/src/aom/aom_ports/bitops.h
Line | Count | Source (jump to first uncovered line) |
1 | | /* |
2 | | * Copyright (c) 2016, Alliance for Open Media. All rights reserved |
3 | | * |
4 | | * This source code is subject to the terms of the BSD 2 Clause License and |
5 | | * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License |
6 | | * was not distributed with this source code in the LICENSE file, you can |
7 | | * obtain it at www.aomedia.org/license/software. If the Alliance for Open |
8 | | * Media Patent License 1.0 was not distributed with this source code in the |
9 | | * PATENTS file, you can obtain it at www.aomedia.org/license/patent. |
10 | | */ |
11 | | |
12 | | #ifndef AOM_AOM_PORTS_BITOPS_H_ |
13 | | #define AOM_AOM_PORTS_BITOPS_H_ |
14 | | |
15 | | #include <assert.h> |
16 | | #include <stdint.h> |
17 | | |
18 | | #include "aom_ports/msvc.h" |
19 | | #include "config/aom_config.h" |
20 | | |
21 | | #ifdef _MSC_VER |
22 | | #if defined(_M_X64) || defined(_M_IX86) |
23 | | #include <intrin.h> |
24 | | #define USE_MSC_INTRINSICS |
25 | | #endif |
26 | | #endif |
27 | | |
28 | | #ifdef __cplusplus |
29 | | extern "C" { |
30 | | #endif |
31 | | |
32 | | // get_msb: |
33 | | // Returns (int)floor(log2(n)). n must be > 0. |
34 | | // These versions of get_msb() are only valid when n != 0 because all |
35 | | // of the optimized versions are undefined when n == 0: |
36 | | |
37 | | // get_byteswap64: |
38 | | // Returns the number (uint64_t) with byte-positions reversed |
39 | | // e.g. input 0x123456789ABCDEF0 returns 0xF0DEBC9A78563412 |
40 | | |
41 | | // GCC compiler: https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html |
42 | | // MSVC: https://learn.microsoft.com/en-us/cpp/c-runtime-library/ |
43 | | |
44 | | // use GNU builtins where available. |
45 | | #if defined(__GNUC__) && \ |
46 | | ((__GNUC__ == 3 && __GNUC_MINOR__ >= 4) || __GNUC__ >= 4) |
47 | 903M | static INLINE int get_msb(unsigned int n) { |
48 | 903M | assert(n != 0); |
49 | 0 | return 31 ^ __builtin_clz(n); |
50 | 903M | } Unexecuted instantiation: av1_dx_iface.c:get_msb Line | Count | Source | 47 | 426k | static INLINE int get_msb(unsigned int n) { | 48 | 426k | assert(n != 0); | 49 | 0 | return 31 ^ __builtin_clz(n); | 50 | 426k | } |
Unexecuted instantiation: decodemv.c:get_msb Unexecuted instantiation: decoder.c:get_msb Unexecuted instantiation: decodetxb.c:get_msb Line | Count | Source | 47 | 121k | static INLINE int get_msb(unsigned int n) { | 48 | 121k | assert(n != 0); | 49 | 0 | return 31 ^ __builtin_clz(n); | 50 | 121k | } |
Unexecuted instantiation: obu.c:get_msb Unexecuted instantiation: aom_dsp_rtcd.c:get_msb Unexecuted instantiation: av1_rtcd.c:get_msb Unexecuted instantiation: aom_convolve.c:get_msb bitreader_buffer.c:get_msb Line | Count | Source | 47 | 22.1k | static INLINE int get_msb(unsigned int n) { | 48 | 22.1k | assert(n != 0); | 49 | 0 | return 31 ^ __builtin_clz(n); | 50 | 22.1k | } |
Unexecuted instantiation: blend_a64_hmask.c:get_msb Unexecuted instantiation: blend_a64_mask.c:get_msb Unexecuted instantiation: blend_a64_vmask.c:get_msb Unexecuted instantiation: intrapred.c:get_msb binary_codes_reader.c:get_msb Line | Count | Source | 47 | 104k | static INLINE int get_msb(unsigned int n) { | 48 | 104k | assert(n != 0); | 49 | 0 | return 31 ^ __builtin_clz(n); | 50 | 104k | } |
Unexecuted instantiation: bitreader.c:get_msb Line | Count | Source | 47 | 667M | static INLINE int get_msb(unsigned int n) { | 48 | 667M | assert(n != 0); | 49 | 0 | return 31 ^ __builtin_clz(n); | 50 | 667M | } |
Unexecuted instantiation: aom_convolve_copy_sse2.c:get_msb Unexecuted instantiation: aom_asm_stubs.c:get_msb Unexecuted instantiation: intrapred_sse2.c:get_msb Unexecuted instantiation: loopfilter_sse2.c:get_msb Unexecuted instantiation: highbd_convolve_sse2.c:get_msb Unexecuted instantiation: highbd_loopfilter_sse2.c:get_msb Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:get_msb Unexecuted instantiation: intrapred_ssse3.c:get_msb Unexecuted instantiation: highbd_convolve_ssse3.c:get_msb Unexecuted instantiation: blend_a64_hmask_sse4.c:get_msb Unexecuted instantiation: blend_a64_mask_sse4.c:get_msb Unexecuted instantiation: blend_a64_vmask_sse4.c:get_msb Unexecuted instantiation: intrapred_sse4.c:get_msb Unexecuted instantiation: aom_convolve_copy_avx2.c:get_msb Unexecuted instantiation: aom_subpixel_8t_intrin_avx2.c:get_msb Unexecuted instantiation: intrapred_avx2.c:get_msb Unexecuted instantiation: loopfilter_avx2.c:get_msb Unexecuted instantiation: blend_a64_mask_avx2.c:get_msb Unexecuted instantiation: highbd_convolve_avx2.c:get_msb Unexecuted instantiation: highbd_loopfilter_avx2.c:get_msb Unexecuted instantiation: alloccommon.c:get_msb Unexecuted instantiation: av1_inv_txfm2d.c:get_msb Unexecuted instantiation: av1_loopfilter.c:get_msb Unexecuted instantiation: av1_txfm.c:get_msb Unexecuted instantiation: blockd.c:get_msb Unexecuted instantiation: cdef.c:get_msb Line | Count | Source | 47 | 28.2M | static INLINE int get_msb(unsigned int n) { | 48 | 28.2M | assert(n != 0); | 49 | 0 | return 31 ^ __builtin_clz(n); | 50 | 28.2M | } |
Unexecuted instantiation: cfl.c:get_msb Unexecuted instantiation: convolve.c:get_msb Unexecuted instantiation: entropy.c:get_msb Unexecuted instantiation: entropymode.c:get_msb Unexecuted instantiation: entropymv.c:get_msb Unexecuted instantiation: idct.c:get_msb Unexecuted instantiation: mvref_common.c:get_msb Unexecuted instantiation: pred_common.c:get_msb Unexecuted instantiation: quant_common.c:get_msb Unexecuted instantiation: reconinter.c:get_msb Unexecuted instantiation: reconintra.c:get_msb Unexecuted instantiation: resize.c:get_msb Unexecuted instantiation: restoration.c:get_msb Unexecuted instantiation: scale.c:get_msb Unexecuted instantiation: scan.c:get_msb Unexecuted instantiation: seg_common.c:get_msb Unexecuted instantiation: thread_common.c:get_msb Unexecuted instantiation: tile_common.c:get_msb Unexecuted instantiation: txb_common.c:get_msb Line | Count | Source | 47 | 1.05M | static INLINE int get_msb(unsigned int n) { | 48 | 1.05M | assert(n != 0); | 49 | 0 | return 31 ^ __builtin_clz(n); | 50 | 1.05M | } |
Unexecuted instantiation: cdef_block_sse2.c:get_msb Unexecuted instantiation: cfl_sse2.c:get_msb Unexecuted instantiation: convolve_2d_sse2.c:get_msb Unexecuted instantiation: convolve_sse2.c:get_msb Unexecuted instantiation: jnt_convolve_sse2.c:get_msb Unexecuted instantiation: warp_plane_sse2.c:get_msb Unexecuted instantiation: wiener_convolve_sse2.c:get_msb Unexecuted instantiation: av1_inv_txfm_ssse3.c:get_msb Unexecuted instantiation: cdef_block_ssse3.c:get_msb Unexecuted instantiation: cfl_ssse3.c:get_msb Unexecuted instantiation: jnt_convolve_ssse3.c:get_msb Unexecuted instantiation: resize_ssse3.c:get_msb Unexecuted instantiation: highbd_convolve_2d_ssse3.c:get_msb Unexecuted instantiation: highbd_wiener_convolve_ssse3.c:get_msb Unexecuted instantiation: reconinter_ssse3.c:get_msb Unexecuted instantiation: av1_convolve_horiz_rs_sse4.c:get_msb Unexecuted instantiation: av1_convolve_scale_sse4.c:get_msb Unexecuted instantiation: av1_txfm_sse4.c:get_msb Unexecuted instantiation: cdef_block_sse4.c:get_msb Unexecuted instantiation: filterintra_sse4.c:get_msb Unexecuted instantiation: highbd_inv_txfm_sse4.c:get_msb Unexecuted instantiation: intra_edge_sse4.c:get_msb Unexecuted instantiation: reconinter_sse4.c:get_msb Unexecuted instantiation: selfguided_sse4.c:get_msb Unexecuted instantiation: warp_plane_sse4.c:get_msb Unexecuted instantiation: highbd_convolve_2d_sse4.c:get_msb Unexecuted instantiation: highbd_jnt_convolve_sse4.c:get_msb Unexecuted instantiation: highbd_warp_plane_sse4.c:get_msb Unexecuted instantiation: av1_inv_txfm_avx2.c:get_msb cdef_block_avx2.c:get_msb Line | Count | Source | 47 | 205M | static INLINE int get_msb(unsigned int n) { | 48 | 205M | assert(n != 0); | 49 | 0 | return 31 ^ __builtin_clz(n); | 50 | 205M | } |
Unexecuted instantiation: cfl_avx2.c:get_msb Unexecuted instantiation: convolve_2d_avx2.c:get_msb Unexecuted instantiation: convolve_avx2.c:get_msb Unexecuted instantiation: highbd_inv_txfm_avx2.c:get_msb Unexecuted instantiation: jnt_convolve_avx2.c:get_msb Unexecuted instantiation: reconinter_avx2.c:get_msb Unexecuted instantiation: selfguided_avx2.c:get_msb Unexecuted instantiation: warp_plane_avx2.c:get_msb Unexecuted instantiation: wiener_convolve_avx2.c:get_msb Unexecuted instantiation: highbd_convolve_2d_avx2.c:get_msb Unexecuted instantiation: highbd_jnt_convolve_avx2.c:get_msb Unexecuted instantiation: highbd_wiener_convolve_avx2.c:get_msb Unexecuted instantiation: highbd_warp_affine_avx2.c:get_msb Unexecuted instantiation: entcode.c:get_msb Unexecuted instantiation: aom_subpixel_8t_intrin_sse2.c:get_msb Unexecuted instantiation: highbd_intrapred_sse2.c:get_msb Unexecuted instantiation: av1_inv_txfm1d.c:get_msb |
51 | | |
52 | 0 | static INLINE uint64_t get_byteswap64(uint64_t num) { |
53 | 0 | return __builtin_bswap64(num); |
54 | 0 | } Unexecuted instantiation: av1_dx_iface.c:get_byteswap64 Unexecuted instantiation: decodeframe.c:get_byteswap64 Unexecuted instantiation: decodemv.c:get_byteswap64 Unexecuted instantiation: decoder.c:get_byteswap64 Unexecuted instantiation: decodetxb.c:get_byteswap64 Unexecuted instantiation: detokenize.c:get_byteswap64 Unexecuted instantiation: obu.c:get_byteswap64 Unexecuted instantiation: aom_dsp_rtcd.c:get_byteswap64 Unexecuted instantiation: av1_rtcd.c:get_byteswap64 Unexecuted instantiation: aom_convolve.c:get_byteswap64 Unexecuted instantiation: bitreader_buffer.c:get_byteswap64 Unexecuted instantiation: blend_a64_hmask.c:get_byteswap64 Unexecuted instantiation: blend_a64_mask.c:get_byteswap64 Unexecuted instantiation: blend_a64_vmask.c:get_byteswap64 Unexecuted instantiation: intrapred.c:get_byteswap64 Unexecuted instantiation: binary_codes_reader.c:get_byteswap64 Unexecuted instantiation: bitreader.c:get_byteswap64 Unexecuted instantiation: entdec.c:get_byteswap64 Unexecuted instantiation: aom_convolve_copy_sse2.c:get_byteswap64 Unexecuted instantiation: aom_asm_stubs.c:get_byteswap64 Unexecuted instantiation: intrapred_sse2.c:get_byteswap64 Unexecuted instantiation: loopfilter_sse2.c:get_byteswap64 Unexecuted instantiation: highbd_convolve_sse2.c:get_byteswap64 Unexecuted instantiation: highbd_loopfilter_sse2.c:get_byteswap64 Unexecuted instantiation: aom_subpixel_8t_intrin_ssse3.c:get_byteswap64 Unexecuted instantiation: intrapred_ssse3.c:get_byteswap64 Unexecuted instantiation: highbd_convolve_ssse3.c:get_byteswap64 Unexecuted instantiation: blend_a64_hmask_sse4.c:get_byteswap64 Unexecuted instantiation: blend_a64_mask_sse4.c:get_byteswap64 Unexecuted instantiation: blend_a64_vmask_sse4.c:get_byteswap64 Unexecuted instantiation: intrapred_sse4.c:get_byteswap64 Unexecuted instantiation: aom_convolve_copy_avx2.c:get_byteswap64 Unexecuted instantiation: aom_subpixel_8t_intrin_avx2.c:get_byteswap64 Unexecuted instantiation: intrapred_avx2.c:get_byteswap64 Unexecuted instantiation: loopfilter_avx2.c:get_byteswap64 Unexecuted instantiation: blend_a64_mask_avx2.c:get_byteswap64 Unexecuted instantiation: highbd_convolve_avx2.c:get_byteswap64 Unexecuted instantiation: highbd_loopfilter_avx2.c:get_byteswap64 Unexecuted instantiation: alloccommon.c:get_byteswap64 Unexecuted instantiation: av1_inv_txfm2d.c:get_byteswap64 Unexecuted instantiation: av1_loopfilter.c:get_byteswap64 Unexecuted instantiation: av1_txfm.c:get_byteswap64 Unexecuted instantiation: blockd.c:get_byteswap64 Unexecuted instantiation: cdef.c:get_byteswap64 Unexecuted instantiation: cdef_block.c:get_byteswap64 Unexecuted instantiation: cfl.c:get_byteswap64 Unexecuted instantiation: convolve.c:get_byteswap64 Unexecuted instantiation: entropy.c:get_byteswap64 Unexecuted instantiation: entropymode.c:get_byteswap64 Unexecuted instantiation: entropymv.c:get_byteswap64 Unexecuted instantiation: idct.c:get_byteswap64 Unexecuted instantiation: mvref_common.c:get_byteswap64 Unexecuted instantiation: pred_common.c:get_byteswap64 Unexecuted instantiation: quant_common.c:get_byteswap64 Unexecuted instantiation: reconinter.c:get_byteswap64 Unexecuted instantiation: reconintra.c:get_byteswap64 Unexecuted instantiation: resize.c:get_byteswap64 Unexecuted instantiation: restoration.c:get_byteswap64 Unexecuted instantiation: scale.c:get_byteswap64 Unexecuted instantiation: scan.c:get_byteswap64 Unexecuted instantiation: seg_common.c:get_byteswap64 Unexecuted instantiation: thread_common.c:get_byteswap64 Unexecuted instantiation: tile_common.c:get_byteswap64 Unexecuted instantiation: txb_common.c:get_byteswap64 Unexecuted instantiation: warped_motion.c:get_byteswap64 Unexecuted instantiation: cdef_block_sse2.c:get_byteswap64 Unexecuted instantiation: cfl_sse2.c:get_byteswap64 Unexecuted instantiation: convolve_2d_sse2.c:get_byteswap64 Unexecuted instantiation: convolve_sse2.c:get_byteswap64 Unexecuted instantiation: jnt_convolve_sse2.c:get_byteswap64 Unexecuted instantiation: warp_plane_sse2.c:get_byteswap64 Unexecuted instantiation: wiener_convolve_sse2.c:get_byteswap64 Unexecuted instantiation: av1_inv_txfm_ssse3.c:get_byteswap64 Unexecuted instantiation: cdef_block_ssse3.c:get_byteswap64 Unexecuted instantiation: cfl_ssse3.c:get_byteswap64 Unexecuted instantiation: jnt_convolve_ssse3.c:get_byteswap64 Unexecuted instantiation: resize_ssse3.c:get_byteswap64 Unexecuted instantiation: highbd_convolve_2d_ssse3.c:get_byteswap64 Unexecuted instantiation: highbd_wiener_convolve_ssse3.c:get_byteswap64 Unexecuted instantiation: reconinter_ssse3.c:get_byteswap64 Unexecuted instantiation: av1_convolve_horiz_rs_sse4.c:get_byteswap64 Unexecuted instantiation: av1_convolve_scale_sse4.c:get_byteswap64 Unexecuted instantiation: av1_txfm_sse4.c:get_byteswap64 Unexecuted instantiation: cdef_block_sse4.c:get_byteswap64 Unexecuted instantiation: filterintra_sse4.c:get_byteswap64 Unexecuted instantiation: highbd_inv_txfm_sse4.c:get_byteswap64 Unexecuted instantiation: intra_edge_sse4.c:get_byteswap64 Unexecuted instantiation: reconinter_sse4.c:get_byteswap64 Unexecuted instantiation: selfguided_sse4.c:get_byteswap64 Unexecuted instantiation: warp_plane_sse4.c:get_byteswap64 Unexecuted instantiation: highbd_convolve_2d_sse4.c:get_byteswap64 Unexecuted instantiation: highbd_jnt_convolve_sse4.c:get_byteswap64 Unexecuted instantiation: highbd_warp_plane_sse4.c:get_byteswap64 Unexecuted instantiation: av1_inv_txfm_avx2.c:get_byteswap64 Unexecuted instantiation: cdef_block_avx2.c:get_byteswap64 Unexecuted instantiation: cfl_avx2.c:get_byteswap64 Unexecuted instantiation: convolve_2d_avx2.c:get_byteswap64 Unexecuted instantiation: convolve_avx2.c:get_byteswap64 Unexecuted instantiation: highbd_inv_txfm_avx2.c:get_byteswap64 Unexecuted instantiation: jnt_convolve_avx2.c:get_byteswap64 Unexecuted instantiation: reconinter_avx2.c:get_byteswap64 Unexecuted instantiation: selfguided_avx2.c:get_byteswap64 Unexecuted instantiation: warp_plane_avx2.c:get_byteswap64 Unexecuted instantiation: wiener_convolve_avx2.c:get_byteswap64 Unexecuted instantiation: highbd_convolve_2d_avx2.c:get_byteswap64 Unexecuted instantiation: highbd_jnt_convolve_avx2.c:get_byteswap64 Unexecuted instantiation: highbd_wiener_convolve_avx2.c:get_byteswap64 Unexecuted instantiation: highbd_warp_affine_avx2.c:get_byteswap64 Unexecuted instantiation: entcode.c:get_byteswap64 Unexecuted instantiation: aom_subpixel_8t_intrin_sse2.c:get_byteswap64 Unexecuted instantiation: highbd_intrapred_sse2.c:get_byteswap64 Unexecuted instantiation: av1_inv_txfm1d.c:get_byteswap64 |
55 | | #elif defined(USE_MSC_INTRINSICS) |
56 | | #pragma intrinsic(_BitScanReverse) |
57 | | |
58 | | static INLINE int get_msb(unsigned int n) { |
59 | | unsigned long first_set_bit; |
60 | | assert(n != 0); |
61 | | _BitScanReverse(&first_set_bit, n); |
62 | | return first_set_bit; |
63 | | } |
64 | | |
65 | | static INLINE uint64_t get_byteswap64(uint64_t num) { |
66 | | return _byteswap_uint64(num); |
67 | | } |
68 | | #undef USE_MSC_INTRINSICS |
69 | | #else |
70 | | static INLINE int get_msb(unsigned int n) { |
71 | | int log = 0; |
72 | | unsigned int value = n; |
73 | | |
74 | | assert(n != 0); |
75 | | |
76 | | for (int shift = 16; shift != 0; shift >>= 1) { |
77 | | const unsigned int x = value >> shift; |
78 | | if (x != 0) { |
79 | | value = x; |
80 | | log += shift; |
81 | | } |
82 | | } |
83 | | return log; |
84 | | } |
85 | | |
86 | | static INLINE uint64_t get_byteswap64(uint64_t num) { |
87 | | uint64_t out = 0x00; |
88 | | uint64_t mask = 0xFF00000000000000; |
89 | | int bit_shift = 56; // 7 bytes |
90 | | // 4 ms bytes |
91 | | do { |
92 | | out |= (num & mask) >> bit_shift; |
93 | | mask >>= 8; |
94 | | bit_shift -= 16; |
95 | | } while (bit_shift >= 0); |
96 | | // 4 ls bytes |
97 | | bit_shift = 8; // 1 byte |
98 | | do { |
99 | | out |= (num & mask) << bit_shift; |
100 | | mask >>= 8; |
101 | | bit_shift += 16; |
102 | | } while (bit_shift <= 56); |
103 | | return out; |
104 | | } |
105 | | #endif |
106 | | |
107 | | #ifdef __cplusplus |
108 | | } // extern "C" |
109 | | #endif |
110 | | |
111 | | #endif // AOM_AOM_PORTS_BITOPS_H_ |