfix_PCorr2Q32(const int16_t* in, int32_t* logcorQ8) {
|
||||
int16_t scaling,n,k;
|
||||
int32_t ysum32,csum32, lys, lcs;
|
||||
int32_t oneQ8;
|
||||
const int16_t* x;
|
||||
const int16_t* inptr;
|
||||
|
||||
oneQ8 = WEBRTC_SPL_LSHIFT_W32((int32_t)1, 8); // 1.00 in Q8
|
||||
|
||||
x = in + PITCH_MAX_LAG / 2 + 2;
|
||||
scaling = WebRtcSpl_GetScalingSquare((int16_t*)in,
|
||||
PITCH_CORR_LEN2,
|
||||
PITCH_CORR_LEN2);
|
||||
ysum32 = 1;
|
||||
csum32 = 0;
|
||||
x = in + PITCH_MAX_LAG / 2 + 2;
|
||||
for (n = 0; n < PITCH_CORR_LEN2; n++) {
|
||||
ysum32 += WEBRTC_SPL_MUL_16_16_RSFT((int16_t)in[n],
|
||||
(int16_t)in[n],
|
||||
scaling); // Q0
|
||||
csum32 += WEBRTC_SPL_MUL_16_16_RSFT((int16_t)x[n],
|
||||
(int16_t)in[n],
|
||||
scaling); // Q0
|
||||
}
|
||||
logcorQ8 += PITCH_LAG_SPAN2 - 1;
|
||||
lys = WebRtcIsacfix_Log2Q8((uint32_t)ysum32); // Q8
|
||||
lys = WEBRTC_SPL_RSHIFT_W32(lys, 1); //sqrt(ysum);
|
||||
if (csum32 > 0) {
|
||||
lcs = WebRtcIsacfix_Log2Q8((uint32_t)csum32); // 2log(csum) in Q8
|
||||
if (lcs > (lys + oneQ8)) { // csum/sqrt(ysum) > 2 in Q8
|
||||
*logcorQ8 = lcs - lys; // log2(csum/sqrt(ysum))
|
||||
} else {
|
||||
*logcorQ8 = oneQ8; // 1.00
|
||||
}
|
||||
} else {
|
||||
*logcorQ8 = 0;
|
||||
}
|
||||
|
||||
|
||||
for (k = 1; k < PITCH_LAG_SPAN2; k++) {
|
||||
inptr = &in[k];
|
||||
ysum32 -= WEBRTC_SPL_MUL_16_16_RSFT((int16_t)in[k - 1],
|
||||
(int16_t)in[k - 1],
|
||||
scaling);
|
||||
ysum32 += WEBRTC_SPL_MUL_16_16_RSFT((int16_t)in[PITCH_CORR_LEN2 + k - 1],
|
||||
(int16_t)in[PITCH_CORR_LEN2 + k - 1],
|
||||
scaling);
|
||||
#ifdef WEBRTC_ARCH_ARM_NEON
|
||||
{
|
||||
int32_t vbuff[4];
|
||||
int32x4_t int_32x4_sum = vmovq_n_s32(0);
|
||||
// Can't shift a Neon register to right with a non-constant shift value.
|
||||
int32x4_t int_32x4_scale = vdupq_n_s32(-scaling);
|
||||
// Assert a codition used in loop unrolling at compile-time.
|
||||
COMPILE_ASSERT(PITCH_CORR_LEN2 %4 == 0);
|
||||
|
||||
for (n = 0; n < PITCH_CORR_LEN2; n += 4) {
|
||||
int16x4_t int_16x4_x = vld1_s16(&x[n]);
|
||||
int16x4_t int_16x4_in = vld1_s16(&inptr[n]);
|
||||
int32x4_t int_32x4 = vmull_s16(int_16x4_x, int_16x4_in);
|
||||
int_32x4 = vshlq_s32(int_32x4, int_32x4_scale);
|
||||
int_32x4_sum = vaddq_s32(int_32x4_sum, int_32x4);
|
||||
}
|
||||
|
||||
// Use vector store to avoid long stall from data trasferring
|
||||
// from vector to general register.
|
||||
vst1q_s32(vbuff, int_32x4_sum);
|
||||
csum32 = vbuff[0] + vbuff[1];
|
||||
csum32 += vbuff[2];
|
||||
csum32 += vbuff[3];
|
||||
}
|
||||
#else
|
||||
csum32 = 0;
|
||||
if(scaling == 0) {
|
||||
for (n = 0; n < PITCH_CORR_LEN2; n++) {
|
||||
csum32 += x[n] * inptr[n];
|
||||
}
|
||||
} else {
|
||||
for (n = 0; n < PITCH_CORR_LEN2; n++) {
|
||||
csum32 += (x[n] * inptr[n]) >> scaling;
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
logcorQ8--;
|
||||
|
||||
lys = WebRtcIsacfix_Log2Q8((uint32_t)ysum32); // Q8
|
||||
lys = WEBRTC_SPL_RSHIFT_W32(lys, 1); //sqrt(ysum);
|
||||
|
||||
if (csum32 > 0) {
|
||||
lcs = WebRtcIsacfix_Log2Q8((uint32_t)csum32); // 2log(csum) in Q8
|
||||
if (lcs > (lys + oneQ8)) { // csum/sqrt(ysum) > 2
|
||||
*logcorQ8 = lcs - lys; // log2(csum/sqrt(ysum))
|
||||
} else {
|
||||
*logcorQ8 = oneQ8; // 1.00
|
||||
}
|
||||
} else {
|
||||
*logcorQ8 = 0;
|
||||
}
|
||||
}
|
||||
}
|
||||
@ -0,0 +1,196 @@
|
||||
/*
|
||||
* 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.
|
||||
*/
|
||||
|
||||
#include "webrtc/modules/audio_coding/codecs/isac/fix/source/pitch_estimator.h"
|
||||
#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
|
||||
#include "webrtc/system_wrappers/interface/compile_assert_c.h"
|
||||
|
||||
extern int32_t WebRtcIsacfix_Log2Q8(uint32_t x);
|
||||
|
||||
void WebRtcIsacfix_PCorr2Q32(const int16_t* in, int32_t* logcorQ8) {
|
||||
int16_t scaling,n,k;
|
||||
int32_t ysum32,csum32, lys, lcs;
|
||||
int32_t oneQ8;
|
||||
const int16_t* x;
|
||||
const int16_t* inptr;
|
||||
|
||||
oneQ8 = WEBRTC_SPL_LSHIFT_W32((int32_t)1, 8); // 1.00 in Q8
|
||||
x = in + PITCH_MAX_LAG / 2 + 2;
|
||||
scaling = WebRtcSpl_GetScalingSquare((int16_t*)in,
|
||||
PITCH_CORR_LEN2,
|
||||
PITCH_CORR_LEN2);
|
||||
ysum32 = 1;
|
||||
csum32 = 0;
|
||||
x = in + PITCH_MAX_LAG / 2 + 2;
|
||||
{
|
||||
const int16_t* tmp_x = x;
|
||||
const int16_t* tmp_in = in;
|
||||
int32_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
|
||||
n = PITCH_CORR_LEN2;
|
||||
COMPILE_ASSERT(PITCH_CORR_LEN2 % 4 == 0);
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"1: \n\t"
|
||||
"lh %[tmp1], 0(%[tmp_in]) \n\t"
|
||||
"lh %[tmp2], 2(%[tmp_in]) \n\t"
|
||||
"lh %[tmp3], 4(%[tmp_in]) \n\t"
|
||||
"lh %[tmp4], 6(%[tmp_in]) \n\t"
|
||||
"lh %[tmp5], 0(%[tmp_x]) \n\t"
|
||||
"lh %[tmp6], 2(%[tmp_x]) \n\t"
|
||||
"lh %[tmp7], 4(%[tmp_x]) \n\t"
|
||||
"lh %[tmp8], 6(%[tmp_x]) \n\t"
|
||||
"mul %[tmp5], %[tmp1], %[tmp5] \n\t"
|
||||
"mul %[tmp1], %[tmp1], %[tmp1] \n\t"
|
||||
"mul %[tmp6], %[tmp2], %[tmp6] \n\t"
|
||||
"mul %[tmp2], %[tmp2], %[tmp2] \n\t"
|
||||
"mul %[tmp7], %[tmp3], %[tmp7] \n\t"
|
||||
"mul %[tmp3], %[tmp3], %[tmp3] \n\t"
|
||||
"mul %[tmp8], %[tmp4], %[tmp8] \n\t"
|
||||
"mul %[tmp4], %[tmp4], %[tmp4] \n\t"
|
||||
"addiu %[n], %[n], -4 \n\t"
|
||||
"srav %[tmp5], %[tmp5], %[scaling] \n\t"
|
||||
"srav %[tmp1], %[tmp1], %[scaling] \n\t"
|
||||
"srav %[tmp6], %[tmp6], %[scaling] \n\t"
|
||||
"srav %[tmp2], %[tmp2], %[scaling] \n\t"
|
||||
"srav %[tmp7], %[tmp7], %[scaling] \n\t"
|
||||
"srav %[tmp3], %[tmp3], %[scaling] \n\t"
|
||||
"srav %[tmp8], %[tmp8], %[scaling] \n\t"
|
||||
"srav %[tmp4], %[tmp4], %[scaling] \n\t"
|
||||
"addu %[ysum32], %[ysum32], %[tmp1] \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp5] \n\t"
|
||||
"addu %[ysum32], %[ysum32], %[tmp2] \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp6] \n\t"
|
||||
"addu %[ysum32], %[ysum32], %[tmp3] \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp7] \n\t"
|
||||
"addu %[ysum32], %[ysum32], %[tmp4] \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp8] \n\t"
|
||||
"addiu %[tmp_in], %[tmp_in], 8 \n\t"
|
||||
"bgtz %[n], 1b \n\t"
|
||||
" addiu %[tmp_x], %[tmp_x], 8 \n\t"
|
||||
".set pop \n\t"
|
||||
: [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), [tmp3] "=&r" (tmp3),
|
||||
[tmp4] "=&r" (tmp4), [tmp5] "=&r" (tmp5), [tmp6] "=&r" (tmp6),
|
||||
[tmp7] "=&r" (tmp7), [tmp8] "=&r" (tmp8), [tmp_in] "+r" (tmp_in),
|
||||
[ysum32] "+r" (ysum32), [tmp_x] "+r" (tmp_x), [csum32] "+r" (csum32),
|
||||
[n] "+r" (n)
|
||||
: [scaling] "r" (scaling)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
}
|
||||
logcorQ8 += PITCH_LAG_SPAN2 - 1;
|
||||
lys = WebRtcIsacfix_Log2Q8((uint32_t)ysum32); // Q8
|
||||
lys = WEBRTC_SPL_RSHIFT_W32(lys, 1); //sqrt(ysum);
|
||||
if (csum32 > 0) {
|
||||
lcs = WebRtcIsacfix_Log2Q8((uint32_t)csum32); // 2log(csum) in Q8
|
||||
if (lcs > (lys + oneQ8)) { // csum/sqrt(ysum) > 2 in Q8
|
||||
*logcorQ8 = lcs - lys; // log2(csum/sqrt(ysum))
|
||||
} else {
|
||||
*logcorQ8 = oneQ8; // 1.00
|
||||
}
|
||||
} else {
|
||||
*logcorQ8 = 0;
|
||||
}
|
||||
|
||||
for (k = 1; k < PITCH_LAG_SPAN2; k++) {
|
||||
inptr = &in[k];
|
||||
const int16_t* tmp_in1 = &in[k - 1];
|
||||
const int16_t* tmp_in2 = &in[PITCH_CORR_LEN2 + k - 1];
|
||||
const int16_t* tmp_x = x;
|
||||
int32_t tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
|
||||
n = PITCH_CORR_LEN2;
|
||||
csum32 = 0;
|
||||
__asm __volatile (
|
||||
".set push \n\t"
|
||||
".set noreorder \n\t"
|
||||
"lh %[tmp1], 0(%[tmp_in1]) \n\t"
|
||||
"lh %[tmp2], 0(%[tmp_in2]) \n\t"
|
||||
"mul %[tmp1], %[tmp1], %[tmp1] \n\t"
|
||||
"mul %[tmp2], %[tmp2], %[tmp2] \n\t"
|
||||
"srav %[tmp1], %[tmp1], %[scaling] \n\t"
|
||||
"srav %[tmp2], %[tmp2], %[scaling] \n\t"
|
||||
"subu %[ysum32], %[ysum32], %[tmp1] \n\t"
|
||||
"bnez %[scaling], 2f \n\t"
|
||||
" addu %[ysum32], %[ysum32], %[tmp2] \n\t"
|
||||
"1: \n\t"
|
||||
"lh %[tmp1], 0(%[inptr]) \n\t"
|
||||
"lh %[tmp2], 0(%[tmp_x]) \n\t"
|
||||
"lh %[tmp3], 2(%[inptr]) \n\t"
|
||||
"lh %[tmp4], 2(%[tmp_x]) \n\t"
|
||||
"lh %[tmp5], 4(%[inptr]) \n\t"
|
||||
"lh %[tmp6], 4(%[tmp_x]) \n\t"
|
||||
"lh %[tmp7], 6(%[inptr]) \n\t"
|
||||
"lh %[tmp8], 6(%[tmp_x]) \n\t"
|
||||
"mul %[tmp1], %[tmp1], %[tmp2] \n\t"
|
||||
"mul %[tmp2], %[tmp3], %[tmp4] \n\t"
|
||||
"mul %[tmp3], %[tmp5], %[tmp6] \n\t"
|
||||
"mul %[tmp4], %[tmp7], %[tmp8] \n\t"
|
||||
"addiu %[n], %[n], -4 \n\t"
|
||||
"addiu %[inptr], %[inptr], 8 \n\t"
|
||||
"addiu %[tmp_x], %[tmp_x], 8 \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp1] \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp2] \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp3] \n\t"
|
||||
"bgtz %[n], 1b \n\t"
|
||||
" addu %[csum32], %[csum32], %[tmp4] \n\t"
|
||||
"b 3f \n\t"
|
||||
" nop \n\t"
|
||||
"2: \n\t"
|
||||
"lh %[tmp1], 0(%[inptr]) \n\t"
|
||||
"lh %[tmp2], 0(%[tmp_x]) \n\t"
|
||||
"lh %[tmp3], 2(%[inptr]) \n\t"
|
||||
"lh %[tmp4], 2(%[tmp_x]) \n\t"
|
||||
"lh %[tmp5], 4(%[inptr]) \n\t"
|
||||
"lh %[tmp6], 4(%[tmp_x]) \n\t"
|
||||
"lh %[tmp7], 6(%[inptr]) \n\t"
|
||||
"lh %[tmp8], 6(%[tmp_x]) \n\t"
|
||||
"mul %[tmp1], %[tmp1], %[tmp2] \n\t"
|
||||
"mul %[tmp2], %[tmp3], %[tmp4] \n\t"
|
||||
"mul %[tmp3], %[tmp5], %[tmp6] \n\t"
|
||||
"mul %[tmp4], %[tmp7], %[tmp8] \n\t"
|
||||
"addiu %[n], %[n], -4 \n\t"
|
||||
"addiu %[inptr], %[inptr], 8 \n\t"
|
||||
"addiu %[tmp_x], %[tmp_x], 8 \n\t"
|
||||
"srav %[tmp1], %[tmp1], %[scaling] \n\t"
|
||||
"srav %[tmp2], %[tmp2], %[scaling] \n\t"
|
||||
"srav %[tmp3], %[tmp3], %[scaling] \n\t"
|
||||
"srav %[tmp4], %[tmp4], %[scaling] \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp1] \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp2] \n\t"
|
||||
"addu %[csum32], %[csum32], %[tmp3] \n\t"
|
||||
"bgtz %[n], 2b \n\t"
|
||||
" addu %[csum32], %[csum32], %[tmp4] \n\t"
|
||||
"3: \n\t"
|
||||
".set pop \n\t"
|
||||
: [tmp1] "=&r" (tmp1), [tmp2] "=&r" (tmp2), [tmp3] "=&r" (tmp3),
|
||||
[tmp4] "=&r" (tmp4), [tmp5] "=&r" (tmp5), [tmp6] "=&r" (tmp6),
|
||||
[tmp7] "=&r" (tmp7), [tmp8] "=&r" (tmp8), [inptr] "+r" (inptr),
|
||||
[csum32] "+r" (csum32), [tmp_x] "+r" (tmp_x), [ysum32] "+r" (ysum32),
|
||||
[n] "+r" (n)
|
||||
: [tmp_in1] "r" (tmp_in1), [tmp_in2] "r" (tmp_in2),
|
||||
[scaling] "r" (scaling)
|
||||
: "memory", "hi", "lo"
|
||||
);
|
||||
|
||||
logcorQ8--;
|
||||
lys = WebRtcIsacfix_Log2Q8((uint32_t)ysum32); // Q8
|
||||
lys = WEBRTC_SPL_RSHIFT_W32(lys, 1); //sqrt(ysum);
|
||||
if (csum32 > 0) {
|
||||
lcs = WebRtcIsacfix_Log2Q8((uint32_t)csum32); // 2log(csum) in Q8
|
||||
if (lcs > (lys + oneQ8)) { // csum/sqrt(ysum) > 2
|
||||
*logcorQ8 = lcs - lys; // log2(csum/sqrt(ysum))
|
||||
} else {
|
||||
*logcorQ8 = oneQ8; // 1.00
|
||||
}
|
||||
} else {
|
||||
*logcorQ8 = 0;
|
||||
}
|
||||
}
|
||||
}
|
||||
@ -19,89 +19,13 @@
|
||||
#include "webrtc/modules/audio_coding/codecs/isac/fix/source/fft.h"
|
||||
#include "webrtc/modules/audio_coding/codecs/isac/fix/source/settings.h"
|
||||
|
||||
#if (defined WEBRTC_DETECT_ARM_NEON || defined WEBRTC_ARCH_ARM_NEON)
|
||||
/* Tables are defined in ARM assembly files. */
|
||||
/* Tables are defined in transform_tables.c file or ARM assembly files. */
|
||||
/* Cosine table 1 in Q14 */
|
||||
extern const int16_t WebRtcIsacfix_kCosTab1[FRAMESAMPLES/2];
|
||||
/* Sine table 1 in Q14 */
|
||||
extern const int16_t WebRtcIsacfix_kSinTab1[FRAMESAMPLES/2];
|
||||
/* Sine table 2 in Q14 */
|
||||
extern const int16_t WebRtcIsacfix_kSinTab2[FRAMESAMPLES/4];
|
||||
#else
|
||||
/* Cosine table 1 in Q14 */
|
||||
static const int16_t WebRtcIsacfix_kCosTab1[FRAMESAMPLES/2] = {
|
||||
16384, 16383, 16378, 16371, 16362, 16349, 16333, 16315, 16294, 16270,
|
||||
16244, 16214, 16182, 16147, 16110, 16069, 16026, 15980, 15931, 15880,
|
||||
15826, 15769, 15709, 15647, 15582, 15515, 15444, 15371, 15296, 15218,
|
||||
15137, 15053, 14968, 14879, 14788, 14694, 14598, 14500, 14399, 14295,
|
||||
14189, 14081, 13970, 13856, 13741, 13623, 13502, 13380, 13255, 13128,
|
||||
12998, 12867, 12733, 12597, 12458, 12318, 12176, 12031, 11885, 11736,
|
||||
11585, 11433, 11278, 11121, 10963, 10803, 10641, 10477, 10311, 10143,
|
||||
9974, 9803, 9630, 9456, 9280, 9102, 8923, 8743, 8561, 8377,
|
||||
8192, 8006, 7818, 7629, 7438, 7246, 7053, 6859, 6664, 6467,
|
||||
6270, 6071, 5872, 5671, 5469, 5266, 5063, 4859, 4653, 4447,
|
||||
4240, 4033, 3825, 3616, 3406, 3196, 2986, 2775, 2563, 2351,
|
||||
2139, 1926, 1713, 1499, 1285, 1072, 857, 643, 429, 214,
|
||||
0, -214, -429, -643, -857, -1072, -1285, -1499, -1713, -1926,
|
||||
-2139, -2351, -2563, -2775, -2986, -3196, -3406, -3616, -3825, -4033,
|
||||
-4240, -4447, -4653, -4859, -5063, -5266, -5469, -5671, -5872, -6071,
|
||||
-6270, -6467, -6664, -6859, -7053, -7246, -7438, -7629, -7818, -8006,
|
||||
-8192, -8377, -8561, -8743, -8923, -9102, -9280, -9456, -9630, -9803,
|
||||
-9974, -10143, -10311, -10477, -10641, -10803, -10963, -11121, -11278, -11433,
|
||||
-11585, -11736, -11885, -12031, -12176, -12318, -12458, -12597, -12733,
|
||||
-12867, -12998, -13128, -13255, -13380, -13502, -13623, -13741, -13856,
|
||||
-13970, -14081, -14189, -14295, -14399, -14500, -14598, -14694, -14788,
|
||||
-14879, -14968, -15053, -15137, -15218, -15296, -15371, -15444, -15515,
|
||||
-15582, -15647, -15709, -15769, -15826, -15880, -15931, -15980, -16026,
|
||||
-16069, -16110, -16147, -16182, -16214, -16244, -16270, -16294, -16315,
|
||||
-16333, -16349, -16362, -16371, -16378, -16383
|
||||
};
|
||||
|
||||
/* Sine table 1 in Q14 */
|
||||
static const int16_t WebRtcIsacfix_kSinTab1[FRAMESAMPLES/2] = {
|
||||
0, 214, 429, 643, 857, 1072, 1285, 1499, 1713, 1926,
|
||||
2139, 2351, 2563, 2775, 2986, 3196, 3406, 3616, 3825, 4033,
|
||||
4240, 4447, 4653, 4859, 5063, 5266, 5469, 5671, 5872, 6071,
|
||||
6270, 6467, 6664, 6859, 7053, 7246, 7438, 7629, 7818, 8006,
|
||||
8192, 8377, 8561, 8743, 8923, 9102, 9280, 9456, 9630, 9803,
|
||||
9974, 10143, 10311, 10477, 10641, 10803, 10963, 11121, 11278, 11433,
|
||||
11585, 11736, 11885, 12031, 12176, 12318, 12458, 12597, 12733, 12867,
|
||||
12998, 13128, 13255, 13380, 13502, 13623, 13741, 13856, 13970, 14081,
|
||||
14189, 14295, 14399, 14500, 14598, 14694, 14788, 14879, 14968, 15053,
|
||||
15137, 15218, 15296, 15371, 15444, 15515, 15582, 15647, 15709, 15769,
|
||||
15826, 15880, 15931, 15980, 16026, 16069, 16110, 16147, 16182, 16214,
|
||||
16244, 16270, 16294, 16315, 16333, 16349, 16362, 16371, 16378, 16383,
|
||||
16384, 16383, 16378, 16371, 16362, 16349, 16333, 16315, 16294, 16270,
|
||||
16244, 16214, 16182, 16147, 16110, 16069, 16026, 15980, 15931, 15880,
|
||||
15826, 15769, 15709, 15647, 15582, 15515, 15444, 15371, 15296, 15218,
|
||||
15137, 15053, 14968, 14879, 14788, 14694, 14598, 14500, 14399, 14295,
|
||||
14189, 14081, 13970, 13856, 13741, 13623, 13502, 13380, 13255, 13128,
|
||||
12998, 12867, 12733, 12597, 12458, 12318, 12176, 12031, 11885, 11736,
|
||||
11585, 11433, 11278, 11121, 10963, 10803, 10641, 10477, 10311, 10143,
|
||||
9974, 9803, 9630, 9456, 9280, 9102, 8923, 8743, 8561, 8377,
|
||||
8192, 8006, 7818, 7629, 7438, 7246, 7053, 6859, 6664, 6467,
|
||||
6270, 6071, 5872, 5671, 5469, 5266, 5063, 4859, 4653, 4447,
|
||||
4240, 4033, 3825, 3616, 3406, 3196, 2986, 2775, 2563, 2351,
|
||||
2139, 1926, 1713, 1499, 1285, 1072, 857, 643, 429, 214
|
||||
};
|
||||
|
||||
|
||||
/* Sine table 2 in Q14 */
|
||||
static const int16_t WebRtcIsacfix_kSinTab2[FRAMESAMPLES/4] = {
|
||||
16384, -16381, 16375, -16367, 16356, -16342, 16325, -16305, 16283, -16257,
|
||||
16229, -16199, 16165, -16129, 16090, -16048, 16003, -15956, 15906, -15853,
|
||||
15798, -15739, 15679, -15615, 15549, -15480, 15408, -15334, 15257, -15178,
|
||||
15095, -15011, 14924, -14834, 14741, -14647, 14549, -14449, 14347, -14242,
|
||||
14135, -14025, 13913, -13799, 13682, -13563, 13441, -13318, 13192, -13063,
|
||||
12933, -12800, 12665, -12528, 12389, -12247, 12104, -11958, 11810, -11661,
|
||||
11509, -11356, 11200, -11042, 10883, -10722, 10559, -10394, 10227, -10059,
|
||||
9889, -9717, 9543, -9368, 9191, -9013, 8833, -8652, 8469, -8285,
|
||||
8099, -7912, 7723, -7534, 7342, -7150, 6957, -6762, 6566, -6369,
|
||||
6171, -5971, 5771, -5570, 5368, -5165, 4961, -4756, 4550, -4344,
|
||||
4137, -3929, 3720, -3511, 3301, -3091, 2880, -2669, 2457, -2245,
|
||||
2032, -1819, 1606, -1392, 1179, -965, 750, -536, 322, -107
|
||||
};
|
||||
#endif // WEBRTC_DETECT_ARM_NEON || WEBRTC_ARCH_ARM_NEON
|
||||
|
||||
void WebRtcIsacfix_Time2SpecC(int16_t *inre1Q9,
|
||||
int16_t *inre2Q9,
|
||||
|
||||
1287
webrtc/modules/audio_coding/codecs/isac/fix/source/transform_mips.c
Normal file
1287
webrtc/modules/audio_coding/codecs/isac/fix/source/transform_mips.c
Normal file
File diff suppressed because it is too large
Load Diff
@ -0,0 +1,111 @@
|
||||
/*
|
||||
* Copyright (c) 2014 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.
|
||||
*/
|
||||
|
||||
/*
|
||||
* This file contains trigonometric functions look-up tables used in
|
||||
* transform functions WebRtcIsacfix_Time2Spec and WebRtcIsacfix_Spec2Time.
|
||||
*/
|
||||
|
||||
#include "webrtc/modules/audio_coding/codecs/isac/fix/source/settings.h"
|
||||
#include "webrtc/typedefs.h"
|
||||
|
||||
#if !(defined WEBRTC_DETECT_ARM_NEON || defined WEBRTC_ARCH_ARM_NEON)
|
||||
/* Cosine table 1 in Q14. */
|
||||
const int16_t WebRtcIsacfix_kCosTab1[FRAMESAMPLES/2] = {
|
||||
16384, 16383, 16378, 16371, 16362, 16349, 16333, 16315, 16294, 16270,
|
||||
16244, 16214, 16182, 16147, 16110, 16069, 16026, 15980, 15931, 15880,
|
||||
15826, 15769, 15709, 15647, 15582, 15515, 15444, 15371, 15296, 15218,
|
||||
15137, 15053, 14968, 14879, 14788, 14694, 14598, 14500, 14399, 14295,
|
||||
14189, 14081, 13970, 13856, 13741, 13623, 13502, 13380, 13255, 13128,
|
||||
12998, 12867, 12733, 12597, 12458, 12318, 12176, 12031, 11885, 11736,
|
||||
11585, 11433, 11278, 11121, 10963, 10803, 10641, 10477, 10311, 10143,
|
||||
9974, 9803, 9630, 9456, 9280, 9102, 8923, 8743, 8561, 8377,
|
||||
8192, 8006, 7818, 7629, 7438, 7246, 7053, 6859, 6664, 6467,
|
||||
6270, 6071, 5872, 5671, 5469, 5266, 5063, 4859, 4653, 4447,
|
||||
4240, 4033, 3825, 3616, 3406, 3196, 2986, 2775, 2563, 2351,
|
||||
2139, 1926, 1713, 1499, 1285, 1072, 857, 643, 429, 214,
|
||||
0, -214, -429, -643, -857, -1072, -1285, -1499, -1713, -1926,
|
||||
-2139, -2351, -2563, -2775, -2986, -3196, -3406, -3616, -3825, -4033,
|
||||
-4240, -4447, -4653, -4859, -5063, -5266, -5469, -5671, -5872, -6071,
|
||||
-6270, -6467, -6664, -6859, -7053, -7246, -7438, -7629, -7818, -8006,
|
||||
-8192, -8377, -8561, -8743, -8923, -9102, -9280, -9456, -9630, -9803,
|
||||
-9974, -10143, -10311, -10477, -10641, -10803, -10963, -11121, -11278, -11433,
|
||||
-11585, -11736, -11885, -12031, -12176, -12318, -12458, -12597, -12733,
|
||||
-12867, -12998, -13128, -13255, -13380, -13502, -13623, -13741, -13856,
|
||||
-13970, -14081, -14189, -14295, -14399, -14500, -14598, -14694, -14788,
|
||||
-14879, -14968, -15053, -15137, -15218, -15296, -15371, -15444, -15515,
|
||||
-15582, -15647, -15709, -15769, -15826, -15880, -15931, -15980, -16026,
|
||||
-16069, -16110, -16147, -16182, -16214, -16244, -16270, -16294, -16315,
|
||||
-16333, -16349, -16362, -16371, -16378, -16383
|
||||
};
|
||||
|
||||
/* Sine table 1 in Q14. */
|
||||
const int16_t WebRtcIsacfix_kSinTab1[FRAMESAMPLES/2] = {
|
||||
0, 214, 429, 643, 857, 1072, 1285, 1499, 1713, 1926,
|
||||
2139, 2351, 2563, 2775, 2986, 3196, 3406, 3616, 3825, 4033,
|
||||
4240, 4447, 4653, 4859, 5063, 5266, 5469, 5671, 5872, 6071,
|
||||
6270, 6467, 6664, 6859, 7053, 7246, 7438, 7629, 7818, 8006,
|
||||
8192, 8377, 8561, 8743, 8923, 9102, 9280, 9456, 9630, 9803,
|
||||
9974, 10143, 10311, 10477, 10641, 10803, 10963, 11121, 11278, 11433,
|
||||
11585, 11736, 11885, 12031, 12176, 12318, 12458, 12597, 12733, 12867,
|
||||
12998, 13128, 13255, 13380, 13502, 13623, 13741, 13856, 13970, 14081,
|
||||
14189, 14295, 14399, 14500, 14598, 14694, 14788, 14879, 14968, 15053,
|
||||
15137, 15218, 15296, 15371, 15444, 15515, 15582, 15647, 15709, 15769,
|
||||
15826, 15880, 15931, 15980, 16026, 16069, 16110, 16147, 16182, 16214,
|
||||
16244, 16270, 16294, 16315, 16333, 16349, 16362, 16371, 16378, 16383,
|
||||
16384, 16383, 16378, 16371, 16362, 16349, 16333, 16315, 16294, 16270,
|
||||
16244, 16214, 16182, 16147, 16110, 16069, 16026, 15980, 15931, 15880,
|
||||
15826, 15769, 15709, 15647, 15582, 15515, 15444, 15371, 15296, 15218,
|
||||
15137, 15053, 14968, 14879, 14788, 14694, 14598, 14500, 14399, 14295,
|
||||
14189, 14081, 13970, 13856, 13741, 13623, 13502, 13380, 13255, 13128,
|
||||
12998, 12867, 12733, 12597, 12458, 12318, 12176, 12031, 11885, 11736,
|
||||
11585, 11433, 11278, 11121, 10963, 10803, 10641, 10477, 10311, 10143,
|
||||
9974, 9803, 9630, 9456, 9280, 9102, 8923, 8743, 8561, 8377,
|
||||
8192, 8006, 7818, 7629, 7438, 7246, 7053, 6859, 6664, 6467,
|
||||
6270, 6071, 5872, 5671, 5469, 5266, 5063, 4859, 4653, 4447,
|
||||
4240, 4033, 3825, 3616, 3406, 3196, 2986, 2775, 2563, 2351,
|
||||
2139, 1926, 1713, 1499, 1285, 1072, 857, 643, 429, 214
|
||||
};
|
||||
|
||||
|
||||
/* Sine table 2 in Q14. */
|
||||
const int16_t WebRtcIsacfix_kSinTab2[FRAMESAMPLES/4] = {
|
||||
16384, -16381, 16375, -16367, 16356, -16342, 16325, -16305, 16283, -16257,
|
||||
16229, -16199, 16165, -16129, 16090, -16048, 16003, -15956, 15906, -15853,
|
||||
15798, -15739, 15679, -15615, 15549, -15480, 15408, -15334, 15257, -15178,
|
||||
15095, -15011, 14924, -14834, 14741, -14647, 14549, -14449, 14347, -14242,
|
||||
14135, -14025, 13913, -13799, 13682, -13563, 13441, -13318, 13192, -13063,
|
||||
12933, -12800, 12665, -12528, 12389, -12247, 12104, -11958, 11810, -11661,
|
||||
11509, -11356, 11200, -11042, 10883, -10722, 10559, -10394, 10227, -10059,
|
||||
9889, -9717, 9543, -9368, 9191, -9013, 8833, -8652, 8469, -8285,
|
||||
8099, -7912, 7723, -7534, 7342, -7150, 6957, -6762, 6566, -6369,
|
||||
6171, -5971, 5771, -5570, 5368, -5165, 4961, -4756, 4550, -4344,
|
||||
4137, -3929, 3720, -3511, 3301, -3091, 2880, -2669, 2457, -2245,
|
||||
2032, -1819, 1606, -1392, 1179, -965, 750, -536, 322, -107
|
||||
};
|
||||
#endif
|
||||
|
||||
#if defined(MIPS32_LE)
|
||||
/* Cosine table 2 in Q14. Used only on MIPS platforms. */
|
||||
const int16_t WebRtcIsacfix_kCosTab2[FRAMESAMPLES/4] = {
|
||||
107, -322, 536, -750, 965, -1179, 1392, -1606, 1819, -2032,
|
||||
2245, -2457, 2669, -2880, 3091, -3301, 3511, -3720, 3929, -4137,
|
||||
4344, -4550, 4756, -4961, 5165, -5368, 5570, -5771, 5971, -6171,
|
||||
6369, -6566, 6762, -6957, 7150, -7342, 7534, -7723, 7912, -8099,
|
||||
8285, -8469, 8652, -8833, 9013, -9191, 9368, -9543, 9717, -9889,
|
||||
10059, -10227, 10394, -10559, 10722, -10883, 11042, -11200, 11356, -11509,
|
||||
11661, -11810, 11958, -12104, 12247, -12389, 12528, -12665, 12800, -12933,
|
||||
13063, -13192, 13318, -13441, 13563, -13682, 13799, -13913, 14025, -14135,
|
||||
14242, -14347, 14449, -14549, 14647, -14741, 14834, -14924, 15011, -15095,
|
||||
15178, -15257, 15334, -15408, 15480, -15549, 15615, -15679, 15739, -15798,
|
||||
15853, -15906, 15956, -16003, 16048, -16090, 16129, -16165, 16199, -16229,
|
||||
16257, -16283, 16305, -16325, 16342, -16356, 16367, -16375, 16381, -16384
|
||||
};
|
||||
#endif
|
||||
Loading…
x
Reference in New Issue
Block a user