text instancing

cudoku.c

@@ -29,7 +29,7 @@ static const int quad_indices[] = {
 
 static const Color mistake_color = {1.f, 0.f, 0.f, 0.5f};
 
-static const wide_str *win_text = U"You won!";
+static const char *win_text = "You won!";
 
 static const char *help_texts[HELP_TEXT_SIZE] = {
   "F1 - Toggle help",
@@ -58,15 +58,15 @@ void draw_win() {
   /* set_width_constraint(&constraints, window.width, UI_CONSTRAINT_FIXED); */
   /* set_height_constraint(&constraints, window.height, UI_CONSTRAINT_FIXED); */
 
-  Alignment align = ALIGN_CENTER;
-  int font_size = 72.0f;
+  Alignment align = ALIGN_TOP_LEFT;
+  int font_size = 84.0f;
 
-  /* Sizef text_size = calculate_text_size(win_text, font_size); */
+  Sizef text_size = calculate_text_size(win_text, font_size);
 
   set_x_constraint(&constraints, 0, UI_CONSTRAINT_FIXED);
   set_y_constraint(&constraints, 0, UI_CONSTRAINT_FIXED);
-  /* set_width_constraint(&constraints, text_size.width, UI_CONSTRAINT_FIXED); */
-  /* set_height_constraint(&constraints, text_size.height, UI_CONSTRAINT_FIXED); */
+  set_width_constraint(&constraints, text_size.width, UI_CONSTRAINT_FIXED);
+  set_height_constraint(&constraints, text_size.height, UI_CONSTRAINT_FIXED);
   /* set_rotation_constraint(&constraints, 0); */
 
   /* UIConstraints circle_const; */
@@ -76,7 +76,7 @@ void draw_win() {
   /* set_height_constraint(&circle_const, 90, UI_CONSTRAINT_FIXED); */
 
 /*   draw_circle(circle_const, NULL, ALIGN_LEFT_CENTER); */
-  /* draw_quad(constraints, &bg_color, 0.0, align); */
+  draw_quad(constraints, &bg_color, 0.0, align);
 /*   set_x_constraint(&constraints, -50, UI_CONSTRAINT_FIXED); */
 /*   draw_quad(constraints, &bg_color, 25.0, ALIGN_TOP_RIGHT); */
 /*   set_x_constraint(&constraints, 0, UI_CONSTRAINT_FIXED); */
@@ -87,7 +87,17 @@ void draw_win() {
 
   set_width_constraint(&constraints, 1, UI_CONSTRAINT_FIXED);
   set_height_constraint(&constraints, 1, UI_CONSTRAINT_FIXED);
+  /* set_rotation_constraint(&constraints, 0); */
   draw_text(win_text, font_size, constraints, &text_color, align);
+
+  /* for (int i = 0; i < 100; i++) { */
+  /*   int x = rand() % window.width; */
+  /*   int y = rand() % window.height; */
+  /*   set_x_constraint(&constraints, x, UI_CONSTRAINT_FIXED); */
+  /*   set_y_constraint(&constraints, y, UI_CONSTRAINT_FIXED); */
+
+  /*   draw_text(win_text, font_size, constraints, &text_color, align); */
+  /* } */
 }
 
 void set_scale_factor(int width, int height, float *x, float *y) {

helper.c

@@ -70,14 +70,6 @@ Matrix4x4 orthographic_projection_2d(float left, float right, float bottom, floa
     return result;
 }
 
-uint wide_str_len(const wide_str *text) {
-  uint len = 0;
-  for (uint i = 0; text[i] != '\0'; i++) {
-    len++;
-  }
-  return len;
-}
-
 int max(int a, int b) {
     return a > b ? a : b;
 }

helper.h

@@ -4,15 +4,12 @@
 
 #include "z_math.h"
 
-typedef unsigned int wide_str;
-
 Matrix4x4 identity();
 void mult_4x4(Matrix4x4 *multiplicant, Matrix4x4 *multiplier);
 void apply_scale(Matrix4x4 *matrix, Sizef scale);
 void apply_rotation(Matrix4x4 *model, float angle);
 void apply_translation(Matrix4x4 *matrix, Vec2f pos);
 Matrix4x4 orthographic_projection_2d(float left, float right, float bottom, float top);
-uint wide_str_len(const wide_str *text);
 int max(int a, int b);
 float maxf(float a, float b);
 float clampf(float value, float min, float max);

shader.c

@@ -87,6 +87,11 @@ void set_float(Shader shader, const char *name, float val) {
   glUniform1f(loc, val);
 }
 
+void set_vec2f(Shader shader, const char *name, float val1, float val2) {
+  int loc = glGetUniformLocation(shader.program, name);
+  glUniform2f(loc, val1, val2);
+}
+
 void set_vec3f(Shader shader, const char *name, float val1, float val2, float val3) {
   int loc = glGetUniformLocation(shader.program, name);
   glUniform3f(loc, val1, val2, val3);

shader.h

@@ -9,5 +9,6 @@ Shader create_shader(const char *vertex_path, const char *fragment_path);
 void use_shader(Shader shader);
 void set_mat4f(Shader shader, const char *name, float *mat);
 void set_float(Shader shader, const char *name, float val);
+void set_vec2f(Shader shader, const char *name, float val1, float val2);
 void set_vec3f(Shader shader, const char *name, float val1, float val2, float val3);
 void set_vec4f(Shader shader, const char *name, float val1, float val2, float val3, float val4);

shaders/font.vert

@@ -0,0 +1,18 @@
+#version 330 core
+layout (location = 0) in vec2 vertex;
+
+// per instance
+layout (location = 1) in vec4 offset;
+layout (location = 2) in int tex_index;
+
+out vec2 v_TexCoords;
+uniform mat4 projection;
+uniform mat4 model;
+// buffer for all the 96 ascii characters' texcoords
+uniform vec2 texcoords[96 * 4];
+
+void main() {
+  vec2 pos = vec2((vertex.x) * offset.z, (vertex.y) * offset.w);
+  gl_Position = projection * model * vec4(pos + offset.xy, 0.0, 1.0);
+  v_TexCoords = texcoords[tex_index * 4 + gl_VertexID];
+}

todo.txt

@@ -4,8 +4,12 @@
 and replace them with the screen's currently active resolution
 
 text rendering improvements
-  - [x] unicode support (allows for non-ascii characters)
+  - [ ] unicode support (allows for non-ascii characters)
+    - unicode support removed for the foreseeable future
   - [x] atlas texture for text glyphs (this means we only load in one texture for all text)
+    - [x] better estimation of atlas width and height (currently the generated atlas texture
+          has a lot of empty space (over 50% of the texture's height))
+       - ended up just looping over the ascii glyphs and finding the max width and height
 
   - SDF generation allowing for better text scaling (slower but visually better)
     - SDF is too slow if we want to support unicode. it's acceptably fast for ascii only though.
@@ -17,4 +21,24 @@ text rendering improvements
 
       Will probably end up using MSDF (if I can get it to work with C lol)
 
-  - instanced rendering for text (this means we only have to do one draw call for all text)
+  - [x] instanced rendering for text (this means we only have to do one draw call for all text)
+    running into problems. we got the positions to instance correctly but the texture coordinates
+    are more difficult to work with because we can't just send a texcoord per vertex per instance
+    - Ended up passing hardcoded texcoords of all glyphs to the shader using a uniform array which allows us to render infinite
+      instances of text in one draw call without exceeding any buffer limits. Of course this does mean that if we were to support
+      unicode we'd have to make the uniform array bigger and its size dynamic. So unicode is dropped for now
+
+
+text tests / iGPU / Non-instaned rendering
+ - 10 texts / 72 font size
+   0.8-0.9ms per frame (1200-1300fps)
+
+ - 100 texts / 48 font size
+   1.2-1.3ms per frame (770-830fps)
+
+ - 1000 texts / 28 font size
+   9-9.5ms per frame (105-110fps)
+
+ - 10000 texts / 9 font size
+   80-85ms per frame (12-13fps)
+

ui.c

@@ -19,15 +19,20 @@
 Display *display;
 Window window;
 
+typedef struct InstanceData {
+  Vec4f position;
+  int tex_coords_index;
+} InstanceData;
+
 Context zephr_context;
 Shader font_shader;
 Shader ui_shader;
 unsigned int font_vao;
 unsigned int ui_vao;
-unsigned int font_vbo;
+unsigned int font_instance_vbo;
 unsigned int ui_vbo;
 
-// TODO: instanced rendering of ui elements and text??
+// TODO: instanced rendering of ui elements
 
 int init_fonts(const char *font_path) {
   FT_Library ft;
@@ -55,22 +60,38 @@ int init_fonts(const char *font_path) {
 
   glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
 
-  uint max_dim = (1 + (face->size->metrics.height >> 6)) * ceilf(sqrtf(face->num_glyphs));
-  uint tex_width = 1;
-  while (tex_width < max_dim) {
-    tex_width <<= 1;
-  }
-  uint tex_height = tex_width;
+  /* uint max_dim = (1 + (face->size->metrics.height >> 6)) * ceilf(sqrtf(face->num_glyphs)); */
+  /* uint tex_width = 1; */
+  /* while (tex_width < max_dim) { */
+  /*   tex_width <<= 1; */
+  /* } */
+  /* uint tex_height = tex_width; */
 
-  char *pixels = (char *)calloc(tex_width * tex_height, 1);
   int pen_x = 0, pen_y = 0;
 
   FT_UInt glyph_idx;
   FT_ULong c = FT_Get_First_Char(face, &glyph_idx);
 
-  while (glyph_idx) {
-    if (FT_Load_Char(face, c, FT_LOAD_RENDER)) {
-      printf("[ERROR]: failed to load glyph for char '0x%lx'\n", c);
+  uint tex_width = 0, tex_height = 0;
+  for (uint i = 32; i < 128; i++) {
+    if (FT_Load_Char(face, i, FT_LOAD_RENDER)) {
+      printf("[ERROR]: failed to load glyph for char '0x%x'\n", i);
+    }
+
+    /* FT_Render_Glyph(face->glyph, FT_RENDER_MODE_SDF); */
+
+    tex_width += face->glyph->bitmap.width + 1;
+    tex_height = max(tex_height, face->glyph->bitmap.rows);
+  }
+
+  printf("[INFO] Texture size: %d,%d\n", tex_width, tex_height);
+
+  char *pixels = (char *)calloc(tex_width * tex_height, 1);
+
+  for (uint i = 32; i < 128; i++) {
+  /* while (glyph_idx) { */
+    if (FT_Load_Char(face, i, FT_LOAD_RENDER)) {
+      printf("[ERROR]: failed to load glyph for char '0x%x'\n", i);
     }
 
     /* FT_Render_Glyph(face->glyph, FT_RENDER_MODE_SDF); */
@@ -110,11 +131,11 @@ int init_fonts(const char *font_path) {
     character.size = (Size){ .width = face->glyph->bitmap.width, .height = face->glyph->bitmap.rows };
     character.bearing = (Size){ .width = face->glyph->bitmap_left, .height = face->glyph->bitmap_top };
 
-    hmput(zephr_context.font.characters, c, character);
+    zephr_context.font.characters[i] = character;
 
     pen_x += bmp->width + 1;
 
-    c = FT_Get_Next_Char(face, c, &glyph_idx);
+    /* c = FT_Get_Next_Char(face, c, &glyph_idx); */
   }
 
   unsigned int texture;
@@ -140,6 +161,9 @@ int init_fonts(const char *font_path) {
 }
 
 int init_ui(const char* font_path, Size window_size) {
+  uint font_vbo;
+  uint font_ebo;
+
   zephr_context.window_size = window_size;
   zephr_context.projection = orthographic_projection_2d(0.f, window_size.width, window_size.height, 0.f);
 
@@ -152,19 +176,58 @@ int init_ui(const char* font_path, Size window_size) {
     return 1;
   }
 
-  font_shader = create_shader("shaders/ui.vert", "shaders/font.frag");
+  font_shader = create_shader("shaders/font.vert", "shaders/font.frag");
   ui_shader = create_shader("shaders/ui.vert", "shaders/ui.frag");
 
   glGenVertexArrays(1, &font_vao);
   glGenVertexArrays(1, &ui_vao);
   glGenBuffers(1, &font_vbo);
+  glGenBuffers(1, &font_instance_vbo);
+  glGenBuffers(1, &font_ebo);
   glGenBuffers(1, &ui_vbo);
 
+  float quad_vertices[4][2] = {
+    {0.0, 1.0}, // top left
+    {1.0, 1.0}, // top right
+    {1.0, 0.0}, // bottom right
+    {0.0, 0.0}, // bottom left
+  };
+
+  int quad_indices[6] = {
+    0, 1, 2,
+    2, 3, 0
+  };
+
   glBindVertexArray(font_vao);
+
+  // font quad vbo
   glBindBuffer(GL_ARRAY_BUFFER, font_vbo);
-  glBufferData(GL_ARRAY_BUFFER, sizeof(float) * 6 * 4, NULL, GL_DYNAMIC_DRAW);
+  glBufferData(GL_ARRAY_BUFFER, sizeof(float) * 4 * 2, quad_vertices, GL_STATIC_DRAW);
   glEnableVertexAttribArray(0);
-  glVertexAttribPointer(0, 4, GL_FLOAT, GL_FALSE, 4 * sizeof(float), (void *)0);
+  glVertexAttribPointer(0, 2, GL_FLOAT, GL_FALSE, 2 * sizeof(float), (void *)0);
+
+  // font instance vbo
+  glBindBuffer(GL_ARRAY_BUFFER, font_instance_vbo);
+  glEnableVertexAttribArray(1);
+  glVertexAttribPointer(1, 4, GL_FLOAT, GL_FALSE, sizeof(InstanceData), (void *)0);
+  glVertexAttribDivisor(1, 1);
+  glEnableVertexAttribArray(2);
+  glVertexAttribIPointer(2, 1, GL_INT, sizeof(InstanceData), (void *)sizeof(Vec4f));
+  glVertexAttribDivisor(2, 1);
+
+  // font ebo
+  glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, font_ebo);
+  glBufferData(GL_ELEMENT_ARRAY_BUFFER, sizeof(quad_indices), quad_indices, GL_STATIC_DRAW);
+
+  use_shader(font_shader);
+
+  for (int i = 32; i < 128; i++) {
+    for (int j = 0; j < 4; j++) {
+      char text[32];
+      snprintf(text, 32, "texcoords[%d]", (i - 32) * 4 + j);
+      set_vec2f(font_shader, text, zephr_context.font.characters[i].tex_coords[j].x, zephr_context.font.characters[i].tex_coords[j].y);
+    }
+  }
 
   glBindVertexArray(ui_vao);
   glBindBuffer(GL_ARRAY_BUFFER, ui_vbo);
@@ -256,9 +319,16 @@ bool zephr_should_quit() {
   /* update_ui(); */
   audio_update();
 
+
   return zephr_context.should_quit;
 }
 
+// example of how to batch all text drawing in one drawcall
+// this should be called at the end of the frame
+/* void batch_text_draw(&zephr_context.texts) {
+ * do something with zephr_context.texts
+}*/
+
 void zephr_swap_buffers() {
   glXSwapBuffers(display, window);
 }
@@ -267,7 +337,7 @@ Size zephr_get_window_size() {
   return zephr_context.window_size;
 }
 
-Sizef calculate_text_size(const wide_str *text, int font_size) {
+Sizef calculate_text_size(const char *text, int font_size) {
   float scale = (float)font_size / FONT_PIXEL_SIZE;
   Sizef size = { .width = 0, .height = 0 };
   int w = 0;
@@ -275,29 +345,27 @@ Sizef calculate_text_size(const wide_str *text, int font_size) {
   int max_bearing_h = 0;
 
   // NOTE: I don't like looping through the characters twice, but it's fine for now
-  for (uint i = 0; (i < text[i]) != '\0'; i++) {
-    Character ch = hmget(zephr_context.font.characters, (uint)text[i]);
-    /* Character ch = characters[(uint)text[i]]; */
+  for (int i = 0; text[i] != '\0'; i++) {
+    Character ch = zephr_context.font.characters[(uint)text[i]];
     max_bearing_h = max(max_bearing_h, ch.bearing.height);
   }
 
-  for (uint i = 0; (i < text[i]) != '\0'; i++) {
-    Character ch = hmget(zephr_context.font.characters, (uint)text[i]);
-    /* Character ch = characters[(uint)text[i]]; */
+  for (int i = 0; text[i] != '\0'; i++) {
+    Character ch = zephr_context.font.characters[(uint)text[i]];
     w += ch.advance / 64;
 
     // remove bearing of first character
-    if (i == 0 && wide_str_len(text) > 1) {
+    if (i == 0 && strlen(text) > 1) {
       w -= ch.bearing.width;
     }
 
     // remove the trailing width of the last character
-    if (i == wide_str_len(text) - 1) {
+    if (i == strlen(text) - 1) {
       w -= ((ch.advance / 64) - (ch.bearing.width + ch.size.width));
     }
 
     // if we only have one character in the text, then remove the bearing width
-    if (wide_str_len(text) == 1) {
+    if (strlen(text) == 1) {
       w -= (ch.bearing.width);
     }
 
@@ -525,7 +593,7 @@ void draw_triangle(UIConstraints constraints, Color *color, Alignment align) {
   glBindVertexArray(0);
 }
 
-void draw_text(const wide_str* text, int font_size, UIConstraints constraints, Color *color, Alignment alignment) {
+void draw_text(const char* text, int font_size, UIConstraints constraints, Color *color, Alignment alignment) {
   use_shader(font_shader);
   if (color) {
     set_vec4f(font_shader, "textColor", color->r / 255.f, color->g / 255.f, color->b / 255.f, color->a / 255.f);
@@ -556,52 +624,54 @@ void draw_text(const wide_str* text, int font_size, UIConstraints constraints, C
   set_mat4f(font_shader, "model", (float *)model.m);
 
   int max_bearing_h = 0;
-  for (uint i = 0; (i < text[i]) != '\0'; i++) {
-    Character ch = hmget(zephr_context.font.characters, text[i]);
-    /* Character ch = characters[(int)text[i]]; */
+  for (int i = 0; (i < text[i]) != '\0'; i++) {
+    Character ch = zephr_context.font.characters[(int)text[i]];
     max_bearing_h = max(max_bearing_h, ch.bearing.height);
   }
 
-  float first_char_bearing_w = hmget(zephr_context.font.characters, (int)text[0]).bearing.width;
+  float first_char_bearing_w = zephr_context.font.characters[(int)text[0]].bearing.width;
 
   glActiveTexture(GL_TEXTURE0);
   glBindTexture(GL_TEXTURE_2D, zephr_context.font.atlas_texture_id);
   glBindVertexArray(font_vao);
 
+  InstanceData* instance_data = malloc(sizeof(InstanceData) * strlen(text));
+
   // we use the original text and character sizes in the loop and then we just
   // scale up or down the model matrix to get the desired font size.
   // this way everything works out fine and we get to transform the text using the
   // model matrix
   int c = 0;
   int x = 0;
+  /* int line = 0; */
   while (text[c] != '\0') {
-    Character ch = hmget(zephr_context.font.characters, text[c]);
-    /* Character ch = characters[(int)text[c]]; */
+    Character ch = zephr_context.font.characters[(int)text[c]];
     // subtract the bearing width of the first character to remove the extra space
     // at the start of the text and move every char to the left by that width
     float xpos = (x + (ch.bearing.width - first_char_bearing_w));
-    float ypos = (text_size.height - ch.bearing.height - (text_size.height - max_bearing_h));
+    float ypos = /*(line * text_size.height) + */(text_size.height - ch.bearing.height - (text_size.height - max_bearing_h));
 
-    float vertices[6][4] = {
-      {xpos,                 ypos + ch.size.height, ch.tex_coords[0].x, ch.tex_coords[0].y},
-      {xpos,                 ypos,                  ch.tex_coords[3].x, ch.tex_coords[3].y},
-      {xpos + ch.size.width, ypos,                  ch.tex_coords[2].x, ch.tex_coords[2].y},
+    /* if (text[c] == '\n') { */
+    /*   line++; */
+    /*   x = 0; */
+    /*   c++; */
+    /*   continue; */
+    /* } */
 
-      {xpos,                 ypos + ch.size.height, ch.tex_coords[0].x, ch.tex_coords[0].y},
-      {xpos + ch.size.width, ypos,                  ch.tex_coords[2].x, ch.tex_coords[2].y},
-      {xpos + ch.size.width, ypos + ch.size.height, ch.tex_coords[1].x, ch.tex_coords[1].y},
-    };
-
-    glBindBuffer(GL_ARRAY_BUFFER, font_vbo);
-    glBufferSubData(GL_ARRAY_BUFFER, 0, sizeof(vertices), vertices);
-    glBindBuffer(GL_ARRAY_BUFFER, 0);
-
-    glDrawArrays(GL_TRIANGLES, 0, 6);
+    instance_data[c].position = (Vec4f){xpos, ypos, ch.size.width, ch.size.height};
+    instance_data[c].tex_coords_index = (int)text[c] - 32;
 
     x += (ch.advance >> 6); 
     c++;
   }
 
+  glBindBuffer(GL_ARRAY_BUFFER, font_instance_vbo);
+  glBufferData(GL_ARRAY_BUFFER, sizeof(InstanceData) * strlen(text), instance_data, GL_DYNAMIC_DRAW);
+
+  glDrawElementsInstanced(GL_TRIANGLES, 6, GL_UNSIGNED_INT, NULL, strlen(text));
+
+  free(instance_data);
+
   glBindVertexArray(0);
   glBindTexture(GL_TEXTURE_2D, 0);
 }

ui.h

@@ -12,13 +12,8 @@ typedef struct Character {
   Vec2f tex_coords[4]; // the 4 corners of the character quad
 } Character;
 
-typedef struct Characters {
-  uint key;
-  Character value;
-} Characters;
-
 typedef struct ZephFont {
-  Characters *characters;
+  Character characters[128];
   unsigned int atlas_texture_id;
 } ZephFont;
 
@@ -71,5 +66,5 @@ void set_rotation_constraint(UIConstraints *constraints, float angle_d);
 void draw_quad(UIConstraints constraints, Color *color, float border_radius, Alignment align);
 void draw_circle(UIConstraints constraints, Color *color, Alignment align);
 void draw_triangle(UIConstraints constraints, Color *color, Alignment align);
-void draw_text(const uint* text, int font_size, UIConstraints constraints, Color *color, Alignment align);
-Sizef calculate_text_size(const uint *text, int font_size);
+void draw_text(const char* text, int font_size, UIConstraints constraints, Color *color, Alignment align);
+Sizef calculate_text_size(const char *text, int font_size);

z_math.h

@@ -24,6 +24,14 @@ typedef struct Vec2f {
   float y;
 } Vec2f;
 
+typedef struct Vec4f {
+  float x;
+  float y;
+  float z;
+  float w;
+} Vec4f;
+
+
 typedef struct Color {
   float r;
   float g;