1 /*
2 * Copyright 2012 Google Inc.
3 *
4 * Use of this source code is governed by a BSD-style license that can be
5 * found in the LICENSE file.
6 */
7
8 #ifndef SkMathPriv_DEFINED
9 #define SkMathPriv_DEFINED
10
11 #include "SkMath.h"
12
13 #if defined(SK_BUILD_FOR_IOS) && (defined(SK_BUILD_FOR_ARM32) || defined(SK_BUILD_FOR_ARM64))
14 // iOS on ARM starts processes with the Flush-To-Zero (FTZ) and
15 // Denormals-Are-Zero (DAZ) bits in the fpscr register set.
16 // Algorithms that rely on denormalized numbers need alternative implementations.
17 // This can also be controlled in SSE with the MXCSR register,
18 // x87 with FSTCW/FLDCW, and mips with FCSR. This should be detected at runtime,
19 // or the library built one way or the other more generally (by the build).
20 #define SK_CPU_FLUSH_TO_ZERO
21 #endif
22
23 /** Returns -1 if n < 0, else returns 0
24 */
25 #define SkExtractSign(n) ((int32_t)(n) >> 31)
26
27 /** If sign == -1, returns -n, else sign must be 0, and returns n.
28 Typically used in conjunction with SkExtractSign().
29 */
SkApplySign(int32_t n,int32_t sign)30 static inline int32_t SkApplySign(int32_t n, int32_t sign) {
31 SkASSERT(sign == 0 || sign == -1);
32 return (n ^ sign) - sign;
33 }
34
35 /** Return x with the sign of y */
SkCopySign32(int32_t x,int32_t y)36 static inline int32_t SkCopySign32(int32_t x, int32_t y) {
37 return SkApplySign(x, SkExtractSign(x ^ y));
38 }
39
40 /** Given a positive value and a positive max, return the value
41 pinned against max.
42 Note: only works as long as max - value doesn't wrap around
43 @return max if value >= max, else value
44 */
SkClampUMax(unsigned value,unsigned max)45 static inline unsigned SkClampUMax(unsigned value, unsigned max) {
46 if (value > max) {
47 value = max;
48 }
49 return value;
50 }
51
52 ///////////////////////////////////////////////////////////////////////////////
53
54 /** Return a*b/255, truncating away any fractional bits. Only valid if both
55 a and b are 0..255
56 */
SkMulDiv255Trunc(U8CPU a,U8CPU b)57 static inline U8CPU SkMulDiv255Trunc(U8CPU a, U8CPU b) {
58 SkASSERT((uint8_t)a == a);
59 SkASSERT((uint8_t)b == b);
60 unsigned prod = a*b + 1;
61 return (prod + (prod >> 8)) >> 8;
62 }
63
64 /** Return (a*b)/255, taking the ceiling of any fractional bits. Only valid if
65 both a and b are 0..255. The expected result equals (a * b + 254) / 255.
66 */
SkMulDiv255Ceiling(U8CPU a,U8CPU b)67 static inline U8CPU SkMulDiv255Ceiling(U8CPU a, U8CPU b) {
68 SkASSERT((uint8_t)a == a);
69 SkASSERT((uint8_t)b == b);
70 unsigned prod = a*b + 255;
71 return (prod + (prod >> 8)) >> 8;
72 }
73
74 /** Just the rounding step in SkDiv255Round: round(value / 255)
75 */
SkDiv255Round(unsigned prod)76 static inline unsigned SkDiv255Round(unsigned prod) {
77 prod += 128;
78 return (prod + (prod >> 8)) >> 8;
79 }
80
SkPinToUnitFloat(float x)81 static inline float SkPinToUnitFloat(float x) {
82 return SkTMin(SkTMax(x, 0.0f), 1.0f);
83 }
84
85 /**
86 * Swap byte order of a 4-byte value, e.g. 0xaarrggbb -> 0xbbggrraa.
87 */
88 #if defined(_MSC_VER)
89 #include <intrin.h>
SkBSwap32(uint32_t v)90 static inline uint32_t SkBSwap32(uint32_t v) { return _byteswap_ulong(v); }
91 #else
SkBSwap32(uint32_t v)92 static inline uint32_t SkBSwap32(uint32_t v) { return __builtin_bswap32(v); }
93 #endif
94
95 //! Returns the number of leading zero bits (0...32)
96 int SkCLZ_portable(uint32_t);
97
98 #ifndef SkCLZ
99 #if defined(SK_BUILD_FOR_WIN32)
100 #include <intrin.h>
101
SkCLZ(uint32_t mask)102 static inline int SkCLZ(uint32_t mask) {
103 if (mask) {
104 unsigned long index;
105 _BitScanReverse(&index, mask);
106 // Suppress this bogus /analyze warning. The check for non-zero
107 // guarantees that _BitScanReverse will succeed.
108 #pragma warning(suppress : 6102) // Using 'index' from failed function call
109 return index ^ 0x1F;
110 } else {
111 return 32;
112 }
113 }
114 #elif defined(SK_CPU_ARM32) || defined(__GNUC__) || defined(__clang__)
SkCLZ(uint32_t mask)115 static inline int SkCLZ(uint32_t mask) {
116 // __builtin_clz(0) is undefined, so we have to detect that case.
117 return mask ? __builtin_clz(mask) : 32;
118 }
119 #else
120 #define SkCLZ(x) SkCLZ_portable(x)
121 #endif
122 #endif
123
124 /**
125 * Returns the smallest power-of-2 that is >= the specified value. If value
126 * is already a power of 2, then it is returned unchanged. It is undefined
127 * if value is <= 0.
128 */
SkNextPow2(int value)129 static inline int SkNextPow2(int value) {
130 SkASSERT(value > 0);
131 return 1 << (32 - SkCLZ(value - 1));
132 }
133
134 /**
135 * Returns the largest power-of-2 that is <= the specified value. If value
136 * is already a power of 2, then it is returned unchanged. It is undefined
137 * if value is <= 0.
138 */
SkPrevPow2(int value)139 static inline int SkPrevPow2(int value) {
140 SkASSERT(value > 0);
141 return 1 << (32 - SkCLZ(value >> 1));
142 }
143
144 /**
145 * Returns the log2 of the specified value, were that value to be rounded up
146 * to the next power of 2. It is undefined to pass 0. Examples:
147 * SkNextLog2(1) -> 0
148 * SkNextLog2(2) -> 1
149 * SkNextLog2(3) -> 2
150 * SkNextLog2(4) -> 2
151 * SkNextLog2(5) -> 3
152 */
SkNextLog2(uint32_t value)153 static inline int SkNextLog2(uint32_t value) {
154 SkASSERT(value != 0);
155 return 32 - SkCLZ(value - 1);
156 }
157
158 /**
159 * Returns the log2 of the specified value, were that value to be rounded down
160 * to the previous power of 2. It is undefined to pass 0. Examples:
161 * SkPrevLog2(1) -> 0
162 * SkPrevLog2(2) -> 1
163 * SkPrevLog2(3) -> 1
164 * SkPrevLog2(4) -> 2
165 * SkPrevLog2(5) -> 2
166 */
SkPrevLog2(uint32_t value)167 static inline int SkPrevLog2(uint32_t value) {
168 SkASSERT(value != 0);
169 return 32 - SkCLZ(value >> 1);
170 }
171
172 ///////////////////////////////////////////////////////////////////////////////
173
174 /**
175 * Return the next power of 2 >= n.
176 */
GrNextPow2(uint32_t n)177 static inline uint32_t GrNextPow2(uint32_t n) {
178 return n ? (1 << (32 - SkCLZ(n - 1))) : 1;
179 }
180
181 /**
182 * Returns the next power of 2 >= n or n if the next power of 2 can't be represented by size_t.
183 */
GrNextSizePow2(size_t n)184 static inline size_t GrNextSizePow2(size_t n) {
185 constexpr int kNumSizeTBits = 8 * sizeof(size_t);
186 constexpr size_t kHighBitSet = size_t(1) << (kNumSizeTBits - 1);
187
188 if (!n) {
189 return 1;
190 } else if (n >= kHighBitSet) {
191 return n;
192 }
193
194 n--;
195 uint32_t shift = 1;
196 while (shift < kNumSizeTBits) {
197 n |= n >> shift;
198 shift <<= 1;
199 }
200 return n + 1;
201 }
202
203 #endif
204