1 | // Copyright 2015 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 | // SSE4.1 variant of methods for lossless encoder |
11 | // |
12 | // Author: Skal (pascal.massimino@gmail.com) |
13 | |
14 | #include "./dsp.h" |
15 | |
16 | #if defined(WEBP_USE_SSE41) |
17 | #include <assert.h> |
18 | #include <smmintrin.h> |
19 | #include "./lossless.h" |
20 | |
21 | //------------------------------------------------------------------------------ |
22 | // Subtract-Green Transform |
23 | |
24 | static void SubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixels) { |
25 | int i; |
26 | const __m128i kCstShuffle = _mm_set_epi8(-1, 13, -1, 13, -1, 9, -1, 9, |
27 | -1, 5, -1, 5, -1, 1, -1, 1); |
28 | for (i = 0; i + 4 <= num_pixels; i += 4) { |
29 | const __m128i in = _mm_loadu_si128((__m128i*)&argb_data[i]); |
30 | const __m128i in_0g0g = _mm_shuffle_epi8(in, kCstShuffle); |
31 | const __m128i out = _mm_sub_epi8(in, in_0g0g); |
32 | _mm_storeu_si128((__m128i*)&argb_data[i], out); |
33 | } |
34 | // fallthrough and finish off with plain-C |
35 | if (i != num_pixels) { |
36 | VP8LSubtractGreenFromBlueAndRed_C(argb_data + i, num_pixels - i); |
37 | } |
38 | } |
39 | |
40 | //------------------------------------------------------------------------------ |
41 | // Entry point |
42 | |
43 | extern void VP8LEncDspInitSSE41(void); |
44 | |
45 | WEBP_TSAN_IGNORE_FUNCTION void VP8LEncDspInitSSE41(void) { |
46 | VP8LSubtractGreenFromBlueAndRed = SubtractGreenFromBlueAndRed; |
47 | } |
48 | |
49 | #else // !WEBP_USE_SSE41 |
50 | |
51 | WEBP_DSP_INIT_STUB(VP8LEncDspInitSSE41) |
52 | |
53 | #endif // WEBP_USE_SSE41 |
54 | |