Update library.properties - bump to 1.3.2

adding pin cfg for hstx dvi cowbell

.github/ISSUE_TEMPLATE.md

@@ -0,0 +1,46 @@
+Thank you for opening an issue on an Adafruit Arduino library repository.  To
+improve the speed of resolution please review the following guidelines and
+common troubleshooting steps below before creating the issue:
+
+- **Do not use GitHub issues for troubleshooting projects and issues.**  Instead use
+  the forums at http://forums.adafruit.com to ask questions and troubleshoot why
+  something isn't working as expected.  In many cases the problem is a common issue
+  that you will more quickly receive help from the forum community.  GitHub issues
+  are meant for known defects in the code.  If you don't know if there is a defect
+  in the code then start with troubleshooting on the forum first.
+
+- **If following a tutorial or guide be sure you didn't miss a step.** Carefully
+  check all of the steps and commands to run have been followed.  Consult the
+  forum if you're unsure or have questions about steps in a guide/tutorial.
+
+- **For Arduino projects check these very common issues to ensure they don't apply**:
+
+  - For uploading sketches or communicating with the board make sure you're using
+    a **USB data cable** and **not** a **USB charge-only cable**.  It is sometimes
+    very hard to tell the difference between a data and charge cable!  Try using the
+    cable with other devices or swapping to another cable to confirm it is not
+    the problem.
+
+  - **Be sure you are supplying adequate power to the board.**  Check the specs of
+    your board and plug in an external power supply.  In many cases just
+    plugging a board into your computer is not enough to power it and other
+    peripherals.
+
+  - **Double check all soldering joints and connections.**  Flakey connections
+    cause many mysterious problems.  See the [guide to excellent soldering](https://learn.adafruit.com/adafruit-guide-excellent-soldering/tools) for examples of good solder joints.
+
+  - **Ensure you are using an official Arduino or Adafruit board.** We can't
+    guarantee a clone board will have the same functionality and work as expected
+    with this code and don't support them.
+
+If you're sure this issue is a defect in the code and checked the steps above
+please fill in the following fields to provide enough troubleshooting information.
+You may delete the guideline and text above to just leave the following details:
+
+- Arduino board:  **INSERT ARDUINO BOARD NAME/TYPE HERE**
+
+- Arduino IDE version (found in Arduino -> About Arduino menu):  **INSERT ARDUINO
+  VERSION HERE**
+
+- List the steps to reproduce the problem below (if possible attach a sketch or
+  copy the sketch code in too): **LIST REPRO STEPS BELOW**

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,26 @@
+Thank you for creating a pull request to contribute to Adafruit's GitHub code!
+Before you open the request please review the following guidelines and tips to
+help it be more easily integrated:
+
+- **Describe the scope of your change--i.e. what the change does and what parts
+  of the code were modified.**  This will help us understand any risks of integrating
+  the code.
+
+- **Describe any known limitations with your change.**  For example if the change
+  doesn't apply to a supported platform of the library please mention it.
+
+- **Please run any tests or examples that can exercise your modified code.**  We
+  strive to not break users of the code and running tests/examples helps with this
+  process.
+
+Thank you again for contributing!  We will try to test and integrate the change
+as soon as we can, but be aware we have many GitHub repositories to manage and
+can't immediately respond to every request.  There is no need to bump or check in
+on a pull request (it will clutter the discussion of the request).
+
+Also don't be worried if the request is closed or not integrated--sometimes the
+priorities of Adafruit's GitHub code (education, ease of use) might not match the
+priorities of the pull request.  Don't fret, the open source community thrives on
+forks and GitHub makes it easy to keep your changes in a forked repo.
+
+After reviewing the guidelines above you can delete this text from the pull request.

.github/workflows/githubci.yml

@@ -0,0 +1,32 @@
+name: Arduino Library CI
+
+on: [pull_request, push]
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    
+    steps:
+    - uses: actions/setup-python@v4
+      with:
+        python-version: '3.x'
+    - uses: actions/checkout@v3
+    - uses: actions/checkout@v3
+      with:
+         repository: adafruit/ci-arduino
+         path: ci
+
+    - name: pre-install
+      run: bash ci/actions_install.sh
+
+    - name: test platforms
+      run: python3 ci/build_platform.py pico_rp2040
+
+#    - name: clang
+#      run: python3 ci/run-clang-format.py -e "ci/*" -e "bin/*" -r . 
+
+    - name: doxygen
+      env:
+        GH_REPO_TOKEN: ${{ secrets.GH_REPO_TOKEN }}
+        PRETTYNAME : "PicoDVI - Adafruit Fork"
+      run: bash ci/doxy_gen_and_deploy.sh

.gitignore

@@ -1,2 +1,5 @@
+Doxyfile*
+doxygen_sqlite3.db
+html
 .vscode
 build

Readme.md

@@ -1,3 +1,45 @@
+PicoDVI - Adafruit Fork for Arduino IDE + Adafruit_GFX compatibility
+====================================================================
+(Original Readme content follows later)
+
+Implements a few framebuffer types to which Adafruit_GFX drawing operations
+can be made; not every permutation PicoDVI is capable of, but a useful subset.
+
+WARNING: all video modes require overclocking (performed automatically at
+run-time, nothing to select), occasionally over-volting (optionally selected
+in sketch code), and higher resolutions may require reducing the QSPI clock
+rate (Tools menu in future arduino-pico release). The RP2040 microcontroller
+is being run WAY beyond spec and there is a VERY SMALL BUT NONZERO
+POSSIBILITY OF PERMANENT DAMAGE. Please see LICENSE file; usual software
+disclaimers about liability apply, user accepts risk.
+
+Requires Earle Philhower III RP2040 Arduino core (not the "official" Arduino
+RP2040 core).
+
+Changes vs main PicoDVI repo:
+- Add library.properties file, src and examples directories per Arduino
+requirements.
+- A full copy of software/libdvi is made in src (originally was soft-linked but Arduino Library Manager does not approve). If any updates are made in the original PicoDVI libdvi directory, copy them here!
+- The file dvi_serialiser.pio.h, normally not part of the distribution and
+generated during the Pico SDK build process, is provided here for Arduino
+build to work. If any changes are made in dvi_serialiser.pio (either here
+or by bringing in updates from upstream code), this header will need to be
+regenerated.
+- extern "C" around most function defs, to be linkable with C++ (Arduino).
+- A couple compile-time constants have been changed to run-time configurable
+because some color and resolution things aren't known until a constructor is
+called: dvi_vertical_repeat and dvi_monochrome_tmds.
+- DVI_1BPP_BIT_REVERSE is switched '1' by default (to work with bit order
+used by GFXcanvas1 object). Pico SDK-built examples using 1-bit mode are
+fixed by changing the corresponding CMakeLists.txt files to set this to 0.
+Font data has had bit order reversed to match this layout.
+
+All files from the PicoDVI repo are kept even if not used in this build
+(e.g. apps and assets directories, among others) so same repo can still be
+used via Pico SDK if desired.
+
+Original Readme content:
+=======
 RP2350 PicoDVI Preview
 ======================
 

examples/16bit_hello/16bit_hello.ino

@@ -0,0 +1,619 @@
+// Basic full-color PicoDVI test. Provides a 16-bit color video framebuffer to
+// which Adafruit_GFX calls can be made. It's based on the EYESPI_Test.ino sketch.
+
+#include <PicoDVI.h>                  // Core display & graphics library
+#include <Fonts/FreeSansBold18pt7b.h> // A custom font
+
+// Here's how a 320x240 16-bit color framebuffer is declared. Double-buffering
+// is not an option in 16-bit color mode, just not enough RAM; all drawing
+// operations are shown as they occur. Second argument is a hardware
+// configuration -- examples are written for Adafruit Feather RP2040 DVI, but
+// that's easily switched out for boards like the Pimoroni Pico DV (use
+// 'pimoroni_demo_hdmi_cfg') or Pico DVI Sock ('pico_sock_cfg').
+DVIGFX16 display(DVI_RES_320x240p60, adafruit_feather_dvi_cfg);
+
+// A 400x240 mode is possible but pushes overclocking even higher than
+// 320x240 mode. SOME BOARDS MIGHT SIMPLY NOT BE COMPATIBLE WITH THIS.
+// May require selecting QSPI div4 clock (Tools menu) to slow down flash
+// accesses, may require further over-volting the CPU to 1.25 or 1.3 V.
+//DVIGFX16 display(DVI_RES_400x240p60, adafruit_feather_dvi_cfg);
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+}
+
+#define PAUSE 2000  // Delay (milliseconds) between examples
+uint8_t rotate = 0; // Current screen orientation (0-3)
+#define CORNER_RADIUS 0
+
+void loop() {
+  // Each of these functions demonstrates a different Adafruit_GFX concept:
+  show_shapes();
+  show_charts();
+  show_basic_text();
+  show_char_map();
+  show_custom_text();
+  show_bitmap();
+  show_canvas();
+
+  if (++rotate > 3) rotate = 0; // Cycle through screen rotations 0-3
+  display.setRotation(rotate);  // Takes effect on next drawing command
+}
+
+// BASIC SHAPES EXAMPLE ----------------------------------------------------
+
+void show_shapes() {
+  // Draw outlined and filled shapes. This demonstrates:
+  // - Enclosed shapes supported by GFX (points & lines are shown later).
+  // - Adapting to different-sized displays, and to rounded corners.
+
+  const int16_t cx = display.width() / 2;  // Center of screen =
+  const int16_t cy = display.height() / 2; // half of width, height
+  int16_t minor = min(cx, cy);             // Lesser of half width or height
+  // Shapes will be drawn in a square region centered on the screen. But one
+  // particular screen -- rounded 240x280 ST7789 -- has VERY rounded corners
+  // that would clip a couple of shapes if drawn full size. If using that
+  // screen type, reduce area by a few pixels to avoid drawing in corners.
+  if (CORNER_RADIUS > 40) minor -= 4;
+  const uint8_t pad = 5;                   // Space between shapes is 2X this
+  const int16_t size = minor - pad;        // Shapes are this width & height
+  const int16_t half = size / 2;           // 1/2 of shape size
+
+  display.fillScreen(0); // Start by clearing the screen; color 0 = black
+
+  // Draw outline version of basic shapes: rectangle, triangle, circle and
+  // rounded rectangle in different colors. Rather than hardcoded numbers
+  // for position and size, some arithmetic helps adapt to screen dimensions.
+  display.drawRect(cx - minor, cy - minor, size, size, 0xF800);
+  display.drawTriangle(cx + pad, cy - pad, cx + pad + half, cy - minor,
+                       cx + minor - 1, cy - pad, 0x07E0);
+  display.drawCircle(cx - pad - half, cy + pad + half, half, 0x001F);
+  display.drawRoundRect(cx + pad, cy + pad, size, size, size / 5, 0xFFE0);
+  delay(PAUSE);
+
+  // Draw same shapes, same positions, but filled this time.
+  display.fillRect(cx - minor, cy - minor, size, size, 0xF800);
+  display.fillTriangle(cx + pad, cy - pad, cx + pad + half, cy - minor,
+                       cx + minor - 1, cy - pad, 0x07E0);
+  display.fillCircle(cx - pad - half, cy + pad + half, half, 0x001F);
+  display.fillRoundRect(cx + pad, cy + pad, size, size, size / 5, 0xFFE0);
+  delay(PAUSE);
+} // END SHAPE EXAMPLE
+
+// CHART EXAMPLES ----------------------------------------------------------
+
+void show_charts() {
+  // Draw some graphs and charts. GFX library doesn't handle these as native
+  // object types, but it only takes a little code to build them from simple
+  // shapes. This demonstrates:
+  // - Drawing points and horizontal, vertical and arbitrary lines.
+  // - Adapting to different-sized displays.
+  // - Graphics being clipped off edge.
+  // - Use of negative values to draw shapes "backward" from an anchor point.
+  // - C technique for finding array size at runtime (vs hardcoding).
+
+  display.fillScreen(0);  // Clear screen
+
+  const int16_t cx = display.width() / 2;  // Center of screen =
+  const int16_t cy = display.height() / 2; // half of width, height
+  const int16_t minor = min(cx, cy);       // Lesser of half width or height
+  const int16_t major = max(cx, cy);       // Greater of half width or height
+
+  // Let's start with a relatively simple sine wave graph with axes.
+  // Draw graph axes centered on screen. drawFastHLine() and drawFastVLine()
+  // need fewer arguments than normal 2-point line drawing shown later.
+  display.drawFastHLine(0, cy, display.width(), 0x0210);  // Dark blue
+  display.drawFastVLine(cx, 0, display.height(), 0x0210);
+
+  // Then draw some tick marks along the axes. To keep this code simple,
+  // these aren't to any particular scale, but a real program may want that.
+  // The loop here draws them from the center outward and pays no mind
+  // whether the screen is rectangular; any ticks that go off-screen will
+  // be clipped by the library.
+  for (uint8_t i=1; i<=10; i++) {
+    // The Arduino map() function scales an input value (e.g. "i") from an
+    // input range (0-10 here) to an output range (0 to major-1 here).
+    // Very handy for making graphics adjust to different screens!
+    int16_t n = map(i, 0, 10, 0, major - 1); // Tick offset relative to center point
+    display.drawFastVLine(cx - n, cy - 5, 11, 0x210);
+    display.drawFastVLine(cx + n, cy - 5, 11, 0x210);
+    display.drawFastHLine(cx - 5, cy - n, 11, 0x210);
+    display.drawFastHLine(cx - 5, cy + n, 11, 0x210);
+  }
+
+  // Then draw sine wave over this using GFX drawPixel() function.
+  for (int16_t x=0; x<display.width(); x++) { // Each column of screen...
+    // Note the inverted Y axis here (cy-value rather than cy+value)
+    // because GFX, like most graphics libraries, has +Y heading down,
+    // vs. classic Cartesian coords which have +Y heading up.
+    int16_t y = cy - (int16_t)(sin((x - cx) * 0.05) * (float)minor * 0.5);
+    display.drawPixel(x, y, 0xFFFF);
+  }
+
+  delay(PAUSE);
+
+  // Next, let's draw some charts...
+  // NOTE: some other examples in this code take extra steps to avoid placing
+  // anything off in the rounded corners of certain displays. The charts do
+  // not. It's *possible* but would introduce a lot of complexity into code
+  // that's trying to show the basics. We'll leave the clipped charts here as
+  // a teachable moment: not all content suits all displays.
+
+  // A list of data to plot. These are Y values only; X assumed equidistant.
+  const uint8_t data[] = { 31, 42, 36, 58, 67, 88 };       // Percentages, 0-100
+  const uint8_t num_points = sizeof data / sizeof data[0]; // Length of data[] list
+
+  display.fillScreen(0);  // Clear screen
+  display.setFont();      // Use default (built-in) font
+  display.setTextSize(2); // and 2X size for chart label
+  
+  // Chart label is centered manually; 144 is the width in pixels of
+  // "Widget Sales" at 2X scale (12 chars * 6 px * 2 = 144). A later example
+  // shows automated centering based on string.
+  display.setCursor((display.width() - 144) / 2, 0);
+  display.print(F("Widget Sales")); // F("string") is in program memory, not RAM
+  // The chart-drawing code is then written to skip the top 20 rows where
+  // this label is located.
+
+  // First, a line chart, connecting the values point-to-point:
+
+  // Draw a grid of lines to provide scale & an interesting background.
+  for (uint8_t i=0; i<11; i++) {
+    int16_t x = map(i, 0, 10, 0, display.width() - 1);   // Scale grid X to screen
+    display.drawFastVLine(x, 20, display.height(), 0x001F);
+    int16_t y = map(i, 0, 10, 20, display.height() - 1); // Scale grid Y to screen
+    display.drawFastHLine(0, y, display.width(), 0x001F);
+  }
+  // And then draw lines connecting data points. Load up the first point...
+  int16_t prev_x = 0;
+  int16_t prev_y = map(data[0], 0, 100, display.height() - 1, 20);
+  // Then connect lines to each subsequent point...
+  for (uint8_t i=1; i<num_points; i++) {
+    int16_t new_x = map(i, 0, num_points - 1, 0, display.width() - 1);
+    int16_t new_y = map(data[i], 0, 100, display.height() - 1, 20);
+    display.drawLine(prev_x, prev_y, new_x, new_y, 0x07FF);
+    prev_x = new_x;
+    prev_y = new_y;
+  }
+  // For visual interest, let's add a circle around each data point. This is
+  // done in a second pass so the circles are always drawn "on top" of lines.
+  for (uint8_t i=0; i<num_points; i++) {
+    int16_t x = map(i, 0, num_points - 1, 0, display.width() - 1);
+    int16_t y = map(data[i], 0, 100, display.height() - 1, 20);
+    display.drawCircle(x, y, 5, 0xFFFF);
+  }
+
+  delay(PAUSE);
+
+  // Then a bar chart of the same data...
+
+  // Erase the old chart but keep the label at top.
+  display.fillRect(0, 20, display.width(), display.height() - 20, 0);
+
+  // Just draw the Y axis lines; bar chart doesn't really need X lines.
+  for (uint8_t i=0; i<11; i++) {
+    int16_t y = map(i, 0, 10, 20, display.height() - 1);
+    display.drawFastHLine(0, y, display.width(), 0x001F);
+  }
+
+  int bar_width = display.width() / num_points - 4; // 2px pad to either side
+  for (uint8_t i=0; i<num_points; i++) {
+    int16_t x = map(i, 0, num_points, 0, display.width()) + 2; // Left edge of bar
+    int16_t height = map(data[i], 0, 100, 0, display.height() - 20);
+    // Some GFX functions (rects, H/V lines and similar) can accept negative
+    // width/height values. What this does is anchor the shape at the right or
+    // bottom coordinate (rather than the usual left/top) and draw back from
+    // there, hence the -height here (bar is anchored at bottom of screen):
+    display.fillRect(x, display.height() - 1, bar_width, -height, 0xFFE0);
+  }
+
+  delay(PAUSE);
+
+} // END CHART EXAMPLES
+
+// TEXT ALIGN FUNCTIONS ----------------------------------------------------
+
+// Adafruit_GFX only handles left-aligned text. This is normal and by design;
+// it's a rare need that would further strain AVR by incurring a ton of extra
+// code to properly handle, and some details would confuse. If needed, these
+// functions give a fair approximation, with the "gotchas" that multi-line
+// input won't work, and this operates only as a println(), not print()
+// (though, unlike println(), cursor X does not reset to column 0, instead
+// returning to initial column and downward by font's line spacing). If you
+// can work with those constraints, it's a modest amount of code to copy
+// into a project. Or, if your project only needs one or two aligned strings,
+// simply use getTextBounds() for a bounding box and work from there.
+// DO NOT ATTEMPT TO MAKE THIS A GFX-NATIVE FEATURE, EVERYTHING WILL BREAK.
+
+typedef enum { // Alignment options passed to functions below
+  GFX_ALIGN_LEFT,
+  GFX_ALIGN_CENTER,
+  GFX_ALIGN_RIGHT
+} GFXalign;
+
+// Draw text aligned relative to current cursor position. Arguments:
+// gfx   : An Adafruit_GFX-derived type (e.g. display or canvas object).
+// str   : String to print (as a char *).
+// align : One of the GFXalign values declared above.
+//         GFX_ALIGN_LEFT is normal left-aligned println() behavior.
+//         GFX_ALIGN_CENTER prints centered on cursor pos.
+//         GFX_ALIGN_RIGHT prints right-aligned to cursor pos.
+// Cursor advances down one line a la println(). Column is unchanged.
+void print_aligned(Adafruit_GFX &gfx, const char *str,
+                   GFXalign align = GFX_ALIGN_LEFT) {
+  uint16_t w, h;
+  int16_t  x, y, cursor_x, cursor_x_save;
+  cursor_x = cursor_x_save = gfx.getCursorX();
+  gfx.getTextBounds(str, 0, gfx.getCursorY(), &x, &y, &w, &h);
+  if (align == GFX_ALIGN_RIGHT)       cursor_x -= w;
+  else if (align == GFX_ALIGN_CENTER) cursor_x -= w / 2;
+  //gfx.drawRect(cursor_x, y, w, h, 0xF800);      // Debug rect
+  gfx.setCursor(cursor_x - x, gfx.getCursorY());  // Center/right align
+  gfx.println(str);
+  gfx.setCursor(cursor_x_save, gfx.getCursorY()); // Restore cursor X
+}
+
+// Equivalent function for strings in flash memory (e.g. F("Foo")). Body
+// appears identical to above function, but with C++ overloading it it works
+// from flash instead of RAM. Any changes should be made in both places.
+void print_aligned(Adafruit_GFX &gfx, const __FlashStringHelper *str,
+                   GFXalign align = GFX_ALIGN_LEFT) {
+  uint16_t w, h;
+  int16_t  x, y, cursor_x, cursor_x_save;
+  cursor_x = cursor_x_save = gfx.getCursorX();
+  gfx.getTextBounds(str, 0, gfx.getCursorY(), &x, &y, &w, &h);
+  if (align == GFX_ALIGN_RIGHT)       cursor_x -= w;
+  else if (align == GFX_ALIGN_CENTER) cursor_x -= w / 2;
+  //gfx.drawRect(cursor_x, y, w, h, 0xF800);      // Debug rect
+  gfx.setCursor(cursor_x - x, gfx.getCursorY());  // Center/right align
+  gfx.println(str);
+  gfx.setCursor(cursor_x_save, gfx.getCursorY()); // Restore cursor X
+}
+
+// Equivalent function for Arduino Strings; converts to C string (char *)
+// and calls corresponding print_aligned() implementation.
+void print_aligned(Adafruit_GFX &gfx, const String &str,
+                   GFXalign align = GFX_ALIGN_LEFT) {
+  print_aligned(gfx, const_cast<char *>(str.c_str()));
+}
+
+// TEXT EXAMPLES -----------------------------------------------------------
+
+// This section demonstrates:
+// - Using the default 5x7 built-in font, including scaling in each axis.
+// - How to access all characters of this font, including symbols.
+// - Using a custom font, including alignment techniques that aren't a normal
+//   part of the GFX library (uses functions above).
+
+void show_basic_text() {
+  // Show text scaling with built-in font.
+  display.fillScreen(0);
+  display.setFont();                   // Use default font
+  display.setCursor(0, CORNER_RADIUS); // Initial cursor position
+  display.setTextSize(1);              // Default size
+  display.println(F("Standard built-in font"));
+  display.setTextSize(2);
+  display.println(F("BIG TEXT"));
+  display.setTextSize(3);
+  // "BIGGER TEXT" won't fit on narrow screens, so abbreviate there.
+  display.println((display.width() >= 200) ? F("BIGGER TEXT") : F("BIGGER"));
+  display.setTextSize(2, 4);
+  display.println(F("TALL and"));
+  display.setTextSize(4, 2);
+  display.println(F("WIDE"));
+
+  delay(PAUSE);
+} // END BASIC TEXT EXAMPLE
+
+void show_char_map() {
+  // "Code Page 437" is a name given to the original IBM PC character set.
+  // Despite age and limited language support, still seen in small embedded
+  // settings as it has some useful symbols and accented characters. The
+  // default 5x7 pixel font of Adafruit_GFX is modeled after CP437. This
+  // function draws a table of all the characters & explains some issues.
+
+  // There are 256 characters in all. Draw table as 16 rows of 16 columns,
+  // plus hexadecimal row & column labels. How big can each cell be drawn?
+  const int cell_size = min(display.width(), display.height()) / 17;
+  if (cell_size < 8) return; // Screen is too small for table, skip example.
+  const int total_size = cell_size * 17; // 16 cells + 1 row or column label
+
+  // Set up for default 5x7 font at 1:1 scale. Custom fonts are NOT used
+  // here as most are only 128 characters to save space (the "7b" at the
+  // end of many GFX font names means "7 bits," i.e. 128 characters).
+  display.setFont();
+  display.setTextSize(1);
+
+  // Early Adafruit_GFX was missing one symbol, throwing off some indices!
+  // But fixing the library would break MANY existing sketches that relied
+  // on the degrees symbol and others. The default behavior is thus "broken"
+  // to keep older code working. New code can access the CORRECT full CP437
+  // table by calling this function like so:
+  display.cp437(true);
+
+  display.fillScreen(0);
+
+  const int16_t x = (display.width() - total_size) / 2;  // Upper left corner of
+  int16_t       y = (display.height() - total_size) / 2; // table centered on screen
+  if (y >= 4) { // If there's a little extra space above & below, scoot table
+    y += 4;     // down a few pixels and show a message centered at top.
+    display.setCursor((display.width() - 114) / 2, 0); // 114 = pixel width
+    display.print(F("CP437 Character Map"));           //   of this message
+  }
+
+  const int16_t inset_x = (cell_size - 5) / 2; // To center each character within cell,
+  const int16_t inset_y = (cell_size - 8) / 2; // compute X & Y offset from corner.
+
+  for (uint8_t row=0; row<16; row++) { // 16 down...
+    // Draw row and columm headings as hexadecimal single digits. To get the
+    // hex value for a specific character, combine the left & top labels,
+    // e.g. Pi symbol is row E, column 3, thus: display.print((char)0xE3);
+    display.setCursor(x + (row + 1) * cell_size + inset_x, y + inset_y);
+    display.print(row, HEX); // This actually draws column labels
+    display.setCursor(x + inset_x, y + (row + 1) * cell_size + inset_y);
+    display.print(row, HEX); // and THIS is the row labels
+    for (uint8_t col=0; col<16; col++) { // 16 across...
+      if ((row + col) & 1) { // Fill alternating cells w/gray
+        display.fillRect(x + (col + 1) * cell_size, y + (row + 1) * cell_size,
+                         cell_size, cell_size, 0x630C);
+      }
+      // drawChar() bypasses usual cursor positioning to go direct to an X/Y
+      // location. If foreground & background match, it's drawn transparent.
+      display.drawChar(x + (col + 1) * cell_size + inset_x,
+                       y + (row + 1) * cell_size + inset_y, row * 16 + col,
+                       0xFFFF, 0xFFFF, 1);
+    }
+  }
+
+  delay(PAUSE * 2);
+} // END CHAR MAP EXAMPLE
+
+void show_custom_text() {
+  // Show use of custom fonts, plus how to do center or right alignment
+  // using some additional functions provided earlier.
+
+  display.fillScreen(0);
+  display.setFont(&FreeSansBold18pt7b);
+  display.setTextSize(1);
+  display.setTextWrap(false); // Allow text off edges
+
+  // Get "M height" of custom font and move initial base line there:
+  uint16_t w, h;
+  int16_t  x, y;
+  display.getTextBounds("M", 0, 0, &x, &y, &w, &h);
+  // On rounded 240x280 display in tall orientation, "Custom Font" gets
+  // clipped by top corners. Scoot text down a few pixels in that one case.
+  if (CORNER_RADIUS && (display.height() == 280)) h += 20;
+  display.setCursor(display.width() / 2, h);
+
+  if (display.width() >= 200) {
+    print_aligned(display, F("Custom Font"), GFX_ALIGN_CENTER);
+    display.setCursor(0, display.getCursorY() + 10);
+    print_aligned(display, F("Align Left"), GFX_ALIGN_LEFT);
+    display.setCursor(display.width() / 2, display.getCursorY());
+    print_aligned(display, F("Centered"), GFX_ALIGN_CENTER);
+    // Small rounded screen, when oriented the wide way, "Right" gets
+    // clipped by bottom right corner. Scoot left to compensate. 
+    int16_t x_offset = (CORNER_RADIUS && (display.height() < 200)) ? 15 : 0;
+    display.setCursor(display.width() - x_offset, display.getCursorY());
+    print_aligned(display, F("Align Right"), GFX_ALIGN_RIGHT);
+  } else {
+    // On narrow screens, use abbreviated messages
+    print_aligned(display, F("Font &"), GFX_ALIGN_CENTER);
+    print_aligned(display, F("Align"), GFX_ALIGN_CENTER);
+    display.setCursor(0, display.getCursorY() + 10);
+    print_aligned(display, F("Left"), GFX_ALIGN_LEFT);
+    display.setCursor(display.width() / 2, display.getCursorY());
+    print_aligned(display, F("Center"), GFX_ALIGN_CENTER);
+    display.setCursor(display.width(), display.getCursorY());
+    print_aligned(display, F("Right"), GFX_ALIGN_RIGHT);
+  }
+
+  delay(PAUSE);
+} // END CUSTOM FONT EXAMPLE
+
+// BITMAP EXAMPLE ----------------------------------------------------------
+
+// This section demonstrates:
+// - Embedding a small bitmap in the code (flash memory).
+// - Drawing that bitmap in various colors, and transparently (only '1' bits
+//   are drawn; '0' bits are skipped, leaving screen contents in place).
+// - Use of the color565() function to decimate 24-bit RGB to 16 bits.
+
+#define HEX_WIDTH  16 // Bitmap width in pixels
+#define HEX_HEIGHT 16 // Bitmap height in pixels
+// Bitmap data. PROGMEM ensures it's in flash memory (not RAM). And while
+// it would be valid to leave the brackets empty here (i.e. hex_bitmap[]),
+// having dimensions with a little math makes the compiler verify the
+// correct number of bytes are present in the list.
+PROGMEM const uint8_t hex_bitmap[(HEX_WIDTH + 7) / 8 * HEX_HEIGHT] = {
+  0b00000001, 0b10000000,
+  0b00000111, 0b11100000,
+  0b00011111, 0b11111000,
+  0b01111111, 0b11111110,
+  0b01111111, 0b11111110,
+  0b01111111, 0b11111110,
+  0b01111111, 0b11111110,
+  0b01111111, 0b11111110,
+  0b01111111, 0b11111110,
+  0b01111111, 0b11111110,
+  0b01111111, 0b11111110,
+  0b01111111, 0b11111110,
+  0b01111111, 0b11111110,
+  0b00011111, 0b11111000,
+  0b00000111, 0b11100000,
+  0b00000001, 0b10000000,
+};
+#define Y_SPACING (HEX_HEIGHT - 2) // Used by code below for positioning
+
+void show_bitmap() {
+  display.fillScreen(0);
+
+  // Not screen center, but UL coordinates of center hexagon bitmap
+  const int16_t center_x = (display.width() - HEX_WIDTH) / 2;
+  const int16_t center_y = (display.height() - HEX_HEIGHT) / 2;
+  const uint8_t steps = min((display.height() - HEX_HEIGHT) / Y_SPACING,
+                            display.width() / HEX_WIDTH - 1) / 2;
+
+  display.drawBitmap(center_x, center_y, hex_bitmap, HEX_WIDTH, HEX_HEIGHT,
+                     0xFFFF); // Draw center hexagon in white
+
+  // Tile the hexagon bitmap repeatedly in a range of hues. Don't mind the
+  // bit of repetition in the math, the optimizer easily picks this up.
+  // Also, if math looks odd, keep in mind "PEMDAS" operator precedence;
+  // multiplication and division occur before addition and subtraction.
+  for (uint8_t a=0; a<=steps; a++) {
+    for (uint8_t b=1; b<=steps; b++) {
+      display.drawBitmap( // Right section centered red: a = green, b = blue
+        center_x + (a + b) * HEX_WIDTH / 2,
+        center_y + (a - b) * Y_SPACING,
+        hex_bitmap, HEX_WIDTH, HEX_HEIGHT,
+        display.color565(255, 255 - 255 * a / steps, 255 - 255 * b / steps));
+      display.drawBitmap( // UL section centered green: a = blue, b = red
+        center_x - b * HEX_WIDTH + a * HEX_WIDTH / 2,
+        center_y - a * Y_SPACING,
+        hex_bitmap, HEX_WIDTH, HEX_HEIGHT,
+        display.color565(255 - 255 * b / steps, 255, 255 - 255 * a / steps));
+      display.drawBitmap( // LL section centered blue: a = red, b = green
+        center_x - a * HEX_WIDTH + b * HEX_WIDTH / 2,
+        center_y + b * Y_SPACING,
+        hex_bitmap, HEX_WIDTH, HEX_HEIGHT,
+        display.color565(255 - 255 * a / steps, 255 - 255 * b / steps, 255));
+    }
+  }
+
+  delay(PAUSE);
+} // END BITMAP EXAMPLE
+
+// CANVAS EXAMPLE ----------------------------------------------------------
+
+// This section demonstrates:
+// - How to refresh changing values onscreen without erase/redraw flicker.
+// - Using an offscreen canvas. It's similar to a bitmap above, but rather
+//   than a fixed pattern in flash memory, it's drawable like the screen.
+// - More tips on text alignment, and adapting to different screen sizes.
+
+#define PADDING 6 // Pixels between axis label and value
+
+void show_canvas() {
+  // For this example, let's suppose we want to display live readings from a
+  // sensor such as a three-axis accelerometer, something like:
+  //   X: (number)
+  //   Y: (number)
+  //   Z: (number)
+  // To look extra classy, we want a custom font, and the labels for each
+  // axis are right-aligned so the ':' characters line up...
+
+  display.setFont(&FreeSansBold18pt7b); // Use a custom font
+  display.setTextSize(1);               // and reset to 1:1 scale
+
+  char          *label[] = { "X:", "Y:", "Z:" };       // Labels for each axis
+  const uint16_t color[] = { 0xF800, 0x07E0, 0x001F }; // Colors for each value
+
+  // To get the labels right-aligned, one option would be simple trial and
+  // error to find a column that looks good and doesn't clip anything off.
+  // Let's do this dynamically though, so it adapts to any font or labels!
+  // Start by finding the widest of the label strings:
+  uint16_t w, h, max_w = 0;
+  int16_t  x, y;
+  for (uint8_t i=0; i<3; i++) { // For each label...
+    display.getTextBounds(label[i], 0, 0, &x, &y, &w, &h);
+    if (w > max_w) max_w = w; // Keep track of widest label
+  }
+
+  // Rounded corners throwing us a curve again. If needed, scoot everything
+  // to the right a bit on wide displays, down a bit on tall ones.
+  int16_t y_offset = 0;
+  if (display.width() > display.height()) max_w += CORNER_RADIUS;
+  else                                    y_offset = CORNER_RADIUS;
+
+  // Now we have max_w for right-aligning the labels. Before we draw them
+  // though...in order to perform flicker-free updates, the numbers we show
+  // will be rendered in either a GFXcanvas1 or GFXcanvas16 object; a 1-bit
+  // or 16-bit offscreen bitmap, RAM permitting. The correct size for this
+  // canvas could also be trial-and-errored, but again let's make this adapt
+  // automatically. The width of the canvas will span from max_w (plus a few
+  // pixels for padding) to the right edge. But the height? Looking at an
+  // uppercase 'M' can work in many situations, but some fonts have ascenders
+  // and descenders on digits, and in some locales a comma (extending below
+  // the baseline) is the decimal separator. Feed ALL the numeric chars into
+  // getTextBounds() for a cumulative height:
+  display.setTextWrap(false); // Keep on one line
+  display.getTextBounds(F("0123456789.,-"), 0, 0, &x, &y, &w, &h);
+
+  // Now declare a GFXcanvas16 object based on the computed width & height:
+  GFXcanvas16 canvas16(display.width() - max_w - PADDING, h);
+
+  // Small devices (e.g. ATmega328p) will almost certainly lack enough RAM
+  // for the canvas. Check if canvas buffer exists. If not, fall back on
+  // using a 1-bit (rather than 16-bit) canvas. Much more RAM friendly, but
+  // not as fast to draw. If a project doesn't require super interactive
+  // updates, consider just going straight for the more compact Canvas1.
+  if (canvas16.getBuffer()) {
+    // If here, 16-bit canvas allocated successfully! Point of interest,
+    // only one canvas is needed for this example, we can reuse it for all
+    // three numbers because the regions are the same size.
+
+    // display and canvas are independent drawable objects; must explicitly
+    // set the same custom font to use on the canvas now:
+    canvas16.setFont(&FreeSansBold18pt7b);
+
+    // Clear display and print labels. Once drawn, these remain untouched.
+    display.fillScreen(0);
+    display.setCursor(max_w, -y + y_offset); // Set baseline for first row
+    for (uint8_t i=0; i<3; i++) print_aligned(display, label[i], GFX_ALIGN_RIGHT);
+
+    // Last part now is to print numbers on the canvas and copy the canvas to
+    // the display, repeating for several seconds...
+    uint32_t elapsed, startTime = millis();
+    while ((elapsed = (millis() - startTime)) <= PAUSE * 2) {
+      for (uint8_t i=0; i<3; i++) {  // For each label...
+        canvas16.fillScreen(0);    // fillScreen() in this case clears canvas
+        canvas16.setCursor(0, -y); // Reset baseline for custom font
+        canvas16.setTextColor(color[i]);
+        // These aren't real accelerometer readings, just cool-looking numbers.
+        // Notice we print to the canvas, NOT the display:
+        canvas16.print(sin(elapsed / 200.0 + (float)i * M_PI * 2.0 / 3.0), 5);
+        // And HERE is the secret sauce to flicker-free updates. Canvas details
+        // can be passed to the drawRGBBitmap() function, which fully overwrites
+        // prior screen contents in that area. yAdvance is font line spacing.
+        display.drawRGBBitmap(max_w + PADDING, i * FreeSansBold18pt7b.yAdvance +
+                              y_offset, canvas16.getBuffer(), canvas16.width(),
+                              canvas16.height());
+      }
+    }
+  } else {
+    // Insufficient RAM for Canvas16. Try declaring a 1-bit canvas instead...
+    GFXcanvas1 canvas1(display.width() - max_w - PADDING, h);
+    // If even this smaller object fails, can't proceed, cancel this example.
+    if (!canvas1.getBuffer()) return;
+
+    // Remainder here is nearly identical to the code above, simply using a
+    // different canvas type. It's stripped of most comments for brevity.
+    canvas1.setFont(&FreeSansBold18pt7b);
+    display.fillScreen(0);
+    display.setCursor(max_w, -y + y_offset);
+    for (uint8_t i=0; i<3; i++) print_aligned(display, label[i], GFX_ALIGN_RIGHT);
+    uint32_t elapsed, startTime = millis();
+    while ((elapsed = (millis() - startTime)) <= PAUSE * 2) {
+      for (uint8_t i=0; i<3; i++) {
+        canvas1.fillScreen(0);
+        canvas1.setCursor(0, -y);
+        canvas1.print(sin(elapsed / 200.0 + (float)i * M_PI * 2.0 / 3.0), 5);
+        // Here's the secret sauce to flicker-free updates with GFXcanvas1.
+        // Canvas details can be passed to the drawBitmap() function, and by
+        // specifying both a foreground AND BACKGROUND color (0), this will fully
+        // overwrite/erase prior screen contents in that area (vs transparent).
+        display.drawBitmap(max_w + PADDING, i * FreeSansBold18pt7b.yAdvance +
+                           y_offset, canvas1.getBuffer(), canvas1.width(),
+                           canvas1.height(), color[i], 0);
+      }
+    }
+  }
+
+  // Because canvas object was declared locally to this function, it's freed
+  // automatically when the function returns; no explicit delete needed.
+} // END CANVAS EXAMPLE

examples/1bit_double_buffer/1bit_double_buffer.ino

@@ -0,0 +1,58 @@
+// Double-buffered 1-bit Adafruit_GFX-compatible framebuffer for PicoDVI.
+// Animates without redraw flicker. Requires Adafruit_GFX >= 1.11.5
+
+#include <PicoDVI.h>
+
+// Here's how a 640x480 1-bit (black, white) framebuffer is declared.
+// Second argument ('true' here) enables double-buffering for flicker-free
+// animation. Third argument is a hardware configuration -- examples are
+// written for Adafruit Feather RP2040 DVI, but that's easily switched out
+// for boards like the Pimoroni Pico DV (use 'pimoroni_demo_hdmi_cfg') or
+// Pico DVI Sock ('pico_sock_cfg').
+DVIGFX1 display(DVI_RES_640x480p60, true, adafruit_feather_dvi_cfg);
+
+// An 800x480 mode is possible but pushes overclocking even higher than
+// 640x480 mode. SOME BOARDS MIGHT SIMPLY NOT BE COMPATIBLE WITH THIS.
+// May require selecting QSPI div4 clock (Tools menu) to slow down flash
+// accesses, may require further over-volting the CPU to 1.25 or 1.3 V.
+//DVIGFX1 display(DVI_RES_800x480p60, true, adafruit_feather_dvi_cfg);
+
+#define N_BALLS 100 // Number of bouncy balls to draw
+struct {
+  int16_t pos[2]; // Ball position (X,Y)
+  int8_t  vel[2]; // Ball velocity (X,Y)
+} ball[N_BALLS];
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+
+  // Randomize initial ball positions and velocities
+  for (int i=0; i<N_BALLS; i++) {
+    ball[i].pos[0] = 10 + random(display.width() - 20);
+    ball[i].pos[1] = 10 + random(display.height() - 20);
+    do {
+      ball[i].vel[0] = 4 - random(9);
+      ball[i].vel[1] = 4 - random(9);
+    } while ((ball[i].vel[0] == 0) && (ball[i].vel[1] == 0));
+  }
+}
+
+void loop() {
+  display.fillScreen(0); // Clear back framebuffer...
+  // And draw bouncy balls (circles) there
+  for (int i=0; i<N_BALLS; i++) {
+    display.drawCircle(ball[i].pos[0], ball[i].pos[1], 40, 1);
+    // After drawing each one, update positions, bounce off edges.
+    ball[i].pos[0] += ball[i].vel[0];
+    if ((ball[i].pos[0] <= 0) || (ball[i].pos[0] >= display.width())) ball[i].vel[0] *= -1;
+    ball[i].pos[1] += ball[i].vel[1];
+    if ((ball[i].pos[1] <= 0) || (ball[i].pos[1] >= display.height())) ball[i].vel[1] *= -1;
+  }
+
+  // Swap front/back buffers, do not duplicate current screen state to next frame,
+  // we'll draw it new from scratch each time.
+  display.swap();
+}

examples/1bit_single_buffer/1bit_single_buffer.ino

@@ -0,0 +1,31 @@
+// Simple 1-bit Adafruit_GFX-compatible framebuffer for PicoDVI.
+
+#include <PicoDVI.h>
+
+// Here's how a 640x480 1-bit (black, white) framebuffer is declared.
+// Second argument ('false' here) means NO double-buffering; all drawing
+// operations are shown as they occur. Third argument is a hardware
+// configuration -- examples are written for Adafruit Feather RP2040 DVI,
+// but that's easily switched out for boards like the Pimoroni Pico DV
+// (use 'pimoroni_demo_hdmi_cfg') or Pico DVI Sock ('pico_sock_cfg').
+DVIGFX1 display(DVI_RES_640x480p60, false, adafruit_feather_dvi_cfg);
+
+// An 800x480 mode is possible but pushes overclocking even higher than
+// 640x480 mode. SOME BOARDS MIGHT SIMPLY NOT BE COMPATIBLE WITH THIS.
+// May require selecting QSPI div4 clock (Tools menu) to slow down flash
+// accesses, may require further over-volting the CPU to 1.25 or 1.3 V.
+//DVIGFX1 display(DVI_RES_800x480p60, false, adafruit_feather_dvi_cfg);
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+}
+
+void loop() {
+  // Draw random lines
+  display.drawLine(random(display.width()), random(display.height()), // Start X,Y
+                   random(display.width()), random(display.height()), // End X,Y
+                   random(2)); // Color (0 or 1)
+}

examples/1bit_text/1bit_text.ino

@@ -0,0 +1,35 @@
+// 1-bit (black, white) text mode for PicoDVI.
+
+#include <PicoDVI.h>
+
+// Here's how an 80x30 character display is declared. First argument,
+// resolution, is full display pixel count...character cells are 8x8 pixels,
+// yielding the 80x30 result. 640x240 uses "tall" pixels, the result of all
+// this is very reminiscent of IBM VGA mode. Second argument is a hardware
+// configuration -- examples are written for Adafruit Feather RP2040 DVI,
+// but that's easily switched out for boards like the Pimoroni Pico DV
+// (use 'pimoroni_demo_hdmi_cfg') or Pico DVI Sock ('pico_sock_cfg').
+DVItext1 display(DVI_RES_640x240p60, adafruit_feather_dvi_cfg);
+
+// Wider and taller modes are possible. Wider pushes overclocking even
+// higher than 640x480 mode. SOME BOARDS MIGHT SIMPLY NOT BE COMPATIBLE
+// WITH THIS. May require selecting QSPI div4 clock (Tools menu) to slow
+// down flash accesses, may require over-volting the CPU to 1.25 or 1.3 V.
+// Here's how a 100x60 char display might be declared:
+//DVItext1 display(DVI_RES_800x480p60, adafruit_feather_dvi_cfg);
+
+// A reduced refresh rate display doesn't as aggressive an over-clock
+// This timing is verified to work on https://www.adafruit.com/product/2232
+//DVItext1 display(DVI_RES_800x240p30, adafruit_feather_dvi_cfg);
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+}
+
+void loop() {
+  display.print("Hello World!  ");
+  delay(50);
+}

examples/8bit_double_buffer/8bit_double_buffer.ino

@@ -0,0 +1,61 @@
+// Double-buffered 8-bit Adafruit_GFX-compatible framebuffer for PicoDVI.
+// Animates without redraw flicker. Requires Adafruit_GFX >= 1.11.4
+
+#include <PicoDVI.h>
+
+// Here's how a 320x240 8-bit (color-paletted) framebuffer is declared.
+// Second argument ('true' here) enables double-buffering for flicker-free
+// animation. Third argument is a hardware configuration -- examples are
+// written for Adafruit Feather RP2040 DVI, but that's easily switched out
+// for boards like the Pimoroni Pico DV (use 'pimoroni_demo_hdmi_cfg') or
+// Pico DVI Sock ('pico_sock_cfg').
+DVIGFX8 display(DVI_RES_320x240p60, true, adafruit_feather_dvi_cfg);
+
+// A 400x240 mode is possible but pushes overclocking even higher than
+// 320x240 mode. SOME BOARDS MIGHT SIMPLY NOT BE COMPATIBLE WITH THIS.
+// May require selecting QSPI div4 clock (Tools menu) to slow down flash
+// accesses, may require further over-volting the CPU to 1.25 or 1.3 V.
+//DVIGFX8 display(DVI_RES_400x240p60, true, adafruit_feather_dvi_cfg);
+
+
+#define N_BALLS 100 // Number of bouncy balls to draw, 1-254 (not 255)
+struct {
+  int16_t pos[2]; // Ball position (X,Y)
+  int8_t  vel[2]; // Ball velocity (X,Y)
+} ball[N_BALLS];
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+
+  // Randomize initial ball positions, velocities and colors
+  for (int i=0; i<N_BALLS; i++) {
+    display.setColor(i+1, 64 + random(192), 64 + random(192), 64 + random(192));
+    ball[i].pos[0] = 10 + random(display.width() - 20);
+    ball[i].pos[1] = 10 + random(display.height() - 20);
+    do {
+      ball[i].vel[0] = 2 - random(5);
+      ball[i].vel[1] = 2 - random(5);
+    } while ((ball[i].vel[0] == 0) && (ball[i].vel[1] == 0));
+  }
+  display.setColor(255, 0xFFFF); // Last palette entry = white
+  display.swap(false, true); // Duplicate same palette into front & back buffers
+}
+
+void loop() {
+  // Clear back framebuffer and draw balls (circles) there.
+  display.fillScreen(0);
+  for (int i=0; i<N_BALLS; i++) {
+    display.fillCircle(ball[i].pos[0], ball[i].pos[1], 20, i + 1);
+    // After drawing each one, update positions, bounce off edges.
+    ball[i].pos[0] += ball[i].vel[0];
+    if ((ball[i].pos[0] <= 0) || (ball[i].pos[0] >= display.width())) ball[i].vel[0] *= -1;
+    ball[i].pos[1] += ball[i].vel[1];
+    if ((ball[i].pos[1] <= 0) || (ball[i].pos[1] >= display.height())) ball[i].vel[1] *= -1;
+  }
+  // Swap front/back buffers, do not duplicate current screen state to next frame,
+  // we'll draw it new from scratch each time.
+  display.swap();
+}

examples/8bit_single_buffer/8bit_single_buffer.ino

@@ -0,0 +1,33 @@
+// 8-bit Adafruit_GFX-compatible framebuffer for PicoDVI.
+
+#include <PicoDVI.h>
+
+// Here's how a 320x240 8-bit (color-paletted) framebuffer is declared.
+// Second argument ('false' here) means NO double-buffering; all drawing
+// operations are shown as they occur. Third argument is a hardware
+// configuration -- examples are written for Adafruit Feather RP2040 DVI,
+// but that's easily switched out for boards like the Pimoroni Pico DV
+// (use 'pimoroni_demo_hdmi_cfg') or Pico DVI Sock ('pico_sock_cfg').
+DVIGFX8 display(DVI_RES_320x240p60, false, adafruit_feather_dvi_cfg);
+
+// A 400x240 mode is possible but pushes overclocking even higher than
+// 320x240 mode. SOME BOARDS MIGHT SIMPLY NOT BE COMPATIBLE WITH THIS.
+// May require selecting QSPI div4 clock (Tools menu) to slow down flash
+// accesses, may require further over-volting the CPU to 1.25 or 1.3 V.
+//DVIGFX8 display(DVI_RES_400x240p60, false, adafruit_feather_dvi_cfg);
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+
+  // Randomize color palette. First entry is left black, last is set white.
+  for (int i=1; i<255; i++) display.setColor(i, random(65536));
+  display.setColor(255, 0xFFFF);
+}
+
+void loop() {
+  // Draw random lines
+  display.drawLine(random(display.width()), random(display.height()), random(display.width()), random(display.height()), random(256));
+}

examples/Readme.md

@@ -0,0 +1,5 @@
+Arduino Examples for PicoDVI Library
+====================================
+
+This directory does not exist in the original PicoDVI repository.
+It contains examples compatible with the Arduino IDE.

examples/file_access/file_access.ino

@@ -0,0 +1,83 @@
+/*
+This example shows how to use PicoDVI and a board's flash filesystem (i.e.
+CIRCUITPY drive) simultaneously. This can be useful for loading graphics
+files, saving game state in emulators, etc. To keep this simple, it's just
+a mash-up of the existing PicoDVI '1bit_text' and Adafruit_CPFS 'simple'
+examples, but the same principles certainly apply to other modes.
+
+To start, you first need to temporarily install CircuitPython on the board
+to initialize the flash filesystem, which can then be loaded up with the
+files you need. https://circuitpython.org/downloads
+With the filesystem prepared, one can then work with the Arduino code...
+
+SUPER IMPORTANT: you MUST have current versions of several libraries and
+the Earle Philhower arduino-pico core installed. Failure to have all the
+right pieces will wipe out any data stored on the drive!
+
+arduino-pico (via Arduino board manager)  3.3.0 or later
+Adafruit_CPFS (via Library manager)       1.1.0 or later
+Adafruit_SPIFlash (")                     4.2.0 or later
+
+It is wise and STRONGLY RECOMMENDED to keep a backup of any data you install
+on the board. These libraries combined are asking a LOT of the RP2040 chip,
+and despite best efforts there's still the occasional hiccup that can wipe
+the filesystem, making you start over with the CircuitPython install and
+drive setup. See notes below about perhaps adding a boot switch to USB-mount
+CIRCUITPY only when needed; it's more stable if left unmounted.
+*/
+
+#include <PicoDVI.h>       // For DVI video out
+#include <Adafruit_CPFS.h> // For accessing the CIRCUITPY drive
+
+FatVolume *fs = NULL; // CIRCUITPY flash filesystem, as a FAT pointer
+
+// This example uses 80x30 monochrome text mode. See other PicoDVI examples
+// for color, bitmapped graphics, widescreen, alternate boards, etc.
+DVItext1 display(DVI_RES_640x240p60, adafruit_feather_dvi_cfg);
+
+void setup() { // Runs once on startup
+  pinMode(LED_BUILTIN, OUTPUT);
+
+  // Start the CIRCUITPY flash filesystem. SUPER IMPORTANT: NOTICE THE
+  // EXTRA PARAMETERS HERE. This is REQUIRED when using PicoDVI and
+  // Adafruit_CPFS together.
+  fs = Adafruit_CPFS::begin(true, -1, NULL, false);
+  // The initial 'true' argument tells CPFS to make the flash filesystem
+  // available to a USB-connected host computer. Passing 'false' makes it
+  // only available to the Arduino sketch. Given the tenuous stability of
+  // handling so much at once (DVI, flash, USB), one might want to add a
+  // boot-time button or switch to select whether CIRCUITPY is mounted on
+  // host, or is just using USB for power.
+  // Next two arguments are ignored on RP2040; they're specifically for
+  // some 'Haxpress' dev boards with the CPFS library. Last argument should
+  // ALWAYS be set 'false' on RP2040, or there will be...trouble.
+
+  if (!display.begin()) { // Start DVI, slow blink LED if insufficient RAM
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+
+  if (fs == NULL) { // If CIRCUITPY filesystem is missing or malformed...
+    // Show error message & fast blink LED to indicate problem. Full stop.
+    display.println("Can't access board's CIRCUITPY drive.");
+    display.println("Has CircuitPython been previously installed?");
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 250) & 1);
+  } // else valid CIRCUITPY drive, proceed...
+
+  // As in Adafruit_CPFS 'simple' example, allow USB events to settle...
+  delay(2500);
+  Adafruit_CPFS::change_ack();
+
+  // Then access files and directories using any SdFat calls (open(), etc.)
+
+  // Because fs is a pointer, we use "->" indirection rather than "." access.
+  // display pointer is cast to print_t so ls() treats it just like Serial.
+  fs->ls((print_t *)&display, LS_R | LS_SIZE); // List initial drive contents
+}
+
+void loop() {
+  if (Adafruit_CPFS::changed()) { // Anything changed on CIRCUITPY drive?
+    Adafruit_CPFS::change_ack();  // Got it, thanks.
+    display.println("CIRCUITPY drive contents changed.");
+    fs->ls((print_t *)&display, LS_R | LS_SIZE);  // List updated drive contents
+  }
+}

examples/screensavers/aquarium/aquarium.ino

@@ -0,0 +1,153 @@
+// "Aquarium" example for PicoDVI library. If just starting out,
+// see the 8bit_double_buffer which explains the PicoDVI groundwork.
+// Comments in THIS file are mostly distinct & new concepts.
+// The flying toasters example also goes into more detail.
+
+// IF NO OUTPUT OR RED FLICKER SCANLINES: try Tools->Optimize->(-O3)
+
+#include <PicoDVI.h>
+#include "sprites.h" // Graphics data
+
+DVIGFX8 display(DVI_RES_320x240p60, true, adafruit_feather_dvi_cfg);
+
+// See notes in 8bit_double_buffer regarding 400x240 mode.
+//DVIGFX8 display(DVI_RES_400x240p60, true, adafruit_feather_dvi_cfg);
+// Also requires -O3 setting.
+
+// This structure holds pointers to sprite graphics and masks in sprites.h.
+const struct {
+  const uint8_t *sprite[2][2]; // L/R directions and A/B frames
+  const uint8_t *mask[2][2];   // Same
+} spritedata[] = {
+  // There are FOUR sprites/masks for each fish (and kelp):
+  // two left-facing, two right-facing, and A/B frames for each.
+  { sprite0LA , sprite0LB , sprite0RA , sprite0RB , mask0LA , mask0LB , mask0RA , mask0RB  },
+  { sprite1LA , sprite1LB , sprite1RA , sprite1RB , mask1LA , mask1LB , mask1RA , mask1RB  },
+  { sprite2LA , sprite2LB , sprite2RA , sprite2RB , mask2LA , mask2LB , mask2RA , mask2RB  },
+  { sprite3LA , sprite3LB , sprite3RA , sprite3RB,  mask3LA , mask3LB , mask3RA , mask3RB  },
+  { sprite4LA , sprite4LB , sprite4RA , sprite4RB , mask4LA , mask4LB , mask4RA , mask4RB  },
+  { sprite5LA , sprite5LB , sprite5RA , sprite5RB , mask5LA , mask5LB , mask5RA , mask5RB  },
+  { sprite6LA , sprite6LB , sprite6RA , sprite6RB , mask6LA , mask6LB , mask6RA , mask6RB  },
+  { sprite7LA , sprite7LB , sprite7RA , sprite7RB , mask7LA , mask7LB , mask7RA , mask7RB  },
+  { sprite8LA , sprite8LB , sprite8RA , sprite8RB , mask8LA , mask8LB , mask8RA , mask8RB  },
+  { sprite9LA , sprite9LB , sprite9RA , sprite9RB , mask9LA , mask9LB , mask9RA , mask9RB  },
+  { sprite10LA, sprite10LB, sprite10RA, sprite10RB, mask10LA, mask10LB, mask10RA, mask10RB },
+  { sprite11LA, sprite11LB, sprite11RA, sprite11RB, mask11LA, mask11LB, mask11RA, mask11RB },
+  { sprite12LA, sprite12LB, sprite12RA, sprite12RB, mask12LA, mask12LB, mask12RA, mask12RB },
+  { sprite13LA, sprite13LB, sprite13RA, sprite13RB, mask13LA, mask13LB, mask13RA, mask13RB },
+  { sprite14LA, sprite14LB, sprite14RA, sprite14RB, mask14LA, mask14LB, mask14RA, mask14RB },
+  { sprite15LA, sprite15LB, sprite15RA, sprite15RB, mask15LA, mask15LB, mask15RA, mask15RB },
+  { sprite16LA, sprite16LB, sprite16RA, sprite16RB, mask16LA, mask16LB, mask16RA, mask16RB },
+  { sprite17LA, sprite17LB, sprite17RA, sprite17RB, mask17LA, mask17LB, mask17RA, mask17RB },
+  // Bubbles are a special case. No right/left versions, but A/B frames.
+  // To use same struct (not a special case), just duplicate each 2X.
+  { sprite18A , sprite18B , sprite18A , sprite18B , mask18A , mask18B , mask18A , mask18B  },
+};
+
+#define N_SPRITES 12 // MUST be >= 6
+
+// This structure contains positions and other data for the sprites
+// in motion (notice it's not "const", because contents change).
+struct {
+  int16_t pos[2]; // sprite position (X,Y) * 16
+  int8_t  speed;  // sprite speed (-16 to -8 or +8 to +16)
+  uint8_t index;  // which index (in spritedata) to use
+  uint8_t offset; // Timer offset to de-sync each sprite's animation
+} sprite[N_SPRITES];
+
+// Initialize one sprite (index passed to function) to a random offscreen
+// position, also randomizing speed and sprite (fish) type.
+void randomsprite(uint8_t i) {
+  // To move the sprites at slightly different speeds, coordinates are
+  // stored in 1/16 pixel units. When drawing, the stored values get
+  // divided by 16 to yield final pixel coordinates.
+  sprite[i].speed = random(8, 17); // 1/2 to 1 pixel per frame
+  if (random(2)) {                 // 50/50 random chance...
+    sprite[i].speed *= -1;         // Fish moves right-to-left
+    sprite[i].pos[0] = (display.width() + random(64)) * 16; // Start off right edge
+  } else {                         // Fish moves left-to-right
+    sprite[i].pos[0] = random(64, 128) * -16; // Start off left edge
+  }
+  // WHEEL. OF. FISH. -2 here is to ignore last 2 sprites (kelp, bubbles)
+  sprite[i].index = random(sizeof spritedata / sizeof spritedata[0] - 2);
+  if (sprite[i].index == 8) { // Sprite #8 is crab, keep close to ground
+    sprite[i].pos[1] = random(display.height() - 96, display.height() - 64) * 16;
+  } else { // Is a fish, upper part of screen
+    sprite[i].pos[1] = random(display.height() - 96) * 16;
+  }
+  sprite[i].offset = random(256); // De-synchronize sprite animation
+}
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+
+  // Initialize color palette from table in sprites.h. Rather than
+  // calling setColor() for each one, we can just dump it directly...
+  memcpy(display.getPalette(), palette, sizeof palette);
+  display.swap(false, true); // Duplicate same palette to front & back buffers
+
+  // Randomize initial sprite states
+  randomSeed(analogRead(A0)); // Seed randomness from unused analog in
+  int range = display.width() + 64;
+  for (int i=0; i<3; i++) {   // FIRST THREE sprites...
+    sprite[i].index = 17;     // Are always kelp
+    sprite[i].speed = random(2) ? 1 : -1; // 50/50 left/right flip
+    sprite[i].pos[0] = (random(range * i / 3, range * (i + 1) / 3 - 64) - 32) * 16;
+    sprite[i].pos[1] = random(display.height() - 120, display.height() - 100) * 16;
+    sprite[i].offset = random(256);
+  }
+  for (int i=3; i<6; i++) { // NEXT THREE sprites...
+    sprite[i].index = 18;   // Are always bubbles
+    sprite[i].speed = 0;
+    sprite[i].pos[0] = display.width() * 16; // Start them all offscreen
+    sprite[i].pos[1] = random(display.height()) * 8;
+    sprite[i].offset = random(256);
+  }
+  for (int i=6; i<N_SPRITES; i++) randomsprite(i); // Rest are fish
+}
+
+uint8_t frame = 0; // Counter for animation
+
+void loop() { // Runs once every frame
+  display.fillScreen(0);                       // Clear back framebuffer,
+  for (int x=0; x<display.width(); x += 192) { // Tile background sprite
+    // Although DVIGFX8 is a COLOR display type, we leverage GFX's
+    // drawGrayscaleBitmap() function to draw the sprites...it saves us
+    // writing a ton of code this way.
+    display.drawGrayscaleBitmap(x, display.height() - 64, gravel, 192, 64);
+  }
+
+  for (int i=0; i<N_SPRITES; i++) { // and then the rest of the sprites...
+    uint8_t dir = sprite[i].speed > 0;                  // Left/right
+    uint8_t fr = ((frame + sprite[i].offset) >> 4) & 1; // A/B frame
+    if (sprite[i].speed) { // FISH or KELP; 64x64 sprite
+      display.drawGrayscaleBitmap(sprite[i].pos[0] / 16, sprite[i].pos[1] / 16,
+        spritedata[sprite[i].index].sprite[dir][fr],
+        spritedata[sprite[i].index].mask[dir][fr], 64, 64);
+      if (abs(sprite[i].speed) > 1) {        // Not kelp...
+        sprite[i].pos[0] += sprite[i].speed; // Update position, check if offscreen...
+        if (((sprite[i].speed > 0) && (sprite[i].pos[0] > (display.width() * 16))) ||
+            ((sprite[i].speed < 0) && (sprite[i].pos[0] < -64 * 16)))
+          randomsprite(i); // Replace with a new fish
+      }
+    } else { // Is BUBBLES
+      display.drawGrayscaleBitmap(sprite[i].pos[0] / 16, sprite[i].pos[1] / 16,
+        spritedata[sprite[i].index].sprite[0][fr],
+        spritedata[sprite[i].index].mask[0][fr], 64, 16);
+      sprite[i].pos[1] -= 16;                // Move up by 1 pixel
+      if (sprite[i].pos[1] < -256) {         // Off top of screen?
+        int j = random(6, N_SPRITES);        // Pick a random fish,
+        sprite[i].pos[0] = sprite[j].pos[0]; // and move bubbles there
+        sprite[i].pos[1] = sprite[j].pos[1] + 384;
+      }
+    }
+  }
+
+  // Swap front/back buffers, do not duplicate current screen state
+  // to next frame, we'll draw it new from scratch each time.
+  display.swap();
+  frame++; // Increment animation counter; "rolls over" 0-255 automatically.
+}

examples/screensavers/boing/boing.ino

@@ -0,0 +1,121 @@
+// "Boing" ball example for PicoDVI library, If just starting out,
+// see the 8bit_double_buffer which explains the PicoDVI groundwork.
+// Comments in THIS file are mostly distinct & new concepts.
+// Adapted from PyPortal example in Adafruit_ILI9341 library.
+
+#include <PicoDVI.h>
+#include "graphics.h" // Graphics data
+
+// 4:3 aspect for that Amiga flavour:
+DVIGFX8 display(DVI_RES_320x240p60, true, adafruit_feather_dvi_cfg);
+
+#define YBOTTOM   123  // Ball Y coord at bottom
+#define YBOUNCE  -3.5  // Upward velocity on ball bounce
+
+// Ball coordinates are stored floating-point because screen refresh
+// is so quick, whole-pixel movements are just too fast!
+float ballx     = 20.0, bally     = YBOTTOM, // Current ball position
+      ballvx    =  0.8, ballvy    = YBOUNCE, // Ball velocity
+      ballframe = 3;                         // Ball animation frame #
+int   balloldx  = ballx, balloldy = bally;   // Prior ball position
+
+void setup() {
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+
+  // Set up color palette
+  display.setColor(0, 0xAD75); // #0 = Background color
+  display.setColor(1, 0xA815); // #1 = Grid color
+  display.setColor(2, 0x5285); // #2 = Background in shadow
+  display.setColor(3, 0x600C); // #3 = Grid in shadow
+
+  // Draw initial framebuffer contents (grid, no shadow):
+  display.drawBitmap(0, 0, (uint8_t *)background, 320, 240, 1, 0);
+  display.swap(true, true); // Duplicate same bg & palette into both buffers
+}
+
+void loop() {
+  balloldx = (int16_t)ballx; // Save prior position
+  balloldy = (int16_t)bally;
+  ballx   += ballvx;         // Update position
+  bally   += ballvy;
+  ballvy  += 0.06;           // Update Y velocity
+  if((ballx <= 15) || (ballx >= display.width() - BALLWIDTH))
+    ballvx *= -1;            // Left/right bounce
+  if(bally >= YBOTTOM) {     // Hit ground?
+    bally  = YBOTTOM;        // Clip and
+    ballvy = YBOUNCE;        // bounce up
+  }
+
+  // Determine screen area to update.  This is the bounds of the ball's
+  // prior and current positions, so the old ball is fully erased and new
+  // ball is fully drawn.
+  int16_t minx, miny, maxx, maxy, width, height;
+  // Determine bounds of prior and new positions
+  minx = ballx;
+  if(balloldx < minx)                    minx = balloldx;
+  miny = bally;
+  if(balloldy < miny)                    miny = balloldy;
+  maxx = ballx + BALLWIDTH  - 1;
+  if((balloldx + BALLWIDTH  - 1) > maxx) maxx = balloldx + BALLWIDTH  - 1;
+  maxy = bally + BALLHEIGHT - 1;
+  if((balloldy + BALLHEIGHT - 1) > maxy) maxy = balloldy + BALLHEIGHT - 1;
+
+  width  = maxx - minx + 1;
+  height = maxy - miny + 1;
+
+  // Ball animation frame # is incremented opposite the ball's X velocity
+  ballframe -= ballvx * 0.5;
+  if(ballframe < 0)        ballframe += 14; // Constrain from 0 to 13
+  else if(ballframe >= 14) ballframe -= 14;
+
+  // Set 7 palette entries to white, 7 to red, based on frame number.
+  // This makes the ball spin.
+  for(uint8_t i=0; i<14; i++) {
+    display.setColor(i + 4, ((((int)ballframe + i) % 14) < 7) ? 0xFFFF : 0xF800);
+  }
+
+  // Only the changed rectangle is drawn into the 'renderbuf' array...
+  uint8_t  c, *destPtr;
+  int16_t  bx  = minx - (int)ballx, // X relative to ball bitmap (can be negative)
+           by  = miny - (int)bally, // Y relative to ball bitmap (can be negative)
+           bgx = minx,              // X relative to background bitmap (>= 0)
+           bgy = miny,              // Y relative to background bitmap (>= 0)
+           x, y, bx1, bgx1;         // Loop counters and working vars
+  uint8_t  p;                       // 'packed' value of 2 ball pixels
+  int8_t bufIdx = 0;
+
+  uint8_t *buf = display.getBuffer(); // -> back buffer
+
+  for(y=0; y<height; y++) { // For each row...
+    destPtr = &buf[display.width() * (miny + y) + minx];
+    bx1  = bx;  // Need to keep the original bx and bgx values,
+    bgx1 = bgx; // so copies of them are made here (and changed in loop below)
+    for(x=0; x<width; x++) {
+      if((bx1 >= 0) && (bx1 < BALLWIDTH) &&  // Is current pixel row/column
+         (by  >= 0) && (by  < BALLHEIGHT)) { // inside the ball bitmap area?
+        // Yes, do ball compositing math...
+        p = ball[by][bx1 / 2];                // Get packed value (2 pixels)
+        c = (bx1 & 1) ? (p & 0xF) : (p >> 4); // Unpack high or low nybble
+        if(c == 0) { // Outside ball - just draw grid
+          c = background[bgy][bgx1 / 8] & (0x80 >> (bgx1 & 7)) ? 1 : 0;
+        } else if(c > 1) { // In ball area...
+          c += 2; // Convert to color index >= 4
+        } else { // In shadow area, draw shaded grid...
+          c = background[bgy][bgx1 / 8] & (0x80 >> (bgx1 & 7)) ? 3 : 2;
+        }
+      } else { // Outside ball bitmap, just draw background bitmap...
+        c = background[bgy][bgx1 / 8] & (0x80 >> (bgx1 & 7)) ? 1 : 0;
+      }
+      *destPtr++ = c; // Store pixel color
+      bx1++;  // Increment bitmap position counters (X axis)
+      bgx1++;
+    }
+    by++; // Increment bitmap position counters (Y axis)
+    bgy++;
+  }
+
+  display.swap(true, false); // Show & copy current background buffer to next
+}

examples/screensavers/logobounce/logobounce.ino

@@ -0,0 +1,41 @@
+// "Logo bounce" example for PicoDVI library. If just starting out,
+// see the 1bit_double_buffer which explains the PicoDVI groundwork.
+// Comments in THIS file are mostly distinct & new concepts.
+
+// IF NO OUTPUT OR RED FLICKER SCANLINES: try Tools->Optimize->(-O3)
+
+#include <PicoDVI.h>
+#include "sprite.h" // Graphics data
+
+DVIGFX1 display(DVI_RES_640x480p60, true, adafruit_feather_dvi_cfg);
+
+// See notes in 1bit_double_buffer regarding 800x480 mode.
+//DVIGFX1 display(DVI_RES_800x480p60, true, adafruit_feather_dvi_cfg);
+// May also require -O3 setting.
+
+int x  = 0; // Start logo at
+int y  = 0; // top left corner,
+int vx = 1; // moving right
+int vy = 1; // and down
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+}
+
+void loop() { // Runs once every frame
+  display.fillScreen(0); // Clear back framebuffer, then draw sprite:
+  display.drawBitmap(x, y, sprite, SPRITE_WIDTH, SPRITE_HEIGHT, 1);
+
+  // Swap front/back buffers, do not duplicate current screen state
+  // to next frame, we'll draw it new from scratch each time.
+  display.swap();
+
+  // Update sprite position, bouncing off all 4 sides
+  x += vx; // Horizontal
+  if ((x == 0) || (x == (display.width() - SPRITE_WIDTH))) vx *= -1;
+  y += vy; // Vertical
+  if ((y == 0) || (y == (display.height() - SPRITE_HEIGHT))) vy *= -1;
+}

examples/screensavers/logobounce/sprite.h

@@ -0,0 +1,386 @@
+// "DVI video" logo sprite. Generated from a 1-bit PNG using ImageMagick:
+// magick sprite.png -define h:format=gray -depth 1 sprite.h
+// (then edited a bit for this program)
+
+// The sprite was intentionally planned with dimensions that are both
+// prime numbers, and then starts from the upper-left corner. This should
+// make the bounce SUPER annoying, taking hours or days to ever hit an
+// exact corner again!
+#define SPRITE_WIDTH  239
+#define SPRITE_HEIGHT 149
+
+static const uint8_t sprite[] = {
+  0x00, 0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xE0, 0x00,
+  0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1F, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF0, 0x00,
+  0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x1F, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1F, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF0, 0x00,
+  0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1F, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF8, 0x00,
+  0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00,
+  0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0x00,
+  0x00, 0x00, 0x07, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0x00,
+  0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xFC,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x00,
+  0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF, 0xF8,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80,
+  0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x7F, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xEF,
+  0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF,
+  0xFF, 0x80, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xFF, 0x7F, 0xFF, 0xFF, 0x80,
+  0x00, 0x03, 0xFF, 0xFF, 0xFF, 0xDF, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x1F, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xC0, 0x00, 0x07, 0xFF, 0xFF, 0xFF, 0x9F,
+  0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF,
+  0xFF, 0x80, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xC0,
+  0x00, 0x07, 0xFF, 0xFF, 0xFF, 0x1F, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x07, 0xFF,
+  0xFF, 0xFF, 0xBF, 0xFF, 0xFF, 0xC0, 0x00, 0x0F, 0xFF, 0xFF, 0xFE, 0x1F,
+  0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF,
+  0xFF, 0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF, 0xFF, 0xBF, 0xFF, 0xFF, 0xE0,
+  0x00, 0x1F, 0xFF, 0xFF, 0xFC, 0x3F, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0xFF,
+  0xFF, 0xFF, 0xBF, 0xFF, 0xFF, 0xE0, 0x00, 0x3F, 0xFF, 0xFF, 0xF8, 0x3F,
+  0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF,
+  0xFF, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0x9F, 0xFF, 0xFF, 0xE0,
+  0x00, 0x7F, 0xFF, 0xFF, 0xF8, 0x3F, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x03, 0xFF,
+  0xFF, 0xFF, 0x9F, 0xFF, 0xFF, 0xF0, 0x00, 0xFF, 0xFF, 0xFF, 0xF0, 0x3F,
+  0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF,
+  0xFE, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0x9F, 0xFF, 0xFF, 0xF0,
+  0x00, 0xFF, 0xFF, 0xFF, 0xE0, 0x7F, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00, 0x03, 0xFF,
+  0xFF, 0xFF, 0x8F, 0xFF, 0xFF, 0xF0, 0x01, 0xFF, 0xFF, 0xFF, 0xC0, 0x7F,
+  0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF,
+  0xFE, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0x8F, 0xFF, 0xFF, 0xF8,
+  0x03, 0xFF, 0xFF, 0xFF, 0x80, 0x7F, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00, 0x03, 0xFF,
+  0xFF, 0xFF, 0x8F, 0xFF, 0xFF, 0xF8, 0x07, 0xFF, 0xFF, 0xFF, 0x00, 0x7F,
+  0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF,
+  0xFC, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0x07, 0xFF, 0xFF, 0xF8,
+  0x07, 0xFF, 0xFF, 0xFE, 0x00, 0xFF, 0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x03, 0xFF,
+  0xFF, 0xFF, 0x07, 0xFF, 0xFF, 0xFC, 0x0F, 0xFF, 0xFF, 0xFC, 0x00, 0xFF,
+  0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF,
+  0xFC, 0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF, 0xFF, 0x03, 0xFF, 0xFF, 0xFC,
+  0x1F, 0xFF, 0xFF, 0xF8, 0x00, 0xFF, 0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x07, 0xFF,
+  0xFF, 0xFF, 0x03, 0xFF, 0xFF, 0xFC, 0x3F, 0xFF, 0xFF, 0xF8, 0x00, 0xFF,
+  0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF,
+  0xFC, 0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFE, 0x03, 0xFF, 0xFF, 0xFC,
+  0x3F, 0xFF, 0xFF, 0xF0, 0x00, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00, 0x0F, 0xFF,
+  0xFF, 0xFE, 0x01, 0xFF, 0xFF, 0xFE, 0x7F, 0xFF, 0xFF, 0xE0, 0x01, 0xFF,
+  0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1F, 0xFF, 0xFF,
+  0xF8, 0x00, 0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xFC, 0x01, 0xFF, 0xFF, 0xFE,
+  0xFF, 0xFF, 0xFF, 0xC0, 0x01, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00, 0x3F, 0xFF,
+  0xFF, 0xFC, 0x01, 0xFF, 0xFF, 0xFE, 0xFF, 0xFF, 0xFF, 0x80, 0x01, 0xFF,
+  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1F, 0xFF, 0xFF,
+  0xF8, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF, 0xF8, 0x00, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00, 0xFF, 0xFF,
+  0xFF, 0xF8, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0x00, 0x03, 0xFF,
+  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF,
+  0xF0, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xF0, 0x00, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFC, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x03, 0xFF, 0xFF,
+  0xFF, 0xE0, 0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x03, 0xFF,
+  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF,
+  0xF0, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xC0, 0x00, 0x7F, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xF8, 0x00, 0x03, 0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x3F, 0xFF, 0xFF,
+  0xFF, 0x80, 0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xF0, 0x00, 0x07, 0xFF,
+  0xFF, 0xFE, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF,
+  0xE0, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x3F, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xE0, 0x00, 0x07, 0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF, 0xE0, 0x00, 0x07, 0xFF, 0xFF, 0xFF,
+  0xFE, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xFF, 0xFF, 0xC0, 0x00, 0x07, 0xFF,
+  0xFF, 0xFE, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF,
+  0xE0, 0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x1F, 0xFF, 0xFF,
+  0xFF, 0xFF, 0x80, 0x00, 0x07, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xF8, 0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x0F, 0xFF,
+  0xFF, 0xFC, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x1F, 0xFF, 0xFF,
+  0xFF, 0xFE, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xE0, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x0F, 0xFF,
+  0xFF, 0xFC, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x0F, 0xFF, 0xFF,
+  0xFF, 0xFC, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x1F, 0xFF,
+  0xFF, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF,
+  0xFF, 0xF0, 0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF0,
+  0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x1F, 0xFF,
+  0xFF, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF,
+  0xFF, 0xC0, 0x00, 0x00, 0x1F, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x00,
+  0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x1F, 0xFF,
+  0xFF, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF,
+  0xFF, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF0, 0x00,
+  0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x00, 0x3F, 0xFF,
+  0xFF, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF,
+  0xFC, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF8, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x3F, 0xFF,
+  0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF,
+  0xF8, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x07, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x00, 0x7F, 0xFF,
+  0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF,
+  0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xC0, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF,
+  0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F, 0xFE,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x3F, 0xFC, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F, 0xFC,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x1F, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1F, 0xF0,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x1F, 0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0xC0,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x0F, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0x00, 0x00,
+  0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00,
+  0x03, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xE0,
+  0x3F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFC, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF8, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF0,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00, 0x00, 0x07,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xC0,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC,
+  0x3F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0x80, 0x00, 0x00, 0x00, 0x07, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xF8, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x03, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xE0,
+  0x03, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00,
+  0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0x00, 0x00,
+  0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x7F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFF, 0xFF, 0xFF,
+  0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
+  0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x7F, 0x80, 0x01, 0xFE, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFF, 0xFE, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xF0, 0x00, 0x00, 0x0F, 0xFF,
+  0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xC0, 0x01,
+  0xFE, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x3F,
+  0xFF, 0xF0, 0x00, 0x00, 0x3F, 0xFF, 0xFE, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x3F, 0xC0, 0x03, 0xFC, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x3F, 0xFF, 0xF0, 0x00, 0x00, 0xFF, 0xFF,
+  0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F, 0xE0, 0x03,
+  0xFC, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x3F,
+  0xFF, 0xF0, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x1F, 0xE0, 0x07, 0xF8, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFF, 0xFF, 0xFE, 0x00, 0x00, 0x3F, 0xFF, 0xF0, 0x00, 0x03, 0xFF, 0xFF,
+  0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0xF0, 0x07,
+  0xF8, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFE, 0x07, 0xFF, 0x00, 0x00, 0x3F,
+  0x80, 0x00, 0x00, 0x07, 0xFF, 0x00, 0x7F, 0xE0, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x0F, 0xF0, 0x0F, 0xF0, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFE, 0x01, 0xFF, 0x00, 0x00, 0x3F, 0x80, 0x00, 0x00, 0x07, 0xFC, 0x00,
+  0x3F, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x07, 0xF8, 0x0F,
+  0xE0, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFE, 0x00, 0xFF, 0x80, 0x00, 0x3F,
+  0x80, 0x00, 0x00, 0x0F, 0xF8, 0x00, 0x0F, 0xF8, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x07, 0xFC, 0x1F, 0xE0, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFE, 0x00, 0x7F, 0x80, 0x00, 0x3F, 0xFF, 0xE0, 0x00, 0x0F, 0xF0, 0x00,
+  0x0F, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0xFC, 0x1F,
+  0xC0, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFE, 0x00, 0x7F, 0x80, 0x00, 0x3F,
+  0xFF, 0xE0, 0x00, 0x0F, 0xF0, 0x00, 0x07, 0xF8, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x03, 0xFE, 0x3F, 0xC0, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFE, 0x00, 0x3F, 0x80, 0x00, 0x3F, 0xFF, 0xE0, 0x00, 0x1F, 0xE0, 0x00,
+  0x07, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0xFE, 0x3F,
+  0x80, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFE, 0x00, 0x3F, 0x80, 0x00, 0x3F,
+  0xFF, 0xE0, 0x00, 0x1F, 0xE0, 0x00, 0x07, 0xF8, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x01, 0xFF, 0x7F, 0x80, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFE, 0x00, 0x3F, 0x80, 0x00, 0x3F, 0xFF, 0xE0, 0x00, 0x0F, 0xF0, 0x00,
+  0x07, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0x7F,
+  0x00, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFE, 0x00, 0x7F, 0x80, 0x00, 0x3F,
+  0x80, 0x00, 0x00, 0x0F, 0xF0, 0x00, 0x07, 0xF8, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x7F, 0xFF, 0x00, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFE, 0x00, 0x7F, 0x80, 0x00, 0x3F, 0x80, 0x00, 0x00, 0x0F, 0xF8, 0x00,
+  0x0F, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x7F, 0xFE,
+  0x00, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFE, 0x00, 0xFF, 0x80, 0x00, 0x3F,
+  0x80, 0x00, 0x00, 0x0F, 0xFC, 0x00, 0x1F, 0xF0, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x3F, 0xFC, 0x00, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFE, 0x01, 0xFF, 0x00, 0x00, 0x3F, 0x80, 0x00, 0x00, 0x07, 0xFE, 0x00,
+  0x3F, 0xF0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x3F, 0xFC,
+  0x00, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFE, 0x0F, 0xFE, 0x00, 0x00, 0x3F,
+  0x80, 0x00, 0x00, 0x07, 0xFF, 0x80, 0xFF, 0xE0, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x1F, 0xF8, 0x00, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFF, 0xFF, 0xFC, 0x00, 0x00, 0x3F, 0xFF, 0xF0, 0x00, 0x03, 0xFF, 0xFF,
+  0xFF, 0xC0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x1F, 0xF8,
+  0x00, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFF, 0xFF, 0xF8, 0x00, 0x00, 0x3F,
+  0xFF, 0xF0, 0x00, 0x01, 0xFF, 0xFF, 0xFF, 0x80, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x0F, 0xF0, 0x00, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFF, 0xFF, 0xF0, 0x00, 0x00, 0x3F, 0xFF, 0xF0, 0x00, 0x00, 0x7F, 0xFF,
+  0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x07, 0xF0,
+  0x00, 0x00, 0x03, 0xF8, 0x00, 0x01, 0xFF, 0xFF, 0xC0, 0x00, 0x00, 0x3F,
+  0xFF, 0xF0, 0x00, 0x00, 0x3F, 0xFF, 0xFC, 0x00, 0x00, 0x00, 0x00, 0x00,
+  0x00, 0x00, 0x00, 0x00, 0x07, 0xE0, 0x00, 0x00, 0x03, 0xF8, 0x00, 0x01,
+  0xFF, 0xFC, 0x00, 0x00, 0x00, 0x3F, 0xFF, 0xF0, 0x00, 0x00, 0x07, 0xFF,
+  0xF0, 0x00, 0x00, 0x00, 0x00, 0x00
+};

examples/screensavers/toasters/toasters.ino

@@ -0,0 +1,138 @@
+// "Flying toasters" example for PicoDVI library. If just starting out,
+// see the 8bit_double_buffer which explains the PicoDVI groundwork.
+// Comments in THIS file are mostly distinct & new concepts.
+
+// IF NO OUTPUT OR RED FLICKER SCANLINES: try Tools->Optimize->(-O3)
+
+#include <PicoDVI.h>
+#include "sprites.h" // Graphics data
+
+DVIGFX8 display(DVI_RES_320x240p60, true, adafruit_feather_dvi_cfg);
+
+// See notes in 8bit_double_buffer regarding 400x240 mode.
+//DVIGFX8 display(DVI_RES_400x240p60, true, adafruit_feather_dvi_cfg);
+// Also requires -O3 setting.
+
+// This structure holds pointers to sprite graphics and masks in sprites.h.
+// First 8 are toasters, last 3 are toast.
+const struct {
+  const uint8_t *sprite;
+  const uint8_t *mask;
+} spritedata[] = {
+  { toaster0, toaster0_mask}, // There are really only 4 unique frames
+  { toaster1, toaster1_mask}, // of toaster flapping. Instead of code to
+  { toaster2, toaster2_mask}, // handle the back-and-forth motion, just
+  { toaster3, toaster3_mask}, // refer to frames in the desired sequence.
+  { toaster3, toaster3_mask}, // This also doubles up the first and last
+  { toaster2, toaster2_mask}, // frames to create a small pause at those
+  { toaster1, toaster1_mask}, // points in the loop.
+  { toaster0, toaster0_mask},
+  { toast0  , toast_mask},    // Light,
+  { toast1  , toast_mask},    // medium and
+  { toast2  , toast_mask},    // dark toast - all use the same mask.
+};
+
+// 12 sprites seems to be about the limit for this code to maintain
+// consistent 60 frames/sec updates. Double-buffered graphics synchronize
+// to display refresh, and if something can't complete in one frame,
+// everything is stalled until the next. It's especially annoying in
+// edge cases with some frames take 1/60 sec but others take 1/30.
+// See notes at end of file regarding potential improvements.
+#define N_SPRITES 12
+
+// This structure contains positions and other data for the sprites
+// in motion (notice it's not "const", because contents change).
+struct {
+  int16_t pos[2]; // sprite position (X,Y) * 16
+  int8_t  speed;  // sprite speed
+  uint8_t frame;  // for animation
+} sprite[N_SPRITES];
+
+// Initialize one sprite (index passed to function) to a random offscreen
+// position, also randomizing speed and sprite type (toaster or toast).
+void randomsprite(uint8_t i) {
+  // To move the sprites at slightly different speeds, coordinates are
+  // stored in 1/16 pixel units. When drawing, the stored values get
+  // divided by 16 to yield final pixel coordinates.
+  sprite[i].pos[0] = display.width() * 16; // Off right edge
+  sprite[i].pos[1] = random(-display.width() / 2, display.height()) * 16;
+  sprite[i].speed = random(8, 17); // Move 1/2 to 1 pixel per frame
+  // The spritedata array has 8 toaster frames and 3 toasts; just picking
+  // one randomly gives us 8/11 odds of a toaster (with a random initial
+  // wing position) and 3/11 of toast (with random done-ness), good mix.
+  sprite[i].frame = random(sizeof spritedata / sizeof spritedata[0]);
+}
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+
+  // Initialize color palette from table in sprites.h. Rather than
+  // calling setColor() for each one, we can just dump it directly...
+  memcpy(display.getPalette(), palette, sizeof palette);
+  display.swap(false, true); // Duplicate same palette to front & back buffers
+
+  // Randomize initial sprite states
+  randomSeed(analogRead(A0)); // Seed randomness from unused analog in
+  for (int i=0; i<N_SPRITES; i++) randomsprite(i);
+}
+
+uint8_t flap = 0; // Counter for flapping animation
+
+void loop() { // Runs once every frame
+  display.fillScreen(0);            // Clear back framebuffer,
+  for (int i=0; i<N_SPRITES; i++) { // and draw each sprite there...
+    // Although DVIGFX8 is a COLOR display type, we leverage GFX's
+    // drawGrayscaleBitmap() function to draw the sprites...it saves us
+    // writing a ton of code this way. See notes at end of this file.
+    // Also here's the "divide position by 16" mentioned earlier:
+    display.drawGrayscaleBitmap(sprite[i].pos[0] / 16, sprite[i].pos[1] / 16,
+      spritedata[sprite[i].frame].sprite, spritedata[sprite[i].frame].mask,
+      64, 64); // All sprites are 64x64 to simplify this code
+    // After drawing each sprite, update its position
+    sprite[i].pos[0] -= sprite[i].speed;     // Move left
+    sprite[i].pos[1] += sprite[i].speed / 2; // Move down (X:Y 2:1 ratio)
+    // Every fourth video frame, IF this sprite is a toaster (frame is 0-7),
+    // increment the sprite index and wrap around back to 0 if necessary...
+    if ((!(flap & 3)) && (sprite[i].frame < 8)) {
+      sprite[i].frame = (sprite[i].frame + 1) & 7; // Loop 0-7
+    } // else is video frame 1-3 or is toast (not animated)
+    // If the sprite has moved off the left or bottom edges, reassign it:
+    if ((sprite[i].pos[0] < -64 * 16) || (sprite[i].pos[1] > display.height() * 16))
+      randomsprite(i);
+  }
+  // Swap front/back buffers, do not duplicate current screen state
+  // to next frame, we'll draw it new from scratch each time.
+  display.swap();
+  flap++; // Increment animation counter; "rolls over" 0-255 automatically.
+}
+
+/*
+NOTES ON EFFICIENCY
+This was written to be a silly and somewhat simple example. No care is
+taken to minimize program size or speed, so it's NOT a good role model
+for complex situations, but it was quick to produce. Given time and effort,
+what could be improved?
+- As written, every sprite is 64x64 pixels, period. Makes it very easy to
+  draw. Flash space could be saved by cropping each sprite to a minimum
+  bounding rectangle...tradeoff being there would need to be a table of
+  sizes and X/Y offsets to maintain consistent motion when animating.
+  Demo's using ~230K -- just a fraction of available flash -- and life is
+  short so it's just not a priority here.
+- The GFX library's drawGrayscaleBitmap() function is used because it's
+  there and works for our needs, but is not especially efficient (testing
+  clipping on every pixel) and expects 1 byte/pixel data. Since this demo's
+  sprites are all 16-color, "packing" 2 pixels/byte is possible, using half
+  as much flash. A function could be written both to handle clipping more
+  efficiently and to de-pack and draw sprite data straight to framebuffer,
+  but would likely increase source code 2-3X and confuse novices.
+- The sprite data is all stored in flash memory, which is slower to access
+  than RAM. RAM is scarce (perhaps ~64K after PicoDVI claims a framebuffer),
+  but the sprites might still fit, especially if packed 2 pixels/byte. It's
+  not a bottleneck now AS WRITTEN (due to drawGrayscaleBitmap() itself being
+  a bit pokey), but if sprite-drawing were handled as mentioned above, could
+  be possible to increase the sprite count while maintaining frame rate by
+  adding the __not_in_flash attribute to these tables (see Pico SDK docs).
+*/

examples/screensavers/tvhost/tvhost.ino

@@ -0,0 +1,343 @@
+// "TV host" example for PicoDVI library. If just starting out,
+// see the 8bit_double_buffer which explains the PicoDVI groundwork.
+// Comments in THIS file are mostly distinct & new concepts.
+
+// IF NO OUTPUT OR RED FLICKER SCANLINES: try Tools->Optimize->(-O3)
+
+#include <PicoDVI.h>
+#include "sprites.h" // Graphics data
+
+DVIGFX8 display(DVI_RES_320x240p60, true, adafruit_feather_dvi_cfg);
+
+// See notes in 8bit_double_buffer regarding 400x240 mode.
+//DVIGFX8 display(DVI_RES_400x240p60, true, adafruit_feather_dvi_cfg);
+// Also requires -O3 setting.
+
+#define SPRITE_WIDTH  256
+#define SPRITE_HEIGHT 207
+
+// This structure holds pointers to sprite graphics and masks in sprites.h.
+const struct {
+  const uint8_t *sprite;
+  const uint8_t *mask;
+} spritedata[] = {
+  { sprite0 , mask0  },
+  { sprite1 , mask1  },
+  { sprite2 , mask2  },
+  { sprite3 , mask3  },
+  { sprite4 , mask4  },
+  { sprite5 , mask5  },
+  { sprite6 , mask6  },
+  { sprite7 , mask7  },
+  { sprite8 , mask8  },
+  { sprite9 , mask9  },
+  { sprite10, mask10 },
+  { sprite11, mask11 },
+  { sprite12, mask12 },
+  { sprite13, mask13 },
+  { sprite14, mask14 },
+  { sprite15, mask15 },
+  { sprite16, mask16 }
+};
+#define NUM_SPRITES (sizeof spritedata / sizeof spritedata[0])
+
+// Spinning cube implements 3D rotation & projection, but takes every
+// mathematical shortcut; not recommended as a starting point for proper
+// 3D drawing. Check out Wikipedia, Foley & Van Dam, etc.
+#define STEPS 20   // Number of lines between cube edges
+#define CAM_Z 1.8  // Camera position along Z axis; approx sqrt(3) * 1.05
+float scale;       // Calc'd once per frame for in/out bounce effect
+float min_scale = sqrt((display.width() * display.width() + display.height() * display.height()) / 4) / M_SQRT2 * 2.0;
+
+void setup() { // Runs once on startup
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+
+  // Initialize color palette from table in sprites.h. Rather than
+  // calling setColor() for each one, we can just dump it directly...
+  memcpy(display.getPalette(), palette, sizeof palette);
+  // And then override the last few entries for our background cube...
+  display.setColor(0, 0, 0, 0);       // Black background
+  display.setColor(253, 255, 0, 255); // Index 253 = magenta
+  display.setColor(254, 0, 255, 255); // Index 254 = cyan
+  display.setColor(255, 255, 255, 0); // Index 255 = yellow
+  display.swap(false, true); // Duplicate same palette into front & back buffers
+}
+
+const float corners[8][3] = { // Cube vertices in original coord space
+  { -1, -1, -1 },
+  {  1, -1, -1 },
+  { -1,  1, -1 },
+  {  1,  1, -1 },
+  { -1, -1,  1 },
+  {  1, -1,  1 },
+  { -1,  1,  1 },
+  {  1,  1,  1 },
+};
+
+float rotated[8][3]; // Cube vertices after rotation applied
+// (Cam is stationary, cube is transformed, rather than other way around)
+
+// Draw one face of the cube, if it's visible (not back-facing).
+// Some points & doubled-up lines could be avoided with more care,
+// but code would be somewhat longer. Not a bottleneck here, so...
+void face(uint8_t p1, uint8_t p2, uint8_t p3, uint8_t p4, uint8_t p5, uint8_t color) {
+  // p1-p4 are indices of 4 corners, p5 is a perpendicular point (surface normal)
+  float vx = rotated[p1][0];                  // Vector from cam to first point
+  float vy = rotated[p1][1];                  // (not normalized)
+  float vz = rotated[p1][2] - CAM_Z;
+  float nx = rotated[p5][0] - rotated[p1][0]; // Surface normal
+  float ny = rotated[p5][1] - rotated[p1][1]; // (not normalized)
+  float nz = rotated[p5][2] - rotated[p1][2];
+  // Dot product of vectors. Since only the sign of the result is needed
+  // below, the two vectors do not require normalization first.
+  float dot = vx * nx + vy * ny + vz * nz;
+  if (dot < 0) {                 // Facing camera?
+    for(int i=0; i<STEPS; i++) { // Interpolate parallel lines...
+      float s1 = (float)i / (float)(STEPS - 1); // Weighting of p1, p3
+      float s2 = 1.0 - s1;                      // Weighting of p2, p4
+      // Interpolate between p1, p2
+      float t1x = rotated[p1][0] * s1 + rotated[p2][0] * s2;
+      float t1y = rotated[p1][1] * s1 + rotated[p2][1] * s2;
+      float t1z = rotated[p1][2] * s1 + rotated[p2][2] * s2;
+      // Interpolate between p3, p4
+      float t2x = rotated[p3][0] * s1 + rotated[p4][0] * s2;
+      float t2y = rotated[p3][1] * s1 + rotated[p4][1] * s2;
+      float t2z = rotated[p3][2] * s1 + rotated[p4][2] * s2;
+      // Project interpolated endpoints to image plane @ Z=0
+      float dz = t1z - CAM_Z; // Camera to point
+      float x1 = (float)(display.width()  / 2) + t1x / dz * scale + 0.5;
+      float y1 = (float)(display.height() / 2) + t1y / dz * scale + 0.5;
+      dz = t2z - CAM_Z;       // Camera to point
+      float x2 = (float)(display.width()  / 2) + t2x / dz * scale + 0.5;
+      float y2 = (float)(display.height() / 2) + t2y / dz * scale + 0.5;
+      line(x1, y1, x2, y2, color);
+    }
+  }
+}
+
+uint8_t  frame = 0;         // For animation timing
+bool     stutter = false;   // If true, doing A/B stutter animation
+uint32_t duration = 3000;   // Time to hold current mode, in milliseconds
+uint32_t lastModeSwitchTime = 0;
+uint8_t  var[2] = { 4, 4 }; // Mode-specific variables
+uint8_t  spritenum = 0;     // Active sprite index
+
+void loop() { // Runs once every frame
+  display.fillScreen(0); // Clear back framebuffer, will do full redraw...
+
+  // This does the in & out bounce effect (bit of a cheat but less math
+  // than moving camera in & out. Cam stays in one position just far
+  // enough that it's never inside cube, need not handle Z clipping.)
+  scale = min_scale * (1.0 + (float)abs((long)(millis() & 4095) - 2048) / 8192.0); // 1.0-1.25X
+
+  // Rotate cube vertices from corners[] to rotated[]
+  float now = millis() / 1000.0; // Elapsed time in seconds
+  float rx = now * -0.19;        // X, Y, Z rotation
+  float ry = now *  0.23;
+  float rz = now *  0.13;
+  float cx = cos(rx);
+  float cy = cos(ry);
+  float cz = cos(rz);
+  float sx = sin(rx);
+  float sy = sin(ry);
+  float sz = sin(rz);
+  for (int i=0; i<8; i++) {
+    float x1 = corners[i][0];
+    float y1 = corners[i][1];
+    float z1 = corners[i][2];
+    float y2 = y1 * cx - z1 * sx;
+    float z2 = y1 * sx + z1 * cx;
+    float x2 = x1 * cy - z2 * sy;
+    float z3 = x1 * sy + z2 * cy;
+    float x3 = x2 * cz - y2 * sz;
+    float y3 = x2 * sz + y2 * cz;
+    rotated[i][0] = x3;
+    rotated[i][1] = y3;
+    rotated[i][2] = z3;
+  }
+
+  face(0, 1, 4, 5, 2, 253); // Test & draw 6 faces of cube
+  face(0, 2, 4, 6, 1, 254);
+  face(2, 3, 6, 7, 0, 253);
+  face(1, 3, 5, 7, 0, 254);
+  face(0, 1, 2, 3, 4, 255);
+  face(4, 5, 6, 7, 0, 255);
+
+  // Check if switching in/out of stutter mode
+  if ((millis() - lastModeSwitchTime) >= duration) { // Time to switch?
+    lastModeSwitchTime = millis();   // Note time,
+    stutter = !stutter;              // and make the switch
+    if (stutter) {                   // If entering stutter mode...
+      duration = random(250, 500);   // Go for 1/4 to 1/2 second
+      var[0] = random(NUM_SPRITES);  // Pick 2 different sprites
+      var[1] = (var[0] + 1 + random(NUM_SPRITES - 1)) % NUM_SPRITES;
+    } else {                         // Returning to regular mode...
+      duration = random(2000, 5000); // Go for 2-5 seconds
+      var[0] = random(3, 6);         // Hold each pose for 3-5 frames
+    }
+  }
+
+  // Choose sprite to display based on current mode...
+  if (stutter) {
+    // Every second frame, alternate between two sprites
+    spritenum = var[((frame >> 1) & 1)];
+  } else {
+    // Pick among random frames (may repeat, is OK)
+    if (!(frame % var[0])) {
+      spritenum = random(NUM_SPRITES);
+    }
+  }
+
+  // Overlay big sprite. Although DVIGFX8 is a COLOR display type,
+  // we leverage GFX's drawGrayscaleBitmap() function to draw it...
+  // saves us writing a ton of code this way.
+  // Comment this out if you want ONLY the background cube.
+  display.drawGrayscaleBitmap((display.width() - SPRITE_WIDTH) / 2,
+    display.height() - SPRITE_HEIGHT, spritedata[spritenum].sprite,
+    spritedata[spritenum].mask, SPRITE_WIDTH, SPRITE_HEIGHT);
+
+  // Swap front/back buffers, do not duplicate current screen state
+  // to next frame, we'll draw it new from scratch each time.
+  display.swap();
+  frame++; // For animation timing; wraps around 0-255
+}
+
+// A somewhat smooth(er) line drawing function. Not antialiased, but with
+// subpixel positioning that makes the background animation less "jumpy"
+// than with GFX, which is all integer end points.
+
+// Returns bitmask of which edge(s) a point exceeds
+uint8_t xymask(float x, float y) {
+  return ( (x < 0.0      ) | ((x >= (float)display.width() ) << 1) |
+          ((y < 0.0) << 2) | ((y >= (float)display.height()) << 3));
+}
+
+// Returns bitmask of which X edge(s) a point exceeds
+uint8_t xmask(float x) {
+  return (x < 0.0) | ((x >= (float)display.width()) << 1);
+}
+
+void line(float x1, float y1, float x2, float y2, uint8_t color) {
+float ox1 = x1, oy1 = y1, ox2 = x2, oy2 = y2;
+  uint8_t mask1 = xymask(x1, y1); // If both endpoints are
+  uint8_t mask2 = xymask(x2, y2); // off same edge(s),
+  if (mask1 & mask2) return;      // line won't overlap screen
+
+  float dx = x2 - x1;
+  float dy = y2 - y1;
+
+  // Clip top
+  if (mask1 & 4) {
+    x1 -= y1 * dx / dy;
+    y1  = 0.0;
+  } else if (mask2 & 4) {
+    x2 -= y2 * dx / dy;
+    y2  = 0.0;
+  }
+
+  float maxy = (float)(display.height() - 1);
+
+  // Clip bottom
+  if (mask1 & 8) {
+    x1 -= (y1 - maxy) * dx / dy;
+    y1  = maxy;
+  } else if (mask2 & 8) {
+    x2 -= (y2 - maxy) * dx / dy;
+    y2  = maxy;
+  }
+
+  mask1 = xmask(x1); // Repeat the offscreen check after Y clip
+  mask2 = xmask(x2);
+  if (mask1 & mask2) return;
+
+  dx = x2 - x1;
+  dy = y2 - y1;
+
+  // Clip left
+  if (mask1 & 1) {
+    y1 -= x1 * dy / dx;
+    x1  = 0.0;
+  } else if (mask2 & 1) {
+    y2 -= x2 * dy / dx;
+    x2  = 0.0;
+  }
+
+  float maxx = (float)(display.width() - 1);
+
+  // Clip right
+  if (mask1 & 2) {
+    y1 -= (x1 - maxx) * dy / dx;
+    x1  = maxx;
+  } else if (mask2 & 2) {
+    y2 -= (x2 - maxx) * dy / dx;
+    x2  = maxx;
+  }
+
+  // (x1, y1) to (x2, y2) is known to be on-screen and in-bounds now.
+
+  // Handle a couple special cases (horizontal, vertical lines) first,
+  // GFX takes care of these fine and it avoids some divide-by-zero
+  // checks in the diagonal code later.
+  if ((int)y1 == (int)y2) { // Horizontal or single point
+    int16_t x, w;
+    if (x2 >= x1) {
+      x = (int)x1;
+      w = (int)x2 - (int)x1 + 1;
+    } else {
+      x = (int)x2;
+      w = (int)x1 - (int)x2 + 1;
+    }
+    display.drawFastHLine(x, (int)y1, w, color);
+  } else if ((int)x1 == (int)x2) { // Vertical
+    int16_t y, h;
+    if (y2 >= y1) {
+      y = (int)y1;
+      h = (int)y2 - (int)y1 + 1;
+    } else {
+      y = (int)y2;
+      h = (int)y1 - (int)y2 + 1;
+    }
+    display.drawFastVLine((int)x1, y, h, color);
+  } else { // Diagonal
+    dx = x2 - x1;
+    dy = y2 - y1;
+
+    uint8_t *ptr = display.getBuffer();
+
+    // This is a bit ugly in that it uses floating-point math in the line
+    // drawing loop. There are more optimal Bresenham-esque fixed-point
+    // approaches to do this (initializing the error term based on subpixel
+    // endpoints and slopes), but A) I'm out of spoons today, and B) we're
+    // drawing just a few dozen lines and it's simply not a bottleneck in
+    // this demo. Just saying this won't scale up to thousands of lines.
+
+    if (fabs(dx) >= fabs(dy)) { // "Wide" line
+      if (x1 > x2) {            // Put (x1, y1) at left
+        float t = x1; x1 = x2; x2 = t;
+        t = y1; y1 = y2; y2 = t;
+      }
+      uint16_t ix1   = (uint16_t)x1;
+      uint16_t ix2   = (uint16_t)x2;
+      float    slope = dy / dx;
+      for (uint16_t x=ix1; x <= ix2; x++) {
+        uint16_t iy = (uint16_t)(y1 + slope * (float)(x - ix1));
+        if (iy < display.height()) ptr[iy * display.width() + x] = color;
+      }
+    } else {         // "Tall" line
+      if (y1 > y2) { // Put (x1, y1) at top
+        float t = x1; x1 = x2; x2 = t;
+        t = y1; y1 = y2; y2 = t;
+      }
+      uint16_t iy1   = (uint16_t)y1;
+      uint16_t iy2   = (uint16_t)y2;
+      float    slope = dx / dy;
+      for (uint16_t y=iy1; y <= iy2; y++) {
+        uint16_t ix = (uint16_t)(x1 + slope * (float)(y - iy1));
+        if (ix < display.width()) ptr[y * display.width() + ix] = color;
+      }
+    }
+  }
+}

examples/virtual_spitft/fourwire.pio

@@ -0,0 +1,24 @@
+.program fourwire
+; Sample bits using an external clock, and push groups of bits into the RX FIFO.
+; - IN pin 0 is the data pin   (GPIO18)
+; - IN pin 1 is the dc pin     (GPIO19)
+; - IN pin 2 is the clock pin  (GPIO20)
+; - JMP pin is the chip select (GPIO21)
+; - Autopush is enabled, threshold 8
+;
+; This program waits for chip select to be asserted (low) before it begins
+; clocking in data. Whilst chip select is low, data is clocked continuously. If
+; chip select is deasserted part way through a data byte, the partial data is
+; discarded. This makes use of the fact a mov to isr clears the input shift
+; counter.
+flush:
+    mov isr, null         ; Clear ISR and input shift counter
+    jmp check_chip_select ; Poll chip select again
+.wrap_target
+do_bit:
+    wait 0 pin 2          ; Detect rising edge and sample input data
+    wait 1 pin 2          ; (autopush takes care of moving each complete
+    in pins, 2            ; data word to the FIFO)
+check_chip_select:
+    jmp pin, flush        ; Bail out if we see chip select high
+.wrap

examples/virtual_spitft/virtual_spitft.ino

@@ -0,0 +1,251 @@
+// PicoDVI-based "virtual SPITFT" display. Receives graphics commands/data
+// over 4-wire SPI interface, mimicking functionality of displays such as
+// ST7789 or ILI9341, but shown on an HDMI monitor instead.
+
+#include <PicoDVI.h> // Core display & graphics library
+
+// Configurables ----
+
+// GPIO connected to (or shared with) TFT control.
+// Careful not to overlap the DVI pins.
+#define PIN_DATA 9 // 3 contiguous pins start here: data, DC, clk
+#define PIN_CS   6 // Chip-select need not be contiguous
+
+// 320x240 16-bit color display (to match common TFT display resolution):
+DVIGFX16 display(DVI_RES_320x240p60, adafruit_feather_dvi_cfg);
+
+// Output of pioasm ----
+
+#define fourwire_wrap_target 2
+#define fourwire_wrap 5
+
+static const uint16_t fourwire_program_instructions[] = {
+    0xa0c3, //  0: mov    isr, null                  
+    0x0005, //  1: jmp    5                          
+            //     .wrap_target
+    0x2022, //  2: wait   0 pin, 2                   
+    0x20a2, //  3: wait   1 pin, 2                   
+    0x4002, //  4: in     pins, 2                    
+    0x00c0, //  5: jmp    pin, 0                     
+            //     .wrap
+};
+
+static const struct pio_program fourwire_program = {
+    .instructions = fourwire_program_instructions,
+    .length = 6,
+    .origin = -1,
+};
+
+static inline pio_sm_config fourwire_program_get_default_config(uint offset) {
+    pio_sm_config c = pio_get_default_sm_config();
+    sm_config_set_wrap(&c, offset + fourwire_wrap_target, offset + fourwire_wrap);
+    return c;
+}
+
+// end pioasm output ----
+
+PIO pio = pio1; // libdvi uses pio0 (but has 1 avail state machine if you want to use it)
+uint sm;
+uint16_t *framebuf = display.getBuffer();
+uint8_t decode[256];
+#define DECODE(w) (decode[(w & 0x55) | ((w >> 7) & 0xaa)])
+
+#define BIT_DEPOSIT(b, i) ((b) ? (1<<(i)) : 0)
+#define BIT_EXTRACT(b, i) (((b) >> (i)) & 1)
+#define BIT_MOVE(b, src, dest) BIT_DEPOSIT(BIT_EXTRACT(b, src), dest)
+#define ENCODED_COMMAND(x) ( \
+    (BIT_MOVE(x, 0,  0)) | \
+    (BIT_MOVE(x, 1,  2)) | \
+    (BIT_MOVE(x, 2,  4)) | \
+    (BIT_MOVE(x, 3,  6)) | \
+    (BIT_MOVE(x, 4,  8)) | \
+    (BIT_MOVE(x, 5, 10)) | \
+    (BIT_MOVE(x, 6, 12)) | \
+    (BIT_MOVE(x, 7, 14)) \
+)
+
+#define COMMAND_NOP     (0x00)
+#define COMMAND_SWRESET (0x01)
+#define COMMAND_CASET   (0x2a)
+#define COMMAND_PASET   (0x2b)
+#define COMMAND_RAMWR   (0x2c)
+#define COMMAND_MADCTL  (0x36)
+
+#define MADCTL_MY 0x80
+#define MADCTL_MX 0x40
+#define MADCTL_MV 0x20
+#define MADCTL_ML 0x10
+
+void setup() {
+  Serial.begin(115200);
+  //while(!Serial);
+  if (!display.begin()) { // Blink LED if insufficient RAM
+    pinMode(LED_BUILTIN, OUTPUT);
+    for (;;) digitalWrite(LED_BUILTIN, (millis() / 500) & 1);
+  }
+
+  for(int i=0; i<256; i++) {
+    int j = (BIT_MOVE(i, 0, 0)) |
+            (BIT_MOVE(i, 2, 1)) |
+            (BIT_MOVE(i, 4, 2)) |
+            (BIT_MOVE(i, 6, 3)) |
+            (BIT_MOVE(i, 1, 4)) |
+            (BIT_MOVE(i, 3, 5)) |
+            (BIT_MOVE(i, 5, 6)) |
+            (BIT_MOVE(i, 7, 7));
+    decode[i] = j;
+  }
+    
+  uint offset = pio_add_program(pio, &fourwire_program);
+  sm = pio_claim_unused_sm(pio, true);
+
+  pio_sm_config c = fourwire_program_get_default_config(offset);
+
+  // Set the IN base pin to the provided PIN_DATA parameter. This is the data
+  // pin, and the next-numbered GPIO is used as the clock pin.
+  sm_config_set_in_pins(&c, PIN_DATA);
+  sm_config_set_jmp_pin(&c, PIN_CS);
+  // Set the pin directions to input at the PIO
+  pio_sm_set_consecutive_pindirs(pio, sm, PIN_DATA, 3, false);
+  pio_sm_set_consecutive_pindirs(pio, sm, PIN_CS, 1, false);
+  // Connect GPIOs to PIO block, set pulls
+  for (uint8_t i=0; i<3; i++) {
+    pio_gpio_init(pio, PIN_DATA + i);
+    gpio_set_pulls(PIN_DATA + i, true, false);
+  }
+  pio_gpio_init(pio, PIN_CS);
+  gpio_set_pulls(PIN_CS, true, false);
+
+  // Shifting to left matches the customary MSB-first ordering of SPI.
+  sm_config_set_in_shift(
+      &c,
+      false, // Shift-to-right = false (i.e. shift to left)
+      true,  // Autopush enabled
+      16     // Autopush threshold
+  );
+
+  // We only receive, so disable the TX FIFO to make the RX FIFO deeper.
+  sm_config_set_fifo_join(&c, PIO_FIFO_JOIN_RX);
+
+  // Load our configuration, and start the program from the beginning
+  pio_sm_init(pio, sm, offset, &c);
+  pio_sm_set_enabled(pio, sm, true);
+
+  // State machine should handle malformed requests somewhat,
+  // e.g. RAMWR writes that wrap around or drop mid-data.
+
+  uint8_t cmd = COMMAND_NOP;                  // Last received command
+  uint16_t X0 = 0, X1 = display.width() - 1;  // Address window X
+  uint16_t Y0 = 0, Y1 = display.height() - 1; // Address window Y
+  uint16_t x = 0, y = 0;                      // Current pixel pos.
+  union { // Data receive buffer sufficient for implemented commands
+    uint8_t  b[4];
+    uint16_t w[2];
+    uint32_t l;
+  } buf;
+  int8_t bufidx = -1; // Current pos. in buf.b[] array (or -1 = full)
+
+  for (;;) {
+    uint16_t ww = pio_sm_get_blocking(pio, sm); // Read next word (data & DC interleaved)
+    if ((ww & 0x2)) { // DC bit is set, that means it's data (most common case, hence 1st)
+      if (bufidx >= 0) { // Decode & process only if recv buffer isn't full
+        buf.b[bufidx] = DECODE(ww);
+        // Buffer is filled in reverse so byte swaps aren't needed on uint16_t values
+        if (--bufidx < 0) { // Receive threshold reached?
+          switch (cmd) {
+           case COMMAND_CASET:
+            // Clipping is not performed here because framebuffer
+            // may be a different size than implied SPI device.
+            // That occurs in the RAMWR condition later.
+            X0 = buf.w[1]; // [sic.] 1 because buffer is loaded in reverse
+            X1 = buf.w[0];
+            if (X0 > X1) {
+              uint16_t tmp = X0;
+              X0 = X1;
+              X1 = tmp;
+            }
+            break;
+           case COMMAND_PASET:
+            Y0 = buf.w[1]; // [sic.] 1 because buffer is loaded in reverse
+            Y1 = buf.w[0];
+            if (Y0 > Y1) {
+              uint16_t tmp = Y0;
+              Y0 = Y1;
+              Y1 = tmp;
+            }
+            break;
+           case COMMAND_RAMWR:
+            // Write pixel to screen, increment X/Y, wrap around as needed.
+            // drawPixel() is used as it handles both clipping & rotation,
+            // saves a lot of bother here. However, this only handles rotation,
+            // NOT full MADCTL mapping, but the latter is super rare, I think
+            // it's only used in some eye code to mirror one of two screens.
+            // If it's required, then rotation, mirroring and clipping will
+            // all need to be handled in this code...but, can write direct to
+            // framebuffer then, might save some cycles.
+            display.drawPixel(x, y, buf.w[0]);
+            if (++x > X1) {
+              x = X0;
+              if (++y > Y1) {
+                y = Y0;
+              }
+            }
+            bufidx = 1; // Reset buffer counter for next pixel
+            // Buflen is left as-is, so more pixels can be processed
+            break;
+           case COMMAND_MADCTL:
+            switch (buf.b[0] & 0xF0) {
+             case MADCTL_MX | MADCTL_MV:             // ST77XX
+             case MADCTL_MX | MADCTL_MY | MADCTL_MV: // ILI9341
+              display.setRotation(0);
+              break;
+             case MADCTL_MX | MADCTL_MY: // ST77XX
+             case MADCTL_MX:             // ILI9341
+              display.setRotation(1);
+              break;
+             case MADCTL_MY | MADCTL_MV: // ST77XX
+             case MADCTL_MV:             // ILI9341
+              display.setRotation(2);
+              break;
+             case 0:         // ST77XX
+             case MADCTL_MY: // ILI9341
+              display.setRotation(3);
+              break;
+            }
+            break;
+          }
+        }
+      }
+    } else { // Is command
+      cmd = DECODE(ww);
+      switch (cmd) {
+       case COMMAND_SWRESET:
+        display.setRotation(0);
+        x = y = X0 = Y0 = 0;
+        X1 = display.width() - 1;
+        Y1 = display.height() - 1;
+        break;
+       case COMMAND_CASET:
+        bufidx = 3; // Expecting two 16-bit values (X0, X1)
+        break;
+       case COMMAND_PASET:
+        bufidx = 3; // Expecting two 16-bit values (Y0, Y1)
+        break;
+       case COMMAND_RAMWR:
+        bufidx = 1; // Expecting one 16-bit value (or more)
+        x = X0;     // Start at UL of address window
+        y = Y0;
+        break;
+       case COMMAND_MADCTL:
+        bufidx = 0; // Expecting one 8-bit value
+        break;
+       default:
+        // Unknown or unimplemented command, discard any data that follows
+        bufidx = -1;
+      }
+    }
+  }
+}
+
+void loop() {
+}

library.properties

@@ -0,0 +1,10 @@
+name=PicoDVI - Adafruit Fork
+version=1.3.2
+author=Luke Wren (Wren6991)
+maintainer=Adafruit <info@adafruit.com>
+sentence=Arduino library for RP2040 DVI output, based on PicoDVI
+paragraph=Arduino library for RP2040 DVI output, based on PicoDVI
+category=Display
+url=https://github.com/adafruit/PicoDVI
+architectures=rp2040
+depends=Adafruit GFX Library, Adafruit CPFS

software/apps/bad_apple/CMakeLists.txt

@@ -23,6 +23,7 @@ target_compile_definitions(bad_apple PRIVATE
 	DVI_VERTICAL_REPEAT=1
 	DVI_N_TMDS_BUFFERS=3
 	DVI_MONOCHROME_TMDS
+	DVI_1BPP_BIT_REVERSE=0
 )
 
 target_link_libraries(bad_apple

software/apps/colour_terminal/tmds_encode_font_2bpp.h

@@ -3,6 +3,11 @@
 
 #include "pico/types.h"
 
+#if defined(__cplusplus)
+extern "C"
+{
+#endif
+
 // Render characters using an 8px-wide font and a per-character 2bpp
 // foreground/background colour. This function is fast enough to run 3 times
 // per scanline on one core, so RGB222 coloured text can be rendered (with
@@ -20,4 +25,8 @@
 void tmds_encode_font_2bpp(const uint8_t *charbuf, const uint32_t	*colourbuf,
 	uint32_t *tmdsbuf, uint n_pix, const uint8_t *font_line);
 
+#if defined(__cplusplus)
+}
+#endif
+
 #endif

software/apps/dht_logging/CMakeLists.txt

@@ -13,6 +13,7 @@ target_compile_definitions(dht_logging PRIVATE
 	DVI_DEFAULT_SERIAL_CONFIG=${DVI_DEFAULT_SERIAL_CONFIG}
 	DVI_VERTICAL_REPEAT=1
 	DVI_N_TMDS_BUFFERS=3
+	DVI_1BPP_BIT_REVERSE=0
 	)
 
 

software/apps/moon/CMakeLists.txt

@@ -11,6 +11,7 @@ target_compile_definitions(moon PRIVATE
 	DVI_VERTICAL_REPEAT=1
 	DVI_N_TMDS_BUFFERS=3
 	DVI_MONOCHROME_TMDS
+	DVI_1BPP_BIT_REVERSE=0
 )
 
 target_link_libraries(moon

software/apps/terminal/CMakeLists.txt

@@ -14,6 +14,7 @@ target_compile_definitions(terminal PRIVATE
 	DVI_VERTICAL_REPEAT=1
 	DVI_N_TMDS_BUFFERS=3
 	DVI_MONOCHROME_TMDS=1
+	DVI_1BPP_BIT_REVERSE=0
 	)
 
 target_link_libraries(terminal

software/apps/terminal/main.c

@@ -24,11 +24,12 @@
 
 
 // Pick one:
-#define MODE_640x480_60Hz
+
+//#define MODE_640x480_60Hz
 // #define MODE_720x480_60Hz
 // #define MODE_800x600_60Hz
 // #define MODE_960x540p_60Hz
-// #define MODE_1280x720_30Hz
+#define MODE_1280x720_30Hz
 
 #if defined(MODE_640x480_60Hz)
 // DVDD 1.2V (1.1V seems ok too)

software/include/common_dvi_pin_configs.h

@@ -5,7 +5,11 @@
 // developing on. It's not a particularly important file -- just saves some
 // copy + paste.
 
+#if defined(ARDUINO)
+#include "../libdvi/dvi_serialiser.h"
+#else
 #include "dvi_serialiser.h"
+#endif
 
 #ifndef DVI_DEFAULT_SERIAL_CONFIG
 #define DVI_DEFAULT_SERIAL_CONFIG pico_sock_cfg
@@ -101,13 +105,42 @@ static const struct dvi_serialiser_cfg not_hdmi_featherwing_cfg = {
 
 // Adafruit Feather RP2040 DVI
 static const struct dvi_serialiser_cfg adafruit_feather_dvi_cfg = {
-	.pio = pio0,
+	.pio = DVI_DEFAULT_PIO_INST,
 	.sm_tmds = {0, 1, 2},
 	.pins_tmds = {18, 20, 22},
 	.pins_clk = 16,
 	.invert_diffpairs = true
 };
 
+static const struct dvi_serialiser_cfg adafruit_dvibell_cfg = {
+  .pio = DVI_DEFAULT_PIO_INST,
+  .sm_tmds = {0, 1, 2},
+  .pins_tmds = {10, 8, 6},
+  .pins_clk = 12,
+  .invert_diffpairs = false,
+};
+
+// hstx dvi cowbell
+static const struct dvi_serialiser_cfg adafruit_hstxdvibell_cfg = {
+  .pio = DVI_DEFAULT_PIO_INST,
+  .sm_tmds = {0, 1, 2},
+  .pins_tmds = {12, 18, 16},
+  .pins_clk = 14,
+  .invert_diffpairs = false,
+};
+
+
+// Adafruit 'Fruit Jam' computer board
+static const struct dvi_serialiser_cfg adafruit_fruitjam_cfg = {
+	.pio = DVI_DEFAULT_PIO_INST,
+	.sm_tmds = {0, 1, 2},
+	.pins_tmds = {14, 16, 18},
+	.pins_clk = 12,
+	.invert_diffpairs = true,
+};
+
+
+
 // Waveshare RP2040-PiZero
 static const struct dvi_serialiser_cfg waveshare_rp2040_pizero = {
 	.pio = DVI_DEFAULT_PIO_INST,
@@ -117,4 +150,13 @@ static const struct dvi_serialiser_cfg waveshare_rp2040_pizero = {
 	.invert_diffpairs = false
 };
 
+// Olimex RP2040PC, RP2040-Pico-PC, Neo6502, 
+static struct dvi_serialiser_cfg olimex_rp2040_cfg = {
+    .pio = DVI_DEFAULT_PIO_INST,
+    .sm_tmds = {0, 1, 2},
+    .pins_tmds = {14, 18, 16},
+    .pins_clk = 12,
+    .invert_diffpairs = true
+};
+
 #endif

software/libdvi/dvi.c

@@ -7,6 +7,10 @@
 #include "dvi_serialiser.h"
 #include "tmds_encode.h"
 
+// Adafruit PicoDVI fork requires a couple global items run-time configurable:
+uint8_t dvi_vertical_repeat = DVI_VERTICAL_REPEAT;
+bool    dvi_monochrome_tmds = DVI_MONOCHROME_TMDS;
+
 // Time-critical functions pulled into RAM but each in a unique section to
 // allow garbage collection
 #define __dvi_func(f) __not_in_flash_func(f)
@@ -37,16 +41,19 @@ void dvi_init(struct dvi_inst *inst, uint spinlock_tmds_queue, uint spinlock_col
 
 	dvi_setup_scanline_for_vblank(inst->timing, inst->dma_cfg, true, &inst->dma_list_vblank_sync);
 	dvi_setup_scanline_for_vblank(inst->timing, inst->dma_cfg, false, &inst->dma_list_vblank_nosync);
+#if defined(ARDUINO)
+	dvi_setup_scanline_for_active(inst->timing, inst->dma_cfg, (uint32_t*)SRAM_BASE, &inst->dma_list_active);
+#else
 	dvi_setup_scanline_for_active(inst->timing, inst->dma_cfg, (void*)SRAM_BASE, &inst->dma_list_active);
+#endif
 	dvi_setup_scanline_for_active(inst->timing, inst->dma_cfg, NULL, &inst->dma_list_error);
 
 	for (int i = 0; i < DVI_N_TMDS_BUFFERS; ++i) {
 		void *tmdsbuf;
-#if DVI_MONOCHROME_TMDS
-		tmdsbuf = malloc(inst->timing->h_active_pixels / DVI_SYMBOLS_PER_WORD * sizeof(uint32_t));
-#else
-		tmdsbuf = malloc(3 * inst->timing->h_active_pixels / DVI_SYMBOLS_PER_WORD * sizeof(uint32_t));
-#endif
+		if (dvi_monochrome_tmds)
+			tmdsbuf = malloc(inst->timing->h_active_pixels / DVI_SYMBOLS_PER_WORD * sizeof(uint32_t));
+		else
+			tmdsbuf = malloc(3 * inst->timing->h_active_pixels / DVI_SYMBOLS_PER_WORD * sizeof(uint32_t));
 		if (!tmdsbuf)
 			panic("TMDS buffer allocation failed");
 		queue_add_blocking_u32(&inst->q_tmds_free, &tmdsbuf);
@@ -201,7 +208,7 @@ static void __dvi_func(dvi_dma_irq_handler)(struct dvi_inst *inst) {
 		tmdsbuf = NULL;
 	}
 	else if (queue_try_peek_u32(&inst->q_tmds_valid, &tmdsbuf)) {
-		if (inst->timing_state.v_ctr % DVI_VERTICAL_REPEAT == DVI_VERTICAL_REPEAT - 1) {
+		if (inst->timing_state.v_ctr % dvi_vertical_repeat == dvi_vertical_repeat - 1) {
 			queue_remove_blocking_u32(&inst->q_tmds_valid, &tmdsbuf);
 			inst->tmds_buf_release_next = tmdsbuf;
 		}
@@ -209,7 +216,7 @@ static void __dvi_func(dvi_dma_irq_handler)(struct dvi_inst *inst) {
 	else {
 		// No valid scanline was ready (generates solid red scanline)
 		tmdsbuf = NULL;
-		if (inst->timing_state.v_ctr % DVI_VERTICAL_REPEAT == DVI_VERTICAL_REPEAT - 1)
+		if (inst->timing_state.v_ctr % dvi_vertical_repeat == dvi_vertical_repeat - 1)
 			++inst->late_scanline_ctr;
 	}
 
@@ -222,7 +229,7 @@ static void __dvi_func(dvi_dma_irq_handler)(struct dvi_inst *inst) {
 			else {
 				_dvi_load_dma_op(inst->dma_cfg, &inst->dma_list_error);
 			}
-			if (inst->scanline_callback && inst->timing_state.v_ctr % DVI_VERTICAL_REPEAT == DVI_VERTICAL_REPEAT - 1) {
+			if (inst->scanline_callback && inst->timing_state.v_ctr % dvi_vertical_repeat == dvi_vertical_repeat - 1) {
 				inst->scanline_callback();
 			}
 			break;

software/libdvi/dvi.h

@@ -52,6 +52,11 @@ struct dvi_inst {
 
 };
 
+#if defined(__cplusplus)
+extern "C"
+{
+#endif
+
 // Set up data structures and hardware for DVI.
 void dvi_init(struct dvi_inst *inst, uint spinlock_tmds_queue, uint spinlock_colour_queue);
 
@@ -73,7 +78,7 @@ void dvi_scanbuf_main_16bpp(struct dvi_inst *inst);
 void dvi_framebuf_main_8bpp(struct dvi_inst *inst);
 void dvi_framebuf_main_16bpp(struct dvi_inst *inst);
 
-#ifdef __cplusplus
+#if defined(__cplusplus)
 }
 #endif
 

software/libdvi/dvi_config_defs.h

@@ -126,7 +126,7 @@
 // If 1, reverse the order of pixels within each byte. Order of bytes within
 // each word is still little-endian.
 #ifndef DVI_1BPP_BIT_REVERSE
-#define DVI_1BPP_BIT_REVERSE 0
+#define DVI_1BPP_BIT_REVERSE 1 // Adafruit_GFX GFXcanvas1 requires this 1
 #endif
 
 // ----------------------------------------------------------------------------

software/libdvi/dvi_serialiser.pio.h

@@ -0,0 +1,101 @@
+// -------------------------------------------------- //
+// This file is autogenerated by pioasm; do not edit! //
+// -------------------------------------------------- //
+
+#pragma once
+
+#if !PICO_NO_HARDWARE
+#include "hardware/pio.h"
+#endif
+
+// -------------- //
+// dvi_serialiser //
+// -------------- //
+
+#define dvi_serialiser_wrap_target 0
+#define dvi_serialiser_wrap 1
+
+static const uint16_t dvi_serialiser_program_instructions[] = {
+            //     .wrap_target
+    0x70a1, //  0: out    pc, 1           side 2     
+    0x68a1, //  1: out    pc, 1           side 1     
+            //     .wrap
+};
+
+#if !PICO_NO_HARDWARE
+static const struct pio_program dvi_serialiser_program = {
+    .instructions = dvi_serialiser_program_instructions,
+    .length = 2,
+    .origin = 0,
+};
+
+static inline pio_sm_config dvi_serialiser_program_get_default_config(uint offset) {
+    pio_sm_config c = pio_get_default_sm_config();
+    sm_config_set_wrap(&c, offset + dvi_serialiser_wrap_target, offset + dvi_serialiser_wrap);
+    sm_config_set_sideset(&c, 2, false, false);
+    return c;
+}
+#endif
+
+// -------------------- //
+// dvi_serialiser_debug //
+// -------------------- //
+
+#define dvi_serialiser_debug_wrap_target 0
+#define dvi_serialiser_debug_wrap 11
+
+static const uint16_t dvi_serialiser_debug_program_instructions[] = {
+            //     .wrap_target
+    0x98e0, //  0: pull   ifempty block   side 1     
+    0xb042, //  1: nop                    side 0     
+    0x6001, //  2: out    pins, 1                    
+    0x6001, //  3: out    pins, 1                    
+    0x6001, //  4: out    pins, 1                    
+    0x6001, //  5: out    pins, 1                    
+    0x6001, //  6: out    pins, 1                    
+    0x6001, //  7: out    pins, 1                    
+    0x6001, //  8: out    pins, 1                    
+    0x6001, //  9: out    pins, 1                    
+    0x6001, // 10: out    pins, 1                    
+    0x6001, // 11: out    pins, 1                    
+            //     .wrap
+};
+
+#if !PICO_NO_HARDWARE
+static const struct pio_program dvi_serialiser_debug_program = {
+    .instructions = dvi_serialiser_debug_program_instructions,
+    .length = 12,
+    .origin = -1,
+};
+
+static inline pio_sm_config dvi_serialiser_debug_program_get_default_config(uint offset) {
+    pio_sm_config c = pio_get_default_sm_config();
+    sm_config_set_wrap(&c, offset + dvi_serialiser_debug_wrap_target, offset + dvi_serialiser_debug_wrap);
+    sm_config_set_sideset(&c, 2, true, false);
+    return c;
+}
+
+#include "dvi_config_defs.h"
+static inline void dvi_serialiser_program_init(PIO pio, uint sm, uint offset, uint data_pins, bool debug) {
+    pio_sm_set_pins_with_mask(pio, sm, 2u << data_pins, 3u << data_pins);
+    pio_sm_set_pindirs_with_mask(pio, sm, ~0u, 3u << data_pins);
+    pio_gpio_init(pio, data_pins);
+    pio_gpio_init(pio, data_pins + 1);
+    pio_sm_config c;
+    if (debug) {
+        c = dvi_serialiser_debug_program_get_default_config(offset);
+    }
+    else {
+        c = dvi_serialiser_program_get_default_config(offset);
+    }
+    sm_config_set_sideset_pins(&c, data_pins);
+    if (debug)
+	    sm_config_set_out_pins(&c, data_pins, 1);
+    sm_config_set_out_shift(&c, true, !debug, 10 * DVI_SYMBOLS_PER_WORD);
+    sm_config_set_fifo_join(&c, PIO_FIFO_JOIN_TX);
+    pio_sm_init(pio, sm, offset, &c);
+    pio_sm_set_enabled(pio, sm, false);
+}
+
+#endif
+

software/libdvi/dvi_timing.c

@@ -7,6 +7,8 @@
 //   achievable bit clock from 12 MHz crystal)
 // - Helper functions for generating DMA lists based on these timings
 
+extern bool dvi_monochrome_tmds; // In dvi.c
+
 // Pull into RAM but apply unique section suffix to allow linker GC
 #define __dvi_func(x) __not_in_flash_func(x)
 #define __dvi_const(x) __not_in_flash_func(x)
@@ -330,11 +332,7 @@ void dvi_setup_scanline_for_active(const struct dvi_timing *t, const struct dvi_
 
 void __dvi_func(dvi_update_scanline_data_dma)(const struct dvi_timing *t, const uint32_t *tmdsbuf, struct dvi_scanline_dma_list *l) {
 	for (int i = 0; i < N_TMDS_LANES; ++i) {
-#if DVI_MONOCHROME_TMDS
-		const uint32_t *lane_tmdsbuf = tmdsbuf;
-#else
-		const uint32_t *lane_tmdsbuf = tmdsbuf + i * t->h_active_pixels / DVI_SYMBOLS_PER_WORD;
-#endif
+		const uint32_t *lane_tmdsbuf = dvi_monochrome_tmds ? tmdsbuf : tmdsbuf + i * t->h_active_pixels / DVI_SYMBOLS_PER_WORD;
 		if (i == TMDS_SYNC_LANE)
 			dvi_lane_from_list(l, i)[3].read_addr = lane_tmdsbuf;
 		else

src/PicoDVI.cpp

@@ -0,0 +1,457 @@
+// SPDX-FileCopyrightText: 2023 P Burgess for Adafruit Industries
+//
+// SPDX-License-Identifier: BSD-3-Clause
+
+/*!
+ * @file PicoDVI.cpp
+ *
+ * Arduino-and-Adafruit-GFX wrapper around Luke Wren's PicoDVI library.
+ */
+
+#include "PicoDVI.h"
+#include "libdvi/tmds_encode.h"
+
+// PicoDVI class encapsulates some of the libdvi functionality -------------
+// Subclasses then implement specific display types.
+
+static struct {
+  const dvi_timing &timing;
+  vreg_voltage v;
+  uint16_t width;
+  uint16_t height;
+  uint8_t v_rep;
+} dvispec[] = {
+    {dvi_timing_640x480p_60hz, VREG_VOLTAGE_1_20, 320, 240, 2},
+    {dvi_timing_800x480p_60hz, VREG_VOLTAGE_1_20, 400, 240, 2},
+    {dvi_timing_800x480p_30hz, VREG_VOLTAGE_1_20, 400, 240, 2},
+    {dvi_timing_640x480p_60hz, VREG_VOLTAGE_1_20, 640, 480, 1},
+    {dvi_timing_800x480p_60hz, VREG_VOLTAGE_1_20, 800, 480, 1},
+    {dvi_timing_800x480p_30hz, VREG_VOLTAGE_1_20, 800, 480, 1},
+    {dvi_timing_640x480p_60hz, VREG_VOLTAGE_1_20, 640, 240, 2},
+    {dvi_timing_800x480p_60hz, VREG_VOLTAGE_1_20, 800, 240, 2},
+    {dvi_timing_800x480p_30hz, VREG_VOLTAGE_1_20, 800, 240, 2},
+    // Additional resolutions might get added here if the overclock issue can
+    // be sorted out. Regardless, always keep this list 1:1 in sync with the
+    // DVIresolution enum in PicoDVI.h.
+    {dvi_timing_1280x720p_30hz, VREG_VOLTAGE_1_30, 1280, 720, 1},
+};
+
+static PicoDVI *dviptr = NULL; // For C access to active C++ object
+
+/*!
+  @brief  Runs on core 1 on startup; this is how Philhower RP2040 handles
+          multiprocessing.
+*/
+void setup1(void) {
+  while (dviptr == NULL) // Wait for PicoDVI::begin() to start on core 0
+    yield();
+  dviptr->_setup();
+}
+
+// Runs on core 1 after dviptr set
+void PicoDVI::_setup(void) {
+  while (wait_begin)
+    ; // Wait for DVIGFX*::begin() to set this
+  dvi_register_irqs_this_core(&dvi0, DMA_IRQ_0);
+  dvi_start(&dvi0);
+
+  // This is disabled for now, for a transitional period while folks
+  // might be mixing in older versions of arduino-pico, Adafruit_CPFS
+  // and/or Adafruit_SPIFlash. Once is seems pretty solid that everyone's
+  // using current versions that correctly locate PicoDVI functions in
+  // RAM, this can be activated. Not urgent but a nice-to-have.
+#if 0
+  // Borrowed from CircuitPython: turn off core 1 flash access. Any attempt
+  // after this will hard fault (not good, but preferable to the alternative
+  // which would corrupt the CIRCUITPY flash filesystem if mounted). Linker
+  // file MUST place the following files in RAM or library won't work:
+  // *interp.c.obj *divider.S.obj *PicoDVI.cpp.o *dvi.c.o
+  // (Earle Philhower arduino-pico >= 3.3.0 does this now.)
+  MPU->CTRL = MPU_CTRL_PRIVDEFENA_Msk | MPU_CTRL_ENABLE_Msk;
+  MPU->RNR  = 6;
+  MPU->RBAR = XIP_MAIN_BASE | MPU_RBAR_VALID_Msk;
+  MPU->RASR = MPU_RASR_XN_Msk | MPU_RASR_ENABLE_Msk |
+              (0x1b << MPU_RASR_SIZE_Pos);
+  MPU->RNR  = 7;
+#endif
+
+  (*mainloop)(&dvi0);
+}
+
+PicoDVI::PicoDVI(const struct dvi_timing &t, const struct dvi_serialiser_cfg &c,
+                 vreg_voltage v)
+    : voltage(v) {
+  dvi0.timing = &t;
+  memcpy(&dvi0.ser_cfg, &c, sizeof dvi0.ser_cfg);
+};
+
+PicoDVI::~PicoDVI(void) { dviptr = NULL; }
+
+void PicoDVI::begin(void) {
+  dviptr = this;
+  vreg_set_voltage(voltage);
+  delay(10);
+  set_sys_clock_khz(dvi0.timing->bit_clk_khz, true); // Run at TMDS bit clock
+  dvi_init(&dvi0, next_striped_spin_lock_num(), next_striped_spin_lock_num());
+}
+
+// DVIGFX16 class provides GFX-compatible 16-bit color framebuffer ---------
+
+static void *gfxptr = NULL; // For C access to active C++ object
+
+DVIGFX16::DVIGFX16(const DVIresolution r, const struct dvi_serialiser_cfg &c,
+                   vreg_voltage v)
+    : PicoDVI(dvispec[r].timing, c, v),
+      GFXcanvas16(dvispec[r].width, dvispec[r].height) {
+  dvi_vertical_repeat = dvispec[r].v_rep;
+  dvi_monochrome_tmds = false;
+}
+
+DVIGFX16::~DVIGFX16(void) { gfxptr = NULL; }
+
+static void scanline_callback_GFX16(void) {
+  ((DVIGFX16 *)gfxptr)->_scanline_callback();
+}
+
+void DVIGFX16::_scanline_callback(void) {
+  // Discard any scanline pointers passed back
+  uint16_t *bufptr;
+  while (queue_try_remove_u32(&dvi0.q_colour_free, &bufptr))
+    ;
+  bufptr = &getBuffer()[WIDTH * scanline];
+  queue_add_blocking_u32(&dvi0.q_colour_valid, &bufptr);
+  scanline = (scanline + 1) % HEIGHT;
+}
+
+bool DVIGFX16::begin(void) {
+  uint16_t *bufptr = getBuffer();
+  if ((bufptr)) {
+    gfxptr = this;
+    mainloop = dvi_scanbuf_main_16bpp; // in libdvi
+    dvi0.scanline_callback = scanline_callback_GFX16;
+    PicoDVI::begin();
+    queue_add_blocking_u32(&dvi0.q_colour_valid, &bufptr);
+    bufptr += WIDTH;
+    queue_add_blocking_u32(&dvi0.q_colour_valid, &bufptr);
+    wait_begin = false; // Set core 1 in motion
+    return true;
+  }
+  return false;
+}
+
+// DVIGFX8 (8-bit, color-indexed framebuffer) is all manner of dirty pool.
+// PicoDVI seems to have some palette support but I couldn't grasp the DMA
+// stuff going on, so just doing a brute force thing here for now: in
+// addition to the 8-bit framebuffer, two 16-bit (RGB565) scanlines are
+// allocated...then, in the scanline callback, pixels are mapped from the
+// 8-bit framebuffer through the palette into one of these buffers, allowing
+// use of the same dvi_scanbuf_main_16bpp handler as DVIGFX16 above. Not
+// optimal, sure...but not pessimal either. The allocation of those 16-bit
+// scanlines is weird(tm) though. Rather than a separate malloc (which
+// creates a nasty can of worms if that fails after a successful framebuffer
+// allocation...unlikely but not impossible), the framebuffer size is
+// tweaked so that W*H is always an even number, plus 4 extra rows are
+// added: thus two 16-bit scanlines, word-aligned. That extra memory is for
+// us, but allocated by GFX as part of the framebuffer all at once. The
+// HEIGHT value is de-tweaked to the original value so clipping won't allow
+// any drawing operations to spill into the 16-bit scanlines.
+// This requires latest Adafruit_GFX as double-buffered mode plays games with
+// the canvas pointer, wasn't possible until that was made protected (vs
+// private). Drawing and palette-setting operations ONLY apply to the "back"
+// state. Call swap() to switch the front/back buffers at the next vertical
+// sync, for flicker-free and tear-free animation.
+
+DVIGFX8::DVIGFX8(const DVIresolution r, const bool d,
+                 const struct dvi_serialiser_cfg &c, vreg_voltage v)
+    : PicoDVI(dvispec[r].timing, c, v),
+      GFXcanvas8(
+          dvispec[r].width,
+          (d ? (dvispec[r].height * 2) : ((dvispec[r].height + 1) & ~1)) + 4),
+      dbuf(d) {
+  HEIGHT = _height = dvispec[r].height;
+  buffer_save = buffer;
+  dvi_vertical_repeat = dvispec[r].v_rep;
+  dvi_monochrome_tmds = false;
+}
+
+DVIGFX8::~DVIGFX8(void) {
+  buffer = buffer_save; // Restore pointer so canvas destructor works
+  gfxptr = NULL;
+}
+
+static void __not_in_flash_func(scanline_callback_GFX8)(void) {
+  ((DVIGFX8 *)gfxptr)->_scanline_callback();
+}
+
+void __not_in_flash_func(DVIGFX8::_scanline_callback)(void) {
+  uint16_t *b16;
+  while (queue_try_remove_u32(&dvi0.q_colour_free, &b16))
+    ;                   // Discard returned pointer(s)
+  b16 = row565[rowidx]; // Next row to send
+  queue_add_blocking_u32(&dvi0.q_colour_valid, &b16); // Send it
+
+  if (++scanline >= HEIGHT) {      // Next scanline...end of screen reached?
+    if (swap_wait) {               // Swap buffers?
+      back_index = 1 - back_index; // Yes plz
+      buffer = buffer_save + WIDTH * HEIGHT * back_index;
+      swap_wait = 0;
+    }
+    scanline = 0;
+  }
+
+  // Refresh from front buffer
+  uint8_t *b8 = buffer_save + WIDTH * scanline; // New src
+  if (dbuf)
+    b8 += WIDTH * HEIGHT * (1 - back_index);
+  rowidx = (rowidx + 1) & 1; // Swap row565[] bufs
+  b16 = row565[rowidx];      // New dest
+  uint16_t *p16 = dbuf ? palette[1 - back_index] : palette[0];
+  for (int i = 0; i < WIDTH; i++)
+    b16[i] = p16[b8[i]];
+}
+
+bool DVIGFX8::begin(void) {
+  uint8_t *bufptr = getBuffer();
+  if ((bufptr)) {
+    gfxptr = this;
+    row565[0] = (uint16_t *)&bufptr[dbuf ? WIDTH * HEIGHT * 2
+                                         : (WIDTH * HEIGHT + 1) & ~1];
+    row565[1] = row565[0] + WIDTH;
+    memset(palette, 0, sizeof palette);
+    // mainloop = mainloop8;
+    mainloop = dvi_scanbuf_main_16bpp; // in libdvi
+    dvi0.scanline_callback = scanline_callback_GFX8;
+    PicoDVI::begin();
+    // No need to initialize the row565 buffer contents as that memory is
+    // cleared on canvas alloc, and the initial palette state is also all 0.
+    uint16_t *b16 = row565[0];
+    queue_add_blocking_u32(&dvi0.q_colour_valid, &b16);
+    b16 = row565[1];
+    queue_add_blocking_u32(&dvi0.q_colour_valid, &b16);
+    wait_begin = false; // Set core 1 in motion
+    return true;
+  }
+  return false;
+}
+
+void DVIGFX8::swap(bool copy_framebuffer, bool copy_palette) {
+  if (dbuf) {
+    // Request buffer swap at next frame end, wait for it to happen.
+    for (swap_wait = 1; swap_wait;)
+      ;
+
+    if ((copy_framebuffer)) {
+      uint32_t bufsize = WIDTH * HEIGHT;
+      memcpy(buffer_save + bufsize * back_index,
+             buffer_save + bufsize * (1 - back_index), bufsize);
+    }
+
+    if ((copy_palette)) {
+      memcpy(palette[back_index], palette[1 - back_index], sizeof(palette[0]));
+    }
+  }
+}
+
+DVIGFX1::DVIGFX1(const DVIresolution r, const bool d,
+                 const struct dvi_serialiser_cfg &c, vreg_voltage v)
+    : PicoDVI(dvispec[r].timing, c, v),
+      GFXcanvas1(dvispec[r].width,
+                 d ? (dvispec[r].height * 2) : dvispec[r].height),
+      dbuf(d) {
+  dvi_vertical_repeat = dvispec[r].v_rep;
+  dvi_monochrome_tmds = true;
+  HEIGHT = _height = dvispec[r].height;
+  buffer_save = buffer;
+}
+
+DVIGFX1::~DVIGFX1(void) {
+  buffer = buffer_save; // Restore pointer so canvas destructor works
+  gfxptr = NULL;
+}
+
+static void mainloop1(struct dvi_inst *inst) {
+  (void)inst; // Avoid compiler warning
+  ((DVIGFX1 *)gfxptr)->_mainloop();
+}
+
+void __not_in_flash_func(DVIGFX1::_mainloop)(void) {
+  for (;;) {
+    uint8_t *b8 = buffer_save;
+    if (dbuf)
+      b8 += ((WIDTH + 7) / 8) * HEIGHT * (1 - back_index);
+    for (int y = 0; y < HEIGHT; y++) {
+      const uint32_t *colourbuf =
+          (const uint32_t *)(b8 + y * ((WIDTH + 7) / 8));
+      uint32_t *tmdsbuf;
+      queue_remove_blocking_u32(&dvi0.q_tmds_free, &tmdsbuf);
+      tmds_encode_1bpp(colourbuf, tmdsbuf, WIDTH);
+      queue_add_blocking_u32(&dvi0.q_tmds_valid, &tmdsbuf);
+    }
+    if (swap_wait) {               // Swap buffers?
+      back_index = 1 - back_index; // Yes plz
+      buffer = buffer_save + ((WIDTH + 7) / 8) * HEIGHT * back_index;
+      swap_wait = 0;
+    }
+  }
+}
+
+bool DVIGFX1::begin(void) {
+  if ((getBuffer())) {
+    gfxptr = this;
+    mainloop = mainloop1;
+    PicoDVI::begin();
+    wait_begin = false; // Set core 1 in motion
+    return true;
+  }
+  return false;
+}
+
+void DVIGFX1::swap(bool copy_framebuffer) {
+  if (dbuf) {
+    // Request buffer swap at next frame end, wait for it to happen.
+    for (swap_wait = 1; swap_wait;)
+      ;
+
+    if ((copy_framebuffer)) {
+      uint32_t bufsize = ((WIDTH + 7) / 8) * HEIGHT;
+      memcpy(buffer_save + bufsize * back_index,
+             buffer_save + bufsize * (1 - back_index), bufsize);
+    }
+  }
+}
+
+// TEXT MODE IS A WORK IN PROGRESS
+
+#define FONT_CHAR_WIDTH 8  ///< Character block width, pixels
+#define FONT_CHAR_HEIGHT 8 ///< Character block height, pixels
+#define FONT_N_CHARS 256   ///< Number of symbols in font
+#define FONT_FIRST_ASCII 0 ///< Full CP437 is there, start at 0
+#include "font_8x8.h"
+
+DVItext1::DVItext1(const DVIresolution r, const struct dvi_serialiser_cfg &c,
+                   vreg_voltage v)
+    : PicoDVI(dvispec[r].timing, c, v),
+      GFXcanvas16(dvispec[r].width / 8, dvispec[r].height / 8) {
+  dvi_vertical_repeat = dvispec[r].v_rep;
+  dvi_monochrome_tmds = true;
+}
+
+DVItext1::~DVItext1(void) { gfxptr = NULL; }
+
+// Character framebuffer is actually a small GFXcanvas16, so...
+size_t DVItext1::write(uint8_t c) {
+  if (c == '\r') { // Carriage return
+    cursor_x = 0;
+  } else if ((c == '\n') || (cursor_x >= WIDTH)) { // Newline OR right edge
+    cursor_x = 0;
+    if (cursor_y >= (HEIGHT - 1)) { // Vert scroll?
+      memmove(getBuffer(), getBuffer() + WIDTH, WIDTH * (HEIGHT - 1) * 2);
+      drawFastHLine(0, HEIGHT - 1, WIDTH, ' '); // Clear bottom line
+      cursor_y = HEIGHT - 1;
+    } else {
+      cursor_y++;
+    }
+  }
+  if ((c != '\r') && (c != '\n')) {
+    drawPixel(cursor_x, cursor_y, c);
+    cursor_x++;
+  }
+  return 1;
+}
+
+// TO DO: alloc this dynamically as part of object (maybe part of canvas)
+static uint8_t scanbuf[1280 / 8] __attribute__((aligned(4)));
+
+#ifdef TERM_USE_INTERRUPT
+
+void inline __not_in_flash_func(DVItext1::_prepare_scanline)(uint16_t y) {
+  uint16_t *row = getBuffer() + (y / FONT_CHAR_HEIGHT) * WIDTH;
+  uint32_t offset = (y & 7) * FONT_N_CHARS;
+
+  // Blit font into 1bpp scanline buffer, then encode scanbuf into tmdsbuf
+  for (uint16_t x = 0; x < WIDTH; x++) {
+    // Though this only handles 8-bit character output (e.g. ASCII), the
+    // character buffer uses 16-bit cells. The LOWER BYTE contains the 8-bit
+    // character value, while (for now) the UPPER BYTE is either 0 (display
+    // normally) or 255 (inverse). Eventually the upper byte might be
+    // switched to bitfields, e.g. bit 0 = inverse, bit 1 = blink, maybe an
+    // underscore cursor or something, etc. This is slightly wasteful, but
+    // greatly simplifies keeping character values and attributes always
+    // moving together (and the display buffer really isn't THAT big).
+    uint8_t mask = row[x] >> 8;
+    uint8_t c = (row[x] & 255) - FONT_FIRST_ASCII;
+    scanbuf[x] = font_8x8[offset + c] ^ mask;
+  }
+  uint32_t *tmdsbuf;
+  queue_remove_blocking_u32(&dvi0.q_tmds_free, &tmdsbuf);
+  tmds_encode_1bpp((const uint32_t *)scanbuf, tmdsbuf, WIDTH * 8);
+  queue_add_blocking_u32(&dvi0.q_tmds_valid, &tmdsbuf);
+}
+
+void __not_in_flash_func(text1_scanline_callback)(void) {
+  static uint y = 1;
+  ((DVItext1 *)gfxptr)->_prepare_scanline(y);
+  y = (y + 1) % (((DVItext1 *)gfxptr)->height() * 8);
+}
+
+static void mainlooptext1(struct dvi_inst *inst) {
+  // Idle func, everything happens in interrupt
+  for (;;)
+    delay(1000);
+}
+
+#else
+
+// Old way, without interrupt
+// This is a little simpler and might stick with it
+// since nothing important to do in idle func above.
+
+static void __not_in_flash_func(mainlooptext1)(struct dvi_inst *inst) {
+  (void)inst; // Avoid compiler warning
+  ((DVItext1 *)gfxptr)->_mainloop();
+}
+
+void __not_in_flash_func(DVItext1::_mainloop)(void) {
+  for (;;) {
+    for (uint16_t y = 0; y < HEIGHT; y++) {
+      uint16_t *row = getBuffer() + y * WIDTH;
+      for (uint8_t y1 = 0; y1 < 8; y1++) {
+        uint32_t offset = y1 * FONT_N_CHARS;
+        for (uint16_t x = 0; x < WIDTH; x++) {
+          // See notes above
+          uint8_t mask = row[x] >> 8;
+          uint8_t c = (row[x] & 255) - FONT_FIRST_ASCII;
+          scanbuf[x] = font_8x8[offset + c] ^ mask;
+        }
+        uint32_t *tmdsbuf;
+        queue_remove_blocking_u32(&dvi0.q_tmds_free, &tmdsbuf);
+        tmds_encode_1bpp((const uint32_t *)scanbuf, tmdsbuf, WIDTH * 8);
+        queue_add_blocking_u32(&dvi0.q_tmds_valid, &tmdsbuf);
+      }
+    }
+  }
+}
+
+#endif // end TERM_USE_INTERRUPT
+
+bool DVItext1::begin(void) {
+  if ((getBuffer())) {
+    fillScreen(' ');
+    gfxptr = this;
+#ifdef TERM_USE_INTERRUPT
+    dvi0.scanline_callback = text1_scanline_callback;
+#endif
+    mainloop = mainlooptext1;
+    PicoDVI::begin();
+
+    // Must do this AFTER begin because tmdsbuf (accessed in func)
+    // doesn't exist yet until dvi_init (in begin) is called.
+#ifdef TERM_USE_INTERRUPT
+    _prepare_scanline(0);
+#endif
+
+    wait_begin = false; // Set core 1 in motion
+    return true;
+  }
+  return false;
+}

src/PicoDVI.h

@@ -0,0 +1,346 @@
+// SPDX-FileCopyrightText: 2023 P Burgess for Adafruit Industries
+//
+// SPDX-License-Identifier: BSD-3-Clause
+
+/*!
+ * @file PicoDVI.h
+ *
+ * Arduino-and-Adafruit-GFX wrapper around Luke Wren's PicoDVI library.
+ */
+
+#pragma once
+
+#include "../software/include/common_dvi_pin_configs.h"
+#include "hardware/vreg.h" // In Pico SDK
+#include "libdvi/dvi.h"
+#include "pico/stdlib.h" // In Pico SDK
+#include <Adafruit_GFX.h>
+
+/** A set list of resolutions keeps things manageable for users,
+    avoids some possibly-incompatible argument combos to constructors. */
+enum DVIresolution {
+  DVI_RES_320x240p60 = 0,
+  DVI_RES_400x240p60,
+  DVI_RES_400x240p30, // Reduced refresh rate, less overclock required
+  DVI_RES_640x480p60,
+  DVI_RES_800x480p60,
+  DVI_RES_800x480p30, // Reduced refresh rate, less overclock required
+  DVI_RES_640x240p60, // "Tall" pixels, e.g. for 80-column text mode
+  DVI_RES_800x240p60, // Same, 100-column
+  DVI_RES_800x240p30, // Reduced refresh rate, less overclock required
+  DVI_RES_1280x720p30 // Experimenting, not working, plz don't use
+};
+
+extern uint8_t dvi_vertical_repeat; ///< In libdvi/dvi.c
+extern bool dvi_monochrome_tmds;    ///< In libdvi/dvi.c
+
+/*!
+  @brief  A base class used by some of the framebuffer classes that follow.
+          Wraps essential functionality of the underlying PicoDVI library.
+          Maybe only rarely (if ever) needed in Arduino sketch code, but
+          it's here if a project requires working at a lower level to
+          provide features not present in the framebuffer modes.
+*/
+class PicoDVI {
+public:
+  /*!
+    @brief  PicoDVI constructor. Instantiates a new PicoDVI object (must
+            follow with a begin() call to alloc buffers and init hardware).
+    @param  t  Pointer to dvi_timing struct, one of the supported
+               resolutions declared in libdvi/dvi_timing.c. Default if not
+               specified is 640x480p60.
+    @param  c  Pointer to dvi_serialiser_cfg struct, which defines pinout.
+               These are in ../software/include/common_dvi_pin_configs.h.
+               Default if not specified is adafruit_feather_dvi_cfg.
+    @param  v  Core voltage; higher resolution modes require faster
+               overclocking and potentially higher voltages. This is an
+               enumeration declared in Pico SDK hardware/vreg.h. Normal
+               system operation would be 1.1V, default here if not
+               specified is 1.2V.
+  */
+  PicoDVI(const struct dvi_timing &t = dvi_timing_640x480p_60hz,
+          const struct dvi_serialiser_cfg &c = adafruit_feather_dvi_cfg,
+          vreg_voltage v = VREG_VOLTAGE_1_20);
+  ~PicoDVI(void);
+  /*!
+    @brief  Internal function used during DVI initialization. User code
+            should not touch this, but it does need to remain public for
+            C init code to access protected elements.
+  */
+  void _setup(void);
+
+protected:
+  /*!
+    @brief  Initialize hardware and start video thread.
+  */
+  void begin(void);
+  vreg_voltage voltage;                ///< Core voltage
+  struct dvi_inst dvi0;                ///< DVI instance
+  void (*mainloop)(dvi_inst *) = NULL; ///< Video mode loop handler
+  volatile bool wait_begin = true;     ///< For synchronizing cores
+};
+
+/*!
+  @brief  DVIGFX16 provides a 16-bit color (RGB565) Adafruit_GFX-compatible
+          framebuffer. Double-buffering is NOT provided even as an option,
+          there simply isn't adequate RAM.
+*/
+class DVIGFX16 : public PicoDVI, public GFXcanvas16 {
+public:
+  /*!
+    @brief  DVIGFX16 constructor.
+    @param  res   Video mode, from the DVIresolution enumeration. This will
+                  usually be 320x240 or 400x240, there just isn't RAM to
+                  handle anything larger as a pixel-accessible framebuffer.
+    @param  c     Pointer to dvi_serialiser_cfg struct, which defines pinout.
+                  These are in ../software/include/common_dvi_pin_configs.h.
+                  Default if not specified is adafruit_feather_dvi_cfg.
+    @param  v     Core voltage; higher resolution modes require faster
+                  overclocking and potentially higher voltages. This is an
+                  enumeration declared in Pico SDK hardware/vreg.h. Normal
+                  system operation would be 1.1V, default here if not
+                  specified is 1.2V.
+  */
+  DVIGFX16(const DVIresolution res = DVI_RES_320x240p60,
+           const struct dvi_serialiser_cfg &c = adafruit_feather_dvi_cfg,
+           vreg_voltage v = VREG_VOLTAGE_1_20);
+  ~DVIGFX16(void);
+  /*!
+    @brief   Initialize DVI output for 16-bit color display.
+    @return  true on success, false otherwise (buffer allocation usually).
+  */
+  bool begin(void);
+  /*!
+    @brief   Convert a 24-bit RGB color (as three 8-bit values) to a packed
+             16-bit "RGB565" color, as native to PicoDVI and Adafruit_GFX.
+    @param   red    Red component, 0-255.
+    @param   green  Green component, 0-255.
+    @param   blue   Blue component, 0-255.
+    @return  Nearest 16-bit RGB565 equivalent.
+  */
+  uint16_t color565(uint8_t red, uint8_t green, uint8_t blue) {
+    return ((red & 0xF8) << 8) | ((green & 0xFC) << 3) | (blue >> 3);
+  }
+  /*!
+    @brief  Scanline-processing function used internally during video
+            generation. User sketch code should never call this, but it
+            needs to be public so that C code can access it.
+  */
+  void _scanline_callback(void);
+
+protected:
+  uint16_t scanline = 2; ///< First 2 scanlines are set up before DVI start
+};
+
+/*!
+  @brief  DVIGFX8 provides a 8-bit color-mapped Adafruit_GFX-compatible
+          framebuffer, with optional double-buffering.
+*/
+class DVIGFX8 : public PicoDVI, public GFXcanvas8 {
+public:
+  /*!
+    @brief  DVIGFX8 constructor.
+    @param  res   Video mode, from the DVIresolution enumeration.
+    @param  dbuf  Double-buffered flag. If true, two 8-bit framebuffers
+                  (each with associated color palette) are allocated, code
+                  can alternate between these for flicker-free animation.
+    @param  c     Pointer to dvi_serialiser_cfg struct, which defines pinout.
+                  These are in ../software/include/common_dvi_pin_configs.h.
+                  Default if not specified is adafruit_feather_dvi_cfg.
+    @param  v     Core voltage; higher resolution modes require faster
+                  overclocking and potentially higher voltages. This is an
+                  enumeration declared in Pico SDK hardware/vreg.h. Normal
+                  system operation would be 1.1V, default here if not
+                  specified is 1.2V.
+  */
+  DVIGFX8(const DVIresolution res = DVI_RES_320x240p60, const bool dbuf = false,
+          const struct dvi_serialiser_cfg &c = adafruit_feather_dvi_cfg,
+          vreg_voltage v = VREG_VOLTAGE_1_20);
+  ~DVIGFX8(void);
+  /*!
+    @brief   Initialize DVI output for 8-bit colormapped display.
+    @return  true on success, false otherwise (buffer allocation usually).
+  */
+  bool begin(void);
+  /*!
+    @brief   Get base address of 'back' palette (or single palette in
+             single-buffered mode), for code that wants to access this.
+             directly rather than through setColor() and getColor().
+    @return  uint16_t* pointer to 256-entry palette (RGB565 colors).
+  */
+  uint16_t *getPalette(void) { return palette[back_index]; }
+  /*!
+    @brief   Set one color in the 'back' color palette (or single palette in
+             single-buffered mode).
+    @param   idx    Color index to set, 0-255.
+    @param   color  16-bit "RGB565" color.
+  */
+  void setColor(uint8_t idx, uint16_t color) {
+    palette[back_index][idx] = color;
+  }
+  /*!
+    @brief   Set one color in the 'back' color palette (or single palette in
+             single-buffered mode). Accepts three 8-bit values, but COLOR
+             WILL BE QUANTIZED TO 16-BIT RGB565 because that's what PicoDVI
+             uses. Original 24-bit value is LOST and cannot be accurately
+             read back by getColor().
+    @param   idx    Color index to set, 0-255.
+    @param   red    Red component, 0-255.
+    @param   green  Green component, 0-255.
+    @param   blue   Blue component, 0-255.
+  */
+  void setColor(uint8_t idx, uint8_t red, uint8_t green, uint8_t blue) {
+    palette[back_index][idx] =
+        ((red & 0xF8) << 8) | ((green & 0xFC) << 3) | (blue >> 3);
+  }
+  /*!
+    @brief   Read one color from the 'back' color palette (or single palette
+             in single-buffered mode).
+    @param   idx  Color index to read, 0-255.
+    @return  16-bit "RGB565" color.
+  */
+  uint16_t getColor(uint8_t idx) { return palette[back_index][idx]; }
+  /*!
+    @brief  Swap front/back framebuffers on next vertical sync.
+    @param  copy_framebuffer  If true, contents of new 'front' framebuffer
+                              will be duplicated into new 'back' buffer for
+                              updates (useful for code that modifies screen
+                              frame-to-frame, rather than complete regen).
+    @param  copy_palette      If true, copies 'front' palette to 'back'.
+                              This is obscure; most code will set up a global
+                              palette once, swap-with-copy once (so front
+                              and back are both using same palette), and not
+                              touch this...but it's there if palette-
+                              animating code needs it.
+  */
+  void swap(bool copy_framebuffer = false, bool copy_palette = false);
+
+  /*!
+    @brief  Scanline-processing function used internally during video
+            generation. User sketch code should never call this, but it
+            needs to be public so that C code can access it.
+  */
+  void _scanline_callback(void);
+
+protected:
+  uint16_t palette[2][256];    ///< [2] for double-buffering
+  uint16_t *row565[2];         ///< 2 scanlines of 16-bit RGB565 data
+  uint16_t scanline = 2;       ///< 1st 2 scanlines loaded before DVI start
+  uint8_t rowidx = 1;          ///< Alternate 0/1 for which row565[] is active
+  bool dbuf = false;           ///< True if double-buffered
+  uint8_t *buffer_save;        ///< Original canvas buffer pointer
+  uint8_t back_index = 0;      ///< Which of 2 buffers receives draw ops
+  volatile bool swap_wait = 0; ///< For synchronizing fromt/back buffer swap
+};
+
+/*!
+  @brief  DVIGFX1 provides a 1-bit monochrome (black, white) Adafruit_GFX-
+          compatible framebuffer, with optional double-buffering.
+*/
+class DVIGFX1 : public PicoDVI, public GFXcanvas1 {
+public:
+  /*!
+    @brief  DVIGFX1 constructor.
+    @param  res   Video mode, from the DVIresolution enumeration. This will
+                  usually be 640x480 or 800x480. Higher settings are not
+                  currently supported.
+    @param  dbuf  Double-buffered flag. If true, two 1-bit framebuffers are
+                  allocated, code can alternate between these for flicker-
+                  free animation.
+    @param  c     Pointer to dvi_serialiser_cfg struct, which defines pinout.
+                  These are in ../software/include/common_dvi_pin_configs.h.
+                  Default if not specified is adafruit_feather_dvi_cfg.
+    @param  v     Core voltage; higher resolution modes require faster
+                  overclocking and potentially higher voltages. This is an
+                  enumeration declared in Pico SDK hardware/vreg.h. Normal
+                  system operation would be 1.1V, default here if not
+                  specified is 1.2V.
+  */
+  DVIGFX1(const DVIresolution res = DVI_RES_640x480p60, const bool dbuf = false,
+          const struct dvi_serialiser_cfg &c = adafruit_feather_dvi_cfg,
+          vreg_voltage v = VREG_VOLTAGE_1_20);
+  ~DVIGFX1(void);
+  /*!
+    @brief   Initialize DVI output for 1-bit monochrome graphics display.
+    @return  true on success, false otherwise (buffer allocation usually).
+  */
+  bool begin(void);
+  /*!
+    @brief  Swap front/back framebuffers on next vertical sync.
+    @param  copy_framebuffer  If true, contents of new 'front' framebuffer
+                              will be duplicated into new 'back' buffer for
+                              updates (useful for code that modifies screen
+                              frame-to-frame, rather than complete regen).
+  */
+  void swap(bool copy_framebuffer = false);
+
+  /*!
+    @brief  Video main loop handler invoked by underlying PicoDVI code.
+            User sketch code should never call this, but it needs to be
+            public so that C code can access it.
+  */
+  void _mainloop(void);
+
+private:
+  bool dbuf = false;           // True if double-buffered
+  uint8_t *buffer_save;        // Original canvas buffer pointer
+  uint8_t back_index = 0;      // Which of 2 buffers receives draw ops
+  volatile bool swap_wait = 0; // For syncronizing front/back buffer swap
+};
+
+/*!
+  @brief  DVItext1 provides a 1-bit monochrome (black, white) text-only
+          display. print() and related calls to this display will vertical
+          scroll when needed. It's very basic and does not provide any ANSI-
+          like terminal functionality.
+*/
+class DVItext1 : public PicoDVI, public GFXcanvas16 {
+public:
+  /*!
+    @brief  DVItext1 constructor.
+    @param  res  Video mode, from the DVIresolution enumeration. This is the
+                 pixel resolution; text cells are 8x8 pixels and thus the
+                 number of rows and columns will be less (e.g. 640x240 mode
+                 yields 80x30 characters).
+    @param  c    Pointer to dvi_serialiser_cfg struct, which defines pinout.
+                 These are in ../software/include/common_dvi_pin_configs.h.
+                 Default if not specified is adafruit_feather_dvi_cfg.
+    @param  v    Core voltage; higher resolution modes require faster
+                 overclocking and potentially higher voltages. This is an
+                 enumeration declared in Pico SDK hardware/vreg.h. Normal
+                 system operation would be 1.1V, default here if not
+                 specified is 1.2V.
+  */
+  DVItext1(const DVIresolution res = DVI_RES_640x240p60,
+           const struct dvi_serialiser_cfg &c = adafruit_feather_dvi_cfg,
+           vreg_voltage v = VREG_VOLTAGE_1_20);
+  ~DVItext1(void);
+  /*!
+    @brief   Initialize DVI output for text display.
+    @return  true on success, false otherwise (buffer allocation usually).
+  */
+  bool begin(void);
+  /*!
+    @brief   Base character-writing function, allows Arduino print() and
+             println() calls to work with display. Text will scroll upward
+             as needed.
+    @param   c  ASCII character to write.
+    @return  Number of characters written; always 1.
+  */
+  size_t write(uint8_t c);
+  /*!
+    @brief  Scanline-processing function used internally during video
+            generation. User sketch code should never call this, but it
+            needs to be public so that C code can access it.
+    @param  y  Scanline index; 0 to height-1.
+  */
+  void _prepare_scanline(uint16_t y);
+  /*!
+    @brief  Video main loop handler invoked by underlying PicoDVI code.
+            User sketch code should never call this, but it needs to be
+            public so that C code can access it.
+  */
+  void _mainloop(void);
+
+protected:
+};

src/Readme.md

@@ -0,0 +1,5 @@
+Arduino Wrapper for PicoDVI Library
+===================================
+
+This directory does not exist in the original PicoDVI repository.
+It contains files and soft link(s) to provide Arduino-like functionality.

src/font_8x8.h

@@ -0,0 +1,177 @@
+// Adapted from IBM VGA 8x8 font, thx to Stec Dose github.com/spacerace/romfont
+// To simplify the scanline conversion function, font data has been reordered
+// so all the row 0 bytes for each of 256 chars appear first, then all row 1
+// bytes for each, row 2, etc. Individual character data is not contiguous.
+
+uint8_t __not_in_flash_func(font_8x8)[] = {
+    0x00, 0x7E, 0x7E, 0x6C, 0x10, 0x38, 0x10, 0x00, 0xFF, 0x00, 0xFF, 0x0F,
+    0x3C, 0x3F, 0x7F, 0x99, 0x80, 0x02, 0x18, 0x66, 0x7F, 0x3E, 0x00, 0x18,
+    0x18, 0x18, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x30, 0x6C, 0x6C,
+    0x30, 0x00, 0x38, 0x60, 0x18, 0x60, 0x00, 0x00, 0x00, 0x00, 0x00, 0x06,
+    0x7C, 0x30, 0x78, 0x78, 0x1C, 0xFC, 0x38, 0xFC, 0x78, 0x78, 0x00, 0x00,
+    0x18, 0x00, 0x60, 0x78, 0x7C, 0x30, 0xFC, 0x3C, 0xF8, 0xFE, 0xFE, 0x3C,
+    0xCC, 0x78, 0x1E, 0xE6, 0xF0, 0xC6, 0xC6, 0x38, 0xFC, 0x78, 0xFC, 0x78,
+    0xFC, 0xCC, 0xCC, 0xC6, 0xC6, 0xCC, 0xFE, 0x78, 0xC0, 0x78, 0x10, 0x00,
+    0x30, 0x00, 0xE0, 0x00, 0x1C, 0x00, 0x38, 0x00, 0xE0, 0x30, 0x0C, 0xE0,
+    0x70, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x1C, 0x18, 0xE0, 0x76, 0x00, 0x78, 0x00, 0x1C, 0x7E,
+    0xCC, 0xE0, 0x30, 0x00, 0x7E, 0xCC, 0xE0, 0xCC, 0x7C, 0xE0, 0xC6, 0x30,
+    0x1C, 0x00, 0x3E, 0x78, 0x00, 0x00, 0x78, 0x00, 0x00, 0xC3, 0xCC, 0x18,
+    0x38, 0xCC, 0xF8, 0x0E, 0x1C, 0x38, 0x00, 0x00, 0x00, 0xFC, 0x3C, 0x38,
+    0x30, 0x00, 0x00, 0xC3, 0xC3, 0x18, 0x00, 0x00, 0x22, 0x55, 0xDB, 0x18,
+    0x18, 0x18, 0x36, 0x00, 0x00, 0x36, 0x36, 0x00, 0x36, 0x36, 0x18, 0x00,
+    0x18, 0x18, 0x00, 0x18, 0x00, 0x18, 0x18, 0x36, 0x36, 0x00, 0x36, 0x00,
+    0x36, 0x00, 0x36, 0x18, 0x36, 0x00, 0x00, 0x36, 0x18, 0x00, 0x00, 0x36,
+    0x18, 0x18, 0x00, 0xFF, 0x00, 0xF0, 0x0F, 0xFF, 0x00, 0x00, 0x00, 0x00,
+    0xFC, 0x00, 0x00, 0x00, 0xFC, 0x38, 0x38, 0x1C, 0x00, 0x06, 0x38, 0x78,
+    0x00, 0x30, 0x60, 0x18, 0x0E, 0x18, 0x30, 0x00, 0x38, 0x00, 0x00, 0x0F,
+    0x78, 0x70, 0x00, 0x00, 0x00, 0x81, 0xFF, 0xFE, 0x38, 0x7C, 0x10, 0x00,
+    0xFF, 0x3C, 0xC3, 0x07, 0x66, 0x33, 0x63, 0x5A, 0xE0, 0x0E, 0x3C, 0x66,
+    0xDB, 0x63, 0x00, 0x3C, 0x3C, 0x18, 0x18, 0x30, 0x00, 0x24, 0x18, 0xFF,
+    0x00, 0x78, 0x6C, 0x6C, 0x7C, 0xC6, 0x6C, 0x60, 0x30, 0x30, 0x66, 0x30,
+    0x00, 0x00, 0x00, 0x0C, 0xC6, 0x70, 0xCC, 0xCC, 0x3C, 0xC0, 0x60, 0xCC,
+    0xCC, 0xCC, 0x30, 0x30, 0x30, 0x00, 0x30, 0xCC, 0xC6, 0x78, 0x66, 0x66,
+    0x6C, 0x62, 0x62, 0x66, 0xCC, 0x30, 0x0C, 0x66, 0x60, 0xEE, 0xE6, 0x6C,
+    0x66, 0xCC, 0x66, 0xCC, 0xB4, 0xCC, 0xCC, 0xC6, 0xC6, 0xCC, 0xC6, 0x60,
+    0x60, 0x18, 0x38, 0x00, 0x30, 0x00, 0x60, 0x00, 0x0C, 0x00, 0x6C, 0x00,
+    0x60, 0x00, 0x00, 0x60, 0x30, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x30, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x30, 0x18, 0x30, 0xDC, 0x10,
+    0xCC, 0xCC, 0x00, 0xC3, 0x00, 0x00, 0x30, 0x00, 0xC3, 0x00, 0x00, 0x00,
+    0xC6, 0x00, 0x38, 0x30, 0x00, 0x00, 0x6C, 0xCC, 0xCC, 0xE0, 0xCC, 0xE0,
+    0xCC, 0x18, 0x00, 0x18, 0x6C, 0xCC, 0xCC, 0x1B, 0x00, 0x00, 0x1C, 0x1C,
+    0xF8, 0x00, 0x6C, 0x6C, 0x00, 0x00, 0x00, 0xC6, 0xC6, 0x18, 0x33, 0xCC,
+    0x88, 0xAA, 0x77, 0x18, 0x18, 0x18, 0x36, 0x00, 0x00, 0x36, 0x36, 0x00,
+    0x36, 0x36, 0x18, 0x00, 0x18, 0x18, 0x00, 0x18, 0x00, 0x18, 0x18, 0x36,
+    0x36, 0x00, 0x36, 0x00, 0x36, 0x00, 0x36, 0x18, 0x36, 0x00, 0x00, 0x36,
+    0x18, 0x00, 0x00, 0x36, 0x18, 0x18, 0x00, 0xFF, 0x00, 0xF0, 0x0F, 0xFF,
+    0x00, 0x78, 0xFC, 0xFE, 0xCC, 0x00, 0x66, 0x76, 0x30, 0x6C, 0x6C, 0x30,
+    0x00, 0x0C, 0x60, 0xCC, 0xFC, 0x30, 0x30, 0x30, 0x1B, 0x18, 0x30, 0x76,
+    0x6C, 0x00, 0x00, 0x0C, 0x6C, 0x18, 0x00, 0x00, 0x00, 0xA5, 0xDB, 0xFE,
+    0x7C, 0x38, 0x38, 0x18, 0xE7, 0x66, 0x99, 0x0F, 0x66, 0x3F, 0x7F, 0x3C,
+    0xF8, 0x3E, 0x7E, 0x66, 0xDB, 0x38, 0x00, 0x7E, 0x7E, 0x18, 0x0C, 0x60,
+    0xC0, 0x66, 0x3C, 0xFF, 0x00, 0x78, 0x6C, 0xFE, 0xC0, 0xCC, 0x38, 0xC0,
+    0x60, 0x18, 0x3C, 0x30, 0x00, 0x00, 0x00, 0x18, 0xCE, 0x30, 0x0C, 0x0C,
+    0x6C, 0xF8, 0xC0, 0x0C, 0xCC, 0xCC, 0x30, 0x30, 0x60, 0xFC, 0x18, 0x0C,
+    0xDE, 0xCC, 0x66, 0xC0, 0x66, 0x68, 0x68, 0xC0, 0xCC, 0x30, 0x0C, 0x6C,
+    0x60, 0xFE, 0xF6, 0xC6, 0x66, 0xCC, 0x66, 0xE0, 0x30, 0xCC, 0xCC, 0xC6,
+    0x6C, 0xCC, 0x8C, 0x60, 0x30, 0x18, 0x6C, 0x00, 0x18, 0x78, 0x60, 0x78,
+    0x0C, 0x78, 0x60, 0x76, 0x6C, 0x70, 0x0C, 0x66, 0x30, 0xCC, 0xF8, 0x78,
+    0xDC, 0x76, 0xDC, 0x7C, 0x7C, 0xCC, 0xCC, 0xC6, 0xC6, 0xCC, 0xFC, 0x30,
+    0x18, 0x30, 0x00, 0x38, 0xC0, 0x00, 0x78, 0x3C, 0x78, 0x78, 0x78, 0x78,
+    0x3C, 0x78, 0x78, 0x70, 0x38, 0x70, 0x6C, 0x00, 0xFC, 0x7F, 0xCC, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x3C, 0xCC, 0x7E, 0x64, 0x78, 0xCC, 0x18,
+    0x78, 0x70, 0x00, 0x00, 0x00, 0xCC, 0x6C, 0x6C, 0x30, 0x00, 0x00, 0xCC,
+    0xCC, 0x00, 0x66, 0x66, 0x22, 0x55, 0xDB, 0x18, 0x18, 0xF8, 0x36, 0x00,
+    0xF8, 0xF6, 0x36, 0xFE, 0xF6, 0x36, 0xF8, 0x00, 0x18, 0x18, 0x00, 0x18,
+    0x00, 0x18, 0x1F, 0x36, 0x37, 0x3F, 0xF7, 0xFF, 0x37, 0xFF, 0xF7, 0xFF,
+    0x36, 0xFF, 0x00, 0x36, 0x1F, 0x1F, 0x00, 0x36, 0xFF, 0x18, 0x00, 0xFF,
+    0x00, 0xF0, 0x0F, 0xFF, 0x76, 0xCC, 0xCC, 0x6C, 0x60, 0x7E, 0x66, 0xDC,
+    0x78, 0xC6, 0xC6, 0x18, 0x7E, 0x7E, 0xC0, 0xCC, 0x00, 0xFC, 0x18, 0x60,
+    0x1B, 0x18, 0x00, 0xDC, 0x6C, 0x00, 0x00, 0x0C, 0x6C, 0x30, 0x3C, 0x00,
+    0x00, 0x81, 0xFF, 0xFE, 0xFE, 0xFE, 0x7C, 0x3C, 0xC3, 0x42, 0xBD, 0x7D,
+    0x66, 0x30, 0x63, 0xE7, 0xFE, 0xFE, 0x18, 0x66, 0x7B, 0x6C, 0x00, 0x18,
+    0x18, 0x18, 0xFE, 0xFE, 0xC0, 0xFF, 0x7E, 0x7E, 0x00, 0x30, 0x00, 0x6C,
+    0x78, 0x18, 0x76, 0x00, 0x60, 0x18, 0xFF, 0xFC, 0x00, 0xFC, 0x00, 0x30,
+    0xDE, 0x30, 0x38, 0x38, 0xCC, 0x0C, 0xF8, 0x18, 0x78, 0x7C, 0x00, 0x00,
+    0xC0, 0x00, 0x0C, 0x18, 0xDE, 0xCC, 0x7C, 0xC0, 0x66, 0x78, 0x78, 0xC0,
+    0xFC, 0x30, 0x0C, 0x78, 0x60, 0xFE, 0xDE, 0xC6, 0x7C, 0xCC, 0x7C, 0x70,
+    0x30, 0xCC, 0xCC, 0xD6, 0x38, 0x78, 0x18, 0x60, 0x18, 0x18, 0xC6, 0x00,
+    0x00, 0x0C, 0x7C, 0xCC, 0x7C, 0xCC, 0xF0, 0xCC, 0x76, 0x30, 0x0C, 0x6C,
+    0x30, 0xFE, 0xCC, 0xCC, 0x66, 0xCC, 0x76, 0xC0, 0x30, 0xCC, 0xCC, 0xD6,
+    0x6C, 0xCC, 0x98, 0xE0, 0x00, 0x1C, 0x00, 0x6C, 0xCC, 0xCC, 0xCC, 0x06,
+    0x0C, 0x0C, 0x0C, 0xC0, 0x66, 0xCC, 0xCC, 0x30, 0x18, 0x30, 0xC6, 0x78,
+    0x60, 0x0C, 0xFE, 0x78, 0x78, 0x78, 0xCC, 0xCC, 0xCC, 0x66, 0xCC, 0xC0,
+    0xF0, 0xFC, 0xFA, 0x3C, 0x0C, 0x30, 0x78, 0xCC, 0xF8, 0xEC, 0x3E, 0x38,
+    0x60, 0xFC, 0xFC, 0xDE, 0xDB, 0x18, 0xCC, 0x33, 0x88, 0xAA, 0xEE, 0x18,
+    0x18, 0x18, 0x36, 0x00, 0x18, 0x06, 0x36, 0x06, 0x06, 0x36, 0x18, 0x00,
+    0x18, 0x18, 0x00, 0x18, 0x00, 0x18, 0x18, 0x36, 0x30, 0x30, 0x00, 0x00,
+    0x30, 0x00, 0x00, 0x00, 0x36, 0x00, 0x00, 0x36, 0x18, 0x18, 0x00, 0x36,
+    0x18, 0x18, 0x00, 0xFF, 0x00, 0xF0, 0x0F, 0xFF, 0xDC, 0xF8, 0xC0, 0x6C,
+    0x30, 0xD8, 0x66, 0x18, 0xCC, 0xFE, 0xC6, 0x7C, 0xDB, 0xDB, 0xF8, 0xCC,
+    0xFC, 0x30, 0x30, 0x30, 0x18, 0x18, 0xFC, 0x00, 0x38, 0x18, 0x00, 0x0C,
+    0x6C, 0x60, 0x3C, 0x00, 0x00, 0xBD, 0xC3, 0x7C, 0x7C, 0xFE, 0xFE, 0x3C,
+    0xC3, 0x42, 0xBD, 0xCC, 0x3C, 0x30, 0x63, 0xE7, 0xF8, 0x3E, 0x18, 0x66,
+    0x1B, 0x6C, 0x7E, 0x7E, 0x18, 0x7E, 0x0C, 0x60, 0xC0, 0x66, 0xFF, 0x3C,
+    0x00, 0x30, 0x00, 0xFE, 0x0C, 0x30, 0xDC, 0x00, 0x60, 0x18, 0x3C, 0x30,
+    0x00, 0x00, 0x00, 0x60, 0xF6, 0x30, 0x60, 0x0C, 0xFE, 0x0C, 0xCC, 0x30,
+    0xCC, 0x0C, 0x00, 0x00, 0x60, 0x00, 0x18, 0x30, 0xDE, 0xFC, 0x66, 0xC0,
+    0x66, 0x68, 0x68, 0xCE, 0xCC, 0x30, 0xCC, 0x6C, 0x62, 0xD6, 0xCE, 0xC6,
+    0x60, 0xDC, 0x6C, 0x1C, 0x30, 0xCC, 0xCC, 0xFE, 0x38, 0x30, 0x32, 0x60,
+    0x0C, 0x18, 0x00, 0x00, 0x00, 0x7C, 0x66, 0xC0, 0xCC, 0xFC, 0x60, 0xCC,
+    0x66, 0x30, 0x0C, 0x78, 0x30, 0xFE, 0xCC, 0xCC, 0x66, 0xCC, 0x66, 0x78,
+    0x30, 0xCC, 0xCC, 0xFE, 0x38, 0xCC, 0x30, 0x30, 0x18, 0x30, 0x00, 0xC6,
+    0x78, 0xCC, 0xFC, 0x3E, 0x7C, 0x7C, 0x7C, 0xC0, 0x7E, 0xFC, 0xFC, 0x30,
+    0x18, 0x30, 0xFE, 0xCC, 0x78, 0x7F, 0xCC, 0xCC, 0xCC, 0xCC, 0xCC, 0xCC,
+    0xCC, 0x66, 0xCC, 0xC0, 0x60, 0x30, 0xC6, 0x18, 0x7C, 0x30, 0xCC, 0xCC,
+    0xCC, 0xFC, 0x00, 0x00, 0xC0, 0xC0, 0x0C, 0x33, 0x37, 0x18, 0x66, 0x66,
+    0x22, 0x55, 0xDB, 0x18, 0xF8, 0xF8, 0xF6, 0xFE, 0xF8, 0xF6, 0x36, 0xF6,
+    0xFE, 0xFE, 0xF8, 0xF8, 0x1F, 0xFF, 0xFF, 0x1F, 0xFF, 0xFF, 0x1F, 0x37,
+    0x3F, 0x37, 0xFF, 0xF7, 0x37, 0xFF, 0xF7, 0xFF, 0xFF, 0xFF, 0xFF, 0x3F,
+    0x1F, 0x1F, 0x3F, 0xFF, 0xFF, 0xF8, 0x1F, 0xFF, 0xFF, 0xF0, 0x0F, 0x00,
+    0xC8, 0xCC, 0xC0, 0x6C, 0x60, 0xD8, 0x66, 0x18, 0xCC, 0xC6, 0x6C, 0xCC,
+    0xDB, 0xDB, 0xC0, 0xCC, 0x00, 0x30, 0x60, 0x18, 0x18, 0x18, 0x00, 0x76,
+    0x00, 0x18, 0x18, 0xEC, 0x6C, 0x78, 0x3C, 0x00, 0x00, 0x99, 0xE7, 0x38,
+    0x38, 0x7C, 0x7C, 0x18, 0xE7, 0x66, 0x99, 0xCC, 0x18, 0x70, 0x67, 0x3C,
+    0xE0, 0x0E, 0x7E, 0x00, 0x1B, 0x38, 0x7E, 0x3C, 0x18, 0x3C, 0x18, 0x30,
+    0xFE, 0x24, 0xFF, 0x18, 0x00, 0x00, 0x00, 0x6C, 0xF8, 0x66, 0xCC, 0x00,
+    0x30, 0x30, 0x66, 0x30, 0x30, 0x00, 0x30, 0xC0, 0xE6, 0x30, 0xCC, 0xCC,
+    0x0C, 0xCC, 0xCC, 0x30, 0xCC, 0x18, 0x30, 0x30, 0x30, 0xFC, 0x30, 0x00,
+    0xC0, 0xCC, 0x66, 0x66, 0x6C, 0x62, 0x60, 0x66, 0xCC, 0x30, 0xCC, 0x66,
+    0x66, 0xC6, 0xC6, 0x6C, 0x60, 0x78, 0x66, 0xCC, 0x30, 0xCC, 0x78, 0xEE,
+    0x6C, 0x30, 0x66, 0x60, 0x06, 0x18, 0x00, 0x00, 0x00, 0xCC, 0x66, 0xCC,
+    0xCC, 0xC0, 0x60, 0x7C, 0x66, 0x30, 0xCC, 0x6C, 0x30, 0xD6, 0xCC, 0xCC,
+    0x7C, 0x7C, 0x60, 0x0C, 0x34, 0xCC, 0x78, 0xFE, 0x6C, 0x7C, 0x64, 0x30,
+    0x18, 0x30, 0x00, 0xC6, 0x18, 0xCC, 0xC0, 0x66, 0xCC, 0xCC, 0xCC, 0x78,
+    0x60, 0xC0, 0xC0, 0x30, 0x18, 0x30, 0xC6, 0xFC, 0x60, 0xCC, 0xCC, 0xCC,
+    0xCC, 0xCC, 0xCC, 0xCC, 0x7C, 0x3C, 0xCC, 0x7E, 0xE6, 0xFC, 0xCF, 0x18,
+    0xCC, 0x30, 0xCC, 0xCC, 0xCC, 0xDC, 0x7E, 0x7C, 0xCC, 0xC0, 0x0C, 0x66,
+    0x6F, 0x18, 0x33, 0xCC, 0x88, 0xAA, 0x77, 0x18, 0x18, 0x18, 0x36, 0x36,
+    0x18, 0x36, 0x36, 0x36, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x18, 0x18,
+    0x00, 0x18, 0x18, 0x36, 0x00, 0x36, 0x00, 0x36, 0x36, 0x00, 0x36, 0x00,
+    0x00, 0x18, 0x36, 0x00, 0x00, 0x18, 0x36, 0x36, 0x18, 0x00, 0x18, 0xFF,
+    0xFF, 0xF0, 0x0F, 0x00, 0xDC, 0xF8, 0xC0, 0x6C, 0xCC, 0xD8, 0x7C, 0x18,
+    0x78, 0x6C, 0x6C, 0xCC, 0x7E, 0x7E, 0x60, 0xCC, 0xFC, 0x00, 0x00, 0x00,
+    0x18, 0xD8, 0x30, 0xDC, 0x00, 0x00, 0x00, 0x6C, 0x00, 0x00, 0x3C, 0x00,
+    0x00, 0x81, 0xFF, 0x10, 0x10, 0x38, 0x38, 0x00, 0xFF, 0x3C, 0xC3, 0xCC,
+    0x7E, 0xF0, 0xE6, 0x5A, 0x80, 0x02, 0x3C, 0x66, 0x1B, 0xCC, 0x7E, 0x18,
+    0x18, 0x18, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x30, 0x00, 0x6C,
+    0x30, 0xC6, 0x76, 0x00, 0x18, 0x60, 0x00, 0x00, 0x30, 0x00, 0x30, 0x80,
+    0x7C, 0xFC, 0xFC, 0x78, 0x1E, 0x78, 0x78, 0x30, 0x78, 0x70, 0x30, 0x30,
+    0x18, 0x00, 0x60, 0x30, 0x78, 0xCC, 0xFC, 0x3C, 0xF8, 0xFE, 0xF0, 0x3E,
+    0xCC, 0x78, 0x78, 0xE6, 0xFE, 0xC6, 0xC6, 0x38, 0xF0, 0x1C, 0xE6, 0x78,
+    0x78, 0xFC, 0x30, 0xC6, 0xC6, 0x78, 0xFE, 0x78, 0x02, 0x78, 0x00, 0x00,
+    0x00, 0x76, 0xDC, 0x78, 0x76, 0x78, 0xF0, 0x0C, 0xE6, 0x78, 0xCC, 0xE6,
+    0x78, 0xC6, 0xCC, 0x78, 0x60, 0x0C, 0xF0, 0xF8, 0x18, 0x76, 0x30, 0x6C,
+    0xC6, 0x0C, 0xFC, 0x1C, 0x18, 0xE0, 0x00, 0xFE, 0x0C, 0x7E, 0x78, 0x3F,
+    0x7E, 0x7E, 0x7E, 0x0C, 0x3C, 0x78, 0x78, 0x78, 0x3C, 0x78, 0xC6, 0xCC,
+    0xFC, 0x7F, 0xCE, 0x78, 0x78, 0x78, 0x7E, 0x7E, 0x0C, 0x18, 0x78, 0x18,
+    0xFC, 0x30, 0xC6, 0xD8, 0x7E, 0x78, 0x78, 0x7E, 0xCC, 0xCC, 0x00, 0x00,
+    0x78, 0x00, 0x00, 0xCC, 0xCF, 0x18, 0x00, 0x00, 0x22, 0x55, 0xDB, 0x18,
+    0x18, 0x18, 0x36, 0x36, 0x18, 0x36, 0x36, 0x36, 0x00, 0x00, 0x00, 0x18,
+    0x00, 0x00, 0x18, 0x18, 0x00, 0x18, 0x18, 0x36, 0x00, 0x36, 0x00, 0x36,
+    0x36, 0x00, 0x36, 0x00, 0x00, 0x18, 0x36, 0x00, 0x00, 0x18, 0x36, 0x36,
+    0x18, 0x00, 0x18, 0xFF, 0xFF, 0xF0, 0x0F, 0x00, 0x76, 0xC0, 0xC0, 0x6C,
+    0xFC, 0x70, 0x60, 0x18, 0x30, 0x38, 0xEE, 0x78, 0x00, 0x60, 0x38, 0xCC,
+    0x00, 0xFC, 0xFC, 0xFC, 0x18, 0xD8, 0x30, 0x00, 0x00, 0x00, 0x00, 0x3C,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x7E, 0x7E, 0x00, 0x00, 0x7C, 0x7C, 0x00,
+    0xFF, 0x00, 0xFF, 0x78, 0x18, 0xE0, 0xC0, 0x99, 0x00, 0x00, 0x18, 0x00,
+    0x00, 0x78, 0x00, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x60, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x60, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xF8,
+    0x00, 0x00, 0x78, 0x00, 0x00, 0x00, 0x00, 0x00, 0xF0, 0x1E, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0xF8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0x78, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x38, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    0xF8, 0x00, 0x00, 0x18, 0x00, 0x30, 0xC7, 0x70, 0x00, 0x00, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0F, 0x03, 0x00, 0x00, 0x00,
+    0x88, 0xAA, 0xEE, 0x18, 0x18, 0x18, 0x36, 0x36, 0x18, 0x36, 0x36, 0x36,
+    0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x18, 0x18, 0x00, 0x18, 0x18, 0x36,
+    0x00, 0x36, 0x00, 0x36, 0x36, 0x00, 0x36, 0x00, 0x00, 0x18, 0x36, 0x00,
+    0x00, 0x18, 0x36, 0x36, 0x18, 0x00, 0x18, 0xFF, 0xFF, 0xF0, 0x0F, 0x00,
+    0x00, 0xC0, 0x00, 0x00, 0x00, 0x00, 0xC0, 0x00, 0xFC, 0x00, 0x00, 0x00,
+    0x00, 0xC0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x18, 0x70, 0x00, 0x00,
+    0x00, 0x00, 0x00, 0x1C, 0x00, 0x00, 0x00, 0x00};

src/libdvi/CMakeLists.txt

@@ -0,0 +1,33 @@
+# Note we are using INTERFACE so that the library can be configured per-app
+# with compile-time defines
+
+add_library(libdvi INTERFACE)
+
+target_sources(libdvi INTERFACE
+	${CMAKE_CURRENT_LIST_DIR}/dvi.c
+	${CMAKE_CURRENT_LIST_DIR}/dvi.h
+	${CMAKE_CURRENT_LIST_DIR}/dvi_config_defs.h
+	${CMAKE_CURRENT_LIST_DIR}/dvi_serialiser.c
+	${CMAKE_CURRENT_LIST_DIR}/dvi_serialiser.h
+	${CMAKE_CURRENT_LIST_DIR}/dvi_timing.c
+	${CMAKE_CURRENT_LIST_DIR}/dvi_timing.h
+	${CMAKE_CURRENT_LIST_DIR}/tmds_encode.S
+	${CMAKE_CURRENT_LIST_DIR}/tmds_encode.c
+	${CMAKE_CURRENT_LIST_DIR}/tmds_encode.h
+	${CMAKE_CURRENT_LIST_DIR}/tmds_table.h
+	${CMAKE_CURRENT_LIST_DIR}/tmds_table_fullres.h
+	${CMAKE_CURRENT_LIST_DIR}/util_queue_u32_inline.h
+	)
+
+target_include_directories(libdvi INTERFACE ${CMAKE_CURRENT_LIST_DIR})
+target_link_libraries(libdvi INTERFACE
+	pico_base_headers
+	pico_util
+	hardware_dma
+	hardware_interp
+	hardware_pio
+	hardware_pwm
+	)
+
+pico_generate_pio_header(libdvi ${CMAKE_CURRENT_LIST_DIR}/dvi_serialiser.pio)
+pico_generate_pio_header(libdvi ${CMAKE_CURRENT_LIST_DIR}/tmds_encode_1bpp.pio)

src/libdvi/dvi.c

@@ -0,0 +1,259 @@
+#include <stdlib.h>
+#include "hardware/dma.h"
+#include "hardware/irq.h"
+
+#include "dvi.h"
+#include "dvi_timing.h"
+#include "dvi_serialiser.h"
+#include "tmds_encode.h"
+
+// Adafruit PicoDVI fork requires a couple global items run-time configurable:
+uint8_t dvi_vertical_repeat = DVI_VERTICAL_REPEAT;
+bool    dvi_monochrome_tmds = DVI_MONOCHROME_TMDS;
+
+// Time-critical functions pulled into RAM but each in a unique section to
+// allow garbage collection
+#define __dvi_func(f) __not_in_flash_func(f)
+#define __dvi_func_x(f) __scratch_x(__STRING(f)) f
+
+// We require exclusive use of a DMA IRQ line. (you wouldn't want to share
+// anyway). It's possible in theory to hook both IRQs and have two DVI outs.
+static struct dvi_inst *dma_irq_privdata[2];
+static void dvi_dma0_irq();
+static void dvi_dma1_irq();
+
+void dvi_init(struct dvi_inst *inst, uint spinlock_tmds_queue, uint spinlock_colour_queue) {
+	dvi_timing_state_init(&inst->timing_state);
+	dvi_serialiser_init(&inst->ser_cfg);
+	for (int i = 0; i < N_TMDS_LANES; ++i) {
+		inst->dma_cfg[i].chan_ctrl = dma_claim_unused_channel(true);
+		inst->dma_cfg[i].chan_data = dma_claim_unused_channel(true);
+		inst->dma_cfg[i].tx_fifo = (void*)&inst->ser_cfg.pio->txf[inst->ser_cfg.sm_tmds[i]];
+		inst->dma_cfg[i].dreq = pio_get_dreq(inst->ser_cfg.pio, inst->ser_cfg.sm_tmds[i], true);
+	}
+	inst->late_scanline_ctr = 0;
+	inst->tmds_buf_release_next = NULL;
+	inst->tmds_buf_release = NULL;
+	queue_init_with_spinlock(&inst->q_tmds_valid,   sizeof(void*),  8, spinlock_tmds_queue);
+	queue_init_with_spinlock(&inst->q_tmds_free,    sizeof(void*),  8, spinlock_tmds_queue);
+	queue_init_with_spinlock(&inst->q_colour_valid, sizeof(void*),  8, spinlock_colour_queue);
+	queue_init_with_spinlock(&inst->q_colour_free,  sizeof(void*),  8, spinlock_colour_queue);
+
+	dvi_setup_scanline_for_vblank(inst->timing, inst->dma_cfg, true, &inst->dma_list_vblank_sync);
+	dvi_setup_scanline_for_vblank(inst->timing, inst->dma_cfg, false, &inst->dma_list_vblank_nosync);
+#if defined(ARDUINO)
+	dvi_setup_scanline_for_active(inst->timing, inst->dma_cfg, (uint32_t*)SRAM_BASE, &inst->dma_list_active);
+#else
+	dvi_setup_scanline_for_active(inst->timing, inst->dma_cfg, (void*)SRAM_BASE, &inst->dma_list_active);
+#endif
+	dvi_setup_scanline_for_active(inst->timing, inst->dma_cfg, NULL, &inst->dma_list_error);
+
+	for (int i = 0; i < DVI_N_TMDS_BUFFERS; ++i) {
+		void *tmdsbuf;
+		if (dvi_monochrome_tmds)
+			tmdsbuf = malloc(inst->timing->h_active_pixels / DVI_SYMBOLS_PER_WORD * sizeof(uint32_t));
+		else
+			tmdsbuf = malloc(3 * inst->timing->h_active_pixels / DVI_SYMBOLS_PER_WORD * sizeof(uint32_t));
+		if (!tmdsbuf)
+			panic("TMDS buffer allocation failed");
+		queue_add_blocking_u32(&inst->q_tmds_free, &tmdsbuf);
+	}
+}
+
+// The IRQs will run on whichever core calls this function (this is why it's
+// called separately from dvi_init)
+void dvi_register_irqs_this_core(struct dvi_inst *inst, uint irq_num) {
+	uint32_t mask_sync_channel = 1u << inst->dma_cfg[TMDS_SYNC_LANE].chan_data;
+	uint32_t mask_all_channels = 0;
+	for (int i = 0; i < N_TMDS_LANES; ++i)
+		mask_all_channels |= 1u << inst->dma_cfg[i].chan_ctrl | 1u << inst->dma_cfg[i].chan_data;
+
+	dma_hw->ints0 = mask_sync_channel;
+	if (irq_num == DMA_IRQ_0) {
+		hw_write_masked(&dma_hw->inte0, mask_sync_channel, mask_all_channels);
+		dma_irq_privdata[0] = inst;
+		irq_set_exclusive_handler(DMA_IRQ_0, dvi_dma0_irq);
+	}
+	else {
+		hw_write_masked(&dma_hw->inte1, mask_sync_channel, mask_all_channels);
+		dma_irq_privdata[1] = inst;
+		irq_set_exclusive_handler(DMA_IRQ_1, dvi_dma1_irq);
+	}
+	irq_set_enabled(irq_num, true);
+}
+
+// Set up control channels to make transfers to data channels' control
+// registers (but don't trigger the control channels -- this is done either by
+// data channel CHAIN_TO or an initial write to MULTI_CHAN_TRIGGER)
+static inline void __attribute__((always_inline)) _dvi_load_dma_op(const struct dvi_lane_dma_cfg dma_cfg[], struct dvi_scanline_dma_list *l) {
+	for (int i = 0; i < N_TMDS_LANES; ++i) {
+		dma_channel_config cfg = dma_channel_get_default_config(dma_cfg[i].chan_ctrl);
+		channel_config_set_ring(&cfg, true, 4); // 16-byte write wrap
+		channel_config_set_read_increment(&cfg, true);
+		channel_config_set_write_increment(&cfg, true);
+		dma_channel_configure(
+			dma_cfg[i].chan_ctrl,
+			&cfg,
+			&dma_hw->ch[dma_cfg[i].chan_data],
+			dvi_lane_from_list(l, i),
+			4, // Configure all 4 registers then halt until next CHAIN_TO
+			false
+		);
+	}
+}
+
+// Setup first set of control block lists, configure the control channels, and
+// trigger them. Control channels will subsequently be triggered only by DMA
+// CHAIN_TO on data channel completion. IRQ handler *must* be prepared before
+// calling this. (Hooked to DMA IRQ0)
+void dvi_start(struct dvi_inst *inst) {
+	_dvi_load_dma_op(inst->dma_cfg, &inst->dma_list_vblank_nosync);
+	dma_start_channel_mask(
+		(1u << inst->dma_cfg[0].chan_ctrl) |
+		(1u << inst->dma_cfg[1].chan_ctrl) |
+		(1u << inst->dma_cfg[2].chan_ctrl));
+
+	// We really don't want the FIFOs to bottom out, so wait for full before
+	// starting the shift-out.
+	for (int i = 0; i < N_TMDS_LANES; ++i)
+		while (!pio_sm_is_tx_fifo_full(inst->ser_cfg.pio, inst->ser_cfg.sm_tmds[i]))
+			tight_loop_contents();
+	dvi_serialiser_enable(&inst->ser_cfg, true);
+}
+
+static inline void __dvi_func_x(_dvi_prepare_scanline_8bpp)(struct dvi_inst *inst, uint32_t *scanbuf) {
+	uint32_t *tmdsbuf;
+	queue_remove_blocking_u32(&inst->q_tmds_free, &tmdsbuf);
+	uint pixwidth = inst->timing->h_active_pixels;
+	uint words_per_channel = pixwidth / DVI_SYMBOLS_PER_WORD;
+	// Scanline buffers are half-resolution; the functions take the number of *input* pixels as parameter.
+	tmds_encode_data_channel_8bpp(scanbuf, tmdsbuf + 0 * words_per_channel, pixwidth / 2, DVI_8BPP_BLUE_MSB,  DVI_8BPP_BLUE_LSB );
+	tmds_encode_data_channel_8bpp(scanbuf, tmdsbuf + 1 * words_per_channel, pixwidth / 2, DVI_8BPP_GREEN_MSB, DVI_8BPP_GREEN_LSB);
+	tmds_encode_data_channel_8bpp(scanbuf, tmdsbuf + 2 * words_per_channel, pixwidth / 2, DVI_8BPP_RED_MSB,   DVI_8BPP_RED_LSB  );
+	queue_add_blocking_u32(&inst->q_tmds_valid, &tmdsbuf);
+}
+
+static inline void __dvi_func_x(_dvi_prepare_scanline_16bpp)(struct dvi_inst *inst, uint32_t *scanbuf) {
+	uint32_t *tmdsbuf;
+	queue_remove_blocking_u32(&inst->q_tmds_free, &tmdsbuf);
+	uint pixwidth = inst->timing->h_active_pixels;
+	uint words_per_channel = pixwidth / DVI_SYMBOLS_PER_WORD;
+	tmds_encode_data_channel_16bpp(scanbuf, tmdsbuf + 0 * words_per_channel, pixwidth / 2, DVI_16BPP_BLUE_MSB,  DVI_16BPP_BLUE_LSB );
+	tmds_encode_data_channel_16bpp(scanbuf, tmdsbuf + 1 * words_per_channel, pixwidth / 2, DVI_16BPP_GREEN_MSB, DVI_16BPP_GREEN_LSB);
+	tmds_encode_data_channel_16bpp(scanbuf, tmdsbuf + 2 * words_per_channel, pixwidth / 2, DVI_16BPP_RED_MSB,   DVI_16BPP_RED_LSB  );
+	queue_add_blocking_u32(&inst->q_tmds_valid, &tmdsbuf);
+}
+
+// "Worker threads" for TMDS encoding (core enters and never returns, but still handles IRQs)
+
+// Version where each record in q_colour_valid is one scanline:
+void __dvi_func(dvi_scanbuf_main_8bpp)(struct dvi_inst *inst) {
+	uint y = 0;
+	while (1) {
+		uint32_t *scanbuf;
+		queue_remove_blocking_u32(&inst->q_colour_valid, &scanbuf);
+		_dvi_prepare_scanline_8bpp(inst, scanbuf);
+		queue_add_blocking_u32(&inst->q_colour_free, &scanbuf);
+		++y;
+		if (y == inst->timing->v_active_lines) {
+			y = 0;
+		}
+	}
+	__builtin_unreachable();
+}
+
+// Ugh copy/paste but it lets us garbage collect the TMDS stuff that is not being used from .scratch_x
+void __dvi_func(dvi_scanbuf_main_16bpp)(struct dvi_inst *inst) {
+	uint y = 0;
+	while (1) {
+		uint32_t *scanbuf;
+		queue_remove_blocking_u32(&inst->q_colour_valid, &scanbuf);
+		_dvi_prepare_scanline_16bpp(inst, scanbuf);
+		queue_add_blocking_u32(&inst->q_colour_free, &scanbuf);
+		++y;
+		if (y == inst->timing->v_active_lines) {
+			y = 0;
+		}
+	}
+	__builtin_unreachable();
+}
+
+static void __dvi_func(dvi_dma_irq_handler)(struct dvi_inst *inst) {
+	// Every fourth interrupt marks the start of the horizontal active region. We
+	// now have until the end of this region to generate DMA blocklist for next
+	// scanline.
+	dvi_timing_state_advance(inst->timing, &inst->timing_state);
+	if (inst->tmds_buf_release && !queue_try_add_u32(&inst->q_tmds_free, &inst->tmds_buf_release))
+		panic("TMDS free queue full in IRQ!");
+	inst->tmds_buf_release = inst->tmds_buf_release_next;
+	inst->tmds_buf_release_next = NULL;
+
+	// Make sure all three channels have definitely loaded their last block
+	// (should be within a few cycles of one another)
+	for (int i = 0; i < N_TMDS_LANES; ++i) {
+#if PICO_SDK_VERSION_MAJOR == 2
+		while (dma_debug_hw->ch[inst->dma_cfg[i].chan_data].dbg_tcr != inst->timing->h_active_pixels / DVI_SYMBOLS_PER_WORD)
+#else
+		while (dma_debug_hw->ch[inst->dma_cfg[i].chan_data].tcr != inst->timing->h_active_pixels / DVI_SYMBOLS_PER_WORD)
+#endif
+			tight_loop_contents();
+	}
+
+	uint32_t *tmdsbuf;
+	while (inst->late_scanline_ctr > 0 && queue_try_remove_u32(&inst->q_tmds_valid, &tmdsbuf)) {
+		// If we displayed this buffer then it would be in the wrong vertical
+		// position on-screen. Just pass it back.
+		queue_add_blocking_u32(&inst->q_tmds_free, &tmdsbuf);
+		--inst->late_scanline_ctr;
+	}
+
+	if (inst->timing_state.v_state != DVI_STATE_ACTIVE) {
+		// Don't care
+		tmdsbuf = NULL;
+	}
+	else if (queue_try_peek_u32(&inst->q_tmds_valid, &tmdsbuf)) {
+		if (inst->timing_state.v_ctr % dvi_vertical_repeat == dvi_vertical_repeat - 1) {
+			queue_remove_blocking_u32(&inst->q_tmds_valid, &tmdsbuf);
+			inst->tmds_buf_release_next = tmdsbuf;
+		}
+	}
+	else {
+		// No valid scanline was ready (generates solid red scanline)
+		tmdsbuf = NULL;
+		if (inst->timing_state.v_ctr % dvi_vertical_repeat == dvi_vertical_repeat - 1)
+			++inst->late_scanline_ctr;
+	}
+
+	switch (inst->timing_state.v_state) {
+		case DVI_STATE_ACTIVE:
+			if (tmdsbuf) {
+				dvi_update_scanline_data_dma(inst->timing, tmdsbuf, &inst->dma_list_active);
+				_dvi_load_dma_op(inst->dma_cfg, &inst->dma_list_active);
+			}
+			else {
+				_dvi_load_dma_op(inst->dma_cfg, &inst->dma_list_error);
+			}
+			if (inst->scanline_callback && inst->timing_state.v_ctr % dvi_vertical_repeat == dvi_vertical_repeat - 1) {
+				inst->scanline_callback();
+			}
+			break;
+		case DVI_STATE_SYNC:
+			_dvi_load_dma_op(inst->dma_cfg, &inst->dma_list_vblank_sync);
+			break;
+		default:
+			_dvi_load_dma_op(inst->dma_cfg, &inst->dma_list_vblank_nosync);
+			break;
+	}
+}
+
+static void __dvi_func(dvi_dma0_irq)() {
+	struct dvi_inst *inst = dma_irq_privdata[0];
+	dma_hw->ints0 = 1u << inst->dma_cfg[TMDS_SYNC_LANE].chan_data;
+	dvi_dma_irq_handler(inst);
+}
+
+static void __dvi_func(dvi_dma1_irq)() {
+	struct dvi_inst *inst = dma_irq_privdata[1];
+	dma_hw->ints1 = 1u << inst->dma_cfg[TMDS_SYNC_LANE].chan_data;
+	dvi_dma_irq_handler(inst);
+}

src/libdvi/dvi.h

@@ -0,0 +1,81 @@
+#ifndef _DVI_H
+#define _DVI_H
+
+#define N_TMDS_LANES 3
+#define TMDS_SYNC_LANE 0 // blue!
+
+#include "pico/util/queue.h"
+
+#include "dvi_config_defs.h"
+#include "dvi_timing.h"
+#include "dvi_serialiser.h"
+#include "util_queue_u32_inline.h"
+
+typedef void (*dvi_callback_t)(void);
+
+struct dvi_inst {
+	// Config ---
+	const struct dvi_timing *timing;
+	struct dvi_lane_dma_cfg dma_cfg[N_TMDS_LANES];
+	struct dvi_timing_state timing_state;
+	struct dvi_serialiser_cfg ser_cfg;
+	// Called in the DMA IRQ once per scanline -- careful with the run time!
+	dvi_callback_t scanline_callback;
+
+	// State ---
+	struct dvi_scanline_dma_list dma_list_vblank_sync;
+	struct dvi_scanline_dma_list dma_list_vblank_nosync;
+	struct dvi_scanline_dma_list dma_list_active;
+	struct dvi_scanline_dma_list dma_list_error;
+
+	// After a TMDS buffer has been enqueue via a control block for the last
+	// time, two IRQs must go by before freeing. The first indicates the control
+	// block for this buf has been loaded, and the second occurs some time after
+	// the actual data DMA transfer has completed.
+	uint32_t *tmds_buf_release_next;
+	uint32_t *tmds_buf_release;
+	// Remember how far behind the source is on TMDS scanlines, so we can output
+	// solid colour until they catch up (rather than dying spectacularly)
+	uint late_scanline_ctr;
+
+	// Encoded scanlines:
+	queue_t q_tmds_valid;
+	queue_t q_tmds_free;
+
+	// Either scanline buffers or frame buffers:
+	queue_t q_colour_valid;
+	queue_t q_colour_free;
+
+};
+
+#if defined(__cplusplus)
+extern "C"
+{
+#endif
+
+// Set up data structures and hardware for DVI.
+void dvi_init(struct dvi_inst *inst, uint spinlock_tmds_queue, uint spinlock_colour_queue);
+
+// Call this after calling dvi_init(). DVI DMA interrupts will be routed to
+// whichever core called this function. Registers an exclusive IRQ handler.
+void dvi_register_irqs_this_core(struct dvi_inst *inst, uint irq_num);
+
+// Start actually wiggling TMDS pairs. Call this once you have initialised the
+// DVI, have registered the IRQs, and are producing rendered scanlines.
+void dvi_start(struct dvi_inst *inst);
+
+// TMDS encode worker function: core enters and doesn't leave, but still
+// responds to IRQs. Repeatedly pop a scanline buffer from q_colour_valid,
+// TMDS encode it, and pass it to the tmds valid queue.
+void dvi_scanbuf_main_8bpp(struct dvi_inst *inst);
+void dvi_scanbuf_main_16bpp(struct dvi_inst *inst);
+
+// Same as above, but each q_colour_valid entry is a framebuffer
+void dvi_framebuf_main_8bpp(struct dvi_inst *inst);
+void dvi_framebuf_main_16bpp(struct dvi_inst *inst);
+
+#if defined(__cplusplus)
+}
+#endif
+
+#endif

src/libdvi/dvi_config_defs.h

@@ -0,0 +1,154 @@
+#ifndef _DVI_CONFIG_DEFS_H
+#define _DVI_CONFIG_DEFS_H
+
+// Compile-time configuration definitions for libdvi. This file provides
+// defaults -- you can override using a board header, or setting compile
+// definitions directly from the commandline (e.g. using CMake
+// target_compile_definitions())
+
+// Pull in base headers to make sure board definitions override the
+// definitions provided here. Note this file is included in asm and C.
+// added ifndef after Philhower 4.6.0 release to fix compilation issue
+#ifndef __ASSEMBLER__
+#include "hardware/platform_defs.h"
+#include "pico/config.h"
+#endif
+
+// ----------------------------------------------------------------------------
+// General DVI defines
+
+// How many times to output the same TMDS buffer before recyling it onto the
+// free queue. Pixels are repeated vertically if this is >1.
+#ifndef DVI_VERTICAL_REPEAT
+#define DVI_VERTICAL_REPEAT 2
+#endif
+
+// Number of TMDS buffers to allocate (malloc()) in DVI init. You can set this
+// to 0 if you want to allocate your own (e.g. if you want static buffers)
+#ifndef DVI_N_TMDS_BUFFERS
+#define DVI_N_TMDS_BUFFERS 3
+#endif
+
+// If 1, replace the DVI serialiser with a 10n1 UART (1 start bit, 10 data
+// bits, 1 stop bit) so the stream can be dumped and analysed easily.
+#ifndef DVI_SERIAL_DEBUG
+#define DVI_SERIAL_DEBUG 0
+#endif
+
+// If 1, the same TMDS symbols are sent to all 3 lanes during the horizontal
+// active period. This means only monochrome colour is available, but the TMDS
+// buffers are 3 times smaller as a result, and the performance requirements
+// for encode are also cut by 3.
+#ifndef DVI_MONOCHROME_TMDS
+#define DVI_MONOCHROME_TMDS 0
+#endif
+
+// By default, we assume each 32-bit word written to a PIO FIFO contains 2x
+// 10-bit TMDS symbols, concatenated into the lower 20 bits, least-significant
+// first. This is convenient if you are generating two or more pixels at once,
+// e.g. using the pixel-doubling TMDS encode. You can change this value to 1
+// (so each word contains 1 symbol) for e.g. full resolution RGB encode. Note
+// that this value needs to divide the DVI horizontal timings, so is limited
+// to 1 or 2.
+#ifndef DVI_SYMBOLS_PER_WORD
+#define DVI_SYMBOLS_PER_WORD 2
+#endif
+
+#if DVI_SYMBOLS_PER_WORD != 1 && DVI_SYMBOLS_PER_WORD !=2
+#error "Unsupported value for DVI_SYMBOLS_PER_WORD"
+#endif
+
+// ----------------------------------------------------------------------------
+// Pixel component layout
+
+// By default we go R, G, B from MSB -> LSB. Override to e.g. swap RGB <-> BGR
+
+// Default 8bpp layout: RGB332, {r[1:0], g[2:0], b[1:0]}
+
+#ifndef DVI_8BPP_RED_MSB
+#define DVI_8BPP_RED_MSB 7
+#endif
+
+#ifndef DVI_8BPP_RED_LSB
+#define DVI_8BPP_RED_LSB 5
+#endif
+
+#ifndef DVI_8BPP_GREEN_MSB
+#define DVI_8BPP_GREEN_MSB 4
+#endif
+
+#ifndef DVI_8BPP_GREEN_LSB
+#define DVI_8BPP_GREEN_LSB 2
+#endif
+
+#ifndef DVI_8BPP_BLUE_MSB
+#define DVI_8BPP_BLUE_MSB 1
+#endif
+
+#ifndef DVI_8BPP_BLUE_LSB
+#define DVI_8BPP_BLUE_LSB 0
+#endif
+
+// Default 16bpp layout: RGB565, {r[4:0], g[5:0], b[4:0]}
+
+#ifndef DVI_16BPP_RED_MSB
+#define DVI_16BPP_RED_MSB 15
+#endif
+
+#ifndef DVI_16BPP_RED_LSB
+#define DVI_16BPP_RED_LSB 11
+#endif
+
+#ifndef DVI_16BPP_GREEN_MSB
+#define DVI_16BPP_GREEN_MSB 10
+#endif
+
+#ifndef DVI_16BPP_GREEN_LSB
+#define DVI_16BPP_GREEN_LSB 5
+#endif
+
+#ifndef DVI_16BPP_BLUE_MSB
+#define DVI_16BPP_BLUE_MSB 4
+#endif
+
+#ifndef DVI_16BPP_BLUE_LSB
+#define DVI_16BPP_BLUE_LSB 0
+#endif
+
+// Default 1bpp layout: bitwise little-endian, i.e. least significant bit of
+// each word is the first (leftmost) of a block of 32 pixels.
+
+// If 1, reverse the order of pixels within each byte. Order of bytes within
+// each word is still little-endian.
+#ifndef DVI_1BPP_BIT_REVERSE
+#define DVI_1BPP_BIT_REVERSE 1 // Adafruit_GFX GFXcanvas1 requires this 1
+#endif
+
+// ----------------------------------------------------------------------------
+// TMDS encode controls
+
+// Number of TMDS loop bodies between branches. cmp + branch costs 3 cycles,
+// so you can easily save 10% of encode time by bumping this. Note that body
+// will *already* produce multiple pixels, and total symbols per iteration
+// must cleanly divide symbols per scanline, else the loop won't terminate.
+// Point gun away from foot.
+#ifndef TMDS_ENCODE_UNROLL
+#define TMDS_ENCODE_UNROLL 1
+#endif
+
+// If 1, don't save/restore the interpolators on full-resolution TMDS encode.
+// Speed hack. The TMDS code uses both interpolators, for each of the 3 data
+// channels, so this define avoids 6 save/restores per scanline.
+#ifndef TMDS_FULLRES_NO_INTERP_SAVE
+#define TMDS_FULLRES_NO_INTERP_SAVE 0
+#endif
+
+// If 1, don't DC-balance the output of full resolution encode. Hilariously
+// noncompliant, but Dell Ultrasharp -- the honey badger of computer monitors
+// -- does not seem to mind (it helps that we DC-couple). Another speed hack,
+// useful when you are trying to get everything else up to speed.
+#ifndef TMDS_FULLRES_NO_DC_BALANCE
+#define TMDS_FULLRES_NO_DC_BALANCE 0
+#endif
+
+#endif

src/libdvi/dvi_serialiser.c

@@ -0,0 +1,73 @@
+#include "pico.h"
+#include "hardware/pio.h"
+#include "hardware/gpio.h"
+#include "hardware/pwm.h"
+#include "hardware/structs/padsbank0.h"
+
+#include "dvi.h"
+#include "dvi_serialiser.h"
+#include "dvi_serialiser.pio.h"
+
+static void dvi_configure_pad(uint gpio, bool invert) {
+	// 2 mA drive, enable slew rate limiting (this seems fine even at 720p30, and
+	// the 3V3 LDO doesn't get warm like when turning all the GPIOs up to 11).
+	// Also disable digital receiver.
+	hw_write_masked(
+		&padsbank0_hw->io[gpio],
+		(0 << PADS_BANK0_GPIO0_DRIVE_LSB),
+		PADS_BANK0_GPIO0_DRIVE_BITS | PADS_BANK0_GPIO0_SLEWFAST_BITS | PADS_BANK0_GPIO0_IE_BITS
+	);
+	gpio_set_outover(gpio, invert ? GPIO_OVERRIDE_INVERT : GPIO_OVERRIDE_NORMAL);
+}
+
+void dvi_serialiser_init(struct dvi_serialiser_cfg *cfg) {
+#if DVI_SERIAL_DEBUG
+	uint offset = pio_add_program(cfg->pio, &dvi_serialiser_debug_program);
+#else
+	uint offset = pio_add_program(cfg->pio, &dvi_serialiser_program);
+#endif
+	cfg->prog_offs = offset;
+
+	for (int i = 0; i < N_TMDS_LANES; ++i) {
+		pio_sm_claim(cfg->pio, cfg->sm_tmds[i]);
+		dvi_serialiser_program_init(
+			cfg->pio,
+			cfg->sm_tmds[i],
+			offset,
+			cfg->pins_tmds[i],
+			DVI_SERIAL_DEBUG
+		);
+		dvi_configure_pad(cfg->pins_tmds[i], cfg->invert_diffpairs);
+		dvi_configure_pad(cfg->pins_tmds[i] + 1, cfg->invert_diffpairs);
+	}
+
+	// Use a PWM slice to drive the pixel clock. Both GPIOs must be on the same
+	// slice (lower-numbered GPIO must be even).
+	assert(cfg->pins_clk % 2 == 0);
+	uint slice = pwm_gpio_to_slice_num(cfg->pins_clk);
+	// 5 cycles high, 5 low. Invert one channel so that we get complementary outputs.
+	pwm_config pwm_cfg = pwm_get_default_config();
+	pwm_config_set_output_polarity(&pwm_cfg, true, false);
+	pwm_config_set_wrap(&pwm_cfg, 9);
+	pwm_init(slice, &pwm_cfg, false);
+	pwm_set_both_levels(slice, 5, 5);
+
+	for (uint i = cfg->pins_clk; i <= cfg->pins_clk + 1; ++i) {
+		gpio_set_function(i, GPIO_FUNC_PWM);
+		dvi_configure_pad(i, cfg->invert_diffpairs);
+	}
+}
+
+void dvi_serialiser_enable(struct dvi_serialiser_cfg *cfg, bool enable) {
+	uint mask = 0;
+	for (int i = 0; i < N_TMDS_LANES; ++i)
+		mask |= 1u << (cfg->sm_tmds[i] + PIO_CTRL_SM_ENABLE_LSB);
+	if (enable) {
+		hw_set_bits(&cfg->pio->ctrl, mask);
+		pwm_set_enabled(pwm_gpio_to_slice_num(cfg->pins_clk), true);
+	}
+	else {
+		hw_clear_bits(&cfg->pio->ctrl, mask);
+		pwm_set_enabled(pwm_gpio_to_slice_num(cfg->pins_clk), false);
+	}
+}

src/libdvi/dvi_serialiser.h

@@ -0,0 +1,22 @@
+#ifndef _DVI_SERIALISER_H
+#define _DVI_SERIALISER_H
+
+#include "hardware/pio.h"
+#include "dvi_config_defs.h"
+
+#define N_TMDS_LANES 3
+
+struct dvi_serialiser_cfg {
+	PIO pio;
+	uint sm_tmds[N_TMDS_LANES];
+	uint pins_tmds[N_TMDS_LANES];
+	uint pins_clk;
+	bool invert_diffpairs;
+	uint prog_offs;
+};
+
+void dvi_serialiser_init(struct dvi_serialiser_cfg *cfg);
+void dvi_serialiser_enable(struct dvi_serialiser_cfg *cfg, bool enable);
+uint32_t dvi_single_to_diff(uint32_t in);
+
+#endif

src/libdvi/dvi_serialiser.pio

@@ -0,0 +1,53 @@
+.program dvi_serialiser
+.side_set 2
+.origin 0
+
+; Single-ended -> differential serial
+
+	out pc, 1    side 0b10
+	out pc, 1    side 0b01
+
+.program dvi_serialiser_debug
+.side_set 1 opt
+
+; The debug variant behaves as a UART with 1 start bit, 10 data bits, 1 stop
+; bit, and 5/6ths the data throughput of the TMDS version.
+
+	pull ifempty  side 1 ; Extend stop bit with FIFO stall
+	nop           side 0
+	out pins, 1          ; Unrolled because we require 1 bit / clk
+	out pins, 1
+	out pins, 1
+	out pins, 1
+	out pins, 1
+	out pins, 1
+	out pins, 1
+	out pins, 1
+	out pins, 1
+	out pins, 1
+	
+% c-sdk {
+#include "dvi_config_defs.h"
+
+static inline void dvi_serialiser_program_init(PIO pio, uint sm, uint offset, uint data_pins, bool debug) {
+    pio_sm_set_pins_with_mask(pio, sm, 2u << data_pins, 3u << data_pins);
+    pio_sm_set_pindirs_with_mask(pio, sm, ~0u, 3u << data_pins);
+    pio_gpio_init(pio, data_pins);
+    pio_gpio_init(pio, data_pins + 1);
+
+    pio_sm_config c;
+    if (debug) {
+        c = dvi_serialiser_debug_program_get_default_config(offset);
+    }
+    else {
+        c = dvi_serialiser_program_get_default_config(offset);
+    }
+    sm_config_set_sideset_pins(&c, data_pins);
+    if (debug)
+	    sm_config_set_out_pins(&c, data_pins, 1);
+    sm_config_set_out_shift(&c, true, !debug, 10 * DVI_SYMBOLS_PER_WORD);
+    sm_config_set_fifo_join(&c, PIO_FIFO_JOIN_TX);
+    pio_sm_init(pio, sm, offset, &c);
+    pio_sm_set_enabled(pio, sm, false);
+}
+%}

src/libdvi/dvi_serialiser.pio.h

@@ -0,0 +1,101 @@
+// -------------------------------------------------- //
+// This file is autogenerated by pioasm; do not edit! //
+// -------------------------------------------------- //
+
+#pragma once
+
+#if !PICO_NO_HARDWARE
+#include "hardware/pio.h"
+#endif
+
+// -------------- //
+// dvi_serialiser //
+// -------------- //
+
+#define dvi_serialiser_wrap_target 0
+#define dvi_serialiser_wrap 1
+
+static const uint16_t dvi_serialiser_program_instructions[] = {
+            //     .wrap_target
+    0x70a1, //  0: out    pc, 1           side 2     
+    0x68a1, //  1: out    pc, 1           side 1     
+            //     .wrap
+};
+
+#if !PICO_NO_HARDWARE
+static const struct pio_program dvi_serialiser_program = {
+    .instructions = dvi_serialiser_program_instructions,
+    .length = 2,
+    .origin = 0,
+};
+
+static inline pio_sm_config dvi_serialiser_program_get_default_config(uint offset) {
+    pio_sm_config c = pio_get_default_sm_config();
+    sm_config_set_wrap(&c, offset + dvi_serialiser_wrap_target, offset + dvi_serialiser_wrap);
+    sm_config_set_sideset(&c, 2, false, false);
+    return c;
+}
+#endif
+
+// -------------------- //
+// dvi_serialiser_debug //
+// -------------------- //
+
+#define dvi_serialiser_debug_wrap_target 0
+#define dvi_serialiser_debug_wrap 11
+
+static const uint16_t dvi_serialiser_debug_program_instructions[] = {
+            //     .wrap_target
+    0x98e0, //  0: pull   ifempty block   side 1     
+    0xb042, //  1: nop                    side 0     
+    0x6001, //  2: out    pins, 1                    
+    0x6001, //  3: out    pins, 1                    
+    0x6001, //  4: out    pins, 1                    
+    0x6001, //  5: out    pins, 1                    
+    0x6001, //  6: out    pins, 1                    
+    0x6001, //  7: out    pins, 1                    
+    0x6001, //  8: out    pins, 1                    
+    0x6001, //  9: out    pins, 1                    
+    0x6001, // 10: out    pins, 1                    
+    0x6001, // 11: out    pins, 1                    
+            //     .wrap
+};
+
+#if !PICO_NO_HARDWARE
+static const struct pio_program dvi_serialiser_debug_program = {
+    .instructions = dvi_serialiser_debug_program_instructions,
+    .length = 12,
+    .origin = -1,
+};
+
+static inline pio_sm_config dvi_serialiser_debug_program_get_default_config(uint offset) {
+    pio_sm_config c = pio_get_default_sm_config();
+    sm_config_set_wrap(&c, offset + dvi_serialiser_debug_wrap_target, offset + dvi_serialiser_debug_wrap);
+    sm_config_set_sideset(&c, 2, true, false);
+    return c;
+}
+
+#include "dvi_config_defs.h"
+static inline void dvi_serialiser_program_init(PIO pio, uint sm, uint offset, uint data_pins, bool debug) {
+    pio_sm_set_pins_with_mask(pio, sm, 2u << data_pins, 3u << data_pins);
+    pio_sm_set_pindirs_with_mask(pio, sm, ~0u, 3u << data_pins);
+    pio_gpio_init(pio, data_pins);
+    pio_gpio_init(pio, data_pins + 1);
+    pio_sm_config c;
+    if (debug) {
+        c = dvi_serialiser_debug_program_get_default_config(offset);
+    }
+    else {
+        c = dvi_serialiser_program_get_default_config(offset);
+    }
+    sm_config_set_sideset_pins(&c, data_pins);
+    if (debug)
+	    sm_config_set_out_pins(&c, data_pins, 1);
+    sm_config_set_out_shift(&c, true, !debug, 10 * DVI_SYMBOLS_PER_WORD);
+    sm_config_set_fifo_join(&c, PIO_FIFO_JOIN_TX);
+    pio_sm_init(pio, sm, offset, &c);
+    pio_sm_set_enabled(pio, sm, false);
+}
+
+#endif
+

src/libdvi/dvi_timing.c

@@ -0,0 +1,344 @@
+#include "dvi.h"
+#include "dvi_timing.h"
+#include "hardware/dma.h"
+
+// This file contains:
+// - Timing parameters for DVI modes (horizontal + vertical counts, best
+//   achievable bit clock from 12 MHz crystal)
+// - Helper functions for generating DMA lists based on these timings
+
+extern bool dvi_monochrome_tmds; // In dvi.c
+
+// Pull into RAM but apply unique section suffix to allow linker GC
+#define __dvi_func(x) __not_in_flash_func(x)
+#define __dvi_const(x) __not_in_flash_func(x)
+
+// VGA -- we do this mode properly, with a pretty comfortable clk_sys (252 MHz)
+const struct dvi_timing __dvi_const(dvi_timing_640x480p_60hz) = {
+	.h_sync_polarity   = false,
+	.h_front_porch     = 16,
+	.h_sync_width      = 96,
+	.h_back_porch      = 48,
+	.h_active_pixels   = 640,
+
+	.v_sync_polarity   = false,
+	.v_front_porch     = 10,
+	.v_sync_width      = 2,
+	.v_back_porch      = 33,
+	.v_active_lines    = 480,
+
+	.bit_clk_khz       = 252000
+};
+
+// SVGA -- completely by-the-book but requires 400 MHz clk_sys
+const struct dvi_timing __dvi_const(dvi_timing_800x600p_60hz) = {
+	.h_sync_polarity   = false,
+	.h_front_porch     = 44,
+	.h_sync_width      = 128,
+	.h_back_porch      = 88,
+	.h_active_pixels   = 800,
+
+	.v_sync_polarity   = false,
+	.v_front_porch     = 1,
+	.v_sync_width      = 4,
+	.v_back_porch      = 23,
+	.v_active_lines    = 600,
+
+	.bit_clk_khz       = 400000
+};
+
+// 800x480p 60 Hz (note this doesn't seem to be a CEA mode, I just used the
+// output of `cvt 800 480 60`), 295 MHz bit clock
+const struct dvi_timing __dvi_const(dvi_timing_800x480p_60hz) = {
+	.h_sync_polarity = false,
+	.h_front_porch   = 24,
+	.h_sync_width    = 72,
+	.h_back_porch    = 96,
+	.h_active_pixels = 800,
+
+	.v_sync_polarity = true,
+	.v_front_porch   = 3,
+	.v_sync_width    = 10,
+	.v_back_porch    = 7,
+	.v_active_lines  = 480,
+
+	.bit_clk_khz     = 295200
+};
+
+// 800x480p 30 Hz, 148 MHz bit clock. This is not a CEA mode either , but needs
+// lower overclock compared to dvi_timing_800x480p_60hz and the timing is
+// verified to work on https://www.adafruit.com/product/2232
+const struct dvi_timing __dvi_const(dvi_timing_800x480p_30hz) = {
+        .h_sync_polarity = false,
+        .h_front_porch   = 24,
+        .h_sync_width    = 72,
+        .h_back_porch    = 96,
+        .h_active_pixels = 800,
+
+        .v_sync_polarity = true,
+        .v_front_porch   = 3,
+        .v_sync_width    = 10,
+        .v_back_porch    = 4,
+        .v_active_lines  = 480,
+
+        .bit_clk_khz     = 147600
+};
+
+
+// SVGA reduced blanking (355 MHz bit clock) -- valid CVT mode, less common
+// than fully-blanked SVGA, but doesn't require such a high system clock
+const struct dvi_timing __dvi_const(dvi_timing_800x600p_reduced_60hz) = {
+	.h_sync_polarity   = true,
+	.h_front_porch     = 48,
+	.h_sync_width      = 32,
+	.h_back_porch      = 80,
+	.h_active_pixels   = 800,
+
+	.v_sync_polarity   = false,
+	.v_front_porch     = 3,
+	.v_sync_width      = 4,
+	.v_back_porch      = 11,
+	.v_active_lines    = 600,
+
+	.bit_clk_khz       = 354000
+};
+
+// Also known as qHD, bit uncommon, but it's a nice modest-resolution 16:9
+// aspect mode. Pixel clock 37.3 MHz
+const struct dvi_timing __dvi_const(dvi_timing_960x540p_60hz) = {
+	.h_sync_polarity   = true,
+	.h_front_porch     = 16,
+	.h_sync_width      = 32,
+	.h_back_porch      = 96,
+	.h_active_pixels   = 960,
+
+	.v_sync_polarity   = true,
+	.v_front_porch     = 2,
+	.v_sync_width      = 6,
+	.v_back_porch      = 15,
+	.v_active_lines    = 540,
+
+	.bit_clk_khz       = 372000
+};
+
+// Note this is NOT the correct 720p30 CEA mode, but rather 720p60 run at half
+// pixel clock. Seems to be commonly accepted (and is a valid CVT mode). The
+// actual CEA mode is the same pixel clock as 720p60 but with >50% blanking,
+// which would require a clk_sys of 742 MHz!
+const struct dvi_timing __dvi_const(dvi_timing_1280x720p_30hz) = {
+	.h_sync_polarity   = true,
+	.h_front_porch     = 110,
+	.h_sync_width      = 40,
+	.h_back_porch      = 220,
+	.h_active_pixels   = 1280,
+
+	.v_sync_polarity   = true,
+	.v_front_porch     = 5,
+	.v_sync_width      = 5,
+	.v_back_porch      = 20,
+	.v_active_lines    = 720,
+
+	.bit_clk_khz       = 372000
+};
+
+// Reduced-blanking (CVT) 720p. You aren't supposed to use reduced blanking
+// modes below 60 Hz, but I won't tell anyone (and it works on the monitors
+// I've tried). This nets a lower system clock than regular 720p30 (319 MHz)
+const struct dvi_timing __dvi_const(dvi_timing_1280x720p_reduced_30hz) = {
+	.h_sync_polarity   = true,
+	.h_front_porch     = 48,
+	.h_sync_width      = 32,
+	.h_back_porch      = 80,
+	.h_active_pixels   = 1280,
+
+	.v_sync_polarity   = false,
+	.v_front_porch     = 3,
+	.v_sync_width      = 5,
+	.v_back_porch      = 13,
+	.v_active_lines    = 720,
+
+	.bit_clk_khz       = 319200
+};
+
+// This requires a spicy 488 MHz system clock and is illegal in most countries
+// (you need to have a very lucky piece of silicon to run this at 1.3 V, or
+// connect an external supply and give it a bit more juice)
+const struct dvi_timing __dvi_const(dvi_timing_1600x900p_reduced_30hz) = {
+	.h_sync_polarity   = true,
+	.h_front_porch     = 48,
+	.h_sync_width      = 32,
+	.h_back_porch      = 80,
+	.h_active_pixels   = 1600,
+
+	.v_sync_polarity   = false,
+	.v_front_porch     = 3,
+	.v_sync_width      = 5,
+	.v_back_porch      = 18,
+	.v_active_lines    = 900,
+
+	.bit_clk_khz       = 488000
+};
+
+// ----------------------------------------------------------------------------
+
+// The DMA scheme is:
+//
+// - One channel transferring data to each of the three PIO state machines
+//   performing TMDS serialisation
+//
+// - One channel programming the registers of each of these data channels,
+//   triggered (CHAIN_TO) each time the corresponding data channel completes
+//
+// - Lanes 1 and 2 have one block for blanking and one for video data
+//
+// - Lane 0 has one block for each horizontal region (front porch, hsync, back
+//   porch, active)
+//
+// - The IRQ_QUIET flag is used to select which data block on the sync lane is
+//   allowed to generate an IRQ upon completion. This is the block immediately
+//   before the horizontal active region. The IRQ is entered at ~the same time
+//   as the last data transfer starts
+//
+// - The IRQ points the control channels at new blocklists for next scanline.
+//   The DMA starts the new list automatically at end-of-scanline, via
+//   CHAIN_TO.
+//
+// The horizontal active region is the longest continuous transfer, so this
+// gives the most time to handle the IRQ and load new blocklists.
+//
+// Note a null trigger IRQ is not suitable because we get that *after* the
+// last data transfer finishes, and the FIFOs bottom out very shortly
+// afterward. For pure DVI (four blocks per scanline), it works ok to take
+// four regular IRQs per scanline and return early from 3 of them, but this
+// breaks down when you have very short scanline sections like guard bands.
+
+// Each symbol appears twice, concatenated in one word. Note these must be in
+// RAM because they see a lot of DMA traffic
+const uint32_t __dvi_const(dvi_ctrl_syms)[4] = {
+	0xd5354,
+	0x2acab,
+	0x55154,
+	0xaaeab
+};
+
+// Output solid red scanline if we are given NULL for tmdsbuff
+#if DVI_SYMBOLS_PER_WORD == 2
+static uint32_t __dvi_const(empty_scanline_tmds)[3] = {
+	0x7fd00u, // 0x00, 0x00
+	0x7fd00u, // 0x00, 0x00
+	0xbfa01u  // 0xfc, 0xfc
+};
+#else
+static uint32_t __attribute__((aligned(8))) __dvi_const(empty_scanline_tmds)[6] = {
+	0x100u, 0x1ffu, // 0x00, 0x00
+	0x100u, 0x1ffu, // 0x00, 0x00
+	0x201u, 0x2feu  // 0xfc, 0xfc
+};
+#endif
+
+void dvi_timing_state_init(struct dvi_timing_state *t) {
+	t->v_ctr = 0;
+	t->v_state = DVI_STATE_FRONT_PORCH;
+};
+
+void __dvi_func(dvi_timing_state_advance)(const struct dvi_timing *t, struct dvi_timing_state *s) {
+		s->v_ctr++;
+		if ((s->v_state == DVI_STATE_FRONT_PORCH && s->v_ctr == t->v_front_porch) || 
+		    (s->v_state == DVI_STATE_SYNC && s->v_ctr == t->v_sync_width) ||
+		    (s->v_state == DVI_STATE_BACK_PORCH && s->v_ctr == t->v_back_porch) ||
+		    (s->v_state == DVI_STATE_ACTIVE && s->v_ctr == t->v_active_lines)) {
+
+			s->v_state = (s->v_state + 1) % DVI_STATE_COUNT;
+			s->v_ctr = 0;
+		}
+}
+
+void dvi_scanline_dma_list_init(struct dvi_scanline_dma_list *dma_list) {
+	*dma_list = (struct dvi_scanline_dma_list){};	
+}
+
+static const uint32_t *get_ctrl_sym(bool vsync, bool hsync) {
+	return &dvi_ctrl_syms[!!vsync << 1 | !!hsync];
+}
+
+// Make a sequence of paced transfers to the relevant FIFO
+static void _set_data_cb(dma_cb_t *cb, const struct dvi_lane_dma_cfg *dma_cfg,
+		const void *read_addr, uint transfer_count, uint read_ring, bool irq_on_finish) {
+	cb->read_addr = read_addr;
+	cb->write_addr = dma_cfg->tx_fifo;
+	cb->transfer_count = transfer_count;
+	cb->c = dma_channel_get_default_config(dma_cfg->chan_data);
+	channel_config_set_ring(&cb->c, false, read_ring);
+	channel_config_set_dreq(&cb->c, dma_cfg->dreq);
+	// Call back to control channel for reconfiguration:
+	channel_config_set_chain_to(&cb->c, dma_cfg->chan_ctrl);
+	// Note we never send a null trigger, so IRQ_QUIET is an IRQ suppression flag
+	channel_config_set_irq_quiet(&cb->c, !irq_on_finish);
+};
+
+void dvi_setup_scanline_for_vblank(const struct dvi_timing *t, const struct dvi_lane_dma_cfg dma_cfg[],
+		bool vsync_asserted, struct dvi_scanline_dma_list *l) {
+
+	bool vsync = t->v_sync_polarity == vsync_asserted;
+	const uint32_t *sym_hsync_off = get_ctrl_sym(vsync, !t->h_sync_polarity);
+	const uint32_t *sym_hsync_on  = get_ctrl_sym(vsync,  t->h_sync_polarity);
+	const uint32_t *sym_no_sync   = get_ctrl_sym(false,  false             );
+
+	dma_cb_t *synclist = dvi_lane_from_list(l, TMDS_SYNC_LANE);
+	// The symbol table contains each control symbol *twice*, concatenated into 20 LSBs of table word, so we can always do word-repeat.
+	_set_data_cb(&synclist[0], &dma_cfg[TMDS_SYNC_LANE], sym_hsync_off, t->h_front_porch   / DVI_SYMBOLS_PER_WORD, 2, false);
+	_set_data_cb(&synclist[1], &dma_cfg[TMDS_SYNC_LANE], sym_hsync_on,  t->h_sync_width    / DVI_SYMBOLS_PER_WORD, 2, false);
+	_set_data_cb(&synclist[2], &dma_cfg[TMDS_SYNC_LANE], sym_hsync_off, t->h_back_porch    / DVI_SYMBOLS_PER_WORD, 2, true);
+	_set_data_cb(&synclist[3], &dma_cfg[TMDS_SYNC_LANE], sym_hsync_off, t->h_active_pixels / DVI_SYMBOLS_PER_WORD, 2, false);
+
+	for (int i = 0; i < N_TMDS_LANES; ++i) {
+		if (i == TMDS_SYNC_LANE)
+			continue;
+		dma_cb_t *cblist = dvi_lane_from_list(l, i);
+		_set_data_cb(&cblist[0], &dma_cfg[i], sym_no_sync,(t->h_front_porch + t->h_sync_width + t->h_back_porch) / DVI_SYMBOLS_PER_WORD, 2, false);
+		_set_data_cb(&cblist[1], &dma_cfg[i], sym_no_sync, t->h_active_pixels / DVI_SYMBOLS_PER_WORD, 2, false);
+	}
+}
+
+void dvi_setup_scanline_for_active(const struct dvi_timing *t, const struct dvi_lane_dma_cfg dma_cfg[],
+		uint32_t *tmdsbuf, struct dvi_scanline_dma_list *l) {
+
+	const uint32_t *sym_hsync_off = get_ctrl_sym(!t->v_sync_polarity, !t->h_sync_polarity);
+	const uint32_t *sym_hsync_on  = get_ctrl_sym(!t->v_sync_polarity,  t->h_sync_polarity);
+	const uint32_t *sym_no_sync   = get_ctrl_sym(false,                false             );
+
+	dma_cb_t *synclist = dvi_lane_from_list(l, TMDS_SYNC_LANE);
+	_set_data_cb(&synclist[0], &dma_cfg[TMDS_SYNC_LANE], sym_hsync_off, t->h_front_porch / DVI_SYMBOLS_PER_WORD, 2, false);
+	_set_data_cb(&synclist[1], &dma_cfg[TMDS_SYNC_LANE], sym_hsync_on,  t->h_sync_width  / DVI_SYMBOLS_PER_WORD, 2, false);
+	_set_data_cb(&synclist[2], &dma_cfg[TMDS_SYNC_LANE], sym_hsync_off, t->h_back_porch  / DVI_SYMBOLS_PER_WORD, 2, true);
+
+	for (int i = 0; i < N_TMDS_LANES; ++i) {
+		dma_cb_t *cblist = dvi_lane_from_list(l, i);
+		if (i != TMDS_SYNC_LANE) {
+			_set_data_cb(&cblist[0], &dma_cfg[i], sym_no_sync,
+				(t->h_front_porch + t->h_sync_width + t->h_back_porch) / DVI_SYMBOLS_PER_WORD, 2, false);
+		}
+		int target_block = i == TMDS_SYNC_LANE ? DVI_SYNC_LANE_CHUNKS - 1 :  DVI_NOSYNC_LANE_CHUNKS - 1;
+		if (tmdsbuf) {
+			// Non-repeating DMA for the freshly-encoded TMDS buffer
+			_set_data_cb(&cblist[target_block], &dma_cfg[i], tmdsbuf + i * (t->h_active_pixels / DVI_SYMBOLS_PER_WORD),
+				t->h_active_pixels / DVI_SYMBOLS_PER_WORD, 0, false);
+		}
+		else {
+			// Use read ring to repeat the correct DC-balanced symbol pair on blank scanlines (4 or 8 byte period)
+			_set_data_cb(&cblist[target_block], &dma_cfg[i], &empty_scanline_tmds[2 * i / DVI_SYMBOLS_PER_WORD],
+				t->h_active_pixels / DVI_SYMBOLS_PER_WORD, DVI_SYMBOLS_PER_WORD == 2 ? 2 : 3, false);
+		}
+	}
+}
+
+void __dvi_func(dvi_update_scanline_data_dma)(const struct dvi_timing *t, const uint32_t *tmdsbuf, struct dvi_scanline_dma_list *l) {
+	for (int i = 0; i < N_TMDS_LANES; ++i) {
+		const uint32_t *lane_tmdsbuf = dvi_monochrome_tmds ? tmdsbuf : tmdsbuf + i * t->h_active_pixels / DVI_SYMBOLS_PER_WORD;
+		if (i == TMDS_SYNC_LANE)
+			dvi_lane_from_list(l, i)[3].read_addr = lane_tmdsbuf;
+		else
+			dvi_lane_from_list(l, i)[1].read_addr = lane_tmdsbuf;
+	}
+}
+

src/libdvi/dvi_timing.h

@@ -0,0 +1,100 @@
+#ifndef _DVI_TIMING_H
+#define _DVI_TIMING_H
+
+#include "hardware/dma.h"
+#include "pico/util/queue.h"
+
+#include "dvi.h"
+
+struct dvi_timing {
+	bool h_sync_polarity;
+	uint h_front_porch;
+	uint h_sync_width;
+	uint h_back_porch;
+	uint h_active_pixels;
+
+	bool v_sync_polarity;
+	uint v_front_porch;
+	uint v_sync_width;
+	uint v_back_porch;
+	uint v_active_lines;
+
+	uint bit_clk_khz;
+};
+
+enum dvi_line_state {
+	DVI_STATE_FRONT_PORCH = 0,
+	DVI_STATE_SYNC,
+	DVI_STATE_BACK_PORCH,
+	DVI_STATE_ACTIVE,
+	DVI_STATE_COUNT
+};
+
+struct dvi_timing_state {
+	uint v_ctr;
+	enum dvi_line_state v_state;
+};
+
+// This should map directly to DMA register layout, but more convenient types
+// (also this really shouldn't be here... we don't have a dma_cb in the SDK
+// because there are many valid formats due to aliases)
+typedef struct dma_cb {
+	const void *read_addr;
+	void *write_addr;
+	uint32_t transfer_count;
+	dma_channel_config c;
+} dma_cb_t;
+
+static_assert(sizeof(dma_cb_t) == 4 * sizeof(uint32_t), "bad dma layout");
+static_assert(__builtin_offsetof(dma_cb_t, c.ctrl) == __builtin_offsetof(dma_channel_hw_t, ctrl_trig), "bad dma layout");
+
+#define DVI_SYNC_LANE_CHUNKS DVI_STATE_COUNT
+#define DVI_NOSYNC_LANE_CHUNKS 2
+
+struct dvi_scanline_dma_list {
+	dma_cb_t l0[DVI_SYNC_LANE_CHUNKS];
+	dma_cb_t l1[DVI_NOSYNC_LANE_CHUNKS];
+	dma_cb_t l2[DVI_NOSYNC_LANE_CHUNKS];
+};
+
+static inline dma_cb_t* dvi_lane_from_list(struct dvi_scanline_dma_list *l, int i) {
+	return i == 0 ? l->l0 : i == 1 ? l->l1 : l->l2;
+}
+
+// Each TMDS lane uses one DMA channel to transfer data to a PIO state
+// machine, and another channel to load control blocks into this channel.
+struct dvi_lane_dma_cfg {
+	uint chan_ctrl;
+	uint chan_data;
+	void *tx_fifo;
+	uint dreq;
+};
+
+// Note these are already converted to pseudo-differential representation
+extern const uint32_t dvi_ctrl_syms[4];
+
+extern const struct dvi_timing dvi_timing_640x480p_60hz;
+extern const struct dvi_timing dvi_timing_800x480p_60hz;
+extern const struct dvi_timing dvi_timing_800x480p_30hz;
+extern const struct dvi_timing dvi_timing_800x600p_60hz;
+extern const struct dvi_timing dvi_timing_960x540p_60hz;
+extern const struct dvi_timing dvi_timing_1280x720p_30hz;
+
+extern const struct dvi_timing dvi_timing_800x600p_reduced_60hz;
+extern const struct dvi_timing dvi_timing_1280x720p_reduced_30hz;
+
+void dvi_timing_state_init(struct dvi_timing_state *t);
+
+void dvi_timing_state_advance(const struct dvi_timing *t, struct dvi_timing_state *s);
+
+void dvi_scanline_dma_list_init(struct dvi_scanline_dma_list *dma_list);
+
+void dvi_setup_scanline_for_vblank(const struct dvi_timing *t, const struct dvi_lane_dma_cfg dma_cfg[],
+		bool vsync_asserted, struct dvi_scanline_dma_list *l);
+
+void dvi_setup_scanline_for_active(const struct dvi_timing *t, const struct dvi_lane_dma_cfg dma_cfg[],
+		uint32_t *tmdsbuf, struct dvi_scanline_dma_list *l);
+
+void dvi_update_scanline_data_dma(const struct dvi_timing *t, const uint32_t *tmdsbuf, struct dvi_scanline_dma_list *l);
+
+#endif

src/libdvi/tmds_encode.S

@@ -0,0 +1,623 @@
+#include "hardware/regs/addressmap.h"
+#include "hardware/regs/sio.h"
+#include "dvi_config_defs.h"
+
+// Offsets suitable for ldr/str (must be <= 0x7c):
+#define ACCUM0_OFFS     (SIO_INTERP0_ACCUM0_OFFSET     - SIO_INTERP0_ACCUM0_OFFSET)
+#define ACCUM1_OFFS     (SIO_INTERP0_ACCUM1_OFFSET     - SIO_INTERP0_ACCUM0_OFFSET)
+#define ACCUM1_ADD_OFFS (SIO_INTERP0_ACCUM1_ADD_OFFSET - SIO_INTERP0_ACCUM0_OFFSET)
+#define PEEK0_OFFS      (SIO_INTERP0_PEEK_LANE0_OFFSET - SIO_INTERP0_ACCUM0_OFFSET)
+#define PEEK1_OFFS      (SIO_INTERP0_PEEK_LANE1_OFFSET - SIO_INTERP0_ACCUM0_OFFSET)
+#define PEEK2_OFFS      (SIO_INTERP0_PEEK_FULL_OFFSET  - SIO_INTERP0_ACCUM0_OFFSET)
+#define INTERP1         (SIO_INTERP1_ACCUM0_OFFSET     - SIO_INTERP0_ACCUM0_OFFSET)
+// Note the entirety of INTERP0 and INTERP1 fits inside this 5-bit
+// word-addressed space... almost as though it were intentional! :)
+
+.syntax unified
+.cpu cortex-m0plus
+.thumb
+
+.macro decl_func_x name
+.section .scratch_x.\name, "ax"
+.global \name
+.type \name,%function
+.thumb_func
+\name:
+.endm
+
+.macro decl_func_y name
+.section .scratch_y.\name, "ax"
+.global \name
+.type \name,%function
+.thumb_func
+\name:
+.endm
+
+#define decl_func decl_func_x
+
+// ----------------------------------------------------------------------------
+// Pixel-doubling encoders for RGB
+
+// r0: Input buffer (word-aligned)
+// r1: Output buffer (word-aligned)
+// r2: Input size (pixels)
+
+.macro do_channel_16bpp r_ibase r_inout0 r_out1
+	str \r_inout0, [\r_ibase, #ACCUM0_OFFS]
+	ldr \r_inout0, [\r_ibase, #PEEK0_OFFS]
+	ldr \r_inout0, [\r_inout0]
+	ldr \r_out1, [\r_ibase, #PEEK1_OFFS]
+	ldr \r_out1, [\r_out1]
+.endm
+
+decl_func tmds_encode_loop_16bpp
+	push {r4, r5, r6, r7, lr}
+	lsls r2, #2
+	add r2, r1
+	mov ip, r2
+	ldr r2, =(SIO_BASE + SIO_INTERP0_ACCUM0_OFFSET)
+	b 2f
+.align 2
+1:
+.rept TMDS_ENCODE_UNROLL
+	ldmia r0!, {r4, r6}
+	do_channel_16bpp r2, r4, r5
+	do_channel_16bpp r2, r6, r7
+	stmia r1!, {r4, r5, r6, r7}
+.endr
+2:
+	cmp r1, ip
+	bne 1b
+	pop {r4, r5, r6, r7, pc}
+
+// Same as above, but scale data to make up for lack of left shift
+// in interpolator (costs 1 cycle per 2 pixels)
+//
+// r0: Input buffer (word-aligned)
+// r1: Output buffer (word-aligned)
+// r2: Input size (pixels)
+// r3: Left shift amount
+
+decl_func tmds_encode_loop_16bpp_leftshift
+	push {r4, r5, r6, r7, lr}
+	lsls r2, #2
+	add r2, r1
+	mov ip, r2
+	ldr r2, =(SIO_BASE + SIO_INTERP0_ACCUM0_OFFSET)
+	b 2f
+.align 2
+1:
+.rept TMDS_ENCODE_UNROLL
+	ldmia r0!, {r4, r6}
+	lsls r4, r3
+	do_channel_16bpp r2, r4, r5
+	lsls r6, r3
+	do_channel_16bpp r2, r6, r7
+	stmia r1!, {r4, r5, r6, r7}
+.endr
+2:
+	cmp r1, ip
+	bne 1b
+	pop {r4, r5, r6, r7, pc}
+
+// r0: Input buffer (word-aligned)
+// r1: Output buffer (word-aligned)
+// r2: Input size (pixels)
+
+decl_func tmds_encode_loop_8bpp
+	push {r4, r5, r6, r7, lr}
+	lsls r2, #2
+	add r2, r1
+	mov ip, r2
+	ldr r2, =(SIO_BASE + SIO_INTERP0_ACCUM0_OFFSET)
+	b 2f
+.align 2
+1:
+.rept TMDS_ENCODE_UNROLL
+	ldmia  r0!, {r4}
+	str r4, [r2, #ACCUM0_OFFS + INTERP1]
+	str r4, [r2, #ACCUM0_OFFS]
+	ldr r4, [r2, #PEEK0_OFFS]
+	ldr r4, [r4]
+	ldr r5, [r2, #PEEK1_OFFS]
+	ldr r5, [r5]
+	ldr r6, [r2, #PEEK0_OFFS + INTERP1]
+	ldr r6, [r6]
+	ldr r7, [r2, #PEEK1_OFFS + INTERP1]
+	ldr r7, [r7]
+	stmia r1!, {r4, r5, r6, r7}
+.endr
+2:
+	cmp r1, ip
+	bne 1b
+	pop {r4, r5, r6, r7, pc}
+
+// r0: Input buffer (word-aligned)
+// r1: Output buffer (word-aligned)
+// r2: Input size (pixels)
+// r3: Left shift amount
+//
+// Note that only the data written to interp0 (pixel 0, 1) is leftshifted, not
+// the data written to interp1 (pixel 2, 3). Otherwise we always lose MSBs, as
+// the LUT offset MSB is at bit 8, so pixel 0 always requires some left shift,
+// since its channel MSBs are no greater than 7.
+
+decl_func tmds_encode_loop_8bpp_leftshift
+	push {r4, r5, r6, r7, lr}
+	lsls r2, #3
+	add r2, r1
+	mov ip, r2
+	ldr r2, =(SIO_BASE + SIO_INTERP0_ACCUM0_OFFSET)
+	b 2f
+.align 2
+1:
+.rept TMDS_ENCODE_UNROLL
+	ldmia  r0!, {r4}
+	str r4, [r2, #ACCUM0_OFFS + INTERP1]
+	lsls r4, r3
+	str r4, [r2, #ACCUM0_OFFS]
+	ldr r4, [r2, #PEEK0_OFFS]
+	ldr r4, [r4]
+	ldr r5, [r2, #PEEK1_OFFS]
+	ldr r5, [r5]
+	ldr r6, [r2, #PEEK0_OFFS + INTERP1]
+	ldr r6, [r6]
+	ldr r7, [r2, #PEEK1_OFFS + INTERP1]
+	ldr r7, [r7]
+	stmia r1!, {r4, r5, r6, r7}
+.endr
+2:
+	cmp r1, ip
+	bne 1b
+	pop {r4, r5, r6, r7, pc}
+
+// ----------------------------------------------------------------------------
+// Fast 1bpp black/white encoder (full res)
+
+// Taking the encoder from DVI spec, with initial balance 0:
+// 
+// - Encoding either 0x00 or 0xff will produce a running balance of -8, with
+//   output symbol of 0x100 or 0x200
+// 
+// - Subsequently encoding either 0x01 or 0xfe will return the balance to 0, with
+//  output symbol of 0x1ff or 0x2ff
+// 
+// So we can do 1bpp encode with a lookup of x coordinate LSB, and input
+// colour bit. If we process pixels in even-sized blocks, only the colour
+// lookup is needed.
+
+// Encode 8 pixels @ 1bpp (using two table lookups)
+// r3 contains lookup mask (preshifted)
+// r8 contains pointer to encode table
+// 2.125 cyc/pix
+.macro tmds_encode_1bpp_body shift_instr0 shamt0 shift_instr1 shamt1
+	\shift_instr0 r4, r2, #\shamt0
+	ands r4, r3
+	add r4, r8
+	ldmia r4, {r4, r5}
+	\shift_instr1 r6, r2, #\shamt1
+	ands r6, r3
+	add r6, r8
+	ldmia r6, {r6, r7}
+	stmia r1!, {r4, r5, r6, r7}
+.endm
+
+// r0: input buffer (word-aligned)
+// r1: output buffer (word-aligned)
+// r2: output pixel count
+decl_func tmds_encode_1bpp
+	push {r4-r7, lr}
+	mov r7, r8
+	push {r7}
+	lsls r2, #1
+	add r2, r1
+	mov ip, r2
+	adr r4, tmds_1bpp_table
+	mov r8, r4
+	// Mask: 4 bit index, 8 bytes per entry
+	movs r3, #0x78
+	b 2f
+1:
+	ldmia r0!, {r2}
+#if !DVI_1BPP_BIT_REVERSE
+	tmds_encode_1bpp_body lsls 3  lsrs 1
+	tmds_encode_1bpp_body lsrs 5  lsrs 9
+	tmds_encode_1bpp_body lsrs 13 lsrs 17
+	tmds_encode_1bpp_body lsrs 21 lsrs 25
+#else
+	tmds_encode_1bpp_body lsrs 1   lsls 3
+	tmds_encode_1bpp_body lsrs 9   lsrs 5
+	tmds_encode_1bpp_body lsrs 17  lsrs 13
+	tmds_encode_1bpp_body lsrs 25  lsrs 21
+#endif
+2:
+	cmp r1, ip
+	blo 1b
+
+	pop {r7}
+	mov r8, r7
+	pop {r4-r7, pc}
+
+.align 2
+tmds_1bpp_table:
+#if !DVI_1BPP_BIT_REVERSE
+	.word 0x7fd00, 0x7fd00  // 0000
+	.word 0x7fe00, 0x7fd00  // 0001
+	.word 0xbfd00, 0x7fd00  // 0010
+	.word 0xbfe00, 0x7fd00  // 0011
+	.word 0x7fd00, 0x7fe00  // 0100
+	.word 0x7fe00, 0x7fe00  // 0101
+	.word 0xbfd00, 0x7fe00  // 0110
+	.word 0xbfe00, 0x7fe00  // 0111
+	.word 0x7fd00, 0xbfd00  // 1000
+	.word 0x7fe00, 0xbfd00  // 1001
+	.word 0xbfd00, 0xbfd00  // 1010
+	.word 0xbfe00, 0xbfd00  // 1011
+	.word 0x7fd00, 0xbfe00  // 1100
+	.word 0x7fe00, 0xbfe00  // 1101
+	.word 0xbfd00, 0xbfe00  // 1110
+	.word 0xbfe00, 0xbfe00  // 1111
+#else
+	.word 0x7fd00, 0x7fd00  // 0000
+	.word 0x7fd00, 0xbfd00  // 1000
+	.word 0x7fd00, 0x7fe00  // 0100
+	.word 0x7fd00, 0xbfe00  // 1100
+	.word 0xbfd00, 0x7fd00  // 0010
+	.word 0xbfd00, 0xbfd00  // 1010
+	.word 0xbfd00, 0x7fe00  // 0110
+	.word 0xbfd00, 0xbfe00  // 1110
+	.word 0x7fe00, 0x7fd00  // 0001
+	.word 0x7fe00, 0xbfd00  // 1001
+	.word 0x7fe00, 0x7fe00  // 0101
+	.word 0x7fe00, 0xbfe00  // 1101
+	.word 0xbfe00, 0x7fd00  // 0011
+	.word 0xbfe00, 0xbfd00  // 1011
+	.word 0xbfe00, 0x7fe00  // 0111
+	.word 0xbfe00, 0xbfe00  // 1111
+#endif
+
+
+// ----------------------------------------------------------------------------
+// Full-resolution 2bpp encode (for 2bpp grayscale, or bitplaned RGB222)
+
+// Even-x-position pixels are encoded as symbols with imbalance -4, and odd
+// pixels with +4, so that we can mix-and-match our even/odd codewords and
+// always get a properly balanced sequence:
+//
+// level 0: (05 -> 103), then (04 -> 1fc)  (decimal 5, 4)
+// level 1: (50 -> 130), then (51 -> 1cf)  (decimal 80, 81)
+// level 2: (af -> 230), then (ae -> 2cf)  (decimal 175, 174)
+// level 3: (fa -> 203), then (fb -> 2fc)  (decimal 250, 251)
+//
+// These correspond to roughly 255 times (0, 1/3, 2/3, 1).
+//
+// Alternatively we could use symbols with 0 balance, which results in lower
+// contrast but avoids the LSB bobble:
+//
+// level 0: (10 -> 1f0) always
+// level 1: (5a -> 263) always
+// level 2: (a5 -> 163) always
+// level 3: (ef -> 2f0) always
+
+// Table base pointer in r0. Input pixels in r2.
+.macro encode_2bpp_body shift_instr shamt rd
+	\shift_instr \rd, r2, #\shamt
+	ands \rd, r3
+	ldr \rd, [r0, \rd]
+.endm
+
+// r0: input buffer (word-aligned)
+// r1: output buffer (word-aligned)
+// r2: output pixel count
+decl_func tmds_encode_2bpp
+	push {r4-r7, lr}
+	mov r7, r8
+	push {r7}
+	mov r8, r0
+	adr r0, tmds_2bpp_table
+	// Mask: 4-bit index into 4-byte entries.
+	movs r3, #0x3c
+	// Limit pointer: 1 word per 2 pixels
+	lsls r2, #1
+	add r2, r1
+	mov ip, r2
+	b 2f
+1:
+	mov r4, r8
+	ldmia r4!, {r2}
+	mov r8, r4
+	encode_2bpp_body lsls 2  r4
+	encode_2bpp_body lsrs 2  r5
+	encode_2bpp_body lsrs 6  r6
+	encode_2bpp_body lsrs 10 r7
+	stmia r1!, {r4-r7}
+	encode_2bpp_body lsrs 14 r4
+	encode_2bpp_body lsrs 18 r5
+	encode_2bpp_body lsrs 22 r6
+	encode_2bpp_body lsrs 26 r7
+	stmia r1!, {r4-r7}
+2:
+	cmp r1, ip
+	blo 1b
+	pop {r7}
+	mov r8, r7
+	pop {r4-r7, pc}
+
+.align 2
+tmds_2bpp_table:
+	.word 0x7f103 // 00, 00
+	.word 0x7f130 // 01, 00
+	.word 0x7f230 // 10, 00
+	.word 0x7f203 // 11, 00
+	.word 0x73d03 // 00, 01
+	.word 0x73d30 // 01, 01
+	.word 0x73e30 // 10, 01
+	.word 0x73e03 // 11, 01
+	.word 0xb3d03 // 00, 10
+	.word 0xb3d30 // 01, 10
+	.word 0xb3e30 // 10, 10
+	.word 0xb3e03 // 11, 10
+	.word 0xbf103 // 00, 11
+	.word 0xbf130 // 01, 11
+	.word 0xbf230 // 10, 11
+	.word 0xbf203 // 11, 11
+
+// ----------------------------------------------------------------------------
+// Full-resolution RGB encode (not very practical)
+
+// Non-doubled TMDS encode. 8.333 cycles per pixel, no exceptions. (This is
+// taking horizontal blanking (at VGA) and dual core into account, and
+// assuming the 3 channels are encoded individually.)
+//
+// Here is an idea
+// Have a table with a 7 bit lookup. The lookup is the 6 colour data bits (in
+// ACCUM0), concatenated with the sign bit of our running disparity (from
+// ACCUM1). Each table entry is a 20-bit TMDS symbol (pseudodifferential),
+// with the symbol's disparity stored left-justified in the upper 12 bits, as
+// e.g. a 6 bit signed integer.
+//
+// - Load pixel data.                        cyc: 0.75 (ldmia 2 words, every 4 pixels)
+// - Write pixel to ACCUM0.                  cyc: 1
+// - Read address from PEEK2.                cyc: 1
+// - Load encoded pixel from address.        cyc: 2
+// - Write disparity data to ACCUM1_ADD      cyc: 1
+// - Write encoded data to output buffer.    cyc: 1.25 (stmia 4 words, every 4 pixels)
+//
+// With decent register allocation we may be able to load 4 pixels at
+// once (2 words), and write 4 at once (4 words). This gives 7 cyc/pix.
+//
+// One issue is that the TMDS data in the bottom of ACCUM1 will eventually
+// overflow and affect the running disparity, but with 16 zeroes in between,
+// this would take much longer than one scanline, so everything is fine if
+// we clear the accumulator at the start of the scanline.
+//
+// Note that we need to use two interpolators to get the bits from both pixels
+// -- we are not outputting a single DC-balanced stream, but rather two
+// interleaved streams which are each DC-balanced. This is fine electrically,
+// but our output here will *NOT* match the TMDS encoder given in the DVI
+// spec.
+
+// You can define TMDS_FULLRES_NO_DC_BALANCE to disable the running balance
+// feedback. With the feedback enabled (default), the output is DC balanced,
+// but there are just barely enough CPU cycles to do all the encode, so it's
+// essentially a party trick. If you disable DC balancing, the performance is
+// much better, and many monitors will still accept the signals as long as you
+// DC couple your DVI signals.
+
+.macro tmds_fullres_encode_loop_body ra rb
+	str \ra, [r2, #ACCUM0_OFFS + INTERP1]
+	str \ra, [r2, #ACCUM0_OFFS]
+	ldr \ra, [r2, #PEEK2_OFFS]
+	ldr \ra, [\ra]
+#if !TMDS_FULLRES_NO_DC_BALANCE
+	str \ra, [r2, #ACCUM1_ADD_OFFS]
+#endif
+	ldr \rb, [r2, #PEEK2_OFFS + INTERP1]
+	ldr \rb, [\rb]
+#if !TMDS_FULLRES_NO_DC_BALANCE
+	str \rb, [r2, #ACCUM1_ADD_OFFS + INTERP1]
+#endif
+.endm
+
+// r0: Input buffer (word-aligned)
+// r1: Output buffer (word-aligned)
+// r2: Pixel count
+
+.macro tmds_fullres_encode_loop_16bpp
+	push {r4-r7, lr}
+	mov r4, r8
+	push {r4}
+
+
+	lsls r2, #2
+	add r2, r1
+	mov ip, r2
+	ldr r2, =(SIO_BASE + SIO_INTERP0_ACCUM0_OFFSET)
+	// DC balance defined to be 0 at start of scanline:
+	movs r4, #0
+	str r4, [r2, #ACCUM1_OFFS]
+#if TMDS_FULLRES_NO_DC_BALANCE
+	// Alternate parity between odd/even symbols if no feedback
+	mvns r4, r4
+#endif
+	str r4, [r2, #ACCUM1_OFFS + INTERP1]
+
+	// Keep loop start pointer in r8 so we can get a longer backward branch
+	adr r4, 1f
+	adds r4, #1 // god damn thumb bit why is this a thing
+	mov r8, r4
+	b 2f
+	.align 2
+1:
+.rept 16
+	ldmia r0!, {r4, r6}
+	tmds_fullres_encode_loop_body r4 r5
+	tmds_fullres_encode_loop_body r6 r7
+	stmia r1!, {r4, r5, r6, r7}
+.endr
+2:
+	cmp r1, ip
+	beq 1f
+	bx r8
+1:
+	pop {r4}
+	mov r8, r4
+	pop {r4-r7, pc}
+.endm
+
+// One copy each in X and Y, so the two cores don't step on each other
+decl_func_x tmds_fullres_encode_loop_16bpp_x
+	tmds_fullres_encode_loop_16bpp
+decl_func_y tmds_fullres_encode_loop_16bpp_y
+	tmds_fullres_encode_loop_16bpp
+
+
+.macro tmds_fullres_encode_loop_body_leftshift ra rb
+	// Note we apply the leftshift for INTERP0 only
+	str \ra, [r2, #ACCUM0_OFFS + INTERP1]
+	lsls \ra, r3
+	str \ra, [r2, #ACCUM0_OFFS]
+	ldr \ra, [r2, #PEEK2_OFFS]
+	ldr \ra, [\ra]
+#if !TMDS_FULLRES_NO_DC_BALANCE
+	str \ra, [r2, #ACCUM1_ADD_OFFS]
+#endif
+	ldr \rb, [r2, #PEEK2_OFFS + INTERP1]
+	ldr \rb, [\rb]
+#if !TMDS_FULLRES_NO_DC_BALANCE
+	str \rb, [r2, #ACCUM1_ADD_OFFS + INTERP1]
+#endif
+.endm
+
+// r0: Input buffer (word-aligned)
+// r1: Output buffer (word-aligned)
+// r2: Pixel count
+// r3: Left shift amount
+
+.macro tmds_fullres_encode_loop_16bpp_leftshift
+	push {r4-r7, lr}
+	mov r4, r8
+	mov r5, r9
+	push {r4-r5}
+
+	lsls r2, #2
+	add r2, r1
+	mov ip, r2
+	ldr r2, =(SIO_BASE + SIO_INTERP0_ACCUM0_OFFSET)
+	// DC balance defined to be 0 at start of scanline:
+	movs r4, #0
+	str r4, [r2, #ACCUM1_OFFS]
+#if TMDS_FULLRES_NO_DC_BALANCE
+	// Alternate parity between odd/even symbols if there's no balance feedback
+	mvns r4, r4
+#endif
+	str r4, [r2, #ACCUM1_OFFS + INTERP1]
+
+	adr r4, 1f
+	adds r4, #1
+	mov r8, r4
+	b 2f
+	.align 2
+1:
+.rept 16 // 64 pixels per iteration
+	ldmia r0!, {r4, r6}
+	tmds_fullres_encode_loop_body_leftshift r4 r5
+	tmds_fullres_encode_loop_body_leftshift r6 r7
+	stmia r1!, {r4, r5, r6, r7}
+.endr
+2:
+	cmp r1, ip
+	beq 1f
+	bx r8
+1:
+	pop {r4-r5}
+	mov r8, r4
+	mov r9, r5
+	pop {r4-r7, pc}
+.endm
+
+decl_func_x tmds_fullres_encode_loop_16bpp_leftshift_x
+	tmds_fullres_encode_loop_16bpp_leftshift
+decl_func_y tmds_fullres_encode_loop_16bpp_leftshift_y
+	tmds_fullres_encode_loop_16bpp_leftshift
+
+
+// ----------------------------------------------------------------------------
+// Full-resolution 8bpp paletted encode
+
+// Variant of tmds_fullres_encode_loop_16bpp that reads
+// 8-bit wide pixels packed 4 per word.  The interpolator
+// base is set to a reordered list of TMDS symbols based
+// on a user colour palette.
+
+// Two pixels input in rd[17:2]. Two symbols output in rd[19:0]. r2 contains
+// interp base pointer. r7 used as temporary.
+.macro tmds_palette_encode_loop_body rd
+	str \rd, [r2, #ACCUM0_OFFS]
+	str \rd, [r2, #ACCUM0_OFFS + INTERP1]
+	ldr \rd, [r2, #PEEK2_OFFS]
+	ldr \rd, [\rd]
+#if !TMDS_FULLRES_NO_DC_BALANCE
+	str \rd, [r2, #ACCUM1_ADD_OFFS]
+#endif
+	ldr r7, [r2, #PEEK2_OFFS + INTERP1]
+	ldr r7, [r7]
+#if !TMDS_FULLRES_NO_DC_BALANCE
+	str r7, [r2, #ACCUM1_ADD_OFFS + INTERP1]
+#endif
+	lsls r7, #10
+	orrs \rd, r7
+.endm
+
+.macro tmds_palette_encode_loop
+	push {r4-r7, lr}
+	mov r4, r8
+	push {r4}
+
+
+	lsls r2, #1
+	add r2, r1
+	mov ip, r2
+	ldr r2, =(SIO_BASE + SIO_INTERP0_ACCUM0_OFFSET)
+	// DC balance defined to be 0 at start of scanline:
+	movs r4, #0
+	str r4, [r2, #ACCUM1_OFFS]
+#if TMDS_FULLRES_NO_DC_BALANCE
+	// Alternate parity between odd/even symbols if there's no balance feedback
+	mvns r4, r4
+#endif
+	str r4, [r2, #ACCUM1_OFFS + INTERP1]
+
+	// Keep loop start pointer in r8 so we can get a longer backward branch
+	adr r4, 1f
+	adds r4, #1 // god damn thumb bit why is this a thing
+	mov r8, r4
+	b 2f
+	.align 2
+1:
+.rept 10
+	ldmia r0!, {r3, r5}
+	lsrs r4, r3, #14
+	lsls r3, #2
+	lsrs r6, r5, #14
+	lsls r5, #2
+	tmds_palette_encode_loop_body r3
+	tmds_palette_encode_loop_body r4
+	tmds_palette_encode_loop_body r5
+	tmds_palette_encode_loop_body r6
+	stmia r1!, {r3, r4, r5, r6}
+.endr
+2:
+	cmp r1, ip
+	beq 1f
+	bx r8
+1:
+	pop {r4}
+	mov r8, r4
+	pop {r4-r7, pc}
+.endm
+
+decl_func_x tmds_palette_encode_loop_x
+	tmds_palette_encode_loop
+decl_func_y tmds_palette_encode_loop_y
+	tmds_palette_encode_loop

src/libdvi/tmds_encode.c

@@ -0,0 +1,305 @@
+#include "hardware/interp.h"
+#include "tmds_encode.h"
+#include "hardware/gpio.h"
+#include "hardware/sync.h"
+
+static const uint32_t __scratch_x("tmds_table") tmds_table[] = {
+#include "tmds_table.h"
+};
+
+// Fullres table is bandwidth-critical, so gets one copy for each scratch
+// memory. There is a third copy which can go in flash, because it's just used
+// to generate palette LUTs. The ones we don't use will get garbage collected
+// during linking.
+const uint32_t __scratch_x("tmds_table_fullres_x") tmds_table_fullres_x[] = {
+#include "tmds_table_fullres.h"
+};
+
+const uint32_t __scratch_y("tmds_table_fullres_y") tmds_table_fullres_y[] = {
+#include "tmds_table_fullres.h"
+};
+
+// Configure an interpolator to extract a single colour channel from each of a pair
+// of pixels, with the first pixel's lsb at pixel_lsb, and the pixels being
+// pixel_width wide. Produce a LUT address for the first pixel's colour data on
+// LANE0, and the second pixel's colour data on LANE1.
+//
+// Returns nonzero if the *_leftshift variant of the encoder loop must be used
+// (needed for blue channel because I was a stubborn idiot and didn't put
+// signed/bidirectional shift on interpolator, very slightly slower). The
+// return value is the size of left shift required.
+
+static int __not_in_flash_func(configure_interp_for_addrgen)(interp_hw_t *interp, uint channel_msb, uint channel_lsb, uint pixel_lsb, uint pixel_width, uint lut_index_width, const uint32_t *lutbase) {
+	interp_config c;
+	const uint index_shift = 2; // scaled lookup for 4-byte LUT entries
+
+	int shift_channel_to_index = pixel_lsb + channel_msb - (lut_index_width - 1) - index_shift;
+	int oops = 0;
+	if (shift_channel_to_index < 0) {
+		// "It's ok we'll fix it in software"
+		oops = -shift_channel_to_index;
+		shift_channel_to_index = 0;
+	}
+
+	uint index_msb = index_shift + lut_index_width - 1;
+
+	c = interp_default_config();
+	interp_config_set_shift(&c, shift_channel_to_index);
+	interp_config_set_mask(&c, index_msb - (channel_msb - channel_lsb), index_msb);
+	interp_set_config(interp, 0, &c);
+
+	c = interp_default_config();
+	interp_config_set_shift(&c, pixel_width	+ shift_channel_to_index);
+	interp_config_set_mask(&c, index_msb - (channel_msb - channel_lsb), index_msb);
+	interp_config_set_cross_input(&c, true);
+	interp_set_config(interp, 1, &c);
+
+	interp->base[0] = (uint32_t)lutbase;
+	interp->base[1] = (uint32_t)lutbase;
+
+	return oops;
+}
+
+// Extract up to 6 bits from a buffer of 16 bit pixels, and produce a buffer
+// of TMDS symbols from this colour channel. Number of pixels must be even,
+// pixel buffer must be word-aligned.
+
+void __not_in_flash_func(tmds_encode_data_channel_16bpp)(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint channel_msb, uint channel_lsb) {
+	interp_hw_save_t interp0_save;
+	interp_save(interp0_hw, &interp0_save);
+	int require_lshift = configure_interp_for_addrgen(interp0_hw, channel_msb, channel_lsb, 0, 16, 6, tmds_table);
+	if (require_lshift)
+		tmds_encode_loop_16bpp_leftshift(pixbuf, symbuf, n_pix, require_lshift);
+	else
+		tmds_encode_loop_16bpp(pixbuf, symbuf, n_pix);
+	interp_restore(interp0_hw, &interp0_save);
+}
+
+// As above, but 8 bits per pixel, multiple of 4 pixels, and still word-aligned.
+void __not_in_flash_func(tmds_encode_data_channel_8bpp)(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint channel_msb, uint channel_lsb) {
+	interp_hw_save_t interp0_save, interp1_save;
+	interp_save(interp0_hw, &interp0_save);
+	interp_save(interp1_hw, &interp1_save);
+	// Note that for 8bpp, some left shift is always required for pixel 0 (any
+	// channel), which destroys some MSBs of pixel 3. To get around this, pixel
+	// data sent to interp1 is *not left-shifted*
+	int require_lshift = configure_interp_for_addrgen(interp0_hw, channel_msb, channel_lsb, 0, 8, 6, tmds_table);
+	int lshift_upper = configure_interp_for_addrgen(interp1_hw, channel_msb, channel_lsb, 16, 8, 6, tmds_table);
+	assert(!lshift_upper); (void)lshift_upper;
+	if (require_lshift)	
+		tmds_encode_loop_8bpp_leftshift(pixbuf, symbuf, n_pix, require_lshift);
+	else
+		tmds_encode_loop_8bpp(pixbuf, symbuf, n_pix);
+	interp_restore(interp0_hw, &interp0_save);
+	interp_restore(interp1_hw, &interp1_save);
+}
+
+// ----------------------------------------------------------------------------
+// Code for full-resolution TMDS encode (barely possible, utterly impractical):
+
+// Different scheme used for full res as the fun pixel-doubling DC balance
+// trick doesn't work, so we need to actually do running disparity. ACCUM0 has
+// pixel data, ACCUM1 has running disparity. INTERP0 is used to process even
+// pixels, and INTERP1 for odd pixels. Note this means that even and odd
+// symbols have their DC balance handled separately, which is not to spec.
+
+static int __not_in_flash_func(configure_interp_for_addrgen_fullres)(interp_hw_t *interp, uint channel_msb, uint channel_lsb, uint lut_index_width, const uint32_t *lutbase) {
+	const uint index_shift = 2; // scaled lookup for 4-byte LUT entries
+
+	int shift_channel_to_index = channel_msb - (lut_index_width - 1) - index_shift;
+	int oops = 0;
+	if (shift_channel_to_index < 0) {
+		// "It's ok we'll fix it in software"
+		oops = -shift_channel_to_index;
+		shift_channel_to_index = 0;
+	}
+
+	uint index_msb = index_shift + lut_index_width - 1;
+
+	interp_config c;
+	// Shift and mask colour channel to lower 6 bits of LUT index (note lut_index_width excludes disparity sign)
+	c = interp_default_config();
+	interp_config_set_shift(&c, shift_channel_to_index);
+	interp_config_set_mask(&c, index_msb - (channel_msb - channel_lsb), index_msb);
+	interp_set_config(interp, 0, &c);
+
+	// Concatenate disparity (ACCUM1) sign onto the LUT index
+	c = interp_default_config();
+	interp_config_set_shift(&c, 30 - index_msb);
+	interp_config_set_mask(&c, index_msb + 1, index_msb + 1);
+	interp_set_config(interp, 1, &c);
+
+	interp->base[2] = (uint32_t)lutbase;
+
+	return oops;
+}
+
+void __not_in_flash_func(tmds_encode_data_channel_fullres_16bpp)(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint channel_msb, uint channel_lsb) {
+	uint core = get_core_num();
+#if !TMDS_FULLRES_NO_INTERP_SAVE
+	interp_hw_save_t interp0_save, interp1_save;
+	interp_save(interp0_hw, &interp0_save);
+	interp_save(interp1_hw, &interp1_save);
+#endif
+
+	// There is a copy of the inner loop and the LUT in both scratch X and
+	// scratch Y memories. Use X on core 1 and Y on core 0 so the cores don't
+	// tread on each other's toes too much.
+	const uint32_t *lutbase = core ? tmds_table_fullres_x : tmds_table_fullres_y;
+	int lshift_lower = configure_interp_for_addrgen_fullres(interp0_hw, channel_msb, channel_lsb, 6, lutbase);
+	int lshift_upper = configure_interp_for_addrgen_fullres(interp1_hw, channel_msb + 16, channel_lsb + 16, 6, lutbase);
+	assert(!lshift_upper); (void)lshift_upper;
+	if (lshift_lower) {
+		(core ?
+			tmds_fullres_encode_loop_16bpp_leftshift_x :
+			tmds_fullres_encode_loop_16bpp_leftshift_y
+		)(pixbuf, symbuf, n_pix, lshift_lower);
+	}
+	else {
+		(core ?
+			tmds_fullres_encode_loop_16bpp_x :
+			tmds_fullres_encode_loop_16bpp_y
+		)(pixbuf, symbuf, n_pix);
+	}
+#if !TMDS_FULLRES_NO_INTERP_SAVE
+	interp_restore(interp0_hw, &interp0_save);
+	interp_restore(interp1_hw, &interp1_save);
+#endif
+}
+
+static const int8_t imbalance_lookup[16] = { -4, -2, -2, 0, -2, 0, 0, 2, -2, 0, 0, 2, 0, 2, 2, 4 };
+
+static inline int byte_imbalance(uint32_t x)
+{
+	return imbalance_lookup[x >> 4] + imbalance_lookup[x & 0xF];
+}
+
+static void tmds_encode_symbols(uint8_t pixel, uint32_t* negative_balance_sym, uint32_t* positive_balance_sym)
+{
+	int pixel_imbalance = byte_imbalance(pixel);
+	uint32_t sym = pixel & 1;
+	if (pixel_imbalance > 0 || (pixel_imbalance == 0 && sym == 0)) {
+		for (int i = 0; i < 7; ++i) {
+			sym |= (~((sym >> i) ^ (pixel >> (i + 1))) & 1) << (i + 1);
+		}
+	}
+	else {
+		for (int i = 0; i < 7; ++i) {
+			sym |= ( ((sym >> i) ^ (pixel >> (i + 1))) & 1) << (i + 1);
+		}
+		sym |= 0x100;
+	}
+
+	int imbalance = byte_imbalance(sym & 0xFF);
+  if (imbalance == 0) {
+		if ((sym & 0x100) == 0) sym ^= 0x2ff;
+		*positive_balance_sym = sym;
+		*negative_balance_sym = sym;
+		return;
+	}
+	else if (imbalance > 0) {
+		*negative_balance_sym = (sym ^ 0x2ff) | (((-imbalance + imbalance_lookup[2 ^ (sym >> 8)] + 2) & 0x3F) << 26);
+		*positive_balance_sym = sym | ((imbalance + imbalance_lookup[sym >> 8] + 2) << 26);
+	}
+	else {
+		*negative_balance_sym = sym | (((imbalance + imbalance_lookup[sym >> 8] + 2) & 0x3F) << 26);
+		*positive_balance_sym = (sym ^ 0x2ff) | ((-imbalance + imbalance_lookup[2 ^ (sym >> 8)] + 2) << 26);
+	}
+}
+
+// This takes a 16-bit (RGB 565) colour palette and makes palettes of TMDS symbols suitable
+// for performing fullres encode.
+// The TMDS palette buffer should be 6 * n_palette words long.
+// n_palette must be a power of 2 <= 256.
+void tmds_setup_palette_symbols(const uint16_t *palette, uint32_t *tmds_palette, size_t n_palette) {
+	uint32_t* tmds_palette_blue = tmds_palette;
+	uint32_t* tmds_palette_green = tmds_palette + 2 * n_palette;
+	uint32_t* tmds_palette_red = tmds_palette + 4 * n_palette;
+	for (int i = 0; i < n_palette; ++i) {
+		uint16_t blue = (palette[i] << 3) & 0xf8;
+		uint16_t green = (palette[i] >> 3) & 0xfc;
+		uint16_t red = (palette[i] >> 8) & 0xf8;
+		tmds_encode_symbols(blue, &tmds_palette_blue[i], &tmds_palette_blue[i + n_palette]);
+		tmds_encode_symbols(green, &tmds_palette_green[i], &tmds_palette_green[i + n_palette]);
+		tmds_encode_symbols(red, &tmds_palette_red[i], &tmds_palette_red[i + n_palette]);
+	}
+}
+
+// This takes a 24-bit (RGB 888) colour palette and makes palettes of TMDS symbols suitable
+// for performing fullres encode.
+// The TMDS palette buffer should be 6 * n_palette words long.
+// n_palette must be a power of 2 <= 256.
+void tmds_setup_palette24_symbols(const uint32_t *palette, uint32_t *tmds_palette, size_t n_palette) {
+	uint32_t* tmds_palette_blue = tmds_palette;
+	uint32_t* tmds_palette_green = tmds_palette + 2 * n_palette;
+	uint32_t* tmds_palette_red = tmds_palette + 4 * n_palette;
+	for (int i = 0; i < n_palette; ++i) {
+		uint16_t blue = palette[i] & 0xff;
+		uint16_t green = (palette[i] >> 8) & 0xff;
+		uint16_t red = (palette[i] >> 16) & 0xff;
+		tmds_encode_symbols(blue, &tmds_palette_blue[i], &tmds_palette_blue[i + n_palette]);
+		tmds_encode_symbols(green, &tmds_palette_green[i], &tmds_palette_green[i + n_palette]);
+		tmds_encode_symbols(red, &tmds_palette_red[i], &tmds_palette_red[i + n_palette]);
+	}
+}
+
+// Encode palette data for all 3 channels.
+// pixbuf is an array of n_pix 8-bit wide pixels containing palette values (32-bit word aligned)
+// tmds_palette is a palette of TMDS symbols produced by tmds_setup_palette_symbols
+// symbuf is 3*n_pix 32-bit words, this function writes the symbol values for each of the channels to it.
+void __not_in_flash_func(tmds_encode_palette_data)(const uint32_t *pixbuf, const uint32_t *tmds_palette, uint32_t *symbuf, size_t n_pix, uint32_t palette_bits) {
+	uint core = get_core_num();
+#if !TMDS_FULLRES_NO_INTERP_SAVE
+	interp_hw_save_t interp0_save, interp1_save;
+	interp_save(interp0_hw, &interp0_save);
+	interp_save(interp1_hw, &interp1_save);
+#endif
+
+	interp0_hw->base[2] = (uint32_t)tmds_palette;
+	interp1_hw->base[2] = (uint32_t)tmds_palette;
+
+	// Lane 0 on both interpolators masks the palette bits, starting at bit 2,
+	// The second interpolator also shifts to read the 2nd or 4th byte of the word.
+	interp0_hw->ctrl[0] =
+		(2 << SIO_INTERP0_CTRL_LANE0_MASK_LSB_LSB) |
+		((palette_bits + 1) << SIO_INTERP0_CTRL_LANE0_MASK_MSB_LSB);
+	interp1_hw->ctrl[0] =
+		(8 << SIO_INTERP0_CTRL_LANE0_SHIFT_LSB) |
+		(2 << SIO_INTERP0_CTRL_LANE0_MASK_LSB_LSB) |
+		((palette_bits + 1) << SIO_INTERP0_CTRL_LANE0_MASK_MSB_LSB);
+
+	// Lane 1 shifts and masks the sign bit into the right position to add to the symbol
+	// table index to choose the negative disparity symbols if the sign is negative.
+	const uint32_t ctrl_lane_1 =
+		((31 - (palette_bits + 2)) << SIO_INTERP0_CTRL_LANE0_SHIFT_LSB) |
+		(palette_bits + 2) * ((1 << SIO_INTERP0_CTRL_LANE0_MASK_LSB_LSB) | (1 << SIO_INTERP0_CTRL_LANE0_MASK_MSB_LSB));
+	interp0_hw->ctrl[1] = ctrl_lane_1;
+	interp1_hw->ctrl[1] = ctrl_lane_1;
+
+	if (core) {
+		tmds_palette_encode_loop_x(pixbuf, symbuf, n_pix);
+
+		interp0_hw->base[2] = (uint32_t)(tmds_palette + (2 << palette_bits));
+		interp1_hw->base[2] = (uint32_t)(tmds_palette + (2 << palette_bits));
+		tmds_palette_encode_loop_x(pixbuf, symbuf + (n_pix >> 1), n_pix);
+
+		interp0_hw->base[2] = (uint32_t)(tmds_palette + (4 << palette_bits));
+		interp1_hw->base[2] = (uint32_t)(tmds_palette + (4 << palette_bits));
+		tmds_palette_encode_loop_x(pixbuf, symbuf + n_pix, n_pix);
+	} else {
+		tmds_palette_encode_loop_y(pixbuf, symbuf, n_pix);
+
+		interp0_hw->base[2] = (uint32_t)(tmds_palette + (2 << palette_bits));
+		interp1_hw->base[2] = (uint32_t)(tmds_palette + (2 << palette_bits));
+		tmds_palette_encode_loop_y(pixbuf, symbuf + (n_pix >> 1), n_pix);
+
+		interp0_hw->base[2] = (uint32_t)(tmds_palette + (4 << palette_bits));
+		interp1_hw->base[2] = (uint32_t)(tmds_palette + (4 << palette_bits));
+		tmds_palette_encode_loop_y(pixbuf, symbuf + n_pix, n_pix);
+	}
+
+#if !TMDS_FULLRES_NO_INTERP_SAVE
+	interp_restore(interp0_hw, &interp0_save);
+	interp_restore(interp1_hw, &interp1_save);
+#endif
+}

src/libdvi/tmds_encode.h

@@ -0,0 +1,46 @@
+#ifndef _TMDS_ENCODE_H_
+#define _TMDS_ENCODE_H_
+
+#include "hardware/interp.h"
+#include "dvi_config_defs.h"
+
+#if defined(__cplusplus)
+extern "C"
+{
+#endif
+
+// Functions from tmds_encode.c
+void tmds_encode_data_channel_16bpp(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint channel_msb, uint channel_lsb);
+void tmds_encode_data_channel_8bpp(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint channel_msb, uint channel_lsb);
+void tmds_encode_data_channel_fullres_16bpp(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint channel_msb, uint channel_lsb);
+void tmds_setup_palette_symbols(const uint16_t *palette, uint32_t *symbuf, size_t n_palette);
+void tmds_setup_palette24_symbols(const uint32_t *palette, uint32_t *symbuf, size_t n_palette);
+void tmds_encode_palette_data(const uint32_t *pixbuf, const uint32_t *tmds_palette, uint32_t *symbuf, size_t n_pix, uint32_t palette_bits);
+
+// Functions from tmds_encode.S
+
+void tmds_encode_1bpp(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix);
+void tmds_encode_2bpp(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix);
+
+// Uses interp0:
+void tmds_encode_loop_16bpp(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix);
+void tmds_encode_loop_16bpp_leftshift(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint leftshift);
+
+// Uses interp0 and interp1:
+void tmds_encode_loop_8bpp(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix);
+void tmds_encode_loop_8bpp_leftshift(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint leftshift);
+
+// Uses interp0 and interp1:
+// (Note a copy is provided in scratch memories X and Y)
+void tmds_fullres_encode_loop_16bpp_x(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix);
+void tmds_fullres_encode_loop_16bpp_y(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix);
+void tmds_fullres_encode_loop_16bpp_leftshift_x(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint leftshift);
+void tmds_fullres_encode_loop_16bpp_leftshift_y(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix, uint leftshift);
+void tmds_palette_encode_loop_x(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix);
+void tmds_palette_encode_loop_y(const uint32_t *pixbuf, uint32_t *symbuf, size_t n_pix);
+
+#if defined(__cplusplus)
+}
+#endif
+
+#endif

src/libdvi/tmds_encode_1bpp.pio

@@ -0,0 +1,46 @@
+.program tmds_encode_1bpp
+
+; 1bpp black/white pixels go in, TMDS symbols come out.
+; Each output word contains two output symbols, each 10 bits in size,
+; right-justified. The least-significant symbol is displayed first.
+;
+; We can encode using the following LUT: (yes this is compliant)
+;
+; x % 2 | colour | symbol
+; ------+--------+-------
+; 0     | 0      | 0x100
+; 0     | 1      | 0x200
+; 1     | 0      | 0x1ff
+; 1     | 1      | 0x2ff
+;
+; OSR: shift to right, autopull, threshold 32
+; ISR: shift to right, autopush, threshold 24
+;
+; Note the ISR needs to be shifted to *right* so that we can get the first
+; pixel in the less-significant position. Threshold 24 so we can get 8x 0-bits
+; at the LSBs for free :)
+
+even_pixel:
+    out x, 1
+    mov y, ~x
+    in y, 1
+    in x, 1
+
+odd_pixel:
+    mov x, ~null
+    in x, 8
+    out x, 1
+    mov y, ~x
+    in y, 1
+    in x, 13     ; Bring total shift to 24, triggering push.
+
+% c-sdk {
+static inline void tmds_encode_1bpp_init(PIO pio, uint sm) {
+    uint offset = pio_add_program(pio, &tmds_encode_1bpp_program);
+    pio_sm_config c = tmds_encode_1bpp_program_get_default_config(offset);
+    sm_config_set_out_shift(&c, true, true, 32);
+    sm_config_set_in_shift(&c, true, true, 24);
+    pio_sm_init(pio, sm, offset, &c);
+    pio_sm_set_enabled(pio, sm, true);
+}
+%}

src/libdvi/tmds_table.h

@@ -0,0 +1,76 @@
+// Generated from tmds_table_gen.py
+//
+// This table converts a 6 bit data input into a pair of TMDS data symbols
+// with data content *almost* equal (1 LSB off) to input value left shifted by
+// two. The pairs of symbols have a net DC balance of 0.
+//
+// The two symbols are concatenated in the 20 LSBs of a data word, with the
+// first symbol in least-significant position.
+//
+// Note the declaration isn't included here, just the table body. This is in
+// case you want multiple copies of the table in different SRAMs (particularly
+// scratch X/Y).
+0x7fd00u,
+0x40dfcu,
+0x41df8u,
+0x7ed04u,
+0x43df0u,
+0x7cd0cu,
+0x7dd08u,
+0x42df4u,
+0x47de0u,
+0x78d1cu,
+0x79d18u,
+0x46de4u,
+0x7bd10u,
+0x44decu,
+0x45de8u,
+0xafa41u,
+0x4fdc0u,
+0x70d3cu,
+0x71d38u,
+0x4edc4u,
+0x73d30u,
+0x4cdccu,
+0x4ddc8u,
+0xa7a61u,
+0x77d20u,
+0x48ddcu,
+0x49dd8u,
+0xa3a71u,
+0x4bdd0u,
+0xa1a79u,
+0xa0a7du,
+0x9fa81u,
+0x5fd80u,
+0x60d7cu,
+0x61d78u,
+0x5ed84u,
+0x63d70u,
+0x5cd8cu,
+0x5dd88u,
+0xb7a21u,
+0x67d60u,
+0x58d9cu,
+0x59d98u,
+0xb3a31u,
+0x5bd90u,
+0xb1a39u,
+0xb0a3du,
+0x8fac1u,
+0x6fd40u,
+0x50dbcu,
+0x51db8u,
+0xbba11u,
+0x53db0u,
+0xb9a19u,
+0xb8a1du,
+0x87ae1u,
+0x57da0u,
+0xbda09u,
+0xbca0du,
+0x83af1u,
+0xbea05u,
+0x81af9u,
+0x80afdu,
+0xbfa01u,

src/libdvi/tmds_table_fullres.h

@@ -0,0 +1,139 @@
+// Each entry consists of a 10 bit TMDS symbol in pseudo-differential format
+// (10 LSBs) and the symbol's disparity as a 6 bit signed integer (the 6
+// MSBs). There is a 16 bit gap in between them, which is actually vital for
+// the way the TMDS encode works!
+//
+// There are 128 1-word entries. The lookup index should be the concatenation
+// of the sign bit of current running disparity, with 6 bits of colour channel
+// data.
+
+// Non-negative running disparity:
+0xe0000100,
+0xf8000303,
+0x00000307,
+0xe8000104,
+0x000001f0,
+0xf000010c,
+0xe8000108,
+0x0000030b,
+0xf80001e0,
+0xf800011c,
+0xf0000118,
+0x000001e4,
+0xe8000110,
+0x00000313,
+0x000001e8,
+0xf0000241,
+0xf00001c0,
+0x0000013c,
+0xf8000138,
+0xf80001c4,
+0xf0000130,
+0x000001cc,
+0xf80001c8,
+0xf8000261,
+0xe8000120,
+0x00000323,
+0x000001d8,
+0x00000271,
+0xf80001d0,
+0xf0000086,
+0xe8000082,
+0xf0000281,
+0xe8000180,
+0x00000383,
+0x00000178,
+0xf0000184,
+0xf8000170,
+0xf800018c,
+0xf0000188,
+0xf0000221,
+0xf0000160,
+0x0000019c,
+0xf8000198,
+0xf8000231,
+0xf0000190,
+0x00000239,
+0xf00000c2,
+0xf80002c1,
+0xe8000140,
+0x00000343,
+0x000001b8,
+0xf0000211,
+0xf80001b0,
+0xf8000219,
+0x0000021d,
+0x000002e1,
+0xf00001a0,
+0xf0000209,
+0xf800020d,
+0xf000000e,
+0xf0000205,
+0xe8000006,
+0xe0000002,
+0xe8000201,
+// Negative running disparity:
+0x280003ff,
+0x100001fc,
+0x080001f8,
+0x200003fb,
+0x000001f0,
+0x180003f3,
+0x200003f7,
+0x080001f4,
+0x1000031f,
+0x100003e3,
+0x180003e7,
+0x000001e4,
+0x200003ef,
+0x080001ec,
+0x000001e8,
+0x080000be,
+0x1800033f,
+0x0000013c,
+0x100003c7,
+0x1000033b,
+0x180003cf,
+0x000001cc,
+0x10000337,
+0x0000009e,
+0x200003df,
+0x080001dc,
+0x000001d8,
+0x00000271,
+0x1000032f,
+0x08000279,
+0x1000027d,
+0x0800007e,
+0x2000037f,
+0x0800017c,
+0x00000178,
+0x1800037b,
+0x1000038f,
+0x10000373,
+0x18000377,
+0x080000de,
+0x1800039f,
+0x0000019c,
+0x10000367,
+0x000000ce,
+0x1800036f,
+0x00000239,
+0x0800023d,
+0x0000003e,
+0x200003bf,
+0x080001bc,
+0x000001b8,
+0x080000ee,
+0x1000034f,
+0x000000e6,
+0x0000021d,
+0x000002e1,
+0x1800035f,
+0x080000f6,
+0x000000f2,
+0x080002f1,
+0x080000fa,
+0x100002f9,
+0x180002fd,
+0x100000fe,

src/libdvi/tmds_table_gen.py

@@ -0,0 +1,150 @@
+#!/usr/bin/env python3
+
+# The key fact is that, if x is even, and the encoder currently has a running
+# imbalance of 0, encoding x followed by x + 1 produces a symbol pair with a
+# net balance of 0.
+#
+# This is a reasonable constraint, because we only want RGB565 (so 6 valid
+# channel data bits -> data is multiple of 4), and can probably tolerate
+# 0.25LSB of noise :)
+#
+# This means that encoding a half-horizontal-resolution scanline buffer is a
+# simple LUT operation for each colour channel, because we have made the
+# encoding process stateless by guaranteeing 0 balance.
+
+def popcount(x):
+	n = 0
+	while x:
+		n += 1
+		x = x & (x - 1)
+	return n
+
+# Equivalent to N1(q) - N0(q) in the DVI spec
+def byteimbalance(x):
+	return 2 * popcount(x) - 8
+
+# This is a direct translation of "Figure 3-5. T.M.D.S. Encode Algorithm" on
+# page 29 of DVI 1.0 spec
+
+class TMDSEncode:
+	ctrl_syms = {
+		0b00: 0b1101010100,
+		0b01: 0b0010101011,
+		0b10: 0b0101010100,
+		0b11: 0b1010101011
+	}
+	def __init__(self):
+		self.imbalance = 0
+
+	def encode(self, d, c, de):
+		if not de:
+			self.imbalance = 0
+			return self.ctrl_syms[c]
+		# Minimise transitions
+		q_m = d & 0x1
+		if popcount(d) > 4 or (popcount(d) == 4 and not d & 0x1):
+			for i in range(7):
+				q_m = q_m | (~(q_m >> i ^ d >> i + 1) & 0x1) << i + 1
+		else:
+			for i in range(7):
+				q_m = q_m | ( (q_m >> i ^ d >> i + 1) & 0x1) << i + 1
+			q_m = q_m | 0x100
+		# Correct DC balance
+		inversion_mask = 0x2ff
+		q_out = 0
+		if self.imbalance == 0 or byteimbalance(q_m & 0xff) == 0:
+			q_out = q_m ^ (0 if q_m & 0x100 else inversion_mask)
+			if q_m & 0x100:
+				self.imbalance += byteimbalance(q_m & 0xff)
+			else:
+				self.imbalance -= byteimbalance(q_m & 0xff)
+		elif (self.imbalance > 0) == (byteimbalance(q_m & 0xff) > 0):
+			q_out = q_m ^ inversion_mask
+			self.imbalance += ((q_m & 0x100) >> 7) - byteimbalance(q_m & 0xff)
+		else:
+			q_out = q_m
+			self.imbalance += byteimbalance(q_m & 0xff) - ((~q_m & 0x100) >> 7)
+		return q_out
+
+# Turn a bitmap of width n into n pairs of pseudo-differential bits
+def differentialise(x, n):
+	accum = 0
+	for i in range(n):
+		accum <<= 2
+		if x & (1 << (n - 1)):
+			accum |= 0b01
+		else:
+			accum |= 0b10
+		x <<= 1
+	return accum
+
+enc = TMDSEncode()
+
+
+###
+# Pixel-doubled table:
+
+# for i in range(0, 256, 4):
+# 	sym0 = enc.encode(i, 0, 1)
+# 	sym1 = enc.encode(i ^ 1, 0, 1)
+# 	assert(enc.imbalance == 0)
+# 	print(f"0x{sym0 | (sym1 << 10):05x}u,")
+
+###
+# Fullres 1bpp table: (each entry is 2 words, 4 pixels)
+
+# (note trick here is that encoding 0x00 or 0xff sets imbalance to -8, and
+# (encoding 0x01 or 0xfe returns imbalance to 0, so we alternate between these
+# (two pairs of dark/light colours. Creates some fairly subtle vertical
+# (banding, but it's cheap.
+
+# for i in range(1 << 4):
+# 	syms = list(enc.encode((0xff if i & 1 << j else 0) ^ j & 0x01, 0, 1) for j in range(4))
+# 	print(f"0x{syms[0] | syms[1] << 10:05x}, 0x{syms[2] | syms[3] << 10:05x}")
+# 	assert(enc.imbalance == 0)
+
+###
+# Fullres table stuff:
+
+# def disptable_format(sym):
+# 	return sym | ((popcount(sym) * 2 - 10 & 0x3f) << 26)
+
+# print("// Non-negative running disparity:")
+# for i in range(0, 256, 4):
+# 	enc.imbalance = 1
+# 	print("0x{:08x},".format(disptable_format(enc.encode(i, 0, 1))))
+
+# print("// Negative running disparity:")
+# for i in range(0, 256, 4):
+# 	enc.imbalance = -1
+# 	print("0x{:08x},".format(disptable_format(enc.encode(i, 0, 1))))
+
+###
+# Control symbols:
+
+# for i in range(4):
+# 	sym = enc.encode(0, i, 0)
+# 	print(f"0x{sym << 10 | sym:05x},")
+
+
+###
+# Find zero-balance symbols:
+
+# for i in range(256):
+# 	enc.imbalance = 0
+# 	sym = enc.encode(i, 0, 1)
+# 	if enc.imbalance == 0:
+# 		print(f"{i:02x}: {sym:03x}")
+
+###
+# Generate 2bpp table based on above experiment:
+
+levels_2bpp_even = [0x05, 0x50, 0xaf, 0xfa]
+levels_2bpp_odd  = [0x04, 0x51, 0xae, 0xfb]
+
+for i1, p1 in enumerate(levels_2bpp_odd):
+	for i0, p0 in enumerate(levels_2bpp_even):
+		sym0 = enc.encode(p0, 0, 1)
+		sym1 = enc.encode(p1, 0, 1)
+		assert(enc.imbalance == 0)
+		print(f".word 0x{sym1 << 10 | sym0:05x} // {i0:02b}, {i1:02b}")

src/libdvi/util_queue_u32_inline.h

@@ -0,0 +1,83 @@
+#ifndef _UTIL_QUEUE_U32_INLINE_H
+#define _UTIL_QUEUE_U32_INLINE_H
+
+// Faster versions of the functions found in pico/util/queue.h, for the common
+// case of 32-bit-sized elements. Can be used on the same queue data
+// structure, and mixed freely with the generic access methods, as long as
+// element_size == 4.
+
+#include "pico/util/queue.h"
+#include "hardware/sync.h"
+
+static inline uint16_t _queue_inc_index_u32(queue_t *q, uint16_t index) {
+    if (++index > q->element_count) { // > because we have element_count + 1 elements
+        index = 0;
+    }
+    return index;
+}
+
+static inline bool queue_try_add_u32(queue_t *q, void *data) {
+    bool success = false;
+    uint32_t flags = spin_lock_blocking(q->core.spin_lock);
+    if (queue_get_level_unsafe(q) != q->element_count) {
+        ((uint32_t*)q->data)[q->wptr] = *(uint32_t*)data;
+        q->wptr = _queue_inc_index_u32(q, q->wptr);
+        success = true;
+    }
+    spin_unlock(q->core.spin_lock, flags);
+    if (success) __sev();
+    return success;
+}
+
+static inline bool queue_try_remove_u32(queue_t *q, void *data) {
+    bool success = false;
+    uint32_t flags = spin_lock_blocking(q->core.spin_lock);
+    if (queue_get_level_unsafe(q) != 0) {
+        *(uint32_t*)data = ((uint32_t*)q->data)[q->rptr];
+        q->rptr = _queue_inc_index_u32(q, q->rptr);
+        success = true;
+    }
+    spin_unlock(q->core.spin_lock, flags);
+    if (success) __sev();
+    return success;
+}
+
+static inline bool queue_try_peek_u32(queue_t *q, void *data) {
+    bool success = false;
+    uint32_t flags = spin_lock_blocking(q->core.spin_lock);
+    if (queue_get_level_unsafe(q) != 0) {
+        *(uint32_t*)data = ((uint32_t*)q->data)[q->rptr];
+        success = true;
+    }
+    spin_unlock(q->core.spin_lock, flags);
+    return success;
+}
+
+static inline void queue_add_blocking_u32(queue_t *q, void *data) {
+    bool done;
+    do {
+        done = queue_try_add_u32(q, data);
+        if (done) break;
+        __wfe();
+    } while (true);
+}
+
+static inline void queue_remove_blocking_u32(queue_t *q, void *data) {
+    bool done;
+    do {
+        done = queue_try_remove_u32(q, data);
+        if (done) break;
+        __wfe();
+    } while (true);
+}
+
+static inline void queue_peek_blocking_u32(queue_t *q, void *data) {
+    bool done;
+    do {
+        done = queue_try_peek_u32(q, data);
+        if (done) break;
+        __wfe();
+    } while (true);
+}
+
+#endif