static inline void mmx_average_2_U8 (uint8_t * dest,
uint8_t * src1, uint8_t * src2)
{
/* *dest = (*src1 + *src2 + 1)/ 2; */
movq_m2r (*src1, mm1); // load 8 src1 bytes
movq_r2r (mm1, mm2); // copy 8 src1 bytes
movq_m2r (*src2, mm3); // load 8 src2 bytes
movq_r2r (mm3, mm4); // copy 8 src2 bytes
punpcklbw_r2r (mm0, mm1); // unpack low src1 bytes
punpckhbw_r2r (mm0, mm2); // unpack high src1 bytes
punpcklbw_r2r (mm0, mm3); // unpack low src2 bytes
punpckhbw_r2r (mm0, mm4); // unpack high src2 bytes
paddw_r2r (mm3, mm1); // add lows to mm1
paddw_m2r (round1, mm1);
psraw_i2r (1, mm1); // /2
paddw_r2r (mm4, mm2); // add highs to mm2
paddw_m2r (round1, mm2);
psraw_i2r (1, mm2); // /2
packuswb_r2r (mm2, mm1); // pack (w/ saturation)
movq_r2m (mm1, *dest); // store result in dest
}
int main(int ac, char **av)
{
int i, j, k, n;
unsigned char dat0[8] = { 0x01, 0xf2, 0x03, 0x04, 0x05, 0x06, 0xf7, 0x08 };
long long *datp = (long long *)&dat0;
int16_t dat1[8] = { 0x10, 0x20, -0x130, -0x140, 0x50, -0x160, -0x170, 0x80 };
volatile uint8_t *rfp = dat0;
volatile int16_t *bp = dat1;
unsigned char ans1[8], ans2[8];
n = 0;
for( i=-32768; i<32768; ++i ) {
j = 0;
while( j < 256 ) {
for( k=0; k<8; ++k ) {
dat0[k] = i;
dat1[k] = j++;
}
movq_m2r(m_(&rfp[0]),mm1); /* rfp[0..7] */
pxor_r2r(mm3,mm3);
pxor_r2r(mm4,mm4);
movq_m2r(m_(&bp[0]),mm5); /* bp[0..3] */
movq_r2r(mm1,mm2);
movq_m2r(m_(&bp[4]),mm6); /* bp[4..7] */
punpcklbw_r2r(mm3,mm1); /* rfp[0,2,4,6] */
punpckhbw_r2r(mm3,mm2); /* rfp[1,3,5,7] */
paddsw_r2r(mm5,mm1); /* bp[0..3] */
paddsw_r2r(mm6,mm2); /* bp[4..7] */
pcmpgtw_r2r(mm1,mm3);
pcmpgtw_r2r(mm2,mm4);
pandn_r2r(mm1,mm3);
pandn_r2r(mm2,mm4);
packuswb_r2r(mm4,mm3);
movq_r2m(mm3,m_(&ans1[0]));
emms();
ans2[0] = clip(bp[0] + rfp[0]);
ans2[1] = clip(bp[1] + rfp[1]);
ans2[2] = clip(bp[2] + rfp[2]);
ans2[3] = clip(bp[3] + rfp[3]);
ans2[4] = clip(bp[4] + rfp[4]);
ans2[5] = clip(bp[5] + rfp[5]);
ans2[6] = clip(bp[6] + rfp[6]);
ans2[7] = clip(bp[7] + rfp[7]);
if( *(uint64_t *)&ans1[0] != *(uint64_t *)&ans2[0] )
{
printf(" i=%5d %02x %02x %02x %02x %02x %02x %02x %02x\n", i,
ans1[0], ans1[1], ans1[2], ans1[3], ans1[4], ans1[5], ans1[6], ans1[7]);
printf(" j=%5d %02x %02x %02x %02x %02x %02x %02x %02x\n", j,
ans2[0], ans2[1], ans2[2], ans2[3], ans2[4], ans2[5], ans2[6], ans2[7]);
// exit(0);
}
n += 8;
}
}
printf("n=%d\n",n);
return 0;
}
static inline void mmx_end(uint8_t *src3, uint8_t *src5,
uint8_t *dst, int X)
{
punpcklbw_m2r (mm_cpool[0], mm4);
punpckhbw_m2r (mm_cpool[0], mm5);
psubusw_r2r (mm2, mm0);
psubusw_r2r (mm3, mm1);
movq_m2r (src5[X], mm2);
movq_m2r (src5[X], mm3);
punpcklbw_m2r (mm_cpool[0], mm2);
punpckhbw_m2r (mm_cpool[0], mm3);
psubusw_r2r (mm2, mm0);
psubusw_r2r (mm3, mm1);
psrlw_i2r (3, mm0);
psrlw_i2r (3, mm1);
psubw_r2r (mm6, mm4);
psubw_r2r (mm7, mm5);
packuswb_r2r (mm1,mm0);
movq_r2r (mm4, mm6);
movq_r2r (mm5, mm7);
pcmpgtw_m2r (mm_lthr, mm4);
pcmpgtw_m2r (mm_lthr, mm5);
pcmpgtw_m2r (mm_hthr, mm6);
pcmpgtw_m2r (mm_hthr, mm7);
packsswb_r2r (mm5, mm4);
packsswb_r2r (mm7, mm6);
pxor_r2r (mm6, mm4);
movq_r2r (mm4, mm5);
pandn_r2r (mm0, mm4);
pand_m2r (src3[X], mm5);
por_r2r (mm4, mm5);
movq_r2m (mm5, dst[X]);
}
static void scale_uint8_x_4_x_bilinear_mmx(gavl_video_scale_context_t * ctx, int scanline, uint8_t * dest_start)
{
int i;
uint8_t * src, * dst, *src_start;
int32_t * factors;
// mmx_t tmp_mm;
/*
* mm0: Input1
* mm1: Factor mask
* mm2:
* mm3: Output
* mm4:
* mm5: Input2
* mm6: 0
* mm7: Factor
*
*/
// fprintf(stderr, "scale_uint8_x_4_x_bilinear_mmx\n");
src_start = ctx->src + scanline * ctx->src_stride;
pxor_r2r(mm6, mm6);
movq_m2r(factor_mask, mm1);
dst = dest_start;
for(i = 0; i < ctx->dst_size; i++)
{
src = src_start + 4*ctx->table_h.pixels[i].index;
factors = ctx->table_h.pixels[i].factor_i;
/* Load pixels */
movd_m2r(*(src), mm0);
punpcklbw_r2r(mm6, mm0);
psllw_i2r(6, mm0); /* 14 bit */
/* Load pixels */
movd_m2r(*(src+4), mm5);
punpcklbw_r2r(mm6, mm5);
psllw_i2r(6, mm5); /* 14 bit */
/* Load factors */
LOAD_FACTOR_1_4_NOCLIP; /* 14 bit */
/* Subtract */
psubsw_r2r(mm5, mm0); /* s1(mm0) - s2(mm5) -> mm0 (14 bit) */
pmulhw_r2r(mm7, mm0); /* factor * (s2 - s1) -> mm0 (12 bit) */
psllw_i2r(2, mm0); /* (14 bit) */
paddsw_r2r(mm5, mm0);/* (15 bit) */
psraw_i2r(6, mm0);/* (8 bit) */
packuswb_r2r(mm6, mm0);
movd_r2m(mm0, *dst);
dst+=4;
}
ctx->need_emms = 1;
}
static void scale_uint8_x_4_x_generic_mmx(gavl_video_scale_context_t * ctx, int scanline, uint8_t * dest_start)
{
int i, j;
uint8_t * src, * dst, *src_start;
int32_t * factors;
// mmx_t tmp_mm;
/*
* mm0: Input
* mm1: factor_mask
* mm2: Factor
* mm3: Output
* mm4:
* mm5:
* mm6: 0
* mm7: scratch
*
*/
src_start = ctx->src + scanline * ctx->src_stride;
pxor_r2r(mm6, mm6);
movq_m2r(factor_mask, mm1);
dst = dest_start;
for(i = 0; i < ctx->dst_size; i++)
{
src = src_start + 4*ctx->table_h.pixels[i].index;
factors = ctx->table_h.pixels[i].factor_i;
pxor_r2r(mm3, mm3);
for(j = 0; j < ctx->table_h.factors_per_pixel; j++)
{
/* Load pixels */
movd_m2r(*(src), mm0);
punpcklbw_r2r(mm6, mm0);
psllw_i2r(7, mm0);
/* Load factors */
LOAD_FACTOR_1_4;
/* Multiply */
pmulhw_r2r(mm7, mm0);
paddw_r2r(mm0, mm3);
// DUMP_MM("mm3_2", mm3);
src += 4;
factors++;
}
psraw_i2r(5, mm3);
packuswb_r2r(mm6, mm3);
movd_r2m(mm3, *dst);
dst+=4;
}
ctx->need_emms = 1;
}
static inline void mmx_average_4_U8 (uint8_t * dest, const uint8_t * src1,
const uint8_t * src2,
const uint8_t * src3,
const uint8_t * src4)
{
/* *dest = (*src1 + *src2 + *src3 + *src4 + 2)/ 4; */
movq_m2r (*src1, mm1); /* load 8 src1 bytes */
movq_r2r (mm1, mm2); /* copy 8 src1 bytes */
punpcklbw_r2r (mm0, mm1); /* unpack low src1 bytes */
punpckhbw_r2r (mm0, mm2); /* unpack high src1 bytes */
movq_m2r (*src2, mm3); /* load 8 src2 bytes */
movq_r2r (mm3, mm4); /* copy 8 src2 bytes */
punpcklbw_r2r (mm0, mm3); /* unpack low src2 bytes */
punpckhbw_r2r (mm0, mm4); /* unpack high src2 bytes */
paddw_r2r (mm3, mm1); /* add lows */
paddw_r2r (mm4, mm2); /* add highs */
/* now have partials in mm1 and mm2 */
movq_m2r (*src3, mm3); /* load 8 src3 bytes */
movq_r2r (mm3, mm4); /* copy 8 src3 bytes */
punpcklbw_r2r (mm0, mm3); /* unpack low src3 bytes */
punpckhbw_r2r (mm0, mm4); /* unpack high src3 bytes */
paddw_r2r (mm3, mm1); /* add lows */
paddw_r2r (mm4, mm2); /* add highs */
movq_m2r (*src4, mm5); /* load 8 src4 bytes */
movq_r2r (mm5, mm6); /* copy 8 src4 bytes */
punpcklbw_r2r (mm0, mm5); /* unpack low src4 bytes */
punpckhbw_r2r (mm0, mm6); /* unpack high src4 bytes */
paddw_r2r (mm5, mm1); /* add lows */
paddw_r2r (mm6, mm2); /* add highs */
/* now have subtotal in mm1 and mm2 */
paddw_m2r (round4, mm1);
psraw_i2r (2, mm1); /* /4 */
paddw_m2r (round4, mm2);
psraw_i2r (2, mm2); /* /4 */
packuswb_r2r (mm2, mm1); /* pack (w/ saturation) */
movq_r2m (mm1, *dest); /* store result in dest */
}
static inline void mmx_average_4_U8 (uint8_t * dest,
uint8_t * src1, uint8_t * src2,
uint8_t * src3, uint8_t * src4)
{
/* *dest = (*src1 + *src2 + *src3 + *src4 + 2)/ 4; */
movq_m2r (*src1, mm1); // load 8 src1 bytes
movq_r2r (mm1, mm2); // copy 8 src1 bytes
punpcklbw_r2r (mm0, mm1); // unpack low src1 bytes
punpckhbw_r2r (mm0, mm2); // unpack high src1 bytes
movq_m2r (*src2, mm3); // load 8 src2 bytes
movq_r2r (mm3, mm4); // copy 8 src2 bytes
punpcklbw_r2r (mm0, mm3); // unpack low src2 bytes
punpckhbw_r2r (mm0, mm4); // unpack high src2 bytes
paddw_r2r (mm3, mm1); // add lows
paddw_r2r (mm4, mm2); // add highs
/* now have partials in mm1 and mm2 */
movq_m2r (*src3, mm3); // load 8 src3 bytes
movq_r2r (mm3, mm4); // copy 8 src3 bytes
punpcklbw_r2r (mm0, mm3); // unpack low src3 bytes
punpckhbw_r2r (mm0, mm4); // unpack high src3 bytes
paddw_r2r (mm3, mm1); // add lows
paddw_r2r (mm4, mm2); // add highs
movq_m2r (*src4, mm5); // load 8 src4 bytes
movq_r2r (mm5, mm6); // copy 8 src4 bytes
punpcklbw_r2r (mm0, mm5); // unpack low src4 bytes
punpckhbw_r2r (mm0, mm6); // unpack high src4 bytes
paddw_r2r (mm5, mm1); // add lows
paddw_r2r (mm6, mm2); // add highs
/* now have subtotal in mm1 and mm2 */
paddw_m2r (round4, mm1);
psraw_i2r (2, mm1); // /4
paddw_m2r (round4, mm2);
psraw_i2r (2, mm2); // /4
packuswb_r2r (mm2, mm1); // pack (w/ saturation)
movq_r2m (mm1, *dest); // store result in dest
}
static void
deinterlace_line_mmx (uint8_t * dst, uint8_t * lum_m4,
uint8_t * lum_m3, uint8_t * lum_m2,
uint8_t * lum_m1, uint8_t * lum, int size)
{
mmx_t rounder;
rounder.uw[0] = 4;
rounder.uw[1] = 4;
rounder.uw[2] = 4;
rounder.uw[3] = 4;
pxor_r2r (mm7, mm7);
movq_m2r (rounder, mm6);
for (; size > 3; size -= 4) {
movd_m2r (*lum_m4, mm0);
movd_m2r (*lum_m3, mm1);
movd_m2r (*lum_m2, mm2);
movd_m2r (*lum_m1, mm3);
movd_m2r (*lum, mm4);
punpcklbw_r2r (mm7, mm0);
punpcklbw_r2r (mm7, mm1);
punpcklbw_r2r (mm7, mm2);
punpcklbw_r2r (mm7, mm3);
punpcklbw_r2r (mm7, mm4);
paddw_r2r (mm3, mm1);
psllw_i2r (1, mm2);
paddw_r2r (mm4, mm0);
psllw_i2r (2, mm1); // 2
paddw_r2r (mm6, mm2);
paddw_r2r (mm2, mm1);
psubusw_r2r (mm0, mm1);
psrlw_i2r (3, mm1); // 3
packuswb_r2r (mm7, mm1);
movd_r2m (mm1, *dst);
lum_m4 += 4;
lum_m3 += 4;
lum_m2 += 4;
lum_m1 += 4;
lum += 4;
dst += 4;
}
emms ();
/* Handle odd widths */
if (size > 0)
deinterlace_line_c (dst, lum_m4, lum_m3, lum_m2, lum_m1, lum, size);
}
static inline void XDeint8x8MergeMMXEXT( uint8_t *dst, int i_dst,
uint8_t *src1, int i_src1,
uint8_t *src2, int i_src2 )
{
static const uint64_t m_4 = INT64_C(0x0004000400040004);
int y, x;
/* Progressive */
pxor_r2r( mm7, mm7 );
for( y = 0; y < 8; y += 2 )
{
for( x = 0; x < 8; x +=4 )
{
movd_m2r( src1[x], mm0 );
movd_r2m( mm0, dst[x] );
movd_m2r( src2[x], mm1 );
movd_m2r( src1[i_src1+x], mm2 );
punpcklbw_r2r( mm7, mm0 );
punpcklbw_r2r( mm7, mm1 );
punpcklbw_r2r( mm7, mm2 );
paddw_r2r( mm1, mm1 );
movq_r2r( mm1, mm3 );
paddw_r2r( mm3, mm3 );
paddw_r2r( mm2, mm0 );
paddw_r2r( mm3, mm1 );
paddw_m2r( m_4, mm1 );
paddw_r2r( mm1, mm0 );
psraw_i2r( 3, mm0 );
packuswb_r2r( mm7, mm0 );
movd_r2m( mm0, dst[i_dst+x] );
}
dst += 2*i_dst;
src1 += i_src1;
src2 += i_src2;
}
}
static inline void mmx_yuv2rgb (uint8_t * py, uint8_t * pu, uint8_t * pv)
{
static mmx_t mmx_80w = {0x0080008000800080LL};
static mmx_t mmx_U_green = {0xf37df37df37df37dLL};
static mmx_t mmx_U_blue = {0x4093409340934093LL};
static mmx_t mmx_V_red = {0x3312331233123312LL};
static mmx_t mmx_V_green = {0xe5fce5fce5fce5fcLL};
static mmx_t mmx_10w = {0x1010101010101010LL};
static mmx_t mmx_00ffw = {0x00ff00ff00ff00ffLL};
static mmx_t mmx_Y_coeff = {0x253f253f253f253fLL};
movd_m2r (*pu, mm0); /* mm0 = 00 00 00 00 u3 u2 u1 u0 */
movd_m2r (*pv, mm1); /* mm1 = 00 00 00 00 v3 v2 v1 v0 */
movq_m2r (*py, mm6); /* mm6 = Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
pxor_r2r (mm4, mm4); /* mm4 = 0 */
/* XXX might do cache preload for image here */
/*
* Do the multiply part of the conversion for even and odd pixels
* register usage:
* mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels
* mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels
* mm6 -> Y even, mm7 -> Y odd
*/
punpcklbw_r2r (mm4, mm0); /* mm0 = u3 u2 u1 u0 */
punpcklbw_r2r (mm4, mm1); /* mm1 = v3 v2 v1 v0 */
psubsw_m2r (mmx_80w, mm0); /* u -= 128 */
psubsw_m2r (mmx_80w, mm1); /* v -= 128 */
psllw_i2r (3, mm0); /* promote precision */
psllw_i2r (3, mm1); /* promote precision */
movq_r2r (mm0, mm2); /* mm2 = u3 u2 u1 u0 */
movq_r2r (mm1, mm3); /* mm3 = v3 v2 v1 v0 */
pmulhw_m2r (mmx_U_green, mm2); /* mm2 = u * u_green */
pmulhw_m2r (mmx_V_green, mm3); /* mm3 = v * v_green */
pmulhw_m2r (mmx_U_blue, mm0); /* mm0 = chroma_b */
pmulhw_m2r (mmx_V_red, mm1); /* mm1 = chroma_r */
paddsw_r2r (mm3, mm2); /* mm2 = chroma_g */
psubusb_m2r (mmx_10w, mm6); /* Y -= 16 */
movq_r2r (mm6, mm7); /* mm7 = Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 */
pand_m2r (mmx_00ffw, mm6); /* mm6 = Y6 Y4 Y2 Y0 */
psrlw_i2r (8, mm7); /* mm7 = Y7 Y5 Y3 Y1 */
psllw_i2r (3, mm6); /* promote precision */
psllw_i2r (3, mm7); /* promote precision */
pmulhw_m2r (mmx_Y_coeff, mm6); /* mm6 = luma_rgb even */
pmulhw_m2r (mmx_Y_coeff, mm7); /* mm7 = luma_rgb odd */
/*
* Do the addition part of the conversion for even and odd pixels
* register usage:
* mm0 -> Cblue, mm1 -> Cred, mm2 -> Cgreen even pixels
* mm3 -> Cblue, mm4 -> Cred, mm5 -> Cgreen odd pixels
* mm6 -> Y even, mm7 -> Y odd
*/
movq_r2r (mm0, mm3); /* mm3 = chroma_b */
movq_r2r (mm1, mm4); /* mm4 = chroma_r */
movq_r2r (mm2, mm5); /* mm5 = chroma_g */
paddsw_r2r (mm6, mm0); /* mm0 = B6 B4 B2 B0 */
paddsw_r2r (mm7, mm3); /* mm3 = B7 B5 B3 B1 */
paddsw_r2r (mm6, mm1); /* mm1 = R6 R4 R2 R0 */
paddsw_r2r (mm7, mm4); /* mm4 = R7 R5 R3 R1 */
paddsw_r2r (mm6, mm2); /* mm2 = G6 G4 G2 G0 */
paddsw_r2r (mm7, mm5); /* mm5 = G7 G5 G3 G1 */
packuswb_r2r (mm0, mm0); /* saturate to 0-255 */
packuswb_r2r (mm1, mm1); /* saturate to 0-255 */
packuswb_r2r (mm2, mm2); /* saturate to 0-255 */
packuswb_r2r (mm3, mm3); /* saturate to 0-255 */
packuswb_r2r (mm4, mm4); /* saturate to 0-255 */
packuswb_r2r (mm5, mm5); /* saturate to 0-255 */
punpcklbw_r2r (mm3, mm0); /* mm0 = B7 B6 B5 B4 B3 B2 B1 B0 */
punpcklbw_r2r (mm4, mm1); /* mm1 = R7 R6 R5 R4 R3 R2 R1 R0 */
punpcklbw_r2r (mm5, mm2); /* mm2 = G7 G6 G5 G4 G3 G2 G1 G0 */
}
static void deinterlace_line( uint8_t *dst, uint8_t *lum_m4,
uint8_t *lum_m3, uint8_t *lum_m2,
uint8_t *lum_m1, uint8_t *lum, int size )
{
#if defined(__i386__) || defined(__x86_64__)
mmx_t rounder;
rounder.uw[0]=4;
rounder.uw[1]=4;
rounder.uw[2]=4;
rounder.uw[3]=4;
pxor_r2r(mm7,mm7);
movq_m2r(rounder,mm6);
for (;size > 3; size-=4) {
movd_m2r(lum_m4[0],mm0);
movd_m2r(lum_m3[0],mm1);
movd_m2r(lum_m2[0],mm2);
movd_m2r(lum_m1[0],mm3);
movd_m2r(lum[0],mm4);
punpcklbw_r2r(mm7,mm0);
punpcklbw_r2r(mm7,mm1);
punpcklbw_r2r(mm7,mm2);
punpcklbw_r2r(mm7,mm3);
punpcklbw_r2r(mm7,mm4);
paddw_r2r(mm3,mm1);
psllw_i2r(1,mm2);
paddw_r2r(mm4,mm0);
psllw_i2r(2,mm1);// 2
paddw_r2r(mm6,mm2);
paddw_r2r(mm2,mm1);
psubusw_r2r(mm0,mm1);
psrlw_i2r(3,mm1); // 3
packuswb_r2r(mm7,mm1);
movd_r2m(mm1,dst[0]);
lum_m4+=4;
lum_m3+=4;
lum_m2+=4;
lum_m1+=4;
lum+=4;
dst+=4;
}
emms();
#else
/**
* C implementation.
*/
int sum;
for(;size > 0;size--) {
sum = -lum_m4[0];
sum += lum_m3[0] << 2;
sum += lum_m2[0] << 1;
sum += lum_m1[0] << 2;
sum += -lum[0];
dst[0] = (sum + 4) >> 3; // This needs to be clipped at 0 and 255: cm[(sum + 4) >> 3];
lum_m4++;
lum_m3++;
lum_m2++;
lum_m1++;
lum++;
dst++;
}
#endif
}
static inline void mmx_interp_average_4_U8 (uint8_t * dest,
const uint8_t * src1,
const uint8_t * src2,
const uint8_t * src3,
const uint8_t * src4)
{
/* *dest = (*dest + (*src1 + *src2 + *src3 + *src4 + 2)/ 4 + 1)/ 2; */
movq_m2r (*src1, mm1); /* load 8 src1 bytes */
movq_r2r (mm1, mm2); /* copy 8 src1 bytes */
punpcklbw_r2r (mm0, mm1); /* unpack low src1 bytes */
punpckhbw_r2r (mm0, mm2); /* unpack high src1 bytes */
movq_m2r (*src2, mm3); /* load 8 src2 bytes */
movq_r2r (mm3, mm4); /* copy 8 src2 bytes */
punpcklbw_r2r (mm0, mm3); /* unpack low src2 bytes */
punpckhbw_r2r (mm0, mm4); /* unpack high src2 bytes */
paddw_r2r (mm3, mm1); /* add lows */
paddw_r2r (mm4, mm2); /* add highs */
/* now have partials in mm1 and mm2 */
movq_m2r (*src3, mm3); /* load 8 src3 bytes */
movq_r2r (mm3, mm4); /* copy 8 src3 bytes */
punpcklbw_r2r (mm0, mm3); /* unpack low src3 bytes */
punpckhbw_r2r (mm0, mm4); /* unpack high src3 bytes */
paddw_r2r (mm3, mm1); /* add lows */
paddw_r2r (mm4, mm2); /* add highs */
movq_m2r (*src4, mm5); /* load 8 src4 bytes */
movq_r2r (mm5, mm6); /* copy 8 src4 bytes */
punpcklbw_r2r (mm0, mm5); /* unpack low src4 bytes */
punpckhbw_r2r (mm0, mm6); /* unpack high src4 bytes */
paddw_r2r (mm5, mm1); /* add lows */
paddw_r2r (mm6, mm2); /* add highs */
paddw_m2r (round4, mm1);
psraw_i2r (2, mm1); /* /4 */
paddw_m2r (round4, mm2);
psraw_i2r (2, mm2); /* /4 */
/* now have subtotal/4 in mm1 and mm2 */
movq_m2r (*dest, mm3); /* load 8 dest bytes */
movq_r2r (mm3, mm4); /* copy 8 dest bytes */
packuswb_r2r (mm2, mm1); /* pack (w/ saturation) */
movq_r2r (mm1,mm2); /* copy subresult */
pxor_r2r (mm1, mm3); /* xor srcavg and dest */
pand_m2r (mask1, mm3); /* mask lower bits */
psrlq_i2r (1, mm3); /* /2 */
por_r2r (mm2, mm4); /* or srcavg and dest */
psubb_r2r (mm3, mm4); /* subtract subresults */
movq_r2m (mm4, *dest); /* store result in dest */
}
static void scale_uint8_x_1_x_bilinear_mmx(gavl_video_scale_context_t * ctx, int scanline, uint8_t * dest_start)
{
int i, imax, index;
uint8_t * src, * dst, *src_start;
mmx_t tmp_mm;
/*
* mm0: Input1 Input2
* mm1: Factor
* mm2:
* mm3:
* mm4:
* mm5:
* mm6: 0
* mm7: scratch
*
*/
src_start = ctx->src + scanline * ctx->src_stride;
pxor_r2r(mm6, mm6);
dst = dest_start;
imax = ctx->dst_size / 4;
// imax = 0;
index = 0;
for(i = 0; i < imax; i++)
{
/* Load pixels */
src = src_start + ctx->table_h.pixels[index].index;
tmp_mm.uw[0] = *src;
tmp_mm.uw[1] = *(src+1);
src = src_start + ctx->table_h.pixels[index+1].index;
tmp_mm.uw[2] = *src;
tmp_mm.uw[3] = *(src+1);
movq_m2r(tmp_mm, mm0);
/* Load factors */
movq_m2r(ctx->table_h.pixels[index].factor_i[0], mm1);
movq_m2r(ctx->table_h.pixels[index+1].factor_i[0], mm7);
packssdw_r2r(mm7, mm1);
pmaddwd_r2r(mm0, mm1);
index += 2;
/* Load pixels */
src = src_start + ctx->table_h.pixels[index].index;
tmp_mm.uw[0] = *src;
tmp_mm.uw[1] = *(src+1);
src = src_start + ctx->table_h.pixels[index+1].index;
tmp_mm.uw[2] = *src;
tmp_mm.uw[3] = *(src+1);
movq_m2r(tmp_mm, mm0);
/* Load factors */
movq_m2r(ctx->table_h.pixels[index].factor_i[0], mm3);
movq_m2r(ctx->table_h.pixels[index+1].factor_i[0], mm7);
packssdw_r2r(mm7, mm3);
pmaddwd_r2r(mm0, mm3);
psrld_i2r(7, mm3);
psrld_i2r(7, mm1);
packssdw_r2r(mm3, mm1);
psrlw_i2r(7, mm1);
index += 2;
packuswb_r2r(mm6, mm1);
movd_r2m(mm1, *dst);
// *dst = tmp_mm.ub[0];
// *(dst+1) = tmp_mm.ub[4];
dst+=4;
}
ctx->need_emms = 1;
#if 1
imax = ctx->dst_size % 4;
for(i = 0; i < imax; i++)
{
src = (src_start + ctx->table_h.pixels[index].index);
*dst = (ctx->table_h.pixels[index].factor_i[0] * *src +
ctx->table_h.pixels[index].factor_i[1] * *(src+1)) >> 14;
dst++;
index++;
}
#endif
}
void
yuv411planar_to_rgb_mmx (const unsigned char *yuv, unsigned char *rgb,
unsigned int w, unsigned int h)
{
unsigned int xx, yy;
register const unsigned char *yp1, *up, *vp;
unsigned char *dp1;
/* plane pointers */
yp1 = yuv;
up = yuv + (w * h);
vp = up + (w * (h / 4));
/* destination pointers */
dp1 = rgb;
yp1 = yuv;
up = yuv + (w * h);
vp = up + ((w / 2) * (h / 2));
dp1 = rgb;
for (yy = 0; yy < h; yy++)
{
for (xx = 0; xx < w; xx += 8)
{
movq_m2r(*yp1, mm0);
movq_r2r(mm0, mm1);
psrlw_i2r(8, mm0);
psllw_i2r(8, mm1);
psrlw_i2r(8, mm1);
pxor_r2r(mm7, mm7);
movd_m2r(*up, mm3);
movd_m2r(*vp, mm2);
punpcklbw_r2r(mm7, mm2);
punpcklbw_r2r(mm7, mm3);
movq_m2r(CONST_16, mm4);
psubsw_r2r(mm4, mm0);
psubsw_r2r(mm4, mm1);
movq_m2r(CONST_128, mm5);
psubsw_r2r(mm5, mm2);
psubsw_r2r(mm5, mm3);
movq_m2r(CONST_YMUL, mm4);
pmullw_r2r(mm4, mm0);
pmullw_r2r(mm4, mm1);
movq_m2r(CONST_CRVCRV, mm7);
pmullw_r2r(mm3, mm7);
movq_m2r(CONST_CBUCBU, mm6);
pmullw_r2r(mm2, mm6);
movq_m2r(CONST_CGUCGU, mm5);
pmullw_r2r(mm2, mm5);
movq_m2r(CONST_CGVCGV, mm4);
pmullw_r2r(mm3, mm4);
movq_r2r(mm0, mm2);
paddsw_r2r(mm7, mm2);
paddsw_r2r(mm1, mm7);
psraw_i2r(RES, mm2);
psraw_i2r(RES, mm7);
packuswb_r2r(mm7, mm2);
pxor_r2r(mm7, mm7);
movq_r2r(mm2, mm3);
punpckhbw_r2r(mm7, mm2);
punpcklbw_r2r(mm3, mm7);
por_r2r(mm7, mm2);
movq_r2r(mm0, mm3);
psubsw_r2r(mm5, mm3);
psubsw_r2r(mm4, mm3);
paddsw_m2r(CONST_32, mm3);
movq_r2r(mm1, mm7);
psubsw_r2r(mm5, mm7);
psubsw_r2r(mm4, mm7);
paddsw_m2r(CONST_32, mm7);
psraw_i2r(RES, mm3);
psraw_i2r(RES, mm7);
packuswb_r2r(mm7, mm3);
pxor_r2r(mm7, mm7);
movq_r2r(mm3, mm4);
punpckhbw_r2r(mm7, mm3);
punpcklbw_r2r(mm4, mm7);
por_r2r(mm7, mm3);
movq_m2r(CONST_32, mm4);
paddsw_r2r(mm6, mm0);
paddsw_r2r(mm6, mm1);
paddsw_r2r(mm4, mm0);
paddsw_r2r(mm4, mm1);
psraw_i2r(RES, mm0);
psraw_i2r(RES, mm1);
packuswb_r2r(mm1, mm0);
pxor_r2r(mm7, mm7);
movq_r2r(mm0, mm5);
punpckhbw_r2r(mm7, mm0);
punpcklbw_r2r(mm5, mm7);
por_r2r(mm7, mm0);
pxor_r2r(mm1, mm1);
movq_r2r(mm0, mm5);
movq_r2r(mm3, mm6);
movq_r2r(mm2, mm7);
punpckhbw_r2r(mm3, mm2);
punpcklbw_r2r(mm6, mm7);
punpckhbw_r2r(mm1, mm0);
punpcklbw_r2r(mm1, mm5);
movq_r2r(mm7, mm1);
punpckhwd_r2r(mm5, mm7);
punpcklwd_r2r(mm5, mm1);
movq_r2r(mm2, mm4);
punpckhwd_r2r(mm0, mm2);
punpcklwd_r2r(mm0, mm4);
movntq_r2m(mm1, *(dp1));
movntq_r2m(mm7, *(dp1 + 8));
movntq_r2m(mm4, *(dp1 + 16));
movntq_r2m(mm2, *(dp1 + 24));
yp1 += 8;
up += 4;
vp += 4;
dp1 += 8 * 4;
}
if (yy & 0x1)
{
up -= w / 2;
vp -= w / 2;
}
}
emms();
}
static void
_evas_yv12torgb_sse(unsigned char **yuv, unsigned char *rgb, int w, int h)
{
#ifdef BUILD_MMX
int xx, yy;
register unsigned char *yp1, *up, *vp;
unsigned char *dp1;
/* destination pointers */
dp1 = rgb;
for (yy = 0; yy < h; yy++)
{
/* plane pointers */
yp1 = yuv[yy];
up = yuv[h + (yy / 2)];
vp = yuv[h + (h / 2) + (yy / 2)];
for (xx = 0; xx < (w - 7); xx += 8)
{
movd_m2r(*up, mm3);
movd_m2r(*vp, mm2);
movq_m2r(*yp1, mm0);
pxor_r2r(mm7, mm7);
punpcklbw_r2r(mm7, mm2);
punpcklbw_r2r(mm7, mm3);
movq_r2r(mm0, mm1);
psrlw_i2r(8, mm0);
psllw_i2r(8, mm1);
psrlw_i2r(8, mm1);
movq_m2r(CONST_16, mm4);
psubsw_r2r(mm4, mm0);
psubsw_r2r(mm4, mm1);
movq_m2r(CONST_128, mm5);
psubsw_r2r(mm5, mm2);
psubsw_r2r(mm5, mm3);
movq_m2r(CONST_YMUL, mm4);
pmullw_r2r(mm4, mm0);
pmullw_r2r(mm4, mm1);
movq_m2r(CONST_CRVCRV, mm7);
pmullw_r2r(mm3, mm7);
movq_m2r(CONST_CBUCBU, mm6);
pmullw_r2r(mm2, mm6);
movq_m2r(CONST_CGUCGU, mm5);
pmullw_r2r(mm2, mm5);
movq_m2r(CONST_CGVCGV, mm4);
pmullw_r2r(mm3, mm4);
movq_r2r(mm0, mm2);
paddsw_r2r(mm7, mm2);
paddsw_r2r(mm1, mm7);
psraw_i2r(RES, mm2);
psraw_i2r(RES, mm7);
packuswb_r2r(mm7, mm2);
pxor_r2r(mm7, mm7);
movq_r2r(mm2, mm3);
punpckhbw_r2r(mm7, mm2);
punpcklbw_r2r(mm3, mm7);
por_r2r(mm7, mm2);
movq_r2r(mm0, mm3);
psubsw_r2r(mm5, mm3);
psubsw_r2r(mm4, mm3);
paddsw_m2r(CONST_32, mm3);
movq_r2r(mm1, mm7);
psubsw_r2r(mm5, mm7);
psubsw_r2r(mm4, mm7);
paddsw_m2r(CONST_32, mm7);
psraw_i2r(RES, mm3);
psraw_i2r(RES, mm7);
packuswb_r2r(mm7, mm3);
pxor_r2r(mm7, mm7);
movq_r2r(mm3, mm4);
punpckhbw_r2r(mm7, mm3);
punpcklbw_r2r(mm4, mm7);
por_r2r(mm7, mm3);
movq_m2r(CONST_32, mm4);
paddsw_r2r(mm6, mm0);
paddsw_r2r(mm6, mm1);
paddsw_r2r(mm4, mm0);
paddsw_r2r(mm4, mm1);
psraw_i2r(RES, mm0);
psraw_i2r(RES, mm1);
packuswb_r2r(mm1, mm0);
pxor_r2r(mm7, mm7);
movq_r2r(mm0, mm5);
punpckhbw_r2r(mm7, mm0);
punpcklbw_r2r(mm5, mm7);
por_r2r(mm7, mm0);
movq_m2r(CONST_FF, mm1);
movq_r2r(mm0, mm5);
movq_r2r(mm3, mm6);
movq_r2r(mm2, mm7);
punpckhbw_r2r(mm3, mm2);
punpcklbw_r2r(mm6, mm7);
punpckhbw_r2r(mm1, mm0);
punpcklbw_r2r(mm1, mm5);
movq_r2r(mm7, mm1);
punpckhwd_r2r(mm5, mm7);
punpcklwd_r2r(mm5, mm1);
movq_r2r(mm2, mm4);
punpckhwd_r2r(mm0, mm2);
punpcklwd_r2r(mm0, mm4);
movntq_r2m(mm1, *(dp1));
movntq_r2m(mm7, *(dp1 + 8));
movntq_r2m(mm4, *(dp1 + 16));
movntq_r2m(mm2, *(dp1 + 24));
yp1 += 8;
up += 4;
vp += 4;
dp1 += 8 * 4;
}
/* cleanup pixles that arent a multiple of 8 pixels wide */
if (xx < w)
{
int y, u, v, r, g, b;
for (; xx < w; xx += 2)
{
u = (*up++) - 128;
v = (*vp++) - 128;
y = RZ(YMUL) * ((*yp1++) - 16);
r = LUT_CLIP((y + (_crv * v)) >> RES);
g = LUT_CLIP((y - (_cgu * u) - (_cgv * v) + RZ(OFF)) >> RES);
b = LUT_CLIP((y + (_cbu * u) + RZ(OFF)) >> RES);
*((DATA32 *) dp1) = 0xff000000 + RGB_JOIN(r,g,b);
dp1 += 4;
y = RZ(YMUL) * ((*yp1++) - 16);
r = LUT_CLIP((y + (_crv * v)) >> RES);
g = LUT_CLIP((y - (_cgu * u) - (_cgv * v) + RZ(OFF)) >> RES);
b = LUT_CLIP((y + (_cbu * u) + RZ(OFF)) >> RES);
*((DATA32 *) dp1) = 0xff000000 + RGB_JOIN(r,g,b);
dp1 += 4;
}
}
}
int bsad_mmx(uint8_t *pf, uint8_t *pb, uint8_t *p2, int lx, int hxf, int hyf, int hxb, int hyb, int h)
{
uint8_t *pfa,*pfb,*pfc,*pba,*pbb,*pbc;
int s, s1, s2;
pfa = pf + hxf;
pfb = pf + lx * hyf;
pfc = pfb + hxf;
pba = pb + hxb;
pbb = pb + lx * hyb;
pbc = pbb + hxb;
s = 0; /* the accumulator */
if (h > 0)
{
pxor_r2r(mm7, mm7);
pxor_r2r(mm6, mm6);
pcmpeqw_r2r(mm5, mm5);
psubw_r2r(mm5, mm6);
psllw_i2r(1, mm6);
do {
BSAD_LOAD(pf[0],mm0,mm1);
BSAD_LOAD_ACC(pfa[0],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pfb[0],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pfc[0],mm2,mm3,mm0,mm1);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psrlw_i2r(2, mm0);
psrlw_i2r(2, mm1);
BSAD_LOAD(pb[0],mm2,mm3);
BSAD_LOAD_ACC(pba[0],mm4,mm5,mm2,mm3);
BSAD_LOAD_ACC(pbb[0],mm4,mm5,mm2,mm3);
BSAD_LOAD_ACC(pbc[0],mm4,mm5,mm2,mm3);
paddw_r2r(mm6, mm2);
paddw_r2r(mm6, mm3);
psrlw_i2r(2, mm2);
psrlw_i2r(2, mm3);
paddw_r2r(mm2, mm0);
paddw_r2r(mm3, mm1);
psrlw_i2r(1, mm6);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psllw_i2r(1, mm6);
psrlw_i2r(1, mm0);
psrlw_i2r(1, mm1);
packuswb_r2r(mm1, mm0);
movq_m2r(p2[0], mm1);
movq_r2r(mm0, mm2);
psubusb_r2r(mm1, mm0);
psubusb_r2r(mm2, mm1);
por_r2r(mm1, mm0);
movq_r2r(mm0, mm1);
punpcklbw_r2r(mm7, mm0);
punpckhbw_r2r(mm7, mm1);
paddw_r2r(mm1, mm0);
movq_r2r(mm0, mm1);
punpcklwd_r2r(mm7, mm0);
punpckhwd_r2r(mm7, mm1);
paddd_r2r(mm1, mm0);
movd_r2g(mm0, s1);
psrlq_i2r(32, mm0);
movd_r2g(mm0, s2);
s += s1 + s2;
BSAD_LOAD(pf[8],mm0,mm1);
BSAD_LOAD_ACC(pfa[8],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pfb[8],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pfc[8],mm2,mm3,mm0,mm1);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psrlw_i2r(2, mm0);
psrlw_i2r(2, mm1);
BSAD_LOAD(pb[8],mm2,mm3);
BSAD_LOAD_ACC(pba[8],mm4,mm5,mm2,mm3);
BSAD_LOAD_ACC(pbb[8],mm4,mm5,mm2,mm3);
BSAD_LOAD_ACC(pbc[8],mm4,mm5,mm2,mm3);
paddw_r2r(mm6, mm2);
paddw_r2r(mm6, mm3);
psrlw_i2r(2, mm2);
psrlw_i2r(2, mm3);
paddw_r2r(mm2, mm0);
paddw_r2r(mm3, mm1);
psrlw_i2r(1, mm6);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psllw_i2r(1, mm6);
psrlw_i2r(1, mm0);
psrlw_i2r(1, mm1);
packuswb_r2r(mm1, mm0);
movq_m2r(p2[8], mm1);
movq_r2r(mm0, mm2);
psubusb_r2r(mm1, mm0);
psubusb_r2r(mm2, mm1);
por_r2r(mm1, mm0);
movq_r2r(mm0, mm1);
punpcklbw_r2r(mm7, mm0);
punpckhbw_r2r(mm7, mm1);
paddw_r2r(mm1, mm0);
movq_r2r(mm0, mm1);
punpcklwd_r2r(mm7, mm0);
punpckhwd_r2r(mm7, mm1);
paddd_r2r(mm1, mm0);
movd_r2g(mm0, s1);
psrlq_i2r(32, mm0);
movd_r2g(mm0, s2);
s += s1 + s2;
p2 += lx;
pf += lx;
pfa += lx;
pfb += lx;
pfc += lx;
pb += lx;
pba += lx;
pbb += lx;
pbc += lx;
h--;
} while (h > 0);
}
emms();
return s;
}
static int bsad_1quad_mmxe(uint8_t *pf, uint8_t *pb, uint8_t *pb2, uint8_t *p2, int lx, int h)
{
int s;
s = 0; /* the accumulator */
if (h > 0)
{
pcmpeqw_r2r(mm6, mm6);
psrlw_i2r(15, mm6);
paddw_r2r(mm6, mm6);
pxor_r2r(mm7, mm7);
pxor_r2r(mm5, mm5);
do {
BSAD_LOAD(pf[0],mm0,mm1);
BSAD_LOAD_ACC(pf[1],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pf[lx],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pf[lx+1],mm2,mm3,mm0,mm1);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psrlw_i2r(2, mm0);
psrlw_i2r(2, mm1);
packuswb_r2r(mm1, mm0);
movq_m2r(pb2[0],mm1);
pavgb_m2r(pb[0],mm1);
pavgb_r2r(mm1, mm0);
psadbw_m2r(p2[0],mm0);
paddd_r2r(mm0,mm5);
BSAD_LOAD(pf[8],mm0,mm1);
BSAD_LOAD_ACC(pf[9],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pf[lx+8],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pf[lx+9],mm2,mm3,mm0,mm1);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psrlw_i2r(2, mm0);
psrlw_i2r(2, mm1);
packuswb_r2r(mm1, mm0);
movq_m2r(pb2[8],mm1);
pavgb_m2r(pb[8],mm1);
pavgb_r2r(mm1, mm0);
psadbw_m2r(p2[8],mm0);
paddd_r2r(mm0,mm5);
p2 += lx;
pf += lx;
pb += lx;
pb2 += lx;
h--;
} while (h > 0);
}
movd_r2g(mm5,s);
emms();
return s;
}
