SSE2 implementation of memcpy32

src/core/SkUtils.cpp

 
 /*
  * Copyright 2006 The Android Open Source Project
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 
 #include "SkUtils.h"
@@ skipping 16 matching lines @@
     do {                                        \
         *(dst)++ = value;   *(dst)++ = value;   \
         *(dst)++ = value;   *(dst)++ = value;   \
         *(dst)++ = value;   *(dst)++ = value;   \
         *(dst)++ = value;   *(dst)++ = value;   \
         *(dst)++ = value;   *(dst)++ = value;   \
         *(dst)++ = value;   *(dst)++ = value;   \
         *(dst)++ = value;   *(dst)++ = value;   \
         *(dst)++ = value;   *(dst)++ = value;   \
     } while (0)
+
+#define copy_16_longs(dst, src)                       \
+    do {                                              \
+        *(dst)++ = *(src)++;   *(dst)++ = *(src)++;   \
+        *(dst)++ = *(src)++;   *(dst)++ = *(src)++;   \
+        *(dst)++ = *(src)++;   *(dst)++ = *(src)++;   \
+        *(dst)++ = *(src)++;   *(dst)++ = *(src)++;   \
+        *(dst)++ = *(src)++;   *(dst)++ = *(src)++;   \
+        *(dst)++ = *(src)++;   *(dst)++ = *(src)++;   \
+        *(dst)++ = *(src)++;   *(dst)++ = *(src)++;   \
+        *(dst)++ = *(src)++;   *(dst)++ = *(src)++;   \
+    } while (0)
 #endif
 
 ///////////////////////////////////////////////////////////////////////////////
 
 static void sk_memset16_portable(uint16_t dst[], uint16_t value, int count) {
     SkASSERT(dst != NULL && count >= 0);
 
     if (count <= 0) {
         return;
     }
@@ skipping 55 matching lines @@
         count &= 15;
     }
 
     if (count) {
         do {
             *dst++ = value;
         } while (--count != 0);
     }
 }
 
+static void sk_memcpy32_portable(uint32_t dst[], const uint32_t src[], int count) {
+    SkASSERT(dst != NULL && count >= 0);
+
+    int sixteenlongs = count >> 4;
+    if (sixteenlongs) {
+        do {
+            copy_16_longs(dst, src);
+        } while (--sixteenlongs != 0);
+        count &= 15;
+    }
+
+    if (count) {
+        do {
+            *dst++ = *src++;
+        } while (--count != 0);
+    }
+}
+
 static void choose_memset16(SkMemset16Proc* proc) {
     *proc = SkMemset16GetPlatformProc();
     if (NULL == *proc) {
         *proc = &sk_memset16_portable;
     }
 }
 
 void sk_memset16(uint16_t dst[], uint16_t value, int count) {
     SK_DECLARE_STATIC_ONCE(once);
     static SkMemset16Proc proc = NULL;
@@ skipping 12 matching lines @@
 
 void sk_memset32(uint32_t dst[], uint32_t value, int count) {
     SK_DECLARE_STATIC_ONCE(once);
     static SkMemset32Proc proc = NULL;
     SkOnce(&once, choose_memset32, &proc);
     SkASSERT(proc != NULL);
 
     return proc(dst, value, count);
 }
 
+static void choose_memcpy32(SkMemcpy32Proc* proc) {
+    *proc = SkMemcpy32GetPlatformProc();
+    if (NULL == *proc) {
+        *proc = &sk_memcpy32_portable;
+    }
+}
+
+void sk_memcpy32(uint32_t dst[], const uint32_t src[], int count) {
+    SK_DECLARE_STATIC_ONCE(once);
+    static SkMemcpy32Proc proc = NULL;
+    SkOnce(&once, choose_memcpy32, &proc);
+    SkASSERT(proc != NULL);
+
+    return proc(dst, src, count);
+}
+
 ///////////////////////////////////////////////////////////////////////////////
 
 /*  0xxxxxxx    1 total
     10xxxxxx    // never a leading byte
     110xxxxx    2 total
     1110xxxx    3 total
     11110xxx    4 total
 
     11 10 01 01 xx xx xx xx 0...
     0xE5XX0000
@@ skipping 251 matching lines @@
         }
     } else {
         char* start = utf8;
         while (utf16 < stop) {
             utf8 += SkUTF8_FromUnichar(SkUTF16_NextUnichar(&utf16), utf8);
         }
         size = utf8 - start;
     }
     return size;
 }