FixedDiv using integers

BUG=250
TEST=fixed div unittest
R=dingkai@google.com, ryanpetrie@google.com

Review URL: https://webrtc-codereview.appspot.com/1681004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@732 16f28f9a-4ce2-e073-06de-1de4eb20be90

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 731
+Version: 732
 License: BSD
 License File: LICENSE
 

include/libyuv/row.h

@@ -130,6 +130,7 @@ 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
@@ -1516,16 +1517,13 @@ void SobelXYRow_SSE2(const uint8* src_sobelx, const uint8* src_sobely,
                      uint8* dst_argb, int width);
 void SobelXYRow_NEON(const uint8* src_sobelx, const uint8* src_sobely,
                      uint8* dst_argb, int width);
-extern const float kRecipTable[4097];
 
-// Divide num by div and return value as 16.16 fixed point.
-#ifdef __cplusplus
-static __inline int FixedDiv(int num, int div) {
-  if (static_cast<unsigned int>(div) <= 4096u) {
-    return static_cast<int>(num * kRecipTable[div]);
-  }
-  return static_cast<int>((static_cast<int64>(num) << 16) / div);
-}
+// Divide num by div and return as 16.16 fixed point result.
+int FixedDiv_C(int num, int div);
+#ifdef HAS_FIXEDDIV
+int FixedDiv(int num, int div);
+#else
+#define FixedDiv FixedDiv_C
 #endif
 
 #ifdef __cplusplus

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_  // NOLINT
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 731
+#define LIBYUV_VERSION 732
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_  NOLINT

source/row_win.cc

@@ -6602,6 +6602,30 @@ void I422ToUYVYRow_SSE2(const uint8* src_y,
     ret
   }
 }
+
+// Fixed point 0.32 reciprocal table.
+extern const uint32 kRecipTable[4097];
+
+// Divide num by div and return as 16.16 fixed point result.
+__declspec(naked) __declspec(align(16))
+int FixedDiv(int num, int div) {
+  __asm {
+    mov        eax, [esp + 4]    // num
+    mov        ecx, [esp + 8]    // div
+    cmp        ecx, 4096
+    ja         largediv
+    mul        dword ptr kRecipTable[ecx * 4]
+    shrd       eax, edx, 16
+    ret
+
+  largediv:
+    cwd                          // extend num to 64 bits
+    shld       edx, eax, 16
+    shl        eax, 16
+    idiv       ecx
+    ret
+  }
+}
 #endif  // !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
 
 #ifdef __cplusplus

unit_test/math_test.cc

@@ -12,16 +12,12 @@
 #include <string.h>
 
 #include "libyuv/basic_types.h"
+#include "libyuv/cpu_id.h"
 #include "libyuv/row.h"
 #include "../unit_test/unit_test.h"
 
 namespace libyuv {
 
-// Divide num by div and return value as 16.16 fixed point.
-static int FixedDiv_C(int num, int div) {
-  return static_cast<int>((static_cast<int64>(num) << 16) / div);
-}
-
 TEST_F(libyuvTest, TestFixedDiv) {
   int num[256];
   int div[256];
@@ -43,7 +39,7 @@ TEST_F(libyuvTest, TestFixedDiv) {
   }
   for (int j = 0; j < 256; ++j) {
     result_c[j] = libyuv::FixedDiv_C(num[j], div[j]);
-    EXPECT_NEAR(result_c[j], result_opt[j], 3);
+    EXPECT_NEAR(result_c[j], result_opt[j], 1);
   }
 }
 
@@ -64,8 +60,7 @@ TEST_F(libyuvTest, TestFixedDiv_Opt) {
   EXPECT_EQ(0x20000, libyuv::FixedDiv(960 * 2, 960));
   EXPECT_EQ(0x08000, libyuv::FixedDiv(640 / 2, 640));
   EXPECT_EQ(0x04000, libyuv::FixedDiv(640 / 4, 640));
-  // TODO(fbarchard): Improve accuracy for divides should be exact.
-  EXPECT_NEAR(0x20000, libyuv::FixedDiv(1080 * 2, 1080), 1);
+  EXPECT_EQ(0x20000, libyuv::FixedDiv(1080 * 2, 1080));
 
   srandom(time(NULL));
   MemRandomize(reinterpret_cast<uint8*>(&num[0]), sizeof(num));
@@ -77,14 +72,22 @@ TEST_F(libyuvTest, TestFixedDiv_Opt) {
       div[j] = 1280;
     }
   }
+
+  int has_x86 = TestCpuFlag(kCpuHasX86);
   for (int i = 0; i < benchmark_pixels_div256_; ++i) {
-    for (int j = 0; j < 256; ++j) {
-      result_opt[j] = libyuv::FixedDiv(num[j], div[j]);
+    if (has_x86) {
+      for (int j = 0; j < 256; ++j) {
+        result_opt[j] = libyuv::FixedDiv(num[j], div[j]);
+      }
+    } else {
+      for (int j = 0; j < 256; ++j) {
+        result_opt[j] = libyuv::FixedDiv_C(num[j], div[j]);
+      }
     }
   }
   for (int j = 0; j < 256; ++j) {
     result_c[j] = libyuv::FixedDiv_C(num[j], div[j]);
-    EXPECT_NEAR(result_c[j], result_opt[j], 3);
+    EXPECT_NEAR(result_c[j], result_opt[j], 1);
   }
 }