1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/gfx/skia/patches/archive/arm-fixes.patch Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,191 @@
1.4 +diff --git a/gfx/skia/include/core/SkMath.h b/gfx/skia/include/core/SkMath.h
1.5 +--- a/gfx/skia/include/core/SkMath.h
1.6 ++++ b/gfx/skia/include/core/SkMath.h
1.7 +@@ -148,20 +148,17 @@ static inline bool SkIsPow2(int value) {
1.8 + }
1.9 +
1.10 + ///////////////////////////////////////////////////////////////////////////////
1.11 +
1.12 + /** SkMulS16(a, b) multiplies a * b, but requires that a and b are both int16_t.
1.13 + With this requirement, we can generate faster instructions on some
1.14 + architectures.
1.15 + */
1.16 +-#if defined(__arm__) \
1.17 +- && !defined(__thumb__) \
1.18 +- && !defined(__ARM_ARCH_4T__) \
1.19 +- && !defined(__ARM_ARCH_5T__)
1.20 ++#ifdef SK_ARM_HAS_EDSP
1.21 + static inline int32_t SkMulS16(S16CPU x, S16CPU y) {
1.22 + SkASSERT((int16_t)x == x);
1.23 + SkASSERT((int16_t)y == y);
1.24 + int32_t product;
1.25 + asm("smulbb %0, %1, %2 \n"
1.26 + : "=r"(product)
1.27 + : "r"(x), "r"(y)
1.28 + );
1.29 +diff --git a/gfx/skia/include/core/SkPostConfig.h b/gfx/skia/include/core/SkPostConfig.h
1.30 +--- a/gfx/skia/include/core/SkPostConfig.h
1.31 ++++ b/gfx/skia/include/core/SkPostConfig.h
1.32 +@@ -300,8 +300,53 @@
1.33 + #endif
1.34 + #endif
1.35 +
1.36 + //////////////////////////////////////////////////////////////////////
1.37 +
1.38 + #ifndef SK_ALLOW_STATIC_GLOBAL_INITIALIZERS
1.39 + #define SK_ALLOW_STATIC_GLOBAL_INITIALIZERS 1
1.40 + #endif
1.41 ++
1.42 ++//////////////////////////////////////////////////////////////////////
1.43 ++// ARM defines
1.44 ++
1.45 ++#if defined(__GNUC__) && defined(__arm__)
1.46 ++
1.47 ++# define SK_ARM_ARCH 3
1.48 ++
1.49 ++# if defined(__ARM_ARCH_4__) || defined(__ARM_ARCH_4T__) \
1.50 ++ || defined(_ARM_ARCH_4)
1.51 ++# undef SK_ARM_ARCH
1.52 ++# define SK_ARM_ARCH 4
1.53 ++# endif
1.54 ++
1.55 ++# if defined(__ARM_ARCH_5__) || defined(__ARM_ARCH_5T__) \
1.56 ++ || defined(__ARM_ARCH_5E__) || defined(__ARM_ARCH_5TE__) \
1.57 ++ || defined(__ARM_ARCH_5TEJ__) || defined(_ARM_ARCH_5)
1.58 ++# undef SK_ARM_ARCH
1.59 ++# define SK_ARM_ARCH 5
1.60 ++# endif
1.61 ++
1.62 ++# if defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) \
1.63 ++ || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) \
1.64 ++ || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) \
1.65 ++ || defined(__ARM_ARCH_6M__) || defined(_ARM_ARCH_6)
1.66 ++# undef SK_ARM_ARCH
1.67 ++# define SK_ARM_ARCH 6
1.68 ++# endif
1.69 ++
1.70 ++# if defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) \
1.71 ++ || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) \
1.72 ++ || defined(__ARM_ARCH_7EM__) || defined(_ARM_ARCH_7)
1.73 ++# undef SK_ARM_ARCH
1.74 ++# define SK_ARM_ARCH 7
1.75 ++# endif
1.76 ++
1.77 ++# undef SK_ARM_HAS_EDSP
1.78 ++# if defined(__thumb2__) && (SK_ARM_ARCH >= 6) \
1.79 ++ || !defined(__thumb__) \
1.80 ++ && ((SK_ARM_ARCH > 5) || defined(__ARM_ARCH_5E__) \
1.81 ++ || defined(__ARM_ARCH_5TE__) || defined(__ARM_ARCH_5TEJ__))
1.82 ++# define SK_ARM_HAS_EDSP 1
1.83 ++# endif
1.84 ++
1.85 ++#endif
1.86 +diff --git a/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp b/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp
1.87 +--- a/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp
1.88 ++++ b/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp
1.89 +@@ -6,17 +6,17 @@
1.90 + * found in the LICENSE file.
1.91 + */
1.92 +
1.93 +
1.94 + #include "SkBitmapProcState.h"
1.95 + #include "SkColorPriv.h"
1.96 + #include "SkUtils.h"
1.97 +
1.98 +-#if __ARM_ARCH__ >= 6 && !defined(SK_CPU_BENDIAN)
1.99 ++#if SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN)
1.100 + void SI8_D16_nofilter_DX_arm(
1.101 + const SkBitmapProcState& s,
1.102 + const uint32_t* SK_RESTRICT xy,
1.103 + int count,
1.104 + uint16_t* SK_RESTRICT colors) __attribute__((optimize("O1")));
1.105 +
1.106 + void SI8_D16_nofilter_DX_arm(const SkBitmapProcState& s,
1.107 + const uint32_t* SK_RESTRICT xy,
1.108 +@@ -177,17 +177,17 @@ void SI8_opaque_D32_nofilter_DX_arm(cons
1.109 + : [xx] "+r" (xx), [count] "+r" (count), [colors] "+r" (colors)
1.110 + : [table] "r" (table), [srcAddr] "r" (srcAddr)
1.111 + : "memory", "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11"
1.112 + );
1.113 + }
1.114 +
1.115 + s.fBitmap->getColorTable()->unlockColors(false);
1.116 + }
1.117 +-#endif //__ARM_ARCH__ >= 6 && !defined(SK_CPU_BENDIAN)
1.118 ++#endif // SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN)
1.119 +
1.120 + ///////////////////////////////////////////////////////////////////////////////
1.121 +
1.122 + /* If we replace a sampleproc, then we null-out the associated shaderproc,
1.123 + otherwise the shader won't even look at the matrix/sampler
1.124 + */
1.125 + void SkBitmapProcState::platformProcs() {
1.126 + bool doFilter = fDoFilter;
1.127 +@@ -195,17 +195,17 @@ void SkBitmapProcState::platformProcs()
1.128 + bool justDx = false;
1.129 +
1.130 + if (fInvType <= (SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask)) {
1.131 + justDx = true;
1.132 + }
1.133 +
1.134 + switch (fBitmap->config()) {
1.135 + case SkBitmap::kIndex8_Config:
1.136 +-#if __ARM_ARCH__ >= 6 && !defined(SK_CPU_BENDIAN)
1.137 ++#if SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN)
1.138 + if (justDx && !doFilter) {
1.139 + #if 0 /* crashing on android device */
1.140 + fSampleProc16 = SI8_D16_nofilter_DX_arm;
1.141 + fShaderProc16 = NULL;
1.142 + #endif
1.143 + if (isOpaque) {
1.144 + // this one is only very slighty faster than the C version
1.145 + fSampleProc32 = SI8_opaque_D32_nofilter_DX_arm;
1.146 +diff --git a/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp b/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp
1.147 +--- a/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp
1.148 ++++ b/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp
1.149 +@@ -669,18 +669,23 @@ static void __attribute((noinline,optimi
1.150 + /* Double Loop */
1.151 + "1: \n\t" /* <double loop> */
1.152 + "ldm %[src]!, {r5, r6} \n\t" /* loading src pointers into r5 and r6 */
1.153 + "ldm %[dst], {r7, r8} \n\t" /* loading dst pointers into r7 and r8 */
1.154 +
1.155 + /* dst1_scale and dst2_scale*/
1.156 + "lsr r9, r5, #24 \n\t" /* src >> 24 */
1.157 + "lsr r10, r6, #24 \n\t" /* src >> 24 */
1.158 ++#ifdef SK_ARM_HAS_EDSP
1.159 + "smulbb r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */
1.160 + "smulbb r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */
1.161 ++#else
1.162 ++ "mul r9, r9, %[alpha] \n\t" /* r9 = SkMulS16 r9 with src_scale */
1.163 ++ "mul r10, r10, %[alpha] \n\t" /* r10 = SkMulS16 r10 with src_scale */
1.164 ++#endif
1.165 + "lsr r9, r9, #8 \n\t" /* r9 >> 8 */
1.166 + "lsr r10, r10, #8 \n\t" /* r10 >> 8 */
1.167 + "rsb r9, r9, #256 \n\t" /* dst1_scale = r9 = 255 - r9 + 1 */
1.168 + "rsb r10, r10, #256 \n\t" /* dst2_scale = r10 = 255 - r10 + 1 */
1.169 +
1.170 + /* ---------------------- */
1.171 +
1.172 + /* src1, src1_scale */
1.173 +@@ -739,17 +744,21 @@ static void __attribute((noinline,optimi
1.174 + /* else get into the single loop */
1.175 + /* Single Loop */
1.176 + "2: \n\t" /* <single loop> */
1.177 + "ldr r5, [%[src]], #4 \n\t" /* loading src pointer into r5: r5=src */
1.178 + "ldr r7, [%[dst]] \n\t" /* loading dst pointer into r7: r7=dst */
1.179 +
1.180 + "lsr r6, r5, #24 \n\t" /* src >> 24 */
1.181 + "and r8, r12, r5, lsr #8 \n\t" /* ag = r8 = r5 masked by r12 lsr by #8 */
1.182 ++#ifdef SK_ARM_HAS_EDSP
1.183 + "smulbb r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */
1.184 ++#else
1.185 ++ "mul r6, r6, %[alpha] \n\t" /* r6 = SkMulS16 with src_scale */
1.186 ++#endif
1.187 + "and r9, r12, r5 \n\t" /* rb = r9 = r5 masked by r12 */
1.188 + "lsr r6, r6, #8 \n\t" /* r6 >> 8 */
1.189 + "mul r8, r8, %[alpha] \n\t" /* ag = r8 times scale */
1.190 + "rsb r6, r6, #256 \n\t" /* r6 = 255 - r6 + 1 */
1.191 +
1.192 + /* src, src_scale */
1.193 + "mul r9, r9, %[alpha] \n\t" /* rb = r9 times scale */
1.194 + "and r8, r8, r12, lsl #8 \n\t" /* ag masked by reverse mask (r12) */
