Add ARM optimizations to the build.

Also had to fix a problem in the ARM memset code that was
causing some tests and bench to fail.
Review URL: http://codereview.appspot.com/5522052

git-svn-id: http://skia.googlecode.com/svn/trunk@2989 2bbb7eff-a529-9590-31e7-b0007b416f81

5 files changed, 10 insertions, 20 deletions

diff --git a/gyp/core.gyp b/gyp/core.gyp
index 9e5c53635a..f83f482bbc 100644
--- a/gyp/core.gyp
+++ b/gyp/core.gyp
@@ -148,8 +148,6 @@
         '../src/core/SkWriter32.cpp',
         '../src/core/SkXfermode.cpp',
 
-        '../src/opts/opts_check_SSE2.cpp',
-
         '../include/core/Sk64.h',
         '../include/core/SkAdvancedTypefaceMetrics.h',
         '../include/core/SkAutoKern.h',
@@ -302,9 +300,6 @@
           ],
         }],
         [ 'skia_os == "android"', {
-          'sources!': [
-            '../src/opts/opts_check_SSE2.cpp',
-          ],
           'dependencies': [
              'android_system.gyp:ft2',
           ],
diff --git a/gyp/opts.gyp b/gyp/opts.gyp
index ae364816d8..1f67e006d2 100644
--- a/gyp/opts.gyp
+++ b/gyp/opts.gyp
@@ -28,6 +28,7 @@
         '../include/config',
         '../include/core',
         '../src/core',
+        '../src/opts',
       ],
       'conditions': [
         [ 'skia_os in ["linux", "freebsd", "openbsd", "solaris"]', {
@@ -37,6 +38,7 @@
         }],
         [ 'skia_target_arch != "arm"', {
           'sources': [
+            '../src/opts/opts_check_SSE2.cpp',
             '../src/opts/SkBitmapProcState_opts_SSE2.cpp',
             '../src/opts/SkBlitRow_opts_SSE2.cpp',
             '../src/opts/SkUtils_opts_SSE2.cpp',
@@ -52,9 +54,12 @@
             '-fomit-frame-pointer',
           ],
           'sources': [
+            '../src/opts/opts_check_arm.cpp',
+            '../src/opts/memset.arm.S',
+            '../src/opts/memset16_neon.S',
+            '../src/opts/memset32_neon.S',
             '../src/opts/SkBitmapProcState_opts_arm.cpp',
             '../src/opts/SkBlitRow_opts_arm.cpp',
-            '../src/opts/SkUtils_opts_none.cpp',
           ],
         }],
         [ 'skia_target_arch == "arm" and armv7 != 1', {
diff --git a/include/core/SkUtils.h b/include/core/SkUtils.h
index 37dd755b7d..b700b96276 100644
--- a/include/core/SkUtils.h
+++ b/include/core/SkUtils.h
@@ -32,13 +32,6 @@ void sk_memset32_portable(uint32_t dst[], uint32_t value, int count);
 typedef void (*SkMemset32Proc)(uint32_t dst[], uint32_t value, int count);
 SkMemset32Proc SkMemset32GetPlatformProc();
 
-#if defined(SK_BUILD_FOR_ANDROID) && !defined(SK_BUILD_FOR_ANDROID_NDK)
-    #include "cutils/memory.h"
-    
-    #define sk_memset16(dst, value, count)    android_memset16(dst, value, (count) << 1)
-    #define sk_memset32(dst, value, count)    android_memset32(dst, value, (count) << 2)
-#endif
-
 #ifndef sk_memset16
 extern SkMemset16Proc sk_memset16;
 #endif
diff --git a/src/core/SkUtils.cpp b/src/core/SkUtils.cpp
index 9161609de4..3f1c65e777 100644
--- a/src/core/SkUtils.cpp
+++ b/src/core/SkUtils.cpp
@@ -108,7 +108,6 @@ void sk_memset32_portable(uint32_t dst[], uint32_t value, int count) {
     }
 }
 
-#if !defined(SK_BUILD_FOR_ANDROID) || defined(SK_BUILD_FOR_ANDROID_NDK)
 static void sk_memset16_stub(uint16_t dst[], uint16_t value, int count) {
     SkMemset16Proc proc = SkMemset16GetPlatformProc();
     sk_memset16 = proc ? proc : sk_memset16_portable;
@@ -125,8 +124,6 @@ static void sk_memset32_stub(uint32_t dst[], uint32_t value, int count) {
 
 SkMemset32Proc sk_memset32 = sk_memset32_stub;
 
-#endif
-
 ///////////////////////////////////////////////////////////////////////////////
 
 /*  0xxxxxxx    1 total
diff --git a/src/opts/memset.arm.S b/src/opts/memset.arm.S
index 7732174881..bc0c060ac9 100644
--- a/src/opts/memset.arm.S
+++ b/src/opts/memset.arm.S
@@ -32,6 +32,10 @@ arm_memset16:
         .fnstart
         push        {lr}
 
+        /* Multiply count by 2 - go from the number of 16-bit shorts
+         * to the number of bytes desired. */
+        mov         r2, r2, lsl #1
+
         /* expand the data to 32 bits */
         orr         r1, r1, lsl #16
 
@@ -40,10 +44,6 @@ arm_memset16:
         strneh      r1, [r0], #2
         subne       r2, r2, #2
 
-        /* Multiply count by 2 - go from the number of 16-bit shorts
-         * to the number of bytes desired. */
-        mov         r2, r2, lsl #1
-
         /* Now jump into the main loop below. */
         b           .Lwork_32
         .fnend