The 2D GDI For Windows XP Through Windows 7, In Detail
XP: Clear Sailing without Competition
Up until (and including) Windows XP, GDI played a key role in rendering 2D graphics. The presence of easily-simplified procedures makes this patently obvious. The mouse movements used to draw a line are transmitted to win32.sys, the central clearinghouse for graphics input. It doesn’t matter whether we’re using the mouse, keystrokes, or other graphical inputs; all such data congregates in this routine and goes directly to 2D graphics rendering modules form there. Our user actions include only 2D graphics information, which gets translated immediately into GDI drawing instructions. These are forwarded to the GDI, as illustrated by the purple arrows in the following diagram.
These simple procedures used to handle 2D graphics in software also explain why it’s so easy to convert them to hardware acceleration, provided that the graphics card offers the necessary capabilities to render them independently. The blue arrow in the preceding figure shows how information returns to the calling application, so that it may be notified that window contents have changed (for example, when other windows may no longer obscure some of its visible content), thereby forcing a redraw.
Windows Vista: CPU instead of GPU, and Buffering instead of Direct Delivery
As we explained in Part 1 of this story, Vista introduces a completely new path for graphics data through the OS. Using the GDI, all versions of Windows up to and including XP handled 2D drawing through outputs from win32k.sys to manage window contents on-screen.
In Vista, the DWM (dynamic window manager) takes over this role. As a consequence, Vista uses only Direct3D to manage windows instead. Every windows for every application is written to the texture storage as a 3D texture map on the graphics card. This is a practical evolution for more modern graphics cards, but it also means that GDI can no longer read from or write to this data. The communications chain appears to be broken in this situation.
At this point, the double buffering of window content that we explored in Part 1 of this story comes into play.
What exactly is going on here? Look at the red arrows in the preceding diagram. In the place of a unified graphics driver (in XP this is called the DDI Display Driver) the new CDD (Canonical Display Driver) is addressed instead. This module is independent from the graphics card in Vista. While the pending window content is stored as a texture map in the graphics card RAM, each window must also be stored in an equivalent buffer in the system RAM as well (its size equals window width times window height times four bytes for 32-bit color data).
The most current rendering of each window get transformed into a bitmap in the system RAM buffer, after which it is converted into a 3D texture map in the video RAM on the graphics card. Throughout, the DWM manages all windows and moves their contents around using Direct3D. The DWM also contains data about which portions of each window are visible on-screen, so that when any region in a window becomes obscured or gets revealed it can be redrawn (shown by the blue arrow in the preceding figure). At this moment, the DWM copies the contents of system RAM into the video RAM, re-rendering the window using Direct3D. The applications no longer need to redraw the window (in contrast to the way things worked under Windows XP).
The aforementioned approach effectively disables 2D hardware acceleration, resulting in a significant performance reduction compared to Windows XP. This manifests itself most clearly in Vista’s well-documented tendency to drag on 2D graphics and to consume large amounts of RAM.
Windows 7: Hardware Acceleration in Miniscule Doses
Even in our initial testing for Part 1 of this story, we could tell that Windows 7 once again offered at least partial support for hardware acceleration of GDI commands—that is, for cards with WDDM 1.1 drivers. Where such drivers are not available (for example, on some Intel graphics chipsets), Windows 7 behaves more or less like Vista. What does this mean for us exactly? Let’s take a look at a diagram of graphics flow in Windows 7:
At first glance, things look pretty much the same as they did under Vista. We can see, however, that it’s no longer necessary to double-buffer every widow’s contents. Instead of system RAM, the term aperture memory now comes into play. This refers to a specific region within the normal system memory that the graphics card can access directly. If a window area changes because of movement or overlays, those window contents may be copied directly from this memory range to the video RAM on the graphics card.
By comparison with Windows XP, only a subset of the GDI commands are supported in the GPU—namely, ClearType, ColorFill, BitBlt, AlphaBlend, TransparentBlt, and StretchBlt. Here’s the skinny for those not already in the know: this means direct text output, surface area fills with simple colors, and copying of image contents and transparent overlays. Whereas rendering of complex geometrical figures isn’t supported at all, copies of image contents and area fills can easily be transferred from aperture memory directly to video RAM.
Windows 7 reduces memory usage by eliminating most of the double buffering of window contents. Even Vista benefits from some of the same effects, thanks to advances resulting from the newer WDDM driver model. That’s why hardware acceleration is once again possible, thanks to the new platform update (which occurred in tandem with the introduction of DirectX 11) for Windows 7. Those specifics are what we hope to chase down in the rest of this story.