Optimize term_feed scalar dispatch, branchless range search, and invalid subparts

1 files changed, 33 insertions, 0 deletions

diff --git a/src/core.h b/src/core.h
index 79e6eb9..aab2cde 100644
--- a/src/core.h
+++ b/src/core.h
@@ -30,6 +30,39 @@ static inline void sc_print_ascii(struct term *t, const char *s, int n);
 /* Print one glyph (single codepoint or cluster handle). */
 void term_sc_print_glyph(struct term *t, uint32_t cp, int width, uint16_t flags);
 
+static inline void sc_print_cell(struct term *t, uint32_t cp, int width, uint16_t flags)
+{
+    if (TERM_EXPECT(width == 1 && !t->pending_wrap && !t->insert_mode &&
+                    !t->lr_margin_mode && t->cursor_x < t->cols, 1)) {
+        struct srow *srow = term_row(t, t->cursor_y);
+        int cx = t->cursor_x;
+        int len = srow->len;
+        if (TERM_EXPECT(srow->kind == ROW_CELLS && cx <= len, 1)) {
+            struct cell *row = srow->cells;
+            if (TERM_EXPECT(!((cx < len && (row[cx].flags & CELL_TRAIL)) ||
+                              (cx + 1 < len && (row[cx + 1].flags & CELL_TRAIL))), 1)) {
+                row[cx] = (struct cell){cp,
+                    t->inverse ? t->cur_bg : t->cur_fg,
+                    t->inverse ? t->cur_fg : t->cur_bg,
+                    flags, t->cur_link};
+                t->last_char = cp;
+                cx++;
+                if (cx > len) srow->len = (uint16_t)cx;
+                srow->version++;
+                if (cx < t->cols) {
+                    t->cursor_x = cx;
+                    t->pending_wrap = false;
+                } else {
+                    t->cursor_x = t->cols - 1;
+                    t->pending_wrap = t->autowrap;
+                }
+                return;
+            }
+        }
+    }
+    term_sc_print_glyph(t, cp, width, flags);
+}
+
 /* Extend the last printed glyph into a grapheme cluster. */
 bool term_sc_extend_cluster(struct term *t, uint32_t cp, int klass);