Recherche avancée

Sur d’autres sites (7805)

• A systematic approach to making Web Applications accessible

  22 février 2012, par silvia

  With the latest developments in HTML5 and the still fairly new ARIA (Accessible Rich Interface Applications) attributes introduced by the W3C WAI (Web Accessibility Initiative), browsers have now implemented many features that allow you to make your JavaScript-heavy Web applications accessible.

  Since I began working on making a complex web application accessible just over a year ago, I discovered that there was no step-by-step guide to approaching the changes necessary for creating an accessible Web application. Therefore, many people believe that it is still hard, if not impossible, to make Web applications accessible. In fact, it can be approached systematically, as this article will describe.

  This post is based on a talk that Alice Boxhall and I gave at the recent Linux.conf.au titled “Developing accessible Web apps – how hard can it be ?” (slides, video), which in turn was based on a Google Developer Day talk by Rachel Shearer (slides).

  These talks, and this article, introduce a process that you can follow to make your Web applications accessible : each step will take you closer to having an application that can be accessed using a keyboard alone, and by users of screenreaders and other accessibility technology (AT).

  The recommendations here only roughly conform to the requirements of WCAG (Web Content Accessibility Guidelines), which is the basis of legal accessibility requirements in many jurisdictions. The steps in this article may or may not be sufficient to meet a legal requirement. It is focused on the practical outcome of ensuring users with disabilities can use your Web application.

  Step-by-step Approach

  The steps to follow to make your Web apps accessible are as follows :

  1. Use native HTML tags wherever possible
  2. Make interactive elements keyboard accessible
  3. Provide extra markup for AT (accessibility technology)

  If you are a total newcomer to accessibility, I highly recommend installing a screenreader and just trying to read/navigate some Web pages. On Windows you can install the free NVDA screenreader, on Mac you can activate the pre-installed VoiceOver screenreader, on Linux you can use Orca, and if you just want a browser plugin for Chrome try installing ChromeVox.

  1. Use native HTML tags

  As you implement your Web application with interactive controls, try to use as many native HTML tags as possible.

  HTML5 provides a rich set of elements which can be used to both add functionality and provide semantic context to your page. HTML4 already included many useful interactive controls, like <a>, <button>, <input> and <select>, and semantic landmark elements like <h1>. HTML5 adds richer <input> controls, and a more sophisticated set of semantic markup elements like such as <time>, <progress>, <meter>, <nav>, <header>, <article> and <aside>. (Note : check browser support for browser support of the new tags).

  Using as much of the rich HTML5 markup as possible means that you get all of the accessibility features which have been implemented in the browser for those elements, such as keyboard support, short-cut keys and accessibility metadata, for free. For generic tags you have to implement them completely from scratch.

  What exactly do you miss out on when you use a generic tag such as <div> over a specific semantic one such as <button> ?

  1. Generic tags are not focusable. That means you cannot reach them through using the [tab] on the keyboard.
  2. You cannot activate them with the space bar or enter key or perform any other keyboard interaction that would be regarded as typical with such a control.
  3. Since the role that the control represents is not specified in code but is only exposed through your custom visual styling, screenreaders cannot express to their users what type of control it is, e.g. button or link.
  4. Neither can screenreaders add the control to the list of controls on the page that are of a certain type, e.g. to navigate to all headers of a certain level on the page.
  5. And finally you need to manually style the element in order for it to look distinctive compared to other elements on the page ; using a default control will allow the browser to provide the default style for the platform, which you can still override using CSS if you want.

  Example :

  Compare these two buttons. The first one is implemented using a <div> tag, the second one using a <button> tag. Try using a screenreader to experience the difference.

  Send	<style> .custombutton cursor : pointer ; border : 1px solid #000 ; background-color : #F6F6F6 ; display : inline-block ; padding : 2px 5px ; </style> <div class="custombutton" onclick="alert(’sent !’)"> Send </div>
  Send	<button onclick="alert(’sent !’)"> Send </button>

  2. Make interactive elements keyboard accessible

  Many sophisticated web applications have some interactive controls that just have no appropriate HTML tag equivalent. In this case, you will have had to build an interactive element with JavaScript and <div> and/or <span> tags and lots of custom styling. The good news is, it’s possible to make even these custom controls accessible, and as a side benefit you will also make your application smoother to use for power users.

  The first thing you can do to test usability of your control, or your Web app, is to unplug the mouse and try to use only the [TAB] and [ENTER] keys to interact with your application.

  

  Try the following :

  • Can you reach all interactive elements with [TAB] ?
  • Can you activate interactive elements with [ENTER] (or [SPACE]) ?
  • Are the elements in the right tab order ?
  • After interaction : is the right element in focus ?
  • Is there a keyboard shortcut that activates the element (accesskey) ?

  No ? Let’s fix it.

  2.1. Reaching interactive elements

  If you have an element on your page that cannot be reached with [TAB], put a @tabindex attribute on it.

  Example :

  Here we have a <span> tag that works as a link (don’t do this – it’s just a simple example). The first one cannot be reached using [TAB] but the second one has a tabindex and is thus part of the tab order of the HTML page.

  (Note : since we experiment lots with the tabindex in this article, to avoid confusion, click on some text in this paragraph and then hit the [TAB] key to see where it goes next. The click will set your keyboard focus in the DOM.)

  Click	<style> .customlink text-decoration : underline ; cursor : pointer ; </style> <span class="customlink" onclick="alert(’activated !’)"> Click </span>
  Click	<style> .customlink text-decoration : underline ; cursor : pointer ; </style> <span class="customlink" onclick="alert(’activated !’)" tabindex="0"> Click </span>

  You set @tabindex=0 to add an element into the native tab order of the page, which is the DOM order.

  2.2. Activating interactive elements

  Next, you typically want to be able to use the [ENTER] and [SPACE] keys to activate your custom control. To do so, you will need to implement an onkeydown event handler. Note that the keyCode for [ENTER] is 13 and for [SPACE] is 32.

  Example :

  Let’s add this functionality to the <span> tag from before. Try tabbing to it and hit the [ENTER] or [SPACE] key.

  Click	<span class="customlink" onclick="alert(’activated !’)" tabindex="0"> Click </span>
  &lt;script&gt;<br /> function handlekey(event) {<br /> var target = event.target || event.srcElement;<br /> if (event.keyCode == 13 || event.keyCode == 32) { target.onclick(); }<br /> }<br /> &lt;/script&gt; Click	<span class="customlink" onclick="alert(’activated !’)" tabindex="0" onkeydown="handlekey(event) ;"> Click </span> <script> function handlekey(event) var target = event.target || event.srcElement ; if (event.keyCode == 13 || event.keyCode == 32) target.onclick() ; </script>

  Note that there are some controls that might need support for keys other than [tab] or [enter] to be able to use them from the keyboard alone, for example a custom list box, menu or slider should respond to arrow keys.

  2.3. Elements in the right tab order

  Have you tried tabbing to all the elements on your page that you care about ? If so, check if the order of tab stops seems right. The default order is given by the order in which interactive elements appear in the DOM. For example, if your page’s code has a right column that is coded before the main article, then the links in the right column will receive tab focus first before the links in the main article.

  You could change this by re-ordering your DOM, but oftentimes this is not possible. So, instead give the elements that should be the first ones to receive tab focus a positive @tabindex. The tab access will start at the smallest non-zero @tabindex value. If multiple elements share the same @tabindex value, these controls receive tab focus in DOM order. After that, interactive elements and those with @tabindex=0 will receive tab focus in DOM order.

  Example :

  The one thing that always annoys me the most is if the tab order in forms that I am supposed to fill in is illogical. Here is an example where the first and last name are separated by the address because they are in a table. We could fix it by moving to a <div> based layout, but let’s use @tabindex to demonstrate the change.

  Firstname : Address : Lastname : City :	<table class="customtabs"> <tr> <td>Firstname : <input type="text" id="firstname"> </td> <td>Address : <input type="text" id="address"> </td> </tr> <tr> <td>Lastname : <input type="text" id="lastname"> </td> <td>City : <input type="text" id="city"> </td> </tr> </table>
  Click here to test this form, then [TAB] : Firstname : Address : Lastname : City :	<table class="customtabs"> <tr> <td>Firstname : <input type="text" id="firstname" tabindex="10"> </td> <td>Address : <input type="text" id="address" tabindex="30"> </td> </tr> <tr> <td>Lastname : <input type="text" id="lastname" tabindex="20"> </td> <td>City : <input type="text" id="city" tabindex="40"> </td> </tr> </table>

  Be very careful with using non-zero tabindex values. Since they change the tab order on the page, you may get side effects that you might not have intended, such as having to give other elements on the page a non-zero tabindex value to avoid skipping too many other elements as I would need to do here.

  2.4. Focus on the right element

  Some of the controls that you create may be rather complex and open elements on the page that were previously hidden. This is particularly the case for drop-downs, pop-ups, and menus in general. Oftentimes the hidden element is not defined in the DOM right after the interactive control, such that a [TAB] will not put your keyboard focus on the next element that you are interacting with.

  The solution is to manage your keyboard focus from JavaScript using the .focus() method.

  Example :

  Here is a menu that is declared ahead of the menu button. If you tab onto the button and hit enter, the menu is revealed. But your tab focus is still on the menu button, so your next [TAB] will take you somewhere else. We fix it by setting the focus on the first menu item after opening the menu.

  Menu item1 Menu item2 Menu &lt;script&gt;<br /> function displayMenu(value) {<br /> document.getElementById(&quot;custommenu&quot;).style.display=value;<br /> }<br /> &lt;/script&gt;	<div id="custommenu" style="display:none ;"> <button id="item1" onclick="displayMenu(’none’) ;">Menu item1</button> <button id="item2" onclick="displayMenu(’none’) ;">Menu item2</button> </div> <button onclick="displayMenu(’block’) ;">Menu</button> <script> function displayMenu(value) document.getElementById("custommenu").style.display=value ; </script>
  Menu item1 Menu item2 Menu &lt;script&gt;<br /> function displayMenu2(value) {<br /> document.getElementById(&quot;custommenu2&quot;).style.display=value;<br /> document.getElementById(&quot;item1&quot;).focus();<br /> }<br /> &lt;/script&gt;	<div id="custommenu" style="display:none ;"> <button id="item1" onclick="displayMenu(’none’) ;">Menu item1</button> <button id="item2" onclick="displayMenu(’none’) ;">Menu item2</button> </div> <button onclick="displayMenu(’block’) ;">Menu</button> <script> function displayMenu(value) document.getElementById("custommenu").style.display=value ; document.getElementById("item1").focus() ; </script>

  You will notice that there are still some things you can improve on here. For example, after you close the menu again with one of the menu items, the focus does not move back onto the menu button.

  Also, after opening the menu, you may prefer not to move the focus onto the first menu item but rather just onto the menu <div>. You can do so by giving that div a @tabindex and then calling .focus() on it. If you do not want to make the div part of the normal tabbing order, just give it a @tabindex=-1 value. This will allow your div to receive focus from script, but be exempt from accidental tabbing onto (though usually you just want to use @tabindex=0).

  Bonus : If you want to help keyboard users even more, you can also put outlines on the element that is currently in focus using CSS”s outline property. If you want to avoid the outlines for mouse users, you can dynamically add a class that removes the outline in mouseover events but leaves it for :focus.

  2.5. Provide sensible keyboard shortcuts

  At this stage your application is actually keyboard accessible. Congratulations !

  However, it’s still not very efficient : like power-users, screenreader users love keyboard shortcuts : can you imagine if you were forced to tab through an entire page, or navigate back to a menu tree at the top of the page, to reach each control you were interested in ? And, obviously, anything which makes navigating the app via the keyboard more efficient for screenreader users will benefit all power users as well, like the ubiquitous keyboard shortcuts for cut, copy and paste.

  HTML4 introduced so-called accesskeys for this. In HTML5 @accesskey is now allowed on all elements.

  The @accesskey attribute takes the value of a keyboard key (e.g. @accesskey="x") and is activated through platform- and browser-specific activation keys. For example, on the Mac it’s generally the [Ctrl] key, in IE it’ the [Alt] key, in Firefox on Windows [Shift]-[Alt], and in Opera on Windows [Shift]-[ESC]. You press the activation key and the accesskey together which either activates or focuses the element with the @accesskey attribute.

  Example :

  Send &lt;script&gt;<br /> var button = document.getElementById('accessbutton');<br /> if (button.accessKeyLabel) {<br />  button.innerHTML += ' (' + button.accessKeyLabel + ')';<br /> }<br /> &lt;/script&gt;

  <button id="accessbutton" onclick="alert(’sent !’)" accesskey="e"> Send </button> <script> var button = document.getElementById(’accessbutton’) ; if (button.accessKeyLabel) button.innerHTML += ’ (’ + button.accessKeyLabel + ’)’ ; </script>

  Now, the idea behind this is clever, but the execution is pretty poor. Firstly, the different activation keys between different platforms and browsers make it really hard for people to get used to the accesskeys. Secondly, the key combinations can conflict with browser and screenreader shortcut keys, the first of which will render browser shortcuts unusable and the second will effectively remove the accesskeys.

  In the end it is up to the Web application developer whether to use the accesskey attribute or whether to implement explicit shortcut keys for the application through key event handlers on the window object. In either case, make sure to provide a help list for your shortcut keys.

  Also note that a page with a really good hierarchical heading layout and use of ARIA landmarks can help to eliminate the need for accesskeys to jump around the page, since there are typically default navigations available in screen readers to jump directly to headings, hyperlinks, and ARIA landmarks.

  3. Provide markup for AT

  Having made the application keyboard accessible also has advantages for screenreaders, since they can now reach the controls individually and activate them. So, next we will use a screenreader and close our eyes to find out where we only provide visual cues to understand the necessary interaction.

  Here are some of the issues to consider :

  • Role may need to get identified
  • States may need to be kept track of
  • Properties may need to be made explicit
  • Labels may need to be provided for elements

  This is where the W3C’s ARIA (Accessible Rich Internet Applications) standard comes in. ARIA attributes provide semantic information to screen readers and other AT that is otherwise conveyed only visually.

  Note that using ARIA does not automatically implement the standard widget behavior – you’ll still need to add focus management, keyboard navigation, and change aria attribute values in script.

  3.1. ARIA roles

  After implementing a custom interactive widget, you need to add a @role attribute to indicate what type of controls it is, e.g. that it is playing the role of a standard tag such as a button.

  Example :

  This menu button is implemented as a <div>, but with a role of “button” it is announced as a button by a screenreader.

  Menu	<div tabindex="0" role="button">Menu</div>

  ARIA roles also describe composite controls that do not have a native HTML equivalent.

  Example :

  This menu with menu items is implemented as a set of <div> tags, but with a role of “menu” and “menuitem” items.

  Cut Copy Paste

  <div role="menu"> <div tabindex="0" role="menuitem">Cut</div> <div tabindex="0" role="menuitem">Copy</div> <div tabindex="0" role="menuitem">Paste</div> </div>

  3.2. ARIA states

  Some interactive controls represent different states, e.g. a checkbox can be checked or unchecked, or a menu can be expanded or collapsed.

  Example :

  The following menu has states on the menu items, which are here not just used to give an aural indication through the screenreader, but also a visual one through CSS.

  Left Center Right

  <style> .custombutton[aria-checked=true]:before content : "\2713 " ; </style> <div role="menu"> <div tabindex="0" role="menuitem" aria-checked="true">Left</div> <div tabindex="0" role="menuitem" aria-checked="false">Center</div> <div tabindex="0" role="menuitem" aria-checked="false">Right</div> </div>

  3.3. ARIA properties

  Some of the functionality of interactive controls cannot be captured by the role attribute alone. We have ARIA properties to add features that the screenreader needs to announce, such as aria-label, aria-haspopup, aria-activedescendant, or aria-live.

  Example :

  The following drop-down menu uses aria-haspopup to tell the screenreader that there is a popup hidden behind the menu button together with an ARIA state of aria-expanded to track whether it’s open or closed.

  Justify Left Center Right &lt;script&gt;<br /> var button = document.getElementById(&quot;button&quot;);<br /> var menu = document.getElementById(&quot;menu&quot;);<br /> var items = document.getElementsByClassName(&quot;menuitem&quot;);<br /> var focused = 0;<br /> function showMenu(evt) {<br />    evt.stopPropagation();<br />    menu.style.visibility = 'visible';<br />    button.setAttribute('aria-expanded','true');<br />    focused = getSelected();<br />    items[focused].focus();<br />  }<br />  function hideMenu(evt) {<br />    evt.stopPropagation();<br />    menu.style.visibility = 'hidden';<br />    button.setAttribute('aria-expanded','false');<br />    button.focus();<br />  }<br />  function getSelected() {<br />    for (var i=0; i &lt; items.length; i++) {<br />      if (items[i].getAttribute('aria-checked') == 'true') {<br />        return i;<br />      }<br />    }<br />  }<br />  function setSelected(elem) {<br />    var curSelected = getSelected();<br />    items[curSelected].setAttribute('aria-checked', 'false');<br />    elem.setAttribute('aria-checked', 'true');<br />  }<br />  function selectItem(evt) {<br />    setSelected(evt.target);<br />    hideMenu(evt);<br />  }<br /> function getPrevItem(index) {<br />    var prev = index - 1;<br />    if (prev &lt; 0) {<br />      prev = items.length - 1;<br />    }<br />    return prev;<br />  }<br />  function getNextItem(index) {<br />    var next = index + 1;<br />    if (next == items.length) {<br />      next = 0;<br />    }<br />    return next;<br />  }<br /> function handleButtonKeys(evt) {<br />    evt.stopPropagation();<br />    var key = evt.keyCode;<br />    switch(key) {<br />      case (13): /* ENTER */<br />      case (32): /* SPACE */<br />        showMenu(evt);<br />      default:<br />    }<br />  }<br />  function handleMenuKeys(evt) {<br />    evt.stopPropagation();<br />    var key = evt.keyCode;<br />    switch(key) {<br />      case (38): /* UP */<br />        focused = getPrevItem(focused);<br />        items[focused].focus();<br />        break;<br />      case (40): /* DOWN */<br />        focused = getNextItem(focused);<br />        items[focused].focus();<br />        break;<br />      case (13): /* ENTER */<br />      case (32): /* SPACE */<br />        setSelected(evt.target);<br />          hideMenu(evt);<br />          break;<br />      case (27): /* ESC */<br />        hideMenu(evt);<br />         break;<br />      default:<br />    }<br />  }<br />  button.addEventListener('click', showMenu, false);<br />  button.addEventListener('keydown', handleButtonKeys, false);<br />  for (var i = 0;  i &lt; items.length; i++) {<br />    items[i].addEventListener('click', selectItem, false);<br />    items[i].addEventListener('keydown', handleMenuKeys, false);<br />  }<br /> &lt;/script&gt;

  <div class="custombutton" id="button" tabindex="0" role="button" aria-expanded="false" aria-haspopup="true"> <span>Justify</span> </div> <div role="menu" class="menu" id="menu" style="display : none ;"> <div tabindex="0" role="menuitem" class="menuitem" aria-checked="true"> Left </div> <div tabindex="0" role="menuitem" class="menuitem" aria-checked="false"> Center </div> <div tabindex="0" role="menuitem" class="menuitem" aria-checked="false"> Right </div> </div> [CSS and JavaScript for example omitted]

  3.4. Labelling

  The main issue that people know about accessibility seems to be that they have to put alt text onto images. This is only one means to provide labels to screenreaders for page content. Labels are short informative pieces of text that provide a name to a control.

  There are actually several ways of providing labels for controls :

  • on img elements use @alt
  • on input elements use the label element
  • use @aria-labelledby if there is another element that contains the label
  • use @title if you also want a label to be used as a tooltip
  • otherwise use @aria-label

  I’ll provide examples for the first two use cases - the other use cases are simple to deduce.

  Example :

  The following two images show the rough concept for providing alt text for images : images that provide information should be transcribed, images that are just decorative should receive an empty @alt attribute.

  Image by Noah Sussman

  <img src="texture.jpg" alt=""> <img src="lolcat.jpg" alt="shocked lolcat titled ’HTML cannot do that !"> <img src="texture.jpg" alt="">

  When marking up decorative images with an empty @alt attribute, the image is actually completely removed from the accessibility tree and does not confuse the blind user. This is a desired effect, so do remember to mark up all your images with @alt attributes, even those that don’t contain anything of interest to AT.

  Example :

  In the example form above in Section 2.3, when tabbing directly on the input elements, the screen reader will only say "edit text" without announcing what meaning that text has. That’s not very useful. So let’s introduce a label element for the input elements. We’ll also add checkboxes with a label.

  Doctor title : Firstname : Lastname : Address : City : Remember me :

  <label>Doctor title :</label> <input type="checkbox" id="doctor"/> <label>Firstname :</label> <input type="text" id="firstname2"/> <label for="lastname2">Lastname :</label> <input type="text" id="lastname2"/> <label>Address : <input type="text" id="address2"> </label> <label for="city2">City : <input type="text" id="city2"> </label> <label for="remember">Remember me :</label> <input type="checkbox" id="remember">

  In this example we use several different approaches to show what a different it makes to use the <label> element to mark up input boxes.

  The first two fields just have a <label> element next to a <input> element. When using a screenreader you will not notice a difference between this and not using the <label> element because there is no connection between the <label> and the <input> element.

  In the third field we use the @for attribute to create that link. Now the input field isn’t just announced as "edit text", but rather as "Lastname edit text", which is much more useful. Also, the screenreader can now skip the labels and get straight on the input element.

  In the fourth and fifth field we actually encapsulate the <input> element inside the <label> element, thus avoiding the need for a @for attribute, though it doesn’t hurt to explicity add it.

  Finally we look at the checkbox. By including a referenced <label> element with the checkbox, we change the screenreaders announcement from just "checkbox not checked" to "Remember me checkbox not checked". Also notice that the click target now includes the label, making the checkbox not only more usable to screenreaders, but also for mouse users.

  4. Conclusions

  This article introduced a process that you can follow to make your Web applications accessible. As you do that, you will noticed that there are other things that you may need to do in order to give the best experience to a power user on a keyboard, a blind user using a screenreader, or a vision-impaired user using a screen magnifier. But once you’ve made a start, you will notice that it’s not all black magic and a lot can be achieved with just a little markup.

  You will find more markup in the WAI ARIA specification and many more resources at Mozilla’s ARIA portal. Now go and change the world !

  Many thanks to Alice Boxhall and Dominic Mazzoni for their proof-reading and suggested changes that really helped improve the article !

• Announcing TMPGEnc 4 : now with x264 !

  26 novembre 2010, par Dark Shikari — commercial, japan, licensing, x264

  A few months ago, we announced a commercial licensing program so that even companies unable to use GPL software in their products have a chance to use the open source x264 instead of proprietary alternatives. The system worked on two basic concepts. First, all licensees would still be required to give their changes to x264 back to us : x264 must forever remain free, with no useful contributions kept hidden from the community. Second, all the profits would go directly back to x264, primarily to the developers who’ve made the most significant contributions to x264 over the years, but also to funding future development, bounties for new features, as well as contributing to other related projects (e.g. Videolan and ffmpeg).

  Over the past couple of months, we’ve gotten an enormous response ; over 40 companies have inquired about licensing, with more contacting us every day. Due to the sheer volume of interest, we’ve partnered with CoreCodec, the creators of the free Matroska container format and developers of CoreAVC, to make x264 as widely available as possible in the world of commercial software as it is in the world of open source. All of this is already filtering back to benefiting x264 users, with many bugs being reported by commercial licensees as well as some code contributed.

  Today, we announce the first commercial consumer encoding software to switch to x264 : Pegasys Inc.’s TMPGEnc. Expect many more to follow : with x264 now available commercially as well as freely, there are few excuses left to use any other H.264 encoder. Vendors of overpriced, underpowered proprietary competitors should begin looking for new jobs.

  (Pegasys press release : English, Japanese)

• Bit-field badness

  30 janvier 2010, par Mans — Compilers, Optimisation

  Consider the following C code which is based on an real-world situation.

  struct bf1_31 
      unsigned a:1 ;
      unsigned b:31 ;
   ;
  void func(struct bf1_31 *p, int n, int a)
  
  
  
      int i = 0 ;
  
      do 
  
          if (p[i].a)
  
              p[i].b += a ;
  
       while (++i < n) ;
  
  
  

  How would we best write this in ARM assembler ? This is how I would do it :
  

  func :
          ldr     r3,  [r0], #4
          tst     r3,  #1
          add     r3,  r3,  r2,  lsl #1
          strne   r3,  [r0, #-4]
          subs    r1,  r1,  #1
          bgt     func
          bx      lr
  

  The add instruction is unconditional to avoid a dependency on the comparison. Unrolling the loop would mask the latency of the ldr instruction as well, but that is outside the scope of this experiment.

  Now compile this code with gcc -march=armv5te -O3 and watch in horror :

  func :
          push    r4
          mov     ip, #0
          mov     r4, r2
  loop :
          ldrb    r3, [r0]
          add     ip, ip, #1
          tst     r3, #1
          ldrne   r3, [r0]
          andne   r2, r3, #1
          addne   r3, r4, r3, lsr #1
          orrne   r2, r2, r3, lsl #1
          strne   r2, [r0]
          cmp     ip, r1
          add     r0, r0, #4
          blt     loop
          pop     r4
          bx      lr
  

  This is nothing short of awful :

  • The same value is loaded from memory twice.
  • A complicated mask/shift/or operation is used where a simple shifted add would suffice.
  • Write-back addressing is not used.
  • The loop control counts up and compares instead of counting down.
  • Useless mov in the prologue ; swapping the roles or r2 and r4 would avoid this.
  • Using lr in place of r4 would allow the return to be done with pop {pc}, saving one instruction (ignoring for the moment that no callee-saved registers are needed at all).

  Even for this trivial function the gcc-generated code is more than twice the optimal size and slower by approximately the same factor.

  The main issue I wanted to illustrate is the poor handling of bit-fields by gcc. When accessing bitfields from memory, gcc issues a separate load for each field even when they are contained in the same aligned memory word. Although each load after the first will most likely hit L1 cache, this is still bad for several reasons :

  • Loads have typically two or three cycles result latency compared to one cycle for data processing instructions. Any bit-field can be extracted from a register with two shifts, and on ARM the second of these can generally be achieved using a shifted second operand to a following instruction. The ARMv6T2 instruction set also adds the SBFX and UBFX instructions for extracting any signed or unsigned bit-field in one cycle.
  • Most CPUs have more data processing units than load/store units. It is thus more likely for an ALU instruction than a load/store to issue without delay on a superscalar processor.
  • Redundant memory accesses can trigger early flushing of store buffers rendering these less efficient.

  No gcc bashing is complete without a comparison with another compiler, so without further ado, here is the ARM RVCT output (armcc --cpu 5te -O3) :

  func :
          mov     r3, #0
          push    r4, lr
  loop :
          ldr     ip, [r0, r3, lsl #2]
          tst     ip, #1
          addne   ip, ip, r2, lsl #1
          strne   ip, [r0, r3, lsl #2]
          add     r3, r3, #1
          cmp     r3, r1
          blt     loop
          pop     r4, pc
  

  This is much better, the core loop using only one instruction more than my version. The loop control is counting up, but at least this register is reused as offset for the memory accesses. More remarkable is the push/pop of two registers that are never used. I had not expected to see this from RVCT.

  Even the best compilers are still no match for a human.