comparison ppc/dsputil_altivec.c @ 1033:b4172ff70d27 libavcodec

Altivec on non darwin systems patch by Romain Dolbeau

1032:8f440ca8e0b0

 */
 
#include "../dsputil.h"
#include "dsputil_altivec.h"

#if CONFIG_DARWIN
#include <sys/sysctl.h>
#endif

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)(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)(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]

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)(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)(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

int pix_abs16x16_xy2_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    uint8_t *pix3 = pix2 + line_size;
    const vector unsigned char zero = (const vector unsigned char)(0);
    const vector unsigned short two = (const vector unsigned short)(2);
    vector unsigned char *tv, avgv, t5;
    vector unsigned char pix1v, pix2v, pix3v, pix2iv, pix3iv;
    vector unsigned short pix2lv, pix2hv, pix2ilv, pix2ihv;
    vector unsigned short pix3lv, pix3hv, pix3ilv, pix3ihv;
    vector unsigned short avghv, avglv;
    vector unsigned short t1, t2, t3, t4;
    vector unsigned int sad;
    vector signed int sumdiffs;

    sad = (vector unsigned int)(0);
    
    s = 0;

    /*
       Due to the fact that pix3 = pix2 + line_size, the pix3 of one

int pix_abs16x16_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)(0);
    vector unsigned char perm1, perm2, *pix1v, *pix2v;
    vector unsigned char t1, t2, t3,t4, t5;
    vector unsigned int sad;
    vector signed int sumdiffs;
    
    sad = (vector unsigned int) (0);


    for(i=0;i<16;i++) {
	/* Read potentially unaligned pixels into t1 and t2 */
        perm1 = vec_lvsl(0, pix1);

int pix_abs8x8_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)(0);
    vector unsigned char perm1, perm2, permclear, *pix1v, *pix2v;
    vector unsigned char t1, t2, t3,t4, t5;
    vector unsigned int sad;
    vector signed int sumdiffs;

    sad = (vector unsigned int)(0);
    permclear = (vector unsigned char) (255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0);

    for(i=0;i<8;i++) {
	/* Read potentially unaligned pixels into t1 and t2
	   Since we're reading 16 pixels, and actually only want 8,
	   mask out the last 8 pixels. The 0s don't change the sum. */

int pix_norm1_altivec(uint8_t *pix, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)(0);
    vector unsigned char *tv;
    vector unsigned char pixv;
    vector unsigned int sv;
    vector signed int sum;
    
    sv = (vector unsigned int)(0);
    
    s = 0;
    for (i = 0; i < 16; i++) {
        /* Read in the potentially unaligned pixels */
        tv = (vector unsigned char *) pix;

 */
int sse8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)(0);
    vector unsigned char perm1, perm2, permclear, *pix1v, *pix2v;
    vector unsigned char t1, t2, t3,t4, t5;
    vector unsigned int sum;
    vector signed int sumsqr;
    
    sum = (vector unsigned int)(0);
    permclear = (vector unsigned char)(0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0xFF,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00);
    
    for(i=0;i<8;i++) {
	/* Read potentially unaligned pixels into t1 and t2
	   Since we're reading 16 pixels, and actually only want 8,
	   mask out the last 8 pixels. The 0s don't change the sum. */

 */
int sse16_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)(0);
    vector unsigned char perm1, perm2, *pix1v, *pix2v;
    vector unsigned char t1, t2, t3,t4, t5;
    vector unsigned int sum;
    vector signed int sumsqr;
    
    sum = (vector unsigned int)(0);
    
    for(i=0;i<16;i++) {
	/* Read potentially unaligned pixels into t1 and t2 */
        perm1 = vec_lvsl(0, pix1);
        pix1v = (vector unsigned char *) pix1;

    return s;
}

int pix_sum_altivec(UINT8 * pix, int line_size)
{
    const vector unsigned int zero = (const vector unsigned int)(0);
    vector unsigned char perm, *pixv;
    vector unsigned char t1;
    vector unsigned int sad;
    vector signed int sumdiffs;

    int i;
    int s __attribute__((aligned(16)));
    
    sad = (vector unsigned int) (0);
    
    for (i = 0; i < 16; i++) {
	/* Read the potentially unaligned 16 pixels into t1 */
        perm = vec_lvsl(0, pix);
        pixv = (vector unsigned char *) pix;

void get_pixels_altivec(DCTELEM *restrict block, const UINT8 *pixels, int line_size)
{
    int i;
    vector unsigned char perm, bytes, *pixv;
    const vector unsigned char zero = (const vector unsigned char) (0);
    vector signed short shorts;

    for(i=0;i<8;i++)
    {
        // Read potentially unaligned pixels.

void diff_pixels_altivec(DCTELEM *restrict block, const UINT8 *s1,
        const UINT8 *s2, int stride)
{
    int i;
    vector unsigned char perm, bytes, *pixv;
    const vector unsigned char zero = (const vector unsigned char) (0);
    vector signed short shorts1, shorts2;

    for(i=0;i<4;i++)
    {
        // Read potentially unaligned pixels

     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)(0);
   register const vector unsigned short vctwo = (const vector unsigned short)(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)

     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)(0);
   register const vector unsigned short vcone = (const vector unsigned short)(1);
   register const vector unsigned short vctwo = (const vector unsigned short)(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)

   register vector unsigned char
     blockv, temp1, temp2;
   register vector unsigned short
     pixelssum1, pixelssum2, temp3,
     pixelssum3, pixelssum4, temp4;
   register const vector unsigned char vczero = (const vector unsigned char)(0);
   register const vector unsigned short vctwo = (const vector unsigned short)(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)

   register vector unsigned char
     blockv, temp1, temp2;
   register vector unsigned short
     pixelssum1, pixelssum2, temp3,
     pixelssum3, pixelssum4, temp4;
   register const vector unsigned char vczero = (const vector unsigned char)(0);
   register const vector unsigned short vcone = (const vector unsigned short)(1);
   register const vector unsigned short vctwo = (const vector unsigned short)(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)

#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

int has_altivec(void)
{
#if CONFIG_DARWIN
    int sels[2] = {CTL_HW, HW_VECTORUNIT};
    int has_vu = 0;
    size_t len = sizeof(has_vu);
    int err;

    err = sysctl(sels, 2, &has_vu, &len, NULL, 0);

    if (err == 0) return (has_vu != 0);
#endif
    return 0;
}

1033:b4172ff70d27

 */
 
#include "../dsputil.h"
#include "dsputil_altivec.h"

#ifdef CONFIG_DARWIN
#include <sys/sysctl.h>
#else /* CONFIG_DARWIN */
#include <signal.h>
#include <setjmp.h>

static sigjmp_buf jmpbuf;
static volatile sig_atomic_t canjump = 0;

static void sigill_handler (int sig)
{
    if (!canjump) {
        signal (sig, SIG_DFL);
        raise (sig);
    }
    
    canjump = 0;
    siglongjmp (jmpbuf, 1);
}
#endif /* CONFIG_DARWIN */

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]

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

int pix_abs16x16_xy2_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    uint8_t *pix3 = pix2 + line_size;
    const vector unsigned char zero = (const vector unsigned char)vec_splat_u8(0);
    const vector unsigned short two = (const vector unsigned short)vec_splat_u16(2);
    vector unsigned char *tv, avgv, t5;
    vector unsigned char pix1v, pix2v, pix3v, pix2iv, pix3iv;
    vector unsigned short pix2lv, pix2hv, pix2ilv, pix2ihv;
    vector unsigned short pix3lv, pix3hv, pix3ilv, pix3ihv;
    vector unsigned short avghv, avglv;
    vector unsigned short t1, t2, t3, t4;
    vector unsigned int sad;
    vector signed int sumdiffs;

    sad = (vector unsigned int)vec_splat_u32(0);
    
    s = 0;

    /*
       Due to the fact that pix3 = pix2 + line_size, the pix3 of one

int pix_abs16x16_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)vec_splat_u32(0);
    vector unsigned char perm1, perm2, *pix1v, *pix2v;
    vector unsigned char t1, t2, t3,t4, t5;
    vector unsigned int sad;
    vector signed int sumdiffs;
    
    sad = (vector unsigned int)vec_splat_u32(0);


    for(i=0;i<16;i++) {
	/* Read potentially unaligned pixels into t1 and t2 */
        perm1 = vec_lvsl(0, pix1);

int pix_abs8x8_altivec(uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)vec_splat_u32(0);
    vector unsigned char perm1, perm2, permclear, *pix1v, *pix2v;
    vector unsigned char t1, t2, t3,t4, t5;
    vector unsigned int sad;
    vector signed int sumdiffs;

    sad = (vector unsigned int)vec_splat_u32(0);
#ifdef CONFIG_DARWIN
    permclear = (vector unsigned char)(255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0);
#else
    permclear = (vector unsigned char){255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0};
#endif

    for(i=0;i<8;i++) {
	/* Read potentially unaligned pixels into t1 and t2
	   Since we're reading 16 pixels, and actually only want 8,
	   mask out the last 8 pixels. The 0s don't change the sum. */

int pix_norm1_altivec(uint8_t *pix, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)vec_splat_u32(0);
    vector unsigned char *tv;
    vector unsigned char pixv;
    vector unsigned int sv;
    vector signed int sum;
    
    sv = (vector unsigned int)vec_splat_u32(0);
    
    s = 0;
    for (i = 0; i < 16; i++) {
        /* Read in the potentially unaligned pixels */
        tv = (vector unsigned char *) pix;

 */
int sse8_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)vec_splat_u32(0);
    vector unsigned char perm1, perm2, permclear, *pix1v, *pix2v;
    vector unsigned char t1, t2, t3,t4, t5;
    vector unsigned int sum;
    vector signed int sumsqr;
    
    sum = (vector unsigned int)vec_splat_u32(0);
#ifdef CONFIG_DARWIN
    permclear = (vector unsigned char)(255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0);
#else
    permclear = (vector unsigned char){255,255,255,255,255,255,255,255,0,0,0,0,0,0,0,0};
#endif
    
    for(i=0;i<8;i++) {
	/* Read potentially unaligned pixels into t1 and t2
	   Since we're reading 16 pixels, and actually only want 8,
	   mask out the last 8 pixels. The 0s don't change the sum. */

 */
int sse16_altivec(void *v, uint8_t *pix1, uint8_t *pix2, int line_size)
{
    int i;
    int s __attribute__((aligned(16)));
    const vector unsigned int zero = (const vector unsigned int)vec_splat_u32(0);
    vector unsigned char perm1, perm2, *pix1v, *pix2v;
    vector unsigned char t1, t2, t3,t4, t5;
    vector unsigned int sum;
    vector signed int sumsqr;
    
    sum = (vector unsigned int)vec_splat_u32(0);
    
    for(i=0;i<16;i++) {
	/* Read potentially unaligned pixels into t1 and t2 */
        perm1 = vec_lvsl(0, pix1);
        pix1v = (vector unsigned char *) pix1;

    return s;
}

int pix_sum_altivec(UINT8 * pix, int line_size)
{
    const vector unsigned int zero = (const vector unsigned int)vec_splat_u32(0);
    vector unsigned char perm, *pixv;
    vector unsigned char t1;
    vector unsigned int sad;
    vector signed int sumdiffs;

    int i;
    int s __attribute__((aligned(16)));
    
    sad = (vector unsigned int)vec_splat_u32(0);
    
    for (i = 0; i < 16; i++) {
	/* Read the potentially unaligned 16 pixels into t1 */
        perm = vec_lvsl(0, pix);
        pixv = (vector unsigned char *) pix;

void get_pixels_altivec(DCTELEM *restrict block, const UINT8 *pixels, int line_size)
{
    int i;
    vector unsigned char perm, bytes, *pixv;
    const vector unsigned char zero = (const vector unsigned char)vec_splat_u8(0);
    vector signed short shorts;

    for(i=0;i<8;i++)
    {
        // Read potentially unaligned pixels.

void diff_pixels_altivec(DCTELEM *restrict block, const UINT8 *s1,
        const UINT8 *s2, int stride)
{
    int i;
    vector unsigned char perm, bytes, *pixv;
    const vector unsigned char zero = (const vector unsigned char)vec_splat_u8(0);
    vector signed short shorts1, shorts2;

    for(i=0;i<4;i++)
    {
        // Read potentially unaligned pixels

     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 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)

     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)

   register vector unsigned char
     blockv, temp1, temp2;
   register vector unsigned short
     pixelssum1, pixelssum2, temp3,
     pixelssum3, pixelssum4, temp4;
   register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
   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)

   register vector unsigned char
     blockv, temp1, temp2;
   register vector unsigned short
     pixelssum1, pixelssum2, temp3,
     pixelssum3, pixelssum4, temp4;
   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)

#endif /* ALTIVEC_USE_REFERENCE_C_CODE */
}

int has_altivec(void)
{
#ifdef CONFIG_DARWIN
    int sels[2] = {CTL_HW, HW_VECTORUNIT};
    int has_vu = 0;
    size_t len = sizeof(has_vu);
    int err;

    err = sysctl(sels, 2, &has_vu, &len, NULL, 0);

    if (err == 0) return (has_vu != 0);
#else /* CONFIG_DARWIN */
/* no Darwin, do it the brute-force way */
/* this is borrowed from the libmpeg2 library */
    {
      signal (SIGILL, sigill_handler);
      if (sigsetjmp (jmpbuf, 1)) {
        signal (SIGILL, SIG_DFL);
      } else {
        canjump = 1;
        
        asm volatile ("mtspr 256, %0\n\t"
                      "vand %%v0, %%v0, %%v0"
                      :
                      : "r" (-1));
        
        signal (SIGILL, SIG_DFL);
        return 1;
      }
    }
#endif /* CONFIG_DARWIN */
    return 0;
}