diff options
author | Vito Caputo <vcaputo@pengaru.com> | 2018-12-31 17:14:48 -0800 |
---|---|---|
committer | Vito Caputo <vcaputo@pengaru.com> | 2018-12-31 17:14:48 -0800 |
commit | fda032d3e3445f449cd82053505a399a404fa6d3 (patch) | |
tree | a2625a954b5f6b4069386662b1069ab62a3001b5 /src/modules/submit | |
parent | 6a643daad79440988845f9242acf50055fa17c52 (diff) |
modules/submit: add bilinearly-interpolated variant
This substantially reworks the cell sampling in submit.
As a result, it's now threaded in the rendering phase which now
resembles a texture mapper sans transformations.
This produces a full-screen rendering rather than a potentially
smaller one when the resolution wasn't cleanly divisable by the grid
size.
A new module, named submit-softly has also been added to expose the
bilinearly interpolated variant. The transition between cells is also
employing a smoothstep so it's not actually linear.
The original non-interpolated version is retained as well, at the same
submit module name.
Some minor cleanups happened as well, nothing worth mentioning, except
perhaps that the cells are now a uint8_t which is fine unless someone
tries to redefine NUM_PLAYERS > 255.
Diffstat (limited to 'src/modules/submit')
-rw-r--r-- | src/modules/submit/submit.c | 224 |
1 files changed, 195 insertions, 29 deletions
diff --git a/src/modules/submit/submit.c b/src/modules/submit/submit.c index 18886d8..7c4a9b6 100644 --- a/src/modules/submit/submit.c +++ b/src/modules/submit/submit.c @@ -30,47 +30,180 @@ #define GRID_SIZE 60 #define TICKS_PER_FRAME 8000 -static uint32_t colors[NUM_PLAYERS + 1] = { - 0x00000000, /* uninitialized cell starts black, becomes winner colors */ - 0xffff5000, /* orange */ - 0xff1020ff, /* blue */ - 0xffe00000, /* red */ - 0xff2ad726, /* green */ - 0xff00e0d0, /* cyan */ - 0xffd000ff, /* purple */ - 0xffe7ef00, /* yellow */ - 0x00000000, /* black */ +typedef struct color_t { + float r, g, b; +} color_t; + +static color_t colors[NUM_PLAYERS + 1] = { + {}, /* uninitialized cell starts black, becomes winner colors */ + {1.f, .317f, 0.f }, /* orange */ + {.627f, .125f, 1.f }, /* blue */ + {.878f, 0.f, 0.f }, /* red */ + {.165f, .843f, .149f }, /* green */ + {0.f, .878f, .815f }, /* cyan */ + {.878f, 0.f, 1.f }, /* purple */ + {.906f, .937f, 0.f }, /* yellow */ + {}, /* black */ }; + typedef struct submit_context_t { grid_t *grid; grid_player_t *players[NUM_PLAYERS]; uint32_t seq; uint32_t game_winner; - fb_fragment_t *fragment; - uint32_t cells[GRID_SIZE * GRID_SIZE]; + uint8_t cells[GRID_SIZE * GRID_SIZE]; } submit_context_t; -/* TODO: drawing is not optimized at all */ -static void draw_cell(fb_fragment_t *fragment, int x, int y, int w, int h, uint32_t color) +/* convert a color into a packed, 32-bit rgb pixel value (taken from libs/ray/ray_color.h) */ +static inline uint32_t color_to_uint32(color_t color) { + uint32_t pixel; + + if (color.r > 1.0f) color.r = 1.0f; + if (color.g > 1.0f) color.g = 1.0f; + if (color.b > 1.0f) color.b = 1.0f; + + pixel = (uint32_t)(color.r * 255.0f); + pixel <<= 8; + pixel |= (uint32_t)(color.g * 255.0f); + pixel <<= 8; + pixel |= (uint32_t)(color.b * 255.0f); + + return pixel; +} + + +static inline float clamp(float x, float lowerlimit, float upperlimit) { + if (x < lowerlimit) + x = lowerlimit; + + if (x > upperlimit) + x = upperlimit; + + return x; +} + + +/* taken from https://en.wikipedia.org/wiki/Smoothstep#Variations */ +static inline float smootherstep(float edge0, float edge1, float x) { + x = clamp((x - edge0) / (edge1 - edge0), 0.f, 1.f); + + return x * x * x * (x * (x * 6.f - 15.f) + 10.f); +} + + +/* linearly interpolate colors */ +static color_t color_lerp(color_t *a, color_t *b, float t) +{ + color_t res = { + .r = a->r * (1.f - t) + b->r * t, + .g = a->g * (1.f - t) + b->g * t, + .b = a->b * (1.f - t) + b->b * t, + }; + + return res; +} + + +/* bilinearly interpolate colors from 4 cells */ +static inline uint32_t sample_grid_bilerp(submit_context_t *ctxt, float x, float y) +{ + int i, ix = x, iy = y; + float x_t, y_t; + uint8_t corners[2][2]; + color_t x1, x2; + + /* FIXME TODO: this short-circuit produces unsmoothed cells at the + * periphery, and could be simply avoided by putting the peripheral + * cells off-screen so all the samples are within the boundaries. + * Ignoring for now. + */ + if (ix < 1 || ix > GRID_SIZE - 2 || + iy < 1 || iy > GRID_SIZE - 2) + return color_to_uint32(colors[ctxt->cells[iy * GRID_SIZE + ix]]); + + i = iy * GRID_SIZE + ix; + + /* ix,iy forms the corner of a 2x2 kernel, determine which corner */ + if (x > ix + .5f) { + x_t = x - ((float)ix + .5f); + + if (y > iy + .5f) { + /* NW corner */ + y_t = y - ((float)iy + .5f); + + corners[0][0] = ctxt->cells[i]; + corners[0][1] = ctxt->cells[i + 1]; + corners[1][0] = ctxt->cells[i + GRID_SIZE]; + corners[1][1] = ctxt->cells[i + GRID_SIZE + 1]; + } else { + /* SW corner */ + y_t = 1.f - (((float)iy + .5f) - y); + + corners[1][0] = ctxt->cells[i]; + corners[1][1] = ctxt->cells[i + 1]; + corners[0][0] = ctxt->cells[i - GRID_SIZE]; + corners[0][1] = ctxt->cells[i - GRID_SIZE + 1]; + } + } else { + x_t = 1.f - (((float)ix + .5f) - x); + + if (y > iy + .5f) { + /* NE corner */ + y_t = y - ((float)iy + .5f); + + corners[0][1] = ctxt->cells[i]; + corners[0][0] = ctxt->cells[i - 1]; + corners[1][1] = ctxt->cells[i + GRID_SIZE]; + corners[1][0] = ctxt->cells[i + GRID_SIZE - 1]; + } else { + /* SE corner */ + y_t = 1.f - (((float)iy + .5f) - y); + + corners[1][1] = ctxt->cells[i]; + corners[1][0] = ctxt->cells[i - 1]; + corners[0][1] = ctxt->cells[i - GRID_SIZE]; + corners[0][0] = ctxt->cells[i - GRID_SIZE - 1]; + } + } + + /* short-circuit cases where interpolation obviously wouldn't do anything */ + if (corners[0][0] == corners[0][1] && + corners[0][1] == corners[1][1] && + corners[1][1] == corners[1][0]) + return color_to_uint32(colors[corners[0][0]]); + + x_t = smootherstep(0.f, 1.f, x_t); + y_t = smootherstep(0.f, 1.f, y_t); + + x1 = color_lerp(&colors[corners[0][0]], &colors[corners[0][1]], x_t); + x2 = color_lerp(&colors[corners[1][0]], &colors[corners[1][1]], x_t); + + return color_to_uint32(color_lerp(&x1, &x2, y_t)); +} + + +static inline uint32_t sample_grid(submit_context_t *ctxt, float x, float y) { - for (int yy = 0; yy < h; yy++) - for (int xx = 0; xx < w; xx++) - fb_fragment_put_pixel_checked(fragment, x + xx, y + yy, color); + return color_to_uint32(colors[ctxt->cells[(int)y * GRID_SIZE + (int)x]]); } -static void draw_grid(submit_context_t *ctxt, fb_fragment_t *fragment) +static void draw_grid(submit_context_t *ctxt, fb_fragment_t *fragment, uint32_t (*sampler)(submit_context_t *ctxt, float x, float y)) { - int w = fragment->width / GRID_SIZE; - int h = fragment->height / GRID_SIZE; - int xoff = (fragment->width - w * GRID_SIZE) / 2; - int yoff = (fragment->height - h * GRID_SIZE) / 2; + float xscale = (float)GRID_SIZE / (float)fragment->frame_width; + float yscale = (float)GRID_SIZE / (float)fragment->frame_height; - for (int y = 0; y < GRID_SIZE; y++) - for (int x = 0; x < GRID_SIZE; x++) - draw_cell(fragment, xoff + x * w, yoff + y * h, w, h, colors[ctxt->cells[y * GRID_SIZE + x]]); + for (int y = 0; y < fragment->height; y++) { + for (int x = 0; x < fragment->width; x++) { + uint32_t color; + + /* TODO: this could be optimized a bit! i.e. don't recompute the y for every x etc. */ + color = sampler(ctxt, ((float)(fragment->x + x)) * xscale, ((float)(fragment->y + y)) * yscale); + fb_fragment_put_pixel_unchecked(fragment, fragment->x + x, fragment->y + y, color); + } + } } @@ -139,11 +272,17 @@ static void submit_destroy_context(void *context) } -static void submit_render_fragment(void *context, fb_fragment_t *fragment) +static int submit_fragmenter(void *context, const fb_fragment_t *fragment, unsigned num, fb_fragment_t *res_fragment) +{ + return fb_fragment_tile_single(fragment, 32, num, res_fragment); +} + + +static void submit_prepare_frame(void *context, unsigned n_cpus, fb_fragment_t *fragment, rototiller_fragmenter_t *res_fragmenter) { submit_context_t *ctxt = context; - ctxt->fragment = fragment; + *res_fragmenter = submit_fragmenter; if (ctxt->game_winner) setup_grid(ctxt); @@ -157,17 +296,44 @@ static void submit_render_fragment(void *context, fb_fragment_t *fragment) for (int j = 0; j < TICKS_PER_FRAME; j++) grid_tick(ctxt->grid); +} + + +static void submit_render_fragment(void *context, fb_fragment_t *fragment) +{ + submit_context_t *ctxt = context; + + draw_grid(ctxt, fragment, sample_grid); +} + + +static void submit_softly_render_fragment(void *context, fb_fragment_t *fragment) +{ + submit_context_t *ctxt = context; - draw_grid(ctxt, fragment); + draw_grid(ctxt, fragment, sample_grid_bilerp); } rototiller_module_t submit_module = { .create_context = submit_create_context, .destroy_context = submit_destroy_context, + .prepare_frame = submit_prepare_frame, .render_fragment = submit_render_fragment, .name = "submit", - .description = "Cellular automata conquest game sim", + .description = "Cellular automata conquest game sim (threaded (poorly))", + .author = "Vito Caputo <vcaputo@pengaru.com>", + .license = "GPLv3", +}; + + +rototiller_module_t submit_softly_module = { + .create_context = submit_create_context, + .destroy_context = submit_destroy_context, + .prepare_frame = submit_prepare_frame, + .render_fragment = submit_softly_render_fragment, + .name = "submit-softly", + .description = "Cellular automata conquest game sim w/bilinear interpolation (threaded (poorly))", .author = "Vito Caputo <vcaputo@pengaru.com>", .license = "GPLv3", }; |