diff options
author | Djordje Pesut <djordje.pesut@imgtec.com> | 2015-07-20 14:36:19 +0300 |
---|---|---|
committer | Michael Niedermayer <michael@niedermayer.cc> | 2015-07-22 22:51:28 +0300 |
commit | 5fd81cf6f082ed00878a5898f47550cb1646d219 (patch) | |
tree | 307a630e3bedb5a186367f08377f5ea3d22b854a /libavcodec | |
parent | 631496e057a203e656d0a952acecf217a83bb26b (diff) |
avcodec: Implementation of AAC_fixed_decoder (PS-module)
Add fixed point implementation.
Signed-off-by: Nedeljko Babic <nedeljko.babic@imgtec.com>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Diffstat (limited to 'libavcodec')
-rw-r--r-- | libavcodec/Makefile | 14 | ||||
-rw-r--r-- | libavcodec/aac_defines.h | 36 | ||||
-rw-r--r-- | libavcodec/aacps.c | 255 | ||||
-rw-r--r-- | libavcodec/aacps.h | 32 | ||||
-rw-r--r-- | libavcodec/aacps_fixed.c | 24 | ||||
-rw-r--r-- | libavcodec/aacps_fixed_tablegen.h | 2 | ||||
-rw-r--r-- | libavcodec/aacps_float.c | 24 | ||||
-rw-r--r-- | libavcodec/aacpsdata.c | 6 | ||||
-rw-r--r-- | libavcodec/aacpsdsp.c | 216 | ||||
-rw-r--r-- | libavcodec/aacpsdsp.h | 30 | ||||
-rw-r--r-- | libavcodec/aacpsdsp_fixed.c | 23 | ||||
-rw-r--r-- | libavcodec/aacpsdsp_float.c | 23 | ||||
-rw-r--r-- | libavcodec/aacpsdsp_template.c | 228 | ||||
-rw-r--r-- | libavcodec/aacsbr_template.c | 8 |
14 files changed, 571 insertions, 350 deletions
diff --git a/libavcodec/Makefile b/libavcodec/Makefile index 88e3ac2d31..eea9eade2c 100644 --- a/libavcodec/Makefile +++ b/libavcodec/Makefile @@ -123,12 +123,12 @@ OBJS-$(CONFIG_WMV2DSP) += wmv2dsp.o OBJS-$(CONFIG_ZERO12V_DECODER) += 012v.o OBJS-$(CONFIG_A64MULTI_ENCODER) += a64multienc.o elbg.o OBJS-$(CONFIG_A64MULTI5_ENCODER) += a64multienc.o elbg.o -OBJS-$(CONFIG_AAC_DECODER) += aacdec.o aactab.o aacsbr.o aacps.o \ +OBJS-$(CONFIG_AAC_DECODER) += aacdec.o aactab.o aacsbr.o aacps_float.o \ aacadtsdec.o mpeg4audio.o kbdwin.o \ - sbrdsp.o aacpsdsp.o -OBJS-$(CONFIG_AAC_FIXED_DECODER) += aacdec_fixed.o aactab.o aacsbr_fixed.o \ + sbrdsp.o aacpsdsp_float.o +OBJS-$(CONFIG_AAC_FIXED_DECODER) += aacdec_fixed.o aactab.o aacsbr_fixed.o aacps_fixed.o \ aacadtsdec.o mpeg4audio.o kbdwin.o \ - sbrdsp_fixed.o + sbrdsp_fixed.o aacpsdsp_fixed.o OBJS-$(CONFIG_AAC_ENCODER) += aacenc.o aaccoder.o \ aacpsy.o aactab.o \ psymodel.o mpeg4audio.o kbdwin.o @@ -932,6 +932,7 @@ TOOLS = fourcc2pixfmt HOSTPROGS = aac_tablegen \ aacps_tablegen \ + aacps_fixed_tablegen \ aacsbr_tablegen \ aacsbr_fixed_tablegen \ cabac_tablegen \ @@ -964,7 +965,7 @@ else $(SUBDIR)%_tablegen$(HOSTEXESUF): HOSTCFLAGS += -DCONFIG_SMALL=0 endif -GEN_HEADERS = cabac_tables.h cbrt_tables.h cbrt_fixed_tables.h aacps_tables.h aacsbr_tables.h \ +GEN_HEADERS = cabac_tables.h cbrt_tables.h cbrt_fixed_tables.h aacps_tables.h aacps_fixed_tables.h aacsbr_tables.h \ aacsbr_fixed_tables.h aac_tables.h dsd_tables.h dv_tables.h \ sinewin_tables.h sinewin_fixed_tables.h mpegaudio_tables.h motionpixels_tables.h \ pcm_tables.h qdm2_tables.h @@ -976,7 +977,8 @@ $(GEN_HEADERS): $(SUBDIR)%_tables.h: $(SUBDIR)%_tablegen$(HOSTEXESUF) ifdef CONFIG_HARDCODED_TABLES $(SUBDIR)aacdec.o: $(SUBDIR)cbrt_tables.h $(SUBDIR)aacdec_fixed.o: $(SUBDIR)cbrt_fixed_tables.h -$(SUBDIR)aacps.o: $(SUBDIR)aacps_tables.h +$(SUBDIR)aacps_float.o: $(SUBDIR)aacps_tables.h +$(SUBDIR)aacps_fixed.o: $(SUBDIR)aacps_fixed_tables.h $(SUBDIR)aacsbr.o: $(SUBDIR)aacsbr_tables.h $(SUBDIR)aacsbr_fixed.o: $(SUBDIR)aacsbr_fixed_tables.h $(SUBDIR)aactab.o: $(SUBDIR)aac_tables.h diff --git a/libavcodec/aac_defines.h b/libavcodec/aac_defines.h index 0f3905fe0b..3c45742ee1 100644 --- a/libavcodec/aac_defines.h +++ b/libavcodec/aac_defines.h @@ -35,6 +35,7 @@ #define AAC_RENAME(x) x ## _fixed #define AAC_RENAME_32(x) x ## _fixed_32 #define INTFLOAT int +#define INT64FLOAT int64_t #define SHORTFLOAT int16_t #define AAC_FLOAT SoftFloat #define AAC_SIGNE int @@ -45,9 +46,33 @@ #define Q31(x) (int)((x)*2147483648.0 + 0.5) #define RANGE15(x) x #define GET_GAIN(x, y) (-(y) << (x)) + 1024 +#define AAC_MUL16(x, y) (int)(((int64_t)(x) * (y) + 0x8000) >> 16) #define AAC_MUL26(x, y) (int)(((int64_t)(x) * (y) + 0x2000000) >> 26) #define AAC_MUL30(x, y) (int)(((int64_t)(x) * (y) + 0x20000000) >> 30) #define AAC_MUL31(x, y) (int)(((int64_t)(x) * (y) + 0x40000000) >> 31) +#define AAC_MADD28(x, y, a, b) (int)((((int64_t)(x) * (y)) + \ + ((int64_t)(a) * (b)) + \ + 0x8000000) >> 28) +#define AAC_MADD30(x, y, a, b) (int)((((int64_t)(x) * (y)) + \ + ((int64_t)(a) * (b)) + \ + 0x20000000) >> 30) +#define AAC_MADD30_V8(x, y, a, b, c, d, e, f) (int)((((int64_t)(x) * (y)) + \ + ((int64_t)(a) * (b)) + \ + ((int64_t)(c) * (d)) + \ + ((int64_t)(e) * (f)) + \ + 0x20000000) >> 30) +#define AAC_MSUB30(x, y, a, b) (int)((((int64_t)(x) * (y)) - \ + ((int64_t)(a) * (b)) + \ + 0x20000000) >> 30) +#define AAC_MSUB30_V8(x, y, a, b, c, d, e, f) (int)((((int64_t)(x) * (y)) + \ + ((int64_t)(a) * (b)) - \ + ((int64_t)(c) * (d)) - \ + ((int64_t)(e) * (f)) + \ + 0x20000000) >> 30) +#define AAC_MSUB31_V3(x, y, z) (int)((((int64_t)(x) * (z)) - \ + ((int64_t)(y) * (z)) + \ + 0x40000000) >> 31) +#define AAC_HALF_SUM(x, y) (x) >> 1 + (y) >> 1 #define AAC_SRA_R(x, y) (int)(((x) + (1 << ((y) - 1))) >> (y)) #else @@ -58,6 +83,7 @@ #define AAC_RENAME(x) x #define AAC_RENAME_32(x) x #define INTFLOAT float +#define INT64FLOAT float #define SHORTFLOAT float #define AAC_FLOAT float #define AAC_SIGNE unsigned @@ -68,9 +94,19 @@ #define Q31(x) x #define RANGE15(x) (32768.0 * (x)) #define GET_GAIN(x, y) powf((x), -(y)) +#define AAC_MUL16(x, y) ((x) * (y)) #define AAC_MUL26(x, y) ((x) * (y)) #define AAC_MUL30(x, y) ((x) * (y)) #define AAC_MUL31(x, y) ((x) * (y)) +#define AAC_MADD28(x, y, a, b) ((x) * (y) + (a) * (b)) +#define AAC_MADD30(x, y, a, b) ((x) * (y) + (a) * (b)) +#define AAC_MADD30_V8(x, y, a, b, c, d, e, f) ((x) * (y) + (a) * (b) + \ + (c) * (d) + (e) * (f)) +#define AAC_MSUB30(x, y, a, b) ((x) * (y) - (a) * (b)) +#define AAC_MSUB30_V8(x, y, a, b, c, d, e, f) ((x) * (y) + (a) * (b) - \ + (c) * (d) - (e) * (f)) +#define AAC_MSUB31_V3(x, y, z) ((x) - (y)) * (z) +#define AAC_HALF_SUM(x, y) ((x) + (y)) * 0.5f #define AAC_SRA_R(x, y) (x) #endif /* USE_FIXED */ diff --git a/libavcodec/aacps.c b/libavcodec/aacps.c index ea5a5d2331..bf6047594b 100644 --- a/libavcodec/aacps.c +++ b/libavcodec/aacps.c @@ -17,16 +17,23 @@ * You should have received a copy of the GNU Lesser General Public * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Note: Rounding-to-nearest used unless otherwise stated + * */ #include <stdint.h> #include "libavutil/common.h" -#include "libavutil/internal.h" #include "libavutil/mathematics.h" #include "avcodec.h" #include "get_bits.h" #include "aacps.h" +#if USE_FIXED +#include "aacps_fixed_tablegen.h" +#else +#include "libavutil/internal.h" #include "aacps_tablegen.h" +#endif /* USE_FIXED */ #include "aacpsdata.c" #define PS_BASELINE 0 ///< Operate in Baseline PS mode @@ -148,7 +155,7 @@ static void ipdopd_reset(int8_t *ipd_hist, int8_t *opd_hist) } } -int ff_ps_read_data(AVCodecContext *avctx, GetBitContext *gb_host, PSContext *ps, int bits_left) +int AAC_RENAME(ff_ps_read_data)(AVCodecContext *avctx, GetBitContext *gb_host, PSContext *ps, int bits_left) { int e; int bit_count_start = get_bits_count(gb_host); @@ -302,35 +309,41 @@ err: /** Split one subband into 2 subsubbands with a symmetric real filter. * The filter must have its non-center even coefficients equal to zero. */ -static void hybrid2_re(float (*in)[2], float (*out)[32][2], const float filter[8], int len, int reverse) +static void hybrid2_re(INTFLOAT (*in)[2], INTFLOAT (*out)[32][2], const INTFLOAT filter[8], int len, int reverse) { int i, j; for (i = 0; i < len; i++, in++) { - float re_in = filter[6] * in[6][0]; //real inphase - float re_op = 0.0f; //real out of phase - float im_in = filter[6] * in[6][1]; //imag inphase - float im_op = 0.0f; //imag out of phase + INT64FLOAT re_in = AAC_MUL31(filter[6], in[6][0]); //real inphase + INT64FLOAT re_op = 0.0f; //real out of phase + INT64FLOAT im_in = AAC_MUL31(filter[6], in[6][1]); //imag inphase + INT64FLOAT im_op = 0.0f; //imag out of phase for (j = 0; j < 6; j += 2) { - re_op += filter[j+1] * (in[j+1][0] + in[12-j-1][0]); - im_op += filter[j+1] * (in[j+1][1] + in[12-j-1][1]); + re_op += (INT64FLOAT)filter[j+1] * (in[j+1][0] + in[12-j-1][0]); + im_op += (INT64FLOAT)filter[j+1] * (in[j+1][1] + in[12-j-1][1]); } - out[ reverse][i][0] = re_in + re_op; - out[ reverse][i][1] = im_in + im_op; - out[!reverse][i][0] = re_in - re_op; - out[!reverse][i][1] = im_in - im_op; + +#if USE_FIXED + re_op = (re_op + 0x40000000) >> 31; + im_op = (im_op + 0x40000000) >> 31; +#endif /* USE_FIXED */ + + out[ reverse][i][0] = (INTFLOAT)(re_in + re_op); + out[ reverse][i][1] = (INTFLOAT)(im_in + im_op); + out[!reverse][i][0] = (INTFLOAT)(re_in - re_op); + out[!reverse][i][1] = (INTFLOAT)(im_in - im_op); } } /** Split one subband into 6 subsubbands with a complex filter */ -static void hybrid6_cx(PSDSPContext *dsp, float (*in)[2], float (*out)[32][2], - TABLE_CONST float (*filter)[8][2], int len) +static void hybrid6_cx(PSDSPContext *dsp, INTFLOAT (*in)[2], INTFLOAT (*out)[32][2], + TABLE_CONST INTFLOAT (*filter)[8][2], int len) { int i; int N = 8; - LOCAL_ALIGNED_16(float, temp, [8], [2]); + LOCAL_ALIGNED_16(INTFLOAT, temp, [8], [2]); for (i = 0; i < len; i++, in++) { - dsp->hybrid_analysis(temp, in, (const float (*)[8][2]) filter, 1, N); + dsp->hybrid_analysis(temp, in, (const INTFLOAT (*)[8][2]) filter, 1, N); out[0][i][0] = temp[6][0]; out[0][i][1] = temp[6][1]; out[1][i][0] = temp[7][0]; @@ -347,18 +360,18 @@ static void hybrid6_cx(PSDSPContext *dsp, float (*in)[2], float (*out)[32][2], } static void hybrid4_8_12_cx(PSDSPContext *dsp, - float (*in)[2], float (*out)[32][2], - TABLE_CONST float (*filter)[8][2], int N, int len) + INTFLOAT (*in)[2], INTFLOAT (*out)[32][2], + TABLE_CONST INTFLOAT (*filter)[8][2], int N, int len) { int i; for (i = 0; i < len; i++, in++) { - dsp->hybrid_analysis(out[0] + i, in, (const float (*)[8][2]) filter, 32, N); + dsp->hybrid_analysis(out[0] + i, in, (const INTFLOAT (*)[8][2]) filter, 32, N); } } -static void hybrid_analysis(PSDSPContext *dsp, float out[91][32][2], - float in[5][44][2], float L[2][38][64], +static void hybrid_analysis(PSDSPContext *dsp, INTFLOAT out[91][32][2], + INTFLOAT in[5][44][2], INTFLOAT L[2][38][64], int is34, int len) { int i, j; @@ -387,8 +400,8 @@ static void hybrid_analysis(PSDSPContext *dsp, float out[91][32][2], } } -static void hybrid_synthesis(PSDSPContext *dsp, float out[2][38][64], - float in[91][32][2], int is34, int len) +static void hybrid_synthesis(PSDSPContext *dsp, INTFLOAT out[2][38][64], + INTFLOAT in[91][32][2], int is34, int len) { int i, n; if (is34) { @@ -429,7 +442,7 @@ static void hybrid_synthesis(PSDSPContext *dsp, float out[2][38][64], } /// All-pass filter decay slope -#define DECAY_SLOPE 0.05f +#define DECAY_SLOPE Q30(0.05f) /// Number of frequency bands that can be addressed by the parameter index, b(k) static const int NR_PAR_BANDS[] = { 20, 34 }; static const int NR_IPDOPD_BANDS[] = { 11, 17 }; @@ -483,28 +496,43 @@ static void map_idx_34_to_20(int8_t *par_mapped, const int8_t *par, int full) } } -static void map_val_34_to_20(float par[PS_MAX_NR_IIDICC]) +static void map_val_34_to_20(INTFLOAT par[PS_MAX_NR_IIDICC]) { +#if USE_FIXED + par[ 0] = (int)(((int64_t)(par[ 0] + (par[ 1]>>1)) * 1431655765 + \ + 0x40000000) >> 31); + par[ 1] = (int)(((int64_t)((par[ 1]>>1) + par[ 2]) * 1431655765 + \ + 0x40000000) >> 31); + par[ 2] = (int)(((int64_t)(par[ 3] + (par[ 4]>>1)) * 1431655765 + \ + 0x40000000) >> 31); + par[ 3] = (int)(((int64_t)((par[ 4]>>1) + par[ 5]) * 1431655765 + \ + 0x40000000) >> 31); +#else par[ 0] = (2*par[ 0] + par[ 1]) * 0.33333333f; par[ 1] = ( par[ 1] + 2*par[ 2]) * 0.33333333f; par[ 2] = (2*par[ 3] + par[ 4]) * 0.33333333f; par[ 3] = ( par[ 4] + 2*par[ 5]) * 0.33333333f; - par[ 4] = ( par[ 6] + par[ 7]) * 0.5f; - par[ 5] = ( par[ 8] + par[ 9]) * 0.5f; +#endif /* USE_FIXED */ + par[ 4] = AAC_HALF_SUM(par[ 6], par[ 7]); + par[ 5] = AAC_HALF_SUM(par[ 8], par[ 9]); par[ 6] = par[10]; par[ 7] = par[11]; - par[ 8] = ( par[12] + par[13]) * 0.5f; - par[ 9] = ( par[14] + par[15]) * 0.5f; + par[ 8] = AAC_HALF_SUM(par[12], par[13]); + par[ 9] = AAC_HALF_SUM(par[14], par[15]); par[10] = par[16]; par[11] = par[17]; par[12] = par[18]; par[13] = par[19]; - par[14] = ( par[20] + par[21]) * 0.5f; - par[15] = ( par[22] + par[23]) * 0.5f; - par[16] = ( par[24] + par[25]) * 0.5f; - par[17] = ( par[26] + par[27]) * 0.5f; + par[14] = AAC_HALF_SUM(par[20], par[21]); + par[15] = AAC_HALF_SUM(par[22], par[23]); + par[16] = AAC_HALF_SUM(par[24], par[25]); + par[17] = AAC_HALF_SUM(par[26], par[27]); +#if USE_FIXED + par[18] = (((par[28]+2)>>2) + ((par[29]+2)>>2) + ((par[30]+2)>>2) + ((par[31]+2)>>2)); +#else par[18] = ( par[28] + par[29] + par[30] + par[31]) * 0.25f; - par[19] = ( par[32] + par[33]) * 0.5f; +#endif /* USE_FIXED */ + par[19] = AAC_HALF_SUM(par[32], par[33]); } static void map_idx_10_to_34(int8_t *par_mapped, const int8_t *par, int full) @@ -589,7 +617,7 @@ static void map_idx_20_to_34(int8_t *par_mapped, const int8_t *par, int full) par_mapped[ 0] = par[ 0]; } -static void map_val_20_to_34(float par[PS_MAX_NR_IIDICC]) +static void map_val_20_to_34(INTFLOAT par[PS_MAX_NR_IIDICC]) { par[33] = par[19]; par[32] = par[19]; @@ -620,27 +648,29 @@ static void map_val_20_to_34(float par[PS_MAX_NR_IIDICC]) par[ 7] = par[ 4]; par[ 6] = par[ 4]; par[ 5] = par[ 3]; - par[ 4] = (par[ 2] + par[ 3]) * 0.5f; + par[ 4] = AAC_HALF_SUM(par[ 2], par[ 3]); par[ 3] = par[ 2]; par[ 2] = par[ 1]; - par[ 1] = (par[ 0] + par[ 1]) * 0.5f; + par[ 1] = AAC_HALF_SUM(par[ 0], par[ 1]); } -static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[32][2], int is34) +static void decorrelation(PSContext *ps, INTFLOAT (*out)[32][2], const INTFLOAT (*s)[32][2], int is34) { - LOCAL_ALIGNED_16(float, power, [34], [PS_QMF_TIME_SLOTS]); - LOCAL_ALIGNED_16(float, transient_gain, [34], [PS_QMF_TIME_SLOTS]); - float *peak_decay_nrg = ps->peak_decay_nrg; - float *power_smooth = ps->power_smooth; - float *peak_decay_diff_smooth = ps->peak_decay_diff_smooth; - float (*delay)[PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2] = ps->delay; - float (*ap_delay)[PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2] = ps->ap_delay; - const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20; - const float peak_decay_factor = 0.76592833836465f; + LOCAL_ALIGNED_16(INTFLOAT, power, [34], [PS_QMF_TIME_SLOTS]); + LOCAL_ALIGNED_16(INTFLOAT, transient_gain, [34], [PS_QMF_TIME_SLOTS]); + INTFLOAT *peak_decay_nrg = ps->peak_decay_nrg; + INTFLOAT *power_smooth = ps->power_smooth; + INTFLOAT *peak_decay_diff_smooth = ps->peak_decay_diff_smooth; + INTFLOAT (*delay)[PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2] = ps->delay; + INTFLOAT (*ap_delay)[PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2] = ps->ap_delay; +#if !USE_FIXED const float transient_impact = 1.5f; const float a_smooth = 0.25f; ///< Smoothing coefficient +#endif /* USE_FIXED */ + const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20; int i, k, m, n; int n0 = 0, nL = 32; + const INTFLOAT peak_decay_factor = Q31(0.76592833836465f);; memset(power, 0, 34 * sizeof(*power)); @@ -658,6 +688,33 @@ static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[3 } //Transient detection +#if USE_FIXED + for (i = 0; i < NR_PAR_BANDS[is34]; i++) { + for (n = n0; n < nL; n++) { + int decayed_peak; + int denom; + + decayed_peak = (int)(((int64_t)peak_decay_factor * \ + peak_decay_nrg[i] + 0x40000000) >> 31); + peak_decay_nrg[i] = FFMAX(decayed_peak, power[i][n]); + power_smooth[i] += (power[i][n] - power_smooth[i] + 2) >> 2; + peak_decay_diff_smooth[i] += (peak_decay_nrg[i] - power[i][n] - \ + peak_decay_diff_smooth[i] + 2) >> 2; + denom = peak_decay_diff_smooth[i] + (peak_decay_diff_smooth[i] >> 1); + if (denom > power_smooth[i]) { + int p = power_smooth[i]; + while (denom < 0x40000000) { + denom <<= 1; + p <<= 1; + } + transient_gain[i][n] = p / (denom >> 16); + } + else { + transient_gain[i][n] = 1 << 16; + } + } + } +#else for (i = 0; i < NR_PAR_BANDS[is34]; i++) { for (n = n0; n < nL; n++) { float decayed_peak = peak_decay_factor * peak_decay_nrg[i]; @@ -671,6 +728,7 @@ static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[3 } } +#endif /* USE_FIXED */ //Decorrelation and transient reduction // PS_AP_LINKS - 1 // ----- @@ -681,8 +739,22 @@ static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[3 //d[k][z] (out) = transient_gain_mapped[k][z] * H[k][z] * s[k][z] for (k = 0; k < NR_ALLPASS_BANDS[is34]; k++) { int b = k_to_i[k]; +#if USE_FIXED + int g_decay_slope; + + if (k - DECAY_CUTOFF[is34] <= 0) { + g_decay_slope = 1 << 30; + } + else if (k - DECAY_CUTOFF[is34] >= 20) { + g_decay_slope = 0; + } + else { + g_decay_slope = (1 << 30) - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]); + } +#else float g_decay_slope = 1.f - DECAY_SLOPE * (k - DECAY_CUTOFF[is34]); g_decay_slope = av_clipf(g_decay_slope, 0.f, 1.f); +#endif /* USE_FIXED */ memcpy(delay[k], delay[k]+nL, PS_MAX_DELAY*sizeof(delay[k][0])); memcpy(delay[k]+PS_MAX_DELAY, s[k], numQMFSlots*sizeof(delay[k][0])); for (m = 0; m < PS_AP_LINKS; m++) { @@ -690,7 +762,7 @@ static void decorrelation(PSContext *ps, float (*out)[32][2], const float (*s)[3 } ps->dsp.decorrelate(out[k], delay[k] + PS_MAX_DELAY - 2, ap_delay[k], phi_fract[is34][k], - (const float (*)[2]) Q_fract_allpass[is34][k], + (const INTFLOAT (*)[2]) Q_fract_allpass[is34][k], transient_gain[b], g_decay_slope, nL - n0); } for (; k < SHORT_DELAY_BAND[is34]; k++) { @@ -749,14 +821,14 @@ static void remap20(int8_t (**p_par_mapped)[PS_MAX_NR_IIDICC], } } -static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2], int is34) +static void stereo_processing(PSContext *ps, INTFLOAT (*l)[32][2], INTFLOAT (*r)[32][2], int is34) { int e, b, k; - float (*H11)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H11; - float (*H12)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H12; - float (*H21)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H21; - float (*H22)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H22; + INTFLOAT (*H11)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H11; + INTFLOAT (*H12)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H12; + INTFLOAT (*H21)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H21; + INTFLOAT (*H22)[PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC] = ps->H22; int8_t *opd_hist = ps->opd_hist; int8_t *ipd_hist = ps->ipd_hist; int8_t iid_mapped_buf[PS_MAX_NUM_ENV][PS_MAX_NR_IIDICC]; @@ -768,7 +840,7 @@ static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2 int8_t (*ipd_mapped)[PS_MAX_NR_IIDICC] = ipd_mapped_buf; int8_t (*opd_mapped)[PS_MAX_NR_IIDICC] = opd_mapped_buf; const int8_t *k_to_i = is34 ? k_to_i_34 : k_to_i_20; - TABLE_CONST float (*H_LUT)[8][4] = (PS_BASELINE || ps->icc_mode < 3) ? HA : HB; + TABLE_CONST INTFLOAT (*H_LUT)[8][4] = (PS_BASELINE || ps->icc_mode < 3) ? HA : HB; //Remapping if (ps->num_env_old) { @@ -823,7 +895,7 @@ static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2 //Mixing for (e = 0; e < ps->num_env; e++) { for (b = 0; b < NR_PAR_BANDS[is34]; b++) { - float h11, h12, h21, h22; + INTFLOAT h11, h12, h21, h22; h11 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][0]; h12 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][1]; h21 = H_LUT[iid_mapped[e][b] + 7 + 23 * ps->iid_quant][icc_mapped[e][b]][2]; @@ -832,27 +904,27 @@ static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2 if (!PS_BASELINE && ps->enable_ipdopd && b < NR_IPDOPD_BANDS[is34]) { //The spec say says to only run this smoother when enable_ipdopd //is set but the reference decoder appears to run it constantly - float h11i, h12i, h21i, h22i; - float ipd_adj_re, ipd_adj_im; + INTFLOAT h11i, h12i, h21i, h22i; + INTFLOAT ipd_adj_re, ipd_adj_im; int opd_idx = opd_hist[b] * 8 + opd_mapped[e][b]; int ipd_idx = ipd_hist[b] * 8 + ipd_mapped[e][b]; - float opd_re = pd_re_smooth[opd_idx]; - float opd_im = pd_im_smooth[opd_idx]; - float ipd_re = pd_re_smooth[ipd_idx]; - float ipd_im = pd_im_smooth[ipd_idx]; + INTFLOAT opd_re = pd_re_smooth[opd_idx]; + INTFLOAT opd_im = pd_im_smooth[opd_idx]; + INTFLOAT ipd_re = pd_re_smooth[ipd_idx]; + INTFLOAT ipd_im = pd_im_smooth[ipd_idx]; opd_hist[b] = opd_idx & 0x3F; ipd_hist[b] = ipd_idx & 0x3F; - ipd_adj_re = opd_re*ipd_re + opd_im*ipd_im; - ipd_adj_im = opd_im*ipd_re - opd_re*ipd_im; - h11i = h11 * opd_im; - h11 = h11 * opd_re; - h12i = h12 * ipd_adj_im; - h12 = h12 * ipd_adj_re; - h21i = h21 * opd_im; - h21 = h21 * opd_re; - h22i = h22 * ipd_adj_im; - h22 = h22 * ipd_adj_re; + ipd_adj_re = AAC_MADD30(opd_re, ipd_re, opd_im, ipd_im); + ipd_adj_im = AAC_MSUB30(opd_im, ipd_re, opd_re, ipd_im); + h11i = AAC_MUL30(h11, opd_im); + h11 = AAC_MUL30(h11, opd_re); + h12i = AAC_MUL30(h12, ipd_adj_im); + h12 = AAC_MUL30(h12, ipd_adj_re); + h21i = AAC_MUL30(h21, opd_im); + h21 = AAC_MUL30(h21, opd_re); + h22i = AAC_MUL30(h22, ipd_adj_im); + h22 = AAC_MUL30(h22, ipd_adj_re); H11[1][e+1][b] = h11i; H12[1][e+1][b] = h12i; H21[1][e+1][b] = h21i; @@ -864,11 +936,14 @@ static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2 H22[0][e+1][b] = h22; } for (k = 0; k < NR_BANDS[is34]; k++) { - float h[2][4]; - float h_step[2][4]; + INTFLOAT h[2][4]; + INTFLOAT h_step[2][4]; int start = ps->border_position[e]; int stop = ps->border_position[e+1]; - float width = 1.f / (stop - start); + INTFLOAT width = Q30(1.f) / (stop - start); +#if USE_FIXED + width <<= 1; +#endif b = k_to_i[k]; h[0][0] = H11[0][e][b]; h[0][1] = H12[0][e][b]; @@ -889,15 +964,15 @@ static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2 } } //Interpolation - h_step[0][0] = (H11[0][e+1][b] - h[0][0]) * width; - h_step[0][1] = (H12[0][e+1][b] - h[0][1]) * width; - h_step[0][2] = (H21[0][e+1][b] - h[0][2]) * width; - h_step[0][3] = (H22[0][e+1][b] - h[0][3]) * width; + h_step[0][0] = AAC_MSUB31_V3(H11[0][e+1][b], h[0][0], width); + h_step[0][1] = AAC_MSUB31_V3(H12[0][e+1][b], h[0][1], width); + h_step[0][2] = AAC_MSUB31_V3(H21[0][e+1][b], h[0][2], width); + h_step[0][3] = AAC_MSUB31_V3(H22[0][e+1][b], h[0][3], width); if (!PS_BASELINE && ps->enable_ipdopd) { - h_step[1][0] = (H11[1][e+1][b] - h[1][0]) * width; - h_step[1][1] = (H12[1][e+1][b] - h[1][1]) * width; - h_step[1][2] = (H21[1][e+1][b] - h[1][2]) * width; - h_step[1][3] = (H22[1][e+1][b] - h[1][3]) * width; + h_step[1][0] = AAC_MSUB31_V3(H11[1][e+1][b], h[1][0], width); + h_step[1][1] = AAC_MSUB31_V3(H12[1][e+1][b], h[1][1], width); + h_step[1][2] = AAC_MSUB31_V3(H21[1][e+1][b], h[1][2], width); + h_step[1][3] = AAC_MSUB31_V3(H22[1][e+1][b], h[1][3], width); } ps->dsp.stereo_interpolate[!PS_BASELINE && ps->enable_ipdopd]( l[k] + start + 1, r[k] + start + 1, @@ -906,10 +981,10 @@ static void stereo_processing(PSContext *ps, float (*l)[32][2], float (*r)[32][2 } } -int ff_ps_apply(AVCodecContext *avctx, PSContext *ps, float L[2][38][64], float R[2][38][64], int top) +int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx, PSContext *ps, INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top) { - float (*Lbuf)[32][2] = ps->Lbuf; - float (*Rbuf)[32][2] = ps->Rbuf; + INTFLOAT (*Lbuf)[32][2] = ps->Lbuf; + INTFLOAT (*Rbuf)[32][2] = ps->Rbuf; const int len = 32; int is34 = ps->is34bands; @@ -919,7 +994,7 @@ int ff_ps_apply(AVCodecContext *avctx, PSContext *ps, float L[2][38][64], float memset(ps->ap_delay + top, 0, (NR_ALLPASS_BANDS[is34] - top)*sizeof(ps->ap_delay[0])); hybrid_analysis(&ps->dsp, Lbuf, ps->in_buf, L, is34, len); - decorrelation(ps, Rbuf, (const float (*)[32][2]) Lbuf, is34); + decorrelation(ps, Rbuf, (const INTFLOAT (*)[32][2]) Lbuf, is34); stereo_processing(ps, Lbuf, Rbuf, is34); hybrid_synthesis(&ps->dsp, L, Lbuf, is34, len); hybrid_synthesis(&ps->dsp, R, Rbuf, is34, len); @@ -936,7 +1011,7 @@ int ff_ps_apply(AVCodecContext *avctx, PSContext *ps, float L[2][38][64], float #define PS_VLC_ROW(name) \ { name ## _codes, name ## _bits, sizeof(name ## _codes), sizeof(name ## _codes[0]) } -av_cold void ff_ps_init(void) { +av_cold void AAC_RENAME(ff_ps_init)(void) { // Syntax initialization static const struct { const void *ps_codes, *ps_bits; @@ -968,7 +1043,7 @@ av_cold void ff_ps_init(void) { ps_tableinit(); } -av_cold void ff_ps_ctx_init(PSContext *ps) +av_cold void AAC_RENAME(ff_ps_ctx_init)(PSContext *ps) { - ff_psdsp_init(&ps->dsp); + AAC_RENAME(ff_psdsp_init)(&ps->dsp); } diff --git a/libavcodec/aacps.h b/libavcodec/aacps.h index 174770d6e4..54f9d99177 100644 --- a/libavcodec/aacps.h +++ b/libavcodec/aacps.h @@ -61,26 +61,26 @@ typedef struct PSContext { int is34bands; int is34bands_old; - DECLARE_ALIGNED(16, float, in_buf)[5][44][2]; - DECLARE_ALIGNED(16, float, delay)[PS_MAX_SSB][PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2]; - DECLARE_ALIGNED(16, float, ap_delay)[PS_MAX_AP_BANDS][PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2]; - DECLARE_ALIGNED(16, float, peak_decay_nrg)[34]; - DECLARE_ALIGNED(16, float, power_smooth)[34]; - DECLARE_ALIGNED(16, float, peak_decay_diff_smooth)[34]; - DECLARE_ALIGNED(16, float, H11)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC]; - DECLARE_ALIGNED(16, float, H12)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC]; - DECLARE_ALIGNED(16, float, H21)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC]; - DECLARE_ALIGNED(16, float, H22)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC]; - DECLARE_ALIGNED(16, float, Lbuf)[91][32][2]; - DECLARE_ALIGNED(16, float, Rbuf)[91][32][2]; + DECLARE_ALIGNED(16, INTFLOAT, in_buf)[5][44][2]; + DECLARE_ALIGNED(16, INTFLOAT, delay)[PS_MAX_SSB][PS_QMF_TIME_SLOTS + PS_MAX_DELAY][2]; + DECLARE_ALIGNED(16, INTFLOAT, ap_delay)[PS_MAX_AP_BANDS][PS_AP_LINKS][PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2]; + DECLARE_ALIGNED(16, INTFLOAT, peak_decay_nrg)[34]; + DECLARE_ALIGNED(16, INTFLOAT, power_smooth)[34]; + DECLARE_ALIGNED(16, INTFLOAT, peak_decay_diff_smooth)[34]; + DECLARE_ALIGNED(16, INTFLOAT, H11)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC]; + DECLARE_ALIGNED(16, INTFLOAT, H12)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC]; + DECLARE_ALIGNED(16, INTFLOAT, H21)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC]; + DECLARE_ALIGNED(16, INTFLOAT, H22)[2][PS_MAX_NUM_ENV+1][PS_MAX_NR_IIDICC]; + DECLARE_ALIGNED(16, INTFLOAT, Lbuf)[91][32][2]; + DECLARE_ALIGNED(16, INTFLOAT, Rbuf)[91][32][2]; int8_t opd_hist[PS_MAX_NR_IIDICC]; int8_t ipd_hist[PS_MAX_NR_IIDICC]; PSDSPContext dsp; } PSContext; -void ff_ps_init(void); -void ff_ps_ctx_init(PSContext *ps); -int ff_ps_read_data(AVCodecContext *avctx, GetBitContext *gb, PSContext *ps, int bits_left); -int ff_ps_apply(AVCodecContext *avctx, PSContext *ps, float L[2][38][64], float R[2][38][64], int top); +void AAC_RENAME(ff_ps_init)(void); +void AAC_RENAME(ff_ps_ctx_init)(PSContext *ps); +int AAC_RENAME(ff_ps_read_data)(AVCodecContext *avctx, GetBitContext *gb, PSContext *ps, int bits_left); +int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx, PSContext *ps, INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top); #endif /* AVCODEC_PS_H */ diff --git a/libavcodec/aacps_fixed.c b/libavcodec/aacps_fixed.c new file mode 100644 index 0000000000..46af21339a --- /dev/null +++ b/libavcodec/aacps_fixed.c @@ -0,0 +1,24 @@ +/* + * MPEG-4 Parametric Stereo decoding functions + * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com> + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#define USE_FIXED 1 + +#include "aacps.c" diff --git a/libavcodec/aacps_fixed_tablegen.h b/libavcodec/aacps_fixed_tablegen.h index 9474206b70..701a9d2b2b 100644 --- a/libavcodec/aacps_fixed_tablegen.h +++ b/libavcodec/aacps_fixed_tablegen.h @@ -288,7 +288,7 @@ static void ps_tableinit(void) int im_smooth = pd0_im + pd1_im + pd2_im; SoftFloat pd_mag = av_int2sf(((ipdopd_cos[(pd0-pd1)&7]+8)>>4) + ((ipdopd_cos[(pd0-pd2)&7]+4)>>3) + - ((ipdopd_cos[(pd1-pd2)&7]+2)>>2) + 0x15000000, 2); + ((ipdopd_cos[(pd1-pd2)&7]+2)>>2) + 0x15000000, 28); pd_mag = av_div_sf(FLOAT_1, av_sqrt_sf(pd_mag)); shift = 30 - pd_mag.exp; round = 1 << (shift-1); diff --git a/libavcodec/aacps_float.c b/libavcodec/aacps_float.c new file mode 100644 index 0000000000..73259c10fb --- /dev/null +++ b/libavcodec/aacps_float.c @@ -0,0 +1,24 @@ +/* + * MPEG-4 Parametric Stereo decoding functions + * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com> + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#define USE_FIXED 0 + +#include "aacps.c" diff --git a/libavcodec/aacpsdata.c b/libavcodec/aacpsdata.c index 7431caebc6..5c1a1b0f88 100644 --- a/libavcodec/aacpsdata.c +++ b/libavcodec/aacpsdata.c @@ -157,7 +157,7 @@ static const int8_t k_to_i_34[] = { 33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33 }; -static const float g1_Q2[] = { - 0.0f, 0.01899487526049f, 0.0f, -0.07293139167538f, - 0.0f, 0.30596630545168f, 0.5f +static const INTFLOAT g1_Q2[] = { + Q31(0.0f), Q31(0.01899487526049f), Q31(0.0f), Q31(-0.07293139167538f), + Q31(0.0f), Q31(0.30596630545168f), Q31(0.5f) }; diff --git a/libavcodec/aacpsdsp.c b/libavcodec/aacpsdsp.c deleted file mode 100644 index 5dc1a6aba9..0000000000 --- a/libavcodec/aacpsdsp.c +++ /dev/null @@ -1,216 +0,0 @@ -/* - * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com> - * - * This file is part of FFmpeg. - * - * FFmpeg is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; either - * version 2.1 of the License, or (at your option) any later version. - * - * FFmpeg is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -#include "config.h" -#include "libavutil/attributes.h" -#include "aacpsdsp.h" - -static void ps_add_squares_c(float *dst, const float (*src)[2], int n) -{ - int i; - for (i = 0; i < n; i++) - dst[i] += src[i][0] * src[i][0] + src[i][1] * src[i][1]; -} - -static void ps_mul_pair_single_c(float (*dst)[2], float (*src0)[2], float *src1, - int n) -{ - int i; - for (i = 0; i < n; i++) { - dst[i][0] = src0[i][0] * src1[i]; - dst[i][1] = src0[i][1] * src1[i]; - } -} - -static void ps_hybrid_analysis_c(float (*out)[2], float (*in)[2], - const float (*filter)[8][2], - int stride, int n) -{ - int i, j; - - for (i = 0; i < n; i++) { - float sum_re = filter[i][6][0] * in[6][0]; - float sum_im = filter[i][6][0] * in[6][1]; - - for (j = 0; j < 6; j++) { - float in0_re = in[j][0]; - float in0_im = in[j][1]; - float in1_re = in[12-j][0]; - float in1_im = in[12-j][1]; - sum_re += filter[i][j][0] * (in0_re + in1_re) - - filter[i][j][1] * (in0_im - in1_im); - sum_im += filter[i][j][0] * (in0_im + in1_im) + - filter[i][j][1] * (in0_re - in1_re); - } - out[i * stride][0] = sum_re; - out[i * stride][1] = sum_im; - } -} - -static void ps_hybrid_analysis_ileave_c(float (*out)[32][2], float L[2][38][64], - int i, int len) -{ - int j; - - for (; i < 64; i++) { - for (j = 0; j < len; j++) { - out[i][j][0] = L[0][j][i]; - out[i][j][1] = L[1][j][i]; - } - } -} - -static void ps_hybrid_synthesis_deint_c(float out[2][38][64], - float (*in)[32][2], - int i, int len) -{ - int n; - - for (; i < 64; i++) { - for (n = 0; n < len; n++) { - out[0][n][i] = in[i][n][0]; - out[1][n][i] = in[i][n][1]; - } - } -} - -static void ps_decorrelate_c(float (*out)[2], float (*delay)[2], - float (*ap_delay)[PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2], - const float phi_fract[2], const float (*Q_fract)[2], - const float *transient_gain, - float g_decay_slope, - int len) -{ - static const float a[] = { 0.65143905753106f, - 0.56471812200776f, - 0.48954165955695f }; - float ag[PS_AP_LINKS]; - int m, n; - - for (m = 0; m < PS_AP_LINKS; m++) - ag[m] = a[m] * g_decay_slope; - - for (n = 0; n < len; n++) { - float in_re = delay[n][0] * phi_fract[0] - delay[n][1] * phi_fract[1]; - float in_im = delay[n][0] * phi_fract[1] + delay[n][1] * phi_fract[0]; - for (m = 0; m < PS_AP_LINKS; m++) { - float a_re = ag[m] * in_re; - float a_im = ag[m] * in_im; - float link_delay_re = ap_delay[m][n+2-m][0]; - float link_delay_im = ap_delay[m][n+2-m][1]; - float fractional_delay_re = Q_fract[m][0]; - float fractional_delay_im = Q_fract[m][1]; - float apd_re = in_re; - float apd_im = in_im; - in_re = link_delay_re * fractional_delay_re - - link_delay_im * fractional_delay_im - a_re; - in_im = link_delay_re * fractional_delay_im + - link_delay_im * fractional_delay_re - a_im; - ap_delay[m][n+5][0] = apd_re + ag[m] * in_re; - ap_delay[m][n+5][1] = apd_im + ag[m] * in_im; - } - out[n][0] = transient_gain[n] * in_re; - out[n][1] = transient_gain[n] * in_im; - } -} - -static void ps_stereo_interpolate_c(float (*l)[2], float (*r)[2], - float h[2][4], float h_step[2][4], - int len) -{ - float h0 = h[0][0]; - float h1 = h[0][1]; - float h2 = h[0][2]; - float h3 = h[0][3]; - float hs0 = h_step[0][0]; - float hs1 = h_step[0][1]; - float hs2 = h_step[0][2]; - float hs3 = h_step[0][3]; - int n; - - for (n = 0; n < len; n++) { - //l is s, r is d - float l_re = l[n][0]; - float l_im = l[n][1]; - float r_re = r[n][0]; - float r_im = r[n][1]; - h0 += hs0; - h1 += hs1; - h2 += hs2; - h3 += hs3; - l[n][0] = h0 * l_re + h2 * r_re; - l[n][1] = h0 * l_im + h2 * r_im; - r[n][0] = h1 * l_re + h3 * r_re; - r[n][1] = h1 * l_im + h3 * r_im; - } -} - -static void ps_stereo_interpolate_ipdopd_c(float (*l)[2], float (*r)[2], - float h[2][4], float h_step[2][4], - int len) -{ - float h00 = h[0][0], h10 = h[1][0]; - float h01 = h[0][1], h11 = h[1][1]; - float h02 = h[0][2], h12 = h[1][2]; - float h03 = h[0][3], h13 = h[1][3]; - float hs00 = h_step[0][0], hs10 = h_step[1][0]; - float hs01 = h_step[0][1], hs11 = h_step[1][1]; - float hs02 = h_step[0][2], hs12 = h_step[1][2]; - float hs03 = h_step[0][3], hs13 = h_step[1][3]; - int n; - - for (n = 0; n < len; n++) { - //l is s, r is d - float l_re = l[n][0]; - float l_im = l[n][1]; - float r_re = r[n][0]; - float r_im = r[n][1]; - h00 += hs00; - h01 += hs01; - h02 += hs02; - h03 += hs03; - h10 += hs10; - h11 += hs11; - h12 += hs12; - h13 += hs13; - - l[n][0] = h00 * l_re + h02 * r_re - h10 * l_im - h12 * r_im; - l[n][1] = h00 * l_im + h02 * r_im + h10 * l_re + h12 * r_re; - r[n][0] = h01 * l_re + h03 * r_re - h11 * l_im - h13 * r_im; - r[n][1] = h01 * l_im + h03 * r_im + h11 * l_re + h13 * r_re; - } -} - -av_cold void ff_psdsp_init(PSDSPContext *s) -{ - s->add_squares = ps_add_squares_c; - s->mul_pair_single = ps_mul_pair_single_c; - s->hybrid_analysis = ps_hybrid_analysis_c; - s->hybrid_analysis_ileave = ps_hybrid_analysis_ileave_c; - s->hybrid_synthesis_deint = ps_hybrid_synthesis_deint_c; - s->decorrelate = ps_decorrelate_c; - s->stereo_interpolate[0] = ps_stereo_interpolate_c; - s->stereo_interpolate[1] = ps_stereo_interpolate_ipdopd_c; - - if (ARCH_ARM) - ff_psdsp_init_arm(s); - if (ARCH_MIPS) - ff_psdsp_init_mips(s); -} diff --git a/libavcodec/aacpsdsp.h b/libavcodec/aacpsdsp.h index 0ef30236ec..9e3c5aa9a7 100644 --- a/libavcodec/aacpsdsp.h +++ b/libavcodec/aacpsdsp.h @@ -21,33 +21,35 @@ #ifndef LIBAVCODEC_AACPSDSP_H #define LIBAVCODEC_AACPSDSP_H +#include "aac_defines.h" + #define PS_QMF_TIME_SLOTS 32 #define PS_AP_LINKS 3 #define PS_MAX_AP_DELAY 5 typedef struct PSDSPContext { - void (*add_squares)(float *dst, const float (*src)[2], int n); - void (*mul_pair_single)(float (*dst)[2], float (*src0)[2], float *src1, + void (*add_squares)(INTFLOAT *dst, const INTFLOAT (*src)[2], int n); + void (*mul_pair_single)(INTFLOAT (*dst)[2], INTFLOAT (*src0)[2], INTFLOAT *src1, int n); - void (*hybrid_analysis)(float (*out)[2], float (*in)[2], - const float (*filter)[8][2], + void (*hybrid_analysis)(INTFLOAT (*out)[2], INTFLOAT (*in)[2], + const INTFLOAT (*filter)[8][2], int stride, int n); - void (*hybrid_analysis_ileave)(float (*out)[32][2], float L[2][38][64], + void (*hybrid_analysis_ileave)(INTFLOAT (*out)[32][2], INTFLOAT L[2][38][64], int i, int len); - void (*hybrid_synthesis_deint)(float out[2][38][64], float (*in)[32][2], + void (*hybrid_synthesis_deint)(INTFLOAT out[2][38][64], INTFLOAT (*in)[32][2], int i, int len); - void (*decorrelate)(float (*out)[2], float (*delay)[2], - float (*ap_delay)[PS_QMF_TIME_SLOTS+PS_MAX_AP_DELAY][2], - const float phi_fract[2], const float (*Q_fract)[2], - const float *transient_gain, - float g_decay_slope, + void (*decorrelate)(INTFLOAT (*out)[2], INTFLOAT (*delay)[2], + INTFLOAT (*ap_delay)[PS_QMF_TIME_SLOTS+PS_MAX_AP_DELAY][2], + const INTFLOAT phi_fract[2], const INTFLOAT (*Q_fract)[2], + const INTFLOAT *transient_gain, + INTFLOAT g_decay_slope, int len); - void (*stereo_interpolate[2])(float (*l)[2], float (*r)[2], - float h[2][4], float h_step[2][4], + void (*stereo_interpolate[2])(INTFLOAT (*l)[2], INTFLOAT (*r)[2], + INTFLOAT h[2][4], INTFLOAT h_step[2][4], int len); } PSDSPContext; -void ff_psdsp_init(PSDSPContext *s); +void AAC_RENAME(ff_psdsp_init)(PSDSPContext *s); void ff_psdsp_init_arm(PSDSPContext *s); void ff_psdsp_init_mips(PSDSPContext *s); diff --git a/libavcodec/aacpsdsp_fixed.c b/libavcodec/aacpsdsp_fixed.c new file mode 100644 index 0000000000..2413295113 --- /dev/null +++ b/libavcodec/aacpsdsp_fixed.c @@ -0,0 +1,23 @@ +/* + * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com> + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#define USE_FIXED 1 + +#include "aacpsdsp_template.c" diff --git a/libavcodec/aacpsdsp_float.c b/libavcodec/aacpsdsp_float.c new file mode 100644 index 0000000000..99aa650acf --- /dev/null +++ b/libavcodec/aacpsdsp_float.c @@ -0,0 +1,23 @@ +/* + * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com> + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + */ + +#define USE_FIXED 0 + +#include "aacpsdsp_template.c" diff --git a/libavcodec/aacpsdsp_template.c b/libavcodec/aacpsdsp_template.c new file mode 100644 index 0000000000..bfec828cf6 --- /dev/null +++ b/libavcodec/aacpsdsp_template.c @@ -0,0 +1,228 @@ +/* + * Copyright (c) 2010 Alex Converse <alex.converse@gmail.com> + * + * This file is part of FFmpeg. + * + * FFmpeg is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * FFmpeg is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with FFmpeg; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA + * + * Note: Rounding-to-nearest used unless otherwise stated + * + */ +#include <stdint.h> + +#include "config.h" +#include "libavutil/attributes.h" +#include "aacpsdsp.h" + +static void ps_add_squares_c(INTFLOAT *dst, const INTFLOAT (*src)[2], int n) +{ + int i; + for (i = 0; i < n; i++) + dst[i] += AAC_MADD28(src[i][0], src[i][0], src[i][1], src[i][1]); +} + +static void ps_mul_pair_single_c(INTFLOAT (*dst)[2], INTFLOAT (*src0)[2], INTFLOAT *src1, + int n) +{ + int i; + for (i = 0; i < n; i++) { + dst[i][0] = AAC_MUL16(src0[i][0], src1[i]); + dst[i][1] = AAC_MUL16(src0[i][1], src1[i]); + } +} + +static void ps_hybrid_analysis_c(INTFLOAT (*out)[2], INTFLOAT (*in)[2], + const INTFLOAT (*filter)[8][2], + int stride, int n) +{ + int i, j; + + for (i = 0; i < n; i++) { + INT64FLOAT sum_re = (INT64FLOAT)filter[i][6][0] * in[6][0]; + INT64FLOAT sum_im = (INT64FLOAT)filter[i][6][0] * in[6][1]; + + for (j = 0; j < 6; j++) { + INTFLOAT in0_re = in[j][0]; + INTFLOAT in0_im = in[j][1]; + INTFLOAT in1_re = in[12-j][0]; + INTFLOAT in1_im = in[12-j][1]; + sum_re += (INT64FLOAT)filter[i][j][0] * (in0_re + in1_re) - + (INT64FLOAT)filter[i][j][1] * (in0_im - in1_im); + sum_im += (INT64FLOAT)filter[i][j][0] * (in0_im + in1_im) + + (INT64FLOAT)filter[i][j][1] * (in0_re - in1_re); + } +#if USE_FIXED + out[i * stride][0] = (int)((sum_re + 0x40000000) >> 31); + out[i * stride][1] = (int)((sum_im + 0x40000000) >> 31); +#else + out[i * stride][0] = sum_re; + out[i * stride][1] = sum_im; +#endif /* USE_FIXED */ + } +} +static void ps_hybrid_analysis_ileave_c(INTFLOAT (*out)[32][2], INTFLOAT L[2][38][64], + int i, int len) +{ + int j; + + for (; i < 64; i++) { + for (j = 0; j < len; j++) { + out[i][j][0] = L[0][j][i]; + out[i][j][1] = L[1][j][i]; + } + } +} + +static void ps_hybrid_synthesis_deint_c(INTFLOAT out[2][38][64], + INTFLOAT (*in)[32][2], + int i, int len) +{ + int n; + + for (; i < 64; i++) { + for (n = 0; n < len; n++) { + out[0][n][i] = in[i][n][0]; + out[1][n][i] = in[i][n][1]; + } + } +} + +static void ps_decorrelate_c(INTFLOAT (*out)[2], INTFLOAT (*delay)[2], + INTFLOAT (*ap_delay)[PS_QMF_TIME_SLOTS + PS_MAX_AP_DELAY][2], + const INTFLOAT phi_fract[2], const INTFLOAT (*Q_fract)[2], + const INTFLOAT *transient_gain, + INTFLOAT g_decay_slope, + int len) +{ + static const INTFLOAT a[] = { Q31(0.65143905753106f), + Q31(0.56471812200776f), + Q31(0.48954165955695f) }; + INTFLOAT ag[PS_AP_LINKS]; + int m, n; + + for (m = 0; m < PS_AP_LINKS; m++) + ag[m] = AAC_MUL30(a[m], g_decay_slope); + + for (n = 0; n < len; n++) { + INTFLOAT in_re = AAC_MSUB30(delay[n][0], phi_fract[0], delay[n][1], phi_fract[1]); + INTFLOAT in_im = AAC_MADD30(delay[n][0], phi_fract[1], delay[n][1], phi_fract[0]); + for (m = 0; m < PS_AP_LINKS; m++) { + INTFLOAT a_re = AAC_MUL31(ag[m], in_re); + INTFLOAT a_im = AAC_MUL31(ag[m], in_im); + INTFLOAT link_delay_re = ap_delay[m][n+2-m][0]; + INTFLOAT link_delay_im = ap_delay[m][n+2-m][1]; + INTFLOAT fractional_delay_re = Q_fract[m][0]; + INTFLOAT fractional_delay_im = Q_fract[m][1]; + INTFLOAT apd_re = in_re; + INTFLOAT apd_im = in_im; + in_re = AAC_MSUB30(link_delay_re, fractional_delay_re, + link_delay_im, fractional_delay_im); + in_re -= a_re; + in_im = AAC_MADD30(link_delay_re, fractional_delay_im, + link_delay_im, fractional_delay_re); + in_im -= a_im; + ap_delay[m][n+5][0] = apd_re + AAC_MUL31(ag[m], in_re); + ap_delay[m][n+5][1] = apd_im + AAC_MUL31(ag[m], in_im); + } + out[n][0] = AAC_MUL16(transient_gain[n], in_re); + out[n][1] = AAC_MUL16(transient_gain[n], in_im); + } +} + +static void ps_stereo_interpolate_c(INTFLOAT (*l)[2], INTFLOAT (*r)[2], + INTFLOAT h[2][4], INTFLOAT h_step[2][4], + int len) +{ + INTFLOAT h0 = h[0][0]; + INTFLOAT h1 = h[0][1]; + INTFLOAT h2 = h[0][2]; + INTFLOAT h3 = h[0][3]; + INTFLOAT hs0 = h_step[0][0]; + INTFLOAT hs1 = h_step[0][1]; + INTFLOAT hs2 = h_step[0][2]; + INTFLOAT hs3 = h_step[0][3]; + int n; + + for (n = 0; n < len; n++) { + //l is s, r is d + INTFLOAT l_re = l[n][0]; + INTFLOAT l_im = l[n][1]; + INTFLOAT r_re = r[n][0]; + INTFLOAT r_im = r[n][1]; + h0 += hs0; + h1 += hs1; + h2 += hs2; + h3 += hs3; + l[n][0] = AAC_MADD30(h0, l_re, h2, r_re); + l[n][1] = AAC_MADD30(h0, l_im, h2, r_im); + r[n][0] = AAC_MADD30(h1, l_re, h3, r_re); + r[n][1] = AAC_MADD30(h1, l_im, h3, r_im); + } +} + +static void ps_stereo_interpolate_ipdopd_c(INTFLOAT (*l)[2], INTFLOAT (*r)[2], + INTFLOAT h[2][4], INTFLOAT h_step[2][4], + int len) +{ + INTFLOAT h00 = h[0][0], h10 = h[1][0]; + INTFLOAT h01 = h[0][1], h11 = h[1][1]; + INTFLOAT h02 = h[0][2], h12 = h[1][2]; + INTFLOAT h03 = h[0][3], h13 = h[1][3]; + INTFLOAT hs00 = h_step[0][0], hs10 = h_step[1][0]; + INTFLOAT hs01 = h_step[0][1], hs11 = h_step[1][1]; + INTFLOAT hs02 = h_step[0][2], hs12 = h_step[1][2]; + INTFLOAT hs03 = h_step[0][3], hs13 = h_step[1][3]; + int n; + + for (n = 0; n < len; n++) { + //l is s, r is d + INTFLOAT l_re = l[n][0]; + INTFLOAT l_im = l[n][1]; + INTFLOAT r_re = r[n][0]; + INTFLOAT r_im = r[n][1]; + h00 += hs00; + h01 += hs01; + h02 += hs02; + h03 += hs03; + h10 += hs10; + h11 += hs11; + h12 += hs12; + h13 += hs13; + + l[n][0] = AAC_MSUB30_V8(h00, l_re, h02, r_re, h10, l_im, h12, r_im); + l[n][1] = AAC_MADD30_V8(h00, l_im, h02, r_im, h10, l_re, h12, r_re); + r[n][0] = AAC_MSUB30_V8(h01, l_re, h03, r_re, h11, l_im, h13, r_im); + r[n][1] = AAC_MADD30_V8(h01, l_im, h03, r_im, h11, l_re, h13, r_re); + } +} + +av_cold void AAC_RENAME(ff_psdsp_init)(PSDSPContext *s) +{ + s->add_squares = ps_add_squares_c; + s->mul_pair_single = ps_mul_pair_single_c; + s->hybrid_analysis = ps_hybrid_analysis_c; + s->hybrid_analysis_ileave = ps_hybrid_analysis_ileave_c; + s->hybrid_synthesis_deint = ps_hybrid_synthesis_deint_c; + s->decorrelate = ps_decorrelate_c; + s->stereo_interpolate[0] = ps_stereo_interpolate_c; + s->stereo_interpolate[1] = ps_stereo_interpolate_ipdopd_c; + +#if !USE_FIXED + if (ARCH_ARM) + ff_psdsp_init_arm(s); + if (ARCH_MIPS) + ff_psdsp_init_mips(s); +#endif /* !USE_FIXED */ +} diff --git a/libavcodec/aacsbr_template.c b/libavcodec/aacsbr_template.c index dd0ddcf7ee..d31b71e0a8 100644 --- a/libavcodec/aacsbr_template.c +++ b/libavcodec/aacsbr_template.c @@ -64,7 +64,7 @@ av_cold void AAC_RENAME(ff_aac_sbr_init)(void) aacsbr_tableinit(); - ff_ps_init(); + AAC_RENAME(ff_ps_init)(); } /** Places SBR in pure upsampling mode. */ @@ -91,7 +91,7 @@ av_cold void AAC_RENAME(ff_aac_sbr_ctx_init)(AACContext *ac, SpectralBandReplica * and scale back down at synthesis. */ AAC_RENAME_32(ff_mdct_init)(&sbr->mdct, 7, 1, 1.0 / (64 * 32768.0)); AAC_RENAME_32(ff_mdct_init)(&sbr->mdct_ana, 7, 1, -2.0 * 32768.0); - ff_ps_ctx_init(&sbr->ps); + AAC_RENAME(ff_ps_ctx_init)(&sbr->ps); AAC_RENAME(ff_sbrdsp_init)(&sbr->dsp); aacsbr_func_ptr_init(&sbr->c); } @@ -945,7 +945,7 @@ static void read_sbr_extension(AACContext *ac, SpectralBandReplication *sbr, *num_bits_left = 0; } else { #if 1 - *num_bits_left -= ff_ps_read_data(ac->avctx, gb, &sbr->ps, *num_bits_left); + *num_bits_left -= AAC_RENAME(ff_ps_read_data)(ac->avctx, gb, &sbr->ps, *num_bits_left); ac->avctx->profile = FF_PROFILE_AAC_HE_V2; #else avpriv_report_missing_feature(ac->avctx, "Parametric Stereo"); @@ -1501,7 +1501,7 @@ void AAC_RENAME(ff_sbr_apply)(AACContext *ac, SpectralBandReplication *sbr, int if (ac->oc[1].m4ac.ps == 1) { if (sbr->ps.start) { - ff_ps_apply(ac->avctx, &sbr->ps, sbr->X[0], sbr->X[1], sbr->kx[1] + sbr->m[1]); + AAC_RENAME(ff_ps_apply)(ac->avctx, &sbr->ps, sbr->X[0], sbr->X[1], sbr->kx[1] + sbr->m[1]); } else { memcpy(sbr->X[1], sbr->X[0], sizeof(sbr->X[0])); } |