Remove OpenCL device test

Fix access violation in some certain conditions

README.md

@@ -1,64 +1,52 @@
 # MultiPar
 
-### v1.3.3.1 is public
+### v1.3.3.5 is public
 
-  This is a testing version to improve speed of PAR2 calculation. 
-Because the new method isn't tested so much, there may be a bug, failure, or mistake. 
-Be careful to use this non-stable version. 
-When you don't want to test by yourself, you should not use this yet. 
-If you see a problem, please report the incident. 
-I will try to solve as possible as I can.
+  This is a minor update version. 
+If there is no serious problem in this version, 
+next version will be the last of v1.3.3 tree.
 
-  CPU's L3 cache optimization depends on hardware environment. 
-It's difficult to guess the best setting for unknown type. 
-It seems to work well on Intel and AMD 's most CPUs. 
-Thanks Anime Tosho and MikeSW17 for long tests. 
-But, I'm not sure the perfomance of rare strange kind CPUs. 
-If you want to compare speed of different settings on your CPU, 
-you may try samples (TestBlock_2023-08-31.zip) in "MultiPar_sample" folder 
-on [OneDrive](https://1drv.ms/u/s!AtGhNMUyvbWOg0cF2UHcs709Icv4).
+  It's difficult to support multi-monitor of different DPI. 
+Thanks [kwinz for reporting this problem](https://github.com/Yutaka-Sawada/MultiPar/issues/146). 
+Though I could not solve blurry window on multi-monitor case, users can see what it is.
 
-  I improved GPU implementation very much. 
-Thanks [Slava46 and K2M74 for many tests](https://github.com/Yutaka-Sawada/MultiPar/issues/99). 
-While I almost gave up to increase speed, their effort encouraged me to try many ways. 
-Without their aid, I could not implement this GPU function. 
-OpenCL perfomance is varied in every graphics boards. 
-If you have a fast graphics board, enabling "GPU acceleration" would be faster. 
-If it's not so fast (or is slow) on your PC, just un-check the feature.
+  I added an optional feature to set max size of recovery files directly.
+For details, refer `Command_GUI.txt` and `Command_par2j.txt` in `help` folder of MultiPar.
 
-  I saw a new feature of Inno Setup 6, which changes install mode. 
-It shows a dialog to ask which install mode. 
-Then, a user can install MultiPar in "Program Files" directory by selecting "Install for all users". 
-This method may be easier than starting installer by "Run as administrator". 
-I test the selection dialog at this version. 
-If there is no problem nor complaint from users, I use this style in later versions, too.
+  My web-pages on `vector.co.jp` disappered at 2024 December 20. 
+Thanks Vector to support MultiPar for long time. 
+I use [this GitHub page](https://github.com/Yutaka-Sawada/MultiPar) for MultiPar announcement.
 
 
-[ Changes from 1.3.3.0 to 1.3.3.1 ]  
+[ Changes from 1.3.3.4 to 1.3.3.5 ]  
 
 Installer update
-- It shows dialog to select "per user" or "per machine" installation.
+- Inno Setup was updated from v6.3.1 to v6.5.4.
+
+GUI update
+- Change
+  - While it supports high DPI, it looks blurry on multi-monitor of different DPI.
+  - It's possible to set maximum packet repetition from 1 time to 6 times.
 
 PAR2 client update
 - Change
-  - Max number of threads to read files on SSD was increased to 6.
-
-- Improvement
-  - GPU acceleration would become faster.
+  - It's possible to set max size of recovery files directly.
 
 
 [ Hash value ]  
 
-MultiPar1331.zip  
-MD5: ECFC1570C839DD30A2492A7B05C2AD6E  
-SHA1: 5E0E4CC38DAA995294A93ECA10AEB3AE84596170  
+  MultiPar1335.zip  
+MD5: 22F3840CDCA56CDBFA39E872D337B460  
+SHA1: A332E230137F6BDF594CD0EB7B6A5130D1CFE71C  
+
+  MultiPar1335_setup.exe  
+MD5: 2D178ED0E66C6AD0AE5C3507A6643EF5  
+SHA1: BA04FD62391351F0698B1BCAD384DD95DEA1E8C3  
 
-MultiPar1331_setup.exe  
-MD5: A55E6FA5A6853CB42E3410F35706BAD9  
-SHA1: 8D46BD6702E82ABA9ACCFA5223B2763B4DCEFE9E  
   To install under "Program Files" or "Program Files (x86)" directory, 
 you must select "Install for all users" at the first dialog.
 
   Old versions and source code packages are available at 
 [GitHub](https://github.com/Yutaka-Sawada/MultiPar/releases) or 
-[OneDrive](https://1drv.ms/u/s!AtGhNMUyvbWOaSo1n_R8awJ_hg0).
+[OneDrive](https://1drv.ms/f/c/8eb5bd32c534a1d1/QtGhNMUyvbUggI5pAAAAAAAAKjWf9HxrAn-GDQ).
+

alpha/ReadMe_English.txt

@@ -25,7 +25,13 @@ Be careful to use those special features.
 
 [ System requirement ]
 
- MultiPar requires a PC with Windows Vista or later (Windows 7, 8, 10).
+ MultiPar requires a PC with Windows 7 or later (Windows 8, 10, 11).
+
+[ Usage manual or Help documents ]
+
+ There are some usage manual or help documents in "help" folder.
+English pages exists in "help/0409" folder.
+You may open the manual by pushing "F1-key", while using MultiPar.
 
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
@@ -140,9 +146,6 @@ so you need to move "MultiPar.ini" into the folder, too.
 They are written by Yutaka Sawada.
 Though console applications are open source (PAR clients are GPL),
 GUI application is closed source.
-Some article are available at my web site.
-(URL: "https://hp.vector.co.jp/authors/VA021385/")
-If you want source code, contact with me by e-mail.
 
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
@@ -168,10 +171,8 @@ mail server may delete it automatically, and I won't see it.
 
 [ Link ]
 
- I use Vector 's author page to introduce MultiPar.
-(URL: "https://hp.vector.co.jp/authors/VA021385/")
-Because there is another official download page,
+ You may download the latest version of MultiPar on GitHub.
+(URL: "https://github.com/Yutaka-Sawada/MultiPar")
+There is a Japanese download page on Vector.
 (URL: "https://www.vector.co.jp/soft/dl/winnt/util/se460801.html")
-using direct link to files on the page isn't preferable.
-When you write a link on somewhere, please don't include filename.
 

alpha/ReadMe_Japanese.txt

@@ -54,7 +54,13 @@ QuickPar など他の PAR クライアントはコメント機能に対応して
 
 [ 動作環境 ]
 
-　Windows Vista かそれ以降 (Windows 7, 8, 10) のパソコンが必要です。
+　Windows 7 かそれ以降 (Windows 8, 10, 11) のパソコンが必要です。
+
+[ 使い方や解説文章 ]
+
+　ほとんど英語ですが、help フォルダーの中に使い方や解説文章が入ってます。
+日本語のページは help\0411 フォルダーの中にあります。
+MultiPar の使用中に、F1-key を押すと、マニュアルが表示されます。
 
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
@@ -213,9 +219,9 @@ GitHubのissueページに書き込んでもいいです。
 
 [ リンク ]
 
-　ベクターの作者ページでは外国人向けに MultiPar を紹介しています。
-(URL: "https://hp.vector.co.jp/authors/VA021385/")
-ベクターの正式なダウンロード・ページは別にあります。
+　MultiPar の最新版は GitHub からダウンロードできます。
+(URL: "https://github.com/Yutaka-Sawada/MultiPar")
+ベクターの日本語ダウンロード・ページもあります。
 (URL: "https://www.vector.co.jp/soft/dl/winnt/util/se460801.html")
 日本人向けにリンクを載せる際は、ベクターの作品紹介ページにしてください。
 (URL: "https://www.vector.co.jp/soft/winnt/util/se460801.html")

alpha/Update_English.txt

@@ -1,5 +1,70 @@
 Release note of v1.3.3 tree
 
+ par2j's "lc" option was changed to support more threads.
+ Windows Vista was removed from supported OS.
+
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+[ Changes from 1.3.3.5 to 1.3.3.6 ] (2025/10/25)
+
+GUI update
+ Change
+  GPU option is enabled always, even when GPU device doesn't exist.
+
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+[ Changes from 1.3.3.4 to 1.3.3.5 ] (2025/10/10)
+
+Installer update
+ Inno Setup was updated from v6.3.1 to v6.5.4.
+
+GUI update
+ Change
+  While it supports high DPI, it looks blurry on multi-monitor of different DPI.
+  It's possible to set maximum packet repetition from 1 time to 6 times.
+
+PAR2 client update
+ Change
+  It's possible to set max size of recovery files directly.
+
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+[ Changes from 1.3.3.3 to 1.3.3.4 ] (2024/11/30)
+
+Installer update
+ Korean language is available at installation.
+
+GUI update
+ New
+  Korean language UI is added.
+  It's possible to add options for par2j on MultiPar.ini file.
+
+ Change
+  Link of Vector's author page is removed. Use GitHub's MultiPar page.
+
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+[ Changes from 1.3.3.2 to 1.3.3.3 ] (2024/06/15)
+
+Installer update
+ Inno Setup was updated from v6.2.2 to v6.3.1.
+
+PAR2 client update
+ Bug fix
+  Fixed a bug in GPU acceleration, when there are many OpenCL devices.
+  Failure of splitting source files with numerical extension was fixed.
+  Faulty prediction of the last block in a file with repeated data was fixed.
+
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
+[ Changes from 1.3.3.1 to 1.3.3.2 ] (2024/01/10)
+
+PAR2 client update
+ Improvement
+  GPU acceleration will work well on AMD graphics boards.
+
+/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
+
 [ Changes from 1.3.3.0 to 1.3.3.1 ] (2023/11/11)
 
 Installer update

alpha/Update_Japanese.txt

@@ -1,6 +1,8 @@
-v1.3.3 の更新情報 (2023/11/11)
+v1.3.3 の更新情報 (2025/10/10)
 
-　まだ動作実験中ですので、不安な人は前のバージョンを使ってください。
+　Windows Vista を動作対象の OS から外しました。
+　Vector のページが無くなるので、GitHub のページを使ってください。
+https://github.com/Yutaka-Sawada/MultiPar
 
 [ 1.3.2 から 1.3.3 への変更点 ]
 

alpha/help/0409/batch.htm

@@ -16,7 +16,7 @@ textarea{width:100%;}
 
 <p>&nbsp
 Because MultiPar consists of PAR clients and GUI, 
-it is possible to use a PAR client on Command prompt. 
+it is possible to use a PAR client on Command Prompt. 
 Read a manual of command line for the details of command and option. 
 It's available by batch file (or command script).
 </p>
@@ -27,13 +27,57 @@ Modify the options for PAR2 files, and set the path of <tt>par2j.exe</tt>.
 Write absolute path like;<br>
 <code>SET par2_path="C:\something directory\MultiPar\par2j.exe"</code><br>
 Read manuals (<tt>Command_***.txt</tt>) to know the detail of options. 
-Read Windows OS 's help for Command prompt's usage.
+Read Windows OS 's help for Command Prompt's usage.
 </p>
 <p>&nbsp
 Save a sample script to a file like <tt>batch.bat</tt> or <tt>batch.cmd</tt>. 
-To specify path of files or folders, Drag & Drop them on the batch file. 
-When you put a shortcut icon of the batch file in "SendTo" menu, 
-you can call the batch file by selecting files then Righ-Click & SendTo.
+There are some ways to specify a file or folder.<br>
+<ol>
+<li>Type everytime by keyboard on Command Prompt
+<p>&nbsp
+If you use the batch file at Command Prompt, change directory to the batch file at first. 
+Next, type <tt>batch.bat "path of the file or folder"</tt>. 
+If there is no space in the path, no need to cover by <tt>""</tt>. 
+The path may be absolute path or relative path from the batch file. 
+Normally absolute path would be safe.
+</p>
+<p>&nbsp
+If you want to specify multiple files or folders, 
+type <tt>batch.bat "path of the first file or folder" "path of second file or folder" "path of third file or folder"</tt>. 
+You may add some pathes after a script filename.
+</p>
+</li>
+<li>Drag & Drop a file or folder by mouse
+<p>&nbsp
+On Windows Explorer, drag a file or folder and drop it on your batch file. 
+If you put the batch file on Desktop, it will be easy to Drag & Drop.
+</p>
+<p>&nbsp
+If you want to specify multiple files or folders, 
+select them at once on Windows Explorer, and Drag & Drop the group on the batch file. 
+While dragging, number of items may be shown.
+</p>
+</li>
+<li>Select SendTo at Windows Explorer's right click menu
+<p>&nbsp
+At first, create Short-cut icon of your batch file. 
+Name it to be something easy to understand like "Create PAR2". 
+Then, put the Short-cut icon in your <tt>SendTo</tt> folder. 
+You can open the <tt>SendTo</tt> folder by typing <tt>shell:sendto</tt> on path-box of Windows Explorer.
+</p>
+<p>&nbsp
+After you select a file or folder on Windows Explorer, click mouse's right button. 
+Select "SendTo" item on the right-click menu. 
+As your created Short-cut item will exist in the SendTo sub-menu, select the Short-cut icon.
+</p>
+<p>&nbsp
+If you want to specify multiple files or folders, 
+select them at once on Windows Explorer. 
+While mouse cursor over there, click mouse's right button. 
+Later action is same as single item.
+</p>
+</li>
+</ol>
 </p>
 <p>&nbsp
 If you want to confirm the result of scripting, 

alpha/help/0409/donate.htm

@@ -13,13 +13,37 @@
 <p>&nbsp
 While MultiPar is available as freeware, I spent a lot of time and effort to develop. 
 Your donation will encourage me to continue my efforts to improve and support MultiPar. 
-I made PayPal account and quick link for donation.<br>
-<ol>
-<li>Click <a href="https://www.paypal.com/cgi-bin/webscr?cmd=_s-xclick&hosted_button_id=NPNAKA32N4GD6" target="_blank" title="PayPal - The safer, easier way to pay online!">this link</a> to go PayPal's web-page.
-<li>On left panel, you can change Quantity, and click Update to multiply the Amount.
-<li>On right panel, you select your PayPal account or Credit card.
-<li>You may write Message to developer at the bottom of right panel.
-</ol>
+I made PayPal account and quick link for donation.
+</p>
+
+<p>
+If you want to send message to me, use below web-form.
+Select how much amount, write message, and click Doante button.
+<center>
+<form action="https://www.paypal.com/cgi-bin/webscr" method="post" target="_top">
+<input type="hidden" name="cmd" value="_s-xclick">
+<input type="hidden" name="hosted_button_id" value="2PP6DCPP5ZGNJ">
+<table border=0 cellspacing=0 cellpadding=0>
+<tr><td><input type="hidden" name="on0" value="How much">How much:<br>
+<select name="os0">
+<option value="Cheer">Cheer $10.00 USD</option>
+<option value="Encouragement">Encouragement $20.00 USD</option>
+<option value="Premium">Premium $30.00 USD</option>
+<option value="Subsidy">Subsidy $40.00 USD</option>
+<option value="Surprise">Surprise $50.00 USD</option>
+</select></td><td>&nbsp&nbsp</td>
+<td><input type="hidden" name="on1" value="Short message">Short message:<br>
+<input type="text" name="os1" maxlength="200" size="48"></td><td>&nbsp&nbsp</td>
+<td><input type="image" src="paypal.gif" border="0" name="submit" alt="PayPal - The safer, easier way to pay online!"></td></tr>
+</table>
+<input type="hidden" name="currency_code" value="USD">
+</form>
+</center>
+</p>
+
+<p>
+If you don't write message, click <a href="https://www.paypal.com/cgi-bin/webscr?cmd=_s-xclick&hosted_button_id=NPNAKA32N4GD6" target="_blank" title="PayPal - The safer, easier way to pay online!">this link</a> to go PayPal's web-page.
+You may change Quantity to multiply the total amount.
 </p>
 
 <p>

alpha/help/0409/index.htm

@@ -51,7 +51,7 @@ There are command-line manuals in "<tt>help</tt>" folder.
 </table>
 
 <hr>
-<small>last update 2023/11/11 for version 1.3.3.1</small>
+<small>last update 2025/01/20 for version 1.3.3.5</small>
 
 </body>
 </html>

alpha/help/0409/python.htm

@@ -174,15 +174,19 @@ It's possible to stop queue on GUI.
 <tr><td>Script file<td><tt>queue_verify.py</tt>
 <tr><td>Caution<td>You must select MultiPar Option: "Re-use verification result" to be "For 3 days" or longer, 
 and you should check "Don't search subfolders" in "Verification and Repair options" section on "Client behavior" tab. 
+Furthermore, you must set proper <tt>save_path</tt> in the script.
 </table>
 </p>
 <p>&nbsp
-In this sample, it tries verification only. 
-If you want to repair a damaged set automatically, use <tt>repair</tt> command instead of <tt>verify</tt> of par2j. 
+In this sample, it tries verification at first. 
+If you want to repair a damaged set automatically, check "Repair" item on the GUI panel. 
 When you want to confirm result by MultiPar, select a PAR2 set and push "Open with MultiPar" button. 
-You may repair on the MultiPar GUI.
+You may repair on the MultiPar GUI manually.
 </p>
 <p>&nbsp
+After you start verification queue, you may stop next verification by pushing "Stop" button. 
+After the queue was stopped, you may re-start queue by pushing "Start" button again. 
+Or, you may select another folder by pushing "Folder" button. 
 While verification, it shows details on Command Prompt window. 
 You can control the progress by pushing a key on the Command Prompt. 
 You may pause the process by pushing P-key. 
@@ -193,7 +197,8 @@ Even when you stop the verification task, it may save result partially.
 <p>&nbsp
 When you push "Folder" button, it searches PAR2 sets in the selected folder. 
 By default, It doesn't search files recursively, because it may be slow. 
-If you want to search all sub-directories, you need to edit the script. 
+If you want to search all sub-directories, select "Recursive" item on the GUI panel. 
+While you don't start verification yet, you may select another folder by pushing "Stop" button. 
 You may change behavior of starting verification automatically or not by editing the script.
 </p>
 

alpha/help/0409/readme1.htm

@@ -31,7 +31,14 @@ Be careful to use those special features.
 
 <h3>System requirement</h3>
 <p>&nbsp
-MultiPar requires a PC with Windows Vista or later (Windows 7, 8, 10).
+MultiPar requires a PC with Windows 7 or later (Windows 8, 10, 11).
+</p>
+
+<h3>Usage manual or Help documents</h3>
+<p>&nbsp
+There are some usage manual or help documents in "help" folder.
+English pages exists in "help/0409" folder.
+You may open the manual by pushing "F1-key", while using MultiPar.
 </p>
 
 </body>

alpha/help/0409/readme4.htm

@@ -13,9 +13,6 @@ MultiPar consists of PAR clients and GUI to control them.
 They are written by Yutaka Sawada. 
 Though console applications are open source (PAR clients are GPL), 
 GUI application is closed source. 
-Some article are available at 
-<a href="https://hp.vector.co.jp/authors/VA021385/" target="_blank" title="Announcement page on Vector">my web site</a>. 
-There are source code packages on <a href="https://github.com/Yutaka-Sawada/MultiPar" target="_blank" title="GitHub page">GitHub</a>.
 </p>
 
 <hr>
@@ -42,10 +39,8 @@ mail server may delete it automatically, and I won't see it.
 
 <h3>Link</h3>
 <p>&nbsp
-I use <i>vector.co.jp</i>'s <a href="https://hp.vector.co.jp/authors/VA021385/" target="_blank" title="Announcement page on Vector">author page</a> to introduce MultiPar. 
-Because there is another <a href="https://www.vector.co.jp/soft/dl/winnt/util/se460801.html" target="_blank" title="Download page on Vector">official download page</a>, 
-using direct link to files on the page isn't preferable. 
-When you write a link on somewhere, please don't include filename. 
+You may download the latest version of <a href="https://github.com/Yutaka-Sawada/MultiPar" target="_blank">MultiPar on GitHub</a>. 
+There is <a href="https://www.vector.co.jp/soft/dl/winnt/util/se460801.html" target="_blank">a Japanese download page on Vector</a>.
 </p>
 
 </body>

alpha/help/0411/donate.htm

@@ -0,0 +1,48 @@
+<html lang="ja">
+<head>
+<meta http-equiv="Content-Type" content="text/html; charset=Shift_JIS">
+<link rel="stylesheet" href="style.css">
+<title><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>̂<EFBFBD><EFBFBD>肢</title>
+</head>
+<body>
+
+<font size=5><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>̂<EFBFBD><EFBFBD>肢</font>
+<hr>
+
+<h3>PayPal <20>ł̊<C582><CC8A>t<EFBFBD><74><EFBFBD><EFBFBD><EFBFBD>肢<EFBFBD><E882A2><EFBFBD>܂<EFBFBD></h3>
+<p>&nbsp
+MultiPar <20>͖<EFBFBD><CD96><EFBFBD><EFBFBD>ŗ<EFBFBD><C597>p<EFBFBD>ł<EFBFBD><C582>܂<EFBFBD><DC82><EFBFBD><EFBFBD>A<EFBFBD>J<EFBFBD><4A><EFBFBD>ɂ͎<C982><CD8E>ԂƎ<D482><C68E>Ԃ<EFBFBD><D482><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă܂<C482><DC82>B
+<EFBFBD><EFBFBD><EFBFBD>t<EFBFBD><EFBFBD><EFBFBD>Ē<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ƁA<EFBFBD>J<EFBFBD><EFBFBD><EFBFBD>𑱂<EFBFBD><EFBFBD><EFBFBD><EFBFBD>C<EFBFBD>͂<EFBFBD><EFBFBD>o<EFBFBD>܂<EFBFBD><EFBFBD>B
+PayPal <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>p<EFBFBD>ӂ<EFBFBD><D382><EFBFBD><EFBFBD>̂ŁA<C581><41><EFBFBD>Њ<EFBFBD><D08A>t<EFBFBD><74><EFBFBD><EFBFBD><EFBFBD>肢<EFBFBD><E882A2><EFBFBD>܂<EFBFBD><DC82>B
+</p>
+
+<p>
+<EFBFBD><EFBFBD><EFBFBD>z<EFBFBD><EFBFBD><EFBFBD>I<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>āA<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>Z<EFBFBD>[<5B>W<EFBFBD><57><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>āADoante <20>{<7B>^<5E><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><C482><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>B
+<center>
+<form action="https://www.paypal.com/cgi-bin/webscr" method="post" target="_top">
+<input type="hidden" name="cmd" value="_s-xclick">
+<input type="hidden" name="hosted_button_id" value="2PP6DCPP5ZGNJ">
+<table border=0 cellspacing=0 cellpadding=0>
+<tr><td><input type="hidden" name="on0" value="How much"><EFBFBD><EFBFBD><EFBFBD>z:<br>
+<select name="os0">
+<option value="Cheer">Cheer $10.00 USD</option>
+<option value="Encouragement">Encouragement $20.00 USD</option>
+<option value="Premium">Premium $30.00 USD</option>
+<option value="Subsidy">Subsidy $40.00 USD</option>
+<option value="Surprise">Surprise $50.00 USD</option>
+</select></td><td>&nbsp&nbsp</td>
+<td><input type="hidden" name="on1" value="Short message"><EFBFBD><EFBFBD><EFBFBD>҂ւ̃<EFBFBD><EFBFBD>b<EFBFBD>Z<EFBFBD>[<5B>W:<br>
+<input type="text" name="os1" maxlength="200" size="48"></td><td>&nbsp&nbsp</td>
+<td><input type="image" src="../0409/paypal.gif" border="0" name="submit" alt="PayPal - The safer, easier way to pay online!"></td></tr>
+</table>
+<input type="hidden" name="currency_code" value="USD">
+</form>
+</center>
+</p>
+
+<p>
+<EFBFBD>V<EFBFBD>@<40>\<5C>̎<EFBFBD><CC8E><EFBFBD><EFBFBD><EFBFBD><EFBFBD>A<EFBFBD>o<EFBFBD>O<EFBFBD>񍐁A<F18D9081><41><EFBFBD>P<EFBFBD>̏<EFBFBD><CC8F><EFBFBD><EFBFBD>ȂǁA<C781><41><EFBFBD>̌`<60>ł̍v<CC8D><76><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>肪<EFBFBD><E882AA><EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><C582>B
+</p>
+
+</body>
+</html>

alpha/help/0411/index.htm

@@ -37,6 +37,7 @@
 <a href="howto3.htm"><EFBFBD><EFBFBD><EFBFBD>J<EFBFBD>o<EFBFBD><EFBFBD><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>̐݌v</a><br>
 <a href="howto4.htm"><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>̌<EFBFBD><CC8C><EFBFBD><EFBFBD>ƏC<C68F><43></a><br>
 <a href="problem.htm"><EFBFBD><EFBFBD><EFBFBD>m<EFBFBD>̖<EFBFBD><EFBFBD><EFBFBD><EFBFBD>_</a><br>
+<a href="donate.htm"><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>̂<EFBFBD><EFBFBD>肢</a>
 </td></tr>
 <tr><td colspan=3><font size=1>&nbsp</font></td></tr>
 <tr valign="top"><td>
@@ -51,7 +52,7 @@
 </table>
 
 <hr>
-<small><EFBFBD>ŏI<EFBFBD>X<EFBFBD>V 2023/11/11 (<28>o<EFBFBD>[<5B>W<EFBFBD><57><EFBFBD><EFBFBD> 1.3.3.1)</small>
+<small><EFBFBD>ŏI<EFBFBD>X<EFBFBD>V 2025/01/21 (<28>o<EFBFBD>[<5B>W<EFBFBD><57><EFBFBD><EFBFBD> 1.3.3.5)</small>
 
 </body>
 </html>

alpha/help/0411/readme1.htm

@@ -25,7 +25,14 @@ QuickPar
 
 <h3><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>‹<EFBFBD></h3>
 <p>&nbsp
-Windows Vista <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȍ~ (Windows 7, 8, 10) <20>̃p<CC83>\<5C>R<EFBFBD><52><EFBFBD><EFBFBD><EFBFBD>K<EFBFBD>v<EFBFBD>ł<EFBFBD><C582>B
+Windows 7 <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȍ~ (Windows 8, 10, 11) <20>̃p<CC83>\<5C>R<EFBFBD><52><EFBFBD><EFBFBD><EFBFBD>K<EFBFBD>v<EFBFBD>ł<EFBFBD><C582>B
+</p>
+
+<h3><EFBFBD>g<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD></h3>
+<p>&nbsp
+<EFBFBD>قƂ<EFBFBD><EFBFBD>ljp<EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ahelp <20>t<EFBFBD>H<EFBFBD><48><EFBFBD>_<EFBFBD>[<5B>̒<EFBFBD><CC92>Ɏg<C98E><67><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>͂<EFBFBD><CD82><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă܂<C482><DC82>B
+<EFBFBD><EFBFBD><EFBFBD>{<7B><><EFBFBD>̃y<CC83>[<5B>W<EFBFBD><57> help\0411 <20>t<EFBFBD>H<EFBFBD><48><EFBFBD>_<EFBFBD>[<5B>̒<EFBFBD><CC92>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
+MultiPar <20>̎g<CC8E>p<EFBFBD><70><EFBFBD>ɁAF1-key <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ƁA<C681>}<7D>j<EFBFBD><6A><EFBFBD>A<EFBFBD><41><EFBFBD><EFBFBD><EFBFBD>\<5C><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
 </p>
 
 </body>

alpha/help/0411/readme4.htm

@@ -14,7 +14,7 @@ PAR
 <EFBFBD><EFBFBD><EFBFBD>̑<EFBFBD><EFBFBD>̃R<EFBFBD><EFBFBD><EFBFBD>\<5C>[<5B><><EFBFBD>E<EFBFBD>A<EFBFBD>v<EFBFBD><76><EFBFBD>P<EFBFBD>[<5B>V<EFBFBD><56><EFBFBD><EFBFBD><EFBFBD><EFBFBD>
 GUI <20>A<EFBFBD>v<EFBFBD><76><EFBFBD>P<EFBFBD>[<5B>V<EFBFBD><56><EFBFBD><EFBFBD> (<tt>MultiPar.exe</tt>) <20>͕<EFBFBD><CD95>ʂ̃t<CC83><74><EFBFBD>[<5B>\<5C>t<EFBFBD>g<EFBFBD>ł<EFBFBD><C582>B
 MultiPar <20><><EFBFBD><EFBFBD><EFBFBD>̂܂܎g<DC8E><67><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882>A<EFBFBD><41><EFBFBD>ꂼ<EFBFBD><EA82BC><EFBFBD>̃<EFBFBD><CC83>C<EFBFBD>Z<EFBFBD><5A><EFBFBD>X<EFBFBD>̈Ⴂ<CC88><E182A2><EFBFBD>C<EFBFBD>ɂ<EFBFBD><C982>Ȃ<EFBFBD><C882>Ă<EFBFBD><C482><EFBFBD><EFBFBD>ł<EFBFBD><C582>B
-<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>R<EFBFBD>[<5B>h<EFBFBD><68><a href="https://github.com/Yutaka-Sawada/MultiPar" target="_blank" title="GitHub page">GitHub</a><EFBFBD>ɒu<EFBFBD><EFBFBD><EFBFBD>Ă܂<EFBFBD><EFBFBD>B
+<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>R<EFBFBD>[<5B>h<EFBFBD><68> <a href="https://github.com/Yutaka-Sawada/MultiPar" target="_blank" title="GitHub page">GitHub</a> <EFBFBD>ɒu<EFBFBD><EFBFBD><EFBFBD>Ă܂<EFBFBD><EFBFBD>B
 </p>
 
 <hr>
@@ -33,9 +33,9 @@ MultiPar
 
 <h3><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>N</h3>
 <p>&nbsp
-<EFBFBD>x<EFBFBD>N<EFBFBD>^<5E>[<5B><><a href="https://hp.vector.co.jp/authors/VA021385/" target="_blank" title="<22>x<EFBFBD>N<EFBFBD>^<5E>[<5B>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>L<EFBFBD><4C><EFBFBD>y<EFBFBD>[<5B>W"><EFBFBD><EFBFBD><EFBFBD>҃y<EFBFBD>[<5B>W</a><EFBFBD>ł͊O<EFBFBD><EFBFBD><EFBFBD>l<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> MultiPar <20><><EFBFBD>Љ<D089>Ă<EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
-<EFBFBD>x<EFBFBD>N<EFBFBD>^<5E>[<5B><><a href="https://www.vector.co.jp/soft/dl/winnt/util/se460801.html" target="_blank" title="<22>x<EFBFBD>N<EFBFBD>^<5E>[<5B>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>_<EFBFBD>E<EFBFBD><45><EFBFBD><EFBFBD><EFBFBD>[<5B>h<EFBFBD>E<EFBFBD>y<EFBFBD>[<5B>W"><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȃ_<EFBFBD>E<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B>h<EFBFBD>E<EFBFBD>y<EFBFBD>[<5B>W</a><EFBFBD>͕ʂɂ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
-<EFBFBD><EFBFBD><EFBFBD>{<7B>l<EFBFBD><6C><EFBFBD><EFBFBD><EFBFBD>Ƀ<EFBFBD><C983><EFBFBD><EFBFBD>N<EFBFBD><4E><EFBFBD>ڂ<EFBFBD><DA82><EFBFBD><EFBFBD>ۂ́A<CD81>x<EFBFBD>N<EFBFBD>^<5E>[<5B><><a href="https://www.vector.co.jp/soft/winnt/util/se460801.html" target="_blank" title="<22>x<EFBFBD>N<EFBFBD>^<5E>[<5B>ɂ<EFBFBD><C982><EFBFBD><EFBFBD><EFBFBD><EFBFBD>i<EFBFBD>Љ<EFBFBD><D089>y<EFBFBD>[<5B>W"><EFBFBD><EFBFBD><EFBFBD>i<EFBFBD>Љ<EFBFBD><EFBFBD>y<EFBFBD>[<5B>W</a><EFBFBD>ɂ<EFBFBD><EFBFBD>Ă<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>B
+MultiPar <20>̍ŐV<C590>ł<EFBFBD> <a href="https://github.com/Yutaka-Sawada/MultiPar" target="_blank">GitHub <20><><EFBFBD><EFBFBD></a><EFBFBD>_<EFBFBD>E<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B>h<EFBFBD>ł<EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
+<EFBFBD>x<EFBFBD>N<EFBFBD>^<5E>[<5B><><a href="https://www.vector.co.jp/soft/dl/winnt/util/se460801.html" target="_blank"><EFBFBD><EFBFBD><EFBFBD>{<7B><EFBFBD><EFBFBD>_<EFBFBD>E<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B>h<EFBFBD>E<EFBFBD>y<EFBFBD>[<5B>W</a><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
+<EFBFBD><EFBFBD><EFBFBD>{<7B>l<EFBFBD><6C><EFBFBD><EFBFBD><EFBFBD>Ƀ<EFBFBD><C983><EFBFBD><EFBFBD>N<EFBFBD><4E><EFBFBD>ڂ<EFBFBD><DA82><EFBFBD><EFBFBD>ۂ́A<CD81>x<EFBFBD>N<EFBFBD>^<5E>[<5B><><a href="https://www.vector.co.jp/soft/winnt/util/se460801.html" target="_blank"><EFBFBD><EFBFBD><EFBFBD>i<EFBFBD>Љ<EFBFBD><EFBFBD>y<EFBFBD>[<5B>W</a><EFBFBD>ɂ<EFBFBD><EFBFBD>Ă<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>B
 </p>
 
 </body>

alpha/help/0804/donate.htm

@@ -11,13 +11,37 @@
 
 <h3>PayPal捐赠</h3>
 <p>&nbsp
-尽管MultiPar可以免费使用，但我花了很多时间和精力来开发。您的捐款将鼓励我继续努力改进和支持MultiPar。我创建了PayPal帐户和捐款的快速链接。<br>
-<ol>
-<li>点击<a href="https://www.paypal.com/cgi-bin/webscr?cmd=_s-xclick&hosted_button_id=NPNAKA32N4GD6" target="_blank" title="PayPal - The safer, easier way to pay online!">此链接</a>转到PayPal的网页。
-<li>在左侧面板上，您可以更改数量，然后单击更新以乘以金额。
-<li>在右侧面板中，您可以选择PayPal帐户或信用卡。
-<li>您可以在右侧面板底部向开发人员写消息。
-</ol>
+尽管MultiPar可以免费使用，但我花了很多时间和精力来开发。您的捐款将鼓励我继续努力改进和支持MultiPar。我创建了PayPal帐户和捐款的快速链接。
+</p>
+
+<p>
+If you want to send message to me, use below web-form.
+Select how much amount, write message, and click Doante button.
+<center>
+<form action="https://www.paypal.com/cgi-bin/webscr" method="post" target="_top">
+<input type="hidden" name="cmd" value="_s-xclick">
+<input type="hidden" name="hosted_button_id" value="2PP6DCPP5ZGNJ">
+<table border=0 cellspacing=0 cellpadding=0>
+<tr><td><input type="hidden" name="on0" value="How much">How much:<br>
+<select name="os0">
+<option value="Cheer">Cheer $10.00 USD</option>
+<option value="Encouragement">Encouragement $20.00 USD</option>
+<option value="Premium">Premium $30.00 USD</option>
+<option value="Subsidy">Subsidy $40.00 USD</option>
+<option value="Surprise">Surprise $50.00 USD</option>
+</select></td><td>&nbsp&nbsp</td>
+<td><input type="hidden" name="on1" value="Short message">Short message:<br>
+<input type="text" name="os1" maxlength="200" size="48"></td><td>&nbsp&nbsp</td>
+<td><input type="image" src="../0409/paypal.gif" border="0" name="submit" alt="PayPal - The safer, easier way to pay online!"></td></tr>
+</table>
+<input type="hidden" name="currency_code" value="USD">
+</form>
+</center>
+</p>
+
+<p>
+If you don't write message, 点击<a href="https://www.paypal.com/cgi-bin/webscr?cmd=_s-xclick&hosted_button_id=NPNAKA32N4GD6" target="_blank" title="PayPal - The safer, easier way to pay online!">此链接</a>转到PayPal的网页。
+You may change Quantity to multiply the total amount.
 </p>
 
 <p>

alpha/help/0804/index.htm

@@ -50,7 +50,7 @@
 </table>
 
 <hr>
-<small>最新更新于2023年11月11日，适用于1.3.3.1版本，简体中文化 Deng Shiqing</small>
+<small>最新更新于2025年01月20日，适用于1.3.3.5版本，简体中文化 Deng Shiqing</small>
 
 </body>
 </html>

alpha/help/0804/readme1.htm

@@ -22,7 +22,7 @@ MultiPar支持PAR1.0和PAR2.0规范。请访问“<a href="http://parchive.sourc
 
 <h3>系统要求</h3>
 <p>&nbsp
-MultiPar需要电脑安装Windows Vista或更高的版本（Windows 7, 8, 10）。
+MultiPar需要电脑安装Windows 7或更高的版本（Windows 8, 10, 11）。
 </p>
 
 </body>

alpha/help/0804/readme4.htm

@@ -7,7 +7,7 @@
 <body>
 <h3>许可证</h3>
 <p>&nbsp
-MultiPar由PAR客户端和用于控制它们的图形用户界面（GUI）组成，由Yutaka Sawada编写。尽管控制台应用程序是开源的（PAR客户端采用GPL协议），但GUI程序是闭源的。有些文章可以在<a href="https://hp.vector.co.jp/authors/VA021385/" target="_blank" title="Announcement page on Vector">我的网站</a>上找到。
+MultiPar由PAR客户端和用于控制它们的图形用户界面（GUI）组成，由Yutaka Sawada编写。尽管控制台应用程序是开源的（PAR客户端采用GPL协议），但GUI程序是闭源的。
 There are source code packages on <a href="https://github.com/Yutaka-Sawada/MultiPar" target="_blank" title="GitHub page">GitHub</a>.
 </p>
 
@@ -26,7 +26,7 @@ I use <a href="https://github.com/Yutaka-Sawada/MultiPar/issues" target="_blank"
 
 <h3>链接</h3>
 <p>&nbsp
-我在<a href="https://hp.vector.co.jp/authors/VA021385/" target="_blank" title="Announcement page on Vector">个人主页</a><i>vector.co.jp</i>有介绍MultiPar。使用这个页面上的文件直链可能不太合适，这还有一个<a href="https://www.vector.co.jp/soft/dl/winnt/util/se460801.html" target="_blank" title="Download page on Vector">官方下载页面</a>。当您在某处输入一个链接时，请不要包含文件名。
+You may download the latest version of <a href="https://github.com/Yutaka-Sawada/MultiPar" target="_blank">MultiPar on GitHub</a>.
 </p>
 
 </body>

alpha/help/Command_GUI.txt

@@ -1,4 +1,4 @@
-[ MultiPar GUI - version 1.3.3.0 or later ]
+[ MultiPar GUI - version 1.3.3.4 or later ]
 
 Usage: MultiPar.exe [command] [/base path] [/list path] [files]
 
@@ -99,6 +99,11 @@ write a line of "RecoveryFileLimit=1" under "[Option]" section.
 It's same as an option:
 "Variable (limited to size of largest data file)" on QuickPar.
 
+ If you want to limit size of Recovery Files by a specific value,
+write a line of "RecoveryFileLimit=2" under "[Option]" section.
+"Limit Size to" value is enabled, even when "Split Files" isn't checked.
+This setting is useful, if you don't want to split source files.
+
  If you want to enable "Most Resent Used List",
 write this line "MRUMax=5" under "[Path]" section.
 You may change the number of items after "MRUMax=".
@@ -112,6 +117,14 @@ Because the name and size are splitted by ":",
 you cannot include ":" in the name.
 The max size is 999999999999 bytes. (931 GB)
 
+ If you want to add more options for par2j (par2j.exe or par2j64.exe),
+write this line "par2jOption=/something" under "[Option]" section.
+You may add multiple options like; "par2jOption=/m1 /lc32"
+The total length must be less than 30 characters.
+When you set an option at MultiPar's Option windows,
+you should not set the same option on .INI file manually.
+If you write wrong or strange text, par2j will fail.
+
 
 Example of lines on "MultiPar.ini";
 
@@ -121,5 +134,6 @@ RedundancyMax=10
 Sort=8
 RecoveryFileLimit=1
 MediaList4=7.9GB DVD:8480000000
+par2jOption=/lc32
 [Path]
 MRUMax=5

alpha/help/Command_par1j.txt

@@ -103,6 +103,16 @@ It means that it is impossible to mix "/" and "-" in options of same command lin
 If PAR file's filename (or relative path) is started by "-",
 you need to use "/" for prefix of options to distinguish PAR file.
 
+ When a option has arbitary number or text,
+don't put space between the option and following part.
+For example, setting "/rr1" is valid, but "/rr 1" is wrong.
+Setting "/dPath" is valid, but "/d Path" is wrong.
+
+ To distinguish options from filenames, order of command-line arguments is important.
+Put all options before <par file>.
+Put <par file> before all [input files].
+
+
  /f, /fu :
  Set this, if you use file-list for file name or file path.
 The file-list is a text file, which contains filenames in each line.
@@ -169,8 +179,10 @@ If you include space in comment, use "" to specify where is end.
 use this setting to set the path as base-directory.
 This is either absolute path or relative path from current directory.
 If you include space in path, use "" to specify where is end.
-Note, setting "/dC:" is same as setting current directory of C-drive.
-If you want to set direct root of C-drive, you must set "/dC:/".
+As you may omit the last "/" in path of a folder,
+both "/dC:/data/" and "/dC:/data" are valid.
+Note, setting "/dD:" is same as setting current directory of D-drive.
+If you want to set direct root of D-drive, you must set "/dD:/".
 
  for example, /dC:/data/ , /d"C:/Program Files/" , /d../data/
 

alpha/help/Command_par2j.txt

@@ -1,4 +1,4 @@
-[ par2j.exe - version 1.3.3.0 or later ]
+[ par2j.exe - version 1.3.3.5 or later ]
 
 Type "par2j.exe" to see version, test integrity, and show usage below.
 
@@ -148,6 +148,16 @@ It means that it is impossible to mix "/" and "-" in options of same command lin
 If PAR file's filename (or relative path) is started by "-",
 you need to use "/" for prefix of options to distinguish PAR file.
 
+ When a option has arbitary number or text,
+don't put space between the option and following part.
+For example, setting "/rr10" is valid, but "/rr 10" is wrong.
+Setting "/dPath" is valid, but "/d Path" is wrong.
+
+ To distinguish options from filenames, order of command-line arguments is important.
+Put all options before <par file>.
+Put <par file> before all [input files] or [external files].
+
+
  /f, /fu :
  Set this, if you use file-list for file name or file path.
 When this option is set, [input files] or [external files] is path of file-list.
@@ -302,7 +312,7 @@ Because it may be difficult to predict recovery file size,
 use trial command to see how is the size.
 When it's "/lr" or "/lr0", block count of the largest input file is used.
 (If the max file is 1000 KB and block size is 10 KB, /lr is same as /lr100.)
-When split size "/ls" is set with "/lr0",
+When both "/lr0" and "/ls(split size)" are set,
 block count of the splited file size is used instead of the largest input file.
 
  for example,  /lr , /lr10 , /lr2000
@@ -354,6 +364,13 @@ This is possible, only when there is one input file of .ZIP or .7z format.
 When recovery files are created in a different directory,
 the protected archive file is made in the directory.
 
+ /ls2 has a special feature to set limit size of recovery files directly.
+When both "/ls2" and "/lr(limit size)" are set,
+setting number of /lr is recognizned as file size instead of number of blocks.
+In this usage, setting limit size must be less than 2 GB.
+If you want max 100 MB each file size, set like these; /ls2 /lr102400000
+Because the calculation ignores packets, you may set a bit smaller size.
+
  for example,  /ls10000 , /ls1048576, /ls1
 
  /lc :
@@ -367,16 +384,43 @@ The format is "/lc#" (# is from 1 to 32 as the number of using threads).
 253: It uses 3/4 number of physical Cores.
 254: It uses one less threads than number of physical Cores.
   0: It uses the number of physical Cores.
-255: It uses one more threads than number of physical Cores.
+255: It tries to use more threads than number of physical Cores.
 
- You may set additional combinations;
-+1024 to disable CLMUL (and use old SSSE3 code), 
-+2048 to disable JIT (for SSE2), 
-+4096 to disable SSSE3,
-+8192 to disable AVX2,
-+256 or +512 (slower device) to enable GPU acceleration.
+ You may set additional combinations for CPU feature;
++1024 to disable CLMUL (and use slower SSSE3 code)
++2048 to disable JIT (for SSE2)
++4096 to disable SSSE3
++8192 to disable AVX2
 
- for example,  /lc1 to use single Core, /lc45 to use half Cores and GPU
+ You may set additional combinations for GPU control;
++256 or +512 (slower device) to enable GPU acceleration
++65536 for classic method
++131072 for 16-byte memory access
++262144 for 4-byte memory access and calculate 2 blocks at once
++524288 for 16-byte memory access and calculate 2 blocks at once
++1048576 for CL_MEM_COPY_HOST_PTR or +2097152 for CL_MEM_USE_HOST_PTR
+(When you set exclusive bits, larger value will be used.)
+
+ for example,  /lc1 to use single Core, /lc508 to use half Cores and GPU
+
+ /lcb :
+ This is for Cache Blocking. (CPU cache optimization)
+By default, this value is set-associative size of CPU L2 cache.
+Maximum value is 32767. It will be multipled by 64 KB.
+To disable cache optimization, set "/lcb0".
+
+ /lcs :
+ This is for large buffer size at calculating multiple files' hashes.
+By default, this value is set-associative size of CPU L3 cache.
+Maximum value is 65535. It will be multipled by 64 KB.
+When you set "/lcs0", large buffer size will become 2 MB by default.
+
+ /lcm :
+ This is for max number of chunks to calculate at once. (CPU shared cache optimization)
+By default, this value may be rate of L3 cache size / L2 cache size.
+The value may be changed by some factors in experimental results.
+Maximum value is 32768. Lower values than 8 will be same as 32768.
+To disable cache optimization, set "/lcm0".
 
  /m :
  Set this, if you want to set memory usage.
@@ -447,8 +491,10 @@ If you include space in comment, use "" to specify where is end.
 use this setting to set the path as base-directory.
 This is either absolute path or relative path from current directory.
 If you include space in path, use "" to specify where is end.
-Note, setting "/dC:" is same as setting current directory of C-drive.
-If you want to set direct root of C-drive, you must set "/dC:/".
+As you may omit the last "/" in path of a folder,
+both "/dC:/data/" and "/dC:/data" are valid.
+Note, setting "/dD:" is same as setting current directory of D-drive.
+If you want to set direct root of D-drive, you must set "/dD:/".
 
  for example, /dC:/data/ , /d"C:/Program Files/" , /d../data/
 

alpha/help/Command_sfv_md5.txt

@@ -94,6 +94,16 @@ It means that it is impossible to mix "/" and "-" in options of same command lin
 If checksum file's filename (or relative path) is started by "-",
 you need to use "/" for prefix of options to distinguish checksum file.
 
+ When a option has arbitary number or text,
+don't put space between the option and following part.
+For example, setting "/vs1" is valid, but "/vs 1" is wrong.
+Setting "/dPath" is valid, but "/d Path" is wrong.
+
+ To distinguish options from filenames, order of command-line arguments is important.
+Put all options before <checksum file>.
+Put <checksum file> before all [input files]
+
+
  /fo :
  Set this, if you want to ignore folders at wildcard matching search.
 This setting disable recursive search into folder also.
@@ -129,8 +139,10 @@ from the PAR client's directory, use this setting to set path.
 use this setting to set the path as base-directory.
 This is either absolute path or relative path from current directory.
 If you include space in path, use "" to specify where is end.
-Note, setting "/dC:" is same as setting current directory of C-drive.
-If you want to set direct root of C-drive, you must set "/dC:/".
+As you may omit the last "/" in path of a folder,
+both "/dC:/data/" and "/dC:/data" are valid.
+Note, setting "/dD:" is same as setting current directory of D-drive.
+If you want to set direct root of D-drive, you must set "/dD:/".
 
  for example, /dC:/data/ , /d"C:/Program Files/" , /d../data/
 

alpha/help/ReadMe1_English.txt

@@ -18,7 +18,7 @@ Parchive 1.0 client by Yutaka Sawada
  You cannot specify Non-Recovery-Set files. (Empty files are set.)
  Number of recovery files are limited to 99.
 
- This requires a PC of Windows Vista or later (Windows 7, 8, 10, 11).
+ This requires a PC of Windows 7 or later (Windows 8, 10, 11).
  This is developed with Visual Studio 2022 on Windows 10.
 
 

alpha/help/ReadMe2_English.txt

@@ -31,7 +31,7 @@ Parchive 2.0 client by Yutaka Sawada
  Verifying file use same memory as the file or double of block size.
  Recovering block use more memory as the number of block is many.
 
- This requires a PC of Windows Vista or later (Windows 7, 8, 10, 11).
+ This requires a PC of Windows 7 or later (Windows 8, 10, 11).
  This is developed with Visual Studio 2022 on Windows 10.
 
 

alpha/tool/queue_verify.py

@@ -43,25 +43,31 @@ def search_par_set(one_path):
             continue
         listbox_list1.insert(tk.END, base_name)
 
-
     # Add found PAR sets in sub-directories
-    # This searches 1 level child only, because recursive search may be slow.
-    for par_path in glob.glob(glob.escape(one_path) + "/*/*.par2"):
-    # If you want to search recursively, use below line instead of above line.
-    #for par_path in glob.glob(glob.escape(one_path) + "/**/*.par2", recursive=True):
-        # Get relative path and convert to UNIX style directory mark
-        rel_path = os.path.relpath(par_path, one_path)
-        rel_path = rel_path.replace('\\', '/')
-        # Remove extension ".par2"
-        one_name = os.path.splitext(rel_path)[0]
-        # Compare filename in case insensitive
-        base_name = one_name.lower()
-        # Remove ".vol#-#", ".vol#+#", or ".vol_#" at the last
-        base_name = re.sub(r'[.]vol\d*[-+_]\d+$', "", base_name)
-        # Ignore same base, if the name exists in the list already.
-        if "'" + base_name + "'" in s_list1.get():
-            continue
-        listbox_list1.insert(tk.END, base_name)
+    s_recursive = s_combo.get()
+    if s_recursive != "No child":
+        if s_recursive == "Recursive":
+            # This will search all sub-directories recursively.
+            search_path = glob.escape(one_path) + "/**/*.par2"
+            recursive_flag = True
+        else:
+            # This searches 1 level child only, because recursive search may be slow.
+            search_path = glob.escape(one_path) + "/*/*.par2"
+            recursive_flag = False
+        for par_path in glob.glob(search_path, recursive=recursive_flag):
+            # Get relative path and convert to UNIX style directory mark
+            rel_path = os.path.relpath(par_path, one_path)
+            rel_path = rel_path.replace('\\', '/')
+            # Remove extension ".par2"
+            one_name = os.path.splitext(rel_path)[0]
+            # Compare filename in case insensitive
+            base_name = one_name.lower()
+            # Remove ".vol#-#", ".vol#+#", or ".vol_#" at the last
+            base_name = re.sub(r'[.]vol\d*[-+_]\d+$', "", base_name)
+            # Ignore same base, if the name exists in the list already.
+            if "'" + base_name + "'" in s_list1.get():
+                continue
+            listbox_list1.insert(tk.END, base_name)
 
     item_count = listbox_list1.size()
     one_name = os.path.basename(one_path)
@@ -70,13 +76,14 @@ def search_par_set(one_path):
         label_status.config(text= "There are no PAR sets in \"" + one_path + "\".")
     else:
         button_folder.config(state=tk.DISABLED)
+        button_stop.config(state=tk.NORMAL)
         button_open2.config(state=tk.DISABLED)
         button_open3.config(state=tk.DISABLED)
+        combo_recursive.config(state=tk.DISABLED)
         # If you want to start manually, use these lines instead of below lines.
-        #label_status.config(text= str(item_count) + " sets were found in \"" + one_path + "\".")
+        label_status.config(text= str(item_count) + " sets were found in \"" + one_path + "\".")
         button_start.config(state=tk.NORMAL)
-        # If you want to start verification automatically, use these lines instead of above lines.
-        #button_stop.config(state=tk.NORMAL)
+        # If you want to start verification automatically, use this line instead of above lines.
         #root.after(100, queue_run)
 
 
@@ -105,6 +112,8 @@ def queue_run():
         button_start.config(state=tk.NORMAL)
         button_open2.config(state=tk.NORMAL)
         button_open3.config(state=tk.NORMAL)
+        combo_recursive.config(state=tk.NORMAL)
+        check_repair.config(state=tk.NORMAL)
         label_status.config(text= "Stopped queue")
         return
 
@@ -121,8 +130,13 @@ def queue_run():
 
     # Set command-line
     # Cover path by " for possible space
-    cmd = "\"" + client_path + "\" v /fo /vs2 /vd\"" + save_path + "\" \"" + one_path + "\""
-    # If you want to repair a damaged set automatically, use "r" command instead of "v".
+    cmd = "\"" + client_path + "\" "
+    if i_check.get() == 0:
+        cmd += "v"
+    else:
+        # If you want to repair a damaged set automatically, use "r" command instead of "v".
+        cmd += "r"
+    cmd += " /fo /vs2 /vd\"" + save_path + "\" \"" + one_path + "\""
     #print(cmd)
 
     # Run PAR2 client
@@ -163,6 +177,8 @@ def queue_result():
         button_stop.config(state=tk.DISABLED)
         button_open2.config(state=tk.NORMAL)
         button_open3.config(state=tk.NORMAL)
+        combo_recursive.config(state=tk.NORMAL)
+        check_repair.config(state=tk.NORMAL)
         label_status.config(text= "Failed queue")
         return
 
@@ -173,6 +189,8 @@ def queue_result():
         button_stop.config(state=tk.DISABLED)
         button_open2.config(state=tk.NORMAL)
         button_open3.config(state=tk.NORMAL)
+        combo_recursive.config(state=tk.NORMAL)
+        check_repair.config(state=tk.NORMAL)
         label_status.config(text= "Canceled queue")
         return
 
@@ -194,6 +212,8 @@ def queue_result():
         button_stop.config(state=tk.DISABLED)
         button_open2.config(state=tk.NORMAL)
         button_open3.config(state=tk.NORMAL)
+        combo_recursive.config(state=tk.NORMAL)
+        check_repair.config(state=tk.NORMAL)
         label_status.config(text= "Verified all PAR sets")
 
     elif "disabled" in button_stop.state():
@@ -201,6 +221,8 @@ def queue_result():
         button_start.config(state=tk.NORMAL)
         button_open2.config(state=tk.NORMAL)
         button_open3.config(state=tk.NORMAL)
+        combo_recursive.config(state=tk.NORMAL)
+        check_repair.config(state=tk.NORMAL)
         label_status.config(text= "Interrupted queue")
 
     else:
@@ -216,6 +238,8 @@ def button_start_clicked():
     button_stop.config(state=tk.NORMAL)
     button_open2.config(state=tk.DISABLED)
     button_open3.config(state=tk.DISABLED)
+    combo_recursive.config(state=tk.DISABLED)
+    check_repair.config(state=tk.DISABLED)
 
     if sub_proc == None:
         queue_run()
@@ -225,8 +249,19 @@ def button_start_clicked():
 
 # Stop running queue
 def button_stop_clicked():
+    global sub_proc
+
     button_stop.config(state=tk.DISABLED)
-    if sub_proc != None:
+    if sub_proc is None:
+        # When verification was not started yet, it's possible to select another folder.
+        button_folder.config(state=tk.NORMAL)
+        button_start.config(state=tk.DISABLED)
+        combo_recursive.config(state=tk.NORMAL)
+        listbox_list1.delete(0, tk.END)
+        label_head1.config(text='? sets in a folder')
+        label_status.config(text='Select a folder to search PAR files.')
+    else:
+        # When it's verifying, it will stop next verification.
         label_status.config(text= "Waiting finish of current task")
 
 
@@ -293,7 +328,7 @@ frame_middle.columnconfigure(2, weight=1)
 frame_list1 = ttk.Frame(frame_middle, padding=(6,2,6,6), relief='groove')
 frame_list1.grid(row=0, column=0, padx=4, sticky=(tk.E,tk.W,tk.S,tk.N))
 frame_list1.columnconfigure(0, weight=1)
-frame_list1.rowconfigure(2, weight=1)
+frame_list1.rowconfigure(3, weight=1)
 
 frame_top1 = ttk.Frame(frame_list1, padding=(0,4,0,3))
 frame_top1.grid(row=0, column=0, columnspan=2, sticky=(tk.E,tk.W))
@@ -307,19 +342,31 @@ button_start.pack(side=tk.LEFT, padx=2)
 button_stop = ttk.Button(frame_top1, text="Stop", width=6, command=button_stop_clicked, state=tk.DISABLED)
 button_stop.pack(side=tk.LEFT, padx=2)
 
+frame_top11 = ttk.Frame(frame_list1, padding=(0,3,0,3))
+frame_top11.grid(row=1, column=0, columnspan=2, sticky=(tk.E,tk.W))
+
+s_combo = tk.StringVar()
+combo_recursive = ttk.Combobox(frame_top11, values=["No child", "Children", "Recursive"], textvariable=s_combo, state="readonly", width=9)
+combo_recursive.current(0)
+combo_recursive.pack(side=tk.LEFT, padx=4)
+
+i_check = tk.IntVar(value=0)
+check_repair = ttk.Checkbutton(frame_top11, text="Repair", variable=i_check)
+check_repair.pack(side=tk.LEFT, padx=10)
+
 label_head1 = ttk.Label(frame_list1, text='? sets in a folder')
-label_head1.grid(row=1, column=0, columnspan=2)
+label_head1.grid(row=2, column=0, columnspan=2)
 
 s_list1 = tk.StringVar()
 listbox_list1 = tk.Listbox(frame_list1, listvariable=s_list1, activestyle='none')
-listbox_list1.grid(row=2, column=0, sticky=(tk.E,tk.W,tk.S,tk.N))
+listbox_list1.grid(row=3, column=0, sticky=(tk.E,tk.W,tk.S,tk.N))
 
 scrollbar_list1 = ttk.Scrollbar(frame_list1, orient=tk.VERTICAL, command=listbox_list1.yview)
-scrollbar_list1.grid(row=2, column=1, sticky=(tk.N, tk.S))
+scrollbar_list1.grid(row=3, column=1, sticky=(tk.N, tk.S))
 listbox_list1["yscrollcommand"] = scrollbar_list1.set
 
 xscrollbar_list1 = ttk.Scrollbar(frame_list1, orient=tk.HORIZONTAL, command=listbox_list1.xview)
-xscrollbar_list1.grid(row=3, column=0, sticky=(tk.E, tk.W))
+xscrollbar_list1.grid(row=4, column=0, sticky=(tk.E, tk.W))
 listbox_list1["xscrollcommand"] = xscrollbar_list1.set
 
 # List of bad files

source/ResourceUI/0412/0412.vcxproj

@@ -0,0 +1,62 @@
+<?xml version="1.0" encoding="utf-8"?>
+<Project DefaultTargets="Build" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
+  <ItemGroup Label="ProjectConfigurations">
+    <ProjectConfiguration Include="Release|Win32">
+      <Configuration>Release</Configuration>
+      <Platform>Win32</Platform>
+    </ProjectConfiguration>
+  </ItemGroup>
+  <PropertyGroup Label="Globals">
+    <VCProjectVersion>17.0</VCProjectVersion>
+    <Keyword>Win32Proj</Keyword>
+    <ProjectGuid>{e8d64ddc-9687-4d16-b2a6-224cab82f8b3}</ProjectGuid>
+    <RootNamespace>My0412</RootNamespace>
+    <WindowsTargetPlatformVersion>10.0</WindowsTargetPlatformVersion>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
+    <ConfigurationType>DynamicLibrary</ConfigurationType>
+    <PlatformToolset>v143</PlatformToolset>
+    <CharacterSet>Unicode</CharacterSet>
+    <WholeProgramOptimization>false</WholeProgramOptimization>
+  </PropertyGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
+  <ImportGroup Label="ExtensionSettings">
+  </ImportGroup>
+  <ImportGroup Label="Shared">
+  </ImportGroup>
+  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
+  </ImportGroup>
+  <PropertyGroup Label="UserMacros" />
+  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <GenerateManifest>false</GenerateManifest>
+  </PropertyGroup>
+  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
+    <ClCompile>
+      <IntrinsicFunctions>true</IntrinsicFunctions>
+      <WarningLevel>Level3</WarningLevel>
+      <FunctionLevelLinking>true</FunctionLevelLinking>
+      <SDLCheck>true</SDLCheck>
+      <PreprocessorDefinitions>WIN32;NDEBUG;MY0412_EXPORTS;_WINDOWS;_USRDLL;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <ConformanceMode>true</ConformanceMode>
+    </ClCompile>
+    <Link>
+      <GenerateDebugInformation>false</GenerateDebugInformation>
+      <SubSystem>Windows</SubSystem>
+      <OptimizeReferences>true</OptimizeReferences>
+      <EnableCOMDATFolding>true</EnableCOMDATFolding>
+      <NoEntryPoint>true</NoEntryPoint>
+      <SetChecksum>true</SetChecksum>
+    </Link>
+  </ItemDefinitionGroup>
+  <ItemGroup>
+    <ClInclude Include="resource.h" />
+  </ItemGroup>
+  <ItemGroup>
+    <ResourceCompile Include="Res0412.rc" />
+  </ItemGroup>
+  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
+  <ImportGroup Label="ExtensionTargets">
+  </ImportGroup>
+</Project>

source/ResourceUI/MultiParUI.sln

@@ -23,52 +23,101 @@ Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "3009", "3009\3009.vcxproj",
 EndProject
 Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "0c0a", "0c0a\0c0a.vcxproj", "{7BC2D258-305F-4BAE-8249-66A4711CB761}"
 EndProject
+Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "0412", "0412\0412.vcxproj", "{E8D64DDC-9687-4D16-B2A6-224CAB82F8B3}"
+EndProject
 Global
 	GlobalSection(SolutionConfigurationPlatforms) = preSolution
 		Debug|Win32 = Debug|Win32
+		Debug|x64 = Debug|x64
 		Release|Win32 = Release|Win32
+		Release|x64 = Release|x64
 	EndGlobalSection
 	GlobalSection(ProjectConfigurationPlatforms) = postSolution
 		{058F9846-C6FE-4BA6-A8B3-F941711876E5}.Debug|Win32.ActiveCfg = Release|Win32
 		{058F9846-C6FE-4BA6-A8B3-F941711876E5}.Debug|Win32.Build.0 = Release|Win32
+		{058F9846-C6FE-4BA6-A8B3-F941711876E5}.Debug|x64.ActiveCfg = Release|Win32
+		{058F9846-C6FE-4BA6-A8B3-F941711876E5}.Debug|x64.Build.0 = Release|Win32
 		{058F9846-C6FE-4BA6-A8B3-F941711876E5}.Release|Win32.ActiveCfg = Release|Win32
 		{058F9846-C6FE-4BA6-A8B3-F941711876E5}.Release|Win32.Build.0 = Release|Win32
+		{058F9846-C6FE-4BA6-A8B3-F941711876E5}.Release|x64.ActiveCfg = Release|Win32
+		{058F9846-C6FE-4BA6-A8B3-F941711876E5}.Release|x64.Build.0 = Release|Win32
 		{A6B4CE94-B0F5-46C2-8D65-2ACBA4444B80}.Debug|Win32.ActiveCfg = Release|Win32
 		{A6B4CE94-B0F5-46C2-8D65-2ACBA4444B80}.Debug|Win32.Build.0 = Release|Win32
+		{A6B4CE94-B0F5-46C2-8D65-2ACBA4444B80}.Debug|x64.ActiveCfg = Release|Win32
+		{A6B4CE94-B0F5-46C2-8D65-2ACBA4444B80}.Debug|x64.Build.0 = Release|Win32
 		{A6B4CE94-B0F5-46C2-8D65-2ACBA4444B80}.Release|Win32.ActiveCfg = Release|Win32
 		{A6B4CE94-B0F5-46C2-8D65-2ACBA4444B80}.Release|Win32.Build.0 = Release|Win32
+		{A6B4CE94-B0F5-46C2-8D65-2ACBA4444B80}.Release|x64.ActiveCfg = Release|Win32
+		{A6B4CE94-B0F5-46C2-8D65-2ACBA4444B80}.Release|x64.Build.0 = Release|Win32
 		{8BE95D56-2B92-40CD-A8D9-506BA9AFD213}.Debug|Win32.ActiveCfg = Release|Win32
 		{8BE95D56-2B92-40CD-A8D9-506BA9AFD213}.Debug|Win32.Build.0 = Release|Win32
+		{8BE95D56-2B92-40CD-A8D9-506BA9AFD213}.Debug|x64.ActiveCfg = Release|Win32
+		{8BE95D56-2B92-40CD-A8D9-506BA9AFD213}.Debug|x64.Build.0 = Release|Win32
 		{8BE95D56-2B92-40CD-A8D9-506BA9AFD213}.Release|Win32.ActiveCfg = Release|Win32
 		{8BE95D56-2B92-40CD-A8D9-506BA9AFD213}.Release|Win32.Build.0 = Release|Win32
+		{8BE95D56-2B92-40CD-A8D9-506BA9AFD213}.Release|x64.ActiveCfg = Release|Win32
+		{8BE95D56-2B92-40CD-A8D9-506BA9AFD213}.Release|x64.Build.0 = Release|Win32
 		{386E588F-363C-4763-830F-4738F947B5D9}.Debug|Win32.ActiveCfg = Release|Win32
 		{386E588F-363C-4763-830F-4738F947B5D9}.Debug|Win32.Build.0 = Release|Win32
+		{386E588F-363C-4763-830F-4738F947B5D9}.Debug|x64.ActiveCfg = Release|Win32
+		{386E588F-363C-4763-830F-4738F947B5D9}.Debug|x64.Build.0 = Release|Win32
 		{386E588F-363C-4763-830F-4738F947B5D9}.Release|Win32.ActiveCfg = Release|Win32
 		{386E588F-363C-4763-830F-4738F947B5D9}.Release|Win32.Build.0 = Release|Win32
+		{386E588F-363C-4763-830F-4738F947B5D9}.Release|x64.ActiveCfg = Release|Win32
+		{386E588F-363C-4763-830F-4738F947B5D9}.Release|x64.Build.0 = Release|Win32
 		{9BFFB67E-6EAB-452B-AB40-10C13A894895}.Debug|Win32.ActiveCfg = Release|Win32
 		{9BFFB67E-6EAB-452B-AB40-10C13A894895}.Debug|Win32.Build.0 = Release|Win32
+		{9BFFB67E-6EAB-452B-AB40-10C13A894895}.Debug|x64.ActiveCfg = Release|Win32
+		{9BFFB67E-6EAB-452B-AB40-10C13A894895}.Debug|x64.Build.0 = Release|Win32
 		{9BFFB67E-6EAB-452B-AB40-10C13A894895}.Release|Win32.ActiveCfg = Release|Win32
 		{9BFFB67E-6EAB-452B-AB40-10C13A894895}.Release|Win32.Build.0 = Release|Win32
+		{9BFFB67E-6EAB-452B-AB40-10C13A894895}.Release|x64.ActiveCfg = Release|Win32
+		{9BFFB67E-6EAB-452B-AB40-10C13A894895}.Release|x64.Build.0 = Release|Win32
 		{41E89B6A-B137-4756-8C7A-E4A4309627DB}.Debug|Win32.ActiveCfg = Release|Win32
 		{41E89B6A-B137-4756-8C7A-E4A4309627DB}.Debug|Win32.Build.0 = Release|Win32
+		{41E89B6A-B137-4756-8C7A-E4A4309627DB}.Debug|x64.ActiveCfg = Release|Win32
+		{41E89B6A-B137-4756-8C7A-E4A4309627DB}.Debug|x64.Build.0 = Release|Win32
 		{41E89B6A-B137-4756-8C7A-E4A4309627DB}.Release|Win32.ActiveCfg = Release|Win32
 		{41E89B6A-B137-4756-8C7A-E4A4309627DB}.Release|Win32.Build.0 = Release|Win32
+		{41E89B6A-B137-4756-8C7A-E4A4309627DB}.Release|x64.ActiveCfg = Release|Win32
+		{41E89B6A-B137-4756-8C7A-E4A4309627DB}.Release|x64.Build.0 = Release|Win32
 		{F5774034-4459-41E4-A34C-5694159088C7}.Debug|Win32.ActiveCfg = Release|Win32
 		{F5774034-4459-41E4-A34C-5694159088C7}.Debug|Win32.Build.0 = Release|Win32
+		{F5774034-4459-41E4-A34C-5694159088C7}.Debug|x64.ActiveCfg = Release|Win32
+		{F5774034-4459-41E4-A34C-5694159088C7}.Debug|x64.Build.0 = Release|Win32
 		{F5774034-4459-41E4-A34C-5694159088C7}.Release|Win32.ActiveCfg = Release|Win32
 		{F5774034-4459-41E4-A34C-5694159088C7}.Release|Win32.Build.0 = Release|Win32
+		{F5774034-4459-41E4-A34C-5694159088C7}.Release|x64.ActiveCfg = Release|Win32
+		{F5774034-4459-41E4-A34C-5694159088C7}.Release|x64.Build.0 = Release|Win32
 		{25EB68DE-0594-4D2F-B31F-50375524D04A}.Debug|Win32.ActiveCfg = Release|Win32
 		{25EB68DE-0594-4D2F-B31F-50375524D04A}.Debug|Win32.Build.0 = Release|Win32
+		{25EB68DE-0594-4D2F-B31F-50375524D04A}.Debug|x64.ActiveCfg = Release|Win32
+		{25EB68DE-0594-4D2F-B31F-50375524D04A}.Debug|x64.Build.0 = Release|Win32
 		{25EB68DE-0594-4D2F-B31F-50375524D04A}.Release|Win32.ActiveCfg = Release|Win32
 		{25EB68DE-0594-4D2F-B31F-50375524D04A}.Release|Win32.Build.0 = Release|Win32
+		{25EB68DE-0594-4D2F-B31F-50375524D04A}.Release|x64.ActiveCfg = Release|Win32
+		{25EB68DE-0594-4D2F-B31F-50375524D04A}.Release|x64.Build.0 = Release|Win32
 		{B316EAF0-4D5E-446C-8520-D720ECD1F3AC}.Debug|Win32.ActiveCfg = Release|Win32
 		{B316EAF0-4D5E-446C-8520-D720ECD1F3AC}.Debug|Win32.Build.0 = Release|Win32
+		{B316EAF0-4D5E-446C-8520-D720ECD1F3AC}.Debug|x64.ActiveCfg = Release|Win32
+		{B316EAF0-4D5E-446C-8520-D720ECD1F3AC}.Debug|x64.Build.0 = Release|Win32
 		{B316EAF0-4D5E-446C-8520-D720ECD1F3AC}.Release|Win32.ActiveCfg = Release|Win32
 		{B316EAF0-4D5E-446C-8520-D720ECD1F3AC}.Release|Win32.Build.0 = Release|Win32
+		{B316EAF0-4D5E-446C-8520-D720ECD1F3AC}.Release|x64.ActiveCfg = Release|Win32
+		{B316EAF0-4D5E-446C-8520-D720ECD1F3AC}.Release|x64.Build.0 = Release|Win32
 		{7BC2D258-305F-4BAE-8249-66A4711CB761}.Debug|Win32.ActiveCfg = Release|Win32
 		{7BC2D258-305F-4BAE-8249-66A4711CB761}.Debug|Win32.Build.0 = Release|Win32
+		{7BC2D258-305F-4BAE-8249-66A4711CB761}.Debug|x64.ActiveCfg = Release|Win32
+		{7BC2D258-305F-4BAE-8249-66A4711CB761}.Debug|x64.Build.0 = Release|Win32
 		{7BC2D258-305F-4BAE-8249-66A4711CB761}.Release|Win32.ActiveCfg = Release|Win32
 		{7BC2D258-305F-4BAE-8249-66A4711CB761}.Release|Win32.Build.0 = Release|Win32
+		{7BC2D258-305F-4BAE-8249-66A4711CB761}.Release|x64.ActiveCfg = Release|Win32
+		{7BC2D258-305F-4BAE-8249-66A4711CB761}.Release|x64.Build.0 = Release|Win32
+		{E8D64DDC-9687-4D16-B2A6-224CAB82F8B3}.Debug|Win32.ActiveCfg = Release|Win32
+		{E8D64DDC-9687-4D16-B2A6-224CAB82F8B3}.Debug|x64.ActiveCfg = Release|Win32
+		{E8D64DDC-9687-4D16-B2A6-224CAB82F8B3}.Release|Win32.ActiveCfg = Release|Win32
+		{E8D64DDC-9687-4D16-B2A6-224CAB82F8B3}.Release|Win32.Build.0 = Release|Win32
+		{E8D64DDC-9687-4D16-B2A6-224CAB82F8B3}.Release|x64.ActiveCfg = Release|Win32
 	EndGlobalSection
 	GlobalSection(SolutionProperties) = preSolution
 		HideSolutionNode = FALSE

source/ShellExt/Command_ShlExt.txt

@@ -64,7 +64,7 @@ append something to the section name like "[0x0409ignore]".
 If you erase these strings or lines, default strings are used.
 
  "CreateHelp" and "VerifyHelp" in old version were removed,
-because they are ignored on recent Windows OS like (Vista, 7, 8).
+because they are ignored on recent Windows OS like (Windows 7, 8, 10).
 
  Strings may contain one "&" for keyboard short-cut.
 Each strings should be reasonably short (under 40 characters).

source/ShellExt/ShellExt.cpp

@@ -1,4 +1,4 @@
-// Copyright : 2023-03-24 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : The MIT license
 
 // ShellExt.cpp : DLL アプリケーション用のエントリ ポイントを定義します。
@@ -11,7 +11,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <windows.h>

source/par1j/Command_par1j.txt

@@ -103,6 +103,16 @@ It means that it is impossible to mix "/" and "-" in options of same command lin
 If PAR file's filename (or relative path) is started by "-",
 you need to use "/" for prefix of options to distinguish PAR file.
 
+ When a option has arbitary number or text,
+don't put space between the option and following part.
+For example, setting "/rr1" is valid, but "/rr 1" is wrong.
+Setting "/dPath" is valid, but "/d Path" is wrong.
+
+ To distinguish options from filenames, order of command-line arguments is important.
+Put all options before <par file>.
+Put <par file> before all [input files].
+
+
  /f, /fu :
  Set this, if you use file-list for file name or file path.
 The file-list is a text file, which contains filenames in each line.
@@ -169,8 +179,10 @@ If you include space in comment, use "" to specify where is end.
 use this setting to set the path as base-directory.
 This is either absolute path or relative path from current directory.
 If you include space in path, use "" to specify where is end.
-Note, setting "/dC:" is same as setting current directory of C-drive.
-If you want to set direct root of C-drive, you must set "/dC:/".
+As you may omit the last "/" in path of a folder,
+both "/dC:/data/" and "/dC:/data" are valid.
+Note, setting "/dD:" is same as setting current directory of D-drive.
+If you want to set direct root of D-drive, you must set "/dD:/".
 
  for example, /dC:/data/ , /d"C:/Program Files/" , /d../data/
 

source/par1j/ReadMe1_English.txt

@@ -18,7 +18,7 @@ Parchive 1.0 client by Yutaka Sawada
  You cannot specify Non-Recovery-Set files. (Empty files are set.)
  Number of recovery files are limited to 99.
 
- This requires a PC of Windows Vista or later (Windows 7, 8, 10, 11).
+ This requires a PC of Windows 7 or later (Windows 8, 10, 11).
  This is developed with Visual Studio 2022 on Windows 10.
 
 

source/par1j/ReadMe1_Japanese.txt

@@ -61,7 +61,7 @@ WinRAR
 <EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>ƃp<C683><70><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>̍<EFBFBD><CC8D>v<EFBFBD><76> 256<35>‚܂łł<C582><C582>B
 <EFBFBD><EFBFBD><EFBFBD>J<EFBFBD>o<EFBFBD><EFBFBD><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>͍ő<CD8D><C591><EFBFBD> 99<39>‚܂łł<C582><C582>B
 
-<EFBFBD>@<40><><EFBFBD><EFBFBD><EFBFBD>‹<EFBFBD><C28B><EFBFBD> Windows Vista <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȍ~ (Windows 7, 8, 10, 11) <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>p<EFBFBD>\<5C>R<EFBFBD><52><EFBFBD>ł<EFBFBD><C582>B
+<EFBFBD>@<40><><EFBFBD><EFBFBD><EFBFBD>‹<EFBFBD><C28B><EFBFBD> Windows 7 <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȍ~ (Windows 8, 10, 11) <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>p<EFBFBD>\<5C>R<EFBFBD><52><EFBFBD>ł<EFBFBD><C582>B
 Win32 API <20><> VC <20><><EFBFBD><EFBFBD><EFBFBD>^<5E>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>g<EFBFBD><67><EFBFBD>Ă<EFBFBD><C482>̂ŁA
 <EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>R<EFBFBD>[<5B>h<EFBFBD><68><EFBFBD>R<EFBFBD><52><EFBFBD>p<EFBFBD>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ɂ<EFBFBD> Visual Studio 2022 <20>ȍ~<7E><><EFBFBD>K<EFBFBD>v<EFBFBD>ł<EFBFBD><C582>B
 

source/par1j/common1.c

@@ -1,5 +1,5 @@
 // common.c
-// Copyright : 2023-03-13 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <conio.h>

source/par1j/ini.c

@@ -1,5 +1,5 @@
 // ini.c
-// Copyright : 2022-01-15 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <malloc.h>

source/par1j/par1.c

@@ -1,5 +1,5 @@
 // par1.c
-// Copyright : 2022-02-16 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <stdio.h>

source/par1j/par1_cmd.c

@@ -1,5 +1,5 @@
 // par1_cmd.c
-// Copyright : 2023-03-14 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <stdio.h>

source/par1j/version.h

@@ -1,2 +1,2 @@
-#define FILE_VERSION "1.3.2.8"	// ファイルのバージョン番号
-#define PRODUCT_VERSION 0x132	// クライアントのバージョン番号
+#define FILE_VERSION "1.3.3.5"	// ファイルのバージョン番号
+#define PRODUCT_VERSION 0x133	// クライアントのバージョン番号

source/par2j/Command_par2j.txt

@@ -1,4 +1,4 @@
-[ par2j.exe - version 1.3.3.1 or later ]
+[ par2j.exe - version 1.3.3.5 or later ]
 
 Type "par2j.exe" to see version, test integrity, and show usage below.
 
@@ -148,6 +148,16 @@ It means that it is impossible to mix "/" and "-" in options of same command lin
 If PAR file's filename (or relative path) is started by "-",
 you need to use "/" for prefix of options to distinguish PAR file.
 
+ When a option has arbitary number or text,
+don't put space between the option and following part.
+For example, setting "/rr10" is valid, but "/rr 10" is wrong.
+Setting "/dPath" is valid, but "/d Path" is wrong.
+
+ To distinguish options from filenames, order of command-line arguments is important.
+Put all options before <par file>.
+Put <par file> before all [input files] or [external files].
+
+
  /f, /fu :
  Set this, if you use file-list for file name or file path.
 When this option is set, [input files] or [external files] is path of file-list.
@@ -302,7 +312,7 @@ Because it may be difficult to predict recovery file size,
 use trial command to see how is the size.
 When it's "/lr" or "/lr0", block count of the largest input file is used.
 (If the max file is 1000 KB and block size is 10 KB, /lr is same as /lr100.)
-When split size "/ls" is set with "/lr0",
+When both "/lr0" and "/ls(split size)" are set,
 block count of the splited file size is used instead of the largest input file.
 
  for example,  /lr , /lr10 , /lr2000
@@ -354,6 +364,13 @@ This is possible, only when there is one input file of .ZIP or .7z format.
 When recovery files are created in a different directory,
 the protected archive file is made in the directory.
 
+ /ls2 has a special feature to set limit size of recovery files directly.
+When both "/ls2" and "/lr(limit size)" are set,
+setting number of /lr is recognizned as file size instead of number of blocks.
+In this usage, setting limit size must be less than 2 GB.
+If you want max 100 MB each file size, set like these; /ls2 /lr102400000
+Because the calculation ignores packets, you may set a bit smaller size.
+
  for example,  /ls10000 , /ls1048576, /ls1
 
  /lc :
@@ -369,14 +386,41 @@ The format is "/lc#" (# is from 1 to 32 as the number of using threads).
   0: It uses the number of physical Cores.
 255: It tries to use more threads than number of physical Cores.
 
- You may set additional combinations;
-+1024 to disable CLMUL (and use slower SSSE3 code), 
-+2048 to disable JIT (for SSE2), 
-+4096 to disable SSSE3,
-+8192 to disable AVX2,
-+256 or +512 (slower device) to enable GPU acceleration.
+ You may set additional combinations for CPU feature;
++1024 to disable CLMUL (and use slower SSSE3 code)
++2048 to disable JIT (for SSE2)
++4096 to disable SSSE3
++8192 to disable AVX2
 
- for example,  /lc1 to use single Core, /lc45 to use half Cores and GPU
+ You may set additional combinations for GPU control;
++256 or +512 (slower device) to enable GPU acceleration
++65536 for classic method
++131072 for 16-byte memory access
++262144 for 4-byte memory access and calculate 2 blocks at once
++524288 for 16-byte memory access and calculate 2 blocks at once
++1048576 for CL_MEM_COPY_HOST_PTR or +2097152 for CL_MEM_USE_HOST_PTR
+(When you set exclusive bits, larger value will be used.)
+
+ for example,  /lc1 to use single Core, /lc508 to use half Cores and GPU
+
+ /lcb :
+ This is for Cache Blocking. (CPU cache optimization)
+By default, this value is set-associative size of CPU L2 cache.
+Maximum value is 32767. It will be multipled by 64 KB.
+To disable cache optimization, set "/lcb0".
+
+ /lcs :
+ This is for large buffer size at calculating multiple files' hashes.
+By default, this value is set-associative size of CPU L3 cache.
+Maximum value is 65535. It will be multipled by 64 KB.
+When you set "/lcs0", large buffer size will become 2 MB by default.
+
+ /lcm :
+ This is for max number of chunks to calculate at once. (CPU shared cache optimization)
+By default, this value may be rate of L3 cache size / L2 cache size.
+The value may be changed by some factors in experimental results.
+Maximum value is 32768. Lower values than 8 will be same as 32768.
+To disable cache optimization, set "/lcm0".
 
  /m :
  Set this, if you want to set memory usage.
@@ -447,8 +491,10 @@ If you include space in comment, use "" to specify where is end.
 use this setting to set the path as base-directory.
 This is either absolute path or relative path from current directory.
 If you include space in path, use "" to specify where is end.
-Note, setting "/dC:" is same as setting current directory of C-drive.
-If you want to set direct root of C-drive, you must set "/dC:/".
+As you may omit the last "/" in path of a folder,
+both "/dC:/data/" and "/dC:/data" are valid.
+Note, setting "/dD:" is same as setting current directory of D-drive.
+If you want to set direct root of D-drive, you must set "/dD:/".
 
  for example, /dC:/data/ , /d"C:/Program Files/" , /d../data/
 

source/par2j/ReadMe2_English.txt

@@ -31,7 +31,7 @@ Parchive 2.0 client by Yutaka Sawada
  Verifying file use same memory as the file or double of block size.
  Recovering block use more memory as the number of block is many.
 
- This requires a PC of Windows Vista or later (Windows 7, 8, 10, 11).
+ This requires a PC of Windows 7 or later (Windows 8, 10, 11).
  This is developed with Visual Studio 2022 on Windows 10.
 
 

source/par2j/ReadMe2_Japanese.txt

@@ -1,140 +1,140 @@
+
+PAR 2.0 の修復用データを作ってファイルの破損や消失に備えます
 
-PAR 2.0 <20>̏C<CC8F><43><EFBFBD>p<EFBFBD>f<EFBFBD>[<5B>^<5E><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ăt<C483>@<40>C<EFBFBD><43><EFBFBD>̔j<CC94><6A><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ɔ<EFBFBD><C994><EFBFBD><EFBFBD>܂<EFBFBD>
-
-<EFBFBD>V<EFBFBD>c <20>L <20>ɂ<EFBFBD><C982><EFBFBD> Parchive 2.0 <20>N<EFBFBD><4E><EFBFBD>C<EFBFBD>A<EFBFBD><41><EFBFBD>g
+澤田 豊 による Parchive 2.0 クライアント
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
-[ <EFBFBD>ȒP<EFBFBD>ȉ<EFBFBD><EFBFBD><EFBFBD> ]
+[ 簡単な解説 ]
 
-<EFBFBD>@<40><><EFBFBD><EFBFBD><EFBFBD>̃\<5C>[<5B>X<EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>烊<EFBFBD>J<EFBFBD>o<EFBFBD><6F><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>쐬<EFBFBD><EC90AC><EFBFBD>āA
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̃t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>̈ꕔ<CC88><EA9594><EFBFBD>j<EFBFBD><6A><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><C482>A
-<EFBFBD>c<EFBFBD><EFBFBD><EFBFBD>ꂽ<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>ƃ<EFBFBD><C683>J<EFBFBD>o<EFBFBD><6F><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>A
-<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ׂ̃\<5C>t<EFBFBD>g<EFBFBD>ł<EFBFBD><C582>B
-PAR 1.0 <EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>P<EFBFBD>ʂŏC<C58F><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂ɔ<CC82><C994>ׂāA
-PAR 2.0 <EFBFBD>̓t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>𕡐<EFBFBD><F095A190>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>ɕ<EFBFBD><C995><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><C482>̃u<CC83><75><EFBFBD>b<EFBFBD>N<EFBFBD>P<EFBFBD>ʂŏC<C58F><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂ŁA
-<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>̈ꕔ<CC88><EA9594><EFBFBD>j<EFBFBD><6A><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><C482>j<EFBFBD><6A><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȊO<C88A>̃u<CC83><75><EFBFBD>b<EFBFBD>N<EFBFBD>͖<EFBFBD><CD96><EFBFBD><EFBFBD>Ŏ<EFBFBD><C58E><EFBFBD><EFBFBD>o<EFBFBD><6F><EFBFBD><EFBFBD><EFBFBD>Ƃ<EFBFBD><C682>ł<EFBFBD><C582>܂<EFBFBD><DC82>B
-<EFBFBD><EFBFBD><EFBFBD>̂<EFBFBD><EFBFBD>߁A<EFBFBD><EFBFBD><EFBFBD>蕡<EFBFBD>G<EFBFBD>ȃG<EFBFBD><EFBFBD><EFBFBD>[<5B>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882>璷<EFBFBD><E792B7><EFBFBD>őΏ<C591><CE8F><EFBFBD><EFBFBD>邱<EFBFBD>Ƃ<EFBFBD><C682>ł<EFBFBD><C582>܂<EFBFBD><DC82>B
+　複数のソース・ファイルからリカバリ・ファイルを作成して、
+それらのファイルの一部が破損したり消失しても、
+残されたソース・ファイルとリカバリ・ファイルから、
+ファイルを修復する為のソフトです。
+PAR 1.0 がファイル単位で修復するのに比べて、
+PAR 2.0 はファイルを複数ブロックに分割してそのブロック単位で修復するので、
+ファイルの一部が破損しても破損部分以外のブロックは無傷で取り出すことができます。
+そのため、より複雑なエラーにも少ない冗長性で対処することができます。
 
-<EFBFBD>@<40><><EFBFBD>J<EFBFBD>o<EFBFBD><6F><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>X<C281>ɓƗ<C993><C697><EFBFBD><EFBFBD><EFBFBD><EFBFBD>p<EFBFBD>P<EFBFBD>b<EFBFBD>g<EFBFBD>ō\<5C><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>\<5C><><EFBFBD>Ȃ̂ŁA
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>I<EFBFBD>Ȕj<EFBFBD><EFBFBD><EFBFBD>ɋ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>āA<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>L<EFBFBD><EFBFBD><EFBFBD>Ɏg<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ƃ<EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
-<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>E<EFBFBD>T<EFBFBD>C<EFBFBD>Y<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>قNjǒn<EFBFBD>I<EFBFBD>ȃG<EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD><EFBFBD><EFBFBD>ɉe<C989><65><EFBFBD><EFBFBD><EFBFBD>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882>A
-<EFBFBD>C<EFBFBD><EFBFBD><EFBFBD>ɕK<EFBFBD>v<EFBFBD>ȃ<EFBFBD><EFBFBD>J<EFBFBD>o<EFBFBD><EFBFBD><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882>čς݂܂<DD82><DC82>B
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>A<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ꂽ<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>𕜌<EFBFBD><F0959C8C><EFBFBD><EFBFBD>邽<EFBFBD>߂ɂ<DF82>
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̃p<EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD><EFBFBD><EFBFBD>K<EFBFBD>v<EFBFBD>ɂȂ<EFBFBD><EFBFBD>Ƃ<EFBFBD><EFBFBD><EFBFBD> PAR <20>̊<EFBFBD><CC8A>{<7B><><EFBFBD><EFBFBD><EFBFBD>͓<EFBFBD><CD93><EFBFBD><EFBFBD>ł<EFBFBD><C582>B
-<EFBFBD>쐬<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>p<EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>̐<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʂŁA
-<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>E<EFBFBD>T<EFBFBD>C<EFBFBD>Y<EFBFBD>͂<EFBFBD><EFBFBD>̕<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>P<EFBFBD>ʂɂȂ<EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
+　リカバリ・ファイルも個々に独立したパケットで構成される構造なので、
+部分的な破損に強くて、より有効に使うことができます。
+ブロック・サイズが小さいほど局地的なエラーが他に影響しにくくなり、
+修復に必要なリカバリ・ファイルも少なくて済みます。
+しかし、失われたソース・ブロックを復元するためには
+同数のパリティ・ブロックが必要になるという PAR の基本原理は同じです。
+作成するパリティ・ブロックの数が復元できる量で、
+ブロック・サイズはその復元する単位になります。
 
-<EFBFBD>@PAR 2.0 <EFBFBD>̓t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>Ƀf<C983>B<EFBFBD><42><EFBFBD>N<EFBFBD>g<EFBFBD><67><EFBFBD><EFBFBD><EFBFBD>܂ނ<DC82><DE82>Ƃ<EFBFBD><C682>ł<EFBFBD><C582><EFBFBD><EFBFBD>̂ŁA
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̃t<EFBFBD>H<EFBFBD><EFBFBD><EFBFBD>_<EFBFBD>ɎU<EFBFBD><EFBFBD><EFBFBD>΂<EFBFBD><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂܂܏C<DC8F><43><EFBFBD><EFBFBD><EFBFBD>邱<EFBFBD>Ƃ<EFBFBD><C682>ł<EFBFBD><C582><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ƁA
-PAR 1.0 <EFBFBD>ɔ<EFBFBD><EFBFBD>ׂĂ<EFBFBD><EFBFBD>葽<EFBFBD><EFBFBD><EFBFBD>̃t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>_<EFBFBD><5F><EFBFBD>ɏC<C98F><43><EFBFBD>ł<EFBFBD><C582>Ďg<C48E><67><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>͂悭<CD82>Ȃ<EFBFBD><C882>Ă܂<C482><DC82>B
-<EFBFBD><EFBFBD><EFBFBD>̔<EFBFBD><EFBFBD>ʁA<EFBFBD>傫<EFBFBD>ȃt<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>⑽<EFBFBD><E291BD><EFBFBD>̃u<CC83><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ƁA
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD><EFBFBD><EF82AA><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>v<EFBFBD>Z<EFBFBD>ɂ₽<C982><E282BD><EFBFBD>Ǝ<EFBFBD><C68E>Ԃ<EFBFBD><D482><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
-<EFBFBD>@<40>\<5C>I<EFBFBD>ɂ<EFBFBD> PAR 2.0 <20>̕<EFBFBD><CC95><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ʔłł<C582><C582><EFBFBD><EFBFBD>A<EFBFBD>\<5C><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>G<EFBFBD>őΉ<C591><CE89>\<5C>t<EFBFBD>g<EFBFBD><67><EFBFBD>قƂ<D982><C682>ǂȂ<C782><C882>A
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>\<5C>ȃp<C883>\<5C>R<EFBFBD><52><EFBFBD>łȂ<C582><C882>Ǝ<EFBFBD><C68E>p<EFBFBD>I<EFBFBD>ȏ<EFBFBD><C88F><EFBFBD><EFBFBD><EFBFBD><EFBFBD>x<EFBFBD>𓾂<EFBFBD><F093BE82><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82><EFBFBD><EFBFBD>B
-<EFBFBD><EFBFBD><EFBFBD>ɔ<EFBFBD><EFBFBD>͂ȃp<EFBFBD>\<5C>R<EFBFBD><52><EFBFBD>ł̓V<CD83><56><EFBFBD>v<EFBFBD><76><EFBFBD>Ōy<C58C><79><EFBFBD><EFBFBD> PAR 1.0 <20>̕<EFBFBD><CC95><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă邱<C482>Ƃ<EFBFBD><C682><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂ŁA
-<EFBFBD>‹<EFBFBD><EFBFBD><EFBFBD><EFBFBD>p<EFBFBD>r<EFBFBD>ɉ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ďg<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>B
+　PAR 2.0 はファイル名にディレクトリも含むことができるので、
+複数のフォルダに散らばったファイルをそのまま修復することができたりと、
+PAR 1.0 に比べてより多くのファイルを柔軟に修復できて使い勝手はよくなってます。
+その反面、大きなファイルや多数のブロックを処理すると、
+メモリー消費が増えたり計算にやたらと時間がかかります。
+機能的には PAR 2.0 の方が上位版ですが、構造が複雑で対応ソフトがほとんどなく、
+高性能なパソコンでないと実用的な処理速度を得られないかもしれません。
+特に非力なパソコンではシンプルで軽快な PAR 1.0 の方が向いてることもあるので、
+環境や用途に応じて使い分けてください。
 
 
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
-[ <EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>d<EFBFBD>l<EFBFBD>Ȃ<EFBFBD> ]
+[ 特徴や仕様など ]
 
-<EFBFBD>@Parity Archive 2.0 <EFBFBD>̖{<7B><> par2cmdline <EFBFBD><EFBFBD> QuickPar <EFBFBD>ɔ<EFBFBD><EFBFBD>ׂāA
-<EFBFBD><EFBFBD><EFBFBD>J<EFBFBD>o<EFBFBD><EFBFBD><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>ɃR<C983><52><EFBFBD><EFBFBD><EFBFBD>g<EFBFBD><67><EFBFBD>t<EFBFBD><74><EFBFBD>邱<EFBFBD>Ƃ<EFBFBD><C682>ł<EFBFBD><C582>܂<EFBFBD><DC82>B
-<EFBFBD><EFBFBD><EFBFBD>j<EFBFBD>R<EFBFBD>[<5B>h<EFBFBD><68><EFBFBD>ǂݏ<C782><DD8F><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂Ń}<7D><><EFBFBD>`<60>o<EFBFBD>C<EFBFBD>g<EFBFBD><67><EFBFBD><EFBFBD><EFBFBD>ɂ<EFBFBD><C982>Ή<EFBFBD><CE89><EFBFBD><EFBFBD>Ă܂<C482><DC82>B
-<EFBFBD>T<EFBFBD>C<EFBFBD>Y<EFBFBD><EFBFBD> 0<>̋<EFBFBD><CC8B>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>ɂ<EFBFBD><C982>Ή<EFBFBD><CE89><EFBFBD><EFBFBD>Ă܂<C482><DC82>B
-par2cmdline <EFBFBD>̓T<EFBFBD>u<EFBFBD>E<EFBFBD>f<EFBFBD>B<EFBFBD><EFBFBD><EFBFBD>N<EFBFBD>g<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>Ƃ<EFBFBD><C682>ĔF<C494><46><EFBFBD><EFBFBD><EFBFBD>A
-QuickPar <EFBFBD>͖<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD><EFBFBD><EFBFBD>A<EFBFBD><EFBFBD><EFBFBD>̃A<EFBFBD>v<EFBFBD><EFBFBD><EFBFBD>P<EFBFBD>[<5B>V<EFBFBD><56><EFBFBD><EFBFBD><EFBFBD><EFBFBD>
-<EFBFBD>t<EFBFBD>H<EFBFBD><EFBFBD><EFBFBD>_<EFBFBD><EFBFBD><EFBFBD>T<EFBFBD>u<EFBFBD>E<EFBFBD>f<EFBFBD>B<EFBFBD><EFBFBD><EFBFBD>N<EFBFBD>g<EFBFBD><EFBFBD><EFBFBD>Ȃǂ̃f<EFBFBD>B<EFBFBD><EFBFBD><EFBFBD>N<EFBFBD>g<EFBFBD><EFBFBD><EFBFBD>\<5C><><EFBFBD>𕜌<EFBFBD><F0959C8C><EFBFBD><EFBFBD>邱<EFBFBD>Ƃ<EFBFBD><C682>ł<EFBFBD><C582>܂<EFBFBD><DC82>B
+　Parity Archive 2.0 の本家 par2cmdline や QuickPar に比べて、
+リカバリ・ファイルにコメントを付けることができます。
+ユニコードを読み書きするのでマルチバイト文字にも対応してます。
+サイズが 0の空ファイルにも対応してます。
+par2cmdline はサブ・ディレクトリをファイル名として認識し、
+QuickPar は無視しますが、このアプリケーションは
+フォルダやサブ・ディレクトリなどのディレクトリ構造を復元することができます。
 
-<EFBFBD>@QuickPar <EFBFBD>͖<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ɃS<EFBFBD>~<7E><><EFBFBD>t<EFBFBD><74><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ԂŔ<D482><C594>[<5B>ȃu<C883><75><EFBFBD>b<EFBFBD>N<EFBFBD>̒<EFBFBD><CC92>O<EFBFBD>̃u<CC83><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E><EFBFBD>j<EFBFBD><6A><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>
-<EFBFBD><EFBFBD><EFBFBD>̃u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD><EFBFBD><EFBFBD>F<EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD>Ȃ<EFBFBD><EFBFBD>Ȃ<EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD><EFBFBD><EFBFBD>A<EFBFBD><EFBFBD><EFBFBD>̃A<EFBFBD>v<EFBFBD><EFBFBD><EFBFBD>P<EFBFBD>[<5B>V<EFBFBD><56><EFBFBD><EFBFBD><EFBFBD>ł͂<C582><CD82><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ɣ<EFBFBD><C694>肵<EFBFBD>܂<EFBFBD><DC82>B
-<EFBFBD>܂<EFBFBD><EFBFBD>A<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>E<EFBFBD>T<EFBFBD>C<EFBFBD>Y<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ē<EFBFBD><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>œ<EFBFBD><C593>e<EFBFBD><65><EFBFBD>d<EFBFBD><64><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><C482>ƁA
-<EFBFBD>j<EFBFBD><EFBFBD><EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>ȍ~<7E>Ɍ<EFBFBD><C98C>̃u<CC83><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E><EFBFBD><EFBFBD><EFBFBD>Ɍ<EFBFBD><C98C>o<EFBFBD><6F><EFBFBD>錻<EFBFBD>ۂ<EFBFBD><DB82>N<EFBFBD><4E><EFBFBD>āA
-<EFBFBD><EFBFBD><EFBFBD>̌<EFBFBD><EFBFBD>Ŗ{<7B><><EFBFBD>̏<EFBFBD><CC8F>Ԃ̃u<CC83><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E><EFBFBD>F<EFBFBD><46><EFBFBD>ł<EFBFBD><C582>Ȃ<EFBFBD><C882>Ȃ<EFBFBD><C882>Ă<EFBFBD><C482>܂<EFBFBD><DC82><EFBFBD><EFBFBD>肪<EFBFBD><E882AA><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82><EFBFBD><EFBFBD>A
-<EFBFBD><EFBFBD><EFBFBD>̃A<EFBFBD>v<EFBFBD><EFBFBD><EFBFBD>P<EFBFBD>[<5B>V<EFBFBD><56><EFBFBD><EFBFBD><EFBFBD>ł͌<C582><CD8C>o<EFBFBD><6F><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ԂɊւ<C98A><D682>炸<EFBFBD><E782B8><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ƔF<C694><46><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
-<EFBFBD>j<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̌<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> 1<>o<EFBFBD>C<EFBFBD>g<EFBFBD><67><EFBFBD>ˆ<EFBFBD><C288>v<EFBFBD><76><EFBFBD><EFBFBD><EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>𒲂ׂĂ<D782><C482><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ߎ<EFBFBD><DF8E>Ԃ͂<D482><CD82><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82><EFBFBD><EFBFBD>A
-<EFBFBD>j<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>葽<EFBFBD><E891BD><EFBFBD>̗<EFBFBD><CC97>p<EFBFBD>”\<5C>ȃu<C883><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E><EFBFBD><EFBFBD><EFBFBD>o<EFBFBD><6F><EFBFBD>邩<EFBFBD><E982A9><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82><EFBFBD><EFBFBD>B
+　QuickPar は末尾にゴミが付いた状態で半端なブロックの直前のブロックが破損すると
+そのブロックを認識できなくなりますが、このアプリケーションではきちんと判定します。
+また、ブロック・サイズが小さくて同一ファイル内で内容が重複してると、
+破損ブロック以降に後のブロックを先に検出する現象が起きて、
+その後で本来の順番のブロックを認識できなくなってしまう問題がありますが、
+このアプリケーションでは検出した順番に関わらずきちんと認識します。
+破損部分の検査で 1バイトずつ一致するブロックを調べていくため時間はかかりますが、
+破損したファイルからより多くの利用可能なブロックを検出するかもしれません。
 
-<EFBFBD>@QuickPar <EFBFBD>͔j<EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ۂɁA<C981><41><EFBFBD><EFBFBD><EFBFBD>̔j<CC94><6A><EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>̈ʒu<CA92>ƏC<C68F><43><EFBFBD>Ɏg<C98E><67>
-<EFBFBD>p<EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>̑g<EFBFBD>ݍ<EFBFBD><EFBFBD>킹<EFBFBD>ɂ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ẮA<EFBFBD>p<EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>̐<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><EFBFBD>Ă<EFBFBD>
-<EFBFBD>C<EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD>Ȃ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ƃ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>́A<EFBFBD>C<EFBFBD><EFBFBD><EFBFBD>Ɏg<EFBFBD><EFBFBD><EFBFBD>p<EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD><EFBFBD>
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ƂŕύX<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ƃ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>߂<EFBFBD><EFBFBD>ǂ<EFBFBD><EFBFBD>ȍ<EFBFBD><EFBFBD>Ƃ<EFBFBD><EFBFBD>K<EFBFBD>v<EFBFBD>ɂȂ<EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
-<EFBFBD><EFBFBD><EFBFBD>̃A<EFBFBD>v<EFBFBD><EFBFBD><EFBFBD>P<EFBFBD>[<5B>V<EFBFBD><56><EFBFBD><EFBFBD><EFBFBD>ł͖<C582><CD96><EFBFBD><EFBFBD>̂<EFBFBD><CC82><EFBFBD><EFBFBD>p<EFBFBD><70><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ď<EFBFBD><C48E><EFBFBD><EFBFBD>I<EFBFBD><49>
-<EFBFBD>ēx<EFBFBD>C<EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD>邩<EFBFBD><EFBFBD><EFBFBD>݂Ă<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂Ŋy<EFBFBD>ł<EFBFBD><EFBFBD>B
+　QuickPar は破損ファイルを修復する際に、特定の破損ブロックの位置と修復に使う
+パリティ・ブロックの組み合わせによっては、パリティ・ブロックの数が足りていても
+修復できないことがあります。そういう時は、修復に使うパリティ・ブロックを
+手作業で変更するというめんどうな作業が必要になります。
+このアプリケーションでは問題のあるパリティ・ブロックを除いて自動的に
+再度修復できるか試みてくれるので楽です。
 
-<EFBFBD>@<40><><EFBFBD><EFBFBD><EFBFBD>ꂽ<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>Ɠ<EFBFBD><C693><EFBFBD><EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>̃u<CC83><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E><EFBFBD><EFBFBD><EFBFBD>݂<EFBFBD><DD82>ė<EFBFBD><C497>p<EFBFBD>ł<EFBFBD><C582>邩<EFBFBD><E982A9>
-<EFBFBD>`<60>F<EFBFBD>b<EFBFBD>N<EFBFBD>T<EFBFBD><54><EFBFBD><EFBFBD><EFBFBD>g<EFBFBD><67><EFBFBD>Č<EFBFBD><C48C><EFBFBD><EFBFBD><EFBFBD><EFBFBD>A<EFBFBD><41><EFBFBD>‚<EFBFBD><C282><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ꍇ<EFBFBD>͒<EFBFBD><CD92>ڃR<DA83>s<EFBFBD>[<5B><><EFBFBD>܂<EFBFBD><DC82>B
-<EFBFBD><EFBFBD><EFBFBD>̂<EFBFBD><EFBFBD>߁A<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>̃\<5C>[<5B>X<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂܂<DC82><DC82>鎞<EFBFBD>́A
-<EFBFBD>C<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̌v<EFBFBD>Z<EFBFBD>ʂ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ƌ<EFBFBD><EFBFBD>ɏ<EFBFBD><EFBFBD>Ȃ<EFBFBD><EFBFBD>p<EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>ł<EFBFBD><EFBFBD>C<EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>I<EFBFBD>ɓ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>̃t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>܂܂<DC82><DC82><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882>AQuickPar <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>C<EFBFBD><43><EFBFBD>ł<EFBFBD><C582><EFBFBD><EFBFBD>”\<5C><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><C582>B
+　失われたソース・ブロックと同じ内容のブロックが存在して利用できるかを
+チェックサムを使って検索し、見つかった場合は直接コピーします。
+そのため、同じ内容のソース・ブロックが複数含まれる時は、
+修復時の計算量が減ると共に少ないパリティ・ブロックでも修復できます。
+部分的に同じ内容のファイルが含まれるなら、QuickPar よりも修復できる可能性が高いです。
 
-<EFBFBD>@<40>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>E<EFBFBD>T<EFBFBD>C<EFBFBD>Y<EFBFBD><59> 4<>o<EFBFBD>C<EFBFBD>g<EFBFBD>̏ꍇ<CC8F>́A
-<EFBFBD>j<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>̃`<60>F<EFBFBD>b<EFBFBD>N<EFBFBD>T<EFBFBD><54><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>e<EFBFBD><65><EFBFBD>t<EFBFBD>Z<EFBFBD><5A><EFBFBD><EFBFBD><EFBFBD>̂ŁA
-<EFBFBD>p<EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD><EFBFBD><EFBFBD>S<EFBFBD><EFBFBD><EFBFBD>g<EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><EFBFBD>ŏC<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>邱<EFBFBD>Ƃ<EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
-<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>E<EFBFBD>T<EFBFBD>C<EFBFBD>Y<EFBFBD><EFBFBD> 8<>o<EFBFBD>C<EFBFBD>g<EFBFBD>̏ꍇ<CC8F>̏<EFBFBD><CC8F><EFBFBD><EFBFBD>́A
-1<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>Z<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂ɉ<EFBFBD><EFBFBD>\<5C><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂Ŏ<CC82><C58E><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><C482>܂<EFBFBD><DC82><EFBFBD><EFBFBD>B
-<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>R<EFBFBD>[<5B>h<EFBFBD><68><EFBFBD>ɂ͂<C982><CD82><EFBFBD><EFBFBD>̂Ŋ<CC82><C58A>]<5D><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> 8<>o<EFBFBD>C<EFBFBD>g<EFBFBD>p<EFBFBD><70><EFBFBD>lj<EFBFBD><C789>ł<EFBFBD><C582>܂<EFBFBD><DC82>B
+　ブロック・サイズが 4バイトの場合は、
+破損したソース・ブロックのチェックサムから内容を逆算するので、
+パリティ・ブロックを全く使わないで修復することができます。
+ブロック・サイズが 8バイトの場合の処理は、
+1ブロックを逆算するのに何十分もかかるので実装していません。
+ソース・コード上にはあるので希望があれば 8バイト用も追加できます。
 
-<EFBFBD>@<40><><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>̍<EFBFBD><CC8D>v<EFBFBD>T<EFBFBD>C<EFBFBD>Y<EFBFBD><59><EFBFBD>󂫃<EFBFBD><F382AB83><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B>ʂ̔<CA82><CC94><EFBFBD><EFBFBD>ȉ<EFBFBD><C889>Ȃ<EFBFBD><C882>΁A
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD> PAR 2.0 <EFBFBD>N<EFBFBD><EFBFBD><EFBFBD>C<EFBFBD>A<EFBFBD><EFBFBD><EFBFBD>g<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ƍ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ɏC<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD>B
-<EFBFBD>}<7D><><EFBFBD>`<60>E<EFBFBD>R<EFBFBD>A CPU <20>ɑΉ<C991><CE89><EFBFBD><EFBFBD>ă}<7D><><EFBFBD>`<60>E<EFBFBD>X<EFBFBD><58><EFBFBD>b<EFBFBD>h<EFBFBD>ōs<C58D>񉉎Z<F189898E><5A>
-<EFBFBD><EFBFBD><EFBFBD>[<5B>h<EFBFBD>E<EFBFBD>\<5C><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̌v<CC8C>Z<EFBFBD><5A><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂ŁACPU <20>R<EFBFBD>A<EFBFBD><41> 2<>ˆȏ゠<C88F><E382A0><EFBFBD>ƍX<C68D>ɑ<EFBFBD><C991><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882>܂<EFBFBD><DC82>B
-<EFBFBD>V<EFBFBD><EFBFBD><EFBFBD>O<EFBFBD><EFBFBD><EFBFBD>E<EFBFBD>R<EFBFBD>A<EFBFBD>ł<EFBFBD> QuickPar <EFBFBD><EFBFBD> 2<>{<7B>A
-<EFBFBD>f<EFBFBD><EFBFBD><EFBFBD>A<EFBFBD><EFBFBD><EFBFBD>E<EFBFBD>R<EFBFBD>A<EFBFBD><EFBFBD><EFBFBD><EFBFBD> QuickPar <20><> 3<>{<7B><><EFBFBD>炢<EFBFBD>̑<EFBFBD><CC91>x<EFBFBD><78><EFBFBD>ł܂<C582><DC82>B
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>A<EFBFBD>󂫃<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>傫<EFBFBD><E582AB><EFBFBD>ƁA
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><EFBFBD>ċɒ[<5B>ɏ<EFBFBD><C98F><EFBFBD><EFBFBD><EFBFBD><EFBFBD>x<EFBFBD><78><EFBFBD>ቺ<EFBFBD><E189BA><EFBFBD>܂<EFBFBD><DC82>B
+　扱うファイルの合計サイズが空きメモリー量の半分以下ならば、
+他の PAR 2.0 クライアントよりもずっと高速に修復します。
+マルチ・コア CPU に対応してマルチ・スレッドで行列演算と
+リード・ソロモン符号の計算をするので、CPU コアが 2個以上あると更に速くなります。
+シングル・コアでも QuickPar の 2倍、
+デュアル・コアだと QuickPar の 3倍ぐらいの速度がでます。
+ただし、空きメモリーが少なかったりファイルが大きいと、
+効率が悪くなって極端に処理速度が低下します。
 
-<EFBFBD>@<40><><EFBFBD>{<7B><><EFBFBD>̊J<CC8A><4A><EFBFBD>‹<EFBFBD><C28B>ō<EFBFBD><C58D><EFBFBD><EFBFBD>Ă<EFBFBD><C482>̂ŁA
-<EFBFBD><EFBFBD><EFBFBD>{<7B><><EFBFBD>̃t<CC83>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>f<EFBFBD>B<EFBFBD><42><EFBFBD>N<EFBFBD>g<EFBFBD><67><EFBFBD>ɑΉ<C991><CE89><EFBFBD><EFBFBD>Ă܂<C482><DC82>B
-4GB <EFBFBD>ȏ<EFBFBD><EFBFBD>̃t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>ɂ<EFBFBD><C982>Ή<EFBFBD><CE89><EFBFBD><EFBFBD>Ă܂<C482><DC82><EFBFBD><EFBFBD>A
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>C<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>x<EFBFBD>̓t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>E<EFBFBD>A<EFBFBD>N<EFBFBD>Z<EFBFBD>X<EFBFBD><58><EFBFBD>x<EFBFBD>ɑ傫<C991><E582AB><EFBFBD>ˑ<EFBFBD><CB91><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
+　日本語の開発環境で作ってるので、
+日本語のファイル名やディレクトリに対応してます。
+4GB 以上のファイルにも対応してますが、
+検査や修復速度はファイル・アクセス速度に大きく依存します。
 
-<EFBFBD>@<40>J<EFBFBD><4A><EFBFBD>Ҍ<EFBFBD><D28C><EFBFBD><EFBFBD>ɁA
-<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>R<EFBFBD>[<5B>h<EFBFBD><68><EFBFBD><EFBFBD><EFBFBD>J<EFBFBD><4A><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><C482>̂ŁA<C581>ׂ<EFBFBD><D782><EFBFBD><EFBFBD>d<EFBFBD>l<EFBFBD><6C><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ŕύX<CF8D>ł<EFBFBD><C582>܂<EFBFBD><DC82>B
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ɕs<EFBFBD><EFBFBD><EFBFBD>ȓ_<EFBFBD><EFBFBD><EFBFBD>o<EFBFBD>O<EFBFBD><EFBFBD><EFBFBD>ۂ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><EFBFBD>A<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ŋm<EFBFBD>F<EFBFBD><EFBFBD><EFBFBD>邱<EFBFBD>Ƃ<EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD>Ĉ<EFBFBD><EFBFBD>S<EFBFBD>ł<EFBFBD><EFBFBD>B
-<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>E<EFBFBD><45><EFBFBD>X<EFBFBD>g<EFBFBD>Ŏw<C58E>肷<EFBFBD>邱<EFBFBD>Ƃ<EFBFBD><C682>ł<EFBFBD><C582>܂<EFBFBD><DC82>B
-<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>ƃ<EFBFBD><C683>J<EFBFBD>o<EFBFBD><6F><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>قȂ<D982><C882>ꏊ<EFBFBD>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>Ă<EFBFBD><C482><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><C582>܂<EFBFBD><DC82>B
+　開発者向けに、
+ソース・コードが公開されてるので、細かい仕様を自分で変更できます。
+動作に不明な点やバグっぽい挙動があっても、自分で確認することができて安心です。
+ソース・ファイルをファイル・リストで指定することができます。
+ソース・ファイルとリカバリ・ファイルが異なる場所にあっても処理できます。
 
-<EFBFBD>@<40><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ƃ<EFBFBD><C682>ẮA
-<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>E<EFBFBD>T<EFBFBD>C<EFBFBD>Y<EFBFBD><EFBFBD> 2 GB <20><><EFBFBD><EFBFBD><EFBFBD>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>Ή<EFBFBD><CE89><EFBFBD><EFBFBD>Ă܂<C482><DC82><EFBFBD><EFBFBD>B
-<EFBFBD>p<EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>B<EFBFBD>v<EFBFBD>Z<EFBFBD>Ɋ܂߂Ȃ<EFBFBD><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43> (<28>m<EFBFBD><6D><EFBFBD>E<EFBFBD><45><EFBFBD>J<EFBFBD>o<EFBFBD><6F><EFBFBD>E<EFBFBD>Z<EFBFBD>b<EFBFBD>g) <20>̎w<CC8E><77><EFBFBD>͂ł<CD82><C582>܂<EFBFBD><DC82><EFBFBD><EFBFBD>B
-<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>͍ő<CD8D><C591><EFBFBD> 32768<36>‚܂łł<C582><C582>B(<28>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD> 65536<EFBFBD>‚܂<EFBFBD>)
-<EFBFBD>쐬<EFBFBD>ł<EFBFBD><EFBFBD><EFBFBD><EFBFBD>p<EFBFBD><EFBFBD><EFBFBD>e<EFBFBD>B<EFBFBD>E<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>͍ő<EFBFBD><EFBFBD><EFBFBD> 65535<33>‚܂łł<C582><C582>B
+　制限事項としては、
+ブロック・サイズは 2 GB 未満にしか対応してません。
+パリティ計算に含めないファイル (ノン・リカバリ・セット) の指定はできません。
+ソース・ブロックは最大で 32768個までです。(ファイル数は 65536個まで)
+作成できるパリティ・ブロックは最大で 65535個までです。
 
-<EFBFBD>@<40><><EFBFBD>ӓ_<D393>̓<EFBFBD><CD83><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD><EFBFBD><EFBFBD>ʂ<EFBFBD><CA82>c<EFBFBD><63><EFBFBD>Ȃ<EFBFBD><C882>Ƃł<C682><C582>B
-<EFBFBD><EFBFBD><EFBFBD>J<EFBFBD>o<EFBFBD><EFBFBD><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>ŏC<C58F><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ۂɂ́A<CD81><41><EFBFBD>[<5B>h<EFBFBD>E<EFBFBD>\<5C><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̍s<CC8D><73><EFBFBD><EFBFBD><EFBFBD>ϊ<EFBFBD><CF8A><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ׂ<EFBFBD>
-<EFBFBD>u <20>\<5C>[<5B>X<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E> * <20><><EFBFBD><EFBFBD><EFBFBD>ꂽ<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E> * 2 <20>v<EFBFBD>̃<EFBFBD><CC83><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD>K<EFBFBD>v<EFBFBD>Ƃ<EFBFBD><C682>܂<EFBFBD><DC82>B
-<EFBFBD>Ⴆ<EFBFBD>΁A<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ő<EFBFBD><EFBFBD><EFBFBD> 32768 <20>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>ƁA25% <20>̃u<CC83><75><EFBFBD>b<EFBFBD>N<EFBFBD>𕜌<EFBFBD><F0959C8C><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ɂ<EFBFBD>
-32768 * 8192 * 2 = 500MB <EFBFBD>ȏ<EFBFBD><EFBFBD>̃<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD>K<EFBFBD>v<EFBFBD>Ƃ<EFBFBD><C682>܂<EFBFBD><DC82>B
-<EFBFBD><EFBFBD><EFBFBD>ʂ͏璷<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>傫<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>邱<EFBFBD>Ƃ͖<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>A<EFBFBD>S<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD><EFBFBD><EFBFBD>j<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>邱<EFBFBD>Ƃ<EFBFBD><EFBFBD>܂<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂ŁA
-<EFBFBD><EFBFBD><EFBFBD>ۂɎg<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>郁<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B>͂<EFBFBD><CD82><EFBFBD><EFBFBD>Ə<EFBFBD><C68F>Ȃ<EFBFBD><C882>ł<EFBFBD><C582><EFBFBD><EFBFBD>A<EFBFBD><41><EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882><EFBFBD><EFBFBD>΃G<CE83><47><EFBFBD>[<5B>ɂȂ<C982><C882>܂<EFBFBD><DC82>B
+　注意点はメモリー消費量が膨大なことです。
+リカバリ・ファイルで修復する際には、リード・ソロモン符号の行列を変換する為に
+「 ソース・ブロック数 * 失われたソース・ブロック数 * 2 」のメモリーを必要とします。
+例えば、ブロック数を最大の 32768 にすると、25% のブロックを復元するには
+32768 * 8192 * 2 = 500MB 以上のメモリーを必要とします。
+普通は冗長性を大きくすることは無いし、全ブロックが破損することもまず無いので、
+実際に使われるメモリーはもっと少ないですが、足りなければエラーになります。
 
-<EFBFBD>@<40>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>E<EFBFBD>A<EFBFBD>N<EFBFBD>Z<EFBFBD>X<EFBFBD>p<EFBFBD>o<EFBFBD>b<EFBFBD>t<EFBFBD>@<40>[<5B>́u (<28>\<5C>[<5B>X<EFBFBD>E<EFBFBD>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E> + 2) * <20>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>E<EFBFBD>T<EFBFBD>C<EFBFBD>Y <20>v
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>{<7B>̑傫<CC91><E582AB><EFBFBD>ɂȂ<C982><C882>A<EFBFBD><41><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882>ꍇ<EFBFBD>̓u<CD83><75><EFBFBD>b<EFBFBD>N<EFBFBD><4E><EFBFBD>f<EFBFBD>Љ<EFBFBD><D089><EFBFBD><EFBFBD>ď<EFBFBD><C48F><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
-<EFBFBD>f<EFBFBD>ЃT<EFBFBD>C<EFBFBD>Y<EFBFBD>́u 1KB<4B>`<60>u<EFBFBD><75><EFBFBD>b<EFBFBD>N<EFBFBD>E<EFBFBD>T<EFBFBD>C<EFBFBD>Y<EFBFBD>̔<EFBFBD><CC94><EFBFBD> <20>v<EFBFBD>ł<EFBFBD><C582>B
-<EFBFBD>s<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>i<EFBFBD>[<5B><><EFBFBD><EFBFBD><EFBFBD>ׂ̃<D782><CC83><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD>D<EFBFBD>悵<EFBFBD>āA<C481>󂫃<EFBFBD><F382AB83><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882>ꍇ<EFBFBD>ɂ́A
-<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>E<EFBFBD>A<EFBFBD>N<EFBFBD>Z<EFBFBD>X<EFBFBD>p<EFBFBD>o<EFBFBD>b<EFBFBD>t<EFBFBD>@<40>[<5B><><EFBFBD><EFBFBD><EFBFBD>炵<EFBFBD>܂<EFBFBD><DC82><EFBFBD><EFBFBD>A<EFBFBD><41><EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><C582><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882><EFBFBD><EFBFBD>΃G<CE83><47><EFBFBD>[<5B><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
-<EFBFBD>܂<EFBFBD><EFBFBD>A<EFBFBD>u<EFBFBD><EFBFBD><EFBFBD>b<EFBFBD>N<EFBFBD>̒f<EFBFBD>Љ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ƌɒ[<5B>ɏ<EFBFBD><C98F><EFBFBD><EFBFBD><EFBFBD><EFBFBD>x<EFBFBD><78><EFBFBD>ቺ<EFBFBD><E189BA><EFBFBD>܂<EFBFBD><DC82>B
+　ファイル・アクセス用バッファーは「 (ソース・ブロック数 + 2) * ブロック・サイズ 」
+が基本の大きさになり、メモリーが足りない場合はブロックを断片化して処理します。
+断片サイズは「 1KB～ブロック・サイズの半分 」です。
+行列を格納する為のメモリーを優先して、空きメモリーが足りない場合には、
+ファイル・アクセス用バッファーを減らしますが、それでも足りなければエラーが発生します。
+また、ブロックの断片化が激しいと極端に処理速度が低下します。
 
-<EFBFBD>@<40><><EFBFBD>J<EFBFBD>o<EFBFBD><6F><EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>̌<EFBFBD><CC8C><EFBFBD><EFBFBD>ł́A
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD>\<5C><><EFBFBD>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>Ă<EFBFBD> CPU <20><><EFBFBD>\<5C><><EFBFBD>t<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>E<EFBFBD>A<EFBFBD>N<EFBFBD>Z<EFBFBD>X<EFBFBD><58><EFBFBD>x<EFBFBD>Ȃǂɂ<C782><C982><EFBFBD><EFBFBD>ẮA
-<EFBFBD>傫<EFBFBD>ȃt<EFBFBD>@<40>C<EFBFBD><43><EFBFBD>̌<EFBFBD><CC8C><EFBFBD><EFBFBD>ɉ<EFBFBD><C989><EFBFBD><EFBFBD><EFBFBD><EFBFBD>҂‚<D282><C282>ƂɂȂ邩<C882><E982A9><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82><EFBFBD><EFBFBD>B
-<EFBFBD>W<EFBFBD><EFBFBD><EFBFBD>ł̓<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B>ʂ<EFBFBD><CA82>\<5C><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ΏڍׂȌ<D782><C88C><EFBFBD><EFBFBD><EFBFBD><EFBFBD>s<EFBFBD><73><EFBFBD>A<EFBFBD><41><EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><C882><EFBFBD><EFBFBD>ΊȈՌ<C888><D58C><EFBFBD><EFBFBD>ɂȂ<C982><C882>܂<EFBFBD><DC82>B
-<EFBFBD>j<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă邩<EFBFBD>ǂ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>̊m<EFBFBD>F<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̏ꍇ<EFBFBD><EFBFBD><EFBFBD>A<EFBFBD>j<EFBFBD><EFBFBD><EFBFBD>̒<EFBFBD><EFBFBD>x<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ꍇ<EFBFBD>́A
-<EFBFBD>ȈՌ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ł<EFBFBD><EFBFBD>\<5C><><EFBFBD>Ȃ̂ŁA<C581>I<EFBFBD>v<EFBFBD>V<EFBFBD><56><EFBFBD><EFBFBD><EFBFBD>ŏڍ׌<DA8D><D78C><EFBFBD><EFBFBD>𖳌<EFBFBD><F096B38C>ɂ<EFBFBD><C982>邱<EFBFBD>Ƃ<EFBFBD><C682>ł<EFBFBD><C582>܂<EFBFBD><DC82>B
+　リカバリ・ファイルやソース・ファイルの検査では、
+メモリーが十分にあっても CPU 性能やファイル・アクセス速度などによっては、
+大きなファイルの検査に何分も待つことになるかもしれません。
+標準ではメモリー量が十分あれば詳細な検査を行い、足りなければ簡易検査になります。
+破損してるかどうかの確認をしたいだけの場合や、破損の程度が小さい場合は、
+簡易検査でも十分なので、オプションで詳細検査を無効にすることもできます。
 
-<EFBFBD>@<40><><EFBFBD><EFBFBD><EFBFBD>‹<EFBFBD><C28B><EFBFBD> Windows Vista <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȍ~ (Windows 7, 8, 10, 11) <EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>p<EFBFBD>\<5C>R<EFBFBD><52><EFBFBD>ł<EFBFBD><C582>B
-Win32 API <EFBFBD><EFBFBD> VC <EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>^<5E>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD>g<EFBFBD><67><EFBFBD>Ă<EFBFBD><C482>̂ŁA
-<EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>R<EFBFBD>[<5B>h<EFBFBD><68><EFBFBD>R<EFBFBD><52><EFBFBD>p<EFBFBD>C<EFBFBD><43><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ɂ<EFBFBD> Visual Studio 2022 <EFBFBD>ȍ~<7E><><EFBFBD>K<EFBFBD>v<EFBFBD>ł<EFBFBD><C582>B
+　動作環境は Windows 7 かそれ以降 (Windows 8, 10, 11) が動くパソコンです。
+Win32 API と VC ランタイムを使ってるので、
+ソース・コードをコンパイルするには Visual Studio 2022 以降が必要です。
 
 
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
-[ <EFBFBD>Q<EFBFBD>l<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>◘<EFBFBD>p<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>\<5C>[<5B>X<EFBFBD>E<EFBFBD>R<EFBFBD>[<5B>h<EFBFBD>A<EFBFBD><41><EFBFBD>C<EFBFBD>u<EFBFBD><75><EFBFBD><EFBFBD><EFBFBD>[<5B>Ȃ<EFBFBD> ]
+[ 参考文献や利用したソース・コード、ライブラリーなど ]
 
 par2-specifications.pdf
 Parity Volume Set Specification 2.0
@@ -220,31 +220,31 @@ This module is Public Domain.
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
-[ <EFBFBD><EFBFBD><EFBFBD>C<EFBFBD>Z<EFBFBD><EFBFBD><EFBFBD>X<EFBFBD>Ȃ<EFBFBD> ]
+[ ライセンスなど ]
 
-<EFBFBD>@PAR 2.0 <EFBFBD>̃t<EFBFBD>H<EFBFBD>[<5B>}<7D>b<EFBFBD>g<EFBFBD>ɂ‚<C982><C282>Ă͐<C482><CD90><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>R<EFBFBD>[<5B>h<EFBFBD><68><EFBFBD><EFBFBD><EFBFBD>āA
-<EFBFBD>݊<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ێ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ĂȂ<EFBFBD><EFBFBD>ׂ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ɂȂ<EFBFBD><EFBFBD>悤<EFBFBD>ɂ<EFBFBD><EFBFBD>Ă܂<EFBFBD><EFBFBD>B
-<EFBFBD><EFBFBD><EFBFBD>̐l<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>R<EFBFBD>[<5B>h<EFBFBD>𗬗p<F097AC97><70><EFBFBD>Ă镔<C482><E99594><EFBFBD>̒<EFBFBD><CC92>쌠<EFBFBD>͂<EFBFBD><CD82>̐l<CC90>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ȊO<EFBFBD>̕<EFBFBD><EFBFBD><EFBFBD><EFBFBD>̒<EFBFBD><EFBFBD>쌠<EFBFBD><EFBFBD> <20>V<EFBFBD>c <20>L <20>ɂ<EFBFBD><C982><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
+　PAR 2.0 のフォーマットについては説明書や実装コードを見て、
+互換性を維持してなるべく同じになるようにしてます。
+他の人が書いたコードを流用してる部分の著作権はその人にあります。
+それ以外の部分の著作権は 澤田 豊 にあります。
 
-<EFBFBD>@<40><><EFBFBD>C<EFBFBD>Z<EFBFBD><5A><EFBFBD>X<EFBFBD><58> GPL <20>Ƃ<EFBFBD><C682><EFBFBD><EFBFBD><EFBFBD><EFBFBD>̂ɏ<CC82><C98F><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
-par2cmdline <EFBFBD><EFBFBD> phpar2 <EFBFBD>̃R<EFBFBD>[<5B>h<EFBFBD>𗬗p<F097AC97><70><EFBFBD>Ă<EFBFBD><C482>āA<C481><41><EFBFBD><EFBFBD><EFBFBD>炪 GPL <20>Ŕz<C594>z<EFBFBD><7A><EFBFBD><EFBFBD><EFBFBD>Ă邩<C482><E982A9><EFBFBD>ł<EFBFBD><C582>B
-GPL <EFBFBD>̃<EFBFBD><EFBFBD>C<EFBFBD>Z<EFBFBD><EFBFBD><EFBFBD>X<EFBFBD>̏ڍׂ͂悭<EFBFBD>킩<EFBFBD><EFBFBD><EFBFBD>Ȃ<EFBFBD><EFBFBD>̂ł<EFBFBD><EFBFBD><EFBFBD><EFBFBD>A
-<EFBFBD><EFBFBD><EFBFBD>p<EFBFBD><EFBFBD><EFBFBD>𕥂<EFBFBD><EFBFBD><EFBFBD><EFBFBD>ƂȂ<EFBFBD><EFBFBD>g<EFBFBD><EFBFBD><EFBFBD>ă\<5C>[<5B>X<EFBFBD>E<EFBFBD>R<EFBFBD>[<5B>h<EFBFBD><68><EFBFBD><EFBFBD><EFBFBD>J<EFBFBD><4A><EFBFBD>邱<EFBFBD>Ƃ<EFBFBD><C682><EFBFBD><EFBFBD><EFBFBD><EFBFBD>݂<EFBFBD><DD82><EFBFBD><EFBFBD>ł<EFBFBD><C582>B
-<EFBFBD>ڂ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ƃ͕t<EFBFBD><EFBFBD><EFBFBD>̉p<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>C<EFBFBD>Z<EFBFBD><EFBFBD><EFBFBD>X<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>ǂނ<EFBFBD><EFBFBD>A
-<EFBFBD><EFBFBD><EFBFBD>{<7B><><EFBFBD>̉<EFBFBD><CC89><EFBFBD><EFBFBD>ł<EFBFBD><C582>T<EFBFBD><54><EFBFBD>ēǂ<C493><C782>ł݂Ă<DD82><C482><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>B
+　ライセンスは GPL というものに準拠します。
+par2cmdline や phpar2 のコードを流用していて、それらが GPL で配布されてるからです。
+GPL のライセンスの詳細はよくわからないのですが、
+利用料を払うことなく使えてソース・コードを公開することが条件みたいです。
+詳しいことは付属の英文ライセンス文を読むか、
+日本語の解説でも探して読んでみてください。
 
 
 
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
-[ <EFBFBD>A<EFBFBD><EFBFBD><EFBFBD><EFBFBD> ]
+[ 連絡先 ]
 
-<EFBFBD>@<40><><EFBFBD>͂<EFBFBD><CD82>܂<EFBFBD> (<28>T<EFBFBD>ꂮ<EFBFBD>炢) <20><><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD>m<EFBFBD>F<EFBFBD>ł<EFBFBD><C582>܂<EFBFBD><DC82>񂵁A
-<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD> (<28>ꃖ<EFBFBD><EA8396><EFBFBD><EFBFBD><EFBFBD>炢) <20>C<EFBFBD><43><EFBFBD>^<5E>[<5B>l<EFBFBD>b<EFBFBD>g<EFBFBD><67><EFBFBD>g<EFBFBD><67><EFBFBD>Ȃ<EFBFBD><C882><EFBFBD><EFBFBD>Ƃ<EFBFBD><C682><EFBFBD><EFBFBD><EFBFBD><EFBFBD>܂<EFBFBD><DC82>B
-<EFBFBD><EFBFBD><EFBFBD>{<7B>I<EFBFBD>ɃT<C983>|<7C>[<5B>g<EFBFBD>̓<EFBFBD><CD83>[<5B><><EFBFBD>ōs<C58D><73><EFBFBD><EFBFBD><EFBFBD>ƂɂȂ<C982><C882>܂<EFBFBD><DC82><EFBFBD><EFBFBD>A
-<EFBFBD>Ԏ<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>–߂<EFBFBD><EFBFBD>Ă<EFBFBD><EFBFBD>邩<EFBFBD>͋C<EFBFBD><EFBFBD><EFBFBD>ɂ<EFBFBD><EFBFBD>҂<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>B
+　私はたまに (週一ぐらい) しかメール確認できませんし、
+長期間 (一ヶ月ぐらい) インターネットを使えないこともあります。
+基本的にサポートはメールで行うことになりますが、
+返事がいつ戻ってくるかは気長にお待ちください。
 
-<EFBFBD><EFBFBD><EFBFBD>O<EFBFBD><EFBFBD> <20>V<EFBFBD>c <20>L <20>ł<EFBFBD><C582>B
-<EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD>E<EFBFBD>A<EFBFBD>h<EFBFBD><68><EFBFBD>X<EFBFBD><58> tenfon (at mark) outlook.jp <EFBFBD>ł<EFBFBD>
-<EFBFBD><EFBFBD><EFBFBD>f<EFBFBD><EFBFBD><EFBFBD>[<5B><><EFBFBD><EFBFBD><EFBFBD>ۂ̕\<5C>L<EFBFBD>Ȃ̂ŁA<C581><41><EFBFBD>ۂ̃<DB82><CC83>[<5B><><EFBFBD>ł<EFBFBD> (at mark) <EFBFBD><EFBFBD> @ <EFBFBD>Ɋ<EFBFBD><EFBFBD><EFBFBD><EFBFBD>Ă<EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD><EFBFBD>B
+名前は 澤田 豊 です。
+メール・アドレスは tenfon (at mark) outlook.jp です
+迷惑メール拒否の表記なので、実際のメールでは (at mark) を @ に換えてください。

source/par2j/com.cpp

@@ -1,5 +1,5 @@
 // com.c
-// Copyright : 2022-01-30 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <windows.h>

source/par2j/common2.c

@@ -1,5 +1,5 @@
 // common2.c
-// Copyright : 2023-10-13 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <conio.h>
@@ -2008,8 +2008,11 @@ void check_cpu(void)
 		if (cache3_size > 0){
 			//printf("L3 cache: %d KB (%d way)\n", cache3_size >> 10 , cache3_way);
 			cpu_cache = cache3_size / cache3_way;	// set-associative のサイズにする
-			if (cpu_cache < 131072)
+			if (cpu_cache < 131072){
 				cpu_cache = 128 << 10;	// 128 KB 以上にする
+			} else {
+				cpu_cache = (cpu_cache + 0xFFFF) & 0xFFFF0000;	// 64 KB の倍数にする
+			}
 		}
 		if (cache2_size > 0){
 			//printf("L2 cache: %d KB (%d way)\n", cache2_size >> 10, cache2_way);
@@ -2030,7 +2033,9 @@ void check_cpu(void)
 						returnLength += returnLength / 2;
 				}
 			}
-			cpu_cache |= returnLength & 0x1FFFF;
+			if (returnLength > 0x8000)
+				returnLength = 0x8000;
+			cpu_cache |= returnLength & 0xFFFF;
 		}
 	}
 
@@ -2038,7 +2043,7 @@ void check_cpu(void)
 		limit_size = 128 << 10;
 	//printf("Limit size of Cache Blocking: %d KB\n", limit_size >> 10);
 	// cpu_flag の上位 16-bit にキャッシュの制限サイズを置く
-	cpu_flag |= limit_size & 0xFFFF0000;	// 64 KB 未満は無視する
+	cpu_flag |= (limit_size + 0xFFFF) & 0xFFFF0000;	// 64 KB の倍数にする
 
 	if (core_count == 0){	// 物理コア数が不明なら、論理コア数と同じにする
 		core_count = cpu_num;

source/par2j/crc.c

@@ -1,9 +1,9 @@
 // crc.c
-// Copyright : 2022-02-09 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <stdio.h>

source/par2j/create.c

@@ -1,5 +1,5 @@
 // create.c
-// Copyright : 2023-10-22 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <process.h>
@@ -26,6 +26,11 @@
 
 //#define TIMER // 実験用
 
+#ifdef TIMER
+#include <time.h>
+static double time_sec, time_speed;
+#endif
+
 // ソート時に項目を比較する
 static int sort_cmp(const void *elem1, const void *elem2)
 {
@@ -196,7 +201,7 @@ int set_common_packet(
 	__int64 prog_now = 0;
 
 #ifdef TIMER
-unsigned int time_start = GetTickCount();
+clock_t time_start = clock();
 #endif
 	print_progress_text(0, "Computing file hash");
 
@@ -305,14 +310,14 @@ unsigned int time_start = GetTickCount();
 	off += (64 + main_packet_size);
 
 #ifdef TIMER
-time_start = GetTickCount() - time_start;
-printf("hash %d.%03d sec", time_start / 1000, time_start % 1000);
-if (time_start > 0){
-	time_start = (int)((total_file_size * 125) / ((__int64)time_start * 131072));
-	printf(", %d MB/s\n", time_start);
+time_start = clock() - time_start;
+time_sec = (double)time_start / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = (double)total_file_size / (time_sec * 1048576);
 } else {
-	printf("\n");
+	time_speed = 0;
 }
+printf("hash %.3f sec, %.0f MB/s\n", time_sec, time_speed);
 #endif
 
 error_end:
@@ -341,7 +346,7 @@ int set_common_packet_multi(
 	FILE_HASH_TH th[MAX_MULTI_READ];
 
 #ifdef TIMER
-unsigned int time_start = GetTickCount();
+clock_t time_start = clock();
 #endif
 
 	memset(hSub, 0, sizeof(HANDLE) * MAX_MULTI_READ);
@@ -545,14 +550,14 @@ unsigned int time_start = GetTickCount();
 	}
 	print_progress_done();	// 改行して行の先頭に戻しておく
 #ifdef TIMER
-time_start = GetTickCount() - time_start;
-printf("hash %d.%03d sec", time_start / 1000, time_start % 1000);
-if (time_start > 0){
-	time_start = (int)((total_file_size * 125) / ((__int64)time_start * 131072));
-	printf(", %d MB/s\n", time_start);
+time_start = clock() - time_start;
+time_sec = (double)time_start / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = (double)total_file_size / (time_sec * 1048576);
 } else {
-	printf("\n");
+	time_speed = 0;
 }
+printf("hash %.3f sec, %.0f MB/s\n", time_sec, time_speed);
 #endif
 
 error_end:
@@ -700,7 +705,7 @@ int set_common_packet_hash(
 	__int64 prog_now = 0;
 
 #ifdef TIMER
-unsigned int time_start = GetTickCount();
+clock_t time_start = clock();
 #endif
 	print_progress_text(0, "Computing file hash");
 
@@ -740,8 +745,8 @@ unsigned int time_start = GetTickCount();
 	print_progress_done();	// 改行して行の先頭に戻しておく
 
 #ifdef TIMER
-time_start = GetTickCount() - time_start;
-printf("hash %d.%03d sec\n", time_start / 1000, time_start % 1000);
+time_start = clock() - time_start;
+printf("hash %.3f sec\n", (double)time_start / CLOCKS_PER_SEC);
 #endif
 	return 0;
 }
@@ -1065,7 +1070,7 @@ int create_recovery_file(
 #endif
 
 #ifdef TIMER
-unsigned int time_start = GetTickCount();
+clock_t time_start = clock();
 #endif
 	print_progress_text(0, "Constructing recovery file");
 	time_last = GetTickCount();
@@ -1258,8 +1263,8 @@ unsigned int time_start = GetTickCount();
 	print_progress_done();	// 改行して行の先頭に戻しておく
 
 #ifdef TIMER
-time_start = GetTickCount() - time_start;
-printf("write %d.%03d sec\n", time_start / 1000, time_start % 1000);
+time_start = clock() - time_start;
+printf("write %.3f sec\n", (double)time_start / CLOCKS_PER_SEC);
 #endif
 
 	return 0;
@@ -1824,10 +1829,12 @@ int split_files(
 		}
 		if (ext_len > 0){	// 全て数字の拡張子を持つソース・ファイルがあるなら
 			//printf_cp("\n risky name = %s \n", file_name);
+			wcscpy(file_path, file_name);	// 比較用に拡張子を取り除く
+			file_path[name_len] = 0;
 			for (num2 = 0; num2 < file_num; num2++){
 				if (num2 == num)
 					continue;
-				if (_wcsnicmp(list_buf + files[num2].name, file_name, name_len) == 0){
+				if (_wcsicmp(list_buf + files[num2].name, file_path) == 0){
 					//printf_cp(" match name = %s \n", list_buf + files[num2].name);
 					num8 = (files[num2].size + (__int64)split_size - 1) / split_size;
 					split_max = (int)num8;
@@ -1838,7 +1845,7 @@ int split_files(
 					if (((split_max < 1000) && (ext_len >= 5)) || ((split_max < 10000) && (ext_len >= 6)))
 						continue;	// 拡張子の桁数が異なる
 					// 上書きする危険性があるのでエラーにする
-					printf_cp("split bad file, %s\n", file_name);
+					printf_cp("split bad file, %s\n", list_buf + files[num2].name);
 					*cur_num = -1;
 					*cur_id = 0;
 					return 1;

source/par2j/gf16.c

@@ -49,7 +49,7 @@ POSSIBILITY OF SUCH DAMAGE.
 
  */
 
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 
 #include <stdlib.h>
 #include <stdio.h>
@@ -2795,7 +2795,7 @@ void galois_align_xor(
 void galois_align16_multiply(
 	unsigned char *r1,	// Region to multiply (must be aligned by 16)
 	unsigned char *r2,	// Products go here
-	unsigned int len,	// Byte length (must be multiple of 32)
+	unsigned int len,	// Byte length (must be multiple of 16)
 	int factor)			// Number to multiply by
 {
 	if (factor <= 1){

source/par2j/gf16.h

@@ -6,7 +6,7 @@ extern "C" {
 #endif
 
 
-extern unsigned short *galois_log_table;
+//extern unsigned short *galois_log_table;
 extern unsigned int cpu_flag;
 
 int galois_create_table(void);	// Returns 0 on success, -1 on failure

source/par2j/ini.c

@@ -1,5 +1,5 @@
 // ini.c
-// Copyright : 2022-10-12 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <malloc.h>

source/par2j/json.c

@@ -1,5 +1,5 @@
 // json.c
-// Copyright : 2023-02-06 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <stdio.h>

source/par2j/lib_opencl.c

@@ -1,9 +1,9 @@
 // lib_opencl.c
-// Copyright : 2023-10-22 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <stdio.h>
@@ -75,7 +75,7 @@ typedef cl_int (CL_API_CALL *API_clEnqueueNDRangeKernel)(cl_command_queue, cl_ke
 extern unsigned int cpu_flag;	// declared in common2.h
 extern int cpu_num;
 
-#define MAX_DEVICE 3
+#define MAX_DEVICE 8
 
 HMODULE hLibOpenCL = NULL;
 
@@ -84,7 +84,7 @@ cl_command_queue OpenCL_command = NULL;
 cl_kernel OpenCL_kernel = NULL;
 cl_mem OpenCL_src = NULL, OpenCL_dst = NULL, OpenCL_buf = NULL;
 size_t OpenCL_group_num;
-int OpenCL_method = 0;	// 正=速い機器を選ぶ, 負=遅い機器を選ぶ
+int OpenCL_method = 0;	// 標準では GPU を使わず、動作は自動選択される
 
 API_clCreateBuffer gfn_clCreateBuffer;
 API_clReleaseMemObject gfn_clReleaseMemObject;
@@ -100,7 +100,11 @@ API_clEnqueueNDRangeKernel gfn_clEnqueueNDRangeKernel;
 
 /*
 入力
-OpenCL_method : どのデバイスを選ぶか
+OpenCL_method : どのデバイスや関数を選ぶか
+    0x100 = 速い機器を選ぶ,          0x200 = 遅い機器を選ぶ
+  0x10000 = 1ブロックずつ計算する, 0x20000 = 2ブロックずつ計算しようとする
+  0x40000 = 4-byte memory access,  0x80000 = try 16-byte memory access
+ 0x100000 = CL_MEM_COPY_HOST_PTR, 0x200000 = CL_MEM_USE_HOST_PTR
 unit_size : ブロックの単位サイズ
 src_max : ソース・ブロック個数
 
@@ -111,11 +115,12 @@ OpenCL_method : 動作フラグいろいろ
 */
 
 // 0=成功, 1～エラー番号
-int init_OpenCL(int unit_size, int *src_max)
+int init_OpenCL(unsigned int unit_size, int *src_max)
 {
 	char buf[2048], *p_source;
 	int err = 0, i, j;
 	int gpu_power, count;
+	int unified_memory;	// non zero = Integrated GPU
 	size_t data_size, alloc_max;
 	//FILE *fp;
 	HRSRC res;
@@ -136,6 +141,7 @@ int init_OpenCL(int unit_size, int *src_max)
 	API_clReleaseProgram fn_clReleaseProgram;
 	API_clCreateKernel fn_clCreateKernel;
 	API_clGetKernelWorkGroupInfo fn_clGetKernelWorkGroupInfo;
+	API_clReleaseKernel fn_clReleaseKernel;
 	cl_int ret;
 	cl_uint num_platforms = 0, num_devices = 0, num_groups, param_value;
 	cl_ulong param_value8;
@@ -215,6 +221,9 @@ int init_OpenCL(int unit_size, int *src_max)
 	fn_clGetKernelWorkGroupInfo = (API_clGetKernelWorkGroupInfo)GetProcAddress(hLibOpenCL, "clGetKernelWorkGroupInfo");
 	if (fn_clGetKernelWorkGroupInfo == NULL)
 		return err;
+	fn_clReleaseKernel = (API_clReleaseKernel)GetProcAddress(hLibOpenCL, "clReleaseKernel");
+	if (fn_clReleaseKernel == NULL)
+		return err;
 	gfn_clFinish = (API_clFinish)GetProcAddress(hLibOpenCL, "clFinish");
 	if (gfn_clFinish == NULL)
 		return err;
@@ -226,12 +235,12 @@ int init_OpenCL(int unit_size, int *src_max)
 	ret = fn_clGetPlatformIDs(MAX_DEVICE, platform_id, &num_platforms);
 	if (ret != CL_SUCCESS)
 		return (ret << 8) | 10;
-	if (OpenCL_method >= 0){	// 選択する順序と初期値を変える
-		OpenCL_method = 1;
-		gpu_power = 0;
-	} else {
-		OpenCL_method = -1;
+	if (num_platforms > MAX_DEVICE)
+		num_platforms = MAX_DEVICE;
+	if (OpenCL_method & 0x200){	// 選択する順序と初期値を変える
 		gpu_power = INT_MIN;
+	} else {
+		gpu_power = 0;
 	}
 	alloc_max = 0;
 
@@ -247,6 +256,8 @@ int init_OpenCL(int unit_size, int *src_max)
 		// 環境内の OpenCL 対応機器の数
 		if (fn_clGetDeviceIDs(platform_id[i], CL_DEVICE_TYPE_GPU, MAX_DEVICE, device_id, &num_devices) != CL_SUCCESS)
 			continue;
+		if (num_devices > MAX_DEVICE)
+			num_devices = MAX_DEVICE;
 
 		for (j = 0; j < (int)num_devices; j++){
 			// デバイスが利用可能か確かめる
@@ -265,45 +276,42 @@ int init_OpenCL(int unit_size, int *src_max)
 			ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_VERSION, sizeof(buf), buf, NULL);
 			if (ret == CL_SUCCESS)
 				printf("Device version = %s\n", buf);
-			ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_LOCAL_MEM_SIZE, sizeof(cl_ulong), &param_value8, NULL);
-			if (ret == CL_SUCCESS)
-				printf("LOCAL_MEM_SIZE = %I64d KB\n", param_value8 >> 10);
-
-			// 無理とは思うけど、一応チェックする
-//#define CL_DEVICE_SVM_CAPABILITIES                  0x1053
-//#define CL_DEVICE_SVM_COARSE_GRAIN_BUFFER           (1 << 0)
-//#define CL_DEVICE_SVM_FINE_GRAIN_BUFFER             (1 << 1)
-//#define CL_DEVICE_SVM_FINE_GRAIN_SYSTEM             (1 << 2)
-//#define CL_DEVICE_SVM_ATOMICS                       (1 << 3)
-//			ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_SVM_CAPABILITIES, sizeof(cl_ulong), &param_value8, NULL);
-//			if (ret == CL_INVALID_VALUE)
-//				printf("Shared Virtual Memory is not supported\n");
-//			if (ret == CL_SUCCESS)
-//				printf("Shared Virtual Memory = 0x%I64X\n", param_value8);
 #endif
 
+			// 取得できなくてもエラーにしない
+			ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_HOST_UNIFIED_MEMORY, sizeof(cl_uint), &param_value, NULL);
+			if (ret == CL_SUCCESS){
+				if (param_value != 0){
+#ifdef DEBUG_OUTPUT
+					printf("HOST_UNIFIED_MEMORY = %d\n", param_value);
+#endif
+					param_value = 1;
+				}
+			} else {	// CL_DEVICE_HOST_UNIFIED_MEMORY は OpenCL 2.0 以降で非推奨になった
+				param_value = 0;
+			}
+
+			// 取得できない場合はエラーにする
 			ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(cl_ulong), &param_value8, NULL);
 			if (ret != CL_SUCCESS)
 				continue;
 #ifdef DEBUG_OUTPUT
 			printf("MAX_MEM_ALLOC_SIZE = %I64d MB\n", param_value8 >> 20);
 #endif
-
 			ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(cl_uint), &num_groups, NULL);
 			if (ret != CL_SUCCESS)
 				continue;
 			ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(size_t), &data_size, NULL);
 			if (ret != CL_SUCCESS)
 				continue;
-			// CL_DEVICE_HOST_UNIFIED_MEMORY は OpenCL 2.0 以降で非推奨になったので、参照しない
-
 #ifdef DEBUG_OUTPUT
 			printf("MAX_COMPUTE_UNITS = %d\n", num_groups);
 			printf("MAX_WORK_GROUP_SIZE = %zd\n", data_size);
 #endif
-			// MAX_COMPUTE_UNITS * MAX_WORK_GROUP_SIZE で計算力を測る
-			count = (int)data_size * num_groups;
-			count *= OpenCL_method;	// 符号を変える
+			// MAX_COMPUTE_UNITS * MAX_WORK_GROUP_SIZE で計算力を測る、外付けGPUなら値を倍にする
+			count = (2 - param_value) * (int)data_size * num_groups;
+			if (OpenCL_method & 0x200)	// Prefer slower device
+				count *= -1;	// 符号を変える
 			//printf("prev = %d, now = %d\n", gpu_power, count);
 			if ((count > gpu_power) && (data_size >= 256) &&	// 256以上ないとテーブルを作れない
 					(param_value8 / 8 > (cl_ulong)unit_size)){	// CL_DEVICE_MAX_MEM_ALLOC_SIZE に収まるか
@@ -312,6 +320,7 @@ int init_OpenCL(int unit_size, int *src_max)
 				selected_platform = platform_id[i];
 				OpenCL_group_num = num_groups;	// ワークグループ数は COMPUTE_UNITS 数にする
 				alloc_max = (size_t)param_value8;
+				unified_memory = param_value;	// 0 = discrete GPU, 1 = integrated GPU
 
 				// AMD や Intel の GPU ではメモリー領域が全体の 1/4 とは限らない
 				ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_GLOBAL_MEM_SIZE, sizeof(cl_ulong), &param_value8, NULL);
@@ -353,31 +362,6 @@ int init_OpenCL(int unit_size, int *src_max)
 	if (ret != CL_SUCCESS)
 		return (ret << 8) | 12;
 
-	// 計算方式を選択する
-	if ((((cpu_flag & 0x101) == 1) || ((cpu_flag & 0x110) == 0x10)) && (sse_unit == 32)){
-		OpenCL_method = 2;	// SSSE3 & ALTMAP または AVX2 ならデータの並び替え対応版を使う
-	} else if (((cpu_flag & 128) != 0) && (sse_unit == 256)){
-		OpenCL_method = 4;	// JIT(SSE2) は bit ごとに上位から 16バイトずつ並ぶ
-		// ローカルのテーブルサイズが異なることに注意
-		// XOR 方式以外は 2KB (4バイト * 256項目 * 2個) 使う
-		// XOR (JIT) は 64バイト (4バイト * 16項目) 使う
-	} else {
-		OpenCL_method = 1;	// 並び替えられてないデータ用
-	}
-
-	// work group 数が必要以上に多い場合は減らす
-	if (OpenCL_method == 2){
-		// work item 一個が 8バイトずつ計算する、256個なら work group ごとに 2KB 担当する
-		data_size = unit_size / 2048;
-	} else {
-		// work item 一個が 4バイトずつ計算する、256個なら work group ごとに 1KB 担当する
-		data_size = unit_size / 1024;
-	}
-	if (OpenCL_group_num > data_size){
-		OpenCL_group_num = data_size;
-		printf("Number of work groups is reduced to %zd\n", OpenCL_group_num);
-	}
-
 	// 最大で何ブロック分のメモリー領域を保持できるのか（ここではまだ確保しない）
 	// 後で実際に確保する量はこれよりも少なくなる
 	count = (int)(alloc_max / unit_size);	// 確保できるメモリー量から逆算する
@@ -389,25 +373,6 @@ int init_OpenCL(int unit_size, int *src_max)
 	printf("src buf : %zd KB (%d blocks), possible\n", data_size >> 10, count);
 #endif
 
-	// 出力先は1ブロック分だけあればいい
-	// CL_MEM_ALLOC_HOST_PTRを使えばpinned memoryになるらしい
-	data_size = unit_size;
-	OpenCL_dst = gfn_clCreateBuffer(OpenCL_context, CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR, data_size, NULL, &ret);
-	if (ret != CL_SUCCESS)
-		return (ret << 8) | 13;
-#ifdef DEBUG_OUTPUT
-	printf("dst buf : %zd KB (%zd Bytes), OK\n", data_size >> 10, data_size);
-#endif
-
-	// factor は最大個数分 (src_max個)
-	data_size = sizeof(unsigned short) * (*src_max);
-	OpenCL_buf = gfn_clCreateBuffer(OpenCL_context, CL_MEM_READ_ONLY, data_size, NULL, &ret);
-	if (ret != CL_SUCCESS)
-		return (ret << 8) | 14;
-#ifdef DEBUG_OUTPUT
-	printf("factor buf : %zd Bytes (%d factors), OK\n", data_size, (*src_max));
-#endif
-
 /*
 	// テキスト形式の OpenCL C ソース・コードを読み込む
 	err = 4;
@@ -508,18 +473,208 @@ int init_OpenCL(int unit_size, int *src_max)
 		return (ret << 8) | 21;
 	}
 
-	// カーネル関数を抽出する
-	wsprintfA(buf, "method%d", OpenCL_method & 7);
-	OpenCL_kernel = fn_clCreateKernel(program, buf, &ret);
-	if (ret != CL_SUCCESS)
-		return (ret << 8) | 22;
+	// 計算方式を選択する
+	if ((((cpu_flag & 0x101) == 1) || ((cpu_flag & 0x110) == 0x10)) && (sse_unit == 32)){
+		int select_method;	// SSSE3 & ALTMAP または AVX2 ならデータの並び替え対応版を使う
+		if (OpenCL_method & 0x80000){	// 16-byte and 2 blocks
+			select_method = 12;
+		} else if (OpenCL_method & 0x40000){	// 4-byte and 2 blocks
+			select_method = 10;
+		} else if (OpenCL_method & 0x20000){	// 16-byte
+			select_method = 4;
+		} else if (OpenCL_method & 0x10000){	// 4-byte
+			select_method = 2;
+		} else {	// kernel を作って詳細を確かめる
+			size_t item2, item4, item10, item12;
+			cl_kernel kernel2, kernel4, kernel10, kernel12;
+			item2 = item4 = item10 = item12 = 0;
+			// まずは一番重くて速い奴を調べる
+			wsprintfA(buf, "method%d", 12);
+			kernel12 = fn_clCreateKernel(program, buf, &ret);
+			if (ret == CL_SUCCESS){
+				ret = fn_clGetKernelWorkGroupInfo(kernel12, selected_device, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, sizeof(size_t), &item12, NULL);
+				if (ret == CL_SUCCESS){
 #ifdef DEBUG_OUTPUT
-	printf("CreateKernel : %s\n", buf);
+					printf("\nTesting %s\n", buf);
+					printf("PREFERRED_WORK_GROUP_SIZE_MULTIPLE = %zu\n", item12);
 #endif
+				}
+			}
+			if (item12 >= 32){	// 32以上あれば余裕で動くとみなす
+				select_method = 12;
+				OpenCL_kernel = kernel12;
+#ifdef DEBUG_OUTPUT
+				printf("\nSelected method%d\n", select_method);
+#endif
+			} else {	// 他の奴と比較する
+				wsprintfA(buf, "method%d", 2);
+				kernel2 = fn_clCreateKernel(program, buf, &ret);
+				if (ret == CL_SUCCESS){
+					ret = fn_clGetKernelWorkGroupInfo(kernel2, selected_device, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, sizeof(size_t), &item2, NULL);
+					if (ret == CL_SUCCESS){
+#ifdef DEBUG_OUTPUT
+						printf("\nTesting %s\n", buf);
+						printf("PREFERRED_WORK_GROUP_SIZE_MULTIPLE = %zu\n", item2);
+#endif
+					}
+				}
+				if (item12 >= item2){
+					select_method = 12;
+					OpenCL_kernel = kernel12;
+					ret = fn_clReleaseKernel(kernel2);
+#ifdef DEBUG_OUTPUT
+					if (ret != CL_SUCCESS)
+						printf("clReleaseKernel : Failed\n");
+					printf("\nSelected method%d\n", select_method);
+#endif
+				} else {
+					ret = fn_clReleaseKernel(kernel12);
+#ifdef DEBUG_OUTPUT
+					if (ret != CL_SUCCESS)
+						printf("clReleaseKernel : Failed\n");
+#endif
+					wsprintfA(buf, "method%d", 10);
+					kernel10 = fn_clCreateKernel(program, buf, &ret);
+					if (ret == CL_SUCCESS){
+						ret = fn_clGetKernelWorkGroupInfo(kernel10, selected_device, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, sizeof(size_t), &item10, NULL);
+						if (ret == CL_SUCCESS){
+#ifdef DEBUG_OUTPUT
+							printf("\nTesting %s\n", buf);
+							printf("PREFERRED_WORK_GROUP_SIZE_MULTIPLE = %zu\n", item10);
+#endif
+						}
+					}
+					if (item10 >= item2){
+						select_method = 10;
+						OpenCL_kernel = kernel10;
+						ret = fn_clReleaseKernel(kernel2);
+#ifdef DEBUG_OUTPUT
+						if (ret != CL_SUCCESS)
+							printf("clReleaseKernel : Failed\n");
+						printf("\nSelected method%d\n", select_method);
+#endif
+					} else {
+						wsprintfA(buf, "method%d", 4);
+						kernel4 = fn_clCreateKernel(program, buf, &ret);
+						if (ret == CL_SUCCESS){
+							ret = fn_clGetKernelWorkGroupInfo(kernel4, selected_device, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, sizeof(size_t), &item4, NULL);
+							if (ret == CL_SUCCESS){
+#ifdef DEBUG_OUTPUT
+								printf("\nTesting %s\n", buf);
+								printf("PREFERRED_WORK_GROUP_SIZE_MULTIPLE = %zu\n", item4);
+#endif
+							}
+						}
+						if (item4 >= item2){
+							select_method = 4;
+							OpenCL_kernel = kernel4;
+							ret = fn_clReleaseKernel(kernel2);
+#ifdef DEBUG_OUTPUT
+							if (ret != CL_SUCCESS)
+								printf("clReleaseKernel : Failed\n");
+							printf("\nSelected method%d\n", select_method);
+#endif
+						} else {
+							select_method = 2;
+							OpenCL_kernel = kernel2;
+							ret = fn_clReleaseKernel(kernel4);
+#ifdef DEBUG_OUTPUT
+							if (ret != CL_SUCCESS)
+								printf("clReleaseKernel : Failed\n");
+							printf("\nSelected method%d\n", select_method);
+#endif
+						}
+					}
+				}
+			}
+		}
+		OpenCL_method |= select_method;
+	} else if (((cpu_flag & 128) != 0) && (sse_unit == 256)){
+		OpenCL_method |= 16;	// JIT(SSE2) は bit ごとに上位から 16バイトずつ並ぶ
+		// ローカルのテーブルサイズが異なることに注意
+		// XOR 方式以外は 2KB (4バイト * 256項目 * 2個) 使う
+		// XOR (JIT) は 64バイト (4バイト * 16項目) 使う
+	} else {
+		int select_method;	// 並び替えられてないデータ用
+		if (OpenCL_method & 0x40000){	// 4-byte and 2 blocks
+			select_method = 9;
+		} else if (OpenCL_method & 0x10000){	// 4-byte
+			select_method = 1;
+		} else {	// kernel を作って詳細を確かめる
+			size_t item1, item9;
+			cl_kernel kernel1, kernel9;
+			item1 = item9 = 0;
+			// まずは一番重くて速い奴を調べる
+			wsprintfA(buf, "method%d", 9);
+			kernel9 = fn_clCreateKernel(program, buf, &ret);
+			if (ret == CL_SUCCESS){
+				ret = fn_clGetKernelWorkGroupInfo(kernel9, selected_device, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, sizeof(size_t), &item9, NULL);
+				if (ret == CL_SUCCESS){
+#ifdef DEBUG_OUTPUT
+					printf("\nTesting %s\n", buf);
+					printf("PREFERRED_WORK_GROUP_SIZE_MULTIPLE = %zu\n", item9);
+#endif
+				}
+			}
+			if (item9 >= 32){	// 32以上あれば余裕で動くとみなす
+				select_method = 9;
+				OpenCL_kernel = kernel9;
+#ifdef DEBUG_OUTPUT
+				printf("\nSelected method%d\n", select_method);
+#endif
+			} else {	// 他の奴と比較する
+				wsprintfA(buf, "method%d", 1);
+				kernel1 = fn_clCreateKernel(program, buf, &ret);
+				if (ret == CL_SUCCESS){
+					ret = fn_clGetKernelWorkGroupInfo(kernel1, selected_device, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, sizeof(size_t), &item1, NULL);
+					if (ret == CL_SUCCESS){
+#ifdef DEBUG_OUTPUT
+						printf("\nTesting %s\n", buf);
+						printf("PREFERRED_WORK_GROUP_SIZE_MULTIPLE = %zu\n", item1);
+#endif
+					}
+				}
+				if (item9 >= item1){
+					select_method = 9;
+					OpenCL_kernel = kernel9;
+					ret = fn_clReleaseKernel(kernel1);
+#ifdef DEBUG_OUTPUT
+					if (ret != CL_SUCCESS)
+						printf("clReleaseKernel : Failed\n");
+					printf("\nSelected method%d\n", select_method);
+#endif
+				} else {
+					select_method = 1;
+					OpenCL_kernel = kernel1;
+					ret = fn_clReleaseKernel(kernel9);
+#ifdef DEBUG_OUTPUT
+					if (ret != CL_SUCCESS)
+						printf("clReleaseKernel : Failed\n");
+					printf("\nSelected method%d\n", select_method);
+#endif
+				}
+			}
+		}
+		OpenCL_method |= select_method;
+	}
+
+	// カーネル関数を抽出する
+	if (OpenCL_kernel == NULL){
+		wsprintfA(buf, "method%d", OpenCL_method & 31);
+		OpenCL_kernel = fn_clCreateKernel(program, buf, &ret);
+		if (ret != CL_SUCCESS)
+			return (ret << 8) | 22;
+#ifdef DEBUG_OUTPUT
+		printf("CreateKernel : %s\n", buf);
+		ret = fn_clGetKernelWorkGroupInfo(OpenCL_kernel, selected_device, CL_KERNEL_PREFERRED_WORK_GROUP_SIZE_MULTIPLE, sizeof(size_t), &data_size, NULL);
+		if (ret == CL_SUCCESS)
+			printf("PREFERRED_WORK_GROUP_SIZE_MULTIPLE = %zu\n", data_size);
+#endif
+	}
 
 	// カーネルが実行できる work item 数を調べる
-	ret = fn_clGetKernelWorkGroupInfo(OpenCL_kernel, NULL, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &data_size, NULL);
-	if ((ret == CL_SUCCESS) && (data_size < 256)){	// 最低でも 256以上は必要
+	ret = fn_clGetKernelWorkGroupInfo(OpenCL_kernel, selected_device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &data_size, NULL);
+	if ((ret == CL_SUCCESS) && (data_size < 256)){	// 最低でも 256 以上は必要
 #ifdef DEBUG_OUTPUT
 		printf("KERNEL_WORK_GROUP_SIZE = %zd\n", data_size);
 #endif
@@ -538,6 +693,62 @@ int init_OpenCL(int unit_size, int *src_max)
 		fn_clUnloadCompiler();
 	}
 
+	// work group 数が必要以上に多い場合は減らす
+	if (OpenCL_method & 4){
+		// work item 一個が 32バイトずつ計算する、256個なら work group ごとに 8KB 担当する
+		data_size = unit_size / 8192;
+	} else if (OpenCL_method & 2){
+		// work item 一個が 8バイトずつ計算する、256個なら work group ごとに 2KB 担当する
+		data_size = unit_size / 2048;
+	} else {
+		// work item 一個が 4バイトずつ計算する、256個なら work group ごとに 1KB 担当する
+		data_size = unit_size / 1024;
+	}
+	if (OpenCL_group_num > data_size){
+		OpenCL_group_num = data_size;
+#ifdef DEBUG_OUTPUT
+		printf("Number of work groups is reduced to %zd\n", OpenCL_group_num);
+#endif
+	}
+
+	// データへのアクセス方法をデバイスによって変える
+	if (OpenCL_method & 0x200000){
+		OpenCL_method |= 32;
+	} else if ((OpenCL_method & 0x100000) == 0){
+		if (unified_memory){
+			OpenCL_method |= 32;	// Integrated GPU なら CL_MEM_USE_HOST_PTR を使う
+		} else {	// Discrete GPU でも Nvidia のは動作を変える
+			ret = fn_clGetDeviceInfo(selected_device, CL_DEVICE_VERSION, sizeof(buf), buf, NULL);
+			if (ret == CL_SUCCESS){
+				if (strstr(buf, "CUDA") != NULL)
+					OpenCL_method |= 32;	// NVIDIA GPU なら CL_MEM_USE_HOST_PTR を使う
+			}
+		}
+	}
+
+	// 出力先は1ブロック分だけあればいい
+	// CL_MEM_ALLOC_HOST_PTRを使えばpinned memoryになるらしい
+	data_size = unit_size;
+	if (OpenCL_method & 8)
+		data_size *= 2;	// 2ブロックずつ計算できるように、2倍確保しておく
+	OpenCL_dst = gfn_clCreateBuffer(OpenCL_context, CL_MEM_ALLOC_HOST_PTR, data_size, NULL, &ret);
+	if (ret != CL_SUCCESS)
+		return (ret << 8) | 13;
+#ifdef DEBUG_OUTPUT
+	printf("dst buf : %zd KB (%zd Bytes), OK\n", data_size >> 10, data_size);
+#endif
+
+	// factor は最大個数分 (src_max個)
+	data_size = sizeof(unsigned short) * (*src_max);
+	if (OpenCL_method & 8)
+		data_size *= 2;	// 2ブロックずつ計算できるように、2倍確保しておく
+	OpenCL_buf = gfn_clCreateBuffer(OpenCL_context, CL_MEM_READ_ONLY, data_size, NULL, &ret);
+	if (ret != CL_SUCCESS)
+		return (ret << 8) | 14;
+#ifdef DEBUG_OUTPUT
+	printf("factor buf : %zd Bytes (%d factors), OK\n", data_size, (*src_max));
+#endif
+
 	// カーネル引数を指定する
 	ret = gfn_clSetKernelArg(OpenCL_kernel, 1, sizeof(cl_mem), &OpenCL_dst);
 	if (ret != CL_SUCCESS)
@@ -545,13 +756,12 @@ int init_OpenCL(int unit_size, int *src_max)
 	ret = gfn_clSetKernelArg(OpenCL_kernel, 2, sizeof(cl_mem), &OpenCL_buf);
 	if (ret != CL_SUCCESS)
 		return (ret << 8) | 102;
-	if (ret != CL_SUCCESS)
-		return (ret << 8) | 103;
 
 #ifdef DEBUG_OUTPUT
 	// ワークアイテム数
 	printf("\nMax number of work items = %zd (256 * %zd)\n", OpenCL_group_num * 256, OpenCL_group_num);
 #endif
+	OpenCL_method &= 0xFF;	// 最後に選択設定を消去する
 
 	return 0;
 }
@@ -663,16 +873,24 @@ void info_OpenCL(char *buf, int buf_size)
 // ソース・ブロックをデバイス側にコピーする
 int gpu_copy_blocks(
 	unsigned char *data,	// ブロックのバッファー (境界は 4096にすること)
-	int unit_size,			// 4096の倍数にすること
+	unsigned int unit_size,	// 4096の倍数にすること
 	int src_num)			// 何ブロックをコピーするのか
 {
 	size_t data_size;
 	cl_int ret;
+	cl_mem_flags flags;
 
 	// Integrated GPU と Discrete GPU の違いに関係なく、使う分だけ毎回メモリー領域を確保する
 	data_size = (size_t)unit_size * src_num;
-	// Intel GPUならZeroCopyできる、GeForce GPUでもメモリー消費量が少なくてコピーが速い
-	OpenCL_src = gfn_clCreateBuffer(OpenCL_context, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR, data_size, data, &ret);
+	if (OpenCL_method & 32){	// AMD's APU や Integrated GPU なら ZeroCopy する
+		// 実際に比較してみると GeForce GPU でもメモリー消費量が少なくてコピーが速い
+		// NVIDIA GPU は CL_MEM_USE_HOST_PTR でも VRAM 上にキャッシュするので速いらしい
+		flags = CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR;
+	} else {	// Discrete GPU ならデータを VRAM にコピーする
+		// AMD GPU は明示的にコピーするよう指定しないといけない
+		flags = CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR;
+	}
+	OpenCL_src = gfn_clCreateBuffer(OpenCL_context, flags, data_size, data, &ret);
 	if (ret != CL_SUCCESS)
 		return (ret << 8) | 1;
 #ifdef DEBUG_OUTPUT
@@ -691,17 +909,31 @@ int gpu_copy_blocks(
 int gpu_multiply_blocks(
 	int src_num,			// Number of multiplying source blocks
 	unsigned short *mat,	// Matrix of numbers to multiply by
+	unsigned short *mat2,	// Set to calculate 2 blocks at once
 	unsigned char *buf,		// Products go here
-	int len)				// Byte length
+	unsigned int len)		// Byte length
 {
 	unsigned __int64 *vram, *src, *dst;
 	size_t global_size, local_size;
 	cl_int ret;
 
 	// 倍率の配列をデバイス側に書き込む
-	ret = gfn_clEnqueueWriteBuffer(OpenCL_command, OpenCL_buf, CL_FALSE, 0, sizeof(short) * src_num, mat, 0, NULL, NULL);
+	if (mat2 == NULL){	// 1ブロック分だけコピーする
+		ret = gfn_clEnqueueWriteBuffer(OpenCL_command, OpenCL_buf, CL_FALSE, 0, sizeof(short) * src_num, mat, 0, NULL, NULL);
+	} else {	// 2ブロックずつ計算する場合は、配列のサイズも２倍になる
+		if ((size_t)mat2 == 1){	// アドレスが 1 になることはあり得ないので、識別できる
+			ret = gfn_clEnqueueWriteBuffer(OpenCL_command, OpenCL_buf, CL_FALSE, 0, sizeof(short) * src_num * 2, mat, 0, NULL, NULL);
+		} else {	// 2回コピーする
+			size_t data_size = sizeof(short) * src_num;
+			ret = gfn_clEnqueueWriteBuffer(OpenCL_command, OpenCL_buf, CL_FALSE, 0, data_size, mat, 0, NULL, NULL);
+			if (ret != CL_SUCCESS)
+				return (ret << 8) | 10;
+			// もう一つの配列は違う場所からコピーする
+			ret = gfn_clEnqueueWriteBuffer(OpenCL_command, OpenCL_buf, CL_FALSE, data_size, data_size, mat2, 0, NULL, NULL);
+		}
+	}
 	if (ret != CL_SUCCESS)
-		return (ret << 8) | 10;
+		return (ret << 8) | 11;
 
 	// 引数を指定する
 	ret = gfn_clSetKernelArg(OpenCL_kernel, 3, sizeof(int), &src_num);
@@ -709,17 +941,17 @@ int gpu_multiply_blocks(
 		return (ret << 8) | 103;
 
 	// カーネル並列実行
-	local_size = 256;	// テーブルやキャッシュのため、work item 数は 256に固定する
+	local_size = 256;	// テーブルやキャッシュのため、work item 数は 256 に固定する
 	global_size = OpenCL_group_num * 256;
-	//printf("group num = %d, global size = %d, local size = 256 \n", OpenCL_group_num, global_size);
+	//printf("group num = %d, global size = %d, local size = %d \n", OpenCL_group_num, global_size, local_size);
 	ret = gfn_clEnqueueNDRangeKernel(OpenCL_command, OpenCL_kernel, 1, NULL, &global_size, &local_size, 0, NULL, NULL);
 	if (ret != CL_SUCCESS)
-		return (ret << 8) | 11;
+		return (ret << 8) | 12;
 
 	// 出力内容をホスト側に反映させる
 	vram = gfn_clEnqueueMapBuffer(OpenCL_command, OpenCL_dst, CL_TRUE, CL_MAP_READ, 0, len, 0, NULL, NULL, &ret);
 	if (ret != CL_SUCCESS)
-		return (ret << 8) | 12;
+		return (ret << 8) | 13;
 
 	// 8バイトごとに XOR する (SSE2 で XOR しても速くならず)
 	src = vram;
@@ -734,7 +966,7 @@ int gpu_multiply_blocks(
 	// ホスト側でデータを変更しなくても、clEnqueueMapBufferと対で呼び出さないといけない
 	ret = gfn_clEnqueueUnmapMemObject(OpenCL_command, OpenCL_dst, vram, 0, NULL, NULL);
 	if (ret != CL_SUCCESS)
-		return (ret << 8) | 13;
+		return (ret << 8) | 14;
 
 	return 0;
 }
@@ -747,12 +979,12 @@ int gpu_finish(void)
 	// 全ての処理が終わるのを待つ
 	ret = gfn_clFinish(OpenCL_command);
 	if (ret != CL_SUCCESS)
-		return (ret << 8) | 20;
+		return (ret << 8) | 30;
 
 	if (OpenCL_src != NULL){	// 確保されてる場合は解除する
 		ret = gfn_clReleaseMemObject(OpenCL_src);
 		if (ret != CL_SUCCESS)
-			return (ret << 8) | 21;
+			return (ret << 8) | 31;
 		OpenCL_src = NULL;
 	}
 

source/par2j/lib_opencl.h

@@ -10,20 +10,21 @@ extern "C" {
 
 extern int OpenCL_method;
 
-int init_OpenCL(int unit_size, int *src_max);
+int init_OpenCL(unsigned int unit_size, int *src_max);
 int free_OpenCL(void);
 void info_OpenCL(char *buf, int buf_size);
 
 int gpu_copy_blocks(
 	unsigned char *data,
-	int unit_size,
+	unsigned int unit_size,
 	int src_num);
 
 int gpu_multiply_blocks(
 	int src_num,			// Number of multiplying source blocks
 	unsigned short *mat,	// Matrix of numbers to multiply by
+	unsigned short *mat2,	// Set to calculate 2 blocks at once
 	unsigned char *buf,		// Products go here
-	int len);				// Byte length
+	unsigned int len);		// Byte length
 
 int gpu_finish(void);
 

source/par2j/list.c

@@ -1,5 +1,5 @@
 // list.c
-// Copyright : 2023-10-15 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <process.h>
@@ -26,6 +26,11 @@
 
 //#define TIMER // 実験用
 
+#ifdef TIMER
+#include <time.h>
+static double time_sec, time_speed;
+#endif
+
 // recovery set のファイルのハッシュ値を調べる (空のファイルは除く)
 // 0x00 = ファイルが存在して完全である
 // 0x01 = ファイルが存在しない
@@ -296,7 +301,7 @@ int check_file_complete(
 {
 	int i, rv;
 #ifdef TIMER
-unsigned int time_start = GetTickCount();
+clock_t time_start = clock();
 #endif
 
 	printf("\nVerifying Input File   :\n");
@@ -332,14 +337,14 @@ unsigned int time_start = GetTickCount();
 	}
 
 #ifdef TIMER
-time_start = GetTickCount() - time_start;
-printf("\n hash %d.%03d sec", time_start / 1000, time_start % 1000);
-if (time_start > 0){
-	time_start = (int)((total_file_size * 125) / ((__int64)time_start * 131072));
-	printf(", %d MB/s\n", time_start);
+time_start = clock() - time_start;
+time_sec = (double)time_start / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = (double)total_file_size / (time_sec * 1048576);
 } else {
-	printf("\n");
+	time_speed = 0;
 }
+printf("\n hash %.3f sec, %.0f MB/s\n", time_sec, time_speed);
 #endif
 	return 0;
 }
@@ -364,7 +369,7 @@ int check_file_complete_multi(
 	HANDLE hSub[MAX_READ_NUM];
 	FILE_CHECK_TH th[MAX_READ_NUM];
 #ifdef TIMER
-unsigned int time_start = GetTickCount();
+clock_t time_start = clock();
 #endif
 
 	memset(hSub, 0, sizeof(HANDLE) * MAX_READ_NUM);
@@ -630,14 +635,14 @@ unsigned int time_start = GetTickCount();
 	}
 
 #ifdef TIMER
-time_start = GetTickCount() - time_start;
-printf("\n hash %d.%03d sec", time_start / 1000, time_start % 1000);
-if (time_start > 0){
-	time_start = (int)((total_file_size * 125) / ((__int64)time_start * 131072));
-	printf(", %d MB/s\n", time_start);
+time_start = clock() - time_start;
+time_sec = (double)time_start / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = (double)total_file_size / (time_sec * 1048576);
 } else {
-	printf("\n");
+	time_speed = 0;
 }
+printf("\n hash %.3f sec, %.0f MB/s\n", time_sec, time_speed);
 #endif
 
 error_end:

source/par2j/md5_crc.c

@@ -1,5 +1,5 @@
 // md5_crc.c
-// Copyright : 2023-10-29 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <stdio.h>
@@ -21,7 +21,6 @@
 #include "phmd5.h"
 #include "md5_crc.h"
 
-
 /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
 // バイト配列の MD5 ハッシュ値を求める
@@ -200,8 +199,10 @@ int file_md5_crc32_block(
 //#define TIMER // 実験用
 
 #ifdef TIMER
-static unsigned int time_start, time1_start;
-static unsigned int time_total = 0, time2_total = 0, time3_total = 0;
+#include <time.h>
+static double time_sec, time_speed;
+static clock_t time_start, time1_start;
+static clock_t time_total = 0, time2_total = 0, time3_total = 0;
 #endif
 
 #define MAX_BUF_SIZE	2097152	// ヒープ領域を使う場合の最大サイズ
@@ -224,7 +225,7 @@ int file_hash_crc(
 	HANDLE hFile;
 	OVERLAPPED ol;
 #ifdef TIMER
-time1_start = GetTickCount();
+time1_start = clock();
 #endif
 
 	// ソース・ファイルを開く
@@ -251,11 +252,11 @@ time1_start = GetTickCount();
 	if (file_left < IO_SIZE)
 		read_size = (unsigned int)file_left;
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	off = ReadFile(hFile, buf1, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 	if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 		print_win32_err();
@@ -281,11 +282,11 @@ time2_total += GetTickCount() - time_start;
 			ol.OffsetHigh = (unsigned int)(file_off >> 32);
 			file_off += IO_SIZE;
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 			off = ReadFile(hFile, buf, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 			if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 				print_win32_err();
@@ -301,7 +302,7 @@ time2_total += GetTickCount() - time_start;
 		}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		off = 0;	// チェックサム計算
 		if (block_left > 0){	// 前回足りなかった分を追加する
@@ -338,7 +339,7 @@ time_start = GetTickCount();
 			}
 		}
 #ifdef TIMER
-time3_total += GetTickCount() - time_start;
+time3_total += clock() - time_start;
 #endif
 
 		// 経過表示
@@ -369,16 +370,17 @@ error_end:
 		CloseHandle(ol.hEvent);
 
 #ifdef TIMER
-time_total += GetTickCount() - time1_start;
+time_total += clock() - time1_start;
 if (*prog_now == total_file_size){
-	printf("\nread  %d.%03d sec\n", time2_total / 1000, time2_total % 1000);
-	printf("main  %d.%03d sec\n", time3_total / 1000, time3_total % 1000);
-	if (time_total > 0){
-		time_start = (int)((total_file_size * 125) / ((__int64)time_total * 131072));
+	printf("\nread  %.3f sec\n", (double)time2_total / CLOCKS_PER_SEC);
+	printf("main  %.3f sec\n", (double)time3_total / CLOCKS_PER_SEC);
+	time_sec = (double)time_total / CLOCKS_PER_SEC;
+	if (time_sec > 0){
+		time_speed = (double)total_file_size / (time_sec * 1048576);
 	} else {
-		time_start = 0;
+		time_speed = 0;
 	}
-	printf("total %d.%03d sec, %d MB/s\n", time_total / 1000, time_total % 1000, time_start);
+	printf("total %.3f sec, %.0f MB/s\n", time_sec, time_speed);
 }
 #endif
 	return err;
@@ -403,7 +405,7 @@ int file_hash_crc(
 	HANDLE hFile;
 	OVERLAPPED ol;
 #ifdef TIMER
-time1_start = GetTickCount();
+time1_start = clock();
 #endif
 
 	// ソース・ファイルを開く
@@ -442,11 +444,11 @@ error_retry_read:
 	if (file_left < IO_SIZE)
 		read_size = (unsigned int)file_left;
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	off = ReadFile(hFile, buf1, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 	if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 		print_win32_err();
@@ -536,11 +538,11 @@ error_retry_pause:
 			ol.OffsetHigh = (unsigned int)(file_off >> 32);
 			file_off += IO_SIZE;
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 			off = ReadFile(hFile, buf, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 			if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 				print_win32_err();
@@ -557,7 +559,7 @@ time2_total += GetTickCount() - time_start;
 		}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		off = 0;	// チェックサム計算
 		if (block_left > 0){	// 前回足りなかった分を追加する
@@ -594,7 +596,7 @@ time_start = GetTickCount();
 			}
 		}
 #ifdef TIMER
-time3_total += GetTickCount() - time_start;
+time3_total += clock() - time_start;
 #endif
 
 		// 経過表示
@@ -625,16 +627,17 @@ error_end:
 		CloseHandle(ol.hEvent);
 
 #ifdef TIMER
-time_total += GetTickCount() - time1_start;
+time_total += clock() - time1_start;
 if (*prog_now == total_file_size){
-	printf("\nread  %d.%03d sec\n", time2_total / 1000, time2_total % 1000);
-	printf("main  %d.%03d sec\n", time3_total / 1000, time3_total % 1000);
-	if (time_total > 0){
-		time_start = (int)((total_file_size * 125) / ((__int64)time_total * 131072));
+	printf("\nread  %.3f sec\n", (double)time2_total / CLOCKS_PER_SEC);
+	printf("main  %.3f sec\n", (double)time3_total / CLOCKS_PER_SEC);
+	time_sec = (double)time_total / CLOCKS_PER_SEC;
+	if (time_sec > 0){
+		time_speed = (double)total_file_size / (time_sec * 1048576);
 	} else {
-		time_start = 0;
+		time_speed = 0;
 	}
-	printf("total %d.%03d sec, %d MB/s\n", time_total / 1000, time_total % 1000, time_start);
+	printf("total %.3f sec, %.0f MB/s\n", time_sec, time_speed);
 }
 #endif
 	return err;
@@ -660,7 +663,7 @@ int file_hash_crc(
 	HANDLE hFile;
 	OVERLAPPED ol;
 #ifdef TIMER
-time1_start = GetTickCount();
+time1_start = clock();
 #endif
 
 	// ソース・ファイルを開く
@@ -674,7 +677,7 @@ time1_start = GetTickCount();
 
 	// バッファー・サイズが大きいのでヒープ領域を使う
 	for (io_size = IO_SIZE; io_size <= MAX_BUF_SIZE; io_size += IO_SIZE){	// IO_SIZE の倍数にする
-		if ((io_size + IO_SIZE > (cpu_cache & 0xFFFE0000)) || ((__int64)(io_size + IO_SIZE) * 4 > file_left))
+		if ((io_size + IO_SIZE > (cpu_cache & 0xFFFF0000)) || ((__int64)(io_size + IO_SIZE) * 4 > file_left))
 			break;
 	}
 	buf1 = _aligned_malloc(io_size * 2, 64);
@@ -699,11 +702,11 @@ time1_start = GetTickCount();
 	if (file_left < io_size)
 		read_size = (unsigned int)file_left;
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	off = ReadFile(hFile, buf1, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 	if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 		print_win32_err();
@@ -729,11 +732,11 @@ time2_total += GetTickCount() - time_start;
 			ol.OffsetHigh = (unsigned int)(file_off >> 32);
 			file_off += io_size;
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 			off = ReadFile(hFile, buf, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 			if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 				print_win32_err();
@@ -749,7 +752,7 @@ time2_total += GetTickCount() - time_start;
 		}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		off = 0;	// チェックサム計算
 		if (block_left > 0){	// 前回足りなかった分を追加する
@@ -786,7 +789,7 @@ time_start = GetTickCount();
 			}
 		}
 #ifdef TIMER
-time3_total += GetTickCount() - time_start;
+time3_total += clock() - time_start;
 #endif
 
 		// 経過表示
@@ -819,16 +822,17 @@ error_end:
 		_aligned_free(buf1);
 
 #ifdef TIMER
-time_total += GetTickCount() - time1_start;
+time_total += clock() - time1_start;
 if (*prog_now == total_file_size){
-	printf("\nread  %d.%03d sec\n", time2_total / 1000, time2_total % 1000);
-	printf("main  %d.%03d sec\n", time3_total / 1000, time3_total % 1000);
-	if (time_total > 0){
-		time_start = (int)((total_file_size * 125) / ((__int64)time_total * 131072));
+	printf("\nread  %.3f sec\n", (double)time2_total / CLOCKS_PER_SEC);
+	printf("main  %.3f sec\n", (double)time3_total / CLOCKS_PER_SEC);
+	time_sec = (double)time_total / CLOCKS_PER_SEC;
+	if (time_sec > 0){
+		time_speed = (double)total_file_size / (time_sec * 1048576);
 	} else {
-		time_start = 0;
+		time_speed = 0;
 	}
-	printf("total %d.%03d sec, %d MB/s\n", time_total / 1000, time_total % 1000, time_start);
+	printf("total %.3f sec, %.0f MB/s\n", time_sec, time_speed);
 }
 #endif
 	return err;
@@ -869,7 +873,7 @@ DWORD WINAPI file_hash_crc2(LPVOID lpParameter)
 	// バッファー・サイズが大きいのでヒープ領域を使う
 	prog_tick = 1;
 	for (io_size = IO_SIZE; io_size <= MAX_BUF_SIZE; io_size += IO_SIZE){	// IO_SIZE の倍数にする
-		if ((io_size + IO_SIZE > (cpu_cache & 0xFFFE0000)) || ((__int64)(io_size + IO_SIZE) * 4 > file_left))
+		if ((io_size + IO_SIZE > (cpu_cache & 0xFFFF0000)) || ((__int64)(io_size + IO_SIZE) * 4 > file_left))
 			break;
 		prog_tick++;
 	}
@@ -1038,7 +1042,7 @@ int file_hash_check(
 	PHMD5 hash_ctx, block_ctx;
 	OVERLAPPED ol;
 #ifdef TIMER
-time1_start = GetTickCount();
+time1_start = clock();
 #endif
 
 	prog_last = -1;	// 検証中のファイル名を毎回表示する
@@ -1062,11 +1066,11 @@ time1_start = GetTickCount();
 		file_left = file_size - 16384;	// 本来のファイル・サイズまでしか検査しない
 	}
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	off = ReadFile(hFile, buf, len, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 	if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 		print_win32_err();
@@ -1141,11 +1145,11 @@ time2_total += GetTickCount() - time_start;
 	if (file_left < IO_SIZE)
 		read_size = (unsigned int)file_left;
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	off = ReadFile(hFile, buf1, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 	if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 		print_win32_err();
@@ -1168,11 +1172,11 @@ time2_total += GetTickCount() - time_start;
 			ol.OffsetHigh = (unsigned int)(file_off >> 32);
 			file_off += IO_SIZE;
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 			off = ReadFile(hFile, buf, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 			if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 				print_win32_err();
@@ -1187,7 +1191,7 @@ time2_total += GetTickCount() - time_start;
 		}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		if (s_blk != NULL){
 			off = 0;
@@ -1230,7 +1234,7 @@ time_start = GetTickCount();
 			Phmd5Process(&hash_ctx, buf, len);	// MD5 計算
 		}
 #ifdef TIMER
-time3_total += GetTickCount() - time_start;
+time3_total += clock() - time_start;
 #endif
 
 		// 経過表示
@@ -1267,15 +1271,16 @@ error_end:
 		CloseHandle(ol.hEvent);
 
 #ifdef TIMER
-time_total += GetTickCount() - time1_start;
-	printf("\nread  %d.%03d sec\n", time2_total / 1000, time2_total % 1000);
-	printf("main  %d.%03d sec\n", time3_total / 1000, time3_total % 1000);
-	if (time_total > 0){
-		time_start = (int)((file_size * 125) / ((__int64)time_total * 131072));
+time_total += clock() - time1_start;
+	printf("\nread  %.3f sec\n", (double)time2_total / CLOCKS_PER_SEC);
+	printf("main  %.3f sec\n", (double)time3_total / CLOCKS_PER_SEC);
+	time_sec = (double)time_total / CLOCKS_PER_SEC;
+	if (time_sec > 0){
+		time_speed = (double)file_size / (time_sec * 1048576);
 	} else {
-		time_start = 0;
+		time_speed = 0;
 	}
-	printf("total %d.%03d sec, %d MB/s\n", time_total / 1000, time_total % 1000, time_start);
+	printf("total %.3f sec, %.0f MB/s\n", time_sec, time_speed);
 #endif
 	return comp_num;
 }
@@ -1306,7 +1311,7 @@ DWORD WINAPI file_hash_background(LPVOID lpParameter)
 
 	// バッファー・サイズが大きいのでヒープ領域を使う
 	for (io_size = IO_SIZE; io_size <= MAX_BUF_SIZE; io_size += IO_SIZE){	// IO_SIZE の倍数にする
-		if ((io_size + IO_SIZE > (cpu_cache & 0xFFFE0000)) || ((__int64)(io_size + IO_SIZE) * 4 > file_size))
+		if ((io_size + IO_SIZE > (cpu_cache & 0xFFFF0000)) || ((__int64)(io_size + IO_SIZE) * 4 > file_size))
 			break;
 	}
 	//printf("\n io_size = %d\n", io_size);
@@ -1536,7 +1541,7 @@ int file_hash_direct(
 	HANDLE hFile;
 	OVERLAPPED ol;
 #ifdef TIMER
-time1_start = GetTickCount();
+time1_start = clock();
 #endif
 
 	prog_last = -1;	// 検証中のファイル名を毎回表示する
@@ -1592,11 +1597,11 @@ time1_start = GetTickCount();
 		file_left = file_size - 16384;	// 本来のファイル・サイズまでしか検査しない
 	}
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	off = ReadFile(hFile, buf, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 	if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 		comp_num = -1;
@@ -1679,11 +1684,11 @@ time2_total += GetTickCount() - time_start;
 		read_size = (read_size + 4095) & ~4095;	// 4KB の倍数にする
 	}
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	off = ReadFile(hFile, buf1, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 	if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 		print_win32_err();
@@ -1710,11 +1715,11 @@ time2_total += GetTickCount() - time_start;
 			ol.OffsetHigh = (unsigned int)(file_off >> 32);
 			file_off += IO_SIZE;
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 			off = ReadFile(hFile, buf, read_size, NULL, &ol);
 #ifdef TIMER
-time2_total += GetTickCount() - time_start;
+time2_total += clock() - time_start;
 #endif
 			if ((off == 0) && (GetLastError() != ERROR_IO_PENDING)){
 				print_win32_err();
@@ -1729,7 +1734,7 @@ time2_total += GetTickCount() - time_start;
 		}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		if (s_blk != NULL){
 			off = 0;
@@ -1771,7 +1776,7 @@ time_start = GetTickCount();
 			Phmd5Process(&hash_ctx, buf, len);	// MD5 計算
 		}
 #ifdef TIMER
-time3_total += GetTickCount() - time_start;
+time3_total += clock() - time_start;
 #endif
 
 		// 経過表示
@@ -1812,10 +1817,16 @@ error_end:
 		_aligned_free(buf1);
 
 #ifdef TIMER
-time_total += GetTickCount() - time1_start;
-	printf("\nread  %d.%03d sec\n", time2_total / 1000, time2_total % 1000);
-	printf("main  %d.%03d sec\n", time3_total / 1000, time3_total % 1000);
-	printf("total %d.%03d sec\n", time_total / 1000, time_total % 1000);
+time_total += clock() - time1_start;
+	printf("\nread  %.3f sec\n", (double)time2_total / CLOCKS_PER_SEC);
+	printf("main  %.3f sec\n", (double)time3_total / CLOCKS_PER_SEC);
+	time_sec = (double)time_total / CLOCKS_PER_SEC;
+	if (time_sec > 0){
+		time_speed = (double)file_size / (time_sec * 1048576);
+	} else {
+		time_speed = 0;
+	}
+	printf("total %.3f sec, %.0f MB/s\n", time_sec, time_speed);
 #endif
 	return comp_num;
 }

source/par2j/par2.c

@@ -1,5 +1,5 @@
 // par2.c
-// Copyright : 2023-10-15 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <malloc.h>

source/par2j/par2_cmd.c

@@ -1,5 +1,5 @@
 // par2_cmd.c
-// Copyright : 2023-10-15 Yutaka Sawada
+// Copyright : 2025-03-12 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <stdio.h>
@@ -86,7 +86,7 @@ static void print_environment(void)
 
 	printf("CPU thread\t: %d / %d\n", cpu_num & 0xFFFF, cpu_num >> 24);
 	cpu_num &= 0xFFFF;	// 利用するコア数だけにしておく
-	printf("CPU cache limit : %d KB, %d KB\n", (cpu_flag & 0xFFFF0000) >> 10, (cpu_cache & 0xFFFE0000) >> 10);
+	printf("CPU cache limit : %d KB, %d KB (%d)\n", (cpu_flag & 0xFFFF0000) >> 10, (cpu_cache & 0xFFFF0000) >> 10, cpu_cache & 0xFFFF);
 #ifndef _WIN64	// 32-bit 版は MMX, SSE2, SSSE3, AVX2 のどれかを表示する
 	printf("CPU extra\t:");
 	if (((cpu_flag & 16) != 0) && ((cpu_flag & 256) == 0)){
@@ -1448,14 +1448,12 @@ ri= switch_set & 0x00040000
 			} else if (wcsncmp(tmp_p, L"lr", 2) == 0){
 				recovery_limit = 0;
 				j = 2;
-				while ((j < 2 + 6) && (tmp_p[j] >= '0') && (tmp_p[j] <= '9')){
+				while ((j < 2 + 10) && (tmp_p[j] >= '0') && (tmp_p[j] <= '9')){
 					recovery_limit = (recovery_limit * 10) + (tmp_p[j] - '0');
 					j++;
 				}
 				if (recovery_limit <= 0)	// -lr0 ならソース・ファイルの最大ブロック数と同じにする
 					recovery_limit = -1;
-				if (recovery_limit > MAX_PARITY_NUM)
-					recovery_limit = MAX_PARITY_NUM;
 			} else if (wcsncmp(tmp_p, L"ls", 2) == 0){
 				__int64 num8 = 0;
 				j = 2;
@@ -1477,50 +1475,77 @@ ri= switch_set & 0x00040000
 					switch_v |= j;
 			// 共通のオプション (数値)
 			} else if (wcsncmp(tmp_p, L"lc", 2) == 0){
-				k = 0;
-				j = 2;
-				while ((j < 2 + 5) && (tmp_p[j] >= '0') && (tmp_p[j] <= '9')){
-					k = (k * 10) + (tmp_p[j] - '0');
-					j++;
-				}
-				if (k & 256){	// GPU を使う
-					OpenCL_method = 1;	// Faster GPU
-				} else if (k & 512){
-					OpenCL_method = -1;	// Slower GPU
-				}
-				if (k & 1024)	// CLMUL と ALTMAP を使わない
-					cpu_flag = (cpu_flag & 0xFFFFFFF7) | 256;
-				if (k & 2048)	// JIT(SSE2) を使わない
-					cpu_flag &= 0xFFFFFF7F;
-				if (k & 4096)	// SSSE3 を使わない
-					cpu_flag &= 0xFFFFFFFE;
-				if (k & 8192)	// AVX2 を使わない
-					cpu_flag &= 0xFFFFFFEF;
-				if (k & 255){	// 使用するコア数を変更する
-					k &= 255;	// 1～255 の範囲
-					// printf("\n lc# = %d , logical = %d, physical = %d \n", k, cpu_num >> 24, (cpu_num & 0x00FF0000) >> 16);
-					if (k == 251){	// 物理コア数の 1/4 にする
-						k = ((cpu_num & 0x00FF0000) >> 16) / 4;
-					} else if (k == 252){	// 物理コア数の半分にする
-						k = ((cpu_num & 0x00FF0000) >> 16) / 2;
-					} else if (k == 253){	// 物理コア数の 3/4 にする
-						k = (((cpu_num & 0x00FF0000) >> 16) * 3) / 4;
-					} else if (k == 254){	// 物理コア数より減らす
-						k = ((cpu_num & 0x00FF0000) >> 16) - 1;
-					} else if (k == 255){	// 物理コア数より増やす
-						k = cpu_num >> 16;
-						k = ((k & 0xFF) + (k >> 8)) / 2;	// 物理コア数と論理コア数の中間にする？
-						// タスクマネージャーにおける CPU使用率は 100%になるけど、速くはならない・・・
-						// k = (k & 0xFF) + ((k >> 8) - (k & 0xFF)) / 4;	// 物理コア数の 5/4 にする？
+				if (tmp_p[2] == 'b'){	// Size of Cache Blocking (CPU's L2 cache optimization)
+					k = 0;
+					j = 3;
+					while ((j < 3 + 5) && (tmp_p[j] >= '0') && (tmp_p[j] <= '9')){
+						k = (k * 10) + (tmp_p[j] - '0');
+						j++;
 					}
-					if (k > MAX_CPU){
-						k = MAX_CPU;
-					} else if (k < 1){
-						k = 1;
-					} else if (k > (cpu_num >> 24)){
-						k = cpu_num >> 24;	// 論理コア数を超えないようにする
+					if (k <= 0x7FFF)	// 上位 16-bit に上書きする
+						cpu_flag = (cpu_flag & 0xFFFF) | (k << 16);
+				} else if (tmp_p[2] == 's'){	// Size of Shared Cache
+					k = 0;
+					j = 3;
+					while ((j < 3 + 5) && (tmp_p[j] >= '0') && (tmp_p[j] <= '9')){
+						k = (k * 10) + (tmp_p[j] - '0');
+						j++;
+					}
+					if (k <= 0xFFFF)	// 上位 16-bit に上書きする
+						cpu_cache = (cpu_cache & 0xFFFF) | (k << 16);
+				} else if (tmp_p[2] == 'm'){	// Max number of chunks (CPU's shared L3 cache optimization)
+					k = 0;
+					j = 3;
+					while ((j < 3 + 5) && (tmp_p[j] >= '0') && (tmp_p[j] <= '9')){
+						k = (k * 10) + (tmp_p[j] - '0');
+						j++;
+					}
+					if (k <= 0x8000)	// CACHE_MIN_NUM 未満なら 0x8000 になる
+						cpu_cache = (cpu_cache & 0xFFFF0000) | k;	// 下位 16-bit に上書きする
+				} else {	// Extra と GPU も別にしてもいいかも？
+					k = 0;
+					j = 2;
+					while ((j < 2 + 7) && (tmp_p[j] >= '0') && (tmp_p[j] <= '9')){
+						k = (k * 10) + (tmp_p[j] - '0');
+						j++;
+					}
+					if (k & 0x300){	// GPU を使う
+						OpenCL_method = k & 0x003F0300;
+					}
+					if (k & 1024)	// CLMUL と ALTMAP を使わない
+						cpu_flag = (cpu_flag & 0xFFFFFFF7) | 256;
+					if (k & 2048)	// JIT(SSE2) を使わない
+						cpu_flag &= 0xFFFFFF7F;
+					if (k & 4096)	// SSSE3 を使わない
+						cpu_flag &= 0xFFFFFFFE;
+					if (k & 8192)	// AVX2 を使わない
+						cpu_flag &= 0xFFFFFFEF;
+					if (k & 255){	// 使用するコア数を変更する
+						k &= 255;	// 1～255 の範囲
+						// printf("\n lc# = %d , logical = %d, physical = %d \n", k, cpu_num >> 24, (cpu_num & 0x00FF0000) >> 16);
+						if (k == 251){	// 物理コア数の 1/4 にする
+							k = ((cpu_num & 0x00FF0000) >> 16) / 4;
+						} else if (k == 252){	// 物理コア数の半分にする
+							k = ((cpu_num & 0x00FF0000) >> 16) / 2;
+						} else if (k == 253){	// 物理コア数の 3/4 にする
+							k = (((cpu_num & 0x00FF0000) >> 16) * 3) / 4;
+						} else if (k == 254){	// 物理コア数より減らす
+							k = ((cpu_num & 0x00FF0000) >> 16) - 1;
+						} else if (k == 255){	// 物理コア数より増やす
+							k = cpu_num >> 16;
+							k = ((k & 0xFF) + (k >> 8)) / 2;	// 物理コア数と論理コア数の中間にする？
+							// タスクマネージャーにおける CPU使用率は 100%になるけど、速くはならない・・・
+							// k = (k & 0xFF) + ((k >> 8) - (k & 0xFF)) / 4;	// 物理コア数の 5/4 にする？
+						}
+						if (k > MAX_CPU){
+							k = MAX_CPU;
+						} else if (k < 1){
+							k = 1;
+						} else if (k > (cpu_num >> 24)){
+							k = cpu_num >> 24;	// 論理コア数を超えないようにする
+						}
+						cpu_num = (cpu_num & 0xFFFF0000) | k;	// 指定されたコア数を下位に配置する
 					}
-					cpu_num = (cpu_num & 0xFFFF0000) | k;	// 指定されたコア数を下位に配置する
 				}
 			} else if (wcsncmp(tmp_p, L"m", 1) == 0){
 				memory_use = 0;
@@ -1950,6 +1975,14 @@ fclose(fp);
 					if (recovery_limit > j)
 						recovery_limit = j;
 				}
+			} else {
+				if (split_size == 2){	// 制限をブロック数ではなく、サイズとして認識する
+					recovery_limit /= block_size;
+					if (recovery_limit < 1)
+						recovery_limit = 1;
+				}
+				if (recovery_limit > MAX_PARITY_NUM)	// パリティ・ブロック数の最大値を超えない
+					recovery_limit = MAX_PARITY_NUM;
 			}
 			// パリティ・ブロックをリカバリ・ファイルに分配する方法
 			i = (switch_set & 0x30000) >> 16;

source/par2j/reedsolomon.c

@@ -1,5 +1,5 @@
 // reedsolomon.c
-// Copyright : 2023-10-26 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <malloc.h>
@@ -27,6 +27,9 @@
 #include "rs_decode.h"
 #include "reedsolomon.h"
 
+#ifdef TIMER
+#include <time.h>
+#endif
 
 // GPU を使う最小データサイズ (MB 単位)
 // GPU の起動には時間がかかるので、データが小さすぎると逆に遅くなる
@@ -739,7 +742,7 @@ int rs_encode(
 	int err = 0;
 	unsigned int len;
 #ifdef TIMER
-unsigned int time_total = GetTickCount();
+clock_t time_total = clock();
 #endif
 
 	if (galois_create_table()){
@@ -755,7 +758,7 @@ unsigned int time_total = GetTickCount();
 	// パリティ計算用の行列演算の準備をする
 	len = sizeof(unsigned short) * source_num;
 	if (OpenCL_method != 0)
-		len *= 2;	// GPU の作業領域も確保しておく
+		len *= 3;	// GPU の作業領域も確保しておく
 	constant = malloc(len);
 	if (constant == NULL){
 		printf("malloc, %d\n", len);
@@ -799,8 +802,8 @@ unsigned int time_total = GetTickCount();
 		err = encode_method2(file_path, header_buf, rcv_hFile, files, s_blk, p_blk, constant);
 #ifdef TIMER
 	if (err != 1){
-		time_total = GetTickCount() - time_total;
-		printf("total  %d.%03d sec\n", time_total / 1000, time_total % 1000);
+		time_total = clock() - time_total;
+		printf("total  %.3f sec\n", (double)time_total / CLOCKS_PER_SEC);
 	}
 #endif
 
@@ -830,7 +833,7 @@ int rs_encode_1pass(
 	int err = 0;
 	unsigned int len;
 #ifdef TIMER
-unsigned int time_total = GetTickCount();
+clock_t time_total = clock();
 #endif
 
 	if (galois_create_table()){
@@ -841,7 +844,7 @@ unsigned int time_total = GetTickCount();
 	// パリティ計算用の行列演算の準備をする
 	len = sizeof(unsigned short) * source_num;
 	if (OpenCL_method != 0)
-		len *= 2;	// GPU の作業領域も確保しておく
+		len *= 3;	// GPU の作業領域も確保しておく
 	constant = malloc(len);
 	if (constant == NULL){
 		printf("malloc, %d\n", len);
@@ -888,8 +891,8 @@ unsigned int time_total = GetTickCount();
 	if (err < 0){
 		printf("switching to 2-pass processing, %d\n", err);
 	} else if (err != 1){
-		time_total = GetTickCount() - time_total;
-		printf("total  %d.%03d sec\n", time_total / 1000, time_total % 1000);
+		time_total = clock() - time_total;
+		printf("total  %.3f sec\n", (double)time_total / CLOCKS_PER_SEC);
 	}
 #endif
 
@@ -913,7 +916,7 @@ int rs_decode(
 	int err = 0, i, j, k;
 	unsigned int len;
 #ifdef TIMER
-unsigned int time_matrix = 0, time_total = GetTickCount();
+clock_t time_matrix = 0, time_total = clock();
 #endif
 
 	if (galois_create_table()){
@@ -948,7 +951,7 @@ unsigned int time_matrix = 0, time_total = GetTickCount();
 	id = mat + (block_lost * source_num);
 
 #ifdef TIMER
-time_matrix = GetTickCount();
+time_matrix = clock();
 #endif
 	// 復元用の行列を計算する
 	print_progress_text(0, "Computing matrix");
@@ -989,7 +992,7 @@ time_matrix = GetTickCount();
 	//for (i = 0; i < block_lost; i++)
 	//	printf("id[%d] = %d\n", i, id[i]);
 #ifdef TIMER
-time_matrix = GetTickCount() - time_matrix;
+time_matrix = clock() - time_matrix;
 #endif
 
 #ifdef TIMER
@@ -1032,9 +1035,9 @@ time_matrix = GetTickCount() - time_matrix;
 		err = decode_method2(file_path, block_lost, rcv_hFile, files, s_blk, p_blk, mat);
 #ifdef TIMER
 	if (err != 1){
-		time_total = GetTickCount() - time_total;
-		printf("total  %d.%03d sec\n", time_total / 1000, time_total % 1000);
-		printf("matrix %d.%03d sec\n", time_matrix / 1000, time_matrix % 1000);
+		time_total = clock() - time_total;
+		printf("total  %.3f sec\n", (double)time_total / CLOCKS_PER_SEC);
+		printf("matrix %.3f sec\n", (double)time_matrix / CLOCKS_PER_SEC);
 	}
 #endif
 

source/par2j/repair.c

@@ -1,5 +1,5 @@
 // repair.c
-// Copyright : 2022-10-14 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <stdio.h>

source/par2j/rs_decode.c

@@ -1,5 +1,5 @@
 // rs_decode.c
-// Copyright : 2023-10-29 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <malloc.h>
@@ -28,7 +28,9 @@
 
 
 #ifdef TIMER
-static unsigned int time_start, time_read = 0, time_write = 0, time_calc = 0;
+#include <time.h>
+static double time_sec, time_speed;
+static clock_t time_start, time_read = 0, time_write = 0, time_calc = 0;
 static unsigned int read_count, write_count = 0, skip_count;
 #endif
 
@@ -60,7 +62,7 @@ static DWORD WINAPI thread_decode2(LPVOID lpParameter)
 	RS_TH *th;
 #ifdef TIMER
 unsigned int loop_count2a = 0, loop_count2b = 0;
-unsigned int time_start2, time_encode2a = 0, time_encode2b = 0;
+clock_t time_start2, time_encode2a = 0, time_encode2b = 0;
 #endif
 
 	th = (RS_TH *)lpParameter;
@@ -78,7 +80,7 @@ unsigned int time_start2, time_encode2a = 0, time_encode2b = 0;
 	WaitForSingleObject(hRun, INFINITE);	// 計算開始の合図を待つ
 	while (th->now < INT_MAX / 2){
 #ifdef TIMER
-time_start2 = GetTickCount();
+time_start2 = clock();
 #endif
 		s_buf = th->buf;
 		factor = th->mat;
@@ -95,7 +97,7 @@ loop_count2a++;
 #endif
 			}
 #ifdef TIMER
-time_encode2a += GetTickCount() - time_start2;
+time_encode2a += clock() - time_start2;
 #endif
 		} else {	// 消失ブロックを部分的に保持する場合
 			// スレッドごとに復元する消失ブロックの chunk を変える
@@ -136,7 +138,7 @@ loop_count2b += src_num;
 #endif
 			}
 #ifdef TIMER
-time_encode2b += GetTickCount() - time_start2;
+time_encode2b += clock() - time_start2;
 #endif
 		}
 		//_mm_sfence();	// メモリーへの書き込みを完了する
@@ -146,19 +148,21 @@ time_encode2b += GetTickCount() - time_start2;
 #ifdef TIMER
 loop_count2b /= chunk_num;	// chunk数で割ってブロック数にする
 printf("sub-thread : total loop = %d\n", loop_count2a + loop_count2b);
-if (time_encode2a > 0){
-	i = (int)((__int64)loop_count2a * unit_size * 125 / ((__int64)time_encode2a * 131072));
+time_sec = (double)time_encode2a / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2a * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
 if (loop_count2a > 0)
-	printf(" 1st decode %d.%03d sec, %d loop, %d MB/s\n", time_encode2a / 1000, time_encode2a % 1000, loop_count2a, i);
-if (time_encode2b > 0){
-	i = (int)((__int64)loop_count2b * unit_size * 125 / ((__int64)time_encode2b * 131072));
+	printf(" 1st decode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2a, time_speed);
+time_sec = (double)time_encode2b / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2b * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
-printf(" 2nd decode %d.%03d sec, %d loop, %d MB/s\n", time_encode2b / 1000, time_encode2b % 1000, loop_count2b, i);
+printf(" 2nd decode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2b, time_speed);
 #endif
 
 	// 終了処理
@@ -178,7 +182,7 @@ static DWORD WINAPI thread_decode3(LPVOID lpParameter)
 	RS_TH *th;
 #ifdef TIMER
 unsigned int loop_count2a = 0, loop_count2b = 0;
-unsigned int time_start2, time_encode2a = 0, time_encode2b = 0;
+clock_t time_start2, time_encode2a = 0, time_encode2b = 0;
 #endif
 
 	th = (RS_TH *)lpParameter;
@@ -197,7 +201,7 @@ unsigned int time_start2, time_encode2a = 0, time_encode2b = 0;
 	WaitForSingleObject(hRun, INFINITE);	// 計算開始の合図を待つ
 	while (th->now < INT_MAX / 2){
 #ifdef TIMER
-time_start2 = GetTickCount();
+time_start2 = clock();
 #endif
 		s_buf = th->buf;
 		factor = th->mat;
@@ -214,7 +218,7 @@ loop_count2a++;
 #endif
 			}
 #ifdef TIMER
-time_encode2a += GetTickCount() - time_start2;
+time_encode2a += clock() - time_start2;
 #endif
 		} else {	// 全ての消失ブロックを保持する場合
 			// スレッドごとに復元する消失ブロックの chunk を変える
@@ -250,7 +254,7 @@ loop_count2b += src_num;
 #endif
 			}
 #ifdef TIMER
-time_encode2b += GetTickCount() - time_start2;
+time_encode2b += clock() - time_start2;
 #endif
 		}
 		//_mm_sfence();	// メモリーへの書き込みを完了する
@@ -260,19 +264,21 @@ time_encode2b += GetTickCount() - time_start2;
 #ifdef TIMER
 loop_count2b /= chunk_num;	// chunk数で割ってブロック数にする
 printf("sub-thread : total loop = %d\n", loop_count2a + loop_count2b);
-if (time_encode2a > 0){
-	i = (int)((__int64)loop_count2a * unit_size * 125 / ((__int64)time_encode2a * 131072));
+time_sec = (double)time_encode2a / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2a * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
 if (loop_count2a > 0)
-	printf(" 1st decode %d.%03d sec, %d loop, %d MB/s\n", time_encode2a / 1000, time_encode2a % 1000, loop_count2a, i);
-if (time_encode2b > 0){
-	i = (int)((__int64)loop_count2b * unit_size * 125 / ((__int64)time_encode2b * 131072));
+	printf(" 1st decode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2a, time_speed);
+time_sec = (double)time_encode2b / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2b * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
-printf(" 2nd decode %d.%03d sec, %d loop, %d MB/s\n", time_encode2b / 1000, time_encode2b % 1000, loop_count2b, i);
+printf(" 2nd decode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2b, time_speed);
 #endif
 
 	// 終了処理
@@ -292,7 +298,8 @@ static DWORD WINAPI thread_decode_gpu(LPVOID lpParameter)
 	HANDLE hRun, hEnd;
 	RS_TH *th;
 #ifdef TIMER
-unsigned int time_start2, time_encode2 = 0, loop_count2 = 0;
+unsigned int loop_count2 = 0;
+clock_t time_start2, time_encode2 = 0;
 #endif
 
 	th = (RS_TH *)lpParameter;
@@ -307,7 +314,7 @@ unsigned int time_start2, time_encode2 = 0, loop_count2 = 0;
 	WaitForSingleObject(hRun, INFINITE);	// 計算開始の合図を待つ
 	while (th->now < INT_MAX / 2){
 #ifdef TIMER
-time_start2 = GetTickCount();
+time_start2 = clock();
 #endif
 		// GPUはソース・ブロック読み込み中に呼ばれない
 		s_buf = th->buf;
@@ -321,22 +328,58 @@ time_start2 = GetTickCount();
 			InterlockedExchange(&(th->now), INT_MAX / 3);	// サブ・スレッドの計算を中断する
 		}
 
-		// スレッドごとに復元する消失ブロックを変える
-		while ((j = InterlockedIncrement(&(th->now))) < block_lost){	// j = ++th_now
-			// 倍率は逆行列から部分的にコピーする
-			i = gpu_multiply_blocks(src_num, factor + source_num * j, g_buf + (size_t)unit_size * j, unit_size);
-			if (i != 0){
-				th->len = i;
-				InterlockedExchange(&(th->now), INT_MAX / 3);	// サブ・スレッドの計算を中断する
-				break;
-			}
+		// 一つの GPUスレッドが全ての消失ブロックを処理する
+		if (OpenCL_method & 8){	// ２ブロックずつ計算する
+			// 消失ブロック数が奇数なら、最初の一個だけ別に計算する
+			if (block_lost & 1){
+				InterlockedIncrement(&(th->now));	// 常に j = 0 となる
+
+				// 倍率は逆行列から部分的にコピーする
+				i = gpu_multiply_blocks(src_num, factor, NULL, g_buf, unit_size);
+				if (i != 0){
+					th->len = i;
+					InterlockedExchange(&(th->now), INT_MAX / 3);	// サブ・スレッドの計算を中断する
+					break;
+				}
 
 #ifdef TIMER
 loop_count2 += src_num;
 #endif
+			}
+
+			// 残りのブロックは二個ずつ計算する
+			while ((j = InterlockedAdd(&(th->now), 2)) < block_lost){	// th_now += 2, j = th_now
+				j--;	// +2 してるから、最初のブロックは -1 する
+
+				// 倍率は逆行列から部分的に２回コピーする
+				i = gpu_multiply_blocks(src_num, factor + source_num * j, factor + source_num * (j + 1), g_buf + (size_t)unit_size * j, unit_size * 2);
+				if (i != 0){
+					th->len = i;
+					InterlockedExchange(&(th->now), INT_MAX / 3);	// サブ・スレッドの計算を中断する
+					break;
+				}
+#ifdef TIMER
+loop_count2 += src_num * 2;
+#endif
+			}
+
+		} else {	// 以前からの１ブロックずつ計算する方式
+			while ((j = InterlockedIncrement(&(th->now))) < block_lost){	// j = ++th_now
+				// 倍率は逆行列から部分的にコピーする（２ブロックずつの場合はブロック数をマイナスにする）
+				i = gpu_multiply_blocks(src_num, factor + source_num * j, NULL, g_buf + (size_t)unit_size * j, unit_size);
+				if (i != 0){
+					th->len = i;
+					InterlockedExchange(&(th->now), INT_MAX / 3);	// サブ・スレッドの計算を中断する
+					break;
+				}
+
+#ifdef TIMER
+loop_count2 += src_num;
+#endif
+			}
 		}
 #ifdef TIMER
-time_encode2 += GetTickCount() - time_start2;
+time_encode2 += clock() - time_start2;
 #endif
 		// 最後にVRAMを解放する
 		i = gpu_finish();
@@ -349,12 +392,13 @@ time_encode2 += GetTickCount() - time_start2;
 	}
 #ifdef TIMER
 printf("gpu-thread :\n");
-if (time_encode2 > 0){
-	i = (int)((__int64)loop_count2 * unit_size * 125 / ((__int64)time_encode2 * 131072));
+time_sec = (double)time_encode2 / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2 * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
-printf(" 2nd decode %d.%03d sec, %d loop, %d MB/s\n", time_encode2 / 1000, time_encode2 % 1000, loop_count2, i);
+printf(" 2nd decode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2, time_speed);
 #endif
 
 	// 終了処理
@@ -430,7 +474,7 @@ int decode_method1(	// ソース・ブロックが一個だけの場合
 	block_off = 0;
 	while (block_off < block_size){
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		// パリティ・ブロックを読み込む
 		len = block_size - block_off;
@@ -447,18 +491,18 @@ time_start = GetTickCount();
 		// パリティ・ブロックのチェックサムを計算する
 		checksum16_altmap(buf, buf + io_size, io_size);
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		// 失われたソース・ブロックを復元する
 		memset(work_buf, 0, unit_size);
 		// factor で割ると元に戻る
 		galois_align_multiply(buf, work_buf, unit_size, galois_divide(1, galois_power(2, id)));
 #ifdef TIMER
-time_calc += GetTickCount() - time_start;
+time_calc += clock() - time_start;
 #endif
 
 		// 経過表示
@@ -472,7 +516,7 @@ time_calc += GetTickCount() - time_start;
 		}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		// 復元されたソース・ブロックのチェックサムを検証する
 		checksum16_return(work_buf, hash, io_size);
@@ -491,7 +535,7 @@ time_start = GetTickCount();
 			goto error_end;
 		}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 
 		block_off += io_size;
@@ -499,9 +543,9 @@ time_write += GetTickCount() - time_start;
 	print_progress_done();	// 末尾ブロックの断片化によっては 100% で完了するとは限らない
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
-printf("write  %d.%03d sec\n", time_write / 1000, time_write % 1000);
-printf("decode %d.%03d sec\n", time_calc / 1000, time_calc % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
+printf("write  %.3f sec\n", (double)time_write / CLOCKS_PER_SEC);
+printf("decode %.3f sec\n", (double)time_calc / CLOCKS_PER_SEC);
 #endif
 
 error_end:
@@ -623,7 +667,7 @@ int decode_method2(	// ソース・データを全て読み込む場合
 #ifdef TIMER
 read_count = 0;
 skip_count = 0;
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		last_file = -1;
 		recv_now = 0;	// 何番目の代替ブロックか
@@ -760,7 +804,7 @@ skip_count++;
 			hFile = NULL;
 		}
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 		WaitForMultipleObjects(cpu_num1, hEnd, TRUE, INFINITE);	// サブ・スレッドの計算終了の合図を待つ
@@ -845,7 +889,7 @@ skip_count++;
 			}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 			// 復元されたブロックを書き込む
 			work_buf = p_buf;
@@ -916,7 +960,7 @@ write_count++;
 				}
 			}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 
 			part_off += part_num;	// 次の消失ブロック位置にする
@@ -930,9 +974,9 @@ time_write += GetTickCount() - time_start;
 	print_progress_done();
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
 j = ((block_size + io_size - 1) / io_size) * block_lost;
-printf("write  %d.%03d sec, count = %d/%d\n", time_write / 1000, time_write % 1000, write_count, j);
+printf("write  %.3f sec, count = %d/%d\n", (double)time_write / CLOCKS_PER_SEC, write_count, j);
 if (prog_num != prog_base)
 	printf(" prog_num = %I64d, prog_base = %I64d\n", prog_num, prog_base);
 #endif
@@ -1063,7 +1107,7 @@ int decode_method3(	// 復元するブロックを全て保持できる場合
 
 #ifdef TIMER
 read_count = 0;
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		last_file = -1;
 		for (i = 0; i < read_num; i++){	// スライスを一個ずつ読み込んでメモリー上に配置していく
@@ -1173,7 +1217,7 @@ read_count++;
 			hFile = NULL;
 		}
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 		WaitForMultipleObjects(cpu_num1, hEnd, TRUE, INFINITE);	// サブ・スレッドの計算終了の合図を待つ
@@ -1238,7 +1282,7 @@ time_read += GetTickCount() - time_start;
 	}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	// 復元されたブロックを書き込む
 	work_buf = p_buf;
@@ -1297,7 +1341,7 @@ time_start = GetTickCount();
 		}
 	}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 	// 最後の書き込みファイルを閉じる
 	CloseHandle(hFile);
@@ -1305,8 +1349,8 @@ time_write += GetTickCount() - time_start;
 	print_progress_done();
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
-printf("write  %d.%03d sec\n", time_write / 1000, time_write % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
+printf("write  %.3f sec\n", (double)time_write / CLOCKS_PER_SEC);
 if (prog_num != prog_base)
 	printf(" prog_num = %I64d, prog_base = %I64d\n", prog_num, prog_base);
 #endif
@@ -1463,7 +1507,7 @@ int decode_method4(	// 全てのブロックを断片的に保持する場合 (G
 #ifdef TIMER
 read_count = 0;
 skip_count = 0;
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		last_file = -1;
 		recv_now = 0;	// 何番目の代替ブロックか
@@ -1600,7 +1644,7 @@ skip_count++;
 			hFile = NULL;
 		}
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 		memset(g_buf, 0, (size_t)unit_size * block_lost);	// 待機中に GPU用の領域をゼロ埋めしておく
@@ -1731,10 +1775,21 @@ skip_count++;
 #endif
 				} else if (src_off + src_num + src_max > source_num){
 					src_num = source_num - src_off - src_max;
-					if ((src_num < src_max) && (src_num + src_max <= vram_max) && (gpu_end * 2 > cpu_end)){
-						src_num += src_max;	// GPU担当量が少なくて、余裕がある場合は、残りも全て任せる
+					if (src_num < src_max){
+						if ((src_num + src_max <= vram_max) && (gpu_end * 2 > cpu_end)){
+							src_num += src_max;	// GPU担当量が少なくて、余裕がある場合は、残りも全て任せる
+#ifdef TIMER
+							printf("GPU last +: src_off = %d, src_num = %d + %d\n", src_off, src_num - src_max, src_max);
+#endif
+						} else if (src_num < src_max / 4){
+							src_num = src_max / 4;	// src_num が小さくなり過ぎないようにする
+#ifdef TIMER
+							printf("GPU last ?: src_off = %d, src_num = %d\n", src_off, src_num);
+						} else {
+							printf("GPU last -: src_off = %d, src_num = %d\n", src_off, src_num);
+#endif
+						}
 #ifdef TIMER
-						printf("GPU last +: src_off = %d, src_num = %d + %d\n", src_off, src_num - src_max, src_max);
 					} else {
 						printf("GPU last 2: src_off = %d, src_num = %d\n", src_off, src_num);
 #endif
@@ -1834,7 +1889,7 @@ skip_count++;
 			prog_num += th->size * block_lost;
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		// 復元されたブロックを書き込む
 		work_buf = p_buf;
@@ -1907,7 +1962,7 @@ write_count++;
 			}
 		}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 
 		block_off += io_size;
@@ -1918,9 +1973,9 @@ time_write += GetTickCount() - time_start;
 	print_progress_done();
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
 j = ((block_size + io_size - 1) / io_size) * block_lost;
-printf("write  %d.%03d sec, count = %d/%d\n", time_write / 1000, time_write % 1000, write_count, j);
+printf("write  %.3f sec, count = %d/%d\n", (double)time_write / CLOCKS_PER_SEC, write_count, j);
 if (prog_num != prog_base)
 	printf(" prog_num = %I64d, prog_base = %I64d\n", prog_num, prog_base);
 #endif
@@ -2085,7 +2140,7 @@ int decode_method5(	// 復元するブロックだけ保持する場合 (GPU対
 
 #ifdef TIMER
 read_count = 0;
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		last_file = -1;
 		for (i = 0; i < read_num; i++){	// スライスを一個ずつ読み込んでメモリー上に配置していく
@@ -2195,7 +2250,7 @@ read_count++;
 			hFile = NULL;
 		}
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 		if (source_off == 0)
@@ -2318,10 +2373,21 @@ time_read += GetTickCount() - time_start;
 #endif
 				} else if (src_off + src_num + src_max > read_num){
 					src_num = read_num - src_off - src_max;
-					if ((src_num < src_max) && (src_num + src_max <= vram_max) && (gpu_end * 2 > cpu_end)){
-						src_num += src_max;	// GPU担当量が少なくて、余裕がある場合は、残りも全て任せる
+					if (src_num < src_max){
+						if ((src_num + src_max <= vram_max) && (gpu_end * 2 > cpu_end)){
+							src_num += src_max;	// GPU担当量が少なくて、余裕がある場合は、残りも全て任せる
+#ifdef TIMER
+							printf("GPU last +: src_off = %d, src_num = %d + %d\n", src_off, src_num - src_max, src_max);
+#endif
+						} else if (src_num < src_max / 4){
+							src_num = src_max / 4;	// src_num が小さくなり過ぎないようにする
+#ifdef TIMER
+							printf("GPU last ?: src_off = %d, src_num = %d\n", src_off, src_num);
+						} else {
+							printf("GPU last -: src_off = %d, src_num = %d\n", src_off, src_num);
+#endif
+						}
 #ifdef TIMER
-						printf("GPU last +: src_off = %d, src_num = %d + %d\n", src_off, src_num - src_max, src_max);
 					} else {
 						printf("GPU last 2: src_off = %d, src_num = %d\n", src_off, src_num);
 #endif
@@ -2424,7 +2490,7 @@ time_read += GetTickCount() - time_start;
 	}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	// 復元されたブロックを書き込む
 	work_buf = p_buf;
@@ -2485,7 +2551,7 @@ time_start = GetTickCount();
 		}
 	}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 	// 最後の書き込みファイルを閉じる
 	CloseHandle(hFile);
@@ -2493,8 +2559,8 @@ time_write += GetTickCount() - time_start;
 	print_progress_done();
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
-printf("write  %d.%03d sec\n", time_write / 1000, time_write % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
+printf("write  %.3f sec\n", (double)time_write / CLOCKS_PER_SEC);
 if (prog_num != prog_base)
 	printf(" prog_num = %I64d, prog_base = %I64d\n", prog_num, prog_base);
 #endif

source/par2j/rs_encode.c

@@ -1,5 +1,5 @@
 // rs_encode.c
-// Copyright : 2023-10-29 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <malloc.h>
@@ -29,7 +29,9 @@
 
 
 #ifdef TIMER
-static unsigned int time_start, time_read = 0, time_write = 0, time_calc = 0;
+#include <time.h>
+static double time_sec, time_speed;
+static clock_t time_start, time_read = 0, time_write = 0, time_calc = 0;
 static unsigned int read_count, skip_count;
 #endif
 
@@ -61,7 +63,7 @@ static DWORD WINAPI thread_encode2(LPVOID lpParameter)
 	RS_TH *th;
 #ifdef TIMER
 unsigned int loop_count2a = 0, loop_count2b = 0;
-unsigned int time_start2, time_encode2a = 0, time_encode2b = 0;
+clock_t time_start2, time_encode2a = 0, time_encode2b = 0;
 #endif
 
 	th = (RS_TH *)lpParameter;
@@ -80,7 +82,7 @@ unsigned int time_start2, time_encode2a = 0, time_encode2b = 0;
 	WaitForSingleObject(hRun, INFINITE);	// 計算開始の合図を待つ
 	while (th->now < INT_MAX / 2){
 #ifdef TIMER
-time_start2 = GetTickCount();
+time_start2 = clock();
 #endif
 		s_buf = th->buf;
 		src_off = th->off;	// ソース・ブロック番号
@@ -98,7 +100,7 @@ loop_count2a++;
 			}
 
 #ifdef TIMER
-time_encode2a += GetTickCount() - time_start2;
+time_encode2a += clock() - time_start2;
 #endif
 		} else {	// パリティ・ブロックを部分的に保持する場合
 			// スレッドごとに作成するパリティ・ブロックの chunk を変える
@@ -143,7 +145,7 @@ loop_count2b += src_num;
 #endif
 			}
 #ifdef TIMER
-time_encode2b += GetTickCount() - time_start2;
+time_encode2b += clock() - time_start2;
 #endif
 		}
 		//_mm_sfence();	// メモリーへの書き込みを完了する
@@ -153,19 +155,21 @@ time_encode2b += GetTickCount() - time_start2;
 #ifdef TIMER
 loop_count2b /= chunk_num;	// chunk数で割ってブロック数にする
 printf("sub-thread : total loop = %d\n", loop_count2a + loop_count2b);
-if (time_encode2a > 0){
-	i = (int)((__int64)loop_count2a * unit_size * 125 / ((__int64)time_encode2a * 131072));
+time_sec = (double)time_encode2a / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2a * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
 if (loop_count2a > 0)
-	printf(" 1st encode %d.%03d sec, %d loop, %d MB/s\n", time_encode2a / 1000, time_encode2a % 1000, loop_count2a, i);
-if (time_encode2b > 0){
-	i = (int)((__int64)loop_count2b * unit_size * 125 / ((__int64)time_encode2b * 131072));
+	printf(" 1st encode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2a, time_speed);
+time_sec = (double)time_encode2b / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2b * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
-printf(" 2nd encode %d.%03d sec, %d loop, %d MB/s\n", time_encode2b / 1000, time_encode2b % 1000, loop_count2b, i);
+printf(" 2nd encode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2b, time_speed);
 #endif
 
 	// 終了処理
@@ -185,7 +189,7 @@ static DWORD WINAPI thread_encode3(LPVOID lpParameter)
 	RS_TH *th;
 #ifdef TIMER
 unsigned int loop_count2a = 0, loop_count2b = 0;
-unsigned int time_start2, time_encode2a = 0, time_encode2b = 0;
+clock_t time_start2, time_encode2a = 0, time_encode2b = 0;
 #endif
 
 	th = (RS_TH *)lpParameter;
@@ -204,7 +208,7 @@ unsigned int time_start2, time_encode2a = 0, time_encode2b = 0;
 	WaitForSingleObject(hRun, INFINITE);	// 計算開始の合図を待つ
 	while (th->now < INT_MAX / 2){
 #ifdef TIMER
-time_start2 = GetTickCount();
+time_start2 = clock();
 #endif
 		s_buf = th->buf;
 		src_off = th->off;	// ソース・ブロック番号
@@ -221,7 +225,7 @@ loop_count2a++;
 #endif
 			}
 #ifdef TIMER
-time_encode2a += GetTickCount() - time_start2;
+time_encode2a += clock() - time_start2;
 #endif
 		} else {	// 全てのパリティ・ブロックを保持する場合
 			// スレッドごとに作成するパリティ・ブロックの chunk を変える
@@ -261,7 +265,7 @@ loop_count2b += src_num;
 #endif
 			}
 #ifdef TIMER
-time_encode2b += GetTickCount() - time_start2;
+time_encode2b += clock() - time_start2;
 #endif
 		}
 		//_mm_sfence();	// メモリーへの書き込みを完了する
@@ -271,19 +275,21 @@ time_encode2b += GetTickCount() - time_start2;
 #ifdef TIMER
 loop_count2b /= chunk_num;	// chunk数で割ってブロック数にする
 printf("sub-thread : total loop = %d\n", loop_count2a + loop_count2b);
-if (time_encode2a > 0){
-	i = (int)((__int64)loop_count2a * unit_size * 125 / ((__int64)time_encode2a * 131072));
+time_sec = (double)time_encode2a / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2a * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
 if (loop_count2a > 0)
-	printf(" 1st encode %d.%03d sec, %d loop, %d MB/s\n", time_encode2a / 1000, time_encode2a % 1000, loop_count2a, i);
-if (time_encode2b > 0){
-	i = (int)((__int64)loop_count2b * unit_size * 125 / ((__int64)time_encode2b * 131072));
+	printf(" 1st encode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2a, time_speed);
+time_sec = (double)time_encode2b / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2b * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
-printf(" 2nd encode %d.%03d sec, %d loop, %d MB/s\n", time_encode2b / 1000, time_encode2b % 1000, loop_count2b, i);
+printf(" 2nd encode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2b, time_speed);
 #endif
 
 	// 終了処理
@@ -303,7 +309,8 @@ static DWORD WINAPI thread_encode_gpu(LPVOID lpParameter)
 	HANDLE hRun, hEnd;
 	RS_TH *th;
 #ifdef TIMER
-unsigned int time_start2, time_encode2 = 0, loop_count2 = 0;
+unsigned int loop_count2 = 0;
+clock_t time_start2, time_encode2 = 0;
 #endif
 
 	th = (RS_TH *)lpParameter;
@@ -320,7 +327,7 @@ unsigned int time_start2, time_encode2 = 0, loop_count2 = 0;
 	WaitForSingleObject(hRun, INFINITE);	// 計算開始の合図を待つ
 	while (th->now < INT_MAX / 2){
 #ifdef TIMER
-time_start2 = GetTickCount();
+time_start2 = clock();
 #endif
 		// GPUはソース・ブロック読み込み中に呼ばれない
 		s_buf = th->buf;
@@ -335,24 +342,71 @@ time_start2 = GetTickCount();
 		}
 
 		// 一つの GPUスレッドが全てのパリティ・ブロックを処理する
-		while ((j = InterlockedIncrement(&(th->now))) < parity_num){	// j = ++th_now
-			// factor は定数行列の乗数になる
-			for (i = 0; i < src_num; i++)
-				factor[i] = galois_power(constant[src_off + i], first_num + j);
+		if (OpenCL_method & 8){	// ２ブロックずつ計算する
+			// パリティ・ブロック数が奇数なら、最初の一個だけ別に計算する
+			if (parity_num & 1){
+				InterlockedIncrement(&(th->now));	// 常に j = 0 となる
 
-			// VRAM上のソース・ブロックごとにパリティを追加していく
-			i = gpu_multiply_blocks(src_num, factor, g_buf + (size_t)unit_size * j, unit_size);
-			if (i != 0){
-				th->len = i;
-				InterlockedExchange(&(th->now), INT_MAX / 3);	// サブ・スレッドの計算を中断する
-				break;
-			}
+				// factor は定数行列の乗数になる
+				for (i = 0; i < src_num; i++)
+					factor[i] = galois_power(constant[src_off + i], first_num);
+
+				// VRAM上のソース・ブロックごとにパリティを追加していく
+				i = gpu_multiply_blocks(src_num, factor, NULL, g_buf, unit_size);
+				if (i != 0){
+					th->len = i;
+					InterlockedExchange(&(th->now), INT_MAX / 3);	// サブ・スレッドの計算を中断する
+					break;
+				}
 #ifdef TIMER
 loop_count2 += src_num;
 #endif
-		}
+			}
+
+			// 残りのブロックは二個ずつ計算する
+			while ((j = InterlockedAdd(&(th->now), 2)) < parity_num){	// th_now += 2, j = th_now
+				j--;	// +2 してるから、最初のブロックは -1 する
+
+				// factor は定数行列の乗数になる
+				for (i = 0; i < src_num; i++){
+					int c = constant[src_off + i];	// 同じ定数だけど、何乗するかが異なる
+					factor[i] = galois_power(c, first_num + j);
+					factor[src_num + i] = galois_power(c, first_num + j + 1);
+				}
+
+				// VRAM上のソース・ブロックごとにパリティを追加していく
+				i = gpu_multiply_blocks(src_num, factor, (void *)1, g_buf + (size_t)unit_size * j, unit_size * 2);
+				if (i != 0){
+					th->len = i;
+					InterlockedExchange(&(th->now), INT_MAX / 3);	// サブ・スレッドの計算を中断する
+					break;
+				}
 #ifdef TIMER
-time_encode2 += GetTickCount() - time_start2;
+loop_count2 += src_num * 2;
+#endif
+			}
+
+		} else {	// 以前からの１ブロックずつ計算する方式
+			while ((j = InterlockedIncrement(&(th->now))) < parity_num){	// j = ++th_now
+				// factor は定数行列の乗数になる
+				for (i = 0; i < src_num; i++)
+					factor[i] = galois_power(constant[src_off + i], first_num + j);
+
+				// VRAM上のソース・ブロックごとにパリティを追加していく
+				i = gpu_multiply_blocks(src_num, factor, NULL, g_buf + (size_t)unit_size * j, unit_size);
+				if (i != 0){
+					th->len = i;
+					InterlockedExchange(&(th->now), INT_MAX / 3);	// サブ・スレッドの計算を中断する
+					break;
+				}
+#ifdef TIMER
+loop_count2 += src_num;
+#endif
+			}
+		}
+
+#ifdef TIMER
+time_encode2 += clock() - time_start2;
 #endif
 		// 最後にVRAMを解放する
 		i = gpu_finish();
@@ -365,12 +419,13 @@ time_encode2 += GetTickCount() - time_start2;
 	}
 #ifdef TIMER
 printf("gpu-thread :\n");
-if (time_encode2 > 0){
-	i = (int)((__int64)loop_count2 * unit_size * 125 / ((__int64)time_encode2 * 131072));
+time_sec = (double)time_encode2 / CLOCKS_PER_SEC;
+if (time_sec > 0){
+	time_speed = ((double)loop_count2 * unit_size) / (time_sec * 1048576);
 } else {
-	i = 0;
+	time_speed = 0;
 }
-printf(" 2nd encode %d.%03d sec, %d loop, %d MB/s\n", time_encode2 / 1000, time_encode2 % 1000, loop_count2, i);
+printf(" 2nd encode %.3f sec, %d loop, %.0f MB/s\n", time_sec, loop_count2, time_speed);
 #endif
 
 	// 終了処理
@@ -452,7 +507,7 @@ int encode_method1(	// ソース・ブロックが一個だけの場合
 	block_off = 0;
 	while (block_off < block_size){
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		// ソース・ブロックを読み込む
 		len = s_blk[0].size - block_off;
@@ -469,7 +524,7 @@ time_start = GetTickCount();
 		s_blk[0].crc = crc_update(s_blk[0].crc, buf, len);	// without pad
 		checksum16_altmap(buf, buf + io_size, io_size);
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 		// リカバリ・ファイルに書き込むサイズ
@@ -482,13 +537,13 @@ time_read += GetTickCount() - time_start;
 		// パリティ・ブロックごとに
 		for (i = 0; i < parity_num; i++){
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 			memset(work_buf, 0, unit_size);
 			// factor は 2の乗数になる
 			galois_align_multiply(buf, work_buf, unit_size, galois_power(2, first_num + i));
 #ifdef TIMER
-time_calc += GetTickCount() - time_start;
+time_calc += clock() - time_start;
 #endif
 
 			// 経過表示
@@ -502,7 +557,7 @@ time_calc += GetTickCount() - time_start;
 			}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 			// パリティ・ブロックのチェックサムを検証する
 			checksum16_return(work_buf, hash, io_size);
@@ -535,7 +590,7 @@ time_start = GetTickCount();
 				goto error_end;
 			}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 		}
 
@@ -565,7 +620,7 @@ time_write += GetTickCount() - time_start;
 		}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		// 最後に Recovery Slice packet のヘッダーを書き込む
 		for (i = 0; i < parity_num; i++){
@@ -581,14 +636,14 @@ time_start = GetTickCount();
 			}
 		}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 	}
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
-printf("write  %d.%03d sec\n", time_write / 1000, time_write % 1000);
-printf("encode %d.%03d sec\n", time_calc / 1000, time_calc % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
+printf("write  %.3f sec\n", (double)time_write / CLOCKS_PER_SEC);
+printf("encode %.3f sec\n", (double)time_calc / CLOCKS_PER_SEC);
 #endif
 
 error_end:
@@ -729,7 +784,7 @@ int encode_method2(	// ソース・データを全て読み込む場合
 #ifdef TIMER
 read_count = 0;
 skip_count = 0;
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		last_file = -1;
 		for (i = 0; i < source_num; i++){
@@ -830,7 +885,7 @@ skip_count++;
 		CloseHandle(hFile);
 		hFile = NULL;
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 		WaitForMultipleObjects(cpu_num1, hEnd, TRUE, INFINITE);	// サブ・スレッドの計算終了の合図を待つ
@@ -930,7 +985,7 @@ skip_count++;
 			}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 			// パリティ・ブロックを書き込む
 			work_buf = p_buf;
@@ -979,7 +1034,7 @@ time_start = GetTickCount();
 				}
 			}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 
 			part_off += part_num;	// 次のパリティ位置にする
@@ -1025,7 +1080,7 @@ time_write += GetTickCount() - time_start;
 		}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		// 最後に Recovery Slice packet のヘッダーを書き込む
 		for (i = 0; i < parity_num; i++){
@@ -1041,13 +1096,13 @@ time_start = GetTickCount();
 			}
 		}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 	}
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
-printf("write  %d.%03d sec\n", time_write / 1000, time_write % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
+printf("write  %.3f sec\n", (double)time_write / CLOCKS_PER_SEC);
 if (prog_num != prog_base)
 	printf(" prog_num = %I64d, prog_base = %I64d\n", prog_num, prog_base);
 #endif
@@ -1186,7 +1241,7 @@ int encode_method3(	// パリティ・ブロックを全て保持して、一度
 		src_off = source_off - 1;	// まだ計算して無い印
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		for (i = 0; i < read_num; i++){	// スライスを一個ずつ読み込んでメモリー上に配置していく
 			// ソース・ブロックを読み込む
@@ -1318,7 +1373,7 @@ time_start = GetTickCount();
 			memcpy(common_buf + packet_off + 16, file_md_ctx.hash, 16);
 		}
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 		WaitForMultipleObjects(cpu_num1, hEnd, TRUE, INFINITE);	// サブ・スレッドの計算終了の合図を待つ
@@ -1393,19 +1448,19 @@ time_read += GetTickCount() - time_start;
 	}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	memcpy(common_buf + common_size, common_buf, common_size);	// 後の半分に前半のをコピーする
 	// 最後にパリティ・ブロックのチェックサムを検証して、リカバリ・ファイルに書き込む
 	err = create_recovery_file_1pass(file_path, recovery_path, packet_limit, block_distri,
 			packet_num, common_buf, common_size, footer_buf, footer_size, rcv_hFile, p_buf, NULL, unit_size);
 #ifdef TIMER
-time_write = GetTickCount() - time_start;
+time_write = clock() - time_start;
 #endif
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
-printf("write  %d.%03d sec\n", time_write / 1000, time_write % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
+printf("write  %.3f sec\n", (double)time_write / CLOCKS_PER_SEC);
 if (prog_num != prog_base - prog_write * parity_num)
 	printf(" prog_num = %I64d != %I64d\n", prog_num, prog_base - prog_write * parity_num);
 #endif
@@ -1577,7 +1632,7 @@ int encode_method4(	// 全てのブロックを断片的に保持する場合 (G
 #ifdef TIMER
 read_count = 0;
 skip_count = 0;
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		last_file = -1;
 		for (i = 0; i < source_num; i++){
@@ -1678,7 +1733,7 @@ skip_count++;
 		CloseHandle(hFile);
 		hFile = NULL;
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 		memset(g_buf, 0, (size_t)unit_size * parity_num);	// 待機中に GPU用の領域をゼロ埋めしておく
@@ -1817,11 +1872,21 @@ skip_count++;
 #endif
 				} else if (src_off + src_num + src_max > source_num){
 					src_num = source_num - src_off - src_max;
-					// src_num が 0にならないように、src_num == src_max なら上の last1 にする
-					if ((src_num < src_max) && (src_num + src_max <= vram_max) && (gpu_end * 2 > cpu_end)){
-						src_num += src_max;	// GPU担当量が少なくて、余裕がある場合は、残りも全て任せる
+					if (src_num < src_max){
+						if ((src_num + src_max <= vram_max) && (gpu_end * 2 > cpu_end)){
+							src_num += src_max;	// GPU担当量が少なくて、余裕がある場合は、残りも全て任せる
+#ifdef TIMER
+							printf("GPU last +: src_off = %d, src_num = %d + %d\n", src_off, src_num - src_max, src_max);
+#endif
+						} else if (src_num < src_max / 4){
+							src_num = src_max / 4;	// src_num が小さくなり過ぎないようにする
+#ifdef TIMER
+							printf("GPU last ?: src_off = %d, src_num = %d\n", src_off, src_num);
+						} else {
+							printf("GPU last -: src_off = %d, src_num = %d\n", src_off, src_num);
+#endif
+						}
 #ifdef TIMER
-						printf("GPU last +: src_off = %d, src_num = %d + %d\n", src_off, src_num - src_max, src_max);
 					} else {
 						printf("GPU last 2: src_off = %d, src_num = %d\n", src_off, src_num);
 #endif
@@ -1921,7 +1986,7 @@ skip_count++;
 			prog_num += th->size * parity_num;
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		// パリティ・ブロックを書き込む
 		work_buf = p_buf;
@@ -1972,7 +2037,7 @@ time_start = GetTickCount();
 			}
 		}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 
 		block_off += io_size;
@@ -2015,7 +2080,7 @@ time_write += GetTickCount() - time_start;
 		}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		// 最後に Recovery Slice packet のヘッダーを書き込む
 		for (i = 0; i < parity_num; i++){
@@ -2031,13 +2096,13 @@ time_start = GetTickCount();
 			}
 		}
 #ifdef TIMER
-time_write += GetTickCount() - time_start;
+time_write += clock() - time_start;
 #endif
 	}
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
-printf("write  %d.%03d sec\n", time_write / 1000, time_write % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
+printf("write  %.3f sec\n", (double)time_write / CLOCKS_PER_SEC);
 if (prog_num != prog_base)
 	printf(" prog_num = %I64d, prog_base = %I64d\n", prog_num, prog_base);
 #endif
@@ -2210,7 +2275,7 @@ int encode_method5(	// ソース・ブロックの一部とパリティ・ブロ
 		src_off = source_off - 1;	// まだ計算して無い印
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 		for (i = 0; i < read_num; i++){	// スライスを一個ずつ読み込んでメモリー上に配置していく
 			// ソース・ブロックを読み込む
@@ -2341,7 +2406,7 @@ time_start = GetTickCount();
 			memcpy(common_buf + packet_off + 16, file_md_ctx.hash, 16);
 		}
 #ifdef TIMER
-time_read += GetTickCount() - time_start;
+time_read += clock() - time_start;
 #endif
 
 		if (source_off == 0)
@@ -2462,10 +2527,21 @@ time_read += GetTickCount() - time_start;
 #endif
 				} else if (src_off + src_num + src_max > read_num){
 					src_num = read_num - src_off - src_max;
-					if ((src_num < src_max) && (src_num + src_max <= vram_max) && (gpu_end * 2 > cpu_end)){
-						src_num += src_max;	// GPU担当量が少なくて、余裕がある場合は、残りも全て任せる
+					if (src_num < src_max){
+						if ((src_num + src_max <= vram_max) && (gpu_end * 2 > cpu_end)){
+							src_num += src_max;	// GPU担当量が少なくて、余裕がある場合は、残りも全て任せる
+#ifdef TIMER
+							printf("GPU last +: src_off = %d, src_num = %d + %d\n", src_off, src_num - src_max, src_max);
+#endif
+						} else if (src_num < src_max / 4){
+							src_num = src_max / 4;	// src_num が小さくなり過ぎないようにする
+#ifdef TIMER
+							printf("GPU last ?: src_off = %d, src_num = %d\n", src_off, src_num);
+						} else {
+							printf("GPU last -: src_off = %d, src_num = %d\n", src_off, src_num);
+#endif
+						}
 #ifdef TIMER
-						printf("GPU last +: src_off = %d, src_num = %d + %d\n", src_off, src_num - src_max, src_max);
 					} else {
 						printf("GPU last 2: src_off = %d, src_num = %d\n", src_off, src_num);
 #endif
@@ -2568,19 +2644,19 @@ time_read += GetTickCount() - time_start;
 	}
 
 #ifdef TIMER
-time_start = GetTickCount();
+time_start = clock();
 #endif
 	memcpy(common_buf + common_size, common_buf, common_size);	// 後の半分に前半のをコピーする
 	// 最後にパリティ・ブロックのチェックサムを検証して、リカバリ・ファイルに書き込む
 	err = create_recovery_file_1pass(file_path, recovery_path, packet_limit, block_distri,
 			packet_num, common_buf, common_size, footer_buf, footer_size, rcv_hFile, p_buf, g_buf, unit_size);
 #ifdef TIMER
-time_write = GetTickCount() - time_start;
+time_write = clock() - time_start;
 #endif
 
 #ifdef TIMER
-printf("read   %d.%03d sec\n", time_read / 1000, time_read % 1000);
-printf("write  %d.%03d sec\n", time_write / 1000, time_write % 1000);
+printf("read   %.3f sec\n", (double)time_read / CLOCKS_PER_SEC);
+printf("write  %.3f sec\n", (double)time_write / CLOCKS_PER_SEC);
 if (prog_num != prog_base - prog_write * parity_num)
 	printf(" prog_num = %I64d != %I64d\n", prog_num, prog_base - prog_write * parity_num);
 #endif

source/par2j/search.c

@@ -1,5 +1,5 @@
 // search.c
-// Copyright : 2022-10-14 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <stdio.h>

source/par2j/source.cl

@@ -1,10 +1,11 @@
 void calc_table(__local uint *mtab, int id, int factor)
 {
-	int i, sum = 0;
+	int i, sum;
 
-	for (i = 0; i < 8; i++){
-		sum = (id & (1 << i)) ? (sum ^ factor) : sum;
-		factor = (factor & 0x8000) ? ((factor << 1) ^ 0x1100B) : (factor << 1);
+	sum = ((id << 31) >> 31) & factor;
+	for (i = 1; i < 8; i++){
+		factor = (factor << 1) ^ (((factor << 16) >> 31) & 0x1100B);
+		sum ^= ((id << (31 - i)) >> 31) & factor;
 	}
 	mtab[id] = sum;
 
@@ -14,6 +15,30 @@ void calc_table(__local uint *mtab, int id, int factor)
 	mtab[id + 256] = sum;
 }
 
+void calc_table2(__local uint *mtab, int id, int factor, int factor2)
+{
+	int i, sum, sum2, mask;
+
+	mask = (id << 31) >> 31;
+	sum = mask & factor;
+	sum2 = mask & factor2;
+	for (i = 1; i < 8; i++){
+		factor = (factor << 1) ^ (((factor << 16) >> 31) & 0x1100B);
+		factor2 = (factor2 << 1) ^ (((factor2 << 16) >> 31) & 0x1100B);
+		mask = (id << (31 - i)) >> 31;
+		sum ^= mask & factor;
+		sum2 ^= mask & factor2;
+	}
+	mtab[id] = sum | (sum2 << 16);
+
+	sum = (sum << 4) ^ (((sum << 16) >> 31) & 0x88058) ^ (((sum << 17) >> 31) & 0x4402C) ^ (((sum << 18) >> 31) & 0x22016) ^ (((sum << 19) >> 31) & 0x1100B);
+	sum = (sum << 4) ^ (((sum << 16) >> 31) & 0x88058) ^ (((sum << 17) >> 31) & 0x4402C) ^ (((sum << 18) >> 31) & 0x22016) ^ (((sum << 19) >> 31) & 0x1100B);
+	sum2 = (sum2 << 4) ^ (((sum2 << 16) >> 31) & 0x88058) ^ (((sum2 << 17) >> 31) & 0x4402C) ^ (((sum2 << 18) >> 31) & 0x22016) ^ (((sum2 << 19) >> 31) & 0x1100B);
+	sum2 = (sum2 << 4) ^ (((sum2 << 16) >> 31) & 0x88058) ^ (((sum2 << 17) >> 31) & 0x4402C) ^ (((sum2 << 18) >> 31) & 0x22016) ^ (((sum2 << 19) >> 31) & 0x1100B);
+
+	mtab[id + 256] = sum | (sum2 << 16);
+}
+
 __kernel void method1(
 	__global uint *src,
 	__global uint *dst,
@@ -31,6 +56,7 @@ __kernel void method1(
 		dst[i] = 0;
 
 	for (blk = 0; blk < blk_num; blk++){
+		barrier(CLK_LOCAL_MEM_FENCE);
 		calc_table(mtab, table_id, factors[blk]);
 		barrier(CLK_LOCAL_MEM_FENCE);
 
@@ -42,7 +68,6 @@ __kernel void method1(
 			dst[i] ^= sum;
 		}
 		src += BLK_SIZE;
-		barrier(CLK_LOCAL_MEM_FENCE);
 	}
 }
 
@@ -65,6 +90,7 @@ __kernel void method2(
 	}
 
 	for (blk = 0; blk < blk_num; blk++){
+		barrier(CLK_LOCAL_MEM_FENCE);
 		calc_table(mtab, table_id, factors[blk]);
 		barrier(CLK_LOCAL_MEM_FENCE);
 
@@ -82,11 +108,182 @@ __kernel void method2(
 			dst[pos + 4] ^= ((sum1 & 0xFF00FF00) >> 8) | (sum2 & 0xFF00FF00);
 		}
 		src += BLK_SIZE;
-		barrier(CLK_LOCAL_MEM_FENCE);
 	}
 }
 
 __kernel void method4(
+	__global uint4 *src,
+	__global uint4 *dst,
+	__global ushort *factors,
+	int blk_num)
+{
+	__local uint mtab[512];
+	int i, blk;
+	uchar4 r0, r1, r2, r3, r4, r5, r6, r7;
+	uchar16 lo, hi;
+	const int work_id = get_global_id(0) * 2;
+	const int work_size = get_global_size(0) * 2;
+	const int table_id = get_local_id(0);
+
+	for (i = work_id; i < BLK_SIZE / 4; i += work_size){
+		dst[i    ] = 0;
+		dst[i + 1] = 0;
+	}
+
+	for (blk = 0; blk < blk_num; blk++){
+		barrier(CLK_LOCAL_MEM_FENCE);
+		calc_table(mtab, table_id, factors[blk]);
+		barrier(CLK_LOCAL_MEM_FENCE);
+
+		for (i = work_id; i < BLK_SIZE / 4; i += work_size){
+			lo = as_uchar16(src[i    ]);
+			hi = as_uchar16(src[i + 1]);
+			r0 = (uchar4)(as_uchar2((ushort)(mtab[lo.s0] ^ mtab[256 + hi.s0])), as_uchar2((ushort)(mtab[lo.s1] ^ mtab[256 + hi.s1])));
+			r1 = (uchar4)(as_uchar2((ushort)(mtab[lo.s2] ^ mtab[256 + hi.s2])), as_uchar2((ushort)(mtab[lo.s3] ^ mtab[256 + hi.s3])));
+			r2 = (uchar4)(as_uchar2((ushort)(mtab[lo.s4] ^ mtab[256 + hi.s4])), as_uchar2((ushort)(mtab[lo.s5] ^ mtab[256 + hi.s5])));
+			r3 = (uchar4)(as_uchar2((ushort)(mtab[lo.s6] ^ mtab[256 + hi.s6])), as_uchar2((ushort)(mtab[lo.s7] ^ mtab[256 + hi.s7])));
+			r4 = (uchar4)(as_uchar2((ushort)(mtab[lo.s8] ^ mtab[256 + hi.s8])), as_uchar2((ushort)(mtab[lo.s9] ^ mtab[256 + hi.s9])));
+			r5 = (uchar4)(as_uchar2((ushort)(mtab[lo.sa] ^ mtab[256 + hi.sa])), as_uchar2((ushort)(mtab[lo.sb] ^ mtab[256 + hi.sb])));
+			r6 = (uchar4)(as_uchar2((ushort)(mtab[lo.sc] ^ mtab[256 + hi.sc])), as_uchar2((ushort)(mtab[lo.sd] ^ mtab[256 + hi.sd])));
+			r7 = (uchar4)(as_uchar2((ushort)(mtab[lo.se] ^ mtab[256 + hi.se])), as_uchar2((ushort)(mtab[lo.sf] ^ mtab[256 + hi.sf])));
+			dst[i    ] ^= as_uint4((uchar16)(r0.x, r0.z, r1.x, r1.z, r2.x, r2.z, r3.x, r3.z, r4.x, r4.z, r5.x, r5.z, r6.x, r6.z, r7.x, r7.z));
+			dst[i + 1] ^= as_uint4((uchar16)(r0.y, r0.w, r1.y, r1.w, r2.y, r2.w, r3.y, r3.w, r4.y, r4.w, r5.y, r5.w, r6.y, r6.w, r7.y, r7.w));
+		}
+		src += BLK_SIZE / 4;
+	}
+}
+
+__kernel void method9(
+	__global uint *src,
+	__global uint *dst,
+	__global ushort *factors,
+	int blk_num)
+{
+	__local uint mtab[512];
+	int i, blk;
+	uint v, sum, sum2;
+	const int work_id = get_global_id(0);
+	const int work_size = get_global_size(0);
+	const int table_id = get_local_id(0);
+
+	for (i = work_id; i < BLK_SIZE; i += work_size){
+		dst[i] = 0;
+		dst[i + BLK_SIZE] = 0;
+	}
+
+	for (blk = 0; blk < blk_num; blk++){
+		barrier(CLK_LOCAL_MEM_FENCE);
+		calc_table2(mtab, table_id, factors[blk], factors[blk_num + blk]);
+		barrier(CLK_LOCAL_MEM_FENCE);
+
+		for (i = work_id; i < BLK_SIZE; i += work_size){
+			v = src[i];
+			sum  = mtab[(uchar)v] ^ mtab[256 + (uchar)(v >> 8)];
+			sum2 = mtab[(uchar)(v >> 16)] ^ mtab[256 + (v >> 24)];
+			dst[i] ^= (sum & 0xFFFF) | (sum2 << 16);
+			dst[i + BLK_SIZE] ^= (sum >> 16) | (sum2 & 0xFFFF0000);
+		}
+		src += BLK_SIZE;
+	}
+}
+
+__kernel void method10(
+	__global uint *src,
+	__global uint *dst,
+	__global ushort *factors,
+	int blk_num)
+{
+	__local uint mtab[512];
+	int i, blk, pos;
+	uint lo, hi, t0, t1, t2, t3;
+	const int work_id = get_global_id(0) * 2;
+	const int work_size = get_global_size(0) * 2;
+	const int table_id = get_local_id(0);
+
+	for (i = work_id; i < BLK_SIZE; i += work_size){
+		dst[i    ] = 0;
+		dst[i + 1] = 0;
+		dst[i + BLK_SIZE    ] = 0;
+		dst[i + BLK_SIZE + 1] = 0;
+	}
+
+	for (blk = 0; blk < blk_num; blk++){
+		barrier(CLK_LOCAL_MEM_FENCE);
+		calc_table2(mtab, table_id, factors[blk], factors[blk_num + blk]);
+		barrier(CLK_LOCAL_MEM_FENCE);
+
+		for (i = work_id; i < BLK_SIZE; i += work_size){
+			pos = (i & ~7) + ((i & 7) >> 1);
+			lo = src[pos    ];
+			hi = src[pos + 4];
+			t0 = mtab[(uchar)lo] ^ mtab[256 + (uchar)hi];
+			t1 = mtab[(uchar)(lo >> 8)] ^ mtab[256 + (uchar)(hi >> 8)];
+			t2 = mtab[(uchar)(lo >> 16)] ^ mtab[256 + (uchar)(hi >> 16)];
+			t3 = mtab[lo >> 24] ^ mtab[256 + (hi >> 24)];
+			dst[pos    ] ^= (uchar)t0 | ((t1 << 8) & 0xFF00) | ((t2 << 16) & 0xFF0000) | (t3 << 24);
+			dst[pos + 4] ^= (uchar)(t0 >> 8) | (t1 & 0xFF00) | ((t2 << 8) & 0xFF0000) | ((t3 << 16) & 0xFF000000);
+			dst[pos + BLK_SIZE    ] ^= (uchar)(t0 >> 16) | ((t1 >> 8) & 0xFF00) | (t2 & 0xFF0000) | ((t3 << 8) & 0xFF000000);
+			dst[pos + BLK_SIZE + 4] ^= (t0 >> 24) | ((t1 >> 16) & 0xFF00) | ((t2 >> 8) & 0xFF0000) | (t3 & 0xFF000000);
+		}
+		src += BLK_SIZE;
+	}
+}
+
+__kernel void method12(
+	__global uint4 *src,
+	__global uint4 *dst,
+	__global ushort *factors,
+	int blk_num)
+{
+	__local uint mtab[512];
+	int i, blk;
+	uchar4 r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, rA, rB, rC, rD, rE, rF;
+	uchar16 lo, hi;
+	const int work_id = get_global_id(0) * 2;
+	const int work_size = get_global_size(0) * 2;
+	const int table_id = get_local_id(0);
+
+	for (i = work_id; i < BLK_SIZE / 4; i += work_size){
+		dst[i    ] = 0;
+		dst[i + 1] = 0;
+		dst[i + BLK_SIZE / 4    ] = 0;
+		dst[i + BLK_SIZE / 4 + 1] = 0;
+	}
+
+	for (blk = 0; blk < blk_num; blk++){
+		barrier(CLK_LOCAL_MEM_FENCE);
+		calc_table2(mtab, table_id, factors[blk], factors[blk_num + blk]);
+		barrier(CLK_LOCAL_MEM_FENCE);
+
+		for (i = work_id; i < BLK_SIZE / 4; i += work_size){
+			lo = as_uchar16(src[i    ]);
+			hi = as_uchar16(src[i + 1]);
+			r0 = as_uchar4(mtab[lo.s0] ^ mtab[256 + hi.s0]);
+			r1 = as_uchar4(mtab[lo.s1] ^ mtab[256 + hi.s1]);
+			r2 = as_uchar4(mtab[lo.s2] ^ mtab[256 + hi.s2]);
+			r3 = as_uchar4(mtab[lo.s3] ^ mtab[256 + hi.s3]);
+			r4 = as_uchar4(mtab[lo.s4] ^ mtab[256 + hi.s4]);
+			r5 = as_uchar4(mtab[lo.s5] ^ mtab[256 + hi.s5]);
+			r6 = as_uchar4(mtab[lo.s6] ^ mtab[256 + hi.s6]);
+			r7 = as_uchar4(mtab[lo.s7] ^ mtab[256 + hi.s7]);
+			r8 = as_uchar4(mtab[lo.s8] ^ mtab[256 + hi.s8]);
+			r9 = as_uchar4(mtab[lo.s9] ^ mtab[256 + hi.s9]);
+			rA = as_uchar4(mtab[lo.sa] ^ mtab[256 + hi.sa]);
+			rB = as_uchar4(mtab[lo.sb] ^ mtab[256 + hi.sb]);
+			rC = as_uchar4(mtab[lo.sc] ^ mtab[256 + hi.sc]);
+			rD = as_uchar4(mtab[lo.sd] ^ mtab[256 + hi.sd]);
+			rE = as_uchar4(mtab[lo.se] ^ mtab[256 + hi.se]);
+			rF = as_uchar4(mtab[lo.sf] ^ mtab[256 + hi.sf]);
+			dst[i    ] ^= as_uint4((uchar16)(r0.x, r1.x, r2.x, r3.x, r4.x, r5.x, r6.x, r7.x, r8.x, r9.x, rA.x, rB.x, rC.x, rD.x, rE.x, rF.x));
+			dst[i + 1] ^= as_uint4((uchar16)(r0.y, r1.y, r2.y, r3.y, r4.y, r5.y, r6.y, r7.y, r8.y, r9.y, rA.y, rB.y, rC.y, rD.y, rE.y, rF.y));
+			dst[i + BLK_SIZE / 4    ] ^= as_uint4((uchar16)(r0.z, r1.z, r2.z, r3.z, r4.z, r5.z, r6.z, r7.z, r8.z, r9.z, rA.z, rB.z, rC.z, rD.z, rE.z, rF.z));
+			dst[i + BLK_SIZE / 4 + 1] ^= as_uint4((uchar16)(r0.w, r1.w, r2.w, r3.w, r4.w, r5.w, r6.w, r7.w, r8.w, r9.w, rA.w, rB.w, rC.w, rD.w, rE.w, rF.w));
+		}
+		src += BLK_SIZE / 4;
+	}
+}
+
+__kernel void method16(
 	__global uint *src,
 	__global uint *dst,
 	__global ushort *factors,
@@ -94,7 +291,7 @@ __kernel void method4(
 {
 	__local int table[16];
 	__local uint cache[256];
-	int i, j, blk, pos, sht, mask;
+	int i, j, blk, pos, mask, tmp;
 	uint sum;
 	const int work_id = get_global_id(0);
 	const int work_size = get_global_size(0);
@@ -104,11 +301,12 @@ __kernel void method4(
 
 	for (blk = 0; blk < blk_num; blk++){
 		if (get_local_id(0) == 0){
-			pos = factors[blk] << 16;
-			table[0] = pos;
+			tmp = factors[blk];
+			table[0] = tmp;
 			for (j = 1; j < 16; j++){
-				pos = (pos << 1) ^ ((pos >> 31) & 0x100B0000);
-				table[j] = pos;
+				mask = (tmp & 0x8000) ? 0x1100B : 0;
+				tmp = (tmp << 1) ^ mask;
+				table[j] = tmp;
 			}
 		}
 		barrier(CLK_LOCAL_MEM_FENCE);
@@ -119,10 +317,11 @@ __kernel void method4(
 			barrier(CLK_LOCAL_MEM_FENCE);
 
 			sum = 0;
-			sht = (i & 60) >> 2;
+			tmp = (i & 60) >> 2;
+			tmp = 0x8000 >> tmp;
 			pos &= ~60;
 			for (j = 15; j >= 0; j--){
-				mask = (table[j] << sht) >> 31;
+				mask = (table[j] & tmp) ? 0xFFFFFFFF : 0;
 				sum ^= mask & cache[pos];
 				pos += 4;
 			}

source/par2j/verify.c

@@ -1,5 +1,5 @@
 // verify.c
-// Copyright : 2022-10-14 Yutaka Sawada
+// Copyright : 2024-11-30 Yutaka Sawada
 // License : GPL
 
 #ifndef _UNICODE
@@ -9,7 +9,7 @@
 #define UNICODE
 #endif
 #ifndef _WIN32_WINNT
-#define _WIN32_WINNT 0x0600	// Windows Vista or later
+#define _WIN32_WINNT 0x0601	// Windows 7 or later
 #endif
 
 #include <process.h>
@@ -1253,21 +1253,22 @@ static int search_block_slide(
 	slice_ctx *sc)
 {
 	unsigned char *buf, hash[16], hash2[16], err_mag, *short_use;
-	int i, j, find_num, find_flag, find_next, find_last, short_next;
+	int i, j, find_num, find_flag, find_next, find_last, short_next, short2_next, tmp_next;
 	int block_count, short_count, tiny_count, tiny_skip, num, i1, i2, i3, i4;
 	int *order, *index, index_shift;
 	unsigned int len, off, end_off, err_off;
 	unsigned int prev_crc, fail_count, rear_off, overlap_count;
 	unsigned int crc, *crcs, *short_crcs;
 	unsigned int time_last, time_slide;
-	__int64 file_off, file_next, short_off, fail_off;
+	__int64 file_off, file_next, short_off, short2_off, tmp_off, fail_off;
 
 	if (file_size + 1 < last_off + (__int64)(sc->min_size))
 		return 0;	// 小さすぎるファイルは調べない
 	find_num = 0;	// このファイル内で何ブロック見つけたか
 	find_next = -1;	// 次に見つかると予想したブロックの番号
 	find_last = -1;	// 最後に見つけたブロックの番号 (-1=不明)
-	short_next = -1;
+	short_next = -1;	// 予想される末尾ブロックの番号
+	short2_next = -1;
 	fail_count = 0;	// CRC は一致したけど MD5 が違った回数
 	fail_off = 0;
 	rear_off = 0;
@@ -1278,24 +1279,32 @@ static int search_block_slide(
 			find_last = find_next - 1;	// 最後に見つけたブロックの番号
 		if ((last_off >= files[num1].size) || (last_off + block_size > file_size + 1))
 			find_next = -1;	// 予想位置がファイル・サイズを超えると駄目
-		if ((last_size < block_size) && (files[num1].b_num >= 2) &&	// 末尾の半端なブロックの番号と想定位置
-				(last_off < files[num1].size) && (files[num1].size <= file_size + 1)){
-			short_next = files[num1].b_off + files[num1].b_num - 1;	// 末尾ブロックの番号
-			if (find_next == short_next)
+		if ((last_size < block_size) && (last_off < files[num1].size)){	// 末尾の半端なブロックの番号と想定位置
+			tmp_next = files[num1].b_off + files[num1].b_num - 1;	// 末尾ブロックの番号
+			if (find_next == tmp_next)
 				find_next = -1;	// 予想が重複したら末尾ブロックとして探す
-			short_off = files[num1].size - last_size;
-		// ファイルサイズが1ブロック未満でも、同じサイズならエラー訂正を試みる
-		} else if ((last_off == 0) && (file_size == files[num1].size) && (file_size < (__int64)block_size)){
-			short_off = 0;
-			short_next = files[num1].b_off;
+			if ((files[num1].b_num >= 2) && (files[num1].size <= file_size + 1)){	// 本来の位置を調べる
+				short_next = tmp_next;
+				short_off = files[num1].size - last_size;
+			} else if ((last_off == 0) && (file_size == last_size)){	// ファイルが1ブロック未満でも、同じサイズならエラー訂正を試みる
+				short_next = tmp_next;
+				short_off = 0;
+			}
+			if (last_size < file_size){	// 末尾を調べる
+				short2_next = tmp_next;
+				short2_off = file_size - last_size;
+			}
 		}
 		if (file_size > files[num1].size){
 			rear_off = (unsigned int)((file_size - files[num1].size) % (__int64)block_size);
 		} else if (file_size < files[num1].size){
 			rear_off = block_size - (unsigned int)((files[num1].size - file_size) % (__int64)block_size);
 		}
-		//printf("file = %d, find_next = %d, find_last = %d\n", num1, find_next, find_last);
-		//printf("short_off = %I64d, short_next = %d, rear_off = %d\n", short_off, short_next, rear_off);
+/*		printf("file = %d, find_next = %d, find_last = %d, rear_off = %d\n", num1, find_next, find_last, rear_off);
+		if (short_next >= 0)
+			printf("short_off  = %I64d, short_next  = %d\n", short_off, short_next);
+		if (short2_next >= 0)
+			printf("short2_off = %I64d, short2_next = %d\n", short2_off, short2_next);*/
 	}
 	file_off = last_off;	// 検査開始位置から調べる
 	buf = sc->buf;
@@ -1374,8 +1383,11 @@ static int search_block_slide(
 						if (last_off < file_off + last_size)
 							last_off = file_off + last_size;	// 一番大きな半端なブロックの終端
 						find_next = -2;	// 小さなファイルが見つかった = ブロック検出の予想が外れた
-						if (i == short_next)
-							short_next = -1;	// 末尾ブロックは検出済み
+						if (i == short_next){	// この末尾ブロックは検出済み
+							short_next = -1;
+						} else if (i == short2_next){
+							short2_next = -1;
+						}
 
 						// 経過表示
 						if (GetTickCount() - time_last >= UPDATE_TIME){
@@ -1416,7 +1428,7 @@ static int search_block_slide(
 
 	// ブロック・サイズごとに探す
 	if (((block_count > 0) && ((file_off + (__int64)block_size <= file_size)
-			|| (find_next >= 0))) || (short_next >= 0)){	// ブロックの位置を予想して探す
+			|| (find_next >= 0))) || (short_next >= 0) || (short2_next >= 0)){	// ブロックの位置を予想して探す
 		// 前からスライドさせながらチェックサムを比較する
 		//printf("slide search from %I64d, file %d, next = %d\n", file_off, num1, find_next);
 		off = 0;	// buf 内でのオフセット
@@ -1445,8 +1457,13 @@ static int search_block_slide(
 			while (off < end_off){
 				find_flag = -2;
 				// 次の番号のブロックがその位置にあるかを先に調べる (発見済みでも)
-				if ((short_next >= 0) && (file_off + off == short_off)){	// 半端なブロックなら
-					i = short_next;
+				if (((short_next >= 0) && (file_off + off == short_off)) ||
+						((short2_next >= 0) && (file_off + off == short2_off))){	// 半端なブロックなら
+					if ((short_next >= 0) && (file_off + off == short_off)){
+						i = short_next;
+					} else {
+						i = short2_next;
+					}
 					num = s_blk[i].file;
 					if ((short_use[num] & 4) == 0){	// パディング部分を取り除いた CRC-32 を逆算する
 						short_crcs[num] = crc_reverse_zero(s_blk[i].crc, block_size - s_blk[i].size);
@@ -1456,7 +1473,8 @@ static int search_block_slide(
 					find_flag = correct_error(buf + off, s_blk[i].size, s_blk[i].hash, short_crcs[num], &err_off, &err_mag);
 					if (find_flag == 0)
 						find_flag = 2;
-				} else if ((find_next >= 0) && (file_off + off == last_off)){	// フルサイズのブロックなら
+				}
+				if ((find_flag < 0) && (find_next >= 0) && (file_off + off == last_off)){	// フルサイズのブロックなら
 					i = find_next;
 					if (crc == s_blk[i].crc){
 						data_md5(buf + off, block_size, hash);
@@ -1661,20 +1679,81 @@ static int search_block_slide(
 					find_next = i + 1;
 					if ((find_next >= source_num) || (s_blk[find_next].file != num)){
 						// 最後までいった、またはファイルが異なる
-						short_next = -1;
 						find_next = -1;
+						if ((short_next >= 0) && ((s_blk[short_next].exist & 0x1000) != 0))
+							short_next = -1;
+						if ((short2_next >= 0) && ((s_blk[short2_next].exist & 0x1000) != 0))
+							short2_next = -1;
 					} else if (s_blk[find_next].size < block_size){	// 半端なブロックは別に調べる
-						short_next = find_next;
-						short_off = file_off + off + block_size;
-						//printf("short_off = %I64d, short_next = %d, file = %d\n", short_off, short_next, num);
+						if (file_off + off + block_size + s_blk[find_next].size <= file_size){	// ファイル内に収まってる時だけ
+							tmp_next = find_next;
+							tmp_off = file_off + off + block_size;
+							if (find_flag <= 3){	// 順当な位置で見つけた場合
+								if ((tmp_next == short_next) && (tmp_off == short_off)){
+									// 予測済みのと一致するなら何もしない
+								} else if ((short_next >= 0) && (short2_next < 0)){	// 予測と異なるけど、別のが空いてるなら、そっちに記録する
+									//printf("short2_off = %I64d, short2_next = %d, file = %d\n", tmp_off, tmp_next, num);
+									short2_next = tmp_next;
+									short2_off = tmp_off;
+								} else {
+									if ((short_next >= 0) && (tmp_next == short2_next) && (tmp_off == short2_off)){	// 既に予測済みのと一致するなら入れ替える
+										short2_next = short_next;
+										short2_off = short_off;
+										//printf("exchange short2_off = %I64d, short2_next = %d\n", short2_off, short2_next);
+									}
+									//printf("short_off = %I64d, short_next = %d, file = %d\n", tmp_off, tmp_next, num);
+									short_next = tmp_next;
+									short_off = tmp_off;
+								}
+							} else if ((short_next < 0) &&
+									(((__int64)block_size * (__int64)(tmp_next - files[num].b_off) == tmp_off) ||
+									(tmp_off + s_blk[tmp_next].size == file_size))){
+								// 検出ブロックが順当でなくても、末尾ブロックの開始位置や末端がファイル・サイズに一致すれば
+								//printf("short_off = %I64d, short_next = %d, file = %d\n", tmp_off, tmp_next, num);
+								short_next = tmp_next;
+								short_off = tmp_off;
+							} else {
+								//printf("short2_off = %I64d, short2_next = %d, file = %d\n", tmp_off, tmp_next, num);
+								short2_next = tmp_next;
+								short2_off = tmp_off;
+							}
+						}
 						find_next = -1;
 					} else {
-						short_next = files[num].b_off + files[num].b_num - 1;	// 末尾ブロックの番号
-						if (s_blk[short_next].size < block_size){	// 半端なブロックは別に調べる
-							short_off = file_off + off + (__int64)(short_next - i) * (__int64)block_size;
-							//printf("short_off = %I64d, short_next = %d, file = %d\n", short_off, short_next, num);
-						} else {
-							short_next = -1;
+						tmp_next = files[num].b_off + files[num].b_num - 1;	// 末尾ブロックの番号
+						if (s_blk[tmp_next].size < block_size){	// 半端なブロックは別に調べる
+							tmp_off = file_off + off + (__int64)(tmp_next - i) * (__int64)block_size;
+							if (tmp_off + s_blk[tmp_next].size <= file_size){	// ファイル内に収まってる時だけ
+								if (find_flag <= 3){	// 順当な位置で見つけた場合
+									if ((tmp_next == short_next) && (tmp_off == short_off)){
+										// 予測済みのと一致するなら何もしない
+									} else if ((short_next >= 0) && (short2_next < 0)){	// 予測と異なるけど、別のが空いてるなら、そっちに記録する
+										//printf("far short2_off = %I64d, short2_next = %d, file = %d\n", tmp_off, tmp_next, num);
+										short2_next = tmp_next;
+										short2_off = tmp_off;
+									} else {
+										if ((short_next >= 0) && (tmp_next == short2_next) && (tmp_off == short2_off)){	// 既に予測済みのと一致するなら入れ替える
+											short2_next = short_next;
+											short2_off = short_off;
+											//printf("exchange short2_off = %I64d, short2_next = %d\n", short2_off, short2_next);
+										}
+										//printf("far short_off = %I64d, short_next = %d, file = %d\n", tmp_off, tmp_next, num);
+										short_next = tmp_next;
+										short_off = tmp_off;
+									}
+								} else if ((short_next < 0) &&
+										(((__int64)block_size * (__int64)(tmp_next - files[num].b_off) == tmp_off) ||
+										(tmp_off + s_blk[tmp_next].size == file_size))){
+									// 検出ブロックが順当でなくても、末尾ブロックの開始位置や末端がファイル・サイズに一致すれば
+									//printf("far short_off = %I64d, short_next = %d, file = %d\n", tmp_off, tmp_next, num);
+									short_next = tmp_next;
+									short_off = tmp_off;
+								} else if ((short2_next != tmp_next) || (short2_off != tmp_off)){
+									//printf("far short2_off = %I64d, short2_next = %d, file = %d\n", tmp_off, tmp_next, num);
+									short2_next = tmp_next;
+									short2_off = tmp_off;
+								}
+							}
 						}
 					}
 					tiny_skip = 0;	// 小さなファイルをブロック直後に一回だけ探す