OLD | NEW |
1 /* | 1 /* |
2 * Copyright 2006 The Android Open Source Project | 2 * Copyright 2006 The Android Open Source Project |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license that can be | 4 * Use of this source code is governed by a BSD-style license that can be |
5 * found in the LICENSE file. | 5 * found in the LICENSE file. |
6 */ | 6 */ |
7 | 7 |
8 #ifndef SkUtils_DEFINED | 8 #ifndef SkUtils_DEFINED |
9 #define SkUtils_DEFINED | 9 #define SkUtils_DEFINED |
10 | 10 |
11 #include "SkTypes.h" | 11 #include "SkTypes.h" |
| 12 #include "../private/SkOpts.h" |
12 | 13 |
13 /////////////////////////////////////////////////////////////////////////////// | 14 /////////////////////////////////////////////////////////////////////////////// |
14 | 15 |
15 // Determined empirically using bench/MemsetBench.cpp on a Nexus 7, Nexus 9, and
desktop. | 16 // The inlining heuristics below were determined using bench/MemsetBench.cpp |
16 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2 || defined(SK_ARM_HAS_NEON) | 17 // on a x86 desktop, a Nexus 7 with and without NEON, and a Nexus 9: |
17 // Platforms where we can assume an autovectorizer will give us a good inlin
e memset. | 18 // - on x86, inlining was never faster, |
18 #define SK_SMALL_MEMSET 1000 | 19 // - on ARMv7, inlining was faster for N<=10. Putting this check inside the N
EON |
19 #else | 20 // code was not helpful; it's got to be here outside. |
20 // Platforms like Chrome on ARMv7 that don't typically compile with NEON glo
bally. | 21 // - NEON code generation for ARMv8 with GCC 4.9 is terrible, |
21 #define SK_SMALL_MEMSET 10 | 22 // making the NEON code ~8x slower that just a serial loop. |
22 #endif | |
23 | |
24 | 23 |
25 /** Similar to memset(), but it assigns a 16bit value into the buffer. | 24 /** Similar to memset(), but it assigns a 16bit value into the buffer. |
26 @param buffer The memory to have value copied into it | 25 @param buffer The memory to have value copied into it |
27 @param value The 16bit value to be copied into buffer | 26 @param value The 16bit value to be copied into buffer |
28 @param count The number of times value should be copied into the buffer. | 27 @param count The number of times value should be copied into the buffer. |
29 */ | 28 */ |
30 void sk_memset16_large(uint16_t dst[], uint16_t value, int count); | 29 static inline void sk_memset16(uint16_t buffer[], uint16_t value, int count) { |
31 inline void sk_memset16(uint16_t dst[], uint16_t value, int count) { | 30 #if defined(SK_CPU_ARM64) |
32 if (count <= SK_SMALL_MEMSET) { | 31 while (count --> 0) { *buffer++ = value; } return; |
33 for (int i = 0; i < count; i++) { | 32 #elif defined(SK_CPU_ARM32) |
34 dst[i] = value; | 33 if (count <= 10) { while (count --> 0) { *buffer++ = value; } return; } |
35 } | 34 #endif |
36 } else { | 35 SkOpts::memset16(buffer, value, count); |
37 sk_memset16_large(dst, value, count); | |
38 } | |
39 } | 36 } |
40 typedef void (*SkMemset16Proc)(uint16_t dst[], uint16_t value, int count); | |
41 SkMemset16Proc SkMemset16GetPlatformProc(); | |
42 | 37 |
43 /** Similar to memset(), but it assigns a 32bit value into the buffer. | 38 /** Similar to memset(), but it assigns a 32bit value into the buffer. |
44 @param buffer The memory to have value copied into it | 39 @param buffer The memory to have value copied into it |
45 @param value The 32bit value to be copied into buffer | 40 @param value The 32bit value to be copied into buffer |
46 @param count The number of times value should be copied into the buffer. | 41 @param count The number of times value should be copied into the buffer. |
47 */ | 42 */ |
48 void sk_memset32_large(uint32_t dst[], uint32_t value, int count); | 43 static inline void sk_memset32(uint32_t buffer[], uint32_t value, int count) { |
49 inline void sk_memset32(uint32_t dst[], uint32_t value, int count) { | 44 #if defined(SK_CPU_ARM64) |
50 if (count <= SK_SMALL_MEMSET) { | 45 while (count --> 0) { *buffer++ = value; } return; |
51 for (int i = 0; i < count; i++) { | 46 #elif defined(SK_CPU_ARM32) |
52 dst[i] = value; | 47 if (count <= 10) { while (count --> 0) { *buffer++ = value; } return; } |
53 } | 48 #endif |
54 } else { | 49 SkOpts::memset32(buffer, value, count); |
55 sk_memset32_large(dst, value, count); | |
56 } | |
57 } | 50 } |
58 | 51 |
59 typedef void (*SkMemset32Proc)(uint32_t dst[], uint32_t value, int count); | |
60 SkMemset32Proc SkMemset32GetPlatformProc(); | |
61 | |
62 #undef SK_SMALL_MEMSET | |
63 | 52 |
64 /////////////////////////////////////////////////////////////////////////////// | 53 /////////////////////////////////////////////////////////////////////////////// |
65 | 54 |
66 #define kMaxBytesInUTF8Sequence 4 | 55 #define kMaxBytesInUTF8Sequence 4 |
67 | 56 |
68 #ifdef SK_DEBUG | 57 #ifdef SK_DEBUG |
69 int SkUTF8_LeadByteToCount(unsigned c); | 58 int SkUTF8_LeadByteToCount(unsigned c); |
70 #else | 59 #else |
71 #define SkUTF8_LeadByteToCount(c) ((((0xE5 << 24) >> ((unsigned)c >> 4 <<
1)) & 3) + 1) | 60 #define SkUTF8_LeadByteToCount(c) ((((0xE5 << 24) >> ((unsigned)c >> 4 <<
1)) & 3) + 1) |
72 #endif | 61 #endif |
(...skipping 58 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
131 } | 120 } |
132 ~SkAutoTrace() { | 121 ~SkAutoTrace() { |
133 SkDebugf("--- trace: %s Leave\n", fLabel); | 122 SkDebugf("--- trace: %s Leave\n", fLabel); |
134 } | 123 } |
135 private: | 124 private: |
136 const char* fLabel; | 125 const char* fLabel; |
137 }; | 126 }; |
138 #define SkAutoTrace(...) SK_REQUIRE_LOCAL_VAR(SkAutoTrace) | 127 #define SkAutoTrace(...) SK_REQUIRE_LOCAL_VAR(SkAutoTrace) |
139 | 128 |
140 #endif | 129 #endif |
OLD | NEW |