#include "gcc_fixes.h"
#include "dsputil_altivec.h"
+#include "util_altivec.h"
static void vector_fmul_altivec(float *dst, const float *src, int len)
{
ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step);
}
-void float_to_int16_altivec(int16_t *dst, const float *src, int len)
+static void vector_fmul_window_altivec(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len)
+{
+ union {
+ vector float v;
+ float s[4];
+ } vadd;
+ vector float vadd_bias, zero, t0, t1, s0, s1, wi, wj;
+ const vector unsigned char reverse = vcprm(3,2,1,0);
+ int i,j;
+
+ dst += len;
+ win += len;
+ src0+= len;
+
+ vadd.s[0] = add_bias;
+ vadd_bias = vec_splat(vadd.v, 0);
+ zero = (vector float)vec_splat_u32(0);
+
+ for(i=-len*4, j=len*4-16; i<0; i+=16, j-=16) {
+ s0 = vec_ld(i, src0);
+ s1 = vec_ld(j, src1);
+ wi = vec_ld(i, win);
+ wj = vec_ld(j, win);
+
+ s1 = vec_perm(s1, s1, reverse);
+ wj = vec_perm(wj, wj, reverse);
+
+ t0 = vec_madd(s0, wj, vadd_bias);
+ t0 = vec_nmsub(s1, wi, t0);
+ t1 = vec_madd(s0, wi, vadd_bias);
+ t1 = vec_madd(s1, wj, t1);
+ t1 = vec_perm(t1, t1, reverse);
+
+ vec_st(t0, i, dst);
+ vec_st(t1, j, dst);
+ }
+}
+
+static void int32_to_float_fmul_scalar_altivec(float *dst, const int *src, float mul, int len)
+{
+ union {
+ vector float v;
+ float s[4];
+ } mul_u;
+ int i;
+ vector float src1, src2, dst1, dst2, mul_v, zero;
+
+ zero = (vector float)vec_splat_u32(0);
+ mul_u.s[0] = mul;
+ mul_v = vec_splat(mul_u.v, 0);
+
+ for(i=0; i<len; i+=8) {
+ src1 = vec_ctf(vec_ld(0, src+i), 0);
+ src2 = vec_ctf(vec_ld(16, src+i), 0);
+ dst1 = vec_madd(src1, mul_v, zero);
+ dst2 = vec_madd(src2, mul_v, zero);
+ vec_st(dst1, 0, dst+i);
+ vec_st(dst2, 16, dst+i);
+ }
+}
+
+
+static vector signed short
+float_to_int16_one_altivec(const float *src)
+{
+ vector float s0 = vec_ld(0, src);
+ vector float s1 = vec_ld(16, src);
+ vector signed int t0 = vec_cts(s0, 0);
+ vector signed int t1 = vec_cts(s1, 0);
+ return vec_packs(t0,t1);
+}
+
+static void float_to_int16_altivec(int16_t *dst, const float *src, long len)
{
int i;
- vector float s0, s1;
- vector signed int t0, t1;
vector signed short d0, d1, d;
vector unsigned char align;
if(((long)dst)&15) //FIXME
for(i=0; i<len-7; i+=8) {
- s0 = vec_ld(0, src+i);
- s1 = vec_ld(16, src+i);
- t0 = vec_cts(s0, 0);
d0 = vec_ld(0, dst+i);
- t1 = vec_cts(s1, 0);
+ d = float_to_int16_one_altivec(src+i);
d1 = vec_ld(15, dst+i);
- d = vec_packs(t0,t1);
d1 = vec_perm(d1, d0, vec_lvsl(0,dst+i));
align = vec_lvsr(0, dst+i);
d0 = vec_perm(d1, d, align);
}
else
for(i=0; i<len-7; i+=8) {
- s0 = vec_ld(0, src+i);
- s1 = vec_ld(16, src+i);
- t0 = vec_cts(s0, 0);
- t1 = vec_cts(s1, 0);
- d = vec_packs(t0,t1);
+ d = float_to_int16_one_altivec(src+i);
vec_st(d, 0, dst+i);
}
}
+static void
+float_to_int16_interleave_altivec(int16_t *dst, const float **src,
+ long len, int channels)
+{
+ int i;
+ vector signed short d0, d1, d2, c0, c1, t0, t1;
+ vector unsigned char align;
+ if(channels == 1)
+ float_to_int16_altivec(dst, src[0], len);
+ else
+ if (channels == 2) {
+ if(((long)dst)&15)
+ for(i=0; i<len-7; i+=8) {
+ d0 = vec_ld(0, dst + i);
+ t0 = float_to_int16_one_altivec(src[0] + i);
+ d1 = vec_ld(31, dst + i);
+ t1 = float_to_int16_one_altivec(src[1] + i);
+ c0 = vec_mergeh(t0, t1);
+ c1 = vec_mergel(t0, t1);
+ d2 = vec_perm(d1, d0, vec_lvsl(0, dst + i));
+ align = vec_lvsr(0, dst + i);
+ d0 = vec_perm(d2, c0, align);
+ d1 = vec_perm(c0, c1, align);
+ vec_st(d0, 0, dst + i);
+ d0 = vec_perm(c1, d2, align);
+ vec_st(d1, 15, dst + i);
+ vec_st(d0, 31, dst + i);
+ dst+=8;
+ }
+ else
+ for(i=0; i<len-7; i+=8) {
+ t0 = float_to_int16_one_altivec(src[0] + i);
+ t1 = float_to_int16_one_altivec(src[1] + i);
+ d0 = vec_mergeh(t0, t1);
+ d1 = vec_mergel(t0, t1);
+ vec_st(d0, 0, dst + i);
+ vec_st(d1, 16, dst + i);
+ dst+=8;
+ }
+ } else {
+ DECLARE_ALIGNED(16, int16_t, tmp[len]);
+ int c, j;
+ for (c = 0; c < channels; c++) {
+ float_to_int16_altivec(tmp, src[c], len);
+ for (i = 0, j = c; i < len; i++, j+=channels) {
+ dst[j] = tmp[i];
+ }
+ }
+ }
+}
+
void float_init_altivec(DSPContext* c, AVCodecContext *avctx)
{
c->vector_fmul = vector_fmul_altivec;
c->vector_fmul_reverse = vector_fmul_reverse_altivec;
c->vector_fmul_add_add = vector_fmul_add_add_altivec;
- if(!(avctx->flags & CODEC_FLAG_BITEXACT))
+ c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_altivec;
+ if(!(avctx->flags & CODEC_FLAG_BITEXACT)) {
+ c->vector_fmul_window = vector_fmul_window_altivec;
c->float_to_int16 = float_to_int16_altivec;
+ c->float_to_int16_interleave = float_to_int16_interleave_altivec;
+ }
}