[feat] 低质量ttf render

This commit is contained in:
2026-06-06 09:00:56 +08:00
Unverified
parent 3d47667c2f
commit 500d302ea9
15 changed files with 914 additions and 17 deletions
+125
View File
@@ -0,0 +1,125 @@
#include <fonts/ttf.h>
#include "ttf_internal.h"
#include <graphics/draw.h>
#include <memory/heap.h>
#include <serial.h>
// Per-glyph scratch — kept static to avoid stack pressure for big Chinese glyphs
static ttf_outline_t s_outline;
static ttf_seg_t s_segs[4096];
static SUINT8 s_coverage[256 * 256];
// Render one glyph bitmap at screen (px, py) where py is the glyph TOP
// (already converted from baseline). (px, py) may be negative — caller
// must have clipped into the visible region.
static void blit_glyph(SSINT32 px, SSINT32 py, SUINT32 w, SUINT32 h,
const SUINT8* coverage, SUINT32 N,
EFI_GRAPHICS_OUTPUT_BLT_PIXEL color)
{
SUINT32 scale = 255 / N;
for (SUINT32 y = 0; y < h; y++) {
SSINT32 sy = py + (SSINT32)y;
if (sy < 0) continue;
for (SUINT32 x = 0; x < w; x++) {
SSINT32 sx = px + (SSINT32)x;
if (sx < 0) continue;
SUINT8 c = coverage[y * w + x];
if (c == 0) continue;
SUINT8 a = (SUINT8)(c * scale);
draw_pixel_alpha(sx, sy, color, a);
}
}
}
// Render a single codepoint at (x, y) = baseline. Returns advance (26.6).
static f26_6 render_codepoint(ttf_face_t* face, SSINT32 cp,
SSINT32 x, SSINT32 y, SUINT32 pixel_size,
EFI_GRAPHICS_OUTPUT_BLT_PIXEL color)
{
SUINT16 gid = ttf_cmap_lookup(face, cp);
if (gid == 0 && cp != 0) gid = 0; // notdef
if (!ttf_load_glyph(face, gid, pixel_size, &s_outline)) {
return 0; // composite / parse error
}
// Bitmap dims in pixels
SSINT32 bbox_w = s_outline.xmax - s_outline.xmin;
SSINT32 bbox_h = s_outline.ymax - s_outline.ymin;
if (bbox_w <= 0 || bbox_h <= 0) {
return s_outline.advance; // whitespace or zero-size
}
SUINT32 pw = (SUINT32)((bbox_w + 63) >> 6);
SUINT32 ph = (SUINT32)((bbox_h + 63) >> 6);
if (pw > 256 || ph > 256) {
return s_outline.advance; // too big for scratch
}
// Translate outline to bitmap-local: bx = fx - xmin, by = ymax - fy
ttf_outline_t local;
local.num_points = s_outline.num_points;
local.num_contours = s_outline.num_contours;
for (SUINT16 i = 0; i < s_outline.num_contours; i++) {
local.first[i] = s_outline.first[i];
local.last[i] = s_outline.last[i];
}
for (SUINT16 i = 0; i < s_outline.num_points; i++) {
local.x[i] = s_outline.x[i] - s_outline.xmin;
local.y[i] = s_outline.ymax - s_outline.y[i];
local.on_curve[i] = s_outline.on_curve[i];
}
local.xmin = 0;
local.ymin = 0;
local.xmax = bbox_w;
local.ymax = bbox_h;
SUINT32 n_segs = 0;
ttf_outline_to_segments(&local, s_segs, &n_segs);
// Clear coverage
for (SUINT32 i = 0; i < pw * ph; i++) s_coverage[i] = 0;
const SUINT32 N = 5; // subsamples per pixel row
ttf_rasterize(s_segs, n_segs, 0, 0, pw, ph, s_coverage, N);
// Screen origin of bitmap
// bitmap (0, 0) is at font (xmin, ymax) which is at screen
// (x + xmin/64, y - ymax/64) since screen y is flipped
SSINT32 px_screen = x + (s_outline.xmin >> 6);
SSINT32 py_screen = y - (s_outline.ymax >> 6);
blit_glyph(px_screen, py_screen, pw, ph, s_coverage, N, color);
return s_outline.advance;
}
SSINT32 ttf_draw_text(ttf_face_t* face, const char* utf8,
SSINT32 x, SSINT32 y, SUINT32 pixel_size,
EFI_GRAPHICS_OUTPUT_BLT_PIXEL color)
{
if (!face || !utf8) return 0;
const char* p = utf8;
f26_6 pen = 0;
while (*p) {
SSINT32 cp = ttf_utf8_decode(&p);
if (cp < 0) continue;
f26_6 adv = render_codepoint(face, cp,
x + (pen >> 6), y, pixel_size, color);
pen += adv;
}
return pen;
}
SSINT32 ttf_text_width(ttf_face_t* face, const char* utf8, SUINT32 pixel_size) {
if (!face || !utf8) return 0;
const char* p = utf8;
f26_6 pen = 0;
while (*p) {
SSINT32 cp = ttf_utf8_decode(&p);
if (cp < 0) continue;
SUINT16 gid = ttf_cmap_lookup(face, cp);
if (!ttf_load_glyph(face, gid, pixel_size, &s_outline)) continue;
pen += s_outline.advance;
}
return pen;
}
+93
View File
@@ -0,0 +1,93 @@
#pragma once
#include <efi.h>
#include <common.h>
#include "ttf_math.h"
typedef struct ttf_face ttf_face_t;
typedef struct ttf_face {
const SUINT8* data;
UINTN size;
// head
SUINT32 units_per_em;
SSINT16 index_to_loc_format; // 0 = short, 1 = long
// hhea
SSINT16 hhea_ascender;
SSINT16 hhea_descender;
SSINT16 hhea_line_gap;
SUINT16 num_long_hor_metrics;
// maxp
SUINT16 num_glyphs;
SUINT16 max_points;
SUINT16 max_contours;
// os/2
SSINT16 os2_ascender;
SSINT16 os2_descender;
SSINT16 os2_line_gap;
// Table directory (sorted by tag at open time)
struct {
char tag[4];
SUINT32 offset;
SUINT32 length;
} tables[32];
SUINT16 num_tables;
// Cached pointers into data[]
const SUINT8* loca; // size depends on num_glyphs and format
const SUINT8* glyf;
SUINT32 glyf_len;
const SUINT8* hmtx;
SUINT32 hmtx_len;
const SUINT8* cmap;
} ttf_face_t;
// Decomposed glyph: contours in pixel space (26.6 fp) ready to scan.
typedef struct ttf_outline {
f26_6 x[1024];
f26_6 y[1024];
SUINT8 on_curve[1024];
SUINT16 first[64];
SUINT16 last[64];
SUINT16 num_contours;
SUINT16 num_points;
// Bounding box in pixel space (26.6 fp).
SSINT32 xmin, ymin, xmax, ymax;
// Pen advance in pixel space (26.6 fp).
f26_6 advance;
} ttf_outline_t;
typedef struct ttf_seg {
f26_6 x0, y0, x1, y1; // line / quad end points
f26_6 cx, cy; // quadratic control (set to x0/y0 if is_line)
SUINT8 is_line; // 1 = line, 0 = quad
} ttf_seg_t;
// Parse the glyf entry for glyph_id. Fills outline already
// scaled to pixel_size_px. Returns false on composite / parse error.
bool ttf_load_glyph(ttf_face_t* face, SUINT16 glyph_id,
SUINT32 pixel_size_px, ttf_outline_t* out);
// UTF-8 decoder. *p advances past the codepoint. Returns -1 on error.
SSINT32 ttf_utf8_decode(const char** p);
// cmap lookup — returns glyph_id (0 = notdef).
SUINT16 ttf_cmap_lookup(ttf_face_t* face, SSINT32 cp);
// Decompose outline into a flat array of line/quadratic segments.
// out must have room for 2*num_points + num_contours entries.
void ttf_outline_to_segments(const ttf_outline_t* outline,
ttf_seg_t* segs, SUINT32* num_segs);
// Scanline rasterize segments into per-pixel coverage.
// coverage[w*h] gets values in [0, N] where N = supersample count.
void ttf_rasterize(const ttf_seg_t* segs, SUINT32 num_segs,
SSINT32 x0, SSINT32 y0, SUINT32 w, SUINT32 h,
SUINT8* coverage, SUINT32 N);
+30
View File
@@ -0,0 +1,30 @@
#pragma once
#include <common.h>
// 26.6 fixed point. 1.0 = 64, -32.0 .. 31.999 range in SSINT32.
typedef SSINT32 f26_6;
#define F26_ONE ((f26_6)64)
#define F26_HALF ((f26_6)32)
#define F26_FROM_INT(x) ((f26_6)((x) * 64))
#define F26_FLOOR(x) ((x) >> 6)
#define F26_ROUND(x) (((x) + 32) >> 6)
#define F26_FRAC(x) ((x) & 63)
static inline f26_6 f26_mul(f26_6 a, f26_6 b) {
return (f26_6)(((SSINT64)a * b) >> 6);
}
static inline f26_6 f26_div(f26_6 a, f26_6 b) {
if (b == 0) return 0;
return (f26_6)(((SSINT64)a << 6) / b);
}
// Linear interpolation. t in [0, 64].
static inline f26_6 f26_lerp(f26_6 a, f26_6 b, f26_6 t) {
return a + f26_mul(b - a, t);
}
// Floor-of-division helper for bezier root solving.
static inline SSINT32 isign(SSINT32 x) { return (x > 0) - (x < 0); }
+331
View File
@@ -0,0 +1,331 @@
#include "ttf_internal.h"
#include <string_utils.h>
#include <memory/heap.h>
#include <serial.h>
// ---- Big-endian readers (TTF is big-endian) ----
static inline SUINT16 rd16(const SUINT8* p) {
return ((SUINT16)p[0] << 8) | p[1];
}
static inline SSINT16 rd16s(const SUINT8* p) {
return (SSINT16)(((SUINT16)p[0] << 8) | p[1]);
}
static inline SUINT32 rd32(const SUINT8* p) {
return ((SUINT32)p[0] << 24) | ((SUINT32)p[1] << 16) |
((SUINT32)p[2] << 8) | (SUINT32)p[3];
}
static const SUINT8* find_table(ttf_face_t* face, const char tag[4]) {
for (SUINT16 i = 0; i < face->num_tables; i++) {
if (face->tables[i].tag[0] == tag[0] &&
face->tables[i].tag[1] == tag[1] &&
face->tables[i].tag[2] == tag[2] &&
face->tables[i].tag[3] == tag[3]) {
return face->data + face->tables[i].offset;
}
}
return NULL;
}
// ---- UTF-8 ----
SSINT32 ttf_utf8_decode(const char** p) {
const SUINT8* s = (const SUINT8*)*p;
SUINT8 b0 = s[0];
if (b0 < 0x80) { (*p)++; return b0; }
if ((b0 & 0xE0) == 0xC0) { (*p) += 2; return ((b0 & 0x1F) << 6) | (s[1] & 0x3F); }
if ((b0 & 0xF0) == 0xE0) { (*p) += 3; return ((b0 & 0x0F) << 12) | ((s[1] & 0x3F) << 6) | (s[2] & 0x3F); }
if ((b0 & 0xF8) == 0xF0) { (*p) += 4; return ((b0 & 0x07) << 18) | ((s[1] & 0x3F) << 12) | ((s[2] & 0x3F) << 6) | (s[3] & 0x3F); }
(*p)++;
return -1;
}
// ---- cmap ----
static const SUINT8* find_cmap_subtable(ttf_face_t* face) {
const SUINT8* cmap = face->cmap;
SUINT16 num = rd16(cmap + 2);
const SUINT8* best = NULL;
SSINT32 best_score = -1;
for (SUINT16 i = 0; i < num; i++) {
const SUINT8* rec = cmap + 4 + i * 8;
SUINT16 platform = rd16(rec + 0);
SUINT16 encoding = rd16(rec + 2);
SUINT32 offset = rd32(rec + 4);
SSINT32 score = -1;
if (platform == 3 && encoding == 10) score = 30;
else if (platform == 3 && encoding == 1) score = 20;
else if (platform == 0 && encoding == 4) score = 10;
else if (platform == 0 && encoding == 3) score = 5;
if (score > best_score) { best_score = score; best = cmap + offset; }
}
return best;
}
static SUINT16 cmap4_lookup(const SUINT8* sub, SSINT32 cp) {
SUINT16 segCountX2 = rd16(sub + 6);
SUINT16 segCount = segCountX2 / 2;
const SUINT8* endCode = sub + 14;
const SUINT8* startCode = endCode + segCountX2 + 2;
const SUINT8* idDelta = startCode + segCountX2;
const SUINT8* idRangeOff = idDelta + segCountX2;
for (SUINT16 i = 0; i < segCount; i++) {
SUINT16 ec = rd16(endCode + i * 2);
SUINT16 sc = rd16(startCode + i * 2);
if (cp < sc) break;
if (cp <= ec) {
SUINT16 dro = rd16(idRangeOff + i * 2);
if (dro == 0) {
return (SUINT16)((SSINT16)cp + (SSINT16)rd16(idDelta + i * 2));
}
SUINT16 gidx = rd16(idRangeOff + i * 2 + dro + (cp - sc) * 2);
if (gidx == 0) return 0;
return (SUINT16)((SSINT16)gidx + (SSINT16)rd16(idDelta + i * 2));
}
}
return 0;
}
static SUINT16 cmap12_lookup(const SUINT8* sub, SSINT32 cp) {
// Format 12 header: format(2) reserved(2) length(4) language(4) nGroups(4)
SUINT32 nGroups = rd32(sub + 12);
const SUINT8* g = sub + 16;
for (SUINT32 i = 0; i < nGroups; i++) {
SUINT32 sc = rd32(g + 0);
SUINT32 ec = rd32(g + 4);
SUINT32 sG = rd32(g + 8);
if (cp < (SSINT32)sc) break;
if (cp <= (SSINT32)ec) return (SUINT16)(sG + (cp - sc));
g += 12;
}
return 0;
}
SUINT16 ttf_cmap_lookup(ttf_face_t* face, SSINT32 cp) {
const SUINT8* sub = find_cmap_subtable(face);
if (!sub) return 0;
SUINT16 fmt = rd16(sub);
if (fmt == 4) return cmap4_lookup(sub, cp);
if (fmt == 12) return cmap12_lookup(sub, cp);
return 0;
}
// ---- Glyf decode ----
bool ttf_load_glyph(ttf_face_t* face, SUINT16 glyph_id,
SUINT32 pixel_size_px, ttf_outline_t* out)
{
mem_set(out, 0, sizeof(*out));
if (glyph_id >= face->num_glyphs) return false;
// Loca
SUINT32 off0, off1;
if (face->index_to_loc_format == 0) {
off0 = ((SUINT32)rd16(face->loca + glyph_id * 2)) * 2;
off1 = ((SUINT32)rd16(face->loca + (glyph_id + 1) * 2)) * 2;
} else {
off0 = rd32(face->loca + glyph_id * 4);
off1 = rd32(face->loca + (glyph_id + 1) * 4);
}
if (off0 >= face->glyf_len) return false;
// Advance width (always readable from hmtx regardless of glyf presence)
{
SUINT16 aw;
if (glyph_id < face->num_long_hor_metrics) {
aw = rd16(face->hmtx + glyph_id * 4);
} else {
aw = rd16(face->hmtx + (face->num_long_hor_metrics - 1) * 4);
}
out->advance = (f26_6)(((SUINT64)aw * (SUINT64)pixel_size_px * 64) / face->units_per_em);
}
if (off0 == off1) {
// Empty glyph (e.g. space)
return true;
}
const SUINT8* g = face->glyf + off0;
SSINT16 numContours = rd16s(g + 0);
if (numContours < 0) {
// Composite — unsupported in v1. Keep advance, no outline.
return true;
}
if (numContours == 0) return true;
if (numContours > 64) return false;
SSINT16 gxMin = rd16s(g + 2);
SSINT16 gyMin = rd16s(g + 4);
SSINT16 gxMax = rd16s(g + 6);
SSINT16 gyMax = rd16s(g + 8);
const SUINT8* p = g + 10;
SUINT16 endPts[64];
for (SSINT16 i = 0; i < numContours; i++) { endPts[i] = rd16(p); p += 2; }
SUINT16 numPoints = (SUINT16)(endPts[numContours - 1] + 1);
if (numPoints > 1024) return false;
SUINT16 instrLen = rd16(p); p += 2;
p += instrLen;
// Decode flags (with REPEAT)
SUINT8 flags[1024];
SUINT16 pi = 0;
while (pi < numPoints) {
SUINT8 f = *p++;
flags[pi++] = f;
if (f & 0x08) {
SUINT8 rep = *p++;
for (SUINT16 k = 1; k < rep && pi < numPoints; k++)
flags[pi++] = f;
}
}
// Decode x-coords into x_raw[]
SSINT32 x_raw[1024];
{
SSINT32 x = 0;
for (SUINT16 i = 0; i < numPoints; i++) {
SUINT8 f = flags[i];
if (f & 0x02) {
SUINT8 b = *p++;
x += (f & 0x10) ? b : -(SSINT32)b;
} else if (!(f & 0x10)) {
x += (SSINT16)((SUINT16)p[0] << 8 | p[1]);
p += 2;
}
x_raw[i] = x;
}
}
// Decode y-coords into y_raw[]
SSINT32 y_raw[1024];
{
SSINT32 y = 0;
for (SUINT16 i = 0; i < numPoints; i++) {
SUINT8 f = flags[i];
if (f & 0x04) {
SUINT8 b = *p++;
y += (f & 0x20) ? b : -(SSINT32)b;
} else if (!(f & 0x20)) {
y += (SSINT16)((SUINT16)p[0] << 8 | p[1]);
p += 2;
}
y_raw[i] = y;
}
}
// Scale to pixel space (f26_6)
SUINT64 scale_num = (SUINT64)pixel_size_px * 64;
for (SUINT16 i = 0; i < numPoints; i++) {
out->x[i] = (f26_6)(((SSINT64)x_raw[i] * (SSINT64)scale_num) / face->units_per_em);
out->y[i] = (f26_6)(((SSINT64)y_raw[i] * (SSINT64)scale_num) / face->units_per_em);
out->on_curve[i] = (flags[i] & 0x01) ? 1 : 0;
}
out->num_contours = (SUINT16)numContours;
out->num_points = numPoints;
SUINT16 start = 0;
for (SSINT16 i = 0; i < numContours; i++) {
out->first[i] = start;
out->last[i] = endPts[i];
start = endPts[i] + 1;
}
out->xmin = (SSINT32)(((SSINT64)gxMin * (SSINT64)scale_num) / face->units_per_em);
out->ymin = (SSINT32)(((SSINT64)gyMin * (SSINT64)scale_num) / face->units_per_em);
out->xmax = (SSINT32)(((SSINT64)gxMax * (SSINT64)scale_num) / face->units_per_em);
out->ymax = (SSINT32)(((SSINT64)gyMax * (SSINT64)scale_num) / face->units_per_em);
return true;
}
// ---- Open / close ----
ttf_face_t* ttf_open(const void* data, UINTN size) {
if (!data || size < 12) return NULL;
const SUINT8* d = (const SUINT8*)data;
SUINT32 sfVersion = rd32(d);
if (sfVersion != 0x00010000 && sfVersion != 0x74727565) {
serial_write("ttf: bad sfVersion\n");
return NULL;
}
SUINT16 numTables = rd16(d + 4);
if (numTables == 0 || numTables > 32) return NULL;
ttf_face_t* face = (ttf_face_t*)kcalloc(1, sizeof(ttf_face_t));
if (!face) return NULL;
face->data = d;
face->size = size;
face->num_tables = numTables;
for (SUINT16 i = 0; i < numTables; i++) {
const SUINT8* r = d + 12 + i * 16;
face->tables[i].tag[0] = r[0];
face->tables[i].tag[1] = r[1];
face->tables[i].tag[2] = r[2];
face->tables[i].tag[3] = r[3];
face->tables[i].offset = rd32(r + 8);
face->tables[i].length = rd32(r + 12);
}
const SUINT8* head = find_table(face, "head");
if (!head) { kfree(face); return NULL; }
face->units_per_em = rd16(head + 18);
face->index_to_loc_format = rd16s(head + 50);
const SUINT8* hhea = find_table(face, "hhea");
if (!hhea) { kfree(face); return NULL; }
face->hhea_ascender = rd16s(hhea + 4);
face->hhea_descender = rd16s(hhea + 6);
face->hhea_line_gap = rd16s(hhea + 8);
face->num_long_hor_metrics = rd16(hhea + 34);
const SUINT8* maxp = find_table(face, "maxp");
if (!maxp) { kfree(face); return NULL; }
face->num_glyphs = rd16(maxp + 4);
face->max_points = rd16(maxp + 6);
face->max_contours = rd16(maxp + 8);
const SUINT8* os2 = find_table(face, "OS/2");
if (os2) {
face->os2_ascender = rd16s(os2 + 68);
face->os2_descender = rd16s(os2 + 70);
face->os2_line_gap = rd16s(os2 + 72);
} else {
face->os2_ascender = face->hhea_ascender;
face->os2_descender = face->hhea_descender;
face->os2_line_gap = face->hhea_line_gap;
}
const SUINT8* loca = find_table(face, "loca");
const SUINT8* glyf = find_table(face, "glyf");
const SUINT8* hmtx = find_table(face, "hmtx");
if (!loca || !glyf || !hmtx) { kfree(face); return NULL; }
face->loca = loca;
face->glyf = glyf;
face->hmtx = hmtx;
for (SUINT16 i = 0; i < numTables; i++) {
if (face->tables[i].tag[0]=='g'&&face->tables[i].tag[1]=='l'&&
face->tables[i].tag[2]=='y'&&face->tables[i].tag[3]=='f')
face->glyf_len = face->tables[i].length;
if (face->tables[i].tag[0]=='h'&&face->tables[i].tag[1]=='m'&&
face->tables[i].tag[2]=='t'&&face->tables[i].tag[3]=='x')
face->hmtx_len = face->tables[i].length;
}
face->cmap = find_table(face, "cmap");
if (!face->cmap) { kfree(face); return NULL; }
return face;
}
void ttf_close(ttf_face_t* face) {
if (face) kfree(face);
}
// ---- Metrics (scaled to pixel_size, returned as 26.6 fp) ----
SSINT32 ttf_ascender (ttf_face_t* face, SUINT32 px) {
return (SSINT32)(((SSINT64)face->os2_ascender * (SSINT64)px * 64) / face->units_per_em);
}
SSINT32 ttf_descender(ttf_face_t* face, SUINT32 px) {
return (SSINT32)(((SSINT64)face->os2_descender * (SSINT64)px * 64) / face->units_per_em);
}
SSINT32 ttf_line_gap (ttf_face_t* face, SUINT32 px) {
return (SSINT32)(((SSINT64)face->os2_line_gap * (SSINT64)px * 64) / face->units_per_em);
}
+214
View File
@@ -0,0 +1,214 @@
#include "ttf_internal.h"
#include <string_utils.h>
// ---- Outline -> segments ----
//
// TrueType contour walk: for each pair of consecutive points, emit either
// a line or a quadratic bezier. Two consecutive off-curve points trigger
// synthesis of an on-curve midpoint.
//
// All coordinates remain in 26.6 fp throughout.
void ttf_outline_to_segments(const ttf_outline_t* outline,
ttf_seg_t* segs, SUINT32* num_segs)
{
*num_segs = 0;
f26_6 syn_x[256];
f26_6 syn_y[256];
int n_syn = 0;
for (SUINT16 ci = 0; ci < outline->num_contours; ci++) {
SUINT16 first = outline->first[ci];
SUINT16 last = outline->last[ci];
int n_pts = last - first + 1;
if (n_pts < 2) continue;
f26_6 lx[1024];
f26_6 ly[1024];
SUINT8 on_c[1024];
for (int i = 0; i < n_pts; i++) {
lx[i] = outline->x[first + i];
ly[i] = outline->y[first + i];
on_c[i] = outline->on_curve[first + i];
}
// First on-curve point
int start = 0;
while (start < n_pts && !on_c[start]) start++;
if (start == n_pts) start = 0; // all off-curve (rare) — keep going
int anchor = start;
int pending = -1;
#define GET_X(i) ((i) < n_pts ? lx[(i)] : syn_x[(i) - n_pts])
#define GET_Y(i) ((i) < n_pts ? ly[(i)] : syn_y[(i) - n_pts])
#define PUSH_LINE(a, b) do { \
ttf_seg_t* __s = &segs[(*num_segs)++]; \
__s->x0 = GET_X(a); __s->y0 = GET_Y(a); \
__s->x1 = GET_X(b); __s->y1 = GET_Y(b); \
__s->cx = __s->x0; __s->cy = __s->y0; __s->is_line = 1; \
} while (0)
#define PUSH_QUAD(a, c, b) do { \
ttf_seg_t* __s = &segs[(*num_segs)++]; \
__s->x0 = GET_X(a); __s->y0 = GET_Y(a); \
__s->x1 = GET_X(b); __s->y1 = GET_Y(b); \
__s->cx = GET_X(c); __s->cy = GET_Y(c); __s->is_line = 0; \
} while (0)
for (int step = 0; step < n_pts; step++) {
int cur = (start + step) % n_pts;
if (step == 0) { anchor = cur; continue; }
if (on_c[cur]) {
if (pending < 0) {
PUSH_LINE(anchor, cur);
} else {
PUSH_QUAD(anchor, pending, cur);
pending = -1;
}
anchor = cur;
} else {
if (pending < 0) {
pending = cur;
} else {
int syn_idx = n_pts + n_syn;
syn_x[n_syn] = (GET_X(pending) + GET_X(cur)) >> 1;
syn_y[n_syn] = (GET_Y(pending) + GET_Y(cur)) >> 1;
n_syn++;
PUSH_QUAD(anchor, pending, syn_idx);
anchor = syn_idx;
pending = cur;
}
}
}
if (pending >= 0) PUSH_QUAD(anchor, pending, start);
#undef GET_X
#undef GET_Y
#undef PUSH_LINE
#undef PUSH_QUAD
}
}
// ---- Integer sqrt (Newton) ----
static SUINT32 isqrt_u64(SUINT64 n) {
if (n == 0) return 0;
SUINT32 x = (n > 0xFFFFFFFFu) ? 0xFFFFu : (SUINT32)n;
SUINT32 y = (x + 1) >> 1;
while (y < x) { x = y; y = (x + (SUINT32)(n / x)) >> 1; }
return x;
}
// ---- Scanline fill with subpixel supersampling ----
//
// For each output row, run N sub-scanlines at offsets (k+0.5)/N. For each
// sub-scanline y, collect all x-intersections, sort, fill alternating
// x-pairs. Sum over N subsamples yields per-pixel coverage in [0, N].
void ttf_rasterize(const ttf_seg_t* segs, SUINT32 num_segs,
SSINT32 x0, SSINT32 y0, SUINT32 w, SUINT32 h,
SUINT8* coverage, SUINT32 N)
{
// Clear coverage
for (SUINT32 i = 0; i < w * h; i++) coverage[i] = 0;
// Intersection x-buf (per scanline, max possible = num_segs)
f26_6 xs[2048];
if (num_segs > 2048) num_segs = 2048;
for (SUINT32 row = 0; row < h; row++) {
for (SUINT32 k = 0; k < N; k++) {
// Sub-scanline y, in 26.6, with y = 0 at glyph top
f26_6 sy = (f26_6)((row * 64) + ((k * 2 + 1) * 64) / (SSINT32)(N * 2));
// ^ y center of subpixel k: (k + 0.5) * 64 / N
// = ((2k+1) * 64) / (2N)
// For N=5: k=0 -> 6.4, k=1 -> 19.2, etc.
SUINT32 nxs = 0;
for (SUINT32 s = 0; s < num_segs; s++) {
const ttf_seg_t* g = &segs[s];
if (g->is_line) {
f26_6 y0s = g->y0;
f26_6 y1s = g->y1;
if (y0s == y1s) continue; // horizontal — skip
// t = (sy - y0s) / (y1s - y0s) in (0, 1)
if (((y0s < sy) && (y1s < sy)) ||
((y0s > sy) && (y1s > sy))) continue;
f26_6 t = f26_div(sy - y0s, y1s - y0s);
if (t <= 0 || t >= F26_ONE) continue;
f26_6 x = g->x0 + f26_mul(t, g->x1 - g->x0);
xs[nxs++] = x;
} else {
// Quadratic: y(t) = (1-t)^2 y0 + 2(1-t)t cy + t^2 y1
// a t^2 + b t + c = 0
// a = y1 - 2 cy + y0
// b = 2 (cy - y0)
// c = y0 - sy
SSINT64 a = (SSINT64)g->y1 - 2 * (SSINT64)g->cy + (SSINT64)g->y0;
SSINT64 b = 2 * ((SSINT64)g->cy - (SSINT64)g->y0);
SSINT64 c = (SSINT64)g->y0 - (SSINT64)sy;
SSINT32 t0_valid = 0, t1_valid = 0;
f26_6 t0 = 0, t1 = 0;
if (a == 0) {
// Linear
if (b == 0) continue;
f26_6 t = f26_div((f26_6)(-c), (f26_6)b);
if (t > 0 && t < F26_ONE) { t0 = t; t0_valid = 1; }
} else {
SSINT64 disc = b * b - 4 * a * c;
if (disc < 0) continue;
SUINT32 sq = isqrt_u64((SUINT64)disc);
// 26.6 roots: t = (-b ± sqrt(disc)) / (2a)
// All in 26.6 fp.
// t = (-b ± sq) / (2a); both numerator and denom in 26.6 units
// Use f26_div: t_26_6 = ((-b ± sq) << 6) / (2a)
SSINT64 denom = 2 * a;
if (denom == 0) continue;
SSINT64 num0 = -b + (SSINT64)sq;
SSINT64 num1 = -b - (SSINT64)sq;
t0 = (f26_6)(num0 == 0 ? 0 : (num0 << 6) / denom);
t1 = (f26_6)(num1 == 0 ? 0 : (num1 << 6) / denom);
if (t0 > 0 && t0 < F26_ONE) t0_valid = 1;
if (t1 > 0 && t1 < F26_ONE) t1_valid = 1;
}
if (t0_valid) {
// x(t) = (1-t)^2 x0 + 2(1-t)t cx + t^2 x1
f26_6 omt = F26_ONE - t0;
f26_6 x = f26_mul(f26_mul(omt, omt), g->x0)
+ f26_mul(2 * f26_mul(omt, t0), g->cx)
+ f26_mul(f26_mul(t0, t0), g->x1);
xs[nxs++] = x;
}
if (t1_valid) {
f26_6 omt = F26_ONE - t1;
f26_6 x = f26_mul(f26_mul(omt, omt), g->x0)
+ f26_mul(2 * f26_mul(omt, t1), g->cx)
+ f26_mul(f26_mul(t1, t1), g->x1);
xs[nxs++] = x;
}
}
}
if (nxs < 2) continue;
// Insertion sort
for (SUINT32 i = 1; i < nxs; i++) {
f26_6 v = xs[i]; SUINT32 j = i;
while (j > 0 && xs[j-1] > v) { xs[j] = xs[j-1]; j--; }
xs[j] = v;
}
// Fill alternating pairs
for (SUINT32 i = 0; i + 1 < nxs; i += 2) {
f26_6 xa = xs[i];
f26_6 xb = xs[i+1];
if (xa == xb) continue;
SSINT32 c0 = F26_FLOOR(xa);
SSINT32 c1 = F26_FLOOR(xb);
if (c0 == c1) continue;
if (c0 < 0) c0 = 0;
if (c1 > (SSINT32)w) c1 = (SSINT32)w;
for (SSINT32 x = c0; x < c1; x++) {
if (x >= 0 && x < (SSINT32)w) coverage[row * w + x]++;
}
}
}
}
// (Alpha conversion happens at the caller via N.)
(void)x0; (void)y0;
}