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