Optimize screen scrolling, SIMD wrap padding, and reflow loops

- Batch full-screen scroll rotations, eliminating row-by-row scroll loops
- Use block memmove for sub-region scrolling, eliminating heap/stack row swaps
- Vectorize CELL_WRAP_PAD flag setting with AVX2/SSE2/NEON in screen.h
- Replace scalar tab-stop loops with 64-bit word stamping
- Optimize active row backwards scan in resize and erase mode 2

1 files changed, 58 insertions, 0 deletions

diff --git a/src/screen.h b/src/screen.h
index 0110049..50e18c6 100644
--- a/src/screen.h
+++ b/src/screen.h
@@ -42,6 +42,64 @@ static inline int phys_row(const struct term *t, int y)
     return p >= t->scr.rows ? p - t->scr.rows : p;
 }
 
+#if defined(TERM_KERNEL_AVX2)
+TERM_TARGET_AVX2
+static inline void set_cells_wrap_pad_avx2(struct cell *cells, int count)
+{
+    const __m256i mask = _mm256_setr_epi32(0, 0, 0, (int)CELL_WRAP_PAD,
+                                            0, 0, 0, (int)CELL_WRAP_PAD);
+    int i = 0;
+    for (; i + 4 <= count; i += 4) {
+        __m256i v0 = _mm256_loadu_si256((const __m256i *)(cells + i));
+        __m256i v1 = _mm256_loadu_si256((const __m256i *)(cells + i + 2));
+        v0 = _mm256_or_si256(v0, mask);
+        v1 = _mm256_or_si256(v1, mask);
+        _mm256_storeu_si256((__m256i *)(cells + i), v0);
+        _mm256_storeu_si256((__m256i *)(cells + i + 2), v1);
+    }
+    for (; i < count; i++)
+        cells[i].flags |= CELL_WRAP_PAD;
+}
+#endif
+
+static inline void set_cells_wrap_pad(struct cell *cells, int count)
+{
+    if (count <= 0)
+        return;
+#if defined(TERM_KERNEL_AVX2)
+    if (term_use_avx2() && count >= 4) {
+        set_cells_wrap_pad_avx2(cells, count);
+        return;
+    }
+#endif
+#if defined(__x86_64__) || defined(_M_X64)
+    const __m128i mask = _mm_setr_epi32(0, 0, 0, (int)CELL_WRAP_PAD);
+    int i = 0;
+    for (; i + 2 <= count; i += 2) {
+        __m128i v0 = _mm_loadu_si128((const __m128i *)(cells + i));
+        __m128i v1 = _mm_loadu_si128((const __m128i *)(cells + i + 1));
+        _mm_storeu_si128((__m128i *)(cells + i), _mm_or_si128(v0, mask));
+        _mm_storeu_si128((__m128i *)(cells + i + 1), _mm_or_si128(v1, mask));
+    }
+    for (; i < count; i++)
+        cells[i].flags |= CELL_WRAP_PAD;
+#elif defined(__aarch64__)
+    const uint32x4_t mask = { 0, 0, 0, CELL_WRAP_PAD };
+    int i = 0;
+    for (; i + 2 <= count; i += 2) {
+        uint32x4_t v0 = vld1q_u32((const uint32_t *)(cells + i));
+        uint32x4_t v1 = vld1q_u32((const uint32_t *)(cells + i + 1));
+        vst1q_u32((uint32_t *)(cells + i), vorrq_u32(v0, mask));
+        vst1q_u32((uint32_t *)(cells + i + 1), vorrq_u32(v1, mask));
+    }
+    for (; i < count; i++)
+        cells[i].flags |= CELL_WRAP_PAD;
+#else
+    for (int i = 0; i < count; i++)
+        cells[i].flags |= CELL_WRAP_PAD;
+#endif
+}
+
 static inline void init_ascii_text_row(struct text_row *tr, uint8_t *text, uint16_t len)
 {
     tr->text = text;