do_imdct(s, s->channels);
if (downmix_output) {
- s->ac3dsp.downmix(s->outptr, s->downmix_coeffs,
+#if USE_FIXED
+ ac3_downmix_c_fixed16(s->outptr, s->downmix_coeffs,
+ s->out_channels, s->fbw_channels, 256);
+#else
+ ff_ac3dsp_downmix(&s->ac3dsp, s->outptr, s->downmix_coeffs,
s->out_channels, s->fbw_channels, 256);
+#endif
}
} else {
if (downmix_output) {
- s->ac3dsp.AC3_RENAME(downmix)(s->xcfptr + 1, s->downmix_coeffs,
- ff_ac3dsp_downmix(&s->ac3dsp, s->xcfptr + 1, s->downmix_coeffs,
- s->out_channels, s->fbw_channels, 256);
++ AC3_RENAME(ff_ac3dsp_downmix)(&s->ac3dsp, s->xcfptr + 1, s->downmix_coeffs,
+ s->out_channels, s->fbw_channels, 256);
}
if (downmix_output && !s->downmixed) {
s->downmixed = 1;
- s->ac3dsp.AC3_RENAME(downmix)(s->dlyptr, s->downmix_coeffs,
- ff_ac3dsp_downmix(&s->ac3dsp, s->dlyptr, s->downmix_coeffs,
- s->out_channels, s->fbw_channels, 128);
++ AC3_RENAME(ff_ac3dsp_downmix)(&s->ac3dsp, s->dlyptr, s->downmix_coeffs,
+ s->out_channels, s->fbw_channels, 128);
}
do_imdct(s, s->out_channels);
}
}
- static void ac3_downmix_c(float **samples, float **matrix,
- int out_ch, int in_ch, int len)
+static void ac3_sum_square_butterfly_int32_c(int64_t sum[4],
+ const int32_t *coef0,
+ const int32_t *coef1,
+ int len)
+{
+ int i;
+
+ sum[0] = sum[1] = sum[2] = sum[3] = 0;
+
+ for (i = 0; i < len; i++) {
+ int lt = coef0[i];
+ int rt = coef1[i];
+ int md = lt + rt;
+ int sd = lt - rt;
+ MAC64(sum[0], lt, lt);
+ MAC64(sum[1], rt, rt);
+ MAC64(sum[2], md, md);
+ MAC64(sum[3], sd, sd);
+ }
+}
+
+static void ac3_sum_square_butterfly_float_c(float sum[4],
+ const float *coef0,
+ const float *coef1,
+ int len)
+{
+ int i;
+
+ sum[0] = sum[1] = sum[2] = sum[3] = 0;
+
+ for (i = 0; i < len; i++) {
+ float lt = coef0[i];
+ float rt = coef1[i];
+ float md = lt + rt;
+ float sd = lt - rt;
+ sum[0] += lt * lt;
+ sum[1] += rt * rt;
+ sum[2] += md * md;
+ sum[3] += sd * sd;
+ }
+}
+
+ static void ac3_downmix_5_to_2_symmetric_c(float **samples, float **matrix,
+ int len)
{
- int **matrix_cmp = (int **)matrix;
- int i, j;
+ int i;
float v0, v1;
+ float front_mix = matrix[0][0];
+ float center_mix = matrix[0][1];
+ float surround_mix = matrix[0][3];
- if (in_ch == 5 && out_ch == 2 &&
- !(matrix_cmp[1][0] | matrix_cmp[0][2] |
- matrix_cmp[1][3] | matrix_cmp[0][4] |
- (matrix_cmp[0][1] ^ matrix_cmp[1][1]) |
- (matrix_cmp[0][0] ^ matrix_cmp[1][2]))) {
- float front_mix = matrix[0][0];
- float center_mix = matrix[0][1];
- float surround_mix = matrix[0][3];
+ for (i = 0; i < len; i++) {
+ v0 = samples[0][i] * front_mix +
+ samples[1][i] * center_mix +
+ samples[3][i] * surround_mix;
- for (i = 0; i < len; i++) {
- v0 = samples[0][i] * front_mix +
- samples[1][i] * center_mix +
- samples[3][i] * surround_mix;
+ v1 = samples[1][i] * center_mix +
+ samples[2][i] * front_mix +
+ samples[4][i] * surround_mix;
- v1 = samples[1][i] * center_mix +
- samples[2][i] * front_mix +
- samples[4][i] * surround_mix;
+ samples[0][i] = v0;
+ samples[1][i] = v1;
+ }
+ }
- samples[0][i] = v0;
- samples[1][i] = v1;
- }
- } else if (in_ch == 5 && out_ch == 1 &&
- matrix_cmp[0][0] == matrix_cmp[0][2] &&
- matrix_cmp[0][3] == matrix_cmp[0][4]) {
- float front_mix = matrix[0][0];
- float center_mix = matrix[0][1];
- float surround_mix = matrix[0][3];
+ static void ac3_downmix_5_to_1_symmetric_c(float **samples, float **matrix,
+ int len)
+ {
+ int i;
+ float front_mix = matrix[0][0];
+ float center_mix = matrix[0][1];
+ float surround_mix = matrix[0][3];
- for (i = 0; i < len; i++) {
- samples[0][i] = samples[0][i] * front_mix +
- samples[1][i] * center_mix +
- samples[2][i] * front_mix +
- samples[3][i] * surround_mix +
- samples[4][i] * surround_mix;
- }
- } else if (out_ch == 2) {
+ for (i = 0; i < len; i++) {
+ samples[0][i] = samples[0][i] * front_mix +
+ samples[1][i] * center_mix +
+ samples[2][i] * front_mix +
+ samples[3][i] * surround_mix +
+ samples[4][i] * surround_mix;
+ }
+ }
+
+ static void ac3_downmix_c(float **samples, float **matrix,
+ int out_ch, int in_ch, int len)
+ {
+ int i, j;
+ float v0, v1;
+
+ if (out_ch == 2) {
for (i = 0; i < len; i++) {
v0 = v1 = 0.0f;
for (j = 0; j < in_ch; j++) {
}
}
+static void ac3_downmix_c_fixed(int32_t **samples, int16_t **matrix,
+ int out_ch, int in_ch, int len)
+{
+ int i, j;
+ int64_t v0, v1;
+ if (out_ch == 2) {
+ for (i = 0; i < len; i++) {
+ v0 = v1 = 0;
+ for (j = 0; j < in_ch; j++) {
+ v0 += (int64_t)samples[j][i] * matrix[0][j];
+ v1 += (int64_t)samples[j][i] * matrix[1][j];
+ }
+ samples[0][i] = (v0+2048)>>12;
+ samples[1][i] = (v1+2048)>>12;
+ }
+ } else if (out_ch == 1) {
+ for (i = 0; i < len; i++) {
+ v0 = 0;
+ for (j = 0; j < in_ch; j++)
+ v0 += (int64_t)samples[j][i] * matrix[0][j];
+ samples[0][i] = (v0+2048)>>12;
+ }
+ }
+}
+
++void ff_ac3dsp_downmix_fixed(AC3DSPContext *c, int32_t **samples, int16_t **matrix,
++ int out_ch, int in_ch, int len)
++{
++ if (c->downmix_fixed)
++ c->downmix_fixed(samples, matrix, len);
++ else
++ ac3_downmix_c_fixed(samples, matrix, out_ch, in_ch, len);
++}
++
static void apply_window_int16_c(int16_t *output, const int16_t *input,
const int16_t *window, unsigned int len)
{
c->update_bap_counts = ac3_update_bap_counts_c;
c->compute_mantissa_size = ac3_compute_mantissa_size_c;
c->extract_exponents = ac3_extract_exponents_c;
- c->downmix = ac3_downmix_c;
- c->downmix_fixed = ac3_downmix_c_fixed;
+ c->sum_square_butterfly_int32 = ac3_sum_square_butterfly_int32_c;
+ c->sum_square_butterfly_float = ac3_sum_square_butterfly_float_c;
+ c->in_channels = 0;
+ c->out_channels = 0;
+ c->downmix = NULL;
++ c->downmix_fixed = NULL;
c->apply_window_int16 = apply_window_int16_c;
if (ARCH_ARM)
void (*extract_exponents)(uint8_t *exp, int32_t *coef, int nb_coefs);
- void (*downmix)(float **samples, float **matrix, int out_ch,
- int in_ch, int len);
-
- void (*downmix_fixed)(int32_t **samples, int16_t **matrix, int out_ch,
- int in_ch, int len);
+ void (*sum_square_butterfly_int32)(int64_t sum[4], const int32_t *coef0,
+ const int32_t *coef1, int len);
+
+ void (*sum_square_butterfly_float)(float sum[4], const float *coef0,
+ const float *coef1, int len);
+
+ int out_channels;
+ int in_channels;
+ void (*downmix)(float **samples, float **matrix, int len);
++ void (*downmix_fixed)(int32_t **samples, int16_t **matrix, int len);
/**
* Apply symmetric window in 16-bit fixed-point.
void ff_ac3dsp_init (AC3DSPContext *c, int bit_exact);
void ff_ac3dsp_init_arm(AC3DSPContext *c, int bit_exact);
void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact);
+void ff_ac3dsp_init_mips(AC3DSPContext *c, int bit_exact);
+ void ff_ac3dsp_downmix(AC3DSPContext *c, float **samples, float **matrix,
+ int out_ch, int in_ch, int len);
++void ff_ac3dsp_downmix_fixed(AC3DSPContext *c, int32_t **samples, int16_t **matrix,
++ int out_ch, int in_ch, int len);
++
+ void ff_ac3dsp_set_downmix_x86(AC3DSPContext *c);
+
#endif /* AVCODEC_AC3DSP_H */
MMX-OBJS-$(CONFIG_VC1DSP) += x86/vc1dsp_mmx.o
# decoders/encoders
-MMX-OBJS-$(CONFIG_MPEG4_DECODER) += x86/xvididct_mmx.o \
- x86/xvididct_sse2.o
+MMX-OBJS-$(CONFIG_SNOW_DECODER) += x86/snowdsp.o
+MMX-OBJS-$(CONFIG_SNOW_ENCODER) += x86/snowdsp.o
# subsystems
- YASM-OBJS-$(CONFIG_AC3DSP) += x86/ac3dsp.o
+ YASM-OBJS-$(CONFIG_AC3DSP) += x86/ac3dsp.o \
+ x86/ac3dsp_downmix.o
YASM-OBJS-$(CONFIG_AUDIODSP) += x86/audiodsp.o
+YASM-OBJS-$(CONFIG_BLOCKDSP) += x86/blockdsp.o
YASM-OBJS-$(CONFIG_BSWAPDSP) += x86/bswapdsp.o
YASM-OBJS-$(CONFIG_DCT) += x86/dct32.o
YASM-OBJS-$(CONFIG_FFT) += x86/fft.o
--- /dev/null
-;* This file is part of Libav.
+ ;*****************************************************************************
+ ;* x86-optimized AC-3 downmixing
+ ;* Copyright (c) 2012 Justin Ruggles
+ ;*
-;* Libav is free software; you can redistribute it and/or
++;* This file is part of FFmpeg.
+ ;*
-;* Libav is distributed in the hope that it will be useful,
++;* 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.
+ ;*
-;* License along with Libav; if not, write to the Free Software
++;* 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
+ ;******************************************************************************
+
+ ;******************************************************************************
+ ;* This is based on the channel mixing asm in libavresample, but it is
+ ;* simplified for only float coefficients and only 3 to 6 channels.
+ ;******************************************************************************
+
+ %include "libavutil/x86/x86util.asm"
+
+ SECTION .text
+
+ ;-----------------------------------------------------------------------------
+ ; functions to downmix from 3 to 6 channels to mono or stereo
+ ; void ff_ac3_downmix_*(float **samples, float **matrix, int len);
+ ;-----------------------------------------------------------------------------
+
+ %macro AC3_DOWNMIX 2 ; %1 = in channels, %2 = out channels
+ ; define some names to make the code clearer
+ %assign in_channels %1
+ %assign out_channels %2
+ %assign stereo out_channels - 1
+
+ ; determine how many matrix elements must go on the stack vs. mmregs
+ %assign matrix_elements in_channels * out_channels
+ %if stereo
+ %assign needed_mmregs 4
+ %else
+ %assign needed_mmregs 3
+ %endif
+ %assign matrix_elements_mm num_mmregs - needed_mmregs
+ %if matrix_elements < matrix_elements_mm
+ %assign matrix_elements_mm matrix_elements
+ %endif
+ %assign total_mmregs needed_mmregs+matrix_elements_mm
+ %if matrix_elements_mm < matrix_elements
+ %assign matrix_elements_stack matrix_elements - matrix_elements_mm
+ %else
+ %assign matrix_elements_stack 0
+ %endif
+
+ cglobal ac3_downmix_%1_to_%2, 3,in_channels+1,total_mmregs,0-matrix_elements_stack*mmsize, src0, src1, len, src2, src3, src4, src5
+
+ ; load matrix pointers
+ %define matrix0q r1q
+ %define matrix1q r3q
+ %if stereo
+ mov matrix1q, [matrix0q+gprsize]
+ %endif
+ mov matrix0q, [matrix0q]
+
+ ; define matrix coeff names
+ %assign %%i 0
+ %assign %%j needed_mmregs
+ %rep in_channels
+ %if %%i >= matrix_elements_mm
+ CAT_XDEFINE mx_stack_0_, %%i, 1
+ CAT_XDEFINE mx_0_, %%i, [rsp+(%%i-matrix_elements_mm)*mmsize]
+ %else
+ CAT_XDEFINE mx_stack_0_, %%i, 0
+ CAT_XDEFINE mx_0_, %%i, m %+ %%j
+ %assign %%j %%j+1
+ %endif
+ %assign %%i %%i+1
+ %endrep
+ %if stereo
+ %assign %%i 0
+ %rep in_channels
+ %if in_channels + %%i >= matrix_elements_mm
+ CAT_XDEFINE mx_stack_1_, %%i, 1
+ CAT_XDEFINE mx_1_, %%i, [rsp+(in_channels+%%i-matrix_elements_mm)*mmsize]
+ %else
+ CAT_XDEFINE mx_stack_1_, %%i, 0
+ CAT_XDEFINE mx_1_, %%i, m %+ %%j
+ %assign %%j %%j+1
+ %endif
+ %assign %%i %%i+1
+ %endrep
+ %endif
+
+ ; load/splat matrix coeffs
+ %assign %%i 0
+ %rep in_channels
+ %if mx_stack_0_ %+ %%i
+ VBROADCASTSS m0, [matrix0q+4*%%i]
+ mova mx_0_ %+ %%i, m0
+ %else
+ VBROADCASTSS mx_0_ %+ %%i, [matrix0q+4*%%i]
+ %endif
+ %if stereo
+ %if mx_stack_1_ %+ %%i
+ VBROADCASTSS m0, [matrix1q+4*%%i]
+ mova mx_1_ %+ %%i, m0
+ %else
+ VBROADCASTSS mx_1_ %+ %%i, [matrix1q+4*%%i]
+ %endif
+ %endif
+ %assign %%i %%i+1
+ %endrep
+
+ lea lenq, [4*r2d]
+ ; load channel pointers to registers
+ %assign %%i 1
+ %rep (in_channels - 1)
+ mov src %+ %%i %+ q, [src0q+%%i*gprsize]
+ add src %+ %%i %+ q, lenq
+ %assign %%i %%i+1
+ %endrep
+ mov src0q, [src0q]
+ add src0q, lenq
+ neg lenq
+ .loop:
+ %if stereo || mx_stack_0_0
+ mova m0, [src0q+lenq]
+ %endif
+ %if stereo
+ mulps m1, m0, mx_1_0
+ %endif
+ %if stereo || mx_stack_0_0
+ mulps m0, m0, mx_0_0
+ %else
+ mulps m0, mx_0_0, [src0q+lenq]
+ %endif
+ %assign %%i 1
+ %rep (in_channels - 1)
+ %define src_ptr src %+ %%i %+ q
+ ; avoid extra load for mono if matrix is in a mm register
+ %if stereo || mx_stack_0_ %+ %%i
+ mova m2, [src_ptr+lenq]
+ %endif
+ %if stereo
+ FMULADD_PS m1, m2, mx_1_ %+ %%i, m1, m3
+ %endif
+ %if stereo || mx_stack_0_ %+ %%i
+ FMULADD_PS m0, m2, mx_0_ %+ %%i, m0, m2
+ %else
+ FMULADD_PS m0, mx_0_ %+ %%i, [src_ptr+lenq], m0, m1
+ %endif
+ %assign %%i %%i+1
+ %endrep
+ mova [src0q+lenq], m0
+ %if stereo
+ mova [src1q+lenq], m1
+ %endif
+
+ add lenq, mmsize
+ jl .loop
+ RET
+ %endmacro
+
+ %macro AC3_DOWNMIX_FUNCS 0
+ %assign %%i 3
+ %rep 4
+ INIT_XMM sse
+ AC3_DOWNMIX %%i, 1
+ AC3_DOWNMIX %%i, 2
+ INIT_YMM avx
+ AC3_DOWNMIX %%i, 1
+ AC3_DOWNMIX %%i, 2
+ %if HAVE_FMA3_EXTERNAL
+ INIT_YMM fma3
+ AC3_DOWNMIX %%i, 1
+ AC3_DOWNMIX %%i, 2
+ %endif
+ %assign %%i %%i+1
+ %endrep
+ %endmacro
+
+ AC3_DOWNMIX_FUNCS