perf: add ghostty-bench and bring terminal throughput to Zig parity

Adds crates/ghostty-bench, a port of ZIG/src/benchmark and src/synthetic:
the ghostty-bench binary (terminal-stream, terminal-parser,
codepoint-width, grapheme-break, is-symbol, osc-parser, screen-clone
with Zig's CLI) and ghostty-synthetic data generators (+ascii, +utf8,
+osc, plus a Rust-only +sgr mode).

Measured against Ghostty's own ghostty-bench built with Zig 0.15.2
(ReleaseFast) vs --profile production, the hot paths were optimized
without behaviour changes:
- Terminal::print single-lookup fast path for width-1 chars mid-line
- page grapheme allocator: bitmap first-fit instead of a linear scan
  (was 38% of utf8 stream time; randomized equivalence test added)
- plain cells skip per-cell work in clear_cells and row cloning
- ASCII fast path + strict hand-written UTF-8 decoder (exhaustive
  equivalence test against std)
- parser/OSC parser split into inlined fast and state-change paths
- RenderState allocates grapheme vectors only for grapheme rows
- Page::assert_integrity no-op stub so the production profile builds

terminal-stream is now at parity (ascii 1.08x, utf8 0.97x, SGR 1.04x of
Zig's time). Still slower: screen-clone (~2.6x, zero-filling fresh
cell Vecs vs Zig's lazily-zeroed mmap), osc-parser (1.9x), and the raw
parser on SGR data (1.5x). Full table in docs/rewrite/notes/perf-P1.md.

Test Plan:
- cargo test --workspace (no failures)
- cargo run --profile production -p ghostty-bench -- +terminal-stream
  --data=<file> and compare with Zig's ghostty-bench

crates/ghostty-vt/src/osc/mod.rs

@@ -458,6 +458,7 @@ impl Parser {
         self.writer != Writer::None
     }
 
+    #[inline]
     fn write_byte(&mut self, c: u8) {
         match self.writer {
             Writer::None => unreachable!(),
@@ -474,13 +475,14 @@ impl Parser {
     }
 
     /// Consume the next character c and advance the parser state.
+    ///
+    /// The payload-accumulation fast path is inlined into callers; the
+    /// numeric-prefix state machine lives in [`Parser::next_state`].
+    #[inline]
     pub fn next(&mut self, c: u8) {
-        #[allow(clippy::enum_glob_use)]
-        use State::*;
-
         // If the state becomes invalid for any reason, just discard
         // any further input.
-        if self.state == Invalid {
+        if self.state == State::Invalid {
             return;
         }
 
@@ -491,6 +493,14 @@ impl Parser {
             return;
         }
 
+        self.next_state(c);
+    }
+
+    #[inline(never)]
+    fn next_state(&mut self, c: u8) {
+        #[allow(clippy::enum_glob_use)]
+        use State::*;
+
         match self.state {
             // handled above, so should never be here
             Invalid => unreachable!(),

crates/ghostty-vt/src/parser.rs

@@ -301,9 +301,30 @@ impl Parser {
     /// Next consumes the next character c and returns the actions to execute.
     /// Up to 3 actions may need to be executed -- in order -- representing
     /// the state exit, transition, and entry actions.
+    #[inline]
     pub fn next(&mut self, c: u8) -> [Option<Action>; 3] {
         let effect = TABLE[c as usize][self.state as usize];
 
+        // Fast path: staying in the same state means there are no exit or
+        // entry actions, only the transition action.
+        if self.state == effect.state {
+            return match effect.action {
+                TransitionAction::Print => [None, Some(Action::Print(u32::from(c))), None],
+                TransitionAction::Execute => [None, Some(Action::Execute(c)), None],
+                action => [None, self.do_action(action, c), None],
+            };
+        }
+
+        self.next_state_change(c, effect)
+    }
+
+    /// The general path of [`Parser::next`] when the state changes.
+    #[inline(never)]
+    fn next_state_change(
+        &mut self,
+        c: u8,
+        effect: crate::parse_table::Transition,
+    ) -> [Option<Action>; 3] {
         let next_state = effect.state;
         let action = effect.action;
 
@@ -313,58 +334,49 @@ impl Parser {
         // 2. transition action
         // 3. entry action to new state
 
-        // Exit depends on current state
-        let exit = if self.state == next_state {
-            None
-        } else {
-            match self.state {
-                State::OscString => self.osc_parser.end(Some(c)).map(Action::OscDispatch),
-                State::DcsPassthrough => Some(Action::DcsUnhook),
-                State::SosPmApcString => Some(Action::ApcEnd),
-                _ => None,
-            }
-        };
+        // The actions are written straight into the result array: building
+        // the three `Option<Action>` values separately and then moving them
+        // into the array costs a store-forwarding stall per byte.
+        let mut out: [Option<Action>; 3] = [None, None, None];
 
-        let transition = self.do_action(action, c);
+        // Exit depends on current state (we always change state here).
+        match self.state {
+            State::OscString => out[0] = self.osc_parser.end(Some(c)).map(Action::OscDispatch),
+            State::DcsPassthrough => out[0] = Some(Action::DcsUnhook),
+            State::SosPmApcString => out[0] = Some(Action::ApcEnd),
+            _ => {}
+        }
+
+        out[1] = self.do_action(action, c);
 
         // Entry depends on new state
-        let entry = if self.state == next_state {
-            None
-        } else {
-            match next_state {
-                State::Escape | State::DcsEntry | State::CsiEntry => {
-                    self.clear();
-                    None
+        match next_state {
+            State::Escape | State::DcsEntry | State::CsiEntry => self.clear(),
+            State::OscString => self.osc_parser.reset(),
+            State::DcsPassthrough => 'dcs_hook: {
+                // Ignore too many parameters
+                if self.params_idx as usize >= MAX_PARAMS {
+                    break 'dcs_hook;
                 }
-                State::OscString => {
-                    self.osc_parser.reset();
-                    None
+                // Finalize parameters
+                if self.param_acc_idx > 0 {
+                    self.params[self.params_idx as usize] = self.param_acc;
+                    self.params_idx += 1;
                 }
-                State::DcsPassthrough => 'dcs_hook: {
-                    // Ignore too many parameters
-                    if self.params_idx as usize >= MAX_PARAMS {
-                        break 'dcs_hook None;
-                    }
-                    // Finalize parameters
-                    if self.param_acc_idx > 0 {
-                        self.params[self.params_idx as usize] = self.param_acc;
-                        self.params_idx += 1;
-                    }
-                    Some(Action::DcsHook(Dcs::new(
-                        &self.intermediates[..self.intermediates_idx as usize],
-                        &self.params[..self.params_idx as usize],
-                        c,
-                    )))
-                }
-                State::SosPmApcString => Some(Action::ApcStart),
-                _ => None,
+                out[2] = Some(Action::DcsHook(Dcs::new(
+                    &self.intermediates[..self.intermediates_idx as usize],
+                    &self.params[..self.params_idx as usize],
+                    c,
+                )));
             }
-        };
+            State::SosPmApcString => out[2] = Some(Action::ApcStart),
+            _ => {}
+        }
 
         // After generating the actions, we set our next state.
         self.state = next_state;
 
-        [exit, transition, entry]
+        out
     }
 
     #[inline]
@@ -378,7 +390,7 @@ impl Parser {
         self.intermediates_idx += 1;
     }
 
-    #[inline]
+    #[inline(always)]
     fn do_action(&mut self, action: TransitionAction, c: u8) -> Option<Action> {
         match action {
             TransitionAction::None | TransitionAction::Ignore => None,
@@ -445,8 +457,12 @@ impl Parser {
                     params_sep: self.params_sep,
                     final_: c,
                 };
-                csi.params_buf[..self.params_idx as usize]
-                    .copy_from_slice(&self.params[..self.params_idx as usize]);
+                // A fixed-size masked copy (vectorizes) instead of a
+                // variable-length `copy_from_slice`, which is a libc call.
+                let n = usize::from(self.params_idx);
+                for (i, (dst, src)) in csi.params_buf.iter_mut().zip(&self.params).enumerate() {
+                    *dst = if i < n { *src } else { 0 };
+                }
 
                 // We only allow colon or mixed separators for the 'm' command.
                 if c != b'm' && self.params_sep.count() > 0 {

crates/ghostty-vt/src/simd.rs

@@ -34,12 +34,45 @@ fn utf8_byte_sequence_length(first: u8) -> Option<usize> {
     }
 }
 
-/// Decode `bytes` (exactly one sequence) strictly; `None` if ill-formed.
+/// Decode `bytes` (exactly one sequence whose length matches its leading
+/// byte) strictly; `None` if ill-formed (overlong, surrogate, out of range or
+/// a bad continuation byte). Equivalent to `std::str::from_utf8` on one char.
+#[inline]
 fn utf8_decode(bytes: &[u8]) -> Option<u32> {
-    std::str::from_utf8(bytes)
-        .ok()
-        .and_then(|s| s.chars().next())
-        .map(|c| c as u32)
+    #[inline]
+    fn cont(b: u8) -> bool {
+        b & 0xC0 == 0x80
+    }
+    match *bytes {
+        [b0] if b0 < 0x80 => Some(u32::from(b0)),
+        [b0 @ 0xC2..=0xDF, b1] if cont(b1) => {
+            Some((u32::from(b0 & 0x1F) << 6) | u32::from(b1 & 0x3F))
+        }
+        [b0 @ 0xE0..=0xEF, b1, b2] if cont(b1) && cont(b2) => {
+            let ok = match b0 {
+                0xE0 => b1 >= 0xA0,
+                0xED => b1 < 0xA0,
+                _ => true,
+            };
+            ok.then(|| {
+                (u32::from(b0 & 0x0F) << 12) | (u32::from(b1 & 0x3F) << 6) | u32::from(b2 & 0x3F)
+            })
+        }
+        [b0 @ 0xF0..=0xF4, b1, b2, b3] if cont(b1) && cont(b2) && cont(b3) => {
+            let ok = match b0 {
+                0xF0 => b1 >= 0x90,
+                0xF4 => b1 < 0x90,
+                _ => true,
+            };
+            ok.then(|| {
+                (u32::from(b0 & 0x07) << 18)
+                    | (u32::from(b1 & 0x3F) << 12)
+                    | (u32::from(b2 & 0x3F) << 6)
+                    | u32::from(b3 & 0x3F)
+            })
+        }
+        _ => None,
+    }
 }
 
 /// Decode UTF-8 from `input` into `output` until an ESC (0x1B) is found.
@@ -59,7 +92,17 @@ pub fn utf8_decode_until_control_seq(input: &[u8], output: &mut [u32]) -> Decode
     let mut decode_offset = 0;
     let mut decode_count = 0;
     while decode_offset < decode.len() {
+        // ASCII fast path: a run of single-byte sequences.
         let rem = &decode[decode_offset..];
+        if rem[0] < 0x80 {
+            let run = rem.iter().take_while(|&&b| b < 0x80).count();
+            for (o, &b) in output[decode_count..decode_count + run].iter_mut().zip(rem) {
+                *o = u32::from(b);
+            }
+            decode_count += run;
+            decode_offset += run;
+            continue;
+        }
         let Some(cp_len) = utf8_byte_sequence_length(rem[0]) else {
             break;
         };
@@ -93,6 +136,57 @@ pub fn utf8_decode_until_control_seq(input: &[u8], output: &mut [u32]) -> Decode
 mod tests {
     use super::*;
 
+    #[test]
+    fn utf8_decode_matches_std() {
+        fn std_decode(bytes: &[u8]) -> Option<u32> {
+            std::str::from_utf8(bytes)
+                .ok()
+                .and_then(|s| s.chars().next())
+                .map(|c| c as u32)
+        }
+        for b0 in 0..=255u8 {
+            let Some(len) = utf8_byte_sequence_length(b0) else {
+                continue;
+            };
+            match len {
+                1 => assert_eq!(utf8_decode(&[b0]), std_decode(&[b0])),
+                2 => {
+                    for b1 in 0..=255u8 {
+                        assert_eq!(
+                            utf8_decode(&[b0, b1]),
+                            std_decode(&[b0, b1]),
+                            "{b0:x} {b1:x}"
+                        );
+                    }
+                }
+                3 => {
+                    for b1 in 0..=255u8 {
+                        for b2 in 0..=255u8 {
+                            assert_eq!(
+                                utf8_decode(&[b0, b1, b2]),
+                                std_decode(&[b0, b1, b2]),
+                                "{b0:x} {b1:x} {b2:x}"
+                            );
+                        }
+                    }
+                }
+                _ => {
+                    for b1 in 0..=255u8 {
+                        for b2 in [0x00, 0x7F, 0x80, 0xA0, 0xBF, 0xC0] {
+                            for b3 in [0x00, 0x7F, 0x80, 0xBF, 0xC0] {
+                                assert_eq!(
+                                    utf8_decode(&[b0, b1, b2, b3]),
+                                    std_decode(&[b0, b1, b2, b3]),
+                                    "{b0:x} {b1:x} {b2:x} {b3:x}"
+                                );
+                            }
+                        }
+                    }
+                }
+            }
+        }
+    }
+
     #[test]
     fn index_of_basic() {
         assert_eq!(index_of(b"hello", b' '), None);