Merge pull request rust-lang#209 from RalfJung/ptrs

Make HashMap insertion work

Author: oli-obk

src/error.rs

@@ -33,6 +33,7 @@ pub enum EvalError<'tcx> {
     ExecuteMemory,
     ArrayIndexOutOfBounds(Span, u64, u64),
     Math(Span, ConstMathErr),
+    Intrinsic(String),
     OverflowingMath,
     InvalidChar(u128),
     OutOfMemory {
@@ -104,6 +105,8 @@ impl<'tcx> Error for EvalError<'tcx> {
                 "array index out of bounds",
             EvalError::Math(..) =>
                 "mathematical operation failed",
+            EvalError::Intrinsic(..) =>
+                "intrinsic failed",
             EvalError::OverflowingMath =>
                 "attempted to do overflowing math",
             EvalError::NoMirFor(..) =>
@@ -168,6 +171,8 @@ impl<'tcx> fmt::Display for EvalError<'tcx> {
                 write!(f, "index out of bounds: the len is {} but the index is {} at {:?}", len, index, span),
             EvalError::Math(span, ref err) =>
                 write!(f, "{:?} at {:?}", err, span),
+            EvalError::Intrinsic(ref err) =>
+                write!(f, "{}", err),
             EvalError::InvalidChar(c) =>
                 write!(f, "tried to interpret an invalid 32-bit value as a char: {}", c),
             EvalError::OutOfMemory { allocation_size, memory_size, memory_usage } =>

src/memory.rs

@@ -546,7 +546,7 @@ impl<'a, 'tcx> Memory<'a, 'tcx> {
                 StaticKind::Immutable => " (immutable)",
                 StaticKind::NotStatic => "",
             };
-            trace!("{}({} bytes){}", msg, alloc.bytes.len(), immutable);
+            trace!("{}({} bytes, alignment {}){}", msg, alloc.bytes.len(), alloc.align, immutable);
 
             if !relocations.is_empty() {
                 msg.clear();

src/operator.rs

@@ -13,7 +13,6 @@ use value::{
     bytes_to_f64,
     f32_to_bytes,
     f64_to_bytes,
-    bytes_to_bool,
 };
 
 impl<'a, 'tcx> EvalContext<'a, 'tcx> {
@@ -190,19 +189,17 @@ impl<'a, 'tcx> EvalContext<'a, 'tcx> {
                     }
                 }
                 // These work if one operand is a pointer, the other an integer
-                Add | Sub
+                Add | BitAnd | Sub
                 if left_kind == right_kind && (left_kind == usize || left_kind == isize)
                 && left.is_ptr() && right.is_bytes() => {
                     // Cast to i128 is fine as we checked the kind to be ptr-sized
-                    let (res, over) = self.ptr_int_arithmetic(bin_op, left.to_ptr()?, right.to_bytes()? as i128, left_kind == isize)?;
-                    return Ok((PrimVal::Ptr(res), over));
+                    return self.ptr_int_arithmetic(bin_op, left.to_ptr()?, right.to_bytes()? as i128, left_kind == isize);
                 }
-                Add
+                Add | BitAnd
                 if left_kind == right_kind && (left_kind == usize || left_kind == isize)
                 && left.is_bytes() && right.is_ptr() => {
                     // This is a commutative operation, just swap the operands
-                    let (res, over) = self.ptr_int_arithmetic(bin_op, right.to_ptr()?, left.to_bytes()? as i128, left_kind == isize)?;
-                    return Ok((PrimVal::Ptr(res), over));
+                    return self.ptr_int_arithmetic(bin_op, right.to_ptr()?, left.to_bytes()? as i128, left_kind == isize);
                 }
                 _ => {}
             }
@@ -287,18 +284,40 @@ impl<'a, 'tcx> EvalContext<'a, 'tcx> {
         left: Pointer,
         right: i128,
         signed: bool,
-    ) -> EvalResult<'tcx, (Pointer, bool)> {
+    ) -> EvalResult<'tcx, (PrimVal, bool)> {
         use rustc::mir::BinOp::*;
 
+        fn map_to_primval((res, over) : (Pointer, bool)) -> (PrimVal, bool) {
+            (PrimVal::Ptr(res), over)
+        }
+
         Ok(match bin_op {
             Sub =>
                 // The only way this can overflow is by underflowing, so signdeness of the right operands does not matter
-                left.overflowing_signed_offset(-right, self.memory.layout),
+                map_to_primval(left.overflowing_signed_offset(-right, self.memory.layout)),
             Add if signed =>
-                left.overflowing_signed_offset(right, self.memory.layout),
+                map_to_primval(left.overflowing_signed_offset(right, self.memory.layout)),
             Add if !signed =>
-                left.overflowing_offset(right as u64, self.memory.layout),
-            _ => bug!("ptr_int_arithmetic called on unsupported operation")
+                map_to_primval(left.overflowing_offset(right as u64, self.memory.layout)),
+
+            BitAnd if !signed => {
+                let base_mask : u64 = !(self.memory.get(left.alloc_id)?.align - 1);
+                let right = right as u64;
+                if right & base_mask == base_mask {
+                    // Case 1: The base address bits are all preserved, i.e., right is all-1 there
+                    (PrimVal::Ptr(Pointer::new(left.alloc_id, left.offset & right)), false)
+                } else if right & base_mask == 0 {
+                    // Case 2: The base address bits are all taken away, i.e., right is all-0 there
+                    (PrimVal::from_u128((left.offset & right) as u128), false)
+                } else {
+                    return Err(EvalError::ReadPointerAsBytes);
+                }
+            }
+
+            _ => {
+                let msg = format!("unimplemented binary op on pointer {:?}: {:?}, {:?} ({})", bin_op, left, right, if signed { "signed" } else { "unsigned" });
+                return Err(EvalError::Unimplemented(msg));
+            }
         })
     }
 }
@@ -314,7 +333,7 @@ pub fn unary_op<'tcx>(
     let bytes = val.to_bytes()?;
 
     let result_bytes = match (un_op, val_kind) {
-        (Not, Bool) => !bytes_to_bool(bytes) as u128,
+        (Not, Bool) => !val.to_bool()? as u128,
 
         (Not, U8)  => !(bytes as u8) as u128,
         (Not, U16) => !(bytes as u16) as u128,

src/terminator/intrinsic.rs

@@ -154,12 +154,21 @@ impl<'a, 'tcx> EvalContext<'a, 'tcx> {
 
             "ctpop" |
             "cttz" |
+            "cttz_nonzero" |
             "ctlz" |
+            "ctlz_nonzero" |
             "bswap" => {
                 let ty = substs.type_at(0);
-                let num = self.value_to_primval(arg_vals[0], ty)?;
+                let num = self.value_to_primval(arg_vals[0], ty)?.to_bytes()?;
                 let kind = self.ty_to_primval_kind(ty)?;
-                let num = numeric_intrinsic(intrinsic_name, num, kind)?;
+                let num = if intrinsic_name.ends_with("_nonzero") {
+                    if num == 0 {
+                        return Err(EvalError::Intrinsic(format!("{} called on 0", intrinsic_name)))
+                    }
+                    numeric_intrinsic(intrinsic_name.trim_right_matches("_nonzero"), num, kind)?
+                } else {
+                    numeric_intrinsic(intrinsic_name, num, kind)?
+                };
                 self.write_primval(dest, num, ty)?;
             }
 
@@ -538,13 +547,11 @@ impl<'a, 'tcx> EvalContext<'a, 'tcx> {
 
 fn numeric_intrinsic<'tcx>(
     name: &str,
-    val: PrimVal,
+    bytes: u128,
     kind: PrimValKind
 ) -> EvalResult<'tcx, PrimVal> {
     macro_rules! integer_intrinsic {
         ($method:ident) => ({
-            let bytes = val.to_bytes()?;
-
             use value::PrimValKind::*;
             let result_bytes = match kind {
                 I8 => (bytes as i8).$method() as u128,
@@ -557,7 +564,7 @@ fn numeric_intrinsic<'tcx>(
                 U64 => (bytes as u64).$method() as u128,
                 I128 => (bytes as i128).$method() as u128,
                 U128 => bytes.$method() as u128,
-                _ => bug!("invalid `{}` argument: {:?}", name, val),
+                _ => bug!("invalid `{}` argument: {:?}", name, bytes),
             };
 
             PrimVal::Bytes(result_bytes)

src/value.rs

@@ -23,12 +23,6 @@ pub(super) fn f64_to_bytes(f: f64) -> u128 {
     unsafe { transmute::<f64, u64>(f) as u128 }
 }
 
-pub(super) fn bytes_to_bool(n: u128) -> bool {
-    // FIXME(solson): Can we reach here due to user error?
-    debug_assert!(n == 0 || n == 1, "bytes interpreted as bool were {}", n);
-    n & 1 == 1
-}
-
 /// A `Value` represents a single self-contained Rust value.
 ///
 /// A `Value` can either refer to a block of memory inside an allocation (`ByRef`) or to a primitve

tests/compile-fail/bitop-beyond-alignment.rs

@@ -0,0 +1,37 @@
+// Copyright 2012-2014 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+#![allow(dead_code)]
+
+use std::mem;
+
+enum Tag<A> {
+    Tag2(A)
+}
+
+struct Rec {
+    c8: u8,
+    t: Tag<u64>
+}
+
+fn mk_rec() -> Rec {
+    return Rec { c8:0, t:Tag::Tag2(0) };
+}
+
+fn is_u64_aligned(u: &Tag<u64>) -> bool {
+    let p: usize = unsafe { mem::transmute(u) };
+    let u64_align = std::mem::align_of::<u64>();
+    return (p & (u64_align + 1)) == 0; //~ ERROR a raw memory access tried to access part of a pointer value as raw bytes
+}
+
+pub fn main() {
+    let x = mk_rec();
+    assert!(is_u64_aligned(&x.t));
+}

tests/compile-fail/ctlz_nonzero.rs

@@ -0,0 +1,15 @@
+#![feature(intrinsics)]
+
+mod rusti {
+    extern "rust-intrinsic" {
+        pub fn ctlz_nonzero<T>(x: T) -> T;
+    }
+}
+
+pub fn main() {
+    unsafe {
+        use rusti::*;
+
+        ctlz_nonzero(0u8); //~ ERROR: ctlz_nonzero called on 0
+    }
+}

tests/compile-fail/cttz_nonzero.rs

@@ -0,0 +1,15 @@
+#![feature(intrinsics)]
+
+mod rusti {
+    extern "rust-intrinsic" {
+        pub fn cttz_nonzero<T>(x: T) -> T;
+    }
+}
+
+pub fn main() {
+    unsafe {
+        use rusti::*;
+
+        cttz_nonzero(0u8); //~ ERROR: cttz_nonzero called on 0
+    }
+}

tests/run-pass-fullmir/hashmap.rs

@@ -2,14 +2,24 @@ use std::collections::{self, HashMap};
 use std::hash::BuildHasherDefault;
 
 fn main() {
-    let map : HashMap<String, i32, BuildHasherDefault<collections::hash_map::DefaultHasher>> = Default::default();
+    let mut map : HashMap<i32, i32, BuildHasherDefault<collections::hash_map::DefaultHasher>> = Default::default();
+    map.insert(0, 0);
     assert_eq!(map.values().fold(0, |x, y| x+y), 0);
 
-    // TODO: This performs bit operations on the least significant bit of a pointer
-//     for i in 0..33 {
-//         map.insert(format!("key_{}", i), i);
-//         assert_eq!(map.values().fold(0, |x, y| x+y), i*(i+1)/2);
-//     }
+    let table_base = map.get(&0).unwrap() as *const _;
+
+    let num = 22; // large enough to trigger a resize
+    for i in 1..num {
+        map.insert(i, i);
+    }
+    assert!(table_base != map.get(&0).unwrap() as *const _); // make sure relocation happened
+    assert_eq!(map.values().fold(0, |x, y| x+y), num*(num-1)/2); // check the right things are in the table now
+
+    // Inserting again replaces the existing entries
+    for i in 0..num {
+        map.insert(i, num-1-i);
+    }
+    assert_eq!(map.values().fold(0, |x, y| x+y), num*(num-1)/2);
 
     // TODO: Test Entry API
 }

tests/run-pass/intrinsics-integer.rs

@@ -14,7 +14,9 @@ mod rusti {
     extern "rust-intrinsic" {
         pub fn ctpop<T>(x: T) -> T;
         pub fn ctlz<T>(x: T) -> T;
+        pub fn ctlz_nonzero<T>(x: T) -> T;
         pub fn cttz<T>(x: T) -> T;
+        pub fn cttz_nonzero<T>(x: T) -> T;
         pub fn bswap<T>(x: T) -> T;
     }
 }
@@ -68,6 +70,21 @@ pub fn main() {
         assert_eq!(ctlz(100u32), 25); assert_eq!(ctlz(100i32), 25);
         assert_eq!(ctlz(100u64), 57); assert_eq!(ctlz(100i64), 57);
 
+        assert_eq!(ctlz_nonzero(1u8), 7); assert_eq!(ctlz_nonzero(1i8), 7);
+        assert_eq!(ctlz_nonzero(1u16), 15); assert_eq!(ctlz_nonzero(1i16), 15);
+        assert_eq!(ctlz_nonzero(1u32), 31); assert_eq!(ctlz_nonzero(1i32), 31);
+        assert_eq!(ctlz_nonzero(1u64), 63); assert_eq!(ctlz_nonzero(1i64), 63);
+
+        assert_eq!(ctlz_nonzero(10u8), 4); assert_eq!(ctlz_nonzero(10i8), 4);
+        assert_eq!(ctlz_nonzero(10u16), 12); assert_eq!(ctlz_nonzero(10i16), 12);
+        assert_eq!(ctlz_nonzero(10u32), 28); assert_eq!(ctlz_nonzero(10i32), 28);
+        assert_eq!(ctlz_nonzero(10u64), 60); assert_eq!(ctlz_nonzero(10i64), 60);
+
+        assert_eq!(ctlz_nonzero(100u8), 1); assert_eq!(ctlz_nonzero(100i8), 1);
+        assert_eq!(ctlz_nonzero(100u16), 9); assert_eq!(ctlz_nonzero(100i16), 9);
+        assert_eq!(ctlz_nonzero(100u32), 25); assert_eq!(ctlz_nonzero(100i32), 25);
+        assert_eq!(ctlz_nonzero(100u64), 57); assert_eq!(ctlz_nonzero(100i64), 57);
+
         assert_eq!(cttz(-1i8 as u8), 0); assert_eq!(cttz(-1i8), 0);
         assert_eq!(cttz(-1i16 as u16), 0); assert_eq!(cttz(-1i16), 0);
         assert_eq!(cttz(-1i32 as u32), 0); assert_eq!(cttz(-1i32), 0);
@@ -93,6 +110,26 @@ pub fn main() {
         assert_eq!(cttz(100u32), 2); assert_eq!(cttz(100i32), 2);
         assert_eq!(cttz(100u64), 2); assert_eq!(cttz(100i64), 2);
 
+        assert_eq!(cttz_nonzero(-1i8 as u8), 0); assert_eq!(cttz_nonzero(-1i8), 0);
+        assert_eq!(cttz_nonzero(-1i16 as u16), 0); assert_eq!(cttz_nonzero(-1i16), 0);
+        assert_eq!(cttz_nonzero(-1i32 as u32), 0); assert_eq!(cttz_nonzero(-1i32), 0);
+        assert_eq!(cttz_nonzero(-1i64 as u64), 0); assert_eq!(cttz_nonzero(-1i64), 0);
+
+        assert_eq!(cttz_nonzero(1u8), 0); assert_eq!(cttz_nonzero(1i8), 0);
+        assert_eq!(cttz_nonzero(1u16), 0); assert_eq!(cttz_nonzero(1i16), 0);
+        assert_eq!(cttz_nonzero(1u32), 0); assert_eq!(cttz_nonzero(1i32), 0);
+        assert_eq!(cttz_nonzero(1u64), 0); assert_eq!(cttz_nonzero(1i64), 0);
+
+        assert_eq!(cttz_nonzero(10u8), 1); assert_eq!(cttz_nonzero(10i8), 1);
+        assert_eq!(cttz_nonzero(10u16), 1); assert_eq!(cttz_nonzero(10i16), 1);
+        assert_eq!(cttz_nonzero(10u32), 1); assert_eq!(cttz_nonzero(10i32), 1);
+        assert_eq!(cttz_nonzero(10u64), 1); assert_eq!(cttz_nonzero(10i64), 1);
+
+        assert_eq!(cttz_nonzero(100u8), 2); assert_eq!(cttz_nonzero(100i8), 2);
+        assert_eq!(cttz_nonzero(100u16), 2); assert_eq!(cttz_nonzero(100i16), 2);
+        assert_eq!(cttz_nonzero(100u32), 2); assert_eq!(cttz_nonzero(100i32), 2);
+        assert_eq!(cttz_nonzero(100u64), 2); assert_eq!(cttz_nonzero(100i64), 2);
+
         assert_eq!(bswap(0x0Au8), 0x0A); // no-op
         assert_eq!(bswap(0x0Ai8), 0x0A); // no-op
         assert_eq!(bswap(0x0A0Bu16), 0x0B0A);

tests/compile-fail/tag-align-dyn-u64.rs renamed to tests/run-pass/tag-align-dyn-u64.rs

@@ -28,7 +28,7 @@ fn mk_rec() -> Rec {
 fn is_u64_aligned(u: &Tag<u64>) -> bool {
     let p: usize = unsafe { mem::transmute(u) };
     let u64_align = std::mem::align_of::<u64>();
-    return (p & (u64_align - 1)) == 0; //~ ERROR a raw memory access tried to access part of a pointer value as raw bytes
+    return (p & (u64_align - 1)) == 0;
 }
 
 pub fn main() {