Skip to content

GitLab

Commit 5917d17c authored by Leon van Stuivenberg's avatar Leon van Stuivenberg Committed by Michael Niedermayer
Browse files

ps2 optimizations update patch by (Leon van Stuivenberg <leonvs at iae dot nl>) 

Originally committed as revision 996 to svn://svn.ffmpeg.org/ffmpeg/trunk
parent a46a3ce4
Hide whitespace changes
Inline Side-by-side
Showing with 533 additions and 258 deletions
libavcodec/Makefile
View file @ 5917d17c
......@@ -77,7 +77,7 @@ OBJS += ppc/dsputil_ppc.o
endif
ifeq ($(TARGET_MMI),yes)
OBJS += ps2/dsputil_mmi.o ps2/idct_mmi.o
OBJS += ps2/dsputil_mmi.o ps2/idct_mmi.o ps2/mpegvideo_mmi.o
endif
ifeq ($(TARGET_ALTIVEC),yes)
......
libavcodec/mpegvideo.c
View file @ 5917d17c
......@@ -216,6 +216,9 @@ int MPV_common_init(MpegEncContext *s)
#ifdef HAVE_MLIB
MPV_common_init_mlib(s);
#endif
#ifdef HAVE_MMI
MPV_common_init_mmi(s);
#endif
/* load & permutate scantables
......
libavcodec/mpegvideo.h
View file @ 5917d17c
......@@ -503,6 +503,9 @@ void MPV_common_init_axp(MpegEncContext *s);
#ifdef HAVE_MLIB
void MPV_common_init_mlib(MpegEncContext *s);
#endif
#ifdef HAVE_MMI
void MPV_common_init_mmi(MpegEncContext *s);
#endif
extern void (*draw_edges)(UINT8 *buf, int wrap, int width, int height, int w);
void ff_conceal_past_errors(MpegEncContext *s, int conceal_all);
void ff_copy_bits(PutBitContext *pb, UINT8 *src, int length);
......
libavcodec/ps2/dsputil_mmi.c
View file @ 5917d17c
......@@ -20,96 +20,113 @@
*/
#include "../dsputil.h"
void ff_mmi_idct(DCTELEM * block);
#include "mmi.h"
/* the provided 'as' in binutils 2.9EE doesn't support
the EE's mips3 instructions properly */
#define AS_BUGGY
static void clear_blocks_mmi(DCTELEM * blocks)
{
/* $4 = blocks */
int i;
for (i = 0; i < 6; i++) {
sq($0, 0, $4);
sq($0, 16, $4);
sq($0, 32, $4);
sq($0, 48, $4);
sq($0, 64, $4);
sq($0, 80, $4);
sq($0, 96, $4);
sq($0, 112, $4);
__asm__ __volatile__("addi $4, $4, 128");
asm volatile(
"sq $0, 0(%0) \n\t"
"sq $0, 16(%0) \n\t"
"sq $0, 32(%0) \n\t"
"sq $0, 48(%0) \n\t"
"sq $0, 64(%0) \n\t"
"sq $0, 80(%0) \n\t"
"sq $0, 96(%0) \n\t"
"sq $0, 112(%0) \n\t" :: "r" (blocks) : "memory" );
blocks += 64;
}
}
static void put_pixels_clamped_mmi(const DCTELEM * block, UINT8 * pixels,
int line_size)
static void get_pixels_mmi(DCTELEM *block, const UINT8 *pixels, int line_size)
{
/* $4 = block, $5 = pixels, $6 = line_size */
__asm__ __volatile__("li $11, 255":::"$11");
lq($4, 0, $12);
pcpyld($11, $11, $11);
pcpyh($11, $11);
#define PUT(rs) \
ppacb($0, $##rs, $##rs); \
sd3(rs, 0, 5); \
__asm__ __volatile__ ("add $5, $5, $6");
pminh($12, $11, $12);
pmaxh($12, $0, $12);
lq($4, 16, $13);
PUT(12);
pminh($13, $11, $13);
pmaxh($13, $0, $13);
lq($4, 32, $12);
PUT(13);
pminh($12, $11, $12);
pmaxh($12, $0, $12);
lq($4, 48, $13);
PUT(12);
pminh($13, $11, $13);
pmaxh($13, $0, $13);
lq($4, 64, $12);
PUT(13);
pminh($12, $11, $12);
pmaxh($12, $0, $12);
lq($4, 80, $13);
PUT(12);
pminh($13, $11, $13);
pmaxh($13, $0, $13);
lq($4, 96, $12);
PUT(13);
int i;
for(i=0;i<8;i++) {
#ifdef AS_BUGGY
ld3(5, 0, 8);
asm volatile(
"add %1, %1, %2 \n\t"
"pextlb $8, $0, $8 \n\t"
"sq $8, 0(%0) \n\t" :: "r" (block), "r" (pixels), "r" (line_size) : "$8", "memory" );
#else
asm volatile(
"ld $8, 0(%1) \n\t"
"add %1, %1, %2 \n\t"
"pextlb $8, $0, $8 \n\t"
"sq $8, 0(%0) \n\t" :: "r" (block), "r" (pixels), "r" (line_size) : "$8", "memory" );
#endif
block += 8;
}
}
pminh($12, $11, $12);
pmaxh($12, $0, $12);
lq($4, 112, $13);
PUT(12);
pminh($13, $11, $13);
pmaxh($13, $0, $13);
PUT(13);
static void put_pixels8_mmi(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
int i;
for(i=0; i<h; i++) {
#ifdef AS_BUGGY
ldr3(5, 0, 8);
ldl3(5, 7, 8);
asm volatile ( "add $5, $5, $6 \n\t" );
sd3(8, 0, 4);
asm volatile ( "add $4, $4, $6 \n\t" );
#else
asm volatile(
"ldr $8, 0(%1) \n\t"
"ldl $8, 7(%1) \n\t"
"add %1, %1, %2 \n\t"
"sd $8, 0(%0) \n\t"
"add %0, %0, %2 \n\t" :: "r" (block), "r" (pixels), "r" (line_size) : "$8", "memory" );
#endif
}
}
/* todo
static void add_pixels_clamped_mmi(const DCTELEM * block, UINT8 * pixels,
int line_size)
static void put_pixels16_mmi(uint8_t *block, const uint8_t *pixels, int line_size, int h)
{
int i;
for(i=0; i<h; i++) {
#ifdef AS_BUGGY
ldr3(5, 0, 8);
ldl3(5, 7, 8);
ldr3(5, 8, 9);
ldl3(5, 15, 9);
asm volatile ( "add $5, $5, $6 \n\t" );
pcpyld($9, $8, $8);
sq($8, 0, $4);
asm volatile ( "add $4, $4, $6 \n\t" );
#else
asm volatile (
"ldr $8, 0(%1) \n\t"
"ldl $8, 7(%1) \n\t"
"ldr $9, 8(%1) \n\t"
"ldl $9, 15(%1) \n\t"
"add %1, %1, %2 \n\t"
"pcpyld $8, $9, $8 \n\t"
"sq $8, 0(%0) \n\t"
"add %0, %0, %2 \n\t" :: "r" (block), "r" (pixels), "r" (line_size) : "$8", "$9", "memory" );
#endif
}
}
*/
void dsputil_init_mmi(void)
{
put_pixels_clamped = put_pixels_clamped_mmi;
//add_pixels_clamped = add_pixels_clamped_mmi;
clear_blocks = clear_blocks_mmi;
ff_idct = ff_mmi_idct;
put_pixels_tab[1][0] = put_pixels8_mmi;
put_no_rnd_pixels_tab[1][0] = put_pixels8_mmi;
put_pixels_tab[0][0] = put_pixels16_mmi;
put_no_rnd_pixels_tab[0][0] = put_pixels16_mmi;
get_pixels = get_pixels_mmi;
}
libavcodec/ps2/idct_mmi.c
View file @ 5917d17c
......@@ -12,217 +12,337 @@
mmi port by leonvs@iae.nl
*/
#include "../common.h"
#include "../dsputil.h"
#include "mmi.h"
#define BITS_INV_ACC 5 // 4 or 5 for IEEE
#define SHIFT_INV_ROW (16 - BITS_INV_ACC)
#define SHIFT_INV_COL (1 + BITS_INV_ACC)
// assume SHIFT_INV_ROW == 11
static int roundertable[8][4] align16 = {
{ 0x103ff, 0x103ff, 0x103ff, 0x103ff },
{ 1023, 1023, 1023, 1023}
};
#define TAB_i_04 0
#define TAB_i_17 64
#define TAB_i_26 128
#define TAB_i_35 192
static short rowtable[4][32] align16 = {
{
16384, 16384, 22725, 12873, 21407, 8867, 19266, 4520,
16384, -16383, 19266, -22724, 8867, -21406, -4519, -12872,
16384, -16383, 12873, 4520, -8866, 21407, -22724, 19266,
16384, 16384, 4520, 19266, -21406, -8866, -12872, -22724},
{
22725, 22725, 31521, 17855, 29692, 12299, 26722, 6270,
22725, -22724, 26722, -31520, 12299, -29691, -6269, -17854,
22725, -22724, 17855, 6270, -12298, 29692, -31520, 26722,
22725, 22725, 6270, 26722, -29691, -12298, -17854, -31520},
{
21407, 21407, 29692, 16819, 27969, 11585, 25172, 5906,
21407, -21406, 25172, -29691, 11585, -27968, -5905, -16818,
21407, -21406, 16819, 5906, -11584, 27969, -29691, 25172,
21407, 21407, 5906, 25172, -27968, -11584, -16818, -29691},
{
19266, 19266, 26722, 15137, 25172, 10426, 22654, 5315,
19266, -19265, 22654, -26721, 10426, -25171, -5314, -15136,
19266, -19265, 15137, 5315, -10425, 25172, -26721, 22654,
19266, 19266, 5315, 22654, -25171, -10425, -15136, -26721}
};
#define TG_3_16_minus_one 0
#define ONE_plus_tg_3_16 16
#define ONE_plus_tg_1_16 32
#define TG_1_16_minus_one 48
#define TG_2_16_minus_one 64
#define ONE_plus_tg_2_16 80
#define ZERO_ocos_4_16 96
#define TG1 6518
#define TG2 13573
#define TG3 21895
#define MN1 -32768
#define PL1 32768
#define CS4 23170
static short coltable[7][8] align16 = {
{ MN1, TG3, MN1, TG3, MN1, TG3, MN1, TG3},
{-TG3, -PL1, -TG3, -PL1, -TG3, -PL1, -TG3, -PL1},
{-TG1, -PL1, -TG1, -PL1, -TG1, -PL1, -TG1, -PL1},
{ MN1, TG1, MN1, TG1, MN1, TG1, MN1, TG1},
{ MN1, TG2, MN1, TG2, MN1, TG2, MN1, TG2},
{-TG2, -PL1, -TG2, -PL1, -TG2, -PL1, -TG2, -PL1},
{ CS4, 0, CS4, 0, CS4, 0, CS4, 0}
#define ROUNDER_0 0
#define ROUNDER_1 16
#define TAB_i_04 (32+0)
#define TAB_i_17 (32+64)
#define TAB_i_26 (32+128)
#define TAB_i_35 (32+192)
#define TG_1_16 (32+256+0)
#define TG_2_16 (32+256+16)
#define TG_3_16 (32+256+32)
#define COS_4_16 (32+256+48)
#define CLIPMAX (32+256+64+0)
static short consttable[] align16 = {
/* rounder 0*/ // assume SHIFT_INV_ROW == 11
0x3ff, 1, 0x3ff, 1, 0x3ff, 1, 0x3ff, 1,
/* rounder 1*/
0x3ff, 0, 0x3ff, 0, 0x3ff, 0, 0x3ff, 0,
/* row 0/4*/
16384, 21407, -16384, -21407, 22725, 19266, -22725, -12873,
8867, 16384, 8867, 16384, 4520, 12873, -4520, 19266,
16384, -8867, 16384, -8867, 12873, -22725, 19266, -22725,
21407, -16384, -21407, 16384, 19266, 4520, -12873, 4520,
/* row 1/7*/
22725, 29692, -22725, -29692, 31521, 26722, -31521, -17855,
12299, 22725, 12299, 22725, 6270, 17855, -6270, 26722,
22725, -12299, 22725, -12299, 17855, -31521, 26722, -31521,
29692, -22725, -29692, 22725, 26722, 6270, -17855, 6270,
/* row 2/6*/
21407, 27969, -21407, -27969, 29692, 25172, -29692, -16819,
11585, 21407, 11585, 21407, 5906, 16819, -5906, 25172,
21407, -11585, 21407, -11585, 16819, -29692, 25172, -29692,
27969, -21407, -27969, 21407, 25172, 5906, -16819, 5906,
/*row 3/5*/
19266, 25172, -19266, -25172, 26722, 22654, -26722, -15137,
10426, 19266, 10426, 19266, 5315, 15137, -5315, 22654,
19266, -10426, 19266, -10426, 15137, -26722, 22654, -26722,
25172, -19266, -25172, 19266, 22654, 5315, -15137, 5315,
/*column constants*/
TG1, TG1, TG1, TG1, TG1, TG1, TG1, TG1,
TG2, TG2, TG2, TG2, TG2, TG2, TG2, TG2,
TG3, TG3, TG3, TG3, TG3, TG3, TG3, TG3,
CS4, CS4, CS4, CS4, CS4, CS4, CS4, CS4,
/* clamp */
255, 255, 255, 255, 255, 255, 255, 255
};
#define noprevh(rt, rd)
#define DCT_8_INV_ROW1(rowoff, taboff, rnd, outreg) { \
\
lq($4, rowoff, $16); /* r16 = x7 x6 x5 x4 x3 x2 x1 x0 */ \
lq($24, 0+taboff, $17); /* r17 = w19 w17 w3 w1 w18 w16 w2 w0 */ \
pinth($16, $16, $16); /* r16 = x7 x3 x6 x2 x5 x1 x4 x0 */ \
phmadh($17, $16, $17); /* r17 = (x7*w19+x3*w17)b0'' (x6*w3+x2*w1)a0'' (x5*w18+x1*w16)b0' (x4*w2+x0*w0)a0' */ \
lq($24, 16+taboff, $18);/* r18 = w23 w21 w7 w5 w22 w20 w6 w4 */ \
lq($24, 32+taboff, $19);/* r19 = w27 w25 w11 w9 w26 w24 w10 w8 */ \
lq($24, 48+taboff, $20);/* r20 = w31 w29 w15 w13 w30 w28 w14 w12 */ \
phmadh($18, $16, $18); /* r18 = (b1'')(a1'')(b1')(a1') */ \
pcpyud($17, $17, $21); /* r21 = (b0'')(a0'')(b0'')(a0'') */ \
paddw($17, $21, $17); /* r17 = (--)(--)(b0)(a0) */ \
phmadh($19, $16, $19); /* r19 = (b2'')(a2'')(b2')(a2') */ \
pcpyud($18, $18, $21); /* r21 = (b1'')(a1'')(b1'')(a1'') */ \
paddw($18, $21, $18); /* r18 = (--)(--)(b1)(a1) */ \
pcpyud($19, $19, $21); \
phmadh($20, $16, $20); /* r12 = (b3'')(a3'')(b3')(a3') */ \
paddw($19, $21, $19); /* r19 = (--)(--)(b2)(a2) */ \
pextlw($19, $17, $16); /* r16 = (b2)(b0)(a2)(a0) */ \
pcpyud($20, $20, $21); \
paddw($20, $21, $20); /* r20 = (--)(--)(b3)(a3) */ \
pextlw($20, $18, $17); /* r17 = (b3)(b1)(a3)(a1) */ \
pextlw($17, $16, $20); /* r20 = (a3)(a2)(a1)(a0)" */ \
pextuw($17, $16, $21); /* r21 = (b3)(b2)(b1)(b0) */ \
paddw($20, rnd, $20); /* r20 = (a3)(a2)(a1)(a0) */\
paddw($20, $21, $17); /* r17 = ()()()(a0+b0) */ \
psubw($20, $21, $18); /* r18 = ()()()(a0-b0) */ \
#define DCT_8_INV_ROW1(blk, rowoff, taboff, rnd, outreg) { \
lq(blk, rowoff, $16); /* r16 = x7 x5 x3 x1 x6 x4 x2 x0 */ \
/*slot*/ \
lq($24, 0+taboff, $17); /* r17 = w */ \
/*delay slot $16*/ \
lq($24, 16+taboff, $18);/* r18 = w */ \
prevh($16, $2); /* r2 = x1 x3 x5 x7 x0 x2 x4 x6 */ \
lq($24, 32+taboff, $19);/* r19 = w */ \
phmadh($17, $16, $17); /* r17 = b1"b0'a1"a0' */ \
lq($24, 48+taboff, $20);/* r20 = w */ \
phmadh($18, $2, $18); /* r18 = b1'b0"a1'a0" */ \
phmadh($19, $16, $19); /* r19 = b3"b2'a3"a2' */ \
phmadh($20, $2, $20); /* r20 = b3'b2"a3'a2" */ \
paddw($17, $18, $17); /* r17 = (b1)(b0)(a1)(a0) */ \
paddw($19, $20, $19); /* r19 = (b3)(b2)(a3)(a2) */ \
pcpyld($19, $17, $18); /* r18 = (a3)(a2)(a1)(a0) */ \
pcpyud($17, $19, $20); /* r20 = (b3)(b2)(b1)(b0) */ \
paddw($18, rnd, $18); /* r18 = (a3)(a2)(a1)(a0) */\
paddw($18, $20, $17); /* r17 = ()()()(a0+b0) */ \
psubw($18, $20, $20); /* r20 = ()()()(a0-b0) */ \
psraw($17, SHIFT_INV_ROW, $17); /* r17 = (y3 y2 y1 y0) */ \
psraw($18, SHIFT_INV_ROW, $18); /* r18 = (y4 y5 y6 y7) */ \
ppach($18, $17, outreg);/* out = y4 y5 y6 y7 y3 y2 y1 y0 Note order */ \
psraw($20, SHIFT_INV_ROW, $20); /* r20 = (y4 y5 y6 y7) */ \
ppach($20, $17, outreg);/* out = y4 y5 y6 y7 y3 y2 y1 y0 Note order */ \
\
prevh(outreg, $2); \
pcpyud($2, $2, $2); \
pcpyld($2, outreg, outreg); \
}
#define DCT_8_INV_COL4(pextop, blkoff, revop) { \
lq($24, TG_3_16_minus_one, $2); /* r2 = (tn3)(-1) x 4 */ \
pextop($11, $13, $3); /* r3 = (x3)(x5) x 4 */ \
lq($24, ONE_plus_tg_3_16, $16); /* r16 = -((+1)(tn3)) x 4 */ \
phmadh($3, $2, $17); /* r17 = (tm35) x 4 */ \
lq($24, ONE_plus_tg_1_16, $2); /* r2 = -((+1)(tn1)) x 4 */ \
phmadh($3, $16, $18); /* r18 = -(tp35) x 4 */ \
lq($24, TG_1_16_minus_one, $16);/* r16 = (tn1)(-1) x 4 */ \
pextop($9, $15, $3); /* r3 = (x1)(x7) x 4 */ \
phmadh($3, $2, $19); \
lq($24, ZERO_ocos_4_16, $2); /* r2 = (0)(cos4) x 4 */ \
phmadh($3, $16, $20); /* r20 = (tm17) x 4 */ \
psubw($0, $19, $19); /* r19 = (tp17) x 4 */ \
paddw($19, $18, $3); /* r3 = t1 */ \
paddw($20, $17, $16); /* r16 = t2 */ \
psubw($20, $17, $23); /* r23 = b3 */ \
psubw($19, $18, $20); /* r20 = b0 */ \
paddw($3, $16, $17); /* (t1+t2) */ \
psubw($3, $16, $18); /* (t1-t2) */ \
psraw($17, 15, $17); \
lq($24, TG_2_16_minus_one, $3); /* r3 = (tn2)(-1) x 4 */ \
pmulth($17, $2, $21); /* r21 = b1 */ \
psraw($18, 15, $18); \
lq($24, ONE_plus_tg_2_16, $16); /* r16 = -((+1)(tn2)) x 4 */ \
pmulth($18, $2, $22); /* r22 = b2 */ \
\
pextop($10, $14, $2); /* r2 = (x2)(x6) x 4 */ \
phmadh($2, $3, $18); /* r18 = (tm26) x 4 */ \
phmadh($2, $16, $19); /* r19 = -(tp26) x 4 */ \
pextop($8, $0, $17); /* r17 = (x0)(0) x 4 */ \
psraw($17, 1, $17); \
pextop($12, $0, $16); /* r16 = (x4)(0) x 4 */ \
psraw($16, 1, $16); \
paddw($17, $16, $2); /* r2 = tp04 */ \
psubw($17, $16, $3); /* r3 = tm04 */ \
psubw($2, $19, $16); /* r16 = a0 */ \
paddw($3, $18, $17); /* r17 = a1 */ \
psubw($3, $18, $18); /* r18 = a2 */ \
paddw($2, $19, $19); /* r19 = a3 */ \
\
paddw($16, $20, $2); /* y0 a0+b0 */ \
psubw($16, $20, $16); /* y7 a0-b0 */ \
psraw($2, SHIFT_INV_COL+15, $2); \
psraw($16, SHIFT_INV_COL+15, $16); \
ppach($0, $2, $2); \
ppach($0, $16, $16); \
revop($2, $2); \
revop($16, $16); \
sd3(2, 0+blkoff, 4); \
sd3(16, 112+blkoff, 4); \
\
paddw($17, $21, $3); /* y1 a1+b1 */ \
psubw($17, $21, $17); /* y6 a1-b1 */ \
psraw($3, SHIFT_INV_COL+15, $3); \
psraw($17, SHIFT_INV_COL+15, $17); \
ppach($0, $3, $3); \
ppach($0, $17, $17); \
revop($3, $3); \
revop($17, $17); \
sd3(3, 16+blkoff, 4); \
sd3(17, 96+blkoff, 4); \
\
paddw($18, $22, $2); /* y2 a2+b2 */ \
psubw($18, $22, $18); /* y5 a2-b2 */ \
psraw($2, SHIFT_INV_COL+15, $2); \
psraw($18, SHIFT_INV_COL+15, $18); \
ppach($0, $2, $2); \
ppach($0, $18, $18); \
revop($2, $2); \
revop($18, $18); \
sd3(2, 32+blkoff, 4); \
sd3(18, 80+blkoff, 4); \
\
paddw($19, $23, $3); /* y3 a3+b3 */ \
psubw($19, $23, $19); /* y4 a3-b3 */ \
psraw($3, SHIFT_INV_COL+15, $3); \
psraw($19, SHIFT_INV_COL+15, $19); \
ppach($0, $3, $3); \
ppach($0, $19, $19); \
revop($3, $3); \
revop($19, $19); \
sd3(3, 48+blkoff, 4); \
sd3(19, 64+blkoff, 4); \
}
#define DCT_8_INV_COL8() \
\
lq($24, TG_3_16, $2); /* r2 = tn3 */ \
\
pmulth($11, $2, $17); /* r17 = x3 * tn3 (6420) */ \
psraw($17, 15, $17); \
pmfhl_uw($3); /* r3 = 7531 */ \
psraw($3, 15, $3); \
pinteh($3, $17, $17); /* r17 = x3 * tn3 */ \
psubh($17, $13, $17); /* r17 = tm35 */ \
\
pmulth($13, $2, $18); /* r18 = x5 * tn3 (6420) */ \
psraw($18, 15, $18); \
pmfhl_uw($3); /* r3 = 7531 */ \
psraw($3, 15, $3); \
pinteh($3, $18, $18); /* r18 = x5 * tn3 */ \
paddh($18, $11, $18); /* r18 = tp35 */ \
\
lq($24, TG_1_16, $2); /* r2 = tn1 */ \
\
pmulth($15, $2, $19); /* r19 = x7 * tn1 (6420) */ \
psraw($19, 15, $19); \
pmfhl_uw($3); /* r3 = 7531 */ \
psraw($3, 15, $3); \
pinteh($3, $19, $19); /* r19 = x7 * tn1 */ \
paddh($19, $9, $19); /* r19 = tp17 */ \
\
pmulth($9, $2, $20); /* r20 = x1 * tn1 (6420) */ \
psraw($20, 15, $20); \
pmfhl_uw($3); /* r3 = 7531 */ \
psraw($3, 15, $3); \
pinteh($3, $20, $20); /* r20 = x1 * tn1 */ \
psubh($20, $15, $20); /* r20 = tm17 */ \
\
psubh($19, $18, $3); /* r3 = t1 */ \
paddh($20, $17, $16); /* r16 = t2 */ \
psubh($20, $17, $23); /* r23 = b3 */ \
paddh($19, $18, $20); /* r20 = b0 */ \
\
lq($24, COS_4_16, $2); /* r2 = cs4 */ \
\
paddh($3, $16, $21); /* r21 = t1+t2 */ \
psubh($3, $16, $22); /* r22 = t1-t2 */ \
\
pmulth($21, $2, $21); /* r21 = cs4 * (t1+t2) 6420 */ \
psraw($21, 15, $21); \
pmfhl_uw($3); /* r3 = 7531 */ \
psraw($3, 15, $3); \
pinteh($3, $21, $21); /* r21 = b1 */ \
\
pmulth($22, $2, $22); /* r22 = cs4 * (t1-t2) 6420 */ \
psraw($22, 15, $22); \
pmfhl_uw($3); /* r3 = 7531 */ \
psraw($3, 15, $3); \
pinteh($3, $22, $22); /* r22 = b2 */ \
\
lq($24, TG_2_16, $2); /* r2 = tn2 */ \
\
pmulth($10, $2, $17); /* r17 = x2 * tn2 (6420) */ \
psraw($17, 15, $17); \
pmfhl_uw($3); /* r3 = 7531 */ \
psraw($3, 15, $3); \
pinteh($3, $17, $17); /* r17 = x3 * tn3 */ \
psubh($17, $14, $17); /* r17 = tm26 */ \
\
pmulth($14, $2, $18); /* r18 = x6 * tn2 (6420) */ \
psraw($18, 15, $18); \
pmfhl_uw($3); /* r3 = 7531 */ \
psraw($3, 15, $3); \
pinteh($3, $18, $18); /* r18 = x6 * tn2 */ \
paddh($18, $10, $18); /* r18 = tp26 */ \
\
paddh($8, $12, $2); /* r2 = tp04 */ \
psubh($8, $12, $3); /* r3 = tm04 */ \
\
paddh($2, $18, $16); /* r16 = a0 */ \
psubh($2, $18, $19); /* r19 = a3 */ \
psubh($3, $17, $18); /* r18 = a2 */ \
paddh($3, $17, $17); /* r17 = a1 */
#define DCT_8_INV_COL8_STORE(blk) \
\
paddh($16, $20, $2); /* y0 a0+b0 */ \
psubh($16, $20, $16); /* y7 a0-b0 */ \
psrah($2, SHIFT_INV_COL, $2); \
psrah($16, SHIFT_INV_COL, $16); \
sq($2, 0, blk); \
sq($16, 112, blk); \
\
paddh($17, $21, $3); /* y1 a1+b1 */ \
psubh($17, $21, $17); /* y6 a1-b1 */ \
psrah($3, SHIFT_INV_COL, $3); \
psrah($17, SHIFT_INV_COL, $17); \
sq($3, 16, blk); \
sq($17, 96, blk); \
\
paddh($18, $22, $2); /* y2 a2+b2 */ \
psubh($18, $22, $18); /* y5 a2-b2 */ \
psrah($2, SHIFT_INV_COL, $2); \
psrah($18, SHIFT_INV_COL, $18); \
sq($2, 32, blk); \
sq($18, 80, blk); \
\
paddh($19, $23, $3); /* y3 a3+b3 */ \
psubh($19, $23, $19); /* y4 a3-b3 */ \
psrah($3, SHIFT_INV_COL, $3); \
psrah($19, SHIFT_INV_COL, $19); \
sq($3, 48, blk); \
sq($19, 64, blk);
#define DCT_8_INV_COL8_PMS() \
paddh($16, $20, $2); /* y0 a0+b0 */ \
psubh($16, $20, $20); /* y7 a0-b0 */ \
psrah($2, SHIFT_INV_COL, $16); \
psrah($20, SHIFT_INV_COL, $20); \
\
paddh($17, $21, $3); /* y1 a1+b1 */ \
psubh($17, $21, $21); /* y6 a1-b1 */ \
psrah($3, SHIFT_INV_COL, $17); \
psrah($21, SHIFT_INV_COL, $21); \
\