tft lcd frame buffer free sample

This library implements a SPI driver for the ILI9341 screen controller providing the ability to display memory framebuffers onto the screen very efficiently.

Differential redraws. The driver compares the framebuffer to be uploaded with the previous one and uploads only the pixels that differ. It does so in a smart way to minimize SPI transactions. Uploading only parts of the screen makes it possible to achieve very high frame rates even with low SPI speeds.

Asynchronous updates via DMA. Upload can be performed directly or asynchronously using DMA so the MCU is free to do other tasks - like generating the next frame - during updates.

adjustable framerate. The screen refresh rate can be adjusted between 40hz and 130Hz and a fixed framerate can be set within the driver. Uploads are timed to meet the requested framerate.

Multiple buffering methods. Support direct upload and double buffering. Partial updates of the screen, with direct or deferred redraw, are also available.

(2) The library"s sole purpose is to perform framebuffer upload from memory to the screen. It does not provide any drawing primitive. You must use another canvas library to draw directly onto the memory framebuffer. To do so, you may use my TGX library which provides optimized drawing primitives for 2D and 3D graphics.

The library can work with any SPI bus. Multiples instances of the driver can manage multiple displays on different SPI buses. A significant speedup is possible when the DC pin from the ILI9341 screen is connected to a hardware CS (chip select) capable pin on the Teensy... Yes, this requirement may seems weird at first ! In that case, the library will use the SPI FIFO and DMA to their full capabilities which increases the framerate (around 35% faster) while reducing CPU usage (by around 20%).

The above buffer has been placed in the upper 512K (dmamem) portion of the memory to preserve the RAM in the faster lower portion (dtcm). Buffers can be placed anywhere in RAM (even in EXTMEM if external ram is present but there will be a speed penalty in that case).

Remark. Not using an internal framebuffer is possible but it will disable asynchronous and differential updates, thus removing most of the library benefit... ADVICE: always use double buffering !

When the library is working in double buffering mode, it can take advantage of the fact that the last internal buffer uploaded to the screen mirrors the screen content. Therefore, when a new framebuffer is to be uploaded, the library can quickly check which pixels have changed and only upload those. This provides a huge speed bump when only a fraction of pixels changes at each frame (which is common when displaying a user interface for instance). The library is said to use differential updates in that case. When performing differential updates, the library first creates a diff log of all the changes between the new framebuffer and the previous one. Once this is accomplished, it can overwrite the internal frame buffer containing the old frame with the new one and, when an update is needed, just read the diff to select the pixels that must be pushed onto the screen. Updates can be carried away asynchronously while the user code draws the next frame.

The template parameter above specifies the size (in bytes) of the buffer. It should range from 1K to 10K depending on the kind of content displayed. The method diff1.printStats() can be used to check the diff buffer memory consumption at runtime to help choose its correct size.

Remark. You can set either 0, 1 or 2 diff buffers. Not providing any diff buffer will disable differential updates. On the other hand, providing 2 diff buffers allows the driver to compute the new diff while the previous one is currently in use for an upload and thus provides a nice speed bump. ADVICE: always use two diff buffers !.

The constructor of the main ILI9341Driver object tft does not initialize anything. We must do that when ready by calling the ubiquitous arduino begin() method (usually within setup()) like so:

This method returns true if the screen was correctly initialized. Note that there is no corresponding end() method so begin() should normally be called only once (but it can still be called again to issue a hard reset). The SPI_WRITE_SPEED and SPI_READ_SPEED parameters can be omitted so default speeds for write/read SPI will be used. The SPI read speed does not really matter and should not be changed. On the other hand, the maximum possible framerate is proportional to the SPI write speed so it should be set as high as possible while still keeping a stable connexion. With short wires, many displays will easily accept speeds upward of 60Mhz...

There are 4 possible orientations for the display (matching, of course, to the four physical orientations). Each one is obtained from the previous one by rotating the screen 90 degrees clockwise. Orientation 0 and 2 are in portrait mode 240x320 whereas orientation 1 and 3 are in landscape mode 320x240. In all cases, the framebuffer layout is in row-major format: if the screen has size LX x LY, then

The "natural" orientation is orientation=0 for which the pixels in the framebuffer are ordered the same way that they are refreshed on the screen. This orientation will give the best possible upload rate and should be favored if possible.

The method can be called again to change the buffering mode on the fly. Calling this method without a parameter removes the current internal framebuffer (and the drivers switches to the inefficient direct upload mode).

Note that, without an internal framebuffer, differential updates cannot be performed so any registered diff buffers will be ignored (until an internal framebuffer is set). Whenever an internal framebuffer and at least 1 diff buffer are registered, differential updates are automatically enabled. As mentioned before: always favor 2 diffs buffers instead of 1 !

The framerate on the other hand is the number of times the screen content is actually updated by the driver every second. When pixel upload and pixel refresh are not synchronized, we end up with the screen displaying part of the old and the new frame simultaneously which create a visual defect called screen tearing.

In order to prevent this, the driver keeps track of the refresh times and tries to upload pixels trailing behind the scanline. However, if the upload rate is much slower than the refresh rate, then the scanline will still eventually catch up with the pixels being uploaded and screen tearing will occur. In order to get prevent visual artifact and insure a stable framerate, the following two conditions should be met:

The first condition depends heavily on the SPI speed. For example, with SPI set at 60Mhz, it is possible to upload up to 45 full frames per second. Being cautious, we can set a refresh rate at 80Hz and a framerate at 40Hz = (80/2) to obtain tear free frames on the screen. However, this computation is a worst case scenario: for most usage, differential updates boost the upload speed tremendously so it is often possible to get "tear free" display at 60Hz framerate (and 120Hz refresh rate) with only 20Mhz SPI speed !

Now, we must tell the driver the actual framerate/vsync method that we want. This is done with the tft.setVsyncSpacing() method. It takes as input a vsync_spacing parameter (what a surprise!)  which has the following meaning:

vsync_spacing = -1. Do not synchronize display updates with screen refreshes (no vsync).  Each new frame is drawn immediately onto the screen or saved into the internal framebuffer to be drawn asap if there is already an update in progress. If the internal framebuffer is in use, the frame is simply dropped...

vsync_spacing = 0. Do not synchronize display updates with screen refreshes (no vsync). Each new frame is drawn immediately onto the screen or saved into the internal framebuffer to be drawn asap if there is already an update in progress. If the internal framebuffer is in use, the method waits for a buffer to become available before returning (same as above but no frame is ever dropped).

vsync_spacing = N > 0. Synchronize uploads with screen refreshes to mitigate screen tearing. Perform upload every N screen refreshes so that the actual framerate is equal to framerate=refreshrate/N (provided, of course, that frames are pushed fast enough to the driver to achieve this rate).

In practice, vsync_spacing=-1 will give the fastest apparent framerate but will usually provide very poor visual quality. Setting vsync_spacing=-0 will give slightly better results (but still with screen tearing) and still leaves the responsibility to the user of setting a stable framerate by pushing frames at regular intervals... In most cases, the best choice is to set vsync_spacing = 2 and then adjust the refresh rate so that uploads can be performed at refreshrate/2 FPS... Using vsync_spacing = 1 should be reserved for drawing very simple frames which can be uploaded very quickly onto the screen (in less than a refresh period). Using vsync_pacing >= 3 can be used to artificially reduce the framerate but I do not know of a really compelling reason to do so.

What actually happens when the method is called depends on the buffering and vsync mode but, as a general rule, the method will try to return to the caller as soon as possible. It will use its internal framebuffer (if present) to save a copy of fb and return while uploading the frame asynchronously via DMA. So, when the method returns, fb may, or not, already be displayed on the screen but, in any case, the framebuffer can be immediately reused for drawing the next frame, changing its content will not affect the display on the screen !.

The setDiffGap() method. When using differential updates, the driver tries to be smart and find a compromise between skipping unchanged pixels but also not fragmenting spi transactions too much because issuing a re-positioning commands also takes times. To adjust this behaviour, the setDiffGap() can be used to specify the number of consecutive unchanged pixels required to break a spi transaction. Typical value should range between 3 and 40. Smaller gaps can provide a speed bump but will require larger diff buffers (possibly up to 10K when using a gap of size 4). It is possible to get statistics on diff buffer memory consumption with the printStats() method. If the diff buffer overflows too often, its size should be increased.

Disabling differential update for a given frame. Differential updates are beneficial in most cases unless almost all pixels change in the frame. In this case, there will be no increase in upload speed. Yet, calculating the diff log takes around 300us to 1ms of CPU time per frame. When using two diff buffers, this computation is done during DMA transfer so it will not slow down the framerate but it can still be beneficial to skip this computation if you already know for sure that the diff will be mostly trivial. You can tell the driver to upload the full frame as is, without computing the diff, by setting the second (facultative) parameter in the update method to true:

Drawing text on the framebuffer The overlayText() method can be used to draw some text onto the framebuffer. This is useful for displaying basic informations or debugging. Similarly, the method overlayFPS() draws the current instantaneous framerate onto a given framebuffer. Both methods overlayText() and overlayFPS() are typically called just before calling update().

diff buffer and memory allocation. The library performs no memory allocation. All the memory needed (framebuffer and diff buffers) are to be provided by the user which keeps complete control over memory allocation. For diff buffers, the StaticDiffBuffer<> template class provides a convenient way to create diff buffers with statically allocated memory. However, if more control is needed, one can use the base DiffBuffer class which is similar but requires the user to provide the memory space at construction time. See the file DiffBuff.h for additional details.

Getting information, additional methods. There are several other methods that can be used to fine tune the driver performance. In particular: resync(), setDiffCompareMask, setLateFrameRatio()... Details about these methods (and more) can be found in the header file ILI9341Driver.h. Each method has a detailed docstring above its declaration explaining its purpose.

Last but not least: drawing things on the frame buffer. The library itself provides not drawing primitive. It simply pushes/mirror a memory frame buffer onto the screen. You can draw on the framebuffer directly "by hand" or use any library you wish to do so... If you want a lightweight, fast, full featured 2D and 3D graphics library optimized for microcontrollers, you should check out my tgx library