performance improvements

Cargo.lock

@@ -175,6 +175,7 @@ dependencies = [
  "anyhow",
  "dyn-clonable",
  "log",
+ "nudge",
  "num-traits",
  "parking_lot",
  "sexp",
@@ -443,6 +444,15 @@ dependencies = [
  "version_check",
 ]
 
+[[package]]
+name = "nudge"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5f379c3eafab28638ffb5cd95f1091dd21f5c6e7699e40b2ee96e9764a87f779"
+dependencies = [
+ "cfg-if",
+]
+
 [[package]]
 name = "num"
 version = "0.1.42"

README.md

@@ -1,6 +1,6 @@
 # CROISSANT VIRTUAL MACHINE
 
-Croissant (or Crsn for short) is an extensible runtime emulating a weird microcomputer (or not so micro, that depends on what extensions you install).
+Croissant (or *crsn* for short) is an extensible runtime emulating a weird microcomputer (or not so micro, that depends on what extensions you install).
 
 ## FAQ
 
@@ -8,6 +8,13 @@ Croissant (or Crsn for short) is an extensible runtime emulating a weird microco
 
 F U N
 
+### How is the performance?
+
+Silly fast, actually. 60fps animations are perfectly doable if that's your thing.
+It's probably faster than you need for most things, actually. 
+
+You can slow it down using the `-C` argument, or using sleep instructions.
+
 #### What if I don't enjoy writing assembly that looks like weird Lisp?
 
 Maybe this is not for you

crsn/Cargo.toml

@@ -13,3 +13,4 @@ dyn-clonable = "0.9.0"
 log = "0.4.11"
 num-traits = "0.2.12"
 parking_lot = "0.11.0"
+nudge = "0.2.1"

crsn/src/asm/data/literal.rs

@@ -7,15 +7,18 @@ pub type DebugMsg = Cow<'static, str>;
 /// Immediate value
 pub type Value = u64;
 
-pub fn is_positive(val: Value) -> bool {
+#[inline(always)]
+pub const fn is_positive(val: Value) -> bool {
     0 == (val & 0x8000_0000_0000_0000)
 }
 
-pub fn is_negative(val: Value) -> bool {
+#[inline(always)]
+pub const fn is_negative(val: Value) -> bool {
     0 != (val & 0x8000_0000_0000_0000)
 }
 
-pub fn as_signed(val: Value) -> i64 {
+#[inline(always)]
+pub const fn as_signed(val: Value) -> i64 {
     i64::from_ne_bytes(val.to_ne_bytes())
 }
 
@@ -24,6 +27,7 @@ pub fn as_signed(val: Value) -> i64 {
 pub struct Addr(pub u64);
 
 impl Addr {
+    #[inline(always)]
     pub fn advance(&mut self, add: i64) {
         if add < 0 {
             self.0 = self.0.wrapping_sub(-add as u64);

crsn/src/asm/data/rd.rs

@@ -12,15 +12,17 @@ impl Rd {
     pub const fn new(src: RdData) -> Self {
         Rd(src, Mask::FULL)
     }
-    pub fn data(self) -> RdData {
+
+    pub const fn data(self) -> RdData {
         self.0
     }
-    pub fn mask(self) -> Mask {
+
+    pub const fn mask(self) -> Mask {
         self.1
     }
 
-    pub fn immediate(val: Value) -> Rd {
-        Rd(RdData::Immediate(val), Mask::default())
+    pub const fn immediate(val: Value) -> Rd {
+        Rd(RdData::Immediate(val), Mask::FULL)
     }
 }
 
@@ -48,7 +50,7 @@ impl Debug for Rd {
 pub struct RdObj(Register);
 
 impl RdObj {
-    pub fn new(reg: Register) -> Self {
+    pub const fn new(reg: Register) -> Self {
         RdObj(reg)
     }
     pub const fn reg(self) -> Register {

crsn/src/asm/mod.rs

@@ -5,8 +5,9 @@ use sexp::SourcePosition;
 
 use crate::asm::instr::flatten::labels_to_skips;
 use crate::asm::parse::{ParserContext, ParserState};
-use crate::module::CrsnExtension;
+use crate::module::{CrsnExtension, CrsnUniq};
 use crate::runtime::program::Program;
+use crate::builtin::BuiltinOps;
 
 pub mod data;
 pub mod error;
@@ -15,7 +16,13 @@ pub mod parse;
 pub mod patches;
 
 /// Parse a program from string and assemble a low level instruction sequence from it.
-pub fn assemble(source: &str, parsers: Arc<Vec<Box<dyn CrsnExtension>>>) -> Result<Arc<Program>, error::CrsnError> {
+pub fn assemble(source: &str, uniq : &CrsnUniq, mut parsers: Vec<Box<dyn CrsnExtension>>) -> Result<Arc<Program>, error::CrsnError> {
+    parsers.insert(0, BuiltinOps::new());
+
+    for p in &mut parsers {
+        p.init(uniq);
+    }
+
     let pcx = ParserContext {
         parsers: &parsers,
         state: RefCell::new(ParserState {
@@ -28,5 +35,5 @@ pub fn assemble(source: &str, parsers: Arc<Vec<Box<dyn CrsnExtension>>>) -> Resu
     let ops = parse::parse(source, &SourcePosition::default(), &pcx)?;
     let ops = labels_to_skips(ops)?;
 
-    Ok(Program::new(ops, parsers)?)
+    Ok(Program::new(ops, Arc::new(parsers))?)
 }

crsn/src/asm/parse/parse_op.rs

@@ -4,12 +4,11 @@ use crate::asm::error::CrsnError;
 use crate::asm::instr::cond::parse_cond;
 use crate::asm::instr::Op;
 use crate::asm::parse::arg_parser::TokenParser;
-use crate::builtin::BuiltinOps;
+
 use crate::module::ParseRes;
 
 pub fn parse_op<'a>(mut keyword: &str, mut arg_tokens: TokenParser<'a>, spos: &SourcePosition) -> Result<Option<Op>, CrsnError> {
     // Include built-in instructions
-    let builtins = [BuiltinOps::new()];
     let mut cond = None;
 
     if let Some(pos) = keyword.find('.') {
@@ -17,7 +16,7 @@ pub fn parse_op<'a>(mut keyword: &str, mut arg_tokens: TokenParser<'a>, spos: &S
         keyword = &keyword[..pos];
     }
 
-    for p in builtins.iter().chain(arg_tokens.pcx.parsers) {
+    for p in arg_tokens.pcx.parsers {
         arg_tokens = match p.parse_op(spos, keyword, arg_tokens) {
             Ok(ParseRes::Parsed(kind)) => return Ok(Some(Op {
                 cond,

crsn/src/module/mod.rs

@@ -14,6 +14,8 @@ use crate::asm::parse::arg_parser::TokenParser;
 use crate::runtime::fault::Fault;
 use crate::runtime::run_thread::state::RunState;
 use crate::runtime::run_thread::ThreadInfo;
+use std::sync::atomic::{AtomicU64, Ordering};
+use std::sync::Arc;
 
 mod eval_res;
 
@@ -41,7 +43,33 @@ pub trait OpTrait: Debug + Send + Sync + 'static {
     fn to_sexp(&self) -> Sexp;
 }
 
+/// CRSN initializer object.
+/// Only one should be created for the lifespan of the parser and runtime.
+#[derive(Default)]
+pub struct CrsnUniq {
+    object_handle_counter : AtomicU64
+}
+
+pub const UNIQ_BASE: u64 = 0x6372_736e_0000_0000;
+
+impl CrsnUniq {
+    pub fn new() -> Arc<Self> {
+        Arc::new(Self {
+            object_handle_counter: AtomicU64::new(UNIQ_BASE),
+        })
+    }
+
+    pub fn unique_handle(&self) -> u64 {
+        self.object_handle_counter.fetch_add(1, Ordering::Relaxed)
+    }
+}
+
+
 pub trait CrsnExtension: Debug + Send + Sync + 'static {
+    fn init(&mut self, uniq: &CrsnUniq) {
+        //
+    }
+
     /// Get name of the module
     fn name(&self) -> &'static str;
 

crsn/src/runtime/exec.rs

@@ -8,7 +8,7 @@ impl RunThread {
         let state = &mut self.state;
         let info = &self.info;
 
-        let op = info.program.read(state.frame.pc);
+        let op = info.program.fetch_instr(state.frame.pc);
 
         trace!("### {:04} : {:?}", state.frame.pc.0, op);
 

crsn/src/runtime/frame.rs

@@ -33,6 +33,7 @@ impl StackFrame {
         sf
     }
 
+    #[inline(always)]
     pub fn set_retvals(&mut self, vals: &[Value]) {
         for n in 0..(vals.len().min(REG_COUNT)) {
             self.res[n] = vals[n];

crsn/src/runtime/frame/status.rs

@@ -27,6 +27,7 @@ pub struct StatusFlags {
 }
 
 impl StatusFlags {
+    #[inline(always)]
     pub fn clear(&mut self) {
         *self = Self::default();
     }
@@ -57,12 +58,14 @@ impl StatusFlags {
         if val & 0x100 != 0 { self.carry = true; }
     }
 
+    #[inline(always)]
     pub fn update(&mut self, val: Value) {
         self.zero = val == 0;
         self.positive = is_positive(val);
         self.negative = is_negative(val);
     }
 
+    #[inline(always)]
     pub fn test(&self, cond: Cond) -> bool {
         match cond {
             Cond::Equal => self.equal,
@@ -86,6 +89,7 @@ impl StatusFlags {
         }
     }
 
+    #[inline(always)]
     pub fn set(&mut self, cond: Cond) {
         match cond {
             Cond::Equal => {

crsn/src/runtime/program.rs

@@ -90,8 +90,10 @@ impl Program {
     }
 
     /// Read a program instruction at address
-    pub fn read(&self, addr: Addr) -> &Op {
-        if addr.0 >= self.ops.len() as u64 {
+    pub fn fetch_instr(&self, addr: Addr) -> &Op {
+        if (addr.0 as usize) < self.ops.len() {
+            unsafe { self.ops.get_unchecked(addr.0 as usize) }
+        } else {
             &Op {
                 kind: OpKind::BuiltIn(BuiltinOp::Halt),
                 pos: SourcePosition {
@@ -101,8 +103,6 @@ impl Program {
                 },
                 cond: None,
             }
-        } else {
-            &self.ops[addr.0 as usize]
         }
     }
 
@@ -116,7 +116,7 @@ impl Program {
             if b0 != b1 {
                 // block barrier that only partially intersects the jump
                 if (*b0 >= from && *b0 <= to) != (*b1 >= from && *b1 <= to) {
-                    if let OpKind::BuiltIn(BuiltinOp::Barrier { msg, .. }) = &self.read(*b0).kind {
+                    if let OpKind::BuiltIn(BuiltinOp::Barrier { msg, .. }) = &self.fetch_instr(*b0).kind {
                         return Err(Fault::JumpThroughBarrier {
                             msg: msg.clone().unwrap_or("BLOCK BARRIER".into())
                         });
@@ -127,7 +127,7 @@ impl Program {
             } else {
                 // point barrier
                 if *b0 >= from && *b0 <= to {
-                    if let OpKind::BuiltIn(BuiltinOp::Barrier { msg, .. }) = &self.read(*b0).kind {
+                    if let OpKind::BuiltIn(BuiltinOp::Barrier { msg, .. }) = &self.fetch_instr(*b0).kind {
                         return Err(Fault::JumpThroughBarrier {
                             msg: msg.clone().unwrap_or("POINT BARRIER".into())
                         });

crsn/src/runtime/run_thread.rs

@@ -1,5 +1,5 @@
 use std::sync::Arc;
-use std::sync::atomic::AtomicU64;
+
 use std::thread::JoinHandle;
 use std::time::Duration;
 
@@ -7,7 +7,7 @@ pub use info::ThreadInfo;
 pub use state::RunState;
 
 use crate::asm::data::literal::Addr;
-use crate::module::EvalRes;
+use crate::module::{EvalRes, CrsnUniq};
 use crate::runtime::fault::Fault;
 use crate::runtime::frame::StackFrame;
 use crate::runtime::program::Program;
@@ -23,13 +23,9 @@ pub struct RunThread {
 pub mod info;
 pub mod state;
 
-pub fn new_uniq() -> Arc<AtomicU64> {
-    Arc::new(AtomicU64::new(info::UNIQ_BASE))
-}
-
 pub struct ThreadParams<'a> {
     pub id: ThreadToken,
-    pub uniq: Option<Arc<AtomicU64>>,
+    pub uniq: Arc<CrsnUniq>,
     pub program: Arc<Program>,
     pub pc: Addr,
     pub cycle_time: Duration,
@@ -42,7 +38,7 @@ impl RunThread {
 
         let ti = Arc::new(ThreadInfo {
             id: params.id,
-            uniq: params.uniq.unwrap_or_else(new_uniq),
+            uniq: params.uniq,
             program: params.program,
             cycle_time: params.cycle_time,
             extensions,
@@ -73,12 +69,11 @@ impl RunThread {
         'run: loop {
             match self.eval_op() {
                 Ok(EvalRes { cycles, advance }) => {
+                    if cycles > 0 {
                         std::thread::sleep(self.info.cycle_time * (cycles as u32));
+                    }
                     trace!("Step {}; Status = {}", advance, self.state.frame.status);
                     self.state.frame.pc.advance(advance);
-                    if self.state.frame.status.invalid {
-                        warn!("Operation failed with INVALID status!");
-                    }
                 }
                 Err(Fault::Halt) => {
                     // TODO implement coordinated shutdown when more threads are running!

crsn/src/runtime/run_thread/info.rs

@@ -1,17 +1,17 @@
 use std::sync::Arc;
-use std::sync::atomic::{AtomicU64, Ordering};
+
 use std::time::Duration;
 
 use crate::asm::data::literal::Value;
 use crate::runtime::program::Program;
 use crate::runtime::run_thread::ThreadToken;
-use crate::module::CrsnExtension;
+use crate::module::{CrsnExtension, CrsnUniq};
 
 pub struct ThreadInfo {
     /// Thread ID
     pub id: ThreadToken,
-    /// Thread ID
-    pub(crate) uniq: Arc<AtomicU64>,
+    /// Initializer, used to generate unique IDs
+    pub(crate) uniq: Arc<CrsnUniq>,
     /// Program to run
     pub program: Arc<Program>,
     /// Program to run
@@ -20,10 +20,8 @@ pub struct ThreadInfo {
     pub extensions: Arc<Vec<Box<dyn CrsnExtension>>>,
 }
 
-pub const UNIQ_BASE: u64 = 0x6372_736e_0000_0000;
-
 impl ThreadInfo {
-    pub fn unique_value(&self) -> Value {
-        self.uniq.fetch_add(1, Ordering::Relaxed)
+    pub fn unique_handle(&self) -> Value {
+        self.uniq.unique_handle()
     }
 }

crsn/src/runtime/run_thread/state.rs

@@ -8,6 +8,7 @@ use crate::runtime::fault::Fault;
 use crate::runtime::frame::{CallStack, REG_COUNT, StackFrame};
 use std::sync::Arc;
 use crate::runtime::run_thread::ThreadInfo;
+use nudge::{likely};
 
 pub struct RunState {
     pub thread_info: Arc<ThreadInfo>,
@@ -49,6 +50,7 @@ impl RunState {
     }
 
     /// Set a status flag. Only supports simple, positive conds (i.e. not GreaterOrEqual)
+    #[inline(always)]
     pub fn set_flag(&mut self, cond: Cond, set: bool) {
         if set {
             self.frame.status.set(cond);
@@ -56,6 +58,7 @@ impl RunState {
     }
 
     /// Check status flags for a condition
+    #[inline(always)]
     pub fn test_cond(&self, cond: Cond) -> bool {
         self.frame.status.test(cond)
     }
@@ -72,12 +75,14 @@ impl RunState {
     }
 
     /// Clear status flags
+    #[inline(always)]
     pub fn clear_status(&mut self) {
         self.frame.status.clear();
     }
 
     /// Update status flags using a variable.
     /// The update is additive - call `clear_status()` first if desired!
+    #[inline(always)]
     pub fn update_status(&mut self, val: Value) {
         self.frame.status.update(val);
     }
@@ -90,29 +95,29 @@ impl RunState {
     /// Read a `Rd` value
     pub fn read(&mut self, rd: Rd) -> Result<Value, Fault> {
         match rd.data() {
-            RdData::Immediate(v) => Ok(v),
-            RdData::Register(Register::Res(rn)) => {
-                if rn >= REG_COUNT as u8 {
-                    Err(Fault::RegisterNotExist { reg: Register::Res(rn) }) // TODO use match after @ when stabilized https://github.com/rust-lang/rust/issues/65490
+            RdData::Register(Register::Gen(rn)) => {
+                if likely(rn < REG_COUNT as u8) {
+                    trace!("Rd {:?} = {}", rd, self.frame.gen[rn as usize]);
+                    Ok(self.frame.gen[rn as usize])
                 } else {
-                    trace!("Rd {:?} = {}", rd, self.frame.res[rn as usize]);
-                    Ok(self.frame.res[rn as usize])
+                    Err(Fault::RegisterNotExist { reg: Register::Gen(rn) })
                 }
             }
+            RdData::Immediate(v) => Ok(v),
             RdData::Register(Register::Arg(rn)) => {
-                if rn >= REG_COUNT as u8 {
-                    Err(Fault::RegisterNotExist { reg: Register::Arg(rn) })
-                } else {
+                if likely(rn < REG_COUNT as u8) {
                     trace!("Rd {:?} = {}", rd, self.frame.arg[rn as usize]);
                     Ok(self.frame.arg[rn as usize])
+                } else {
+                    Err(Fault::RegisterNotExist { reg: Register::Arg(rn) })
                 }
             }
-            RdData::Register(Register::Gen(rn)) => {
-                if rn >= REG_COUNT as u8 {
-                    Err(Fault::RegisterNotExist { reg: Register::Gen(rn) })
+            RdData::Register(Register::Res(rn)) => {
+                if likely(rn < REG_COUNT as u8) {
+                    trace!("Rd {:?} = {}", rd, self.frame.res[rn as usize]);
+                    Ok(self.frame.res[rn as usize])
                 } else {
-                    trace!("Rd {:?} = {}", rd, self.frame.gen[rn as usize]);
-                    Ok(self.frame.gen[rn as usize])
+                    Err(Fault::RegisterNotExist { reg: Register::Res(rn) }) // TODO use match after @ when stabilized https://github.com/rust-lang/rust/issues/65490
                 }
             }
             RdData::RegObject(register) => {
@@ -160,31 +165,31 @@ impl RunState {
         trace!("WR {:?} := {}", wr, val);
 
         match wr.d() {
-            WrData::Discard => {
-                /* Discard */
+            WrData::Register(Register::Gen(rn)) => {
+                if likely(rn < REG_COUNT as u8) {
+                    self.frame.gen[rn as usize] = val;
                     Ok(())
-            }
-            WrData::Register(Register::Res(rn)) => {
-                if rn >= REG_COUNT as u8 {
-                    Err(Fault::RegisterNotExist { reg: Register::Res(rn) }) // TODO use match after @ when stabilized https://github.com/rust-lang/rust/issues/65490
                 } else {
-                    Err(Fault::RegisterNotWritable { reg: Register::Res(rn) })
+                    Err(Fault::RegisterNotExist { reg: Register::Gen(rn) })
                 }
             }
             WrData::Register(Register::Arg(rn)) => {
-                if rn >= REG_COUNT as u8 {
-                    Err(Fault::RegisterNotExist { reg: Register::Arg(rn) })
-                } else {
+                if likely(rn < REG_COUNT as u8) {
                     Err(Fault::RegisterNotWritable { reg: Register::Res(rn) })
+                } else {
+                    Err(Fault::RegisterNotExist { reg: Register::Arg(rn) })
                 }
             }
-            WrData::Register(Register::Gen(rn)) => {
-                if rn >= REG_COUNT as u8 {
-                    Err(Fault::RegisterNotExist { reg: Register::Gen(rn) })
+            WrData::Register(Register::Res(rn)) => {
+                if likely(rn < REG_COUNT as u8) {
+                    Err(Fault::RegisterNotWritable { reg: Register::Res(rn) })
                 } else {
-                    self.frame.gen[rn as usize] = val;
-                    Ok(())
+                    Err(Fault::RegisterNotExist { reg: Register::Res(rn) }) // TODO use match after @ when stabilized https://github.com/rust-lang/rust/issues/65490
+                }
             }
+            WrData::Discard => {
+                /* Discard */
+                Ok(())
             }
             WrData::RegObject(register) => {
                 let reference = self.read(Rd::new(RdData::Register(register)))?;

crsn_arith/src/exec.rs

@@ -64,13 +64,13 @@ impl OpTrait for ArithOp {
                 state.clear_status();
                 let x = state.read(*a)?;
                 let y = state.read(*b)?;
-                let (res, ov) = if let Some(v) = x.checked_mul(y) {
-                    (v, false)
+                let res = if let Some(v) = x.checked_mul(y) {
+                    v
                 } else {
-                    (x.wrapping_mul(y), true)
+                    state.set_flag(Cond::Overflow, true);
+                    x.wrapping_mul(y)
                 };
                 state.update_status(res);
-                state.set_flag(Cond::Overflow, ov);
                 state.write(*dst, res)?;
             }
             ArithOp::Div { dst, rem, a, div } => {

crsn_stacks/src/exec.rs

@@ -22,7 +22,7 @@ impl OpTrait for StackOp {
         let eres = EvalRes::default();
         match self {
             StackOp::NewStack { dst } => {
-                let id = info.unique_value();
+                let id = info.unique_handle();
                 let stacks: &mut Stacks = state.ext_mut();
                 stacks.store.insert(id, VecDeque::new());
 

launcher/src/main.rs

@@ -2,7 +2,7 @@
 extern crate log;
 
 use std::collections::HashMap;
-use std::sync::Arc;
+
 use std::time::Duration;
 
 use clappconfig::{AppConfig, clap};
@@ -10,7 +10,7 @@ use clappconfig::clap::ArgMatches;
 use serde::{Deserialize, Serialize};
 
 use crsn::asm::data::literal::Addr;
-use crsn::module::OpTrait;
+use crsn::module::{OpTrait, CrsnUniq};
 use crsn::runtime::run_thread::{RunThread, ThreadToken, ThreadParams};
 use crsn_arith::ArithOps;
 use crsn_screen::ScreenOps;
@@ -121,13 +121,13 @@ fn main() -> anyhow::Result<()> {
 
     let source = read_file::read_file(&config.program_file)?;
 
-    let parsers = Arc::new(vec![
+    let uniq = CrsnUniq::new();
+
+    let parsed = crsn::asm::assemble(&source, &uniq, vec![
         ArithOps::new(),
         StackOps::new(),
         ScreenOps::new(),
-    ]);
-
-    let parsed = crsn::asm::assemble(&source, parsers)?;
+    ])?;
 
     if config.asm_only {
         for (n, op) in parsed.ops.iter().enumerate() {
@@ -147,7 +147,7 @@ fn main() -> anyhow::Result<()> {
     let args = &[];
     let thread = RunThread::new(ThreadParams {
         id: ThreadToken(0),
-        uniq: None,
+        uniq,
         program: parsed,
         pc: Addr(0),
         cycle_time: config.cycle_time,