| 1 | // Copyright 2018 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 | #ifndef WEBP_DSP_QUANT_H_ |
| 11 | #define WEBP_DSP_QUANT_H_ |
| 12 | |
| 13 | #include <string.h> |
| 14 | |
| 15 | #include "src/dsp/dsp.h" |
| 16 | #include "src/webp/types.h" |
| 17 | |
| 18 | #if defined(WEBP_USE_NEON) && !defined(WEBP_ANDROID_NEON) && \ |
| 19 | !defined(WEBP_HAVE_NEON_RTCD) |
| 20 | #include <arm_neon.h> |
| 21 | |
| 22 | #define IsFlat IsFlat_NEON |
| 23 | |
| 24 | static uint32_t horizontal_add_uint32x4(const uint32x4_t a) { |
| 25 | #if WEBP_AARCH64 |
| 26 | return vaddvq_u32(a); |
| 27 | #else |
| 28 | const uint64x2_t b = vpaddlq_u32(a); |
| 29 | const uint32x2_t c = vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)), |
| 30 | vreinterpret_u32_u64(vget_high_u64(b))); |
| 31 | return vget_lane_u32(c, 0); |
| 32 | #endif |
| 33 | } |
| 34 | |
| 35 | static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks, |
| 36 | int thresh) { |
| 37 | const int16x8_t tst_ones = vdupq_n_s16(-1); |
| 38 | uint32x4_t sum = vdupq_n_u32(0); |
| 39 | |
| 40 | for (int i = 0; i < num_blocks; ++i) { |
| 41 | // Set DC to zero. |
| 42 | const int16x8_t a_0 = vsetq_lane_s16(0, vld1q_s16(levels), 0); |
| 43 | const int16x8_t a_1 = vld1q_s16(levels + 8); |
| 44 | |
| 45 | const uint16x8_t b_0 = vshrq_n_u16(vtstq_s16(a_0, tst_ones), 15); |
| 46 | const uint16x8_t b_1 = vshrq_n_u16(vtstq_s16(a_1, tst_ones), 15); |
| 47 | |
| 48 | sum = vpadalq_u16(sum, b_0); |
| 49 | sum = vpadalq_u16(sum, b_1); |
| 50 | |
| 51 | levels += 16; |
| 52 | } |
| 53 | return thresh >= (int)horizontal_add_uint32x4(sum); |
| 54 | } |
| 55 | |
| 56 | #else |
| 57 | |
| 58 | #define IsFlat IsFlat_C |
| 59 | |
| 60 | static WEBP_INLINE int IsFlat(const int16_t* levels, int num_blocks, |
| 61 | int thresh) { |
| 62 | int score = 0; |
| 63 | while (num_blocks-- > 0) { // TODO(skal): refine positional scoring? |
| 64 | int i; |
| 65 | for (i = 1; i < 16; ++i) { // omit DC, we're only interested in AC |
| 66 | score += (levels[i] != 0); |
| 67 | if (score > thresh) return 0; |
| 68 | } |
| 69 | levels += 16; |
| 70 | } |
| 71 | return 1; |
| 72 | } |
| 73 | |
| 74 | #endif // defined(WEBP_USE_NEON) && !defined(WEBP_ANDROID_NEON) && |
| 75 | // !defined(WEBP_HAVE_NEON_RTCD) |
| 76 | |
| 77 | static WEBP_INLINE int IsFlatSource16(const uint8_t* src) { |
| 78 | const uint32_t v = src[0] * 0x01010101u; |
| 79 | int i; |
| 80 | for (i = 0; i < 16; ++i) { |
| 81 | if (memcmp(src + 0, &v, 4) || memcmp(src + 4, &v, 4) || |
| 82 | memcmp(src + 8, &v, 4) || memcmp(src + 12, &v, 4)) { |
| 83 | return 0; |
| 84 | } |
| 85 | src += BPS; |
| 86 | } |
| 87 | return 1; |
| 88 | } |
| 89 | |
| 90 | #endif // WEBP_DSP_QUANT_H_ |
| 91 | |