FixedDiv port to gcc

BUG=250
TEST=*Div*
R=ryanpetrie@google.com

Review URL: https://webrtc-codereview.appspot.com/1682005

git-svn-id: http://libyuv.googlecode.com/svn/trunk@733 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 732
+Version: 733
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -101,6 +101,7 @@ extern "C" {
 #define HAS_YUY2TOUV422ROW_SSE2
 #define HAS_YUY2TOUVROW_SSE2
 #define HAS_YUY2TOYROW_SSE2
+#define HAS_FIXEDDIV
 
 // Effects
 #define HAS_ARGBADDROW_SSE2
@@ -130,7 +131,6 @@ extern "C" {
 // TODO(fbarchard): Port to gcc.
 #if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
 #define HAS_ARGBCOLORTABLEROW_X86
-#define HAS_FIXEDDIV
 // Visual C 2012 required for AVX2.
 #if _MSC_VER >= 1700
 #define HAS_ARGBSHUFFLEROW_AVX2

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 732
+#define LIBYUV_VERSION 733
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_posix.cc

@@ -5401,6 +5401,38 @@ void I422ToUYVYRow_SSE2(const uint8* src_y,
   );
 }
 
+// Fixed point 0.32 reciprocal table.
+extern const uint32 kRecipTable[4097];
+
+// Divide num by div and return as 16.16 fixed point result.
+int FixedDiv(int num, int div) {
+  asm volatile (
+    "cmp       $0x1000,%%ecx                   \n"
+    "jbe       1f                              \n"
+
+    "cdq                                       \n"
+    "shld      $0x10,%%eax,%%edx               \n"
+    "shl       $0x10,%%eax                     \n"
+    "idiv      %%ecx                           \n"
+    "jmp       9f                              \n"
+
+  "1:                                          \n"
+#if defined(__x86_64__)
+    "mull      (%2,%%rcx,4)                    \n"
+#else
+    "mull      (%2,%%ecx,4)                    \n"
+#endif
+    "shrd      $0x10,%%edx,%%eax               \n"
+
+  "9:                                          \n"
+    "mov       %0, %%eax                       \n"
+    : "+a"(num)  // %0
+    : "c"(div),  // %1
+      "r"(kRecipTable)  // %2
+    : "memory", "cc", "edx"
+  );
+  return num;
+}
 #endif  // defined(__x86_64__) || defined(__i386__)
 
 #ifdef __cplusplus

source/row_win.cc

@@ -6619,7 +6619,7 @@ int FixedDiv(int num, int div) {
     ret
 
   largediv:
-    cwd                          // extend num to 64 bits
+    cdq                          // extend num to 64 bits
     shld       edx, eax, 16
     shl        eax, 16
     idiv       ecx

unit_test/compare_test.cc

@@ -39,7 +39,7 @@ TEST_F(libyuvTest, Djb2_Test) {
       " and feels as if he were in the seventh heaven of typography"
       " together with Hermann Zapf";
   uint32 foxhash = HashDjb2(reinterpret_cast<const uint8*>(fox), 131, 5381);
-  const uint32 kExpectedFoxHash = 2611006483;
+  const uint32 kExpectedFoxHash = 2611006483u;
   EXPECT_EQ(kExpectedFoxHash, foxhash);
 
   for (int i = 0; i < kMaxTest; ++i) {

unit_test/math_test.cc

@@ -24,6 +24,29 @@ TEST_F(libyuvTest, TestFixedDiv) {
   int result_opt[256];
   int result_c[256];
 
+  EXPECT_EQ(0x20000, libyuv::FixedDiv(640 * 2, 640));
+  EXPECT_EQ(0x30000, libyuv::FixedDiv(640 * 3, 640));
+  EXPECT_EQ(0x40000, libyuv::FixedDiv(640 * 4, 640));
+  EXPECT_EQ(0x50000, libyuv::FixedDiv(640 * 5, 640));
+  EXPECT_EQ(0x60000, libyuv::FixedDiv(640 * 6, 640));
+  EXPECT_EQ(0x70000, libyuv::FixedDiv(640 * 7, 640));
+  EXPECT_EQ(0x80000, libyuv::FixedDiv(640 * 8, 640));
+  EXPECT_EQ(0xa0000, libyuv::FixedDiv(640 * 10, 640));
+  EXPECT_EQ(0x20000, libyuv::FixedDiv(960 * 2, 960));
+  EXPECT_EQ(0x08000, libyuv::FixedDiv(640 / 2, 640));
+  EXPECT_EQ(0x04000, libyuv::FixedDiv(640 / 4, 640));
+  EXPECT_EQ(0x20000, libyuv::FixedDiv(1080 * 2, 1080));
+  EXPECT_EQ(0x20000, libyuv::FixedDiv(200000, 100000));
+  EXPECT_EQ(0x18000, libyuv::FixedDiv(150000, 100000));
+  EXPECT_EQ(0x20000, libyuv::FixedDiv(40000, 20000));
+  EXPECT_EQ(0x20000, libyuv::FixedDiv(-40000, -20000));
+  EXPECT_EQ(-0x20000, libyuv::FixedDiv(40000, -20000));
+  EXPECT_EQ(-0x20000, libyuv::FixedDiv(-40000, 20000));
+  EXPECT_NEAR(0x10000, libyuv::FixedDiv(4095, 4095), 1);
+  EXPECT_EQ(0x10000, libyuv::FixedDiv(4096, 4096));
+  EXPECT_EQ(0x10000, libyuv::FixedDiv(4097, 4097));
+  EXPECT_NEAR(123 * 65536, libyuv::FixedDiv(123, 1), 1);
+
   srandom(time(NULL));
   MemRandomize(reinterpret_cast<uint8*>(&num[0]), sizeof(num));
   MemRandomize(reinterpret_cast<uint8*>(&div[0]), sizeof(div));
@@ -49,19 +72,6 @@ TEST_F(libyuvTest, TestFixedDiv_Opt) {
   int result_opt[256];
   int result_c[256];
 
-  EXPECT_EQ(0x20000, libyuv::FixedDiv(640 * 2, 640));
-  EXPECT_EQ(0x30000, libyuv::FixedDiv(640 * 3, 640));
-  EXPECT_EQ(0x40000, libyuv::FixedDiv(640 * 4, 640));
-  EXPECT_EQ(0x50000, libyuv::FixedDiv(640 * 5, 640));
-  EXPECT_EQ(0x60000, libyuv::FixedDiv(640 * 6, 640));
-  EXPECT_EQ(0x70000, libyuv::FixedDiv(640 * 7, 640));
-  EXPECT_EQ(0x80000, libyuv::FixedDiv(640 * 8, 640));
-  EXPECT_EQ(0xa0000, libyuv::FixedDiv(640 * 10, 640));
-  EXPECT_EQ(0x20000, libyuv::FixedDiv(960 * 2, 960));
-  EXPECT_EQ(0x08000, libyuv::FixedDiv(640 / 2, 640));
-  EXPECT_EQ(0x04000, libyuv::FixedDiv(640 / 4, 640));
-  EXPECT_EQ(0x20000, libyuv::FixedDiv(1080 * 2, 1080));
-
   srandom(time(NULL));
   MemRandomize(reinterpret_cast<uint8*>(&num[0]), sizeof(num));
   MemRandomize(reinterpret_cast<uint8*>(&div[0]), sizeof(div));