1 | // Copyright 2014 Google Inc. All Rights Reserved. |
2 | // |
3 | // Use of this source code is governed by a BSD-style license |
4 | // that can be found in the COPYING file in the root of the source |
5 | // tree. An additional intellectual property rights grant can be found |
6 | // in the file PATENTS. All contributing project authors may |
7 | // be found in the AUTHORS file in the root of the source tree. |
8 | // ----------------------------------------------------------------------------- |
9 | // |
10 | // ARGB making functions (SSE2 version). |
11 | // |
12 | // Author: Skal (pascal.massimino@gmail.com) |
13 | |
14 | #include "./dsp.h" |
15 | |
16 | #if defined(WEBP_USE_SSE2) |
17 | |
18 | #include <assert.h> |
19 | #include <emmintrin.h> |
20 | #include <string.h> |
21 | |
22 | static WEBP_INLINE uint32_t MakeARGB32(int a, int r, int g, int b) { |
23 | return (((uint32_t)a << 24) | (r << 16) | (g << 8) | b); |
24 | } |
25 | |
26 | static void PackARGB(const uint8_t* a, const uint8_t* r, const uint8_t* g, |
27 | const uint8_t* b, int len, uint32_t* out) { |
28 | if (g == r + 1) { // RGBA input order. Need to swap R and B. |
29 | int i = 0; |
30 | const int len_max = len & ~3; // max length processed in main loop |
31 | const __m128i red_blue_mask = _mm_set1_epi32(0x00ff00ffu); |
32 | assert(b == r + 2); |
33 | assert(a == r + 3); |
34 | for (; i < len_max; i += 4) { |
35 | const __m128i A = _mm_loadu_si128((const __m128i*)(r + 4 * i)); |
36 | const __m128i B = _mm_and_si128(A, red_blue_mask); // R 0 B 0 |
37 | const __m128i C = _mm_andnot_si128(red_blue_mask, A); // 0 G 0 A |
38 | const __m128i D = _mm_shufflelo_epi16(B, _MM_SHUFFLE(2, 3, 0, 1)); |
39 | const __m128i E = _mm_shufflehi_epi16(D, _MM_SHUFFLE(2, 3, 0, 1)); |
40 | const __m128i F = _mm_or_si128(E, C); |
41 | _mm_storeu_si128((__m128i*)(out + i), F); |
42 | } |
43 | for (; i < len; ++i) { |
44 | out[i] = MakeARGB32(a[4 * i], r[4 * i], g[4 * i], b[4 * i]); |
45 | } |
46 | } else { |
47 | assert(g == b + 1); |
48 | assert(r == b + 2); |
49 | assert(a == b + 3); |
50 | memcpy(out, b, len * 4); |
51 | } |
52 | } |
53 | |
54 | //------------------------------------------------------------------------------ |
55 | // Entry point |
56 | |
57 | extern void VP8EncDspARGBInitSSE2(void); |
58 | |
59 | WEBP_TSAN_IGNORE_FUNCTION void VP8EncDspARGBInitSSE2(void) { |
60 | VP8PackARGB = PackARGB; |
61 | } |
62 | |
63 | #else // !WEBP_USE_SSE2 |
64 | |
65 | WEBP_DSP_INIT_STUB(VP8EncDspARGBInitSSE2) |
66 | |
67 | #endif // WEBP_USE_SSE2 |
68 | |