static void predict_16x16_dc_128_altivec( uint8_t *src )
{
/* test if generating the constant is faster than loading it.
vector unsigned int bc_v = (vector unsigned int)CV(0x80808080, 0x80808080, 0x80808080, 0x80808080);
*/
vec_u8_t bc_v = vec_vslb((vec_u8_t)vec_splat_u8(1),(vec_u8_t)vec_splat_u8(7));
PREDICT_16x16_DC_ALTIVEC(bc_v);
}
int foo(volatile vector float &i, int &j)
{
vector unsigned char zero = vec_splat_u8(0);
vector unsigned char one = vec_splat_u8(1);
i = vec_add( (vector float)zero, (vector float)one );
j = 5;
return 0;
}
// out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
static inline vector unsigned char h264_deblock_q1(register vector unsigned char p0,
register vector unsigned char p1,
register vector unsigned char p2,
register vector unsigned char q0,
register vector unsigned char tc0) {
register vector unsigned char average = vec_avg(p0, q0);
register vector unsigned char temp;
register vector unsigned char uncliped;
register vector unsigned char ones;
register vector unsigned char max;
register vector unsigned char min;
register vector unsigned char newp1;
temp = vec_xor(average, p2);
average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
ones = vec_splat_u8(1);
temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */
uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
max = vec_adds(p1, tc0);
min = vec_subs(p1, tc0);
newp1 = vec_max(min, uncliped);
newp1 = vec_min(max, newp1);
return newp1;
}
void ff_vp3_idct_add_altivec(uint8_t *dst, int stride, DCTELEM block[64])
{
LOAD_ZERO;
vec_u8 t, vdst;
vec_s16 vdst_16;
vec_u8 vdst_mask = vec_mergeh(vec_splat_u8(-1), vec_lvsl(0, dst));
IDCT_START
IDCT_1D(NOP, NOP)
TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7);
IDCT_1D(ADD8, SHIFT4)
#define ADD(a)\
vdst = vec_ld(0, dst);\
vdst_16 = (vec_s16)vec_perm(vdst, zero_u8v, vdst_mask);\
vdst_16 = vec_adds(a, vdst_16);\
t = vec_packsu(vdst_16, vdst_16);\
vec_ste((vec_u32)t, 0, (unsigned int *)dst);\
vec_ste((vec_u32)t, 4, (unsigned int *)dst);
ADD(b0) dst += stride;
ADD(b1) dst += stride;
ADD(b2) dst += stride;
ADD(b3) dst += stride;
ADD(b4) dst += stride;
ADD(b5) dst += stride;
ADD(b6) dst += stride;
ADD(b7)
}
// out: newp1 = clip((p2 + ((p0 + q0 + 1) >> 1)) >> 1, p1-tc0, p1+tc0)
static inline vec_u8_t h264_deblock_q1(register vec_u8_t p0,
register vec_u8_t p1,
register vec_u8_t p2,
register vec_u8_t q0,
register vec_u8_t tc0) {
register vec_u8_t average = vec_avg(p0, q0);
register vec_u8_t temp;
register vec_u8_t uncliped;
register vec_u8_t ones;
register vec_u8_t max;
register vec_u8_t min;
register vec_u8_t newp1;
temp = vec_xor(average, p2);
average = vec_avg(average, p2); /*avg(p2, avg(p0, q0)) */
ones = vec_splat_u8(1);
temp = vec_and(temp, ones); /*(p2^avg(p0, q0)) & 1 */
uncliped = vec_subs(average, temp); /*(p2+((p0+q0+1)>>1))>>1 */
max = vec_adds(p1, tc0);
min = vec_subs(p1, tc0);
newp1 = vec_max(min, uncliped);
newp1 = vec_min(max, newp1);
return newp1;
}
void
transfer_8to16subro_altivec_c(int16_t * dct,
const uint8_t * cur,
const uint8_t * ref,
const uint32_t stride)
{
register vector unsigned char c;
register vector unsigned char r;
register vector unsigned char z;
register vector signed short cs;
register vector signed short rs;
#ifdef DEBUG
/* Check the alignment assumptions if this is on */
if((long)dct & 0xf)
fprintf(stderr, "transfer_8to16subro_altivec_c:incorrect align, dct: %lx\n", (long)dct);
#endif
/* initialize */
z = vec_splat_u8(0);
SUBRO8TO16();
SUBRO8TO16();
SUBRO8TO16();
SUBRO8TO16();
SUBRO8TO16();
SUBRO8TO16();
SUBRO8TO16();
SUBRO8TO16();
}
void
transfer_8to16sub_altivec_c(int16_t * dct,
uint8_t * cur,
uint8_t * ref,
const uint32_t stride)
{
register vector unsigned char c,r;
register vector unsigned char ox00;
register vector unsigned char mask_00ff;
register vector unsigned char mask;
register vector signed short cs,rs;
#ifdef DEBUG
if((long)dct & 0xf)
fprintf(stderr, "transfer_8to16sub_altivec_c:incorrect align, dct: %lx\n", (long)dct);
if((long)cur & 0x7)
fprintf(stderr, "transfer_8to16sub_altivec_c:incorrect align, cur: %lx\n", (long)cur);
if(stride & 0x7)
fprintf(stderr, "transfer_8to16sub_altivec_c:incorrect stride, stride: %lu\n", (long)stride);
#endif
/* initialize */
ox00 = vec_splat_u8(0);
mask_00ff = vec_pack((vector unsigned short)ox00,vec_splat_u16(-1));
SUB8TO16();
SUB8TO16();
SUB8TO16();
SUB8TO16();
SUB8TO16();
SUB8TO16();
SUB8TO16();
SUB8TO16();
}
void ff_idct_add_altivec(uint8_t* dest, int stride, int16_t *blk)
{
vec_s16 *block = (vec_s16*)blk;
vec_u8 tmp;
vec_s16 tmp2, tmp3;
vec_u8 perm0;
vec_u8 perm1;
vec_u8 p0, p1, p;
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)
}
void
transfer_8to16copy_altivec_c(int16_t *dst,
uint8_t * src,
uint32_t stride)
{
register vector unsigned char s;
register vector unsigned char zerovec;
#ifdef DEBUG
/* Check the alignment */
if((long)dst & 0xf)
fprintf(stderr, "transfer_8to16copy_altivec_c:incorrect align, dst: %lx\n", (long)dst);
#endif
/* initialization */
zerovec = vec_splat_u8(0);
COPY8TO16();
COPY8TO16();
COPY8TO16();
COPY8TO16();
COPY8TO16();
COPY8TO16();
COPY8TO16();
COPY8TO16();
}
void imageFilterMean_Altivec(unsigned char *src1, unsigned char *src2, unsigned char *dst, int length)
{
int n = length;
// Compute first few values so we're on a 16-byte boundary in dst
while( (((long)dst & 0xF) > 0) && (n > 0) ) {
MEAN_PIXEL();
--n; ++dst; ++src1; ++src2;
}
// Do bulk of processing using Altivec (find the mean of 16 8-bit unsigned integers, with saturation)
vector unsigned char rshft = vec_splat_u8(0x1);
while(n >= 16) {
vector unsigned char s1 = vec_ld(0,src1);
s1 = vec_sr(s1, rshft); // shift right 1
vector unsigned char s2 = vec_ld(0,src2);
s2 = vec_sr(s2, rshft); // shift right 1
vector unsigned char r = vec_adds(s1, s2);
vec_st(r,0,dst);
n -= 16; src1 += 16; src2 += 16; dst += 16;
}
// If any bytes are left over, deal with them individually
++n;
BASIC_MEAN();
}
int pix_abs16x16_y2_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
{
int i;
int s __attribute__((aligned(16)));
const vector unsigned char zero = (const vector unsigned char)vec_splat_u8(0);
vector unsigned char *tv;
vector unsigned char pix1v, pix2v, pix3v, avgv, t5;
vector unsigned int sad;
vector signed int sumdiffs;
uint8_t *pix3 = pix2 + line_size;
s = 0;
sad = (vector unsigned int)vec_splat_u32(0);
/*
Due to the fact that pix3 = pix2 + line_size, the pix3 of one
iteration becomes pix2 in the next iteration. We can use this
fact to avoid a potentially expensive unaligned read, each
time around the loop.
Read unaligned pixels into our vectors. The vectors are as follows:
pix2v: pix2[0]-pix2[15]
Split the pixel vectors into shorts
*/
tv = (vector unsigned char *) &pix2[0];
pix2v = vec_perm(tv[0], tv[1], vec_lvsl(0, &pix2[0]));
for(i=0;i<16;i++) {
/*
Read unaligned pixels into our vectors. The vectors are as follows:
pix1v: pix1[0]-pix1[15]
pix3v: pix3[0]-pix3[15]
*/
tv = (vector unsigned char *) pix1;
pix1v = vec_perm(tv[0], tv[1], vec_lvsl(0, pix1));
tv = (vector unsigned char *) &pix3[0];
pix3v = vec_perm(tv[0], tv[1], vec_lvsl(0, &pix3[0]));
/* Calculate the average vector */
avgv = vec_avg(pix2v, pix3v);
/* Calculate a sum of abs differences vector */
t5 = vec_sub(vec_max(pix1v, avgv), vec_min(pix1v, avgv));
/* Add each 4 pixel group together and put 4 results into sad */
sad = vec_sum4s(t5, sad);
pix1 += line_size;
pix2v = pix3v;
pix3 += line_size;
}
/* Sum up the four partial sums, and put the result into s */
sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero);
sumdiffs = vec_splat(sumdiffs, 3);
vec_ste(sumdiffs, 0, &s);
return s;
}
/* next one assumes that ((line_size % 8) == 0) */
static void put_no_rnd_pixels8_xy2_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int h)
{
register int i;
register vector unsigned char pixelsv1, pixelsv2, pixelsavg;
register vector unsigned char blockv, temp1, temp2;
register vector unsigned short pixelssum1, pixelssum2, temp3;
register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
register const vector unsigned short vcone = (const vector unsigned short)vec_splat_u16(1);
register const vector unsigned short vctwo = (const vector unsigned short)vec_splat_u16(2);
temp1 = vec_ld(0, pixels);
temp2 = vec_ld(16, pixels);
pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(0, pixels));
if ((((unsigned long)pixels) & 0x0000000F) == 0x0000000F) {
pixelsv2 = temp2;
} else {
pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(1, pixels));
}
pixelsv1 = vec_mergeh(vczero, pixelsv1);
pixelsv2 = vec_mergeh(vczero, pixelsv2);
pixelssum1 = vec_add((vector unsigned short)pixelsv1,
(vector unsigned short)pixelsv2);
pixelssum1 = vec_add(pixelssum1, vcone);
for (i = 0; i < h ; i++) {
int rightside = ((unsigned long)block & 0x0000000F);
blockv = vec_ld(0, block);
temp1 = vec_ld(line_size, pixels);
temp2 = vec_ld(line_size + 16, pixels);
pixelsv1 = vec_perm(temp1, temp2, vec_lvsl(line_size, pixels));
if (((((unsigned long)pixels) + line_size) & 0x0000000F) == 0x0000000F) {
pixelsv2 = temp2;
} else {
pixelsv2 = vec_perm(temp1, temp2, vec_lvsl(line_size + 1, pixels));
}
pixelsv1 = vec_mergeh(vczero, pixelsv1);
pixelsv2 = vec_mergeh(vczero, pixelsv2);
pixelssum2 = vec_add((vector unsigned short)pixelsv1,
(vector unsigned short)pixelsv2);
temp3 = vec_add(pixelssum1, pixelssum2);
temp3 = vec_sra(temp3, vctwo);
pixelssum1 = vec_add(pixelssum2, vcone);
pixelsavg = vec_packsu(temp3, (vector unsigned short) vczero);
if (rightside) {
blockv = vec_perm(blockv, pixelsavg, vcprm(0, 1, s0, s1));
} else {
blockv = vec_perm(blockv, pixelsavg, vcprm(s0, s1, 2, 3));
}
vec_st(blockv, 0, block);
block += line_size;
pixels += line_size;
}
}
uint32_t
quant_h263_inter_altivec_c(int16_t *coeff,
int16_t *data,
const uint32_t quant,
const uint16_t *mpeg_quant_matrices)
{
vector unsigned char zerovec;
vector unsigned short mult;
vector unsigned short quant_m_2;
vector unsigned short quant_d_2;
vector unsigned short sum_short;
vector signed short acLevel;
vector unsigned int even;
vector unsigned int odd;
vector bool short m2_mask;
vector bool short zero_mask;
uint32_t result;
#ifdef DEBUG
if(((unsigned)coeff) & 0x15)
fprintf(stderr, "quant_h263_inter_altivec_c:incorrect align, coeff: %lx\n", (long)coeff);
#endif
/* initialisation stuff */
zerovec = vec_splat_u8(0);
*((unsigned short*)&mult) = (unsigned short)multipliers[quant];
mult = vec_splat(mult, 0);
*((unsigned short*)&quant_m_2) = (unsigned short)quant;
quant_m_2 = vec_splat(quant_m_2, 0);
quant_m_2 = vec_sl(quant_m_2, vec_splat_u16(1));
*((unsigned short*)&quant_d_2) = (unsigned short)quant;
quant_d_2 = vec_splat(quant_d_2, 0);
quant_d_2 = vec_sr(quant_d_2, vec_splat_u16(1));
sum_short = (vector unsigned short)zerovec;
/* Quantize */
QUANT_H263_INTER_ALTIVEC();
QUANT_H263_INTER_ALTIVEC();
QUANT_H263_INTER_ALTIVEC();
QUANT_H263_INTER_ALTIVEC();
QUANT_H263_INTER_ALTIVEC();
QUANT_H263_INTER_ALTIVEC();
QUANT_H263_INTER_ALTIVEC();
QUANT_H263_INTER_ALTIVEC();
/* Calculate the return value */
even = (vector unsigned int)vec_sum4s((vector signed short)sum_short, (vector signed int)zerovec);
even = (vector unsigned int)vec_sums((vector signed int)even, (vector signed int)zerovec);
even = vec_splat(even, 3);
vec_ste(even, 0, &result);
return result;
}
void YV12_422_Altivec( uint8_t *in, uint8_t *out, uint32_t w,uint32_t h)
{
uint8_t *y,*y2,*u,*v,*out2;
uint32_t dx,dy;
vector unsigned char vecy,vecy2,vecu,vecv,MSQ,mask;
vector unsigned char zero;
#define VEC16 vector unsigned short
#define VEC8 vector unsigned char
#define VECS8 vector signed char
out2=out+w*2;
y=in;
y2=in+w;
u=in+w*h;
v=in+((w*h*5)>>2);
zero=vec_splat_u8(0);
if( (long int)out & 15)
{
printf("Alignment issue in yv12 to 422 altivec!\n");
}
for(dy=h>>1;dy>0;dy--)
{
// We do 4 pix in a raw
for(dx=w>>3;dx>0;dx--)
{
LOAD_ALIGN(vecy,y); // expand
LOAD_ALIGN(vecy2,y2); // expand
LOAD_ALIGN(vecu,v); // expand
LOAD_ALIGN(vecv,u); // expand
vecu=(VEC8)vec_mergeh(vecu,vecv);
vecy=(VEC8)vec_mergeh(vecy,vecu);
vecy2=(VEC8)vec_mergeh(vecy2,vecu);
// Store
vec_st(vecy,0,out);
vec_st(vecy2,0,out2);
// next
out2+=16;
out+=16;
y+=8;
y2+=8;
u+=4;
v+=4;
}
out+=w*2;
out2+=w*2;
y+=w;
y2+=w;
}
}
void foo (void)
{
vector bool int boolVec1 = (vector bool int) vec_splat_u32(3);
vector bool short boolVec2 = (vector bool short) vec_splat_u16(3);
vector bool char boolVec3 = (vector bool char) vec_splat_u8(3);
boolVec1 = vec_sld( boolVec1, boolVec1, 4 );
boolVec2 = vec_sld( boolVec2, boolVec2, 2 );
boolVec3 = vec_sld( boolVec3, boolVec3, 1 );
}
static int sad16_x2_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
{
int i;
int s;
const vector unsigned char zero = (const vector unsigned char)vec_splat_u8(0);
vector unsigned char perm1 = vec_lvsl(0, pix2);
vector unsigned char perm2 = vec_add(perm1, vec_splat_u8(1));
vector unsigned char pix2l, pix2r;
vector unsigned char pix1v, pix2v, pix2iv, avgv, t5;
vector unsigned int sad;
vector signed int sumdiffs;
s = 0;
sad = (vector unsigned int)vec_splat_u32(0);
for (i = 0; i < h; i++) {
/* Read unaligned pixels into our vectors. The vectors are as follows:
pix1v: pix1[0]-pix1[15]
pix2v: pix2[0]-pix2[15] pix2iv: pix2[1]-pix2[16] */
pix1v = vec_ld( 0, pix1);
pix2l = vec_ld( 0, pix2);
pix2r = vec_ld(16, pix2);
pix2v = vec_perm(pix2l, pix2r, perm1);
pix2iv = vec_perm(pix2l, pix2r, perm2);
/* Calculate the average vector */
avgv = vec_avg(pix2v, pix2iv);
/* Calculate a sum of abs differences vector */
t5 = vec_sub(vec_max(pix1v, avgv), vec_min(pix1v, avgv));
/* Add each 4 pixel group together and put 4 results into sad */
sad = vec_sum4s(t5, sad);
pix1 += line_size;
pix2 += line_size;
}
/* Sum up the four partial sums, and put the result into s */
sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero);
sumdiffs = vec_splat(sumdiffs, 3);
vec_ste(sumdiffs, 0, &s);
return s;
}
static inline void abcd2cbad_internal( register const vector unsigned char p, unsigned char *data, unsigned int length, unsigned char *newdata ) {
register vector unsigned char d0,d1,d2,z;
z = vec_splat_u8(0);
length = eround16(length);
if( length >= 3 ) {
length -= 3;
d2 = vec_ld(32,data);
d1 = vec_ld(16,data);
d0 = vec_ld(0,data);
while( length >= 3 ) {
d0 = vec_perm(d0,z,p);
d1 = vec_perm(d1,z,p);
d2 = vec_perm(d2,z,p);
vec_st(d0,0,newdata);
vec_st(d1,16,newdata);
vec_st(d2,32,newdata);
length -= 3;
data += 16*3;
newdata += 16*3;
d2 = vec_ld(32,data);
d1 = vec_ld(16,data);
d0 = vec_ld(0,data);
}
d0 = vec_perm(d0,z,p);
d1 = vec_perm(d1,z,p);
d2 = vec_perm(d2,z,p);
vec_st(d0,0,newdata);
vec_st(d1,16,newdata);
vec_st(d2,32,newdata);
}
if( length == 2 ) {
d0 = vec_ld(0,data);
d1 = vec_ld(16,data);
d0 = vec_perm(d0,z,p);
d1 = vec_perm(d1,z,p);
vec_st(d0,0,newdata);
vec_st(d1,16,newdata);
} else if( length == 1 ) {
d0 = vec_ld(0,data);
d0 = vec_perm(d0,d0,z);
vec_st(d0,0,newdata);
}
}
uint32_t
dequant_h263_inter_altivec_c(int16_t *data,
int16_t *coeff,
const uint32_t quant,
const uint16_t *mpeg_quant_matrices)
{
vector signed short acLevel;
vector signed short vec_2048;
vector unsigned short quant_m_2;
vector unsigned short quant_add;
vector unsigned short t;
register vector unsigned int even;
register vector unsigned int odd;
register vector unsigned int high;
register vector unsigned int low;
register vector unsigned char zerovec;
vector bool short equal_zero;
vector bool short less_zero;
vector bool short overflow;
#ifdef DEBUG
/* print alignment errors if this is on */
if(((unsigned)data) & 0x15)
fprintf(stderr, "dequant_h263_inter_altivec_c:incorrect align, data: %lx\n", (long)data);
#endif
/* initialize */
*((unsigned short*)&quant_m_2) = (unsigned short)(quant << 1);
quant_m_2 = vec_splat(quant_m_2,0);
*((unsigned short*)&quant_add) = (unsigned short)(quant & 1 ? quant : quant - 1);
quant_add = vec_splat(quant_add,0);
vec_2048 = vec_sl(vec_splat_s16(1), vec_splat_u16(11));
zerovec = vec_splat_u8(0);
/* dequant */
DEQUANT_H263_INTER_ALTIVEC();
DEQUANT_H263_INTER_ALTIVEC();
DEQUANT_H263_INTER_ALTIVEC();
DEQUANT_H263_INTER_ALTIVEC();
DEQUANT_H263_INTER_ALTIVEC();
DEQUANT_H263_INTER_ALTIVEC();
DEQUANT_H263_INTER_ALTIVEC();
DEQUANT_H263_INTER_ALTIVEC();
return 0;
}
uint32_t
sad16bi_altivec_c(vector unsigned char *cur,
vector unsigned char *ref1,
vector unsigned char *ref2,
uint32_t stride)
{
vector unsigned char t1, t2;
vector unsigned char mask1, mask2;
vector unsigned char sad;
vector unsigned int sum;
uint32_t result;
#ifdef DEBUG
/* print alignment errors if this is on */
if((long)cur & 0xf)
fprintf(stderr, "sad16bi_altivec:incorrect align, cur: %lx\n", (long)cur);
if(stride & 0xf)
fprintf(stderr, "sad16bi_altivec:incorrect align, cur: %lu\n", stride);
#endif
/* Initialisation stuff */
stride >>= 4;
mask1 = vec_lvsl(0, (unsigned char*)ref1);
mask2 = vec_lvsl(0, (unsigned char*)ref2);
sad = vec_splat_u8(0);
sum = (vector unsigned int)sad;
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
SAD16BI();
sum = (vector unsigned int)vec_sums((vector signed int)sum, vec_splat_s32(0));
sum = vec_splat(sum, 3);
vec_ste(sum, 0, (uint32_t*)&result);
return result;
}
uint32_t
quant_h263_intra_altivec_c(int16_t *coeff,
int16_t *data,
const uint32_t quant,
const uint32_t dcscalar,
const uint16_t *mpeg_quant_matrices)
{
vector unsigned char zerovec;
vector unsigned short mult;
vector unsigned short quant_m_2;
vector signed short acLevel;
register vector unsigned int even;
register vector unsigned int odd;
vector bool short zero_mask;
vector bool short m2_mask;
register int16_t *origin_coeff = coeff;
register int16_t *origin_data = data;
#ifdef DEBUG
if(((unsigned)coeff) & 15)
fprintf(stderr, "quant_h263_intra_altivec_c:incorrect align, coeff: %lx\n", (long)coeff);
#endif
zerovec = vec_splat_u8(0);
*((unsigned short*)&mult) = (unsigned short)multipliers[quant];
mult = vec_splat(mult, 0);
*((unsigned short*)&quant_m_2) = (unsigned short)quant;
quant_m_2 = vec_splat(quant_m_2, 0);
quant_m_2 = vec_sl(quant_m_2, vec_splat_u16(1));
QUANT_H263_INTRA_ALTIVEC();
QUANT_H263_INTRA_ALTIVEC();
QUANT_H263_INTRA_ALTIVEC();
QUANT_H263_INTRA_ALTIVEC();
QUANT_H263_INTRA_ALTIVEC();
QUANT_H263_INTRA_ALTIVEC();
QUANT_H263_INTRA_ALTIVEC();
QUANT_H263_INTRA_ALTIVEC();
// noch erstes setzen
origin_coeff[0] = DIV_DIV(origin_data[0], (int32_t)dcscalar);
return 0;
}
void pix_diff :: processRGBA_Altivec(imageStruct &image, imageStruct &right)
{
int datasize = image.xsize * image.ysize / 4;
vector signed short hiImage, loImage, hiRight, loRight;
vector unsigned char zero = vec_splat_u8(0);
vector unsigned char *inData = (vector unsigned char *)image.data;
vector unsigned char *rightData = (vector unsigned char *)right.data;
#ifndef PPC970
UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 );
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
vec_dst( inData+256, prefetchSize, 2 );
vec_dst( rightData+256, prefetchSize, 3 );
#endif
do {
#ifndef PPC970
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
vec_dst( inData+256, prefetchSize, 2 );
vec_dst( rightData+256, prefetchSize, 3 );
#endif
hiImage = (vector signed short)vec_mergeh(zero,inData[0]);
loImage = (vector signed short)vec_mergel(zero,inData[0]);
hiRight = (vector signed short)vec_mergeh(zero,rightData[0]);
loRight = (vector signed short)vec_mergel(zero,rightData[0]);
hiImage = vec_subs(hiImage,hiRight);
loImage = vec_subs(loImage,loRight);
hiImage = vec_abs(hiImage);
loImage = vec_abs(loImage);
inData[0] = vec_packsu(hiImage,loImage);
inData++;
rightData++;
}
while (--datasize);
#ifndef PPC970
vec_dss( 0 );
vec_dss( 1 );
vec_dss( 2 );
vec_dss( 3 );
#endif
}
static av_always_inline
void put_vp8_epel_h_altivec_core(uint8_t *dst, ptrdiff_t dst_stride,
uint8_t *src, ptrdiff_t src_stride,
int h, int mx, int w, int is6tap)
{
LOAD_H_SUBPEL_FILTER(mx-1);
vec_u8 align_vec0, align_vec8, permh0, permh8, filt;
vec_u8 perm_6tap0, perm_6tap8, perml0, perml8;
vec_u8 a, b, pixh, pixl, outer;
vec_s16 f16h, f16l;
vec_s32 filth, filtl;
vec_u8 perm_inner6 = { 1,2,3,4, 2,3,4,5, 3,4,5,6, 4,5,6,7 };
vec_u8 perm_inner4 = { 0,1,2,3, 1,2,3,4, 2,3,4,5, 3,4,5,6 };
vec_u8 perm_inner = is6tap ? perm_inner6 : perm_inner4;
vec_u8 perm_outer = { 4,9, 0,5, 5,10, 1,6, 6,11, 2,7, 7,12, 3,8 };
vec_s32 c64 = vec_sl(vec_splat_s32(1), vec_splat_u32(6));
vec_u16 c7 = vec_splat_u16(7);
align_vec0 = vec_lvsl( -is6tap-1, src);
align_vec8 = vec_lvsl(8-is6tap-1, src);
permh0 = vec_perm(align_vec0, align_vec0, perm_inner);
permh8 = vec_perm(align_vec8, align_vec8, perm_inner);
perm_inner = vec_add(perm_inner, vec_splat_u8(4));
perml0 = vec_perm(align_vec0, align_vec0, perm_inner);
perml8 = vec_perm(align_vec8, align_vec8, perm_inner);
perm_6tap0 = vec_perm(align_vec0, align_vec0, perm_outer);
perm_6tap8 = vec_perm(align_vec8, align_vec8, perm_outer);
while (h --> 0) {
FILTER_H(f16h, 0);
if (w == 16) {
FILTER_H(f16l, 8);
filt = vec_packsu(f16h, f16l);
vec_st(filt, 0, dst);
} else {
filt = vec_packsu(f16h, f16h);
vec_ste((vec_u32)filt, 0, (uint32_t*)dst);
if (w == 8)
vec_ste((vec_u32)filt, 4, (uint32_t*)dst);
}
src += src_stride;
dst += dst_stride;
}
}
int pix_abs16x16_x2_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
{
int i;
int s __attribute__((aligned(16)));
const vector unsigned char zero = (const vector unsigned char)vec_splat_u8(0);
vector unsigned char *tv;
vector unsigned char pix1v, pix2v, pix2iv, avgv, t5;
vector unsigned int sad;
vector signed int sumdiffs;
s = 0;
sad = (vector unsigned int)vec_splat_u32(0);
for(i=0;i<16;i++) {
/*
Read unaligned pixels into our vectors. The vectors are as follows:
pix1v: pix1[0]-pix1[15]
pix2v: pix2[0]-pix2[15] pix2iv: pix2[1]-pix2[16]
*/
tv = (vector unsigned char *) pix1;
pix1v = vec_perm(tv[0], tv[1], vec_lvsl(0, pix1));
tv = (vector unsigned char *) &pix2[0];
pix2v = vec_perm(tv[0], tv[1], vec_lvsl(0, &pix2[0]));
tv = (vector unsigned char *) &pix2[1];
pix2iv = vec_perm(tv[0], tv[1], vec_lvsl(0, &pix2[1]));
/* Calculate the average vector */
avgv = vec_avg(pix2v, pix2iv);
/* Calculate a sum of abs differences vector */
t5 = vec_sub(vec_max(pix1v, avgv), vec_min(pix1v, avgv));
/* Add each 4 pixel group together and put 4 results into sad */
sad = vec_sum4s(t5, sad);
pix1 += line_size;
pix2 += line_size;
}
/* Sum up the four partial sums, and put the result into s */
sumdiffs = vec_sums((vector signed int) sad, (vector signed int) zero);
sumdiffs = vec_splat(sumdiffs, 3);
vec_ste(sumdiffs, 0, &s);
return s;
}
void idct_add_altivec(uint8_t* dest, int stride, int16_t *blk)
{
POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
vec_s16 *block = (vec_s16*)blk;
vec_u8 tmp;
vec_s16 tmp2, tmp3;
vec_u8 perm0;
vec_u8 perm1;
vec_u8 p0, p1, p;
#if CONFIG_POWERPC_PERF
POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
#endif
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)
POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
}
static void vp3_idct_add_altivec(uint8_t *dst, int stride, int16_t block[64])
{
LOAD_ZERO;
vec_u8 t, vdst;
vec_s16 vdst_16;
vec_u8 vdst_mask = vec_mergeh(vec_splat_u8(-1), vec_lvsl(0, dst));
IDCT_START
IDCT_1D(NOP, NOP)
TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7);
IDCT_1D(ADD8, SHIFT4)
#if HAVE_BIGENDIAN
#define GET_VDST16\
vdst = vec_ld(0, dst);\
vdst_16 = (vec_s16)vec_perm(vdst, zero_u8v, vdst_mask);
#else
#define GET_VDST16\
vdst = vec_vsx_ld(0,dst);\
vdst_16 = (vec_s16)vec_mergeh(vdst, zero_u8v);
#endif
#define ADD(a)\
GET_VDST16;\
vdst_16 = vec_adds(a, vdst_16);\
t = vec_packsu(vdst_16, vdst_16);\
vec_ste((vec_u32)t, 0, (unsigned int *)dst);\
vec_ste((vec_u32)t, 4, (unsigned int *)dst);
ADD(b0) dst += stride;
ADD(b1) dst += stride;
ADD(b2) dst += stride;
ADD(b3) dst += stride;
ADD(b4) dst += stride;
ADD(b5) dst += stride;
ADD(b6) dst += stride;
ADD(b7)
memset(block, 0, sizeof(*block) * 64);
}
/* 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)
{
int i;
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, srcvB, srcvC, srcvD;
register vector unsigned short tempB, tempC, tempD;
unsigned long dst_odd = (unsigned long) dst & 0x0000000F;
unsigned long src_really_odd = (unsigned long) src & 0x0000000F;
register vector unsigned short tempA =
vec_ld(0, (const unsigned short *) ABCD);
register vector unsigned short Av = vec_splat(tempA, 0);
register vector unsigned short Bv = vec_splat(tempA, 1);
register vector unsigned short Cv = vec_splat(tempA, 2);
register vector unsigned short Dv = vec_splat(tempA, 3);
register vector unsigned short 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
* reuse 'src + stride' from one iteration as the 'src' of the next. */
register vector unsigned char src_0 = vec_ld(0, src);
register vector unsigned char src_1 = vec_ld(16, src);
register vector unsigned char srcvA = vec_perm(src_0, src_1,
vec_lvsl(0, src));
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
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;
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_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 & 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
srcvD = src_1;
srcvC = vec_mergeh(vczero, srcvC);
srcvD = vec_mergeh(vczero, srcvD);
/* 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);
srcvA = srcvC;
srcvB = srcvD;
tempD = vec_sr(tempD, vcsr8);
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));
vec_st(dstv2, 0, dst);
dst += stride;
src += stride;
}
}
void pix_background :: processYUVAltivec(imageStruct &image)
{
register int h,w,i,j,width;
int pixsize = image.xsize * image.ysize * image.csize;
h = image.ysize;
w = image.xsize/8;
width = image.xsize/8;
//check to see if the buffer isn't 16byte aligned (highly unlikely)
if (image.ysize*image.xsize % 16 != 0){
error("image not properly aligned for Altivec - try something SD or HD maybe?");
return;
}
union{
unsigned short s[8];
vector unsigned short v;
}shortBuffer;
if(m_savedImage.xsize!=image.xsize ||
m_savedImage.ysize!=image.ysize ||
m_savedImage.format!=image.format)m_reset=1;
m_savedImage.xsize=image.xsize;
m_savedImage.ysize=image.ysize;
m_savedImage.setCsizeByFormat(image.format);
m_savedImage.reallocate();
if (m_reset){
memcpy(m_savedImage.data,image.data,pixsize);
m_reset = 0;
}
register vector unsigned short UVres1, Yres1, UVres2, Yres2;//interleave;
register vector unsigned short hiImage, loImage;
register vector unsigned short Yrange, UVrange, Yblank,UVblank,blank;
register vector bool short Ymasklo,Ymaskhi, UVmaskhi;
register vector unsigned short Yhi,Ylo,UVhi,UVlo;
register vector unsigned char one = vec_splat_u8(1);
register vector unsigned short sone = vec_splat_u16(1);
register vector unsigned int Uhi, Ulo, Vhi, Vlo,Ures,Vres;
register vector bool int Umasklo, Umaskhi, Vmaskhi, Vmasklo;
vector unsigned char *inData = (vector unsigned char*) image.data;
vector unsigned char *rightData = (vector unsigned char*) m_savedImage.data;
shortBuffer.s[0] = m_Yrange;
Yrange = shortBuffer.v;
Yrange = vec_splat(Yrange,0);
shortBuffer.s[0] = 128;
shortBuffer.s[1] = 0;
shortBuffer.s[2] = 128;
shortBuffer.s[3] = 0;
shortBuffer.s[4] = 128;
shortBuffer.s[5] = 0;
shortBuffer.s[6] = 128;
shortBuffer.s[7] = 0;
blank = shortBuffer.v;
shortBuffer.s[0] = 0;
Yblank = shortBuffer.v;
Yblank = vec_splat(Yblank,0);
shortBuffer.s[0] = 128;
UVblank = shortBuffer.v;
UVblank = vec_splat(UVblank,0);
shortBuffer.s[0] = m_Urange;
shortBuffer.s[1] = m_Vrange;
shortBuffer.s[2] = m_Urange;
shortBuffer.s[3] = m_Vrange;
shortBuffer.s[4] = m_Urange;
shortBuffer.s[5] = m_Vrange;
shortBuffer.s[6] = m_Urange;
shortBuffer.s[7] = m_Vrange;
UVrange = shortBuffer.v;
//setup the cache prefetch -- A MUST!!!
UInt32 prefetchSize = GetPrefetchConstant( 16, 1, 256 );
#ifndef PPC970
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
vec_dst( inData+32, prefetchSize, 2 );
vec_dst( rightData+32, prefetchSize, 3 );
#endif //PPC970
for ( i=0; i<h; i++){
for (j=0; j<w; j++)
{
#ifndef PPC970
//this function is probably memory bound on most G4's -- what else is new?
vec_dst( inData, prefetchSize, 0 );
vec_dst( rightData, prefetchSize, 1 );
vec_dst( inData+32, prefetchSize, 2 );
vec_dst( rightData+32, prefetchSize, 3 );
#endif
//separate the U and V from Y
UVres1 = (vector unsigned short)vec_mule(one,inData[0]);
UVres2 = (vector unsigned short)vec_mule(one,rightData[0]);
//vec_mulo Y * 1 to short vector Y Y Y Y shorts
Yres1 = (vector unsigned short)vec_mulo(one,inData[0]);
Yres2 = (vector unsigned short)vec_mulo(one,rightData[0]);
Yhi = vec_adds(Yres2,Yrange);
Ylo = vec_subs(Yres2,Yrange);
//go to ints for comparison
UVhi = vec_adds(UVres2,UVrange);
UVlo = vec_subs(UVres2,UVrange);
Uhi = vec_mule(sone,UVhi);
Ulo = vec_mule(sone,UVlo);
Vhi = vec_mulo(sone,UVhi);
Vlo = vec_mulo(sone,UVlo);
Ures = vec_mule(sone,UVres1);
Vres = vec_mulo(sone,UVres1);
Umasklo = vec_cmpgt(Ures,Ulo);
Umaskhi = vec_cmplt(Ures,Uhi);
Vmasklo = vec_cmpgt(Vres,Vlo);
Vmaskhi = vec_cmplt(Vres,Vhi);
Umaskhi = vec_and(Umaskhi,Umasklo);
Vmaskhi = vec_and(Vmaskhi,Vmasklo);
Umasklo = vec_and(Umaskhi,Vmaskhi);
Vmasklo = vec_and(Umaskhi,Vmaskhi);
hiImage = (vector unsigned short)vec_mergeh(Umasklo,Vmasklo);
loImage = (vector unsigned short)vec_mergel(Umasklo,Vmasklo);
//pack it back down to bool short
UVmaskhi = (vector bool short)vec_packsu(hiImage,loImage);
Ymasklo = vec_cmpgt(Yres1,Ylo);
Ymaskhi = vec_cmplt(Yres1,Yhi);
Ymaskhi = vec_and(Ymaskhi,Ymasklo);
Ymaskhi = vec_and(Ymaskhi,UVmaskhi);
UVmaskhi = vec_and(Ymaskhi,UVmaskhi);
//bitwise comparison and move using the result of the comparison as a mask
Yres1 = vec_sel(Yres1,Yblank,Ymaskhi);
//UVres1 = vec_sel(UVres1,UVres2,UVmaskhi);
UVres1 = vec_sel(UVres1,UVblank,UVmaskhi);
//merge the Y and UV back together
hiImage = vec_mergeh(UVres1,Yres1);
loImage = vec_mergel(UVres1,Yres1);
//pack it back down to unsigned char to store
inData[0] = vec_packsu(hiImage,loImage);
inData++;
rightData++;
}
#ifndef PPC970
vec_dss(0);
vec_dss(1);
vec_dss(2);
vec_dss(3);
#endif
}
}
void gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int stride, int h, int x16, int y16, int rounder)
{
POWERPC_PERF_DECLARE(altivec_gmc1_num, GMC1_PERF_COND);
const DECLARE_ALIGNED_16(unsigned short, rounder_a[8]) =
{rounder, rounder, rounder, rounder,
rounder, rounder, rounder, 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;
int i;
unsigned long dst_odd = (unsigned long)dst & 0x0000000F;
unsigned long src_really_odd = (unsigned long)src & 0x0000000F;
POWERPC_PERF_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
tempA = vec_ld(0, (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_ld(0, (unsigned short*)rounder_a);
// 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 re-use 'src+stride' from one iteration
// as the 'src' of the next.
src_0 = vec_ld(0, src);
src_1 = vec_ld(16, src);
srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src));
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
{
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;
dstv = vec_ld(0, dst);
// we 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_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 & 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
{
srcvD = src_1;
}
srcvC = vec_mergeh(vczero, srcvC);
srcvD = vec_mergeh(vczero, srcvD);
// OK, now we (finally) do the math :-)
// those four instructions replaces 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);
srcvA = srcvC;
srcvB = srcvD;
tempD = vec_sr(tempD, vcsr8);
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));
}
vec_st(dstv2, 0, dst);
dst += stride;
src += stride;
}
POWERPC_PERF_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
}
vector unsigned char
testuc_3 ()
{
return vec_splat_u8 (15);
}
vector unsigned char
testuc_2 ()
{
return vec_splat_u8 (-5);
}