michael@0: diff --git a/gfx/skia/include/core/SkMath.h b/gfx/skia/include/core/SkMath.h
michael@0: --- a/gfx/skia/include/core/SkMath.h
michael@0: +++ b/gfx/skia/include/core/SkMath.h
michael@0: @@ -148,20 +148,17 @@ static inline bool SkIsPow2(int value) {
michael@0:  }
michael@0:  
michael@0:  ///////////////////////////////////////////////////////////////////////////////
michael@0:  
michael@0:  /** SkMulS16(a, b) multiplies a * b, but requires that a and b are both int16_t.
michael@0:      With this requirement, we can generate faster instructions on some
michael@0:      architectures.
michael@0:  */
michael@0: -#if defined(__arm__) \
michael@0: -  && !defined(__thumb__) \
michael@0: -  && !defined(__ARM_ARCH_4T__) \
michael@0: -  && !defined(__ARM_ARCH_5T__)
michael@0: +#ifdef SK_ARM_HAS_EDSP
michael@0:      static inline int32_t SkMulS16(S16CPU x, S16CPU y) {
michael@0:          SkASSERT((int16_t)x == x);
michael@0:          SkASSERT((int16_t)y == y);
michael@0:          int32_t product;
michael@0:          asm("smulbb %0, %1, %2 \n"
michael@0:              : "=r"(product)
michael@0:              : "r"(x), "r"(y)
michael@0:              );
michael@0: diff --git a/gfx/skia/include/core/SkPostConfig.h b/gfx/skia/include/core/SkPostConfig.h
michael@0: --- a/gfx/skia/include/core/SkPostConfig.h
michael@0: +++ b/gfx/skia/include/core/SkPostConfig.h
michael@0: @@ -300,8 +300,53 @@
michael@0:  #endif
michael@0:  #endif
michael@0:  
michael@0:  //////////////////////////////////////////////////////////////////////
michael@0:  
michael@0:  #ifndef SK_ALLOW_STATIC_GLOBAL_INITIALIZERS
michael@0:  #define SK_ALLOW_STATIC_GLOBAL_INITIALIZERS 1
michael@0:  #endif
michael@0: +
michael@0: +//////////////////////////////////////////////////////////////////////
michael@0: +// ARM defines
michael@0: +
michael@0: +#if defined(__GNUC__) && defined(__arm__)
michael@0: +
michael@0: +#  define SK_ARM_ARCH 3
michael@0: +
michael@0: +#  if defined(__ARM_ARCH_4__) || defined(__ARM_ARCH_4T__) \
michael@0: +   || defined(_ARM_ARCH_4)
michael@0: +#    undef SK_ARM_ARCH
michael@0: +#    define SK_ARM_ARCH 4
michael@0: +#  endif
michael@0: +
michael@0: +#  if defined(__ARM_ARCH_5__) || defined(__ARM_ARCH_5T__) \
michael@0: +   || defined(__ARM_ARCH_5E__) || defined(__ARM_ARCH_5TE__) \
michael@0: +   || defined(__ARM_ARCH_5TEJ__) || defined(_ARM_ARCH_5)
michael@0: +#    undef SK_ARM_ARCH
michael@0: +#    define SK_ARM_ARCH 5
michael@0: +#  endif
michael@0: + 
michael@0: +#  if defined(__ARM_ARCH_6__) || defined(__ARM_ARCH_6J__) \
michael@0: +   || defined(__ARM_ARCH_6K__) || defined(__ARM_ARCH_6Z__) \
michael@0: +   || defined(__ARM_ARCH_6ZK__) || defined(__ARM_ARCH_6T2__) \
michael@0: +   || defined(__ARM_ARCH_6M__) || defined(_ARM_ARCH_6)
michael@0: +#    undef SK_ARM_ARCH
michael@0: +#    define SK_ARM_ARCH 6
michael@0: +#  endif
michael@0: +
michael@0: +#  if defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) \
michael@0: +   || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) \
michael@0: +   || defined(__ARM_ARCH_7EM__) || defined(_ARM_ARCH_7)
michael@0: +#    undef SK_ARM_ARCH
michael@0: +#    define SK_ARM_ARCH 7
michael@0: +#  endif
michael@0: +
michael@0: +#  undef SK_ARM_HAS_EDSP
michael@0: +#  if defined(__thumb2__) && (SK_ARM_ARCH >= 6) \
michael@0: +   || !defined(__thumb__) \
michael@0: +   && ((SK_ARM_ARCH > 5) || defined(__ARM_ARCH_5E__) \
michael@0: +       || defined(__ARM_ARCH_5TE__) || defined(__ARM_ARCH_5TEJ__))
michael@0: +#    define SK_ARM_HAS_EDSP 1
michael@0: +#  endif
michael@0: +
michael@0: +#endif
michael@0: diff --git a/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp b/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp
michael@0: --- a/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp
michael@0: +++ b/gfx/skia/src/opts/SkBitmapProcState_opts_arm.cpp
michael@0: @@ -6,17 +6,17 @@
michael@0:   * found in the LICENSE file.
michael@0:   */
michael@0:  
michael@0:  
michael@0:  #include "SkBitmapProcState.h"
michael@0:  #include "SkColorPriv.h"
michael@0:  #include "SkUtils.h"
michael@0:  
michael@0: -#if __ARM_ARCH__ >= 6 && !defined(SK_CPU_BENDIAN)
michael@0: +#if SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN)
michael@0:  void SI8_D16_nofilter_DX_arm(
michael@0:      const SkBitmapProcState& s,
michael@0:      const uint32_t* SK_RESTRICT xy,
michael@0:      int count,
michael@0:      uint16_t* SK_RESTRICT colors) __attribute__((optimize("O1")));
michael@0:  
michael@0:  void SI8_D16_nofilter_DX_arm(const SkBitmapProcState& s,
michael@0:                               const uint32_t* SK_RESTRICT xy,
michael@0: @@ -177,17 +177,17 @@ void SI8_opaque_D32_nofilter_DX_arm(cons
michael@0:                        : [xx] "+r" (xx), [count] "+r" (count), [colors] "+r" (colors)
michael@0:                        : [table] "r" (table), [srcAddr] "r" (srcAddr)
michael@0:                        : "memory", "cc", "r4", "r5", "r6", "r7", "r8", "r9", "r10", "r11"
michael@0:                        );
michael@0:      }
michael@0:  
michael@0:      s.fBitmap->getColorTable()->unlockColors(false);
michael@0:  }
michael@0: -#endif //__ARM_ARCH__ >= 6 && !defined(SK_CPU_BENDIAN)
michael@0: +#endif // SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN)
michael@0:  
michael@0:  ///////////////////////////////////////////////////////////////////////////////
michael@0:  
michael@0:  /*  If we replace a sampleproc, then we null-out the associated shaderproc,
michael@0:      otherwise the shader won't even look at the matrix/sampler
michael@0:   */
michael@0:  void SkBitmapProcState::platformProcs() {
michael@0:      bool doFilter = fDoFilter;
michael@0: @@ -195,17 +195,17 @@ void SkBitmapProcState::platformProcs() 
michael@0:      bool justDx = false;
michael@0:  
michael@0:      if (fInvType <= (SkMatrix::kTranslate_Mask | SkMatrix::kScale_Mask)) {
michael@0:          justDx = true;
michael@0:      }
michael@0:  
michael@0:      switch (fBitmap->config()) {
michael@0:          case SkBitmap::kIndex8_Config:
michael@0: -#if __ARM_ARCH__ >= 6 && !defined(SK_CPU_BENDIAN)
michael@0: +#if SK_ARM_ARCH >= 6 && !defined(SK_CPU_BENDIAN)
michael@0:              if (justDx && !doFilter) {
michael@0:  #if 0   /* crashing on android device */
michael@0:                  fSampleProc16 = SI8_D16_nofilter_DX_arm;
michael@0:                  fShaderProc16 = NULL;
michael@0:  #endif
michael@0:                  if (isOpaque) {
michael@0:                      // this one is only very slighty faster than the C version
michael@0:                      fSampleProc32 = SI8_opaque_D32_nofilter_DX_arm;
michael@0: diff --git a/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp b/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp
michael@0: --- a/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp
michael@0: +++ b/gfx/skia/src/opts/SkBlitRow_opts_arm.cpp
michael@0: @@ -669,18 +669,23 @@ static void __attribute((noinline,optimi
michael@0:                    /* Double Loop */
michael@0:                    "1:                                \n\t" /* <double loop> */
michael@0:                    "ldm    %[src]!, {r5, r6}          \n\t" /* loading src pointers into r5 and r6 */
michael@0:                    "ldm    %[dst], {r7, r8}           \n\t" /* loading dst pointers into r7 and r8 */
michael@0:  
michael@0:                    /* dst1_scale and dst2_scale*/
michael@0:                    "lsr    r9, r5, #24                \n\t" /* src >> 24 */
michael@0:                    "lsr    r10, r6, #24               \n\t" /* src >> 24 */
michael@0: +#ifdef SK_ARM_HAS_EDSP
michael@0:                    "smulbb r9, r9, %[alpha]           \n\t" /* r9 = SkMulS16 r9 with src_scale */
michael@0:                    "smulbb r10, r10, %[alpha]         \n\t" /* r10 = SkMulS16 r10 with src_scale */
michael@0: +#else
michael@0: +                  "mul    r9, r9, %[alpha]           \n\t" /* r9 = SkMulS16 r9 with src_scale */
michael@0: +                  "mul    r10, r10, %[alpha]         \n\t" /* r10 = SkMulS16 r10 with src_scale */
michael@0: +#endif
michael@0:                    "lsr    r9, r9, #8                 \n\t" /* r9 >> 8 */
michael@0:                    "lsr    r10, r10, #8               \n\t" /* r10 >> 8 */
michael@0:                    "rsb    r9, r9, #256               \n\t" /* dst1_scale = r9 = 255 - r9 + 1 */
michael@0:                    "rsb    r10, r10, #256             \n\t" /* dst2_scale = r10 = 255 - r10 + 1 */
michael@0:  
michael@0:                    /* ---------------------- */
michael@0:  
michael@0:                    /* src1, src1_scale */
michael@0: @@ -739,17 +744,21 @@ static void __attribute((noinline,optimi
michael@0:                                                             /* else get into the single loop */
michael@0:                    /* Single Loop */
michael@0:                    "2:                                \n\t" /* <single loop> */
michael@0:                    "ldr    r5, [%[src]], #4           \n\t" /* loading src pointer into r5: r5=src */
michael@0:                    "ldr    r7, [%[dst]]               \n\t" /* loading dst pointer into r7: r7=dst */
michael@0:  
michael@0:                    "lsr    r6, r5, #24                \n\t" /* src >> 24 */
michael@0:                    "and    r8, r12, r5, lsr #8        \n\t" /* ag = r8 = r5 masked by r12 lsr by #8 */
michael@0: +#ifdef SK_ARM_HAS_EDSP
michael@0:                    "smulbb r6, r6, %[alpha]           \n\t" /* r6 = SkMulS16 with src_scale */
michael@0: +#else
michael@0: +                  "mul    r6, r6, %[alpha]           \n\t" /* r6 = SkMulS16 with src_scale */
michael@0: +#endif
michael@0:                    "and    r9, r12, r5                \n\t" /* rb = r9 = r5 masked by r12 */
michael@0:                    "lsr    r6, r6, #8                 \n\t" /* r6 >> 8 */
michael@0:                    "mul    r8, r8, %[alpha]           \n\t" /* ag = r8 times scale */
michael@0:                    "rsb    r6, r6, #256               \n\t" /* r6 = 255 - r6 + 1 */
michael@0:  
michael@0:                    /* src, src_scale */
michael@0:                    "mul    r9, r9, %[alpha]           \n\t" /* rb = r9 times scale */
michael@0:                    "and    r8, r8, r12, lsl #8        \n\t" /* ag masked by reverse mask (r12) */