Avoiding the term "mode" in opus_demo
[opus.git] / silk / x86 / main_sse.h
1 /* Copyright (c) 2014, Cisco Systems, INC
2    Written by XiangMingZhu WeiZhou MinPeng YanWang
3
4    Redistribution and use in source and binary forms, with or without
5    modification, are permitted provided that the following conditions
6    are met:
7
8    - Redistributions of source code must retain the above copyright
9    notice, this list of conditions and the following disclaimer.
10
11    - Redistributions in binary form must reproduce the above copyright
12    notice, this list of conditions and the following disclaimer in the
13    documentation and/or other materials provided with the distribution.
14
15    THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16    ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17    LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
18    A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
19    OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
20    EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
21    PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
22    PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
23    LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
24    NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
25    SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26 */
27
28 #ifndef MAIN_SSE_H
29 #define MAIN_SSE_H
30
31 #ifdef HAVE_CONFIG_H
32 #include "config.h"
33 #endif
34
35 # if defined(OPUS_X86_MAY_HAVE_SSE4_1)
36
37 #  define OVERRIDE_silk_VQ_WMat_EC
38
39 void silk_VQ_WMat_EC_sse4_1(
40     opus_int8                   *ind,                           /* O    index of best codebook vector               */
41     opus_int32                  *rate_dist_Q14,                 /* O    best weighted quant error + mu * rate       */
42     opus_int                    *gain_Q7,                       /* O    sum of absolute LTP coefficients            */
43     const opus_int16            *in_Q14,                        /* I    input vector to be quantized                */
44     const opus_int32            *W_Q18,                         /* I    weighting matrix                            */
45     const opus_int8             *cb_Q7,                         /* I    codebook                                    */
46     const opus_uint8            *cb_gain_Q7,                    /* I    codebook effective gain                     */
47     const opus_uint8            *cl_Q5,                         /* I    code length for each codebook vector        */
48     const opus_int              mu_Q9,                          /* I    tradeoff betw. weighted error and rate      */
49     const opus_int32            max_gain_Q7,                    /* I    maximum sum of absolute LTP coefficients    */
50     opus_int                    L                               /* I    number of vectors in codebook               */
51 );
52
53 extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])(
54     opus_int8                   *ind,                           /* O    index of best codebook vector               */
55     opus_int32                  *rate_dist_Q14,                 /* O    best weighted quant error + mu * rate       */
56     opus_int                    *gain_Q7,                       /* O    sum of absolute LTP coefficients            */
57     const opus_int16            *in_Q14,                        /* I    input vector to be quantized                */
58     const opus_int32            *W_Q18,                         /* I    weighting matrix                            */
59     const opus_int8             *cb_Q7,                         /* I    codebook                                    */
60     const opus_uint8            *cb_gain_Q7,                    /* I    codebook effective gain                     */
61     const opus_uint8            *cl_Q5,                         /* I    code length for each codebook vector        */
62     const opus_int              mu_Q9,                          /* I    tradeoff betw. weighted error and rate      */
63     const opus_int32            max_gain_Q7,                    /* I    maximum sum of absolute LTP coefficients    */
64     opus_int                    L                               /* I    number of vectors in codebook               */
65 );
66
67 #  define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \
68                           mu_Q9, max_gain_Q7, L, arch) \
69     ((*SILK_VQ_WMAT_EC_IMPL[(arch) & OPUS_ARCHMASK])(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \
70                           mu_Q9, max_gain_Q7, L))
71
72 #  define OVERRIDE_silk_NSQ
73
74 void silk_NSQ_sse4_1(
75     const silk_encoder_state    *psEncC,                                    /* I/O  Encoder State                   */
76     silk_nsq_state              *NSQ,                                       /* I/O  NSQ state                       */
77     SideInfoIndices             *psIndices,                                 /* I/O  Quantization Indices            */
78     const opus_int32            x_Q3[],                                     /* I    Prefiltered input signal        */
79     opus_int8                   pulses[],                                   /* O    Quantized pulse signal          */
80     const opus_int16            PredCoef_Q12[ 2 * MAX_LPC_ORDER ],          /* I    Short term prediction coefs     */
81     const opus_int16            LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ],    /* I    Long term prediction coefs      */
82     const opus_int16            AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs             */
83     const opus_int              HarmShapeGain_Q14[ MAX_NB_SUBFR ],          /* I    Long term shaping coefs         */
84     const opus_int              Tilt_Q14[ MAX_NB_SUBFR ],                   /* I    Spectral tilt                   */
85     const opus_int32            LF_shp_Q14[ MAX_NB_SUBFR ],                 /* I    Low frequency shaping coefs     */
86     const opus_int32            Gains_Q16[ MAX_NB_SUBFR ],                  /* I    Quantization step sizes         */
87     const opus_int              pitchL[ MAX_NB_SUBFR ],                     /* I    Pitch lags                      */
88     const opus_int              Lambda_Q10,                                 /* I    Rate/distortion tradeoff        */
89     const opus_int              LTP_scale_Q14                               /* I    LTP state scaling               */
90 );
91
92 extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])(
93     const silk_encoder_state    *psEncC,                                    /* I/O  Encoder State                   */
94     silk_nsq_state              *NSQ,                                       /* I/O  NSQ state                       */
95     SideInfoIndices             *psIndices,                                 /* I/O  Quantization Indices            */
96     const opus_int32            x_Q3[],                                     /* I    Prefiltered input signal        */
97     opus_int8                   pulses[],                                   /* O    Quantized pulse signal          */
98     const opus_int16            PredCoef_Q12[ 2 * MAX_LPC_ORDER ],          /* I    Short term prediction coefs     */
99     const opus_int16            LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ],    /* I    Long term prediction coefs      */
100     const opus_int16            AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs             */
101     const opus_int              HarmShapeGain_Q14[ MAX_NB_SUBFR ],          /* I    Long term shaping coefs         */
102     const opus_int              Tilt_Q14[ MAX_NB_SUBFR ],                   /* I    Spectral tilt                   */
103     const opus_int32            LF_shp_Q14[ MAX_NB_SUBFR ],                 /* I    Low frequency shaping coefs     */
104     const opus_int32            Gains_Q16[ MAX_NB_SUBFR ],                  /* I    Quantization step sizes         */
105     const opus_int              pitchL[ MAX_NB_SUBFR ],                     /* I    Pitch lags                      */
106     const opus_int              Lambda_Q10,                                 /* I    Rate/distortion tradeoff        */
107     const opus_int              LTP_scale_Q14                               /* I    LTP state scaling               */
108 );
109
110 #  define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \
111                    HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \
112     ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \
113                    HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14))
114
115 #  define OVERRIDE_silk_NSQ_del_dec
116
117 void silk_NSQ_del_dec_sse4_1(
118     const silk_encoder_state    *psEncC,                                    /* I/O  Encoder State                   */
119     silk_nsq_state              *NSQ,                                       /* I/O  NSQ state                       */
120     SideInfoIndices             *psIndices,                                 /* I/O  Quantization Indices            */
121     const opus_int32            x_Q3[],                                     /* I    Prefiltered input signal        */
122     opus_int8                   pulses[],                                   /* O    Quantized pulse signal          */
123     const opus_int16            PredCoef_Q12[ 2 * MAX_LPC_ORDER ],          /* I    Short term prediction coefs     */
124     const opus_int16            LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ],    /* I    Long term prediction coefs      */
125     const opus_int16            AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs             */
126     const opus_int              HarmShapeGain_Q14[ MAX_NB_SUBFR ],          /* I    Long term shaping coefs         */
127     const opus_int              Tilt_Q14[ MAX_NB_SUBFR ],                   /* I    Spectral tilt                   */
128     const opus_int32            LF_shp_Q14[ MAX_NB_SUBFR ],                 /* I    Low frequency shaping coefs     */
129     const opus_int32            Gains_Q16[ MAX_NB_SUBFR ],                  /* I    Quantization step sizes         */
130     const opus_int              pitchL[ MAX_NB_SUBFR ],                     /* I    Pitch lags                      */
131     const opus_int              Lambda_Q10,                                 /* I    Rate/distortion tradeoff        */
132     const opus_int              LTP_scale_Q14                               /* I    LTP state scaling               */
133 );
134
135 extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])(
136     const silk_encoder_state    *psEncC,                                    /* I/O  Encoder State                   */
137     silk_nsq_state              *NSQ,                                       /* I/O  NSQ state                       */
138     SideInfoIndices             *psIndices,                                 /* I/O  Quantization Indices            */
139     const opus_int32            x_Q3[],                                     /* I    Prefiltered input signal        */
140     opus_int8                   pulses[],                                   /* O    Quantized pulse signal          */
141     const opus_int16            PredCoef_Q12[ 2 * MAX_LPC_ORDER ],          /* I    Short term prediction coefs     */
142     const opus_int16            LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ],    /* I    Long term prediction coefs      */
143     const opus_int16            AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs             */
144     const opus_int              HarmShapeGain_Q14[ MAX_NB_SUBFR ],          /* I    Long term shaping coefs         */
145     const opus_int              Tilt_Q14[ MAX_NB_SUBFR ],                   /* I    Spectral tilt                   */
146     const opus_int32            LF_shp_Q14[ MAX_NB_SUBFR ],                 /* I    Low frequency shaping coefs     */
147     const opus_int32            Gains_Q16[ MAX_NB_SUBFR ],                  /* I    Quantization step sizes         */
148     const opus_int              pitchL[ MAX_NB_SUBFR ],                     /* I    Pitch lags                      */
149     const opus_int              Lambda_Q10,                                 /* I    Rate/distortion tradeoff        */
150     const opus_int              LTP_scale_Q14                               /* I    LTP state scaling               */
151 );
152
153 #  define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \
154                            HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \
155     ((*SILK_NSQ_DEL_DEC_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \
156                            HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14))
157
158 void silk_noise_shape_quantizer(
159     silk_nsq_state      *NSQ,                   /* I/O  NSQ state                       */
160     opus_int            signalType,             /* I    Signal type                     */
161     const opus_int32    x_sc_Q10[],             /* I                                    */
162     opus_int8           pulses[],               /* O                                    */
163     opus_int16          xq[],                   /* O                                    */
164     opus_int32          sLTP_Q15[],             /* I/O  LTP state                       */
165     const opus_int16    a_Q12[],                /* I    Short term prediction coefs     */
166     const opus_int16    b_Q14[],                /* I    Long term prediction coefs      */
167     const opus_int16    AR_shp_Q13[],           /* I    Noise shaping AR coefs          */
168     opus_int            lag,                    /* I    Pitch lag                       */
169     opus_int32          HarmShapeFIRPacked_Q14, /* I                                    */
170     opus_int            Tilt_Q14,               /* I    Spectral tilt                   */
171     opus_int32          LF_shp_Q14,             /* I                                    */
172     opus_int32          Gain_Q16,               /* I                                    */
173     opus_int            Lambda_Q10,             /* I                                    */
174     opus_int            offset_Q10,             /* I                                    */
175     opus_int            length,                 /* I    Input length                    */
176     opus_int            shapingLPCOrder,        /* I    Noise shaping AR filter order   */
177     opus_int            predictLPCOrder         /* I    Prediction filter order         */
178 );
179
180 /**************************/
181 /* Noise level estimation */
182 /**************************/
183 void silk_VAD_GetNoiseLevels(
184     const opus_int32            pX[ VAD_N_BANDS ],  /* I    subband energies                            */
185     silk_VAD_state              *psSilk_VAD         /* I/O  Pointer to Silk VAD state                   */
186 );
187
188 #  define OVERRIDE_silk_VAD_GetSA_Q8
189
190 opus_int silk_VAD_GetSA_Q8_sse4_1(
191     silk_encoder_state *psEnC,
192     const opus_int16   pIn[]
193 );
194
195 #  define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \
196      ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn))
197
198 extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])(
199      silk_encoder_state *psEnC,
200      const opus_int16   pIn[]);
201
202 #  define OVERRIDE_silk_warped_LPC_analysis_filter_FIX
203
204 void silk_warped_LPC_analysis_filter_FIX_sse4_1(
205           opus_int32            state[],                    /* I/O  State [order + 1]                   */
206           opus_int32            res_Q2[],                   /* O    Residual signal [length]            */
207     const opus_int16            coef_Q13[],                 /* I    Coefficients [order]                */
208     const opus_int16            input[],                    /* I    Input signal [length]               */
209     const opus_int16            lambda_Q16,                 /* I    Warping factor                      */
210     const opus_int              length,                     /* I    Length of input signal              */
211     const opus_int              order                       /* I    Filter order (even)                 */
212 );
213
214 extern void (*const SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[OPUS_ARCHMASK + 1])(
215           opus_int32            state[],                    /* I/O  State [order + 1]                   */
216           opus_int32            res_Q2[],                   /* O    Residual signal [length]            */
217     const opus_int16            coef_Q13[],                 /* I    Coefficients [order]                */
218     const opus_int16            input[],                    /* I    Input signal [length]               */
219     const opus_int16            lambda_Q16,                 /* I    Warping factor                      */
220     const opus_int              length,                     /* I    Length of input signal              */
221     const opus_int              order                       /* I    Filter order (even)                 */
222 );
223
224 #  define silk_warped_LPC_analysis_filter_FIX(state, res_Q2, coef_Q13, input, lambda_Q16, length, order, arch) \
225     ((*SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[(arch) & OPUS_ARCHMASK])(state, res_Q2, coef_Q13, input, lambda_Q16, length, order))
226
227 # endif
228 #endif