gfx/skia/patches/archive/arm-fixes.patch@6474c204b198
gfx/skia/patches/archive/arm-fixes.patch
Wed, 31 Dec 2014 06:09:35 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Wed, 31 Dec 2014 06:09:35 +0100
- changeset 0
- 6474c204b198
- permissions
- -rw-r--r--
Cloned upstream origin tor-browser at tor-browser-31.3.0esr-4.5-1-build1
revision ID fc1c9ff7c1b2defdbc039f12214767608f46423f for hacking purpose.
1 diff --git a/gfx/skia/include/core/SkMath.h b/gfx/skia/include/core/SkMath.h
2 --- a/gfx/skia/include/core/SkMath.h
3 +++ b/gfx/skia/include/core/SkMath.h
4 @@ -148,20 +148,17 @@ static inline bool SkIsPow2(int value) {
5 }
7 ///////////////////////////////////////////////////////////////////////////////
9 /** SkMulS16(a, b) multiplies a * b, but requires that a and b are both int16_t.
10 With this requirement, we can generate faster instructions on some
11 architectures.
12 */
13 -#if defined(__arm__) \
14 - && !defined(__thumb__) \
15 - && !defined(__ARM_ARCH_4T__) \
16 - && !defined(__ARM_ARCH_5T__)
17 +#ifdef SK_ARM_HAS_EDSP
18 static inline int32_t SkMulS16(S16CPU x, S16CPU y) {
19 SkASSERT((int16_t)x == x);
20 SkASSERT((int16_t)y == y);
21 int32_t product;
22 asm("smulbb %0, %1, %2 \n"
23 : "=r"(product)
24 : "r"(x), "r"(y)
25 );
26 diff --git a/gfx/skia/include/core/SkPostConfig.h b/gfx/skia/include/core/SkPostConfig.h
27 --- a/gfx/skia/include/core/SkPostConfig.h
28 +++ b/gfx/skia/include/core/SkPostConfig.h
29 @@ -300,8 +300,53 @@
30 #endif
31 #endif
33 //////////////////////////////////////////////////////////////////////
35 #ifndef SK_ALLOW_STATIC_GLOBAL_INITIALIZERS
36 #define SK_ALLOW_STATIC_GLOBAL_INITIALIZERS 1
37 #endif
38 +
39 +//////////////////////////////////////////////////////////////////////
40 +// ARM defines
41 +
42 +#if defined(__GNUC__) && defined(__arm__)
43 +
44 +# define SK_ARM_ARCH 3
45 +
46 +# if defined(__ARM_ARCH_4__) || defined(__ARM_ARCH_4T__) \
47 + || defined(_ARM_ARCH_4)
48 +# undef SK_ARM_ARCH
49 +# define SK_ARM_ARCH 4
50 +# endif
51 +
52 +# if defined(__ARM_ARCH_5__) || defined(__ARM_ARCH_5T__) \
53 + || defined(__ARM_ARCH_5E__) || defined(__ARM_ARCH_5TE__) \
54 + || defined(__ARM_ARCH_5TEJ__) || defined(_ARM_ARCH_5)
55 +# undef SK_ARM_ARCH
56 +# define SK_ARM_ARCH 5
57 +# endif
58 +
59 +# if defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) \
60 + || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) \
61 + || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) \
62 + || defined(__ARM_ARCH_6M__) || defined(_ARM_ARCH_6)
63 +# undef SK_ARM_ARCH
64 +# define SK_ARM_ARCH 6
65 +# endif
66 +
67 +# if defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) \
68 + || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) \
69 + || defined(__ARM_ARCH_7EM__) || defined(_ARM_ARCH_7)
70 +# undef SK_ARM_ARCH
71 +# define SK_ARM_ARCH 7
72 +# endif
73 +
74 +# undef SK_ARM_HAS_EDSP
75 +# if defined(__thumb2__) && (SK_ARM_ARCH >= 6) \
76 + || !defined(__thumb__) \
77 + && ((SK_ARM_ARCH > 5) || defined(__ARM_ARCH_5E__) \
78 + || defined(__ARM_ARCH_5TE__) || defined(__ARM_ARCH_5TEJ__))
79 +# define SK_ARM_HAS_EDSP 1
80 +# endif
81 +
82 +#endif
83 diff --git a/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp b/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp
84 --- a/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp
85 +++ b/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp
86 @@ -6,17 +6,17 @@
87 * found in the LICENSE file.
88 */
91 #include "SkBitmapProcState.h"
92 #include "SkColorPriv.h"
93 #include "SkUtils.h"
95 -#if __ARM_ARCH__ >= 6 && !defined(SK_CPU_BENDIAN)
96 +#if SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN)
97 void SI8_D16_nofilter_DX_arm(
98 const SkBitmapProcState& s,
99 const uint32_t* SK_RESTRICT xy,
100 int count,
101 uint16_t* SK_RESTRICT colors) __attribute__((optimize("O1")));
103 void SI8_D16_nofilter_DX_arm(const SkBitmapProcState& s,
104 const uint32_t* SK_RESTRICT xy,
105 @@ -177,17 +177,17 @@ void SI8_opaque_D32_nofilter_DX_arm(cons
106 : [xx] "+r" (xx), [count] "+r" (count), [colors] "+r" (colors)
107 : [table] "r" (table), [srcAddr] "r" (srcAddr)
108 : "memory", "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11"
109 );
110 }
112 s.fBitmap->getColorTable()->unlockColors(false);
113 }
114 -#endif //__ARM_ARCH__ >= 6 && !defined(SK_CPU_BENDIAN)
115 +#endif // SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN)
117 ///////////////////////////////////////////////////////////////////////////////
119 /* If we replace a sampleproc, then we null-out the associated shaderproc,
120 otherwise the shader won't even look at the matrix/sampler
121 */
122 void SkBitmapProcState::platformProcs() {
123 bool doFilter = fDoFilter;
124 @@ -195,17 +195,17 @@ void SkBitmapProcState::platformProcs()
125 bool justDx = false;
127 if (fInvType <= (SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask)) {
128 justDx = true;
129 }
131 switch (fBitmap->config()) {
132 case SkBitmap::kIndex8_Config:
133 -#if __ARM_ARCH__ >= 6 && !defined(SK_CPU_BENDIAN)
134 +#if SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN)
135 if (justDx && !doFilter) {
136 #if 0 /* crashing on android device */
137 fSampleProc16 = SI8_D16_nofilter_DX_arm;
138 fShaderProc16 = NULL;
139 #endif
140 if (isOpaque) {
141 // this one is only very slighty faster than the C version
142 fSampleProc32 = SI8_opaque_D32_nofilter_DX_arm;
143 diff --git a/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp b/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp
144 --- a/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp
145 +++ b/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp
146 @@ -669,18 +669,23 @@ static void __attribute((noinline,optimi
147 /* Double Loop */
148 "1: \n\t" /* <double loop> */
149 "ldm %[src]!, {r5, r6} \n\t" /* loading src pointers into r5 and r6 */
150 "ldm %[dst], {r7, r8} \n\t" /* loading dst pointers into r7 and r8 */
152 /* dst1_scale and dst2_scale*/
153 "lsr r9, r5, #24 \n\t" /* src >> 24 */
154 "lsr r10, r6, #24 \n\t" /* src >> 24 */
155 +#ifdef SK_ARM_HAS_EDSP
156 "smulbb r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */
157 "smulbb r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */
158 +#else
159 + "mul r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */
160 + "mul r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */
161 +#endif
162 "lsr r9, r9, #8 \n\t" /* r9 >> 8 */
163 "lsr r10, r10, #8 \n\t" /* r10 >> 8 */
164 "rsb r9, r9, #256 \n\t" /* dst1_scale = r9 = 255 - r9 + 1 */
165 "rsb r10, r10, #256 \n\t" /* dst2_scale = r10 = 255 - r10 + 1 */
167 /* ---------------------- */
169 /* src1, src1_scale */
170 @@ -739,17 +744,21 @@ static void __attribute((noinline,optimi
171 /* else get into the single loop */
172 /* Single Loop */
173 "2: \n\t" /* <single loop> */
174 "ldr r5, [%[src]], #4 \n\t" /* loading src pointer into r5: r5=src */
175 "ldr r7, [%[dst]] \n\t" /* loading dst pointer into r7: r7=dst */
177 "lsr r6, r5, #24 \n\t" /* src >> 24 */
178 "and r8, r12, r5, lsr #8 \n\t" /* ag = r8 = r5 masked by r12 lsr by #8 */
179 +#ifdef SK_ARM_HAS_EDSP
180 "smulbb r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */
181 +#else
182 + "mul r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */
183 +#endif
184 "and r9, r12, r5 \n\t" /* rb = r9 = r5 masked by r12 */
185 "lsr r6, r6, #8 \n\t" /* r6 >> 8 */
186 "mul r8, r8, %[alpha] \n\t" /* ag = r8 times scale */
187 "rsb r6, r6, #256 \n\t" /* r6 = 255 - r6 + 1 */
189 /* src, src_scale */
190 "mul r9, r9, %[alpha] \n\t" /* rb = r9 times scale */
191 "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
