mirror of https://git.ffmpeg.org/ffmpeg.git
ac3dsp: Reverse matrix in/out order in downmix()
Also use (float **) instead of (float (*)[2]). This matches the matrix layout in libavresample so we can reuse assembly code between the two. Signed-off-by: Diego Biurrun <diego@biurrun.de>
This commit is contained in:
parent
8ea35af762
commit
43717469f9
|
@ -332,47 +332,57 @@ static int parse_frame_header(AC3DecodeContext *s)
|
|||
* Set stereo downmixing coefficients based on frame header info.
|
||||
* reference: Section 7.8.2 Downmixing Into Two Channels
|
||||
*/
|
||||
static void set_downmix_coeffs(AC3DecodeContext *s)
|
||||
static int set_downmix_coeffs(AC3DecodeContext *s)
|
||||
{
|
||||
int i;
|
||||
float cmix = gain_levels[s-> center_mix_level];
|
||||
float smix = gain_levels[s->surround_mix_level];
|
||||
float norm0, norm1;
|
||||
|
||||
if (!s->downmix_coeffs[0]) {
|
||||
s->downmix_coeffs[0] = av_malloc(2 * AC3_MAX_CHANNELS *
|
||||
sizeof(**s->downmix_coeffs));
|
||||
if (!s->downmix_coeffs[0])
|
||||
return AVERROR(ENOMEM);
|
||||
s->downmix_coeffs[1] = s->downmix_coeffs[0] + AC3_MAX_CHANNELS;
|
||||
}
|
||||
|
||||
for (i = 0; i < s->fbw_channels; i++) {
|
||||
s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
|
||||
s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
|
||||
s->downmix_coeffs[0][i] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
|
||||
s->downmix_coeffs[1][i] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
|
||||
}
|
||||
if (s->channel_mode > 1 && s->channel_mode & 1) {
|
||||
s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
|
||||
s->downmix_coeffs[0][1] = s->downmix_coeffs[1][1] = cmix;
|
||||
}
|
||||
if (s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
|
||||
int nf = s->channel_mode - 2;
|
||||
s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
|
||||
s->downmix_coeffs[0][nf] = s->downmix_coeffs[1][nf] = smix * LEVEL_MINUS_3DB;
|
||||
}
|
||||
if (s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
|
||||
int nf = s->channel_mode - 4;
|
||||
s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
|
||||
s->downmix_coeffs[0][nf] = s->downmix_coeffs[1][nf+1] = smix;
|
||||
}
|
||||
|
||||
/* renormalize */
|
||||
norm0 = norm1 = 0.0;
|
||||
for (i = 0; i < s->fbw_channels; i++) {
|
||||
norm0 += s->downmix_coeffs[i][0];
|
||||
norm1 += s->downmix_coeffs[i][1];
|
||||
norm0 += s->downmix_coeffs[0][i];
|
||||
norm1 += s->downmix_coeffs[1][i];
|
||||
}
|
||||
norm0 = 1.0f / norm0;
|
||||
norm1 = 1.0f / norm1;
|
||||
for (i = 0; i < s->fbw_channels; i++) {
|
||||
s->downmix_coeffs[i][0] *= norm0;
|
||||
s->downmix_coeffs[i][1] *= norm1;
|
||||
s->downmix_coeffs[0][i] *= norm0;
|
||||
s->downmix_coeffs[1][i] *= norm1;
|
||||
}
|
||||
|
||||
if (s->output_mode == AC3_CHMODE_MONO) {
|
||||
for (i = 0; i < s->fbw_channels; i++)
|
||||
s->downmix_coeffs[i][0] = (s->downmix_coeffs[i][0] +
|
||||
s->downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
|
||||
s->downmix_coeffs[0][i] = (s->downmix_coeffs[0][i] +
|
||||
s->downmix_coeffs[1][i]) * LEVEL_MINUS_3DB;
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
/**
|
||||
|
@ -1447,7 +1457,10 @@ static int ac3_decode_frame(AVCodecContext * avctx, void *data,
|
|||
/* set downmixing coefficients if needed */
|
||||
if (s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
|
||||
s->fbw_channels == s->out_channels)) {
|
||||
set_downmix_coeffs(s);
|
||||
if ((ret = set_downmix_coeffs(s)) < 0) {
|
||||
av_log(avctx, AV_LOG_ERROR, "error setting downmix coeffs\n");
|
||||
return ret;
|
||||
}
|
||||
}
|
||||
} else if (!s->channels) {
|
||||
av_log(avctx, AV_LOG_ERROR, "unable to determine channel mode\n");
|
||||
|
@ -1566,6 +1579,7 @@ static av_cold int ac3_decode_end(AVCodecContext *avctx)
|
|||
AC3DecodeContext *s = avctx->priv_data;
|
||||
ff_mdct_end(&s->imdct_512);
|
||||
ff_mdct_end(&s->imdct_256);
|
||||
av_freep(&s->downmix_coeffs[0]);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
|
|
@ -147,7 +147,7 @@ typedef struct AC3DecodeContext {
|
|||
int fbw_channels; ///< number of full-bandwidth channels
|
||||
int channels; ///< number of total channels
|
||||
int lfe_ch; ///< index of LFE channel
|
||||
float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
|
||||
float *downmix_coeffs[2]; ///< stereo downmix coefficients
|
||||
int downmixed; ///< indicates if coeffs are currently downmixed
|
||||
int output_mode; ///< output channel configuration
|
||||
int out_channels; ///< number of output channels
|
||||
|
|
|
@ -171,7 +171,7 @@ static void ac3_extract_exponents_c(uint8_t *exp, int32_t *coef, int nb_coefs)
|
|||
}
|
||||
}
|
||||
|
||||
static void ac3_downmix_c(float **samples, float (*matrix)[2],
|
||||
static void ac3_downmix_c(float **samples, float **matrix,
|
||||
int out_ch, int in_ch, int len)
|
||||
{
|
||||
int i, j;
|
||||
|
@ -180,8 +180,8 @@ static void ac3_downmix_c(float **samples, float (*matrix)[2],
|
|||
for (i = 0; i < len; i++) {
|
||||
v0 = v1 = 0.0f;
|
||||
for (j = 0; j < in_ch; j++) {
|
||||
v0 += samples[j][i] * matrix[j][0];
|
||||
v1 += samples[j][i] * matrix[j][1];
|
||||
v0 += samples[j][i] * matrix[0][j];
|
||||
v1 += samples[j][i] * matrix[1][j];
|
||||
}
|
||||
samples[0][i] = v0;
|
||||
samples[1][i] = v1;
|
||||
|
@ -190,7 +190,7 @@ static void ac3_downmix_c(float **samples, float (*matrix)[2],
|
|||
for (i = 0; i < len; i++) {
|
||||
v0 = 0.0f;
|
||||
for (j = 0; j < in_ch; j++)
|
||||
v0 += samples[j][i] * matrix[j][0];
|
||||
v0 += samples[j][i] * matrix[0][j];
|
||||
samples[0][i] = v0;
|
||||
}
|
||||
}
|
||||
|
|
|
@ -126,7 +126,7 @@ typedef struct AC3DSPContext {
|
|||
|
||||
void (*extract_exponents)(uint8_t *exp, int32_t *coef, int nb_coefs);
|
||||
|
||||
void (*downmix)(float **samples, float (*matrix)[2], int out_ch,
|
||||
void (*downmix)(float **samples, float **matrix, int out_ch,
|
||||
int in_ch, int len);
|
||||
|
||||
/**
|
||||
|
|
|
@ -71,8 +71,8 @@ void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
|
|||
#define MIX5(mono, stereo) \
|
||||
__asm__ volatile ( \
|
||||
"movss 0(%1), %%xmm5 \n" \
|
||||
"movss 8(%1), %%xmm6 \n" \
|
||||
"movss 24(%1), %%xmm7 \n" \
|
||||
"movss 4(%1), %%xmm6 \n" \
|
||||
"movss 12(%1), %%xmm7 \n" \
|
||||
"shufps $0, %%xmm5, %%xmm5 \n" \
|
||||
"shufps $0, %%xmm6, %%xmm6 \n" \
|
||||
"shufps $0, %%xmm7, %%xmm7 \n" \
|
||||
|
@ -97,7 +97,7 @@ void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
|
|||
"add $16, %0 \n" \
|
||||
"jl 1b \n" \
|
||||
: "+&r"(i) \
|
||||
: "r"(matrix), \
|
||||
: "r"(matrix[0]), \
|
||||
"r"(samples[0] + len), \
|
||||
"r"(samples[1] + len), \
|
||||
"r"(samples[2] + len), \
|
||||
|
@ -141,22 +141,22 @@ void ff_apply_window_int16_ssse3_atom(int16_t *output, const int16_t *input,
|
|||
: "memory" \
|
||||
);
|
||||
|
||||
static void ac3_downmix_sse(float **samples, float (*matrix)[2],
|
||||
static void ac3_downmix_sse(float **samples, float **matrix,
|
||||
int out_ch, int in_ch, int len)
|
||||
{
|
||||
int (*matrix_cmp)[2] = (int(*)[2])matrix;
|
||||
int **matrix_cmp = (int **)matrix;
|
||||
intptr_t i, j, k, m;
|
||||
|
||||
i = -len * sizeof(float);
|
||||
if (in_ch == 5 && out_ch == 2 &&
|
||||
!(matrix_cmp[0][1] | matrix_cmp[2][0] |
|
||||
matrix_cmp[3][1] | matrix_cmp[4][0] |
|
||||
(matrix_cmp[1][0] ^ matrix_cmp[1][1]) |
|
||||
(matrix_cmp[0][0] ^ matrix_cmp[2][1]))) {
|
||||
!(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]))) {
|
||||
MIX5(IF0, IF1);
|
||||
} else if (in_ch == 5 && out_ch == 1 &&
|
||||
matrix_cmp[0][0] == matrix_cmp[2][0] &&
|
||||
matrix_cmp[3][0] == matrix_cmp[4][0]) {
|
||||
matrix_cmp[0][0] == matrix_cmp[0][2] &&
|
||||
matrix_cmp[0][3] == matrix_cmp[0][4]) {
|
||||
MIX5(IF1, IF0);
|
||||
} else {
|
||||
DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4];
|
||||
|
@ -166,18 +166,20 @@ static void ac3_downmix_sse(float **samples, float (*matrix)[2],
|
|||
samp[j] = samples[j] + len;
|
||||
|
||||
j = 2 * in_ch * sizeof(float);
|
||||
k = in_ch * sizeof(float);
|
||||
__asm__ volatile (
|
||||
"1: \n"
|
||||
"sub $4, %1 \n"
|
||||
"sub $8, %0 \n"
|
||||
"movss (%2, %0), %%xmm4 \n"
|
||||
"movss 4(%2, %0), %%xmm5 \n"
|
||||
"movss (%3, %1), %%xmm4 \n"
|
||||
"movss (%4, %1), %%xmm5 \n"
|
||||
"shufps $0, %%xmm4, %%xmm4 \n"
|
||||
"shufps $0, %%xmm5, %%xmm5 \n"
|
||||
"movaps %%xmm4, (%1, %0, 4) \n"
|
||||
"movaps %%xmm5, 16(%1, %0, 4) \n"
|
||||
"movaps %%xmm4, (%2, %0, 4) \n"
|
||||
"movaps %%xmm5, 16(%2, %0, 4) \n"
|
||||
"jg 1b \n"
|
||||
: "+&r"(j)
|
||||
: "r"(matrix_simd), "r"(matrix)
|
||||
: "+&r"(j), "+&r"(k)
|
||||
: "r"(matrix_simd), "r"(matrix[0]), "r"(matrix[1])
|
||||
: "memory"
|
||||
);
|
||||
if (out_ch == 2) {
|
||||
|
|
Loading…
Reference in New Issue