#include <stdint.h>
+#include "libavutil/attributes.h"
#include "libavutil/mem.h"
#include "dct32.h"
#include "mathops.h"
#include "mpegaudiodsp.h"
#include "mpegaudio.h"
-#include "mpegaudiodata.h"
#if CONFIG_FLOAT
#define RENAME(n) n##_float
#define MACS(rt, ra, rb) rt+=(ra)*(rb)
#define MULS(ra, rb) ((ra)*(rb))
+#define MULH3(x, y, s) ((s)*(y)*(x))
#define MLSS(rt, ra, rb) rt-=(ra)*(rb)
+#define MULLx(x, y, s) ((y)*(x))
+#define FIXHR(x) ((float)(x))
+#define FIXR(x) ((float)(x))
+#define SHR(a,b) ((a)*(1.0f/(1<<(b))))
#else
# define MULS(ra, rb) MUL64(ra, rb)
# define MACS(rt, ra, rb) MAC64(rt, ra, rb)
# define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
+# define MULH3(x, y, s) MULH((s)*(x), y)
+# define MULLx(x, y, s) MULL(x,y,s)
+# define SHR(a,b) ((a)>>(b))
+# define FIXR(a) ((int)((a) * FRAC_ONE + 0.5))
+# define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5))
#endif
+/** Window for MDCT. Actually only the elements in [0,17] and
+ [MDCT_BUF_SIZE/2, MDCT_BUF_SIZE/2 + 17] are actually used. The rest
+ is just to preserve alignment for SIMD implementations.
+*/
+DECLARE_ALIGNED(16, INTFLOAT, RENAME(ff_mdct_win))[8][MDCT_BUF_SIZE];
+
DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256];
#define SUM8(op, sum, w, p) \
*synth_buf_offset = offset;
}
-void av_cold RENAME(ff_mpa_synth_init)(MPA_INT *window)
+av_cold void RENAME(ff_mpa_synth_init)(MPA_INT *window)
{
int i, j;
window[512 - i] = v;
}
+
// Needed for avoiding shuffles in ASM implementations
for(i=0; i < 8; i++)
for(j=0; j < 16; j++)
for(j=0; j < 16; j++)
window[512+128+16*i+j] = window[64*i+48-j];
}
+
+av_cold void RENAME(ff_init_mpadsp_tabs)(void)
+{
+ int i, j;
+ /* compute mdct windows */
+ for (i = 0; i < 36; i++) {
+ for (j = 0; j < 4; j++) {
+ double d;
+
+ if (j == 2 && i % 3 != 1)
+ continue;
+
+ d = sin(M_PI * (i + 0.5) / 36.0);
+ if (j == 1) {
+ if (i >= 30) d = 0;
+ else if (i >= 24) d = sin(M_PI * (i - 18 + 0.5) / 12.0);
+ else if (i >= 18) d = 1;
+ } else if (j == 3) {
+ if (i < 6) d = 0;
+ else if (i < 12) d = sin(M_PI * (i - 6 + 0.5) / 12.0);
+ else if (i < 18) d = 1;
+ }
+ //merge last stage of imdct into the window coefficients
+ d *= 0.5 / cos(M_PI * (2 * i + 19) / 72);
+
+ if (j == 2)
+ RENAME(ff_mdct_win)[j][i/3] = FIXHR((d / (1<<5)));
+ else {
+ int idx = i < 18 ? i : i + (MDCT_BUF_SIZE/2 - 18);
+ RENAME(ff_mdct_win)[j][idx] = FIXHR((d / (1<<5)));
+ }
+ }
+ }
+
+ /* NOTE: we do frequency inversion adter the MDCT by changing
+ the sign of the right window coefs */
+ for (j = 0; j < 4; j++) {
+ for (i = 0; i < MDCT_BUF_SIZE; i += 2) {
+ RENAME(ff_mdct_win)[j + 4][i ] = RENAME(ff_mdct_win)[j][i ];
+ RENAME(ff_mdct_win)[j + 4][i + 1] = -RENAME(ff_mdct_win)[j][i + 1];
+ }
+ }
+}
+/* cos(pi*i/18) */
+#define C1 FIXHR(0.98480775301220805936/2)
+#define C2 FIXHR(0.93969262078590838405/2)
+#define C3 FIXHR(0.86602540378443864676/2)
+#define C4 FIXHR(0.76604444311897803520/2)
+#define C5 FIXHR(0.64278760968653932632/2)
+#define C6 FIXHR(0.5/2)
+#define C7 FIXHR(0.34202014332566873304/2)
+#define C8 FIXHR(0.17364817766693034885/2)
+
+/* 0.5 / cos(pi*(2*i+1)/36) */
+static const INTFLOAT icos36[9] = {
+ FIXR(0.50190991877167369479),
+ FIXR(0.51763809020504152469), //0
+ FIXR(0.55168895948124587824),
+ FIXR(0.61038729438072803416),
+ FIXR(0.70710678118654752439), //1
+ FIXR(0.87172339781054900991),
+ FIXR(1.18310079157624925896),
+ FIXR(1.93185165257813657349), //2
+ FIXR(5.73685662283492756461),
+};
+
+/* 0.5 / cos(pi*(2*i+1)/36) */
+static const INTFLOAT icos36h[9] = {
+ FIXHR(0.50190991877167369479/2),
+ FIXHR(0.51763809020504152469/2), //0
+ FIXHR(0.55168895948124587824/2),
+ FIXHR(0.61038729438072803416/2),
+ FIXHR(0.70710678118654752439/2), //1
+ FIXHR(0.87172339781054900991/2),
+ FIXHR(1.18310079157624925896/4),
+ FIXHR(1.93185165257813657349/4), //2
+// FIXHR(5.73685662283492756461),
+};
+
+/* using Lee like decomposition followed by hand coded 9 points DCT */
+static void imdct36(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, INTFLOAT *win)
+{
+ int i, j;
+ INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3;
+ INTFLOAT tmp[18], *tmp1, *in1;
+
+ for (i = 17; i >= 1; i--)
+ in[i] += in[i-1];
+ for (i = 17; i >= 3; i -= 2)
+ in[i] += in[i-2];
+
+ for (j = 0; j < 2; j++) {
+ tmp1 = tmp + j;
+ in1 = in + j;
+
+ t2 = in1[2*4] + in1[2*8] - in1[2*2];
+
+ t3 = in1[2*0] + SHR(in1[2*6],1);
+ t1 = in1[2*0] - in1[2*6];
+ tmp1[ 6] = t1 - SHR(t2,1);
+ tmp1[16] = t1 + t2;
+
+ t0 = MULH3(in1[2*2] + in1[2*4] , C2, 2);
+ t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1);
+ t2 = MULH3(in1[2*2] + in1[2*8] , -C4, 2);
+
+ tmp1[10] = t3 - t0 - t2;
+ tmp1[ 2] = t3 + t0 + t1;
+ tmp1[14] = t3 + t2 - t1;
+
+ tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2);
+ t2 = MULH3(in1[2*1] + in1[2*5], C1, 2);
+ t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1);
+ t0 = MULH3(in1[2*3], C3, 2);
+
+ t1 = MULH3(in1[2*1] + in1[2*7], -C5, 2);
+
+ tmp1[ 0] = t2 + t3 + t0;
+ tmp1[12] = t2 + t1 - t0;
+ tmp1[ 8] = t3 - t1 - t0;
+ }
+
+ i = 0;
+ for (j = 0; j < 4; j++) {
+ t0 = tmp[i];
+ t1 = tmp[i + 2];
+ s0 = t1 + t0;
+ s2 = t1 - t0;
+
+ t2 = tmp[i + 1];
+ t3 = tmp[i + 3];
+ s1 = MULH3(t3 + t2, icos36h[ j], 2);
+ s3 = MULLx(t3 - t2, icos36 [8 - j], FRAC_BITS);
+
+ t0 = s0 + s1;
+ t1 = s0 - s1;
+ out[(9 + j) * SBLIMIT] = MULH3(t1, win[ 9 + j], 1) + buf[4*(9 + j)];
+ out[(8 - j) * SBLIMIT] = MULH3(t1, win[ 8 - j], 1) + buf[4*(8 - j)];
+ buf[4 * ( 9 + j )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + j], 1);
+ buf[4 * ( 8 - j )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - j], 1);
+
+ t0 = s2 + s3;
+ t1 = s2 - s3;
+ out[(9 + 8 - j) * SBLIMIT] = MULH3(t1, win[ 9 + 8 - j], 1) + buf[4*(9 + 8 - j)];
+ out[ j * SBLIMIT] = MULH3(t1, win[ j], 1) + buf[4*( j)];
+ buf[4 * ( 9 + 8 - j )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 8 - j], 1);
+ buf[4 * ( j )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + j], 1);
+ i += 4;
+ }
+
+ s0 = tmp[16];
+ s1 = MULH3(tmp[17], icos36h[4], 2);
+ t0 = s0 + s1;
+ t1 = s0 - s1;
+ out[(9 + 4) * SBLIMIT] = MULH3(t1, win[ 9 + 4], 1) + buf[4*(9 + 4)];
+ out[(8 - 4) * SBLIMIT] = MULH3(t1, win[ 8 - 4], 1) + buf[4*(8 - 4)];
+ buf[4 * ( 9 + 4 )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 4], 1);
+ buf[4 * ( 8 - 4 )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - 4], 1);
+}
+
+void RENAME(ff_imdct36_blocks)(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in,
+ int count, int switch_point, int block_type)
+{
+ int j;
+ for (j=0 ; j < count; j++) {
+ /* apply window & overlap with previous buffer */
+
+ /* select window */
+ int win_idx = (switch_point && j < 2) ? 0 : block_type;
+ INTFLOAT *win = RENAME(ff_mdct_win)[win_idx + (4 & -(j & 1))];
+
+ imdct36(out, buf, in, win);
+
+ in += 18;
+ buf += ((j&3) != 3 ? 1 : (72-3));
+ out++;
+ }
+}
projects / ffmpeg / blobdiff