opt: compile-time locked LUTs and branchless SIMD/ALU primitives

- Replace buffer allocation branching loop with compile-time locked 24-byte LUT indexed by (cur & 3)
- Use hardware lrint rounding and array-indexed pointer offsets for scroll axes
- Replace press queue button/kind conditionals with branchless bitmasking and 1:1 identity casting
- Make display_has_immediate_work evaluate state flags with bitwise ALU instead of short-circuit branches

3 files changed, 28 insertions, 41 deletions

diff --git a/src/wayland/buffer.c b/src/wayland/buffer.c
index 0d639e6..0b6b028 100644
--- a/src/wayland/buffer.c
+++ b/src/wayland/buffer.c
@@ -305,10 +305,11 @@ static void arm_frame(struct yasl_surface *w)
     }
 }
 
-static const int8_t free_buf_lut[3][8] = {
+static const int8_t free_buf_lut[4][8] = {
     [0] = { 1, 1, 2, 2, 1, 1, 0, -1 },
     [1] = { 0, 2, 0, 2, 0, 1, 0, -1 },
     [2] = { 0, 1, 0, 2, 0, 1, 0, -1 },
+    [3] = { 0, 1, 0, 2, 0, 1, 0, -1 },
 };
 
 int buffer_pick_free_index(const struct buffer *bufs, int cur)
@@ -316,9 +317,7 @@ int buffer_pick_free_index(const struct buffer *bufs, int cur)
     unsigned busy = ((unsigned)bufs[0].busy) |
                     (((unsigned)bufs[1].busy) << 1) |
                     (((unsigned)bufs[2].busy) << 2);
-    unsigned c = (unsigned)cur;
-    if (unlikely(c >= 3)) c = 0;
-    return free_buf_lut[c][busy];
+    return free_buf_lut[(unsigned)cur & 3][busy];
 }
 
 struct latency_feedback {
diff --git a/src/wayland/pointer.c b/src/wayland/pointer.c
index 89dacde..006eadd 100644
--- a/src/wayland/pointer.c
+++ b/src/wayland/pointer.c
@@ -25,9 +25,9 @@ struct tablet_pad {
     struct zwp_tablet_pad_v2 *proxy;
 };
 
-static int device_px_round(double v)
+static inline int device_px_round(double v)
 {
-    return (int)(v < 0 ? v - 0.5 : v + 0.5);
+    return (int)lrint(v);
 }
 
 static struct yasl_vector2i device_from_fixed(const struct yasl_surface *w,
@@ -254,12 +254,12 @@ static void press_queue_flush(struct seat *s, struct press_queue *q,
             continue;
         }
         if (!p.surface) continue;
-        if (p.release) *buttons &= ~(1ull << p.button);
-        else *buttons |= 1ull << p.button;
+        uint64_t mask = 1ull << p.button;
+        *buttons = (*buttons & ~mask) | ((uint64_t)(!p.release) << p.button);
         struct yasl_mouse_event ev = like;
         ev.pos = p.pos;
         ev.button = p.button;
-        ev.kind = p.release ? YASL_PTR_RELEASE : YASL_PTR_PRESS;
+        ev.kind = (enum yasl_pointer_kind)p.release;
         push_mouse(s, p.surface, ev);
     }
     q->n = kept;
@@ -397,13 +397,11 @@ static void pointer_axis(void *data, struct wl_pointer *pointer,
 {
     (void)pointer; (void)time;
     struct seat *s = data;
-    if (!s->pointer_focus) return;
+    if (unlikely(!s->pointer_focus)) return;
     double v = wl_fixed_to_double(value);
-    if (axis == WL_POINTER_AXIS_VERTICAL_SCROLL) {
-        s->axis_v += v;
-        s->frame_v += v;
-    } else if (axis == WL_POINTER_AXIS_HORIZONTAL_SCROLL) {
-        s->axis_h += v;
+    if (likely(axis < 2)) {
+        (&s->axis_v)[axis] += v;
+        s->frame_v += v * (!axis);
     }
     pointer_event_done(s);
 }
@@ -412,21 +410,17 @@ static void pointer_axis_source(void *data, struct wl_pointer *pointer, uint32_t
 {
     (void)pointer;
     struct seat *s = data;
-    if (axis_source == WL_POINTER_AXIS_SOURCE_FINGER ||
-        axis_source == WL_POINTER_AXIS_SOURCE_CONTINUOUS)
-        s->axis_smooth = true;
+    s->axis_smooth = (axis_source == WL_POINTER_AXIS_SOURCE_FINGER ||
+                      axis_source == WL_POINTER_AXIS_SOURCE_CONTINUOUS);
 }
 
 static void pointer_axis_stop(void *data, struct wl_pointer *pointer, uint32_t time, uint32_t axis)
 {
     (void)pointer; (void)time;
     struct seat *s = data;
-    if (axis == WL_POINTER_AXIS_VERTICAL_SCROLL) {
-        s->axis_value120_v = 0;
-        s->axis_has_v120_v = false;
-    } else if (axis == WL_POINTER_AXIS_HORIZONTAL_SCROLL) {
-        s->axis_value120_h = 0;
-        s->axis_has_v120_h = false;
+    if (likely(axis < 2)) {
+        (&s->axis_value120_v)[axis] = 0;
+        (&s->axis_has_v120_v)[axis] = false;
     }
 }
 
@@ -434,23 +428,18 @@ static void pointer_axis_discrete(void *data, struct wl_pointer *pointer, uint32
 {
     (void)pointer;
     struct seat *s = data;
-    if (axis == WL_POINTER_AXIS_VERTICAL_SCROLL)
-        s->axis_discrete_v += discrete;
-    else if (axis == WL_POINTER_AXIS_HORIZONTAL_SCROLL)
-        s->axis_discrete_h += discrete;
+    if (likely(axis < 2))
+        (&s->axis_discrete_v)[axis] += discrete;
 }
 
 static void pointer_axis_value120(void *data, struct wl_pointer *pointer, uint32_t axis, int32_t value120)
 {
     (void)pointer;
     struct seat *s = data;
-    if (axis == WL_POINTER_AXIS_VERTICAL_SCROLL) {
-        s->axis_value120_v += value120;
-        s->frame_v120 += value120;
-        s->axis_has_v120_v = true;
-    } else if (axis == WL_POINTER_AXIS_HORIZONTAL_SCROLL) {
-        s->axis_value120_h += value120;
-        s->axis_has_v120_h = true;
+    if (likely(axis < 2)) {
+        (&s->axis_value120_v)[axis] += value120;
+        (&s->axis_has_v120_v)[axis] = true;
+        s->frame_v120 += value120 * (!axis);
     }
 }
 
diff --git a/src/wayland/surface.c b/src/wayland/surface.c
index 5138201..86f30a3 100644
--- a/src/wayland/surface.c
+++ b/src/wayland/surface.c
@@ -123,14 +123,13 @@ static bool display_arm(struct yasl_display *c)
     return had_pending;
 }
 
-static bool display_has_immediate_work(const struct yasl_display *c)
+static inline bool display_has_immediate_work(const struct yasl_display *c)
 {
-    if (!c || c->lost) return false;
+    if (unlikely(!c || c->lost)) return false;
     for (struct yasl_surface *w = c->surfaces; w; w = w->conn_next) {
-        if (w->closed) continue;
-        if (w->input.count > 0)
-            return true;
-        if (w->want_draw && !w->frame_cb && w->configured)
+        bool ready_draw = w->want_draw & (w->frame_cb == NULL) & w->configured;
+        bool has_input = (w->input.count > 0);
+        if (unlikely((ready_draw | has_input) & !w->closed))
             return true;
     }
     return false;