Commit 022184a6 authored by Diego Biurrun's avatar Diego Biurrun
Browse files

ppc: dsputil: more K&R formatting cosmetics

parent 30f3f959
This diff is collapsed.
......@@ -27,32 +27,36 @@
/* AltiVec-enhanced gmc1. ATM this code assumes stride is a multiple of 8
* to preserve proper dst alignment. */
void ff_gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int stride, int h, int x16, int y16, int rounder)
void ff_gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */,
int stride, int h, int x16, int y16, int rounder)
{
const DECLARE_ALIGNED(16, unsigned short, rounder_a) = rounder;
const DECLARE_ALIGNED(16, unsigned short, ABCD)[8] =
{
(16-x16)*(16-y16), /* A */
( x16)*(16-y16), /* B */
(16-x16)*( y16), /* C */
( x16)*( y16), /* D */
0, 0, 0, 0 /* padding */
};
register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
register const vector unsigned short vcsr8 = (const vector unsigned short)vec_splat_u16(8);
register vector unsigned char dstv, dstv2, src_0, src_1, srcvA, srcvB, srcvC, srcvD;
register vector unsigned short Av, Bv, Cv, Dv, rounderV, tempA, tempB, tempC, tempD;
const DECLARE_ALIGNED(16, unsigned short, ABCD)[8] = {
(16 - x16) * (16 - y16), /* A */
(x16) * (16 - y16), /* B */
(16 - x16) * (y16), /* C */
(x16) * (y16), /* D */
0, 0, 0, 0 /* padding */
};
register const vector unsigned char vczero =
(const vector unsigned char) vec_splat_u8(0);
register const vector unsigned short vcsr8 =
(const vector unsigned short) vec_splat_u16(8);
register vector unsigned char dstv, dstv2, src_0, src_1,
srcvA, srcvB, srcvC, srcvD;
register vector unsigned short Av, Bv, Cv, Dv, rounderV,
tempA, tempB, tempC, tempD;
int i;
unsigned long dst_odd = (unsigned long)dst & 0x0000000F;
unsigned long src_really_odd = (unsigned long)src & 0x0000000F;
unsigned long dst_odd = (unsigned long) dst & 0x0000000F;
unsigned long src_really_odd = (unsigned long) src & 0x0000000F;
tempA = vec_ld(0, (const unsigned short*)ABCD);
Av = vec_splat(tempA, 0);
Bv = vec_splat(tempA, 1);
Cv = vec_splat(tempA, 2);
Dv = vec_splat(tempA, 3);
tempA = vec_ld(0, (const unsigned short *) ABCD);
Av = vec_splat(tempA, 0);
Bv = vec_splat(tempA, 1);
Cv = vec_splat(tempA, 2);
Dv = vec_splat(tempA, 3);
rounderV = vec_splat((vec_u16)vec_lde(0, &rounder_a), 0);
rounderV = vec_splat((vec_u16) vec_lde(0, &rounder_a), 0);
/* we'll be able to pick-up our 9 char elements at src from those
* 32 bytes we load the first batch here, as inside the loop we can
......@@ -61,36 +65,34 @@ void ff_gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int
src_1 = vec_ld(16, src);
srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src));
if (src_really_odd != 0x0000000F) {
if (src_really_odd != 0x0000000F)
/* If src & 0xF == 0xF, then (src + 1) is properly aligned
* on the second vector. */
srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src));
} else {
else
srcvB = src_1;
}
srcvA = vec_mergeh(vczero, srcvA);
srcvB = vec_mergeh(vczero, srcvB);
for(i=0; i<h; i++) {
dst_odd = (unsigned long)dst & 0x0000000F;
src_really_odd = (((unsigned long)src) + stride) & 0x0000000F;
for (i = 0; i < h; i++) {
dst_odd = (unsigned long) dst & 0x0000000F;
src_really_odd = (((unsigned long) src) + stride) & 0x0000000F;
dstv = vec_ld(0, dst);
/* We'll be able to pick-up our 9 char elements at src + stride from
* those 32 bytes then reuse the resulting 2 vectors srvcC and srcvD
* as the next srcvA and srcvB. */
src_0 = vec_ld(stride + 0, src);
src_0 = vec_ld(stride + 0, src);
src_1 = vec_ld(stride + 16, src);
srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src));
if (src_really_odd != 0x0000000F) {
if (src_really_odd != 0x0000000F)
/* If src & 0xF == 0xF, then (src + 1) is properly aligned
* on the second vector. */
srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src));
} else {
else
srcvD = src_1;
}
srcvC = vec_mergeh(vczero, srcvC);
srcvD = vec_mergeh(vczero, srcvD);
......@@ -98,23 +100,22 @@ void ff_gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int
/* OK, now we (finally) do the math :-)
* Those four instructions replace 32 int muls & 32 int adds.
* Isn't AltiVec nice? */
tempA = vec_mladd((vector unsigned short)srcvA, Av, rounderV);
tempB = vec_mladd((vector unsigned short)srcvB, Bv, tempA);
tempC = vec_mladd((vector unsigned short)srcvC, Cv, tempB);
tempD = vec_mladd((vector unsigned short)srcvD, Dv, tempC);
tempA = vec_mladd((vector unsigned short) srcvA, Av, rounderV);
tempB = vec_mladd((vector unsigned short) srcvB, Bv, tempA);
tempC = vec_mladd((vector unsigned short) srcvC, Cv, tempB);
tempD = vec_mladd((vector unsigned short) srcvD, Dv, tempC);
srcvA = srcvC;
srcvB = srcvD;
tempD = vec_sr(tempD, vcsr8);
dstv2 = vec_pack(tempD, (vector unsigned short)vczero);
dstv2 = vec_pack(tempD, (vector unsigned short) vczero);
if (dst_odd) {
dstv2 = vec_perm(dstv, dstv2, vcprm(0,1,s0,s1));
} else {
dstv2 = vec_perm(dstv, dstv2, vcprm(s0,s1,2,3));
}
if (dst_odd)
dstv2 = vec_perm(dstv, dstv2, vcprm(0, 1, s0, s1));
else
dstv2 = vec_perm(dstv, dstv2, vcprm(s0, s1, 2, 3));
vec_st(dstv2, 0, dst);
......
......@@ -36,147 +36,153 @@
#if HAVE_ALTIVEC_H
#include <altivec.h>
#endif
#include "libavutil/ppc/types_altivec.h"
#include "dsputil_altivec.h"
#define IDCT_HALF \
/* 1st stage */ \
t1 = vec_mradds (a1, vx7, vx1 ); \
t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7)); \
t7 = vec_mradds (a2, vx5, vx3); \
t3 = vec_mradds (ma2, vx3, vx5); \
t1 = vec_mradds(a1, vx7, vx1); \
t8 = vec_mradds(a1, vx1, vec_subs(zero, vx7)); \
t7 = vec_mradds(a2, vx5, vx3); \
t3 = vec_mradds(ma2, vx3, vx5); \
\
/* 2nd stage */ \
t5 = vec_adds (vx0, vx4); \
t0 = vec_subs (vx0, vx4); \
t2 = vec_mradds (a0, vx6, vx2); \
t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6)); \
t6 = vec_adds (t8, t3); \
t3 = vec_subs (t8, t3); \
t8 = vec_subs (t1, t7); \
t1 = vec_adds (t1, t7); \
t5 = vec_adds(vx0, vx4); \
t0 = vec_subs(vx0, vx4); \
t2 = vec_mradds(a0, vx6, vx2); \
t4 = vec_mradds(a0, vx2, vec_subs(zero, vx6)); \
t6 = vec_adds(t8, t3); \
t3 = vec_subs(t8, t3); \
t8 = vec_subs(t1, t7); \
t1 = vec_adds(t1, t7); \
\
/* 3rd stage */ \
t7 = vec_adds (t5, t2); \
t2 = vec_subs (t5, t2); \
t5 = vec_adds (t0, t4); \
t0 = vec_subs (t0, t4); \
t4 = vec_subs (t8, t3); \
t3 = vec_adds (t8, t3); \
t7 = vec_adds(t5, t2); \
t2 = vec_subs(t5, t2); \
t5 = vec_adds(t0, t4); \
t0 = vec_subs(t0, t4); \
t4 = vec_subs(t8, t3); \
t3 = vec_adds(t8, t3); \
\
/* 4th stage */ \
vy0 = vec_adds (t7, t1); \
vy7 = vec_subs (t7, t1); \
vy1 = vec_mradds (c4, t3, t5); \
vy6 = vec_mradds (mc4, t3, t5); \
vy2 = vec_mradds (c4, t4, t0); \
vy5 = vec_mradds (mc4, t4, t0); \
vy3 = vec_adds (t2, t6); \
vy4 = vec_subs (t2, t6);
#define IDCT \
vec_s16 vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7; \
vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \
vec_s16 a0, a1, a2, ma2, c4, mc4, zero, bias; \
vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8; \
vec_u16 shift; \
\
c4 = vec_splat (constants[0], 0); \
a0 = vec_splat (constants[0], 1); \
a1 = vec_splat (constants[0], 2); \
a2 = vec_splat (constants[0], 3); \
mc4 = vec_splat (constants[0], 4); \
ma2 = vec_splat (constants[0], 5); \
bias = (vec_s16)vec_splat ((vec_s32)constants[0], 3); \
\
zero = vec_splat_s16 (0); \
shift = vec_splat_u16 (4); \
\
vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero); \
vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero); \
vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero); \
vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero); \
vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero); \
vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero); \
vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero); \
vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero); \
\
IDCT_HALF \
\
vx0 = vec_mergeh (vy0, vy4); \
vx1 = vec_mergel (vy0, vy4); \
vx2 = vec_mergeh (vy1, vy5); \
vx3 = vec_mergel (vy1, vy5); \
vx4 = vec_mergeh (vy2, vy6); \
vx5 = vec_mergel (vy2, vy6); \
vx6 = vec_mergeh (vy3, vy7); \
vx7 = vec_mergel (vy3, vy7); \
\
vy0 = vec_mergeh (vx0, vx4); \
vy1 = vec_mergel (vx0, vx4); \
vy2 = vec_mergeh (vx1, vx5); \
vy3 = vec_mergel (vx1, vx5); \
vy4 = vec_mergeh (vx2, vx6); \
vy5 = vec_mergel (vx2, vx6); \
vy6 = vec_mergeh (vx3, vx7); \
vy7 = vec_mergel (vx3, vx7); \
\
vx0 = vec_adds (vec_mergeh (vy0, vy4), bias); \
vx1 = vec_mergel (vy0, vy4); \
vx2 = vec_mergeh (vy1, vy5); \
vx3 = vec_mergel (vy1, vy5); \
vx4 = vec_mergeh (vy2, vy6); \
vx5 = vec_mergel (vy2, vy6); \
vx6 = vec_mergeh (vy3, vy7); \
vx7 = vec_mergel (vy3, vy7); \
\
IDCT_HALF \
\
shift = vec_splat_u16 (6); \
vx0 = vec_sra (vy0, shift); \
vx1 = vec_sra (vy1, shift); \
vx2 = vec_sra (vy2, shift); \
vx3 = vec_sra (vy3, shift); \
vx4 = vec_sra (vy4, shift); \
vx5 = vec_sra (vy5, shift); \
vx6 = vec_sra (vy6, shift); \
vx7 = vec_sra (vy7, shift);
vy0 = vec_adds(t7, t1); \
vy7 = vec_subs(t7, t1); \
vy1 = vec_mradds(c4, t3, t5); \
vy6 = vec_mradds(mc4, t3, t5); \
vy2 = vec_mradds(c4, t4, t0); \
vy5 = vec_mradds(mc4, t4, t0); \
vy3 = vec_adds(t2, t6); \
vy4 = vec_subs(t2, t6);
#define IDCT \
vec_s16 vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7; \
vec_s16 vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \
vec_s16 a0, a1, a2, ma2, c4, mc4, zero, bias; \
vec_s16 t0, t1, t2, t3, t4, t5, t6, t7, t8; \
vec_u16 shift; \
\
c4 = vec_splat(constants[0], 0); \
a0 = vec_splat(constants[0], 1); \
a1 = vec_splat(constants[0], 2); \
a2 = vec_splat(constants[0], 3); \
mc4 = vec_splat(constants[0], 4); \
ma2 = vec_splat(constants[0], 5); \
bias = (vec_s16) vec_splat((vec_s32) constants[0], 3); \
\
zero = vec_splat_s16(0); \
shift = vec_splat_u16(4); \
\
vx0 = vec_mradds(vec_sl(block[0], shift), constants[1], zero); \
vx1 = vec_mradds(vec_sl(block[1], shift), constants[2], zero); \
vx2 = vec_mradds(vec_sl(block[2], shift), constants[3], zero); \
vx3 = vec_mradds(vec_sl(block[3], shift), constants[4], zero); \
vx4 = vec_mradds(vec_sl(block[4], shift), constants[1], zero); \
vx5 = vec_mradds(vec_sl(block[5], shift), constants[4], zero); \
vx6 = vec_mradds(vec_sl(block[6], shift), constants[3], zero); \
vx7 = vec_mradds(vec_sl(block[7], shift), constants[2], zero); \
\
IDCT_HALF \
\
vx0 = vec_mergeh(vy0, vy4); \
vx1 = vec_mergel(vy0, vy4); \
vx2 = vec_mergeh(vy1, vy5); \
vx3 = vec_mergel(vy1, vy5); \
vx4 = vec_mergeh(vy2, vy6); \
vx5 = vec_mergel(vy2, vy6); \
vx6 = vec_mergeh(vy3, vy7); \
vx7 = vec_mergel(vy3, vy7); \
\
vy0 = vec_mergeh(vx0, vx4); \
vy1 = vec_mergel(vx0, vx4); \
vy2 = vec_mergeh(vx1, vx5); \
vy3 = vec_mergel(vx1, vx5); \
vy4 = vec_mergeh(vx2, vx6); \
vy5 = vec_mergel(vx2, vx6); \
vy6 = vec_mergeh(vx3, vx7); \
vy7 = vec_mergel(vx3, vx7); \
\
vx0 = vec_adds(vec_mergeh(vy0, vy4), bias); \
vx1 = vec_mergel(vy0, vy4); \
vx2 = vec_mergeh(vy1, vy5); \
vx3 = vec_mergel(vy1, vy5); \
vx4 = vec_mergeh(vy2, vy6); \
vx5 = vec_mergel(vy2, vy6); \
vx6 = vec_mergeh(vy3, vy7); \
vx7 = vec_mergel(vy3, vy7); \
\
IDCT_HALF \
\
shift = vec_splat_u16(6); \
vx0 = vec_sra(vy0, shift); \
vx1 = vec_sra(vy1, shift); \
vx2 = vec_sra(vy2, shift); \
vx3 = vec_sra(vy3, shift); \
vx4 = vec_sra(vy4, shift); \
vx5 = vec_sra(vy5, shift); \
vx6 = vec_sra(vy6, shift); \
vx7 = vec_sra(vy7, shift);
static const vec_s16 constants[5] = {
{23170, 13573, 6518, 21895, -23170, -21895, 32, 31},
{16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725},
{22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521},
{21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692},
{19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722}
{ 23170, 13573, 6518, 21895, -23170, -21895, 32, 31 },
{ 16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725 },
{ 22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521 },
{ 21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692 },
{ 19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722 }
};
void ff_idct_put_altivec(uint8_t* dest, int stride, int16_t *blk)
void ff_idct_put_altivec(uint8_t *dest, int stride, int16_t *blk)
{
vec_s16 *block = (vec_s16*)blk;
vec_s16 *block = (vec_s16 *) blk;
vec_u8 tmp;
IDCT
#define COPY(dest,src) \
tmp = vec_packsu (src, src); \
vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest); \
vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
COPY (dest, vx0) dest += stride;
COPY (dest, vx1) dest += stride;
COPY (dest, vx2) dest += stride;
COPY (dest, vx3) dest += stride;
COPY (dest, vx4) dest += stride;
COPY (dest, vx5) dest += stride;
COPY (dest, vx6) dest += stride;
COPY (dest, vx7)
#define COPY(dest, src) \
tmp = vec_packsu(src, src); \
vec_ste((vec_u32) tmp, 0, (unsigned int *) dest); \
vec_ste((vec_u32) tmp, 4, (unsigned int *) dest);
COPY(dest, vx0)
dest += stride;
COPY(dest, vx1)
dest += stride;
COPY(dest, vx2)
dest += stride;
COPY(dest, vx3)
dest += stride;
COPY(dest, vx4)
dest += stride;
COPY(dest, vx5)
dest += stride;
COPY(dest, vx6)
dest += stride;
COPY(dest, vx7)
}
void ff_idct_add_altivec(uint8_t* dest, int stride, int16_t *blk)
void ff_idct_add_altivec(uint8_t *dest, int stride, int16_t *blk)
{
vec_s16 *block = (vec_s16*)blk;
vec_s16 *block = (vec_s16 *) blk;
vec_u8 tmp;
vec_s16 tmp2, tmp3;
vec_u8 perm0;
......@@ -185,27 +191,34 @@ void ff_idct_add_altivec(uint8_t* dest, int stride, int16_t *blk)
IDCT
p0 = vec_lvsl (0, dest);
p1 = vec_lvsl (stride, dest);
p = vec_splat_u8 (-1);
perm0 = vec_mergeh (p, p0);
perm1 = vec_mergeh (p, p1);
#define ADD(dest,src,perm) \
/* *(uint64_t *)&tmp = *(uint64_t *)dest; */ \
tmp = vec_ld (0, dest); \
tmp2 = (vec_s16)vec_perm (tmp, (vec_u8)zero, perm); \
tmp3 = vec_adds (tmp2, src); \
tmp = vec_packsu (tmp3, tmp3); \
vec_ste ((vec_u32)tmp, 0, (unsigned int *)dest); \
vec_ste ((vec_u32)tmp, 4, (unsigned int *)dest);
ADD (dest, vx0, perm0) dest += stride;
ADD (dest, vx1, perm1) dest += stride;
ADD (dest, vx2, perm0) dest += stride;
ADD (dest, vx3, perm1) dest += stride;
ADD (dest, vx4, perm0) dest += stride;
ADD (dest, vx5, perm1) dest += stride;
ADD (dest, vx6, perm0) dest += stride;
ADD (dest, vx7, perm1)
p0 = vec_lvsl(0, dest);
p1 = vec_lvsl(stride, dest);
p = vec_splat_u8(-1);
perm0 = vec_mergeh(p, p0);
perm1 = vec_mergeh(p, p1);
#define ADD(dest, src, perm) \
/* *(uint64_t *) &tmp = *(uint64_t *) dest; */ \
tmp = vec_ld(0, dest); \
tmp2 = (vec_s16) vec_perm(tmp, (vec_u8) zero, perm); \
tmp3 = vec_adds(tmp2, src); \
tmp = vec_packsu(tmp3, tmp3); \
vec_ste((vec_u32) tmp, 0, (unsigned int *) dest); \
vec_ste((vec_u32) tmp, 4, (unsigned int *) dest);
ADD(dest, vx0, perm0)
dest += stride;
ADD(dest, vx1, perm1)
dest += stride;
ADD(dest, vx2, perm0)
dest += stride;
ADD(dest, vx3, perm1)
dest += stride;
ADD(dest, vx4, perm0)
dest += stride;
ADD(dest, vx5, perm1)
dest += stride;
ADD(dest, vx6, perm0)
dest += stride;
ADD(dest, vx7, perm1)
}
......@@ -31,27 +31,28 @@
#include "libavutil/attributes.h"
#include "libavutil/ppc/types_altivec.h"
#include "libavcodec/dsputil.h"
#include "dsputil_altivec.h"
static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2,
int size) {
int size)
{
int i, size16;
vector signed char vpix1;
vector signed short vpix2, vdiff, vpix1l,vpix1h;
union { vector signed int vscore;
int32_t score[4];
} u;
vector signed short vpix2, vdiff, vpix1l, vpix1h;
union {
vector signed int vscore;
int32_t score[4];
} u;
u.vscore = vec_splat_s32(0);
// XXX lazy way, fix it later
#define vec_unaligned_load(b) \
vec_perm(vec_ld(0,b),vec_ld(15,b),vec_lvsl(0, b));
#define vec_unaligned_load(b) \
vec_perm(vec_ld(0, b), vec_ld(15, b), vec_lvsl(0, b));
size16 = size >> 4;
while(size16) {
// score += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]);
while (size16) {
// score += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
// load pix1 and the first batch of pix2
vpix1 = vec_unaligned_load(pix1);
......@@ -62,20 +63,20 @@ static int ssd_int8_vs_int16_altivec(const int8_t *pix1, const int16_t *pix2,
vdiff = vec_sub(vpix1h, vpix2);
vpix1l = vec_unpackl(vpix1);
// load another batch from pix2
vpix2 = vec_unaligned_load(pix2);
vpix2 = vec_unaligned_load(pix2);
u.vscore = vec_msum(vdiff, vdiff, u.vscore);
vdiff = vec_sub(vpix1l, vpix2);
vdiff = vec_sub(vpix1l, vpix2);
u.vscore = vec_msum(vdiff, vdiff, u.vscore);
pix1 += 16;
pix2 += 8;
pix1 += 16;
pix2 += 8;
size16--;
}
u.vscore = vec_sums(u.vscore, vec_splat_s32(0));
size %= 16;
for (i = 0; i < size; i++) {
u.score[3] += (pix1[i]-pix2[i])*(pix1[i]-pix2[i]);
}
for (i = 0; i < size; i++)
u.score[3] += (pix1[i] - pix2[i]) * (pix1[i] - pix2[i]);
return u.score[3];
}
......@@ -88,56 +89,64 @@ static int32_t scalarproduct_int16_altivec(const int16_t *v1, const int16_t *v2,
register vec_s32 res = vec_splat_s32(0), t;
int32_t ires;
for(i = 0; i < order; i += 8){
for (i = 0; i < order; i += 8) {
vec1 = vec_unaligned_load(v1);
t = vec_msum(vec1, vec_ld(0, v2), zero_s32v);
res = vec_sums(t, res);
v1 += 8;
v2 += 8;
t = vec_msum(vec1, vec_ld(0, v2), zero_s32v);
res = vec_sums(t, res);
v1 += 8;
v2 += 8;
}
res = vec_splat(res, 3);
vec_ste(res, 0, &ires);
return ires;
}
static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1, const int16_t *v2, const int16_t *v3, int order, int mul)
static int32_t scalarproduct_and_madd_int16_altivec(int16_t *v1,
const int16_t *v2,
const int16_t *v3,
int order, int mul)
{
LOAD_ZERO;
vec_s16 *pv1 = (vec_s16*)v1;
register vec_s16 muls = {mul,mul,mul,mul,mul,mul,mul,mul};
vec_s16 *pv1 = (vec_s16 *) v1;
register vec_s16 muls = { mul, mul, mul, mul, mul, mul, mul, mul };
register vec_s16 t0, t1, i0, i1, i4;
register vec_s16 i2 = vec_ld(0, v2), i3 = vec_ld(0, v3);
register vec_s32 res = zero_s32v;
register vec_u8 align = vec_lvsl(0, v2);
int32_t ires;
order >>= 4;
do {
i1 = vec_ld(16, v2);
t0 = vec_perm(i2, i1, align);
i2 = vec_ld(32, v2);
t1 = vec_perm(i1, i2, align);
i0 = pv1[0];
i1 = pv1[1];
res = vec_msum(t0, i0, res);
res = vec_msum(t1, i1, res);
i4 = vec_ld(16, v3);
t0 = vec_perm(i3, i4, align);
i3 = vec_ld(32, v3);
t1 = vec_perm(i4, i3, align);
i1 = vec_ld(16, v2);
t0 = vec_perm(i2, i1, align);
i2 = vec_ld(32, v2);
t1 = vec_perm(i1, i2, align);
i0 = pv1[0];
i1 = pv1[1];
res = vec_msum(t0, i0, res);
res = vec_msum(t1, i1, res);
i4 = vec_ld(16, v3);
t0 = vec_perm(i3, i4, align);
i3 = vec_ld(32, v3);
t1 = vec_perm(i4, i3, align);
pv1[0] = vec_mladd(t0, muls, i0);
pv1[1] = vec_mladd(t1, muls, i1);
pv1 += 2;
v2 += 16;
v3 += 16;