Mercurial > libavcodec.hg

comparison ppc/dsputil_altivec.c @ 3346:052765f11f1c libavcodec

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Cosmetics: should not hurt performance, scream if are
author lu_zero
date Sat, 17 Jun 2006 18:46:06 +0000
parents 0b482ccd7f0e
children 0937cc91b574
comparison
equal deleted inserted replaced
3345:76620d530d9a 3346:052765f11f1c
1309 } 1309 }
1310 1310
1311 int hadamard8_diff8x8_altivec(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){ 1311 int hadamard8_diff8x8_altivec(/*MpegEncContext*/ void *s, uint8_t *dst, uint8_t *src, int stride, int h){
1312 POWERPC_PERF_DECLARE(altivec_hadamard8_diff8x8_num, 1); 1312 POWERPC_PERF_DECLARE(altivec_hadamard8_diff8x8_num, 1);
1313 int sum; 1313 int sum;
1314 POWERPC_PERF_START_COUNT(altivec_hadamard8_diff8x8_num, 1);
1315 register const_vector unsigned char vzero = (const_vector unsigned char)vec_splat_u8(0); 1314 register const_vector unsigned char vzero = (const_vector unsigned char)vec_splat_u8(0);
1316 register vector signed short temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7; 1315 register vector signed short temp0, temp1, temp2, temp3, temp4, temp5, temp6, temp7;
1316 POWERPC_PERF_START_COUNT(altivec_hadamard8_diff8x8_num, 1);
1317 { 1317 {
1318 register const_vector signed short vprod1 = (const_vector signed short)AVV( 1,-1, 1,-1, 1,-1, 1,-1); 1318 register const_vector signed short vprod1 = (const_vector signed short)AVV( 1,-1, 1,-1, 1,-1, 1,-1);
1319 register const_vector signed short vprod2 = (const_vector signed short)AVV( 1, 1,-1,-1, 1, 1,-1,-1); 1319 register const_vector signed short vprod2 = (const_vector signed short)AVV( 1, 1,-1,-1, 1, 1,-1,-1);
1320 register const_vector signed short vprod3 = (const_vector signed short)AVV( 1, 1, 1, 1,-1,-1,-1,-1); 1320 register const_vector signed short vprod3 = (const_vector signed short)AVV( 1, 1, 1, 1,-1,-1,-1,-1);
1321 register const_vector unsigned char perm1 = (const_vector unsigned char) 1321 register const_vector unsigned char perm1 = (const_vector unsigned char)
1336 1336
1337 #define ONEITERBUTTERFLY(i, res) \ 1337 #define ONEITERBUTTERFLY(i, res) \
1338 { \ 1338 { \
1339 register vector unsigned char src1, src2, srcO; \ 1339 register vector unsigned char src1, src2, srcO; \
1340 register vector unsigned char dst1, dst2, dstO; \ 1340 register vector unsigned char dst1, dst2, dstO; \
1341 register vector signed short srcV, dstV; \
1342 register vector signed short but0, but1, but2, op1, op2, op3; \
1341 src1 = vec_ld(stride * i, src); \ 1343 src1 = vec_ld(stride * i, src); \
1342 if ((((stride * i) + (unsigned long)src) & 0x0000000F) > 8) \ 1344 if ((((stride * i) + (unsigned long)src) & 0x0000000F) > 8) \
1343 src2 = vec_ld((stride * i) + 16, src); \ 1345 src2 = vec_ld((stride * i) + 16, src); \
1344 srcO = vec_perm(src1, src2, vec_lvsl(stride * i, src)); \ 1346 srcO = vec_perm(src1, src2, vec_lvsl(stride * i, src)); \
1345 dst1 = vec_ld(stride * i, dst); \ 1347 dst1 = vec_ld(stride * i, dst); \
1346 if ((((stride * i) + (unsigned long)dst) & 0x0000000F) > 8) \ 1348 if ((((stride * i) + (unsigned long)dst) & 0x0000000F) > 8) \
1347 dst2 = vec_ld((stride * i) + 16, dst); \ 1349 dst2 = vec_ld((stride * i) + 16, dst); \
1348 dstO = vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \ 1350 dstO = vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \
1349 /* promote the unsigned chars to signed shorts */ \ 1351 /* promote the unsigned chars to signed shorts */ \
1350 /* we're in the 8x8 function, we only care for the first 8 */ \ 1352 /* we're in the 8x8 function, we only care for the first 8 */ \
1351 register vector signed short srcV = \ 1353 srcV = \
1352 (vector signed short)vec_mergeh((vector signed char)vzero, (vector signed char)srcO); \ 1354 (vector signed short)vec_mergeh((vector signed char)vzero, \
1353 register vector signed short dstV = \ 1355 (vector signed char)srcO); \
1354 (vector signed short)vec_mergeh((vector signed char)vzero, (vector signed char)dstO); \ 1356 dstV = \
1357 (vector signed short)vec_mergeh((vector signed char)vzero, \
1358 (vector signed char)dstO); \
1355 /* substractions inside the first butterfly */ \ 1359 /* substractions inside the first butterfly */ \
1356 register vector signed short but0 = vec_sub(srcV, dstV); \ 1360 but0 = vec_sub(srcV, dstV); \
1357 register vector signed short op1 = vec_perm(but0, but0, perm1); \ 1361 op1 = vec_perm(but0, but0, perm1); \
1358 register vector signed short but1 = vec_mladd(but0, vprod1, op1); \ 1362 but1 = vec_mladd(but0, vprod1, op1); \
1359 register vector signed short op2 = vec_perm(but1, but1, perm2); \ 1363 op2 = vec_perm(but1, but1, perm2); \
1360 register vector signed short but2 = vec_mladd(but1, vprod2, op2); \ 1364 but2 = vec_mladd(but1, vprod2, op2); \
1361 register vector signed short op3 = vec_perm(but2, but2, perm3); \ 1365 op3 = vec_perm(but2, but2, perm3); \
1362 res = vec_mladd(but2, vprod3, op3); \ 1366 res = vec_mladd(but2, vprod3, op3); \
1363 } 1367 }
1364 ONEITERBUTTERFLY(0, temp0); 1368 ONEITERBUTTERFLY(0, temp0);
1365 ONEITERBUTTERFLY(1, temp1); 1369 ONEITERBUTTERFLY(1, temp1);
1366 ONEITERBUTTERFLY(2, temp2); 1370 ONEITERBUTTERFLY(2, temp2);
1479 0x00, 0x01, 0x02, 0x03, 1483 0x00, 0x01, 0x02, 0x03,
1480 0x04, 0x05, 0x06, 0x07); 1484 0x04, 0x05, 0x06, 0x07);
1481 1485
1482 #define ONEITERBUTTERFLY(i, res1, res2) \ 1486 #define ONEITERBUTTERFLY(i, res1, res2) \
1483 { \ 1487 { \
1484 register vector unsigned char src1 REG_v(v22), src2 REG_v(v23); \ 1488 register vector unsigned char src1 REG_v(v22), \
1485 register vector unsigned char dst1 REG_v(v24), dst2 REG_v(v25); \ 1489 src2 REG_v(v23), \
1490 dst1 REG_v(v24), \
1491 dst2 REG_v(v25), \
1492 srcO REG_v(v22), \
1493 dstO REG_v(v23); \
1494 \
1495 register vector signed short srcV REG_v(v24), \
1496 dstV REG_v(v25), \
1497 srcW REG_v(v26), \
1498 dstW REG_v(v27), \
1499 but0 REG_v(v28), \
1500 but0S REG_v(v29), \
1501 op1 REG_v(v30), \
1502 but1 REG_v(v22), \
1503 op1S REG_v(v23), \
1504 but1S REG_v(v24), \
1505 op2 REG_v(v25), \
1506 but2 REG_v(v26), \
1507 op2S REG_v(v27), \
1508 but2S REG_v(v28), \
1509 op3 REG_v(v29), \
1510 op3S REG_v(v30); \
1511 \
1486 src1 = vec_ld(stride * i, src); \ 1512 src1 = vec_ld(stride * i, src); \
1487 src2 = vec_ld((stride * i) + 16, src); \ 1513 src2 = vec_ld((stride * i) + 16, src); \
1488 register vector unsigned char srcO REG_v(v22) = vec_perm(src1, src2, vec_lvsl(stride * i, src)); \ 1514 srcO = vec_perm(src1, src2, vec_lvsl(stride * i, src)); \
1489 dst1 = vec_ld(stride * i, dst); \ 1515 dst1 = vec_ld(stride * i, dst); \
1490 dst2 = vec_ld((stride * i) + 16, dst); \ 1516 dst2 = vec_ld((stride * i) + 16, dst); \
1491 register vector unsigned char dstO REG_v(v23) = vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \ 1517 dstO = vec_perm(dst1, dst2, vec_lvsl(stride * i, dst)); \
1492 /* promote the unsigned chars to signed shorts */ \ 1518 /* promote the unsigned chars to signed shorts */ \
1493 register vector signed short srcV REG_v(v24) = \ 1519 srcV = \
1494 (vector signed short)vec_mergeh((vector signed char)vzero, (vector signed char)srcO); \ 1520 (vector signed short)vec_mergeh((vector signed char)vzero, \
1495 register vector signed short dstV REG_v(v25) = \ 1521 (vector signed char)srcO); \
1496 (vector signed short)vec_mergeh((vector signed char)vzero, (vector signed char)dstO); \ 1522 dstV = \
1497 register vector signed short srcW REG_v(v26) = \ 1523 (vector signed short)vec_mergeh((vector signed char)vzero, \
1498 (vector signed short)vec_mergel((vector signed char)vzero, (vector signed char)srcO); \ 1524 (vector signed char)dstO); \
1499 register vector signed short dstW REG_v(v27) = \ 1525 srcW = \
1500 (vector signed short)vec_mergel((vector signed char)vzero, (vector signed char)dstO); \ 1526 (vector signed short)vec_mergel((vector signed char)vzero, \
1527 (vector signed char)srcO); \
1528 dstW = \
1529 (vector signed short)vec_mergel((vector signed char)vzero, \
1530 (vector signed char)dstO); \
1501 /* substractions inside the first butterfly */ \ 1531 /* substractions inside the first butterfly */ \
1502 register vector signed short but0 REG_v(v28) = vec_sub(srcV, dstV); \ 1532 but0 = vec_sub(srcV, dstV); \
1503 register vector signed short but0S REG_v(v29) = vec_sub(srcW, dstW); \ 1533 but0S = vec_sub(srcW, dstW); \
1504 register vector signed short op1 REG_v(v30) = vec_perm(but0, but0, perm1); \ 1534 op1 = vec_perm(but0, but0, perm1); \
1505 register vector signed short but1 REG_v(v22) = vec_mladd(but0, vprod1, op1); \ 1535 but1 = vec_mladd(but0, vprod1, op1); \
1506 register vector signed short op1S REG_v(v23) = vec_perm(but0S, but0S, perm1); \ 1536 op1S = vec_perm(but0S, but0S, perm1); \
1507 register vector signed short but1S REG_v(v24) = vec_mladd(but0S, vprod1, op1S); \ 1537 but1S = vec_mladd(but0S, vprod1, op1S); \
1508 register vector signed short op2 REG_v(v25) = vec_perm(but1, but1, perm2); \ 1538 op2 = vec_perm(but1, but1, perm2); \
1509 register vector signed short but2 REG_v(v26) = vec_mladd(but1, vprod2, op2); \ 1539 but2 = vec_mladd(but1, vprod2, op2); \
1510 register vector signed short op2S REG_v(v27) = vec_perm(but1S, but1S, perm2); \ 1540 op2S = vec_perm(but1S, but1S, perm2); \
1511 register vector signed short but2S REG_v(v28) = vec_mladd(but1S, vprod2, op2S); \ 1541 but2S = vec_mladd(but1S, vprod2, op2S); \
1512 register vector signed short op3 REG_v(v29) = vec_perm(but2, but2, perm3); \ 1542 op3 = vec_perm(but2, but2, perm3); \
1513 res1 = vec_mladd(but2, vprod3, op3); \ 1543 res1 = vec_mladd(but2, vprod3, op3); \
1514 register vector signed short op3S REG_v(v30) = vec_perm(but2S, but2S, perm3); \ 1544 op3S = vec_perm(but2S, but2S, perm3); \
1515 res2 = vec_mladd(but2S, vprod3, op3S); \ 1545 res2 = vec_mladd(but2S, vprod3, op3S); \
1516 } 1546 }
1517 ONEITERBUTTERFLY(0, temp0, temp0S); 1547 ONEITERBUTTERFLY(0, temp0, temp0S);
1518 ONEITERBUTTERFLY(1, temp1, temp1S); 1548 ONEITERBUTTERFLY(1, temp1, temp1S);
1519 ONEITERBUTTERFLY(2, temp2, temp2S); 1549 ONEITERBUTTERFLY(2, temp2, temp2S);
1524 ONEITERBUTTERFLY(7, temp7, temp7S); 1554 ONEITERBUTTERFLY(7, temp7, temp7S);
1525 } 1555 }
1526 #undef ONEITERBUTTERFLY 1556 #undef ONEITERBUTTERFLY
1527 { 1557 {
1528 register vector signed int vsum; 1558 register vector signed int vsum;
1559 register vector signed short line0S, line1S, line2S, line3S, line4S,
1560 line5S, line6S, line7S, line0BS,line2BS,
1561 line1BS,line3BS,line4BS,line6BS,line5BS,
1562 line7BS,line0CS,line4CS,line1CS,line5CS,
1563 line2CS,line6CS,line3CS,line7CS;
1564
1529 register vector signed short line0 = vec_add(temp0, temp1); 1565 register vector signed short line0 = vec_add(temp0, temp1);
1530 register vector signed short line1 = vec_sub(temp0, temp1); 1566 register vector signed short line1 = vec_sub(temp0, temp1);
1531 register vector signed short line2 = vec_add(temp2, temp3); 1567 register vector signed short line2 = vec_add(temp2, temp3);
1532 register vector signed short line3 = vec_sub(temp2, temp3); 1568 register vector signed short line3 = vec_sub(temp2, temp3);
1533 register vector signed short line4 = vec_add(temp4, temp5); 1569 register vector signed short line4 = vec_add(temp4, temp5);
1560 vsum = vec_sum4s(vec_abs(line4C), vsum); 1596 vsum = vec_sum4s(vec_abs(line4C), vsum);
1561 vsum = vec_sum4s(vec_abs(line5C), vsum); 1597 vsum = vec_sum4s(vec_abs(line5C), vsum);
1562 vsum = vec_sum4s(vec_abs(line6C), vsum); 1598 vsum = vec_sum4s(vec_abs(line6C), vsum);
1563 vsum = vec_sum4s(vec_abs(line7C), vsum); 1599 vsum = vec_sum4s(vec_abs(line7C), vsum);
1564 1600
1565 register vector signed short line0S = vec_add(temp0S, temp1S); 1601 line0S = vec_add(temp0S, temp1S);
1566 register vector signed short line1S = vec_sub(temp0S, temp1S); 1602 line1S = vec_sub(temp0S, temp1S);
1567 register vector signed short line2S = vec_add(temp2S, temp3S); 1603 line2S = vec_add(temp2S, temp3S);
1568 register vector signed short line3S = vec_sub(temp2S, temp3S); 1604 line3S = vec_sub(temp2S, temp3S);
1569 register vector signed short line4S = vec_add(temp4S, temp5S); 1605 line4S = vec_add(temp4S, temp5S);
1570 register vector signed short line5S = vec_sub(temp4S, temp5S); 1606 line5S = vec_sub(temp4S, temp5S);
1571 register vector signed short line6S = vec_add(temp6S, temp7S); 1607 line6S = vec_add(temp6S, temp7S);
1572 register vector signed short line7S = vec_sub(temp6S, temp7S); 1608 line7S = vec_sub(temp6S, temp7S);
1573 1609
1574 register vector signed short line0BS = vec_add(line0S, line2S); 1610 line0BS = vec_add(line0S, line2S);
1575 register vector signed short line2BS = vec_sub(line0S, line2S); 1611 line2BS = vec_sub(line0S, line2S);
1576 register vector signed short line1BS = vec_add(line1S, line3S); 1612 line1BS = vec_add(line1S, line3S);
1577 register vector signed short line3BS = vec_sub(line1S, line3S); 1613 line3BS = vec_sub(line1S, line3S);
1578 register vector signed short line4BS = vec_add(line4S, line6S); 1614 line4BS = vec_add(line4S, line6S);
1579 register vector signed short line6BS = vec_sub(line4S, line6S); 1615 line6BS = vec_sub(line4S, line6S);
1580 register vector signed short line5BS = vec_add(line5S, line7S); 1616 line5BS = vec_add(line5S, line7S);
1581 register vector signed short line7BS = vec_sub(line5S, line7S); 1617 line7BS = vec_sub(line5S, line7S);
1582 1618
1583 register vector signed short line0CS = vec_add(line0BS, line4BS); 1619 line0CS = vec_add(line0BS, line4BS);
1584 register vector signed short line4CS = vec_sub(line0BS, line4BS); 1620 line4CS = vec_sub(line0BS, line4BS);
1585 register vector signed short line1CS = vec_add(line1BS, line5BS); 1621 line1CS = vec_add(line1BS, line5BS);
1586 register vector signed short line5CS = vec_sub(line1BS, line5BS); 1622 line5CS = vec_sub(line1BS, line5BS);
1587 register vector signed short line2CS = vec_add(line2BS, line6BS); 1623 line2CS = vec_add(line2BS, line6BS);
1588 register vector signed short line6CS = vec_sub(line2BS, line6BS); 1624 line6CS = vec_sub(line2BS, line6BS);
1589 register vector signed short line3CS = vec_add(line3BS, line7BS); 1625 line3CS = vec_add(line3BS, line7BS);
1590 register vector signed short line7CS = vec_sub(line3BS, line7BS); 1626 line7CS = vec_sub(line3BS, line7BS);
1591 1627
1592 vsum = vec_sum4s(vec_abs(line0CS), vsum); 1628 vsum = vec_sum4s(vec_abs(line0CS), vsum);
1593 vsum = vec_sum4s(vec_abs(line1CS), vsum); 1629 vsum = vec_sum4s(vec_abs(line1CS), vsum);
1594 vsum = vec_sum4s(vec_abs(line2CS), vsum); 1630 vsum = vec_sum4s(vec_abs(line2CS), vsum);
1595 vsum = vec_sum4s(vec_abs(line3CS), vsum); 1631 vsum = vec_sum4s(vec_abs(line3CS), vsum);