Lines Matching refs:__m128i
30 const __m128i zero = _mm_setzero_si128();
31 const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
32 const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
33 const __m128i C2 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero);
34 const __m128i V1 = _mm_add_epi16(C0, C1);
35 const __m128i V2 = _mm_sub_epi16(V1, C2);
36 const __m128i b = _mm_packus_epi16(V2, V2);
44 const __m128i zero = _mm_setzero_si128();
45 const __m128i C0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c0), zero);
46 const __m128i C1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c1), zero);
47 const __m128i B0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(c2), zero);
48 const __m128i avg = _mm_add_epi16(C1, C0);
49 const __m128i A0 = _mm_srli_epi16(avg, 1);
50 const __m128i A1 = _mm_sub_epi16(A0, B0);
51 const __m128i BgtA = _mm_cmpgt_epi16(B0, A0);
52 const __m128i A2 = _mm_sub_epi16(A1, BgtA);
53 const __m128i A3 = _mm_srai_epi16(A2, 1);
54 const __m128i A4 = _mm_add_epi16(A0, A3);
55 const __m128i A5 = _mm_packus_epi16(A4, A4);
62 const __m128i zero = _mm_setzero_si128();
63 const __m128i A0 = _mm_cvtsi32_si128(a);
64 const __m128i B0 = _mm_cvtsi32_si128(b);
65 const __m128i C0 = _mm_cvtsi32_si128(c);
66 const __m128i AC0 = _mm_subs_epu8(A0, C0);
67 const __m128i CA0 = _mm_subs_epu8(C0, A0);
68 const __m128i BC0 = _mm_subs_epu8(B0, C0);
69 const __m128i CB0 = _mm_subs_epu8(C0, B0);
70 const __m128i AC = _mm_or_si128(AC0, CA0);
71 const __m128i BC = _mm_or_si128(BC0, CB0);
72 const __m128i pa = _mm_unpacklo_epi8(AC, zero); // |a - c|
73 const __m128i pb = _mm_unpacklo_epi8(BC, zero); // |b - c|
74 const __m128i diff = _mm_sub_epi16(pb, pa);
77 _mm_storeu_si128((__m128i*)out, diff);
83 static WEBP_INLINE void Average2_m128i(const __m128i* const a0,
84 const __m128i* const a1,
85 __m128i* const avg) {
87 const __m128i ones = _mm_set1_epi8(1);
88 const __m128i avg1 = _mm_avg_epu8(*a0, *a1);
89 const __m128i one = _mm_and_si128(_mm_xor_si128(*a0, *a1), ones);
95 __m128i* const avg) {
97 const __m128i ones = _mm_set1_epi8(1);
98 const __m128i A0 = _mm_cvtsi32_si128(a0);
99 const __m128i A1 = _mm_cvtsi32_si128(a1);
100 const __m128i avg1 = _mm_avg_epu8(A0, A1);
101 const __m128i one = _mm_and_si128(_mm_xor_si128(A0, A1), ones);
105 static WEBP_INLINE __m128i Average2_uint32_16_SSE2(uint32_t a0, uint32_t a1) {
106 const __m128i zero = _mm_setzero_si128();
107 const __m128i A0 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a0), zero);
108 const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
109 const __m128i sum = _mm_add_epi16(A1, A0);
114 __m128i output;
121 const __m128i zero = _mm_setzero_si128();
122 const __m128i avg1 = Average2_uint32_16_SSE2(a0, a2);
123 const __m128i A1 = _mm_unpacklo_epi8(_mm_cvtsi32_si128(a1), zero);
124 const __m128i sum = _mm_add_epi16(avg1, A1);
125 const __m128i avg2 = _mm_srli_epi16(sum, 1);
126 const __m128i A2 = _mm_packus_epi16(avg2, avg2);
133 const __m128i avg1 = Average2_uint32_16_SSE2(a0, a1);
134 const __m128i avg2 = Average2_uint32_16_SSE2(a2, a3);
135 const __m128i sum = _mm_add_epi16(avg2, avg1);
136 const __m128i avg3 = _mm_srli_epi16(sum, 1);
137 const __m128i A0 = _mm_packus_epi16(avg3, avg3);
187 const __m128i black = _mm_set1_epi32(ARGB_BLACK);
189 const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
190 const __m128i res = _mm_add_epi8(src, black);
191 _mm_storeu_si128((__m128i*)&out[i], res);
202 __m128i prev = _mm_set1_epi32(out[-1]);
205 const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
207 const __m128i shift0 = _mm_slli_si128(src, 4);
209 const __m128i sum0 = _mm_add_epi8(src, shift0);
211 const __m128i shift1 = _mm_slli_si128(sum0, 8);
213 const __m128i sum1 = _mm_add_epi8(sum0, shift1);
214 const __m128i res = _mm_add_epi8(sum1, prev);
215 _mm_storeu_si128((__m128i*)&out[i], res);
231 const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]); \
232 const __m128i other = _mm_loadu_si128((const __m128i*)&(IN)); \
233 const __m128i res = _mm_add_epi8(src, other); \
234 _mm_storeu_si128((__m128i*)&out[i], res); \
260 const __m128i Tother = _mm_loadu_si128((const __m128i*)&(IN)); \
261 const __m128i T = _mm_loadu_si128((const __m128i*)&upper[i]); \
262 const __m128i src = _mm_loadu_si128((const __m128i*)&in[i]); \
263 __m128i avg, res; \
266 _mm_storeu_si128((__m128i*)&out[i], res); \
280 __m128i avgLTL, avg; \
297 __m128i L = _mm_cvtsi32_si128(out[-1]);
299 __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
300 __m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]);
301 const __m128i T = _mm_loadu_si128((const __m128i*)&upper[i]);
302 const __m128i TR = _mm_loadu_si128((const __m128i*)&upper[i + 1]);
303 __m128i avgTTR;
322 const __m128i L_lo = _mm_unpacklo_epi32(L, T); \
323 const __m128i TL_lo = _mm_unpacklo_epi32(TL, T); \
324 const __m128i pb = _mm_sad_epu8(L_lo, TL_lo); /* pb = sum |L-TL|*/ \
325 const __m128i mask = _mm_cmpgt_epi32(pb, pa); \
326 const __m128i A = _mm_and_si128(mask, L); \
327 const __m128i B = _mm_andnot_si128(mask, T); \
328 const __m128i pred = _mm_or_si128(A, B); /* pred = (pa > b)? L : T*/ \
344 __m128i pa;
345 __m128i L = _mm_cvtsi32_si128(out[-1]);
347 __m128i T = _mm_loadu_si128((const __m128i*)&upper[i]);
348 __m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]);
349 __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
354 const __m128i T_lo = _mm_unpacklo_epi32(T, T);
355 const __m128i TL_lo = _mm_unpacklo_epi32(TL, T);
356 const __m128i T_hi = _mm_unpackhi_epi32(T, T);
357 const __m128i TL_hi = _mm_unpackhi_epi32(TL, T);
358 const __m128i s_lo = _mm_sad_epu8(T_lo, TL_lo);
359 const __m128i s_hi = _mm_sad_epu8(T_hi, TL_hi);
379 const __m128i all = _mm_add_epi16(L, (DIFF)); \
380 const __m128i alls = _mm_packus_epi16(all, all); \
381 const __m128i res = _mm_add_epi8(src, alls); \
395 const __m128i zero = _mm_setzero_si128();
396 const __m128i L8 = _mm_cvtsi32_si128(out[-1]);
397 __m128i L = _mm_unpacklo_epi8(L8, zero);
400 __m128i src = _mm_loadu_si128((const __m128i*)&in[i]);
401 const __m128i T = _mm_loadu_si128((const __m128i*)&upper[i]);
402 const __m128i T_lo = _mm_unpacklo_epi8(T, zero);
403 const __m128i T_hi = _mm_unpackhi_epi8(T, zero);
404 const __m128i TL = _mm_loadu_si128((const __m128i*)&upper[i - 1]);
405 const __m128i TL_lo = _mm_unpacklo_epi8(TL, zero);
406 const __m128i TL_hi = _mm_unpackhi_epi8(TL, zero);
407 __m128i diff_lo = _mm_sub_epi16(T_lo, TL_lo);
408 __m128i diff_hi = _mm_sub_epi16(T_hi, TL_hi);
435 const __m128i in = _mm_loadu_si128((const __m128i*)&src[i]); // argb
436 const __m128i A = _mm_srli_epi16(in, 8); // 0 a 0 g
437 const __m128i B = _mm_shufflelo_epi16(A, _MM_SHUFFLE(2, 2, 0, 0));
438 const __m128i C = _mm_shufflehi_epi16(B, _MM_SHUFFLE(2, 2, 0, 0)); // 0g0g
439 const __m128i out = _mm_add_epi8(in, C);
440 _mm_storeu_si128((__m128i*)&dst[i], out);
456 const __m128i mults_rb = _mm_set_epi16(
461 const __m128i mults_b2 = _mm_set_epi16(
465 const __m128i mask_ag = _mm_set1_epi32(0xff00ff00); // alpha-green masks
468 const __m128i in = _mm_loadu_si128((const __m128i*)&src[i]); // argb
469 const __m128i A = _mm_and_si128(in, mask_ag); // a 0 g 0
470 const __m128i B = _mm_shufflelo_epi16(A, _MM_SHUFFLE(2, 2, 0, 0));
471 const __m128i C = _mm_shufflehi_epi16(B, _MM_SHUFFLE(2, 2, 0, 0)); // g0g0
472 const __m128i D = _mm_mulhi_epi16(C, mults_rb); // x dr x db1
473 const __m128i E = _mm_add_epi8(in, D); // x r' x b'
474 const __m128i F = _mm_slli_epi16(E, 8); // r' 0 b' 0
475 const __m128i G = _mm_mulhi_epi16(F, mults_b2); // x db2 0 0
476 const __m128i H = _mm_srli_epi32(G, 8); // 0 x db2 0
477 const __m128i I = _mm_add_epi8(H, F); // r' x b'' 0
478 const __m128i J = _mm_srli_epi16(I, 8); // 0 r' 0 b''
479 const __m128i out = _mm_or_si128(J, A);
480 _mm_storeu_si128((__m128i*)&dst[i], out);
493 const __m128i* in = (const __m128i*)src;
494 __m128i* out = (__m128i*)dst;
498 __m128i in0 = _mm_loadu_si128(in + 0);
499 __m128i in1 = _mm_loadu_si128(in + 1);
500 __m128i in2 = _mm_loadu_si128(in + 2);
501 __m128i in3 = _mm_loadu_si128(in + 3);
502 __m128i in4 = _mm_loadu_si128(in + 4);
503 __m128i in5 = _mm_loadu_si128(in + 5);
504 __m128i in6 = _mm_loadu_si128(in + 6);
505 __m128i in7 = _mm_loadu_si128(in + 7);
529 const __m128i red_blue_mask = _mm_set1_epi32(0x00ff00ffu);
530 const __m128i* in = (const __m128i*)src;
531 __m128i* out = (__m128i*)dst;
533 const __m128i A1 = _mm_loadu_si128(in++);
534 const __m128i A2 = _mm_loadu_si128(in++);
535 const __m128i B1 = _mm_and_si128(A1, red_blue_mask); // R 0 B 0
536 const __m128i B2 = _mm_and_si128(A2, red_blue_mask); // R 0 B 0
537 const __m128i C1 = _mm_andnot_si128(red_blue_mask, A1); // 0 G 0 A
538 const __m128i C2 = _mm_andnot_si128(red_blue_mask, A2); // 0 G 0 A
539 const __m128i D1 = _mm_shufflelo_epi16(B1, _MM_SHUFFLE(2, 3, 0, 1));
540 const __m128i D2 = _mm_shufflelo_epi16(B2, _MM_SHUFFLE(2, 3, 0, 1));
541 const __m128i E1 = _mm_shufflehi_epi16(D1, _MM_SHUFFLE(2, 3, 0, 1));
542 const __m128i E2 = _mm_shufflehi_epi16(D2, _MM_SHUFFLE(2, 3, 0, 1));
543 const __m128i F1 = _mm_or_si128(E1, C1);
544 const __m128i F2 = _mm_or_si128(E2, C2);
557 const __m128i mask_0x0f = _mm_set1_epi8(0x0f);
558 const __m128i mask_0xf0 = _mm_set1_epi8(0xf0);
559 const __m128i* in = (const __m128i*)src;
560 __m128i* out = (__m128i*)dst;
562 const __m128i bgra0 = _mm_loadu_si128(in++); // bgra0|bgra1|bgra2|bgra3
563 const __m128i bgra4 = _mm_loadu_si128(in++); // bgra4|bgra5|bgra6|bgra7
564 const __m128i v0l = _mm_unpacklo_epi8(bgra0, bgra4); // b0b4g0g4r0r4a0a4...
565 const __m128i v0h = _mm_unpackhi_epi8(bgra0, bgra4); // b2b6g2g6r2r6a2a6...
566 const __m128i v1l = _mm_unpacklo_epi8(v0l, v0h); // b0b2b4b6g0g2g4g6...
567 const __m128i v1h = _mm_unpackhi_epi8(v0l, v0h); // b1b3b5b7g1g3g5g7...
568 const __m128i v2l = _mm_unpacklo_epi8(v1l, v1h); // b0...b7 | g0...g7
569 const __m128i v2h = _mm_unpackhi_epi8(v1l, v1h); // r0...r7 | a0...a7
570 const __m128i ga0 = _mm_unpackhi_epi64(v2l, v2h); // g0...g7 | a0...a7
571 const __m128i rb0 = _mm_unpacklo_epi64(v2h, v2l); // r0...r7 | b0...b7
572 const __m128i ga1 = _mm_srli_epi16(ga0, 4); // g0-|g1-|...|a6-|a7-
573 const __m128i rb1 = _mm_and_si128(rb0, mask_0xf0); // -r0|-r1|...|-b6|-a7
574 const __m128i ga2 = _mm_and_si128(ga1, mask_0x0f); // g0-|g1-|...|a6-|a7-
575 const __m128i rgba0 = _mm_or_si128(ga2, rb1); // rg0..rg7 | ba0..ba7
576 const __m128i rgba1 = _mm_srli_si128(rgba0, 8); // ba0..ba7 | 0
578 const __m128i rgba = _mm_unpacklo_epi8(rgba1, rgba0); // barg0...barg7
580 const __m128i rgba = _mm_unpacklo_epi8(rgba0, rgba1); // rgba0...rgba7
593 const __m128i mask_0xe0 = _mm_set1_epi8(0xe0);
594 const __m128i mask_0xf8 = _mm_set1_epi8(0xf8);
595 const __m128i mask_0x07 = _mm_set1_epi8(0x07);
596 const __m128i* in = (const __m128i*)src;
597 __m128i* out = (__m128i*)dst;
599 const __m128i bgra0 = _mm_loadu_si128(in++); // bgra0|bgra1|bgra2|bgra3
600 const __m128i bgra4 = _mm_loadu_si128(in++); // bgra4|bgra5|bgra6|bgra7
601 const __m128i v0l = _mm_unpacklo_epi8(bgra0, bgra4); // b0b4g0g4r0r4a0a4...
602 const __m128i v0h = _mm_unpackhi_epi8(bgra0, bgra4); // b2b6g2g6r2r6a2a6...
603 const __m128i v1l = _mm_unpacklo_epi8(v0l, v0h); // b0b2b4b6g0g2g4g6...
604 const __m128i v1h = _mm_unpackhi_epi8(v0l, v0h); // b1b3b5b7g1g3g5g7...
605 const __m128i v2l = _mm_unpacklo_epi8(v1l, v1h); // b0...b7 | g0...g7
606 const __m128i v2h = _mm_unpackhi_epi8(v1l, v1h); // r0...r7 | a0...a7
607 const __m128i ga0 = _mm_unpackhi_epi64(v2l, v2h); // g0...g7 | a0...a7
608 const __m128i rb0 = _mm_unpacklo_epi64(v2h, v2l); // r0...r7 | b0...b7
609 const __m128i rb1 = _mm_and_si128(rb0, mask_0xf8); // -r0..-r7|-b0..-b7
610 const __m128i g_lo1 = _mm_srli_epi16(ga0, 5);
611 const __m128i g_lo2 = _mm_and_si128(g_lo1, mask_0x07); // g0-...g7-|xx (3b)
612 const __m128i g_hi1 = _mm_slli_epi16(ga0, 3);
613 const __m128i g_hi2 = _mm_and_si128(g_hi1, mask_0xe0); // -g0...-g7|xx (3b)
614 const __m128i b0 = _mm_srli_si128(rb1, 8); // -b0...-b7|0
615 const __m128i rg1 = _mm_or_si128(rb1, g_lo2); // gr0...gr7|xx
616 const __m128i b1 = _mm_srli_epi16(b0, 3);
617 const __m128i gb1 = _mm_or_si128(b1, g_hi2); // bg0...bg7|xx
619 const __m128i rgba = _mm_unpacklo_epi8(gb1, rg1); // rggb0...rggb7
621 const __m128i rgba = _mm_unpacklo_epi8(rg1, gb1); // bgrb0...bgrb7
634 const __m128i mask_l = _mm_set_epi32(0, 0x00ffffff, 0, 0x00ffffff);
635 const __m128i mask_h = _mm_set_epi32(0x00ffffff, 0, 0x00ffffff, 0);
636 const __m128i* in = (const __m128i*)src;
640 const __m128i bgra0 = _mm_loadu_si128(in++); // bgra0|bgra1|bgra2|bgra3
641 const __m128i bgra4 = _mm_loadu_si128(in++); // bgra4|bgra5|bgra6|bgra7
642 const __m128i a0l = _mm_and_si128(bgra0, mask_l); // bgr0|0|bgr0|0
643 const __m128i a4l = _mm_and_si128(bgra4, mask_l); // bgr0|0|bgr0|0
644 const __m128i a0h = _mm_and_si128(bgra0, mask_h); // 0|bgr0|0|bgr0
645 const __m128i a4h = _mm_and_si128(bgra4, mask_h); // 0|bgr0|0|bgr0
646 const __m128i b0h = _mm_srli_epi64(a0h, 8); // 000b|gr00|000b|gr00
647 const __m128i b4h = _mm_srli_epi64(a4h, 8); // 000b|gr00|000b|gr00
648 const __m128i c0 = _mm_or_si128(a0l, b0h); // rgbrgb00|rgbrgb00
649 const __m128i c4 = _mm_or_si128(a4l, b4h); // rgbrgb00|rgbrgb00
650 const __m128i c2 = _mm_srli_si128(c0, 8);
651 const __m128i c6 = _mm_srli_si128(c4, 8);
652 _mm_storel_epi64((__m128i*)(dst + 0), c0);
653 _mm_storel_epi64((__m128i*)(dst + 6), c2);
654 _mm_storel_epi64((__m128i*)(dst + 12), c4);
655 _mm_storel_epi64((__m128i*)(dst + 18), c6);
