ref: 447e5b8fc631ed40b7ad3715de2f8cfe26580960
parent: ef0d5f15c767a03a80bf6cea9c6e6e7cfe473168
author: Jean-Marc Valin <[email protected]>
date: Tue Apr 22 06:13:25 EDT 2008
Separated the pre-rotate from the window+shuffling and further simplified everything.
--- a/libcelt/mdct.c
+++ b/libcelt/mdct.c
@@ -86,6 +86,13 @@
celt_free(l->trig);
}
+/* Only divide by half if float. In fixed-point, it's included in the shift */
+#ifdef FIXED_POINT
+#define FL_HALF(x) (x)
+#else
+#define FL_HALF(x) (.5f*(x))
+#endif
+
void mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * restrict out, const celt_word16_t *window, int overlap)
{int i;
@@ -98,65 +105,54 @@
ALLOC(f, N2, kiss_fft_scalar);
/* Consider the input to be compused of four blocks: [a, b, c, d] */
- /* Shuffle, fold, pre-rotate (part 1) */
+ /* Window, shuffle, fold */
{/* Temp pointers to make it really clear to the compiler what we're doing */
- const kiss_fft_scalar * restrict xp1 = in+overlap/2;
- const kiss_fft_scalar * restrict xp2 = in+N2-1+overlap/2;
+ const kiss_fft_scalar * restrict xp1 = in+(overlap>>1);
+ const kiss_fft_scalar * restrict xp2 = in+N2-1+(overlap>>1);
kiss_fft_scalar * restrict yp = out;
- kiss_fft_scalar *t = &l->trig[0];
- const celt_word16_t * restrict wp1 = window+overlap/2;
- const celt_word16_t * restrict wp2 = window+overlap/2-1;
- for(i=0;i<overlap/4;i++)
+ const celt_word16_t * restrict wp1 = window+(overlap>>1);
+ const celt_word16_t * restrict wp2 = window+(overlap>>1)-1;
+ for(i=0;i<(overlap>>2);i++)
{- kiss_fft_scalar re, im;
/* Real part arranged as -d-cR, Imag part arranged as -b+aR*/
- re = -(MULT16_32_Q16(*wp2, xp1[N2]) + MULT16_32_Q16(*wp1,*xp2));
- im = -(MULT16_32_Q16(*wp1, *xp1) - MULT16_32_Q16(*wp2, xp2[-N2]));
-#ifndef FIXED_POINT
- re *= .5; im *= .5;
-#endif
+ *yp++ = -FL_HALF(MULT16_32_Q16(*wp2, xp1[N2]) + MULT16_32_Q16(*wp1,*xp2));
+ *yp++ = -FL_HALF(MULT16_32_Q16(*wp1, *xp1) - MULT16_32_Q16(*wp2, xp2[-N2]));
xp1+=2;
xp2-=2;
wp1+=2;
wp2-=2;
- /* We could remove the HALF32 above and just use MULT16_32_Q16 below
- (MIXED_PRECISION only) */
- *yp++ = S_MUL(re,t[0]) - S_MUL(im,t[N4]);
- *yp++ = S_MUL(im,t[0]) + S_MUL(re,t[N4]);
- t++;
}
wp1 = window;
wp2 = window+overlap-1;
- for(;i<N4-overlap/4;i++)
+ for(;i<N4-(overlap>>2);i++)
{- kiss_fft_scalar re, im;
/* Real part arranged as a-bR, Imag part arranged as -c-dR */
- re = -HALF32(*xp2);
- im = -HALF32(*xp1);
+ *yp++ = -HALF32(*xp2);
+ *yp++ = -HALF32(*xp1);
xp1+=2;
xp2-=2;
- /* We could remove the HALF32 above and just use MULT16_32_Q16 below
- (MIXED_PRECISION only) */
- *yp++ = S_MUL(re,t[0]) - S_MUL(im,t[N4]);
- *yp++ = S_MUL(im,t[0]) + S_MUL(re,t[N4]);
- t++;
}
for(;i<N4;i++)
{- kiss_fft_scalar re, im;
/* Real part arranged as a-bR, Imag part arranged as -c-dR */
- re = (MULT16_32_Q16(*wp1, xp1[-N2]) - MULT16_32_Q16(*wp2, *xp2));
- im = -(MULT16_32_Q16(*wp2, *xp1) + MULT16_32_Q16(*wp1, xp2[N2]));
-#ifndef FIXED_POINT
- re *= .5; im *= .5;
-#endif
+ *yp++ = FL_HALF(MULT16_32_Q16(*wp1, xp1[-N2]) - MULT16_32_Q16(*wp2, *xp2));
+ *yp++ = -FL_HALF(MULT16_32_Q16(*wp2, *xp1) + MULT16_32_Q16(*wp1, xp2[N2]));
xp1+=2;
xp2-=2;
wp1+=2;
wp2-=2;
- /* We could remove the HALF32 above and just use MULT16_32_Q16 below
- (MIXED_PRECISION only) */
+ }
+ }
+ /* Pre-rotation */
+ {+ kiss_fft_scalar * restrict yp = out;
+ kiss_fft_scalar *t = &l->trig[0];
+ for(i=0;i<N4;i++)
+ {+ kiss_fft_scalar re, im;
+ re = yp[0];
+ im = yp[1];
*yp++ = S_MUL(re,t[0]) - S_MUL(im,t[N4]);
*yp++ = S_MUL(im,t[0]) + S_MUL(re,t[N4]);
t++;