Signed-off-by: Mans Rullgard <mans@mansr.com>tags/n1.0
@@ -620,34 +620,6 @@ static inline void do_imdct(AC3DecodeContext *s, int channels) | |||
} | |||
} | |||
/** | |||
* Downmix the output to mono or stereo. | |||
*/ | |||
void ff_ac3_downmix_c(float (*samples)[256], float (*matrix)[2], | |||
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++) { | |||
v0 += samples[j][i] * matrix[j][0]; | |||
v1 += samples[j][i] * matrix[j][1]; | |||
} | |||
samples[0][i] = v0; | |||
samples[1][i] = v1; | |||
} | |||
} else if (out_ch == 1) { | |||
for (i = 0; i < len; i++) { | |||
v0 = 0.0f; | |||
for (j = 0; j < in_ch; j++) | |||
v0 += samples[j][i] * matrix[j][0]; | |||
samples[0][i] = v0; | |||
} | |||
} | |||
} | |||
/** | |||
* Upmix delay samples from stereo to original channel layout. | |||
*/ | |||
@@ -1266,19 +1238,19 @@ static int decode_audio_block(AC3DecodeContext *s, int blk) | |||
do_imdct(s, s->channels); | |||
if (downmix_output) { | |||
s->dsp.ac3_downmix(s->output, s->downmix_coeffs, | |||
s->out_channels, s->fbw_channels, 256); | |||
s->ac3dsp.downmix(s->output, s->downmix_coeffs, | |||
s->out_channels, s->fbw_channels, 256); | |||
} | |||
} else { | |||
if (downmix_output) { | |||
s->dsp.ac3_downmix(s->transform_coeffs + 1, s->downmix_coeffs, | |||
s->out_channels, s->fbw_channels, 256); | |||
s->ac3dsp.downmix(s->transform_coeffs + 1, s->downmix_coeffs, | |||
s->out_channels, s->fbw_channels, 256); | |||
} | |||
if (downmix_output && !s->downmixed) { | |||
s->downmixed = 1; | |||
s->dsp.ac3_downmix(s->delay, s->downmix_coeffs, s->out_channels, | |||
s->fbw_channels, 128); | |||
s->ac3dsp.downmix(s->delay, s->downmix_coeffs, s->out_channels, | |||
s->fbw_channels, 128); | |||
} | |||
do_imdct(s, s->out_channels); | |||
@@ -221,9 +221,6 @@ int ff_eac3_parse_header(AC3DecodeContext *s); | |||
*/ | |||
void ff_eac3_decode_transform_coeffs_aht_ch(AC3DecodeContext *s, int ch); | |||
void ff_ac3_downmix_c(float (*samples)[256], float (*matrix)[2], | |||
int out_ch, int in_ch, int len); | |||
/** | |||
* Apply spectral extension to each channel by copying lower frequency | |||
* coefficients to higher frequency bins and applying side information to | |||
@@ -171,6 +171,31 @@ static void ac3_extract_exponents_c(uint8_t *exp, int32_t *coef, int nb_coefs) | |||
} | |||
} | |||
static void ac3_downmix_c(float (*samples)[256], float (*matrix)[2], | |||
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++) { | |||
v0 += samples[j][i] * matrix[j][0]; | |||
v1 += samples[j][i] * matrix[j][1]; | |||
} | |||
samples[0][i] = v0; | |||
samples[1][i] = v1; | |||
} | |||
} else if (out_ch == 1) { | |||
for (i = 0; i < len; i++) { | |||
v0 = 0.0f; | |||
for (j = 0; j < in_ch; j++) | |||
v0 += samples[j][i] * matrix[j][0]; | |||
samples[0][i] = v0; | |||
} | |||
} | |||
} | |||
av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact) | |||
{ | |||
c->ac3_exponent_min = ac3_exponent_min_c; | |||
@@ -182,6 +207,7 @@ av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact) | |||
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; | |||
if (ARCH_ARM) | |||
ff_ac3dsp_init_arm(c, bit_exact); | |||
@@ -125,6 +125,9 @@ typedef struct AC3DSPContext { | |||
int (*compute_mantissa_size)(uint16_t mant_cnt[6][16]); | |||
void (*extract_exponents)(uint8_t *exp, int32_t *coef, int nb_coefs); | |||
void (*downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, | |||
int in_ch, int len); | |||
} AC3DSPContext; | |||
void ff_ac3dsp_init (AC3DSPContext *c, int bit_exact); | |||
@@ -36,7 +36,6 @@ | |||
#include "mathops.h" | |||
#include "mpegvideo.h" | |||
#include "config.h" | |||
#include "ac3dec.h" | |||
#include "vorbis.h" | |||
uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP] = {0, }; | |||
@@ -2862,9 +2861,6 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx) | |||
#if CONFIG_VORBIS_DECODER | |||
c->vorbis_inverse_coupling = ff_vorbis_inverse_coupling; | |||
#endif | |||
#if CONFIG_AC3_DECODER | |||
c->ac3_downmix = ff_ac3_downmix_c; | |||
#endif | |||
c->vector_fmul_reverse = vector_fmul_reverse_c; | |||
c->vector_fmul_add = vector_fmul_add_c; | |||
@@ -374,7 +374,6 @@ typedef struct DSPContext { | |||
/* assume len is a multiple of 4, and arrays are 16-byte aligned */ | |||
void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); | |||
void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len); | |||
/* assume len is a multiple of 16, and arrays are 32-byte aligned */ | |||
void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len); | |||
/* assume len is a multiple of 8, and src arrays are 16-byte aligned */ | |||
@@ -19,9 +19,11 @@ | |||
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |||
*/ | |||
#include "libavutil/mem.h" | |||
#include "libavutil/x86/asm.h" | |||
#include "libavutil/x86/cpu.h" | |||
#include "dsputil_mmx.h" | |||
#include "libavcodec/ac3.h" | |||
#include "libavcodec/ac3dsp.h" | |||
extern void ff_ac3_exponent_min_mmx (uint8_t *exp, int num_reuse_blocks, int nb_coefs); | |||
@@ -49,6 +51,119 @@ extern void ff_ac3_extract_exponents_3dnow(uint8_t *exp, int32_t *coef, int nb_c | |||
extern void ff_ac3_extract_exponents_sse2 (uint8_t *exp, int32_t *coef, int nb_coefs); | |||
extern void ff_ac3_extract_exponents_ssse3(uint8_t *exp, int32_t *coef, int nb_coefs); | |||
#if HAVE_SSE_INLINE | |||
#define IF1(x) x | |||
#define IF0(x) | |||
#define MIX5(mono, stereo) \ | |||
__asm__ volatile ( \ | |||
"movss 0(%2), %%xmm5 \n" \ | |||
"movss 8(%2), %%xmm6 \n" \ | |||
"movss 24(%2), %%xmm7 \n" \ | |||
"shufps $0, %%xmm5, %%xmm5 \n" \ | |||
"shufps $0, %%xmm6, %%xmm6 \n" \ | |||
"shufps $0, %%xmm7, %%xmm7 \n" \ | |||
"1: \n" \ | |||
"movaps (%0, %1), %%xmm0 \n" \ | |||
"movaps 0x400(%0, %1), %%xmm1 \n" \ | |||
"movaps 0x800(%0, %1), %%xmm2 \n" \ | |||
"movaps 0xc00(%0, %1), %%xmm3 \n" \ | |||
"movaps 0x1000(%0, %1), %%xmm4 \n" \ | |||
"mulps %%xmm5, %%xmm0 \n" \ | |||
"mulps %%xmm6, %%xmm1 \n" \ | |||
"mulps %%xmm5, %%xmm2 \n" \ | |||
"mulps %%xmm7, %%xmm3 \n" \ | |||
"mulps %%xmm7, %%xmm4 \n" \ | |||
stereo("addps %%xmm1, %%xmm0 \n") \ | |||
"addps %%xmm1, %%xmm2 \n" \ | |||
"addps %%xmm3, %%xmm0 \n" \ | |||
"addps %%xmm4, %%xmm2 \n" \ | |||
mono("addps %%xmm2, %%xmm0 \n") \ | |||
"movaps %%xmm0, (%0, %1) \n" \ | |||
stereo("movaps %%xmm2, 0x400(%0, %1) \n") \ | |||
"add $16, %0 \n" \ | |||
"jl 1b \n" \ | |||
: "+&r"(i) \ | |||
: "r"(samples[0] + len), "r"(matrix) \ | |||
: XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \ | |||
"%xmm4", "%xmm5", "%xmm6", "%xmm7",) \ | |||
"memory" \ | |||
); | |||
#define MIX_MISC(stereo) \ | |||
__asm__ volatile ( \ | |||
"1: \n" \ | |||
"movaps (%3, %0), %%xmm0 \n" \ | |||
stereo("movaps %%xmm0, %%xmm1 \n") \ | |||
"mulps %%xmm4, %%xmm0 \n" \ | |||
stereo("mulps %%xmm5, %%xmm1 \n") \ | |||
"lea 1024(%3, %0), %1 \n" \ | |||
"mov %5, %2 \n" \ | |||
"2: \n" \ | |||
"movaps (%1), %%xmm2 \n" \ | |||
stereo("movaps %%xmm2, %%xmm3 \n") \ | |||
"mulps (%4, %2), %%xmm2 \n" \ | |||
stereo("mulps 16(%4, %2), %%xmm3 \n") \ | |||
"addps %%xmm2, %%xmm0 \n" \ | |||
stereo("addps %%xmm3, %%xmm1 \n") \ | |||
"add $1024, %1 \n" \ | |||
"add $32, %2 \n" \ | |||
"jl 2b \n" \ | |||
"movaps %%xmm0, (%3, %0) \n" \ | |||
stereo("movaps %%xmm1, 1024(%3, %0) \n") \ | |||
"add $16, %0 \n" \ | |||
"jl 1b \n" \ | |||
: "+&r"(i), "=&r"(j), "=&r"(k) \ | |||
: "r"(samples[0] + len), "r"(matrix_simd + in_ch), \ | |||
"g"((intptr_t) - 32 * (in_ch - 1)) \ | |||
: "memory" \ | |||
); | |||
static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], | |||
int out_ch, int in_ch, int len) | |||
{ | |||
int (*matrix_cmp)[2] = (int(*)[2])matrix; | |||
intptr_t i, j, k; | |||
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]))) { | |||
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]) { | |||
MIX5(IF1, IF0); | |||
} else { | |||
DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4]; | |||
j = 2 * in_ch * sizeof(float); | |||
__asm__ volatile ( | |||
"1: \n" | |||
"sub $8, %0 \n" | |||
"movss (%2, %0), %%xmm4 \n" | |||
"movss 4(%2, %0), %%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" | |||
"jg 1b \n" | |||
: "+&r"(j) | |||
: "r"(matrix_simd), "r"(matrix) | |||
: "memory" | |||
); | |||
if (out_ch == 2) { | |||
MIX_MISC(IF1); | |||
} else { | |||
MIX_MISC(IF0); | |||
} | |||
} | |||
} | |||
#endif /* HAVE_SSE_INLINE */ | |||
av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact) | |||
{ | |||
int mm_flags = av_get_cpu_flags(); | |||
@@ -89,4 +204,8 @@ av_cold void ff_ac3dsp_init_x86(AC3DSPContext *c, int bit_exact) | |||
c->extract_exponents = ff_ac3_extract_exponents_ssse3; | |||
} | |||
} | |||
if (INLINE_SSE(mm_flags)) { | |||
c->downmix = ac3_downmix_sse; | |||
} | |||
} |
@@ -28,7 +28,6 @@ | |||
#include "libavcodec/h264dsp.h" | |||
#include "libavcodec/mpegvideo.h" | |||
#include "libavcodec/simple_idct.h" | |||
#include "libavcodec/ac3dec.h" | |||
#include "dsputil_mmx.h" | |||
#include "idct_xvid.h" | |||
@@ -2248,115 +2247,6 @@ static void vorbis_inverse_coupling_sse(float *mag, float *ang, int blocksize) | |||
} | |||
} | |||
#define IF1(x) x | |||
#define IF0(x) | |||
#define MIX5(mono, stereo) \ | |||
__asm__ volatile ( \ | |||
"movss 0(%2), %%xmm5 \n" \ | |||
"movss 8(%2), %%xmm6 \n" \ | |||
"movss 24(%2), %%xmm7 \n" \ | |||
"shufps $0, %%xmm5, %%xmm5 \n" \ | |||
"shufps $0, %%xmm6, %%xmm6 \n" \ | |||
"shufps $0, %%xmm7, %%xmm7 \n" \ | |||
"1: \n" \ | |||
"movaps (%0, %1), %%xmm0 \n" \ | |||
"movaps 0x400(%0, %1), %%xmm1 \n" \ | |||
"movaps 0x800(%0, %1), %%xmm2 \n" \ | |||
"movaps 0xc00(%0, %1), %%xmm3 \n" \ | |||
"movaps 0x1000(%0, %1), %%xmm4 \n" \ | |||
"mulps %%xmm5, %%xmm0 \n" \ | |||
"mulps %%xmm6, %%xmm1 \n" \ | |||
"mulps %%xmm5, %%xmm2 \n" \ | |||
"mulps %%xmm7, %%xmm3 \n" \ | |||
"mulps %%xmm7, %%xmm4 \n" \ | |||
stereo("addps %%xmm1, %%xmm0 \n") \ | |||
"addps %%xmm1, %%xmm2 \n" \ | |||
"addps %%xmm3, %%xmm0 \n" \ | |||
"addps %%xmm4, %%xmm2 \n" \ | |||
mono("addps %%xmm2, %%xmm0 \n") \ | |||
"movaps %%xmm0, (%0, %1) \n" \ | |||
stereo("movaps %%xmm2, 0x400(%0, %1) \n") \ | |||
"add $16, %0 \n" \ | |||
"jl 1b \n" \ | |||
: "+&r"(i) \ | |||
: "r"(samples[0] + len), "r"(matrix) \ | |||
: XMM_CLOBBERS("%xmm0", "%xmm1", "%xmm2", "%xmm3", \ | |||
"%xmm4", "%xmm5", "%xmm6", "%xmm7",) \ | |||
"memory" \ | |||
); | |||
#define MIX_MISC(stereo) \ | |||
__asm__ volatile ( \ | |||
"1: \n" \ | |||
"movaps (%3, %0), %%xmm0 \n" \ | |||
stereo("movaps %%xmm0, %%xmm1 \n") \ | |||
"mulps %%xmm4, %%xmm0 \n" \ | |||
stereo("mulps %%xmm5, %%xmm1 \n") \ | |||
"lea 1024(%3, %0), %1 \n" \ | |||
"mov %5, %2 \n" \ | |||
"2: \n" \ | |||
"movaps (%1), %%xmm2 \n" \ | |||
stereo("movaps %%xmm2, %%xmm3 \n") \ | |||
"mulps (%4, %2), %%xmm2 \n" \ | |||
stereo("mulps 16(%4, %2), %%xmm3 \n") \ | |||
"addps %%xmm2, %%xmm0 \n" \ | |||
stereo("addps %%xmm3, %%xmm1 \n") \ | |||
"add $1024, %1 \n" \ | |||
"add $32, %2 \n" \ | |||
"jl 2b \n" \ | |||
"movaps %%xmm0, (%3, %0) \n" \ | |||
stereo("movaps %%xmm1, 1024(%3, %0) \n") \ | |||
"add $16, %0 \n" \ | |||
"jl 1b \n" \ | |||
: "+&r"(i), "=&r"(j), "=&r"(k) \ | |||
: "r"(samples[0] + len), "r"(matrix_simd + in_ch), \ | |||
"g"((intptr_t) - 32 * (in_ch - 1)) \ | |||
: "memory" \ | |||
); | |||
static void ac3_downmix_sse(float (*samples)[256], float (*matrix)[2], | |||
int out_ch, int in_ch, int len) | |||
{ | |||
int (*matrix_cmp)[2] = (int(*)[2])matrix; | |||
intptr_t i, j, k; | |||
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]))) { | |||
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]) { | |||
MIX5(IF1, IF0); | |||
} else { | |||
DECLARE_ALIGNED(16, float, matrix_simd)[AC3_MAX_CHANNELS][2][4]; | |||
j = 2 * in_ch * sizeof(float); | |||
__asm__ volatile ( | |||
"1: \n" | |||
"sub $8, %0 \n" | |||
"movss (%2, %0), %%xmm4 \n" | |||
"movss 4(%2, %0), %%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" | |||
"jg 1b \n" | |||
: "+&r"(j) | |||
: "r"(matrix_simd), "r"(matrix) | |||
: "memory" | |||
); | |||
if (out_ch == 2) { | |||
MIX_MISC(IF1); | |||
} else { | |||
MIX_MISC(IF0); | |||
} | |||
} | |||
} | |||
#if HAVE_6REGS | |||
static void vector_fmul_window_3dnowext(float *dst, const float *src0, | |||
const float *src1, const float *win, | |||
@@ -2849,7 +2739,6 @@ static void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx, int mm_flags) | |||
} | |||
c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse; | |||
c->ac3_downmix = ac3_downmix_sse; | |||
#if HAVE_6REGS | |||
c->vector_fmul_window = vector_fmul_window_sse; | |||