From 3017fc6fb7f4e098533d3f377b2bb30ccaf44695 Mon Sep 17 00:00:00 2001
From: Victor Zverovich <victor.zverovich@gmail.com>
Date: Mon, 17 Feb 2014 09:12:24 -0800
Subject: [PATCH] Improve 32-bit integer formatting performance by
 parameterizing FormatDecimal on integer type.

---
 format.cc | 29 -----------------------------
 format.h  | 24 ++++++++++++++++++++++--
 2 files changed, 22 insertions(+), 31 deletions(-)

diff --git a/format.cc b/format.cc
index 869046b0..c5473b59 100644
--- a/format.cc
+++ b/format.cc
@@ -169,29 +169,6 @@ typename fmt::BasicWriter<Char>::CharPtr
   return content;
 }
 
-template <typename Char>
-void fmt::internal::FormatDecimal(
-    Char *buffer, uint64_t value, unsigned num_digits) {
-  --num_digits;
-  while (value >= 100) {
-    // Integer division is slow so do it for a group of two digits instead
-    // of for every digit. The idea comes from the talk by Alexandrescu
-    // "Three Optimization Tips for C++". See speed-test for a comparison.
-    unsigned index = (value % 100) * 2;
-    value /= 100;
-    buffer[num_digits] = internal::DIGITS[index + 1];
-    buffer[num_digits - 1] = internal::DIGITS[index];
-    num_digits -= 2;
-  }
-  if (value < 10) {
-    *buffer = static_cast<char>('0' + value);
-    return;
-  }
-  unsigned index = static_cast<unsigned>(value * 2);
-  buffer[1] = internal::DIGITS[index + 1];
-  buffer[0] = internal::DIGITS[index];
-}
-
 template <typename Char>
 typename fmt::BasicWriter<Char>::CharPtr
   fmt::BasicWriter<Char>::PrepareFilledBuffer(
@@ -700,9 +677,6 @@ template fmt::BasicWriter<char>::CharPtr
   fmt::BasicWriter<char>::FillPadding(CharPtr buffer,
     unsigned total_size, std::size_t content_size, wchar_t fill);
 
-template void fmt::internal::FormatDecimal<char>(
-    char *buffer, uint64_t value, unsigned num_digits);
-
 template fmt::BasicWriter<char>::CharPtr
   fmt::BasicWriter<char>::PrepareFilledBuffer(
     unsigned size, const AlignSpec &spec, char sign);
@@ -732,9 +706,6 @@ template fmt::BasicWriter<wchar_t>::CharPtr
   fmt::BasicWriter<wchar_t>::FillPadding(CharPtr buffer,
       unsigned total_size, std::size_t content_size, wchar_t fill);
 
-template void fmt::internal::FormatDecimal<wchar_t>(
-    wchar_t *buffer, uint64_t value, unsigned num_digits);
-
 template fmt::BasicWriter<wchar_t>::CharPtr
   fmt::BasicWriter<wchar_t>::PrepareFilledBuffer(
     unsigned size, const AlignSpec &spec, char sign);
diff --git a/format.h b/format.h
index d8241796..b24b2ad8 100644
--- a/format.h
+++ b/format.h
@@ -276,8 +276,28 @@ extern const char DIGITS[];
 template <typename Char>
 class FormatterProxy;
 
-template <typename Char>
-void FormatDecimal(Char *buffer, uint64_t value, unsigned num_digits);
+// Formats a decimal unsigned integer value writing into buffer.
+template <typename UInt, typename Char>
+void FormatDecimal(Char *buffer, UInt value, unsigned num_digits) {
+    --num_digits;
+  while (value >= 100) {
+    // Integer division is slow so do it for a group of two digits instead
+    // of for every digit. The idea comes from the talk by Alexandrescu
+    // "Three Optimization Tips for C++". See speed-test for a comparison.
+    unsigned index = (value % 100) * 2;
+    value /= 100;
+    buffer[num_digits] = internal::DIGITS[index + 1];
+    buffer[num_digits - 1] = internal::DIGITS[index];
+    num_digits -= 2;
+  }
+  if (value < 10) {
+    *buffer = static_cast<char>('0' + value);
+    return;
+  }
+  unsigned index = static_cast<unsigned>(value * 2);
+  buffer[1] = internal::DIGITS[index + 1];
+  buffer[0] = internal::DIGITS[index];
+}
 }
 
 /**