Generate a constant hash table for recognizing opcodes.

Use a simple quadratically probed, open addressed hash table. We could use a
parfect hash function, but it would take longer to compute in Python, and this
is not in the critical path performancewise.

meta/constant_hash.py

@@ -0,0 +1,76 @@
+"""
+Generate constant hash tables.
+
+The `constant_hash` module can generate constant pre-populated hash tables. We
+don't attempt parfect hashing, but simply generate an open addressed
+quadratically probed hash table.
+"""
+
+def simple_hash(s):
+    """
+    Compute a primitive hash of a string.
+
+    Example:
+        >>> hex(simple_hash("Hello"))
+        '0x2fa70c01'
+        >>> hex(simple_hash("world"))
+        '0x5b0c31d5'
+    """
+    h = 5381
+    for c in s:
+        h = ((h ^ ord(c)) + ((h >> 6) + (h << 26))) & 0xffffffff
+    return h
+
+def next_power_of_two(x):
+    """
+    Compute the next power of two that is greater than `x`:
+        >>> next_power_of_two(0)
+        1
+        >>> next_power_of_two(1)
+        2
+        >>> next_power_of_two(2)
+        4
+        >>> next_power_of_two(3)
+        4
+        >>> next_power_of_two(4)
+        8
+    """
+    s = 1
+    while x & (x + 1) != 0:
+        x |= x >> s
+        s *= 2
+    return x + 1
+
+def compute_quadratic(items, hash_function):
+    """
+    Compute an open addressed, quadratically probed hash table containing
+    `items`. The returned table is a list containing the elements of the
+    iterable `items` and `None` in unused slots.
+
+    :param items: Iterable set of items to place in hash table.
+    :param hash_function: Hash function which takes an item and returns a
+            number.
+
+    Simple example (see hash values above, they collide on slot 1):
+        >>> compute_quadratic(['Hello', 'world'], simple_hash)
+        [None, 'Hello', 'world', None]
+    """
+
+    items = list(items)
+    # Table size must be a power of two. Aim for >20% unused slots.
+    size = next_power_of_two(int(1.20*len(items)))
+    table = [None] * size
+
+    for i in items:
+        h = hash_function(i) % size
+        s = 0
+        while table[h] is not None:
+            s += 1
+            h = (h + s) % size
+        table[h] = i
+
+    return table
+
+if __name__ == "__main__":
+    import doctest
+    doctest.testmod()

meta/gen_instr.py

@@ -3,6 +3,7 @@ Generate sources with instruction info.
 """
 
 import srcgen
+import constant_hash
 
 def collect_instr_groups(targets):
     seen = set()
@@ -24,6 +25,7 @@ def gen_opcodes(groups, out_dir):
     fmt.line('#[derive(Copy, Clone, PartialEq, Eq, Debug)]')
     instrs = []
     with fmt.indented('pub enum Opcode {', '}'):
+        fmt.line('NotAnOpcode,')
         for g in groups:
             for i in g.instructions:
                 instrs.append(i)
@@ -39,9 +41,20 @@ def gen_opcodes(groups, out_dir):
     # Generate a private opcode_name function.
     with fmt.indented('fn opcode_name(opc: Opcode) -> &\'static str {', '}'):
         with fmt.indented('match opc {', '}'):
+            fmt.line('Opcode::NotAnOpcode => "<not an opcode>",')
             for i in instrs:
                 fmt.format('Opcode::{} => "{}",', i.camel_name, i.name)
 
+    # Generate an opcode hash table for looking up opcodes by name.
+    hash_table = constant_hash.compute_quadratic(instrs,
+            lambda i: constant_hash.simple_hash(i.name))
+    with fmt.indented('const OPCODE_HASH_TABLE: [Opcode; {}] = ['.format(len(hash_table)), '];'):
+        for i in hash_table:
+            if i is None:
+                fmt.line('Opcode::NotAnOpcode,')
+            else:
+                fmt.format('Opcode::{},', i.camel_name)
+
     fmt.update_file('opcodes.rs', out_dir)
 
 def generate(targets, out_dir):

src/libcretonne/immediates.rs

@@ -8,8 +8,12 @@
 use std::fmt::{self, Display, Formatter};
 use std::mem;
 
-// The `Opcode` enum and the `opcode_name` function are generated from the meta instruction
-// descriptions.
+// Include code generated by `meta/gen_instr.py`. This file contains:
+//
+// - The `pub enum Opcode` definition with all known opcodes,
+// - The private `fn opcode_name(Opcode) -> &'static str` function, and
+// - The hash table `const OPCODE_HASH_TABLE: [Opcode; N]`.
+//
 include!(concat!(env!("OUT_DIR"), "/opcodes.rs"));
 
 impl Display for Opcode {
@@ -18,6 +22,46 @@ impl Display for Opcode {
     }
 }
 
+// A primitive hash function for matching opcodes.
+// Must match `meta/constant_hash.py`.
+fn simple_hash(s: &str) -> u32 {
+    let mut h: u32 = 5381;
+    for c in s.chars() {
+        h = (h ^ c as u32).wrapping_add(h.rotate_right(6));
+    }
+    h
+}
+
+impl Opcode {
+    /// Parse an Opcode name from a string.
+    pub fn from_str(s: &str) -> Option<Opcode> {
+        let tlen = OPCODE_HASH_TABLE.len();
+        assert!(tlen.is_power_of_two());
+        let mut idx = simple_hash(s) as usize;
+        let mut step: usize = 0;
+        loop {
+            idx = idx % tlen;
+            let entry = OPCODE_HASH_TABLE[idx];
+
+            if entry == Opcode::NotAnOpcode {
+                return None;
+            }
+
+            if *opcode_name(entry) == *s {
+                return Some(entry);
+            }
+
+            // Quadratic probing.
+            step += 1;
+            // When `tlen` is a power of two, it can be proven that idx will visit all entries.
+            // This means that this loop will always terminate if the hash table has even one
+            // unused entry.
+            assert!(step < tlen);
+            idx += step;
+        }
+    }
+}
+
 /// 64-bit immediate integer operand.
 ///
 #[derive(Copy, Clone, PartialEq, Eq, Debug)]
@@ -178,6 +222,13 @@ mod tests {
 
         assert_eq!(format!("{:?}", Opcode::IaddImm), "IaddImm");
         assert_eq!(format!("{}", Opcode::IaddImm), "iadd_imm");
+
+        // Check the matcher.
+        assert_eq!(Opcode::from_str("iadd"), Some(Opcode::Iadd));
+        assert_eq!(Opcode::from_str("iadd_imm"), Some(Opcode::IaddImm));
+        assert_eq!(Opcode::from_str("iadd\0"), None);
+        assert_eq!(Opcode::from_str(""), None);
+        assert_eq!(Opcode::from_str("\0"), None);
     }
 
     #[test]