Commit 588d28ac authored by Måns Rullgård's avatar Måns Rullgård

Remove vestiges of radix-2 FFT

Patch (mostly) by Loren Merritt

Originally committed as revision 23957 to svn://svn.ffmpeg.org/ffmpeg/trunk
parent bf7ba153
......@@ -320,7 +320,7 @@ function ff_fft_permute_neon, export=1
push {r4,lr}
mov r12, #1
ldr r2, [r0] @ nbits
ldr r3, [r0, #20] @ tmp_buf
ldr r3, [r0, #12] @ tmp_buf
ldr r0, [r0, #8] @ revtab
lsl r12, r12, r2
mov r2, r12
......
......@@ -31,8 +31,8 @@ function ff_imdct_half_neon, export=1
push {r4-r8,lr}
mov r12, #1
ldr lr, [r0, #28] @ mdct_bits
ldr r4, [r0, #32] @ tcos
ldr lr, [r0, #20] @ mdct_bits
ldr r4, [r0, #24] @ tcos
ldr r3, [r0, #8] @ revtab
lsl r12, r12, lr @ n = 1 << nbits
lsr lr, r12, #2 @ n4 = n >> 2
......@@ -76,8 +76,8 @@ function ff_imdct_half_neon, export=1
bl ff_fft_calc_neon
mov r12, #1
ldr lr, [r4, #28] @ mdct_bits
ldr r4, [r4, #32] @ tcos
ldr lr, [r4, #20] @ mdct_bits
ldr r4, [r4, #24] @ tcos
lsl r12, r12, lr @ n = 1 << nbits
lsr lr, r12, #3 @ n8 = n >> 3
......@@ -127,7 +127,7 @@ endfunc
function ff_imdct_calc_neon, export=1
push {r4-r6,lr}
ldr r3, [r0, #28]
ldr r3, [r0, #20]
mov r4, #1
mov r5, r1
lsl r4, r4, r3
......@@ -164,8 +164,8 @@ function ff_mdct_calc_neon, export=1
push {r4-r10,lr}
mov r12, #1
ldr lr, [r0, #28] @ mdct_bits
ldr r4, [r0, #32] @ tcos
ldr lr, [r0, #20] @ mdct_bits
ldr r4, [r0, #24] @ tcos
ldr r3, [r0, #8] @ revtab
lsl lr, r12, lr @ n = 1 << nbits
add r7, r2, lr @ in4u
......@@ -253,8 +253,8 @@ function ff_mdct_calc_neon, export=1
bl ff_fft_calc_neon
mov r12, #1
ldr lr, [r4, #28] @ mdct_bits
ldr r4, [r4, #32] @ tcos
ldr lr, [r4, #20] @ mdct_bits
ldr r4, [r4, #24] @ tcos
lsl r12, r12, lr @ n = 1 << nbits
lsr lr, r12, #3 @ n8 = n >> 3
......
......@@ -80,26 +80,21 @@ av_cold void ff_init_ff_cos_tabs(int index)
av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
{
int i, j, m, n;
float alpha, c1, s1, s2;
int av_unused has_vectors;
int i, j, n;
if (nbits < 2 || nbits > 16)
goto fail;
s->nbits = nbits;
n = 1 << nbits;
s->tmp_buf = NULL;
s->exptab = av_malloc((n / 2) * sizeof(FFTComplex));
if (!s->exptab)
goto fail;
s->revtab = av_malloc(n * sizeof(uint16_t));
if (!s->revtab)
goto fail;
s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
if (!s->tmp_buf)
goto fail;
s->inverse = inverse;
s2 = inverse ? 1.0 : -1.0;
s->fft_permute = ff_fft_permute_c;
s->fft_calc = ff_fft_calc_c;
#if CONFIG_MDCT
......@@ -107,104 +102,37 @@ av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
s->imdct_half = ff_imdct_half_c;
s->mdct_calc = ff_mdct_calc_c;
#endif
s->exptab1 = NULL;
s->split_radix = 1;
if (ARCH_ARM) ff_fft_init_arm(s);
if (HAVE_ALTIVEC) ff_fft_init_altivec(s);
if (HAVE_MMX) ff_fft_init_mmx(s);
if (s->split_radix) {
for(j=4; j<=nbits; j++) {
ff_init_ff_cos_tabs(j);
}
for(i=0; i<n; i++)
s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] = i;
s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
} else {
int np, nblocks, np2, l;
FFTComplex *q;
for(i=0; i<(n/2); i++) {
alpha = 2 * M_PI * (float)i / (float)n;
c1 = cos(alpha);
s1 = sin(alpha) * s2;
s->exptab[i].re = c1;
s->exptab[i].im = s1;
}
np = 1 << nbits;
nblocks = np >> 3;
np2 = np >> 1;
s->exptab1 = av_malloc(np * 2 * sizeof(FFTComplex));
if (!s->exptab1)
goto fail;
q = s->exptab1;
do {
for(l = 0; l < np2; l += 2 * nblocks) {
*q++ = s->exptab[l];
*q++ = s->exptab[l + nblocks];
q->re = -s->exptab[l].im;
q->im = s->exptab[l].re;
q++;
q->re = -s->exptab[l + nblocks].im;
q->im = s->exptab[l + nblocks].re;
q++;
}
nblocks = nblocks >> 1;
} while (nblocks != 0);
av_freep(&s->exptab);
/* compute bit reverse table */
for(i=0;i<n;i++) {
m=0;
for(j=0;j<nbits;j++) {
m |= ((i >> j) & 1) << (nbits-j-1);
}
s->revtab[i]=m;
}
}
return 0;
fail:
av_freep(&s->revtab);
av_freep(&s->exptab);
av_freep(&s->exptab1);
av_freep(&s->tmp_buf);
return -1;
}
void ff_fft_permute_c(FFTContext *s, FFTComplex *z)
{
int j, k, np;
FFTComplex tmp;
int j, np;
const uint16_t *revtab = s->revtab;
np = 1 << s->nbits;
if (s->tmp_buf) {
/* TODO: handle split-radix permute in a more optimal way, probably in-place */
for(j=0;j<np;j++) s->tmp_buf[revtab[j]] = z[j];
memcpy(z, s->tmp_buf, np * sizeof(FFTComplex));
return;
}
/* reverse */
for(j=0;j<np;j++) {
k = revtab[j];
if (k < j) {
tmp = z[k];
z[k] = z[j];
z[j] = tmp;
}
}
}
av_cold void ff_fft_end(FFTContext *s)
{
av_freep(&s->revtab);
av_freep(&s->exptab);
av_freep(&s->exptab1);
av_freep(&s->tmp_buf);
}
......
......@@ -33,8 +33,6 @@ struct FFTContext {
int nbits;
int inverse;
uint16_t *revtab;
FFTComplex *exptab;
FFTComplex *exptab1; /* only used by SSE code */
FFTComplex *tmp_buf;
int mdct_size; /* size of MDCT (i.e. number of input data * 2) */
int mdct_bits; /* n = 2^nbits */
......@@ -46,7 +44,6 @@ struct FFTContext {
void (*imdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
void (*imdct_half)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
void (*mdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
int split_radix;
int permutation;
#define FF_MDCT_PERM_NONE 0
#define FF_MDCT_PERM_INTERLEAVE 1
......
......@@ -91,5 +91,4 @@ av_cold void ff_fft_init_altivec(FFTContext *s)
{
if (HAVE_GNU_AS)
s->fft_calc = ff_fft_calc_altivec;
s->split_radix = 1;
}
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment