diff --git a/README.md b/README.md
index 6c2ebb7..fc2dcc7 100644
--- a/README.md
+++ b/README.md
@@ -188,7 +188,11 @@ Objects (`@reg`, `@sym`) can be read or written as if they were a register, but
 Other objects may produce a runtime fault or set the INVALID flag.
 
 In the instruction lists below, I will use the symbols `Rd` for reads, `Wr` for writes, `RW` for read-writes, and `@Obj` for object handles,
-with optional description after a colon, such as: `(add Wr:dst Rd:a Rd:b)`.
+with optional description after an apostrophe, such as: `(add Wr'dst Rd'a Rd'b)`. 
+
+Some instructions use bit offsets and widths. Width is directly attached to the opcode (e.g. `(ld16 …)`); 
+offsets are attached at the respective arguments after a colon: `(ld16 r0:8 r1:32)`  -
+load 16 bits, starting at offset 32 of `r1`, into `r0`, starting at offset 8. The rest if the register is not affected.
 
 ### Conditional branches
 
@@ -314,29 +318,13 @@ Jumping to a label is always safer than a manual skip.
 ; Copy a value
 (ld Wr Rd)
 
-; Copy lower XX bits (the rest is untouched)
-(ldXX RW Rd)
-
-; Copy XX bits to bit YY and up
-; Example: (ld32/32 r1 r0) ; copy lower 32 bits of r0 to upper 32 bits of r1
-; 
-; Rd: ##############<-- X -->
-;                      |<-Y->
-; Wr: ########<-- xY -->######
-(ldXX/YY RW Rd)
-
-; Copy XX bits at ZZ to bit YY and up
-; Example: (ld32/32/16 r1 r0) ; copy bits 16-47 (32 bits) of r0 to bits 32-63
-; 
-;                   |<--Z--->
-; Rd: #####<-- X -->#########
-;                      |<-Y->
-; Wr: ########<-- X -->######
-(ldXX/YY/ZZ RW Rd)
-
+; Copy lower XX bits (the rest is untouched).
+; Offsets can be specified to work with arbitrary bit slices
+(ldXX RW:dst_offset Rd:src_offset)
+(ldXX Wr Rd:dst_offset Rd:src_offset)
 
 ; Copy a value N times. This is useful when used with stream handles or buffers.
-(ldn Wr Rd Rd:count)
+(ldn Wr Rd'src Rd'count)
 
 ; Write a sequence of values, or all codepoints from a string, into the destination.
 ; This is most useful with object handles, such as a buffer or @cout.
@@ -351,10 +339,9 @@ Jumping to a label is always safer than a manual skip.
 ; Exchange two register's values
 (xch RW RW)
 
-; Exchange bits in two registers (see ldXX/YY/ZZ for reference and diagrams)
-(xchXX RW RW)
-(xchXX/YY RW RW)
-(xchXX/YY/ZZ RW RW)
+; Exchange XX bits in two registers
+; Offsets can be specified to work with arbitrary bit slices
+(xchXX RW:offset RW:offset)
 
 ; Store status flags to a register
 (stf Wr)
@@ -407,12 +394,12 @@ Many instructions have two forms:
 
 ; Check if a value is in a range (inclusive).
 ; Sets the EQ, LT and GT flags. Also sets Z, POS and NEG based on the value.
-(rcmp Rd:val Rd:start Rd:end)
+(rcmp Rd'val Rd'start Rd'end)
 
 ; Get a random number
 (rng Wr) ; the value will fill all 64 bits of the target
-(rng Wr Rd:max) ; 0 to max, max is inclusive
-(rng Wr Rd:min Rd:max) ; min to max, both are inclusive
+(rng Wr Rd'max) ; 0 to max, max is inclusive
+(rng Wr Rd'min Rd'max) ; min to max, both are inclusive
 
 ; Add A+B
 (add Wr Rd Rd)
@@ -427,19 +414,19 @@ Many instructions have two forms:
 (mul RW Rd)
 
 ; Divide A/B
-(div Wr Rd Rd:divider)
-(div RW Rd:divider)
+(div Wr Rd Rd'divider)
+(div RW Rd'divider)
 
 ; Divide and get remainder
 ; Both DST and REM are output registers
-(divr Wr:result Wr:remainder Rd Rd:divider)
-(divr RW Wr:remainder Rd:divider)
+(divr Wr'result Wr'remainder Rd Rd'divider)
+(divr RW Wr'remainder Rd'divider)
 
 ; Get remainder A%B
 ; This is equivalent to (divr _ REM A B),
 ; except status flags are updated by the remainder value
-(mod Wr Rd Rd:divider)
-(mod RW Rd:divider)
+(mod Wr Rd Rd'divider)
+(mod RW Rd'divider)
 
 ; Swap the 32-bit halves of a value
 ; 0x01234567_89abcdef -> 0x89abcdef_01234567
@@ -468,21 +455,17 @@ Many instructions have two forms:
 (clz Wr Rd)
 (clz RW)
 ; Count leading zeros in the lower XX bits
-(clzXX Wr Rd)
-(clzXX RW)
-; Count leading zeros in the XX bits starting at YY (e.g. 16/32 is the lower 16 bits of the higher 32 bits)
-(clzXX/YY Wr Rd)
-(clzXX/YY RW)
+; Offsets can be specified to work with arbitrary bit slices
+(clzXX Wr Rd:src_offs)
+(clzXX RW:src_offs)
 
 ; Count leading ones
 (clo Wr Rd)
 (clo RW)
 ; Count leading ones in the lower XX bits
-(cloXX Wr Rd)
-(cloXX RW)
-; Count leading ones in the XX bits starting at YY
-(cloXX/YY Wr Rd)
-(cloXX/YY RW)
+; Offsets can be specified to work with arbitrary bit slices
+(cloXX Wr Rd:src_offs)
+(cloXX RW:src_offs)
 
 ; Sign extend a XX-bit value to 64 bits, XX in range 1..63)
 (seXX Wr Rd)
@@ -509,24 +492,24 @@ Many instructions have two forms:
 (ror RW Rd)
 
 ; Rotate left (wrap around)
-(rol Wr Rd:value Rd:count)
-(rol RW Rd:count)
+(rol Wr Rd'value Rd'count)
+(rol RW Rd'count)
 
 ; Logical shift right (fill with zeros)
-(lsr Wr Rd Rd:count)
-(lsr RW Rd:count)
+(lsr Wr Rd Rd'count)
+(lsr RW Rd'count)
 
 ; Logical shift left (fill with zeros)
-(lsl Wr Rd Rd:count)
-(lsl RW Rd:count)
+(lsl Wr Rd Rd'count)
+(lsl RW Rd'count)
 
 ; Arithmetic shift right (copy sign bit)
-(asr Wr Rd Rd:count)
-(asr RW Rd:count)
+(asr Wr Rd Rd'count)
+(asr RW Rd'count)
 
 ; Arithmetic shift left (this is identical to `lsl`, added for completeness)
-(asl Wr Rd Rd:count)
-(asl RW Rd:count)
+(asl Wr Rd Rd'count)
+(asl RW Rd'count)
 
 ; Delete an object by its handle. Objects are used by some extensions.
 (del @Rd)
diff --git a/crsn/crsn-sexp/src/lib.rs b/crsn/crsn-sexp/src/lib.rs
index 5a360ae..9eaf113 100644
--- a/crsn/crsn-sexp/src/lib.rs
+++ b/crsn/crsn-sexp/src/lib.rs
@@ -95,7 +95,7 @@ fn without_underscores(s: &str) -> Cow<str> {
     }
 }
 
-fn atom_of_string(s: String) -> Atom {
+pub fn atom_of_string(s: String) -> Atom {
     if s.starts_with('#') {
         match u64::from_str_radix(&without_underscores(&s[1..]), 16) {
             Ok(u) => return Atom::U(u),
diff --git a/crsn/src/asm/data/mod.rs b/crsn/src/asm/data/mod.rs
index 4bda70e..0e2677e 100644
--- a/crsn/src/asm/data/mod.rs
+++ b/crsn/src/asm/data/mod.rs
@@ -78,6 +78,52 @@ impl From<WrData> for DataDisp {
     }
 }
 
+impl From<Rd> for DataDisp {
+    fn from(s: Rd) -> Self {
+        s.0.into()
+    }
+}
+
+impl From<Wr> for DataDisp {
+    fn from(s: Wr) -> Self {
+        s.0.into()
+    }
+}
+
+impl From<RdWr> for DataDisp {
+    fn from(s: RdWr) -> Self {
+        s.0.into()
+    }
+}
+
+impl From<&Rd> for DataDisp {
+    fn from(s: &Rd) -> Self {
+        s.0.into()
+    }
+}
+
+impl From<&Wr> for DataDisp {
+    fn from(s: &Wr) -> Self {
+        s.0.into()
+    }
+}
+
+impl From<&RdWr> for DataDisp {
+    fn from(s: &RdWr) -> Self {
+        s.0.into()
+    }
+}
+
+pub trait DataDispEquals : Into<DataDisp> {
+    fn disp_equals<T : Into<DataDisp>>(self, other : T) -> bool;
+}
+
+impl<T : Into<DataDisp>> DataDispEquals for T {
+    fn disp_equals<X : Into<DataDisp>>(self, other : X) -> bool {
+        self.into() == other.into()
+    }
+}
+
 /// Data source disposition
 #[derive(Debug, Clone, Copy, Eq, PartialEq)]
 pub enum RdData {
diff --git a/crsn/src/asm/instr/op.rs b/crsn/src/asm/instr/op.rs
index 9a96bc2..61fdf92 100644
--- a/crsn/src/asm/instr/op.rs
+++ b/crsn/src/asm/instr/op.rs
@@ -18,9 +18,9 @@ pub enum OpKind {
 
 #[derive(Debug)]
 pub struct Op {
+    pub kind: OpKind,
     pub cond: Option<Cond>,
     pub pos: SourcePosition,
-    pub kind: OpKind,
 }
 
 impl OpTrait for Op {
diff --git a/crsn/src/asm/parse/arg_parser.rs b/crsn/src/asm/parse/arg_parser.rs
index 0efeca4..fbb6192 100644
--- a/crsn/src/asm/parse/arg_parser.rs
+++ b/crsn/src/asm/parse/arg_parser.rs
@@ -2,10 +2,12 @@ use sexp::{Sexp, SourcePosition};
 
 use crate::asm::data::{Rd, RdData, RdObj, Wr, RdWr};
 use crate::asm::error::CrsnError;
-use crate::asm::parse::parse_data::{parse_rd, parse_wr, parse_value};
+use crate::asm::parse::parse_data::{parse_rd, parse_wr, parse_value, parse_rdwr_offset, parse_rdwr, parse_rd_offset, parse_wr_offset};
 use crate::asm::parse::ParserContext;
 use crate::asm::parse::sexp_expect::expect_string_atom;
 use crate::asm::data::literal::Value;
+use crate::builtin::defs::BitMask;
+use crate::asm::patches::ErrWithPos;
 
 /// Utility for argument parsing
 #[derive(Debug)]
@@ -21,7 +23,7 @@ impl<'a> IntoIterator for TokenParser<'a> {
     type IntoIter = std::vec::IntoIter<Sexp>;
 
     fn into_iter(mut self) -> Self::IntoIter {
-        // The vec is reversed so we can call "pop", but here we will iterate it as slice so it
+        // The vec is reversed so we can call "pop", but here we will iterate it as a slice, so it
         // must be reversed to its original direction.
         self.args.reverse();
         self.args.into_iter()
@@ -40,6 +42,16 @@ impl<'a> TokenParser<'a> {
         }
     }
 
+    /// Get error if not empty.
+    /// The argument is substituted into the phrase "Instruction needs ...!" - i.e. "one Wr argument and a list or string"
+    pub fn ensure_empty(&self, what_arguments : &str) -> Result<(), CrsnError> {
+        if self.have_more() {
+            Err(CrsnError::Parse(format!("Instruction needs {}!", what_arguments).into(), self.start_pos.clone()))
+        } else {
+            Ok(())
+        }
+    }
+
     pub fn have_more(&self) -> bool {
         self.len() > 0
     }
@@ -121,13 +133,180 @@ impl<'a> TokenParser<'a> {
     /// Get the next entry as write location
     pub fn next_rdwr(&mut self) -> Result<RdWr, CrsnError> {
         let next = self.next_or_err()?;
-        let pos = next.pos().clone();
-        let wr = parse_wr(next, self.pcx)?;
+        parse_rdwr(next, self.pcx)
+    }
 
-        if !wr.is_readable() {
-            return Err(CrsnError::Parse("Argument is not readable!".into(), pos));
+    /// Get the next entry as write location with optional :offset
+    pub fn next_rdwr_offset(&mut self) -> Result<(RdWr, u32), CrsnError> {
+        let next = self.next_or_err()?;
+        parse_rdwr_offset(next, self.pcx)
+    }
+
+    pub fn next_rd_offset(&mut self) -> Result<(Rd, u32), CrsnError> {
+        let next = self.next_or_err()?;
+        parse_rd_offset(next, self.pcx)
+    }
+
+    pub fn next_wr_offset(&mut self) -> Result<(Wr, u32), CrsnError> {
+        let next = self.next_or_err()?;
+        parse_wr_offset(next, self.pcx)
+    }
+
+    /// Parse combining binary instruction operands (i.e. add) with bit masks
+    /// Accepts (Wr, Rd, Rd) and (RdWr, Rd)
+    pub fn parse_masked_wr_rd_rd(&mut self, keyword: &str, prefix: &str) -> Result<Option<(Wr, Rd, Rd, BitMask)>, CrsnError> {
+        if let Some(s) = keyword.strip_prefix(prefix) {
+            let width = if s.is_empty() {
+                (std::mem::size_of::<Value>() as u32) * 8
+            } else {
+                s.parse().err_pos(self.start_pos)?
+            };
+            if self.len() == 2 {
+                let (rw, dst_pos) = self.next_rdwr_offset()?;
+                let (rd, src_pos) = self.next_rd_offset()?;
+
+                let mask = BitMask {
+                    width,
+                    dst_pos,
+                    src_pos: dst_pos,
+                    src2_pos: src_pos,
+                };
+                mask.validate(self.start_pos)?;
+
+                Ok(Some((rw.wr(), rw.rd(), rd, mask)))
+            } else if self.len() == 3 {
+                let (wr, dst_pos) = self.next_wr_offset()?;
+                let (rd1, src_pos) = self.next_rd_offset()?;
+                let (rd2, src2_pos) = self.next_rd_offset()?;
+
+                let mask = BitMask {
+                    width,
+                    dst_pos,
+                    src_pos,
+                    src2_pos,
+                };
+                mask.validate(self.start_pos)?;
+
+                Ok(Some((wr, rd1, rd2, mask)))
+            } else {
+                Err(CrsnError::Parse("Instruction needs 2 (RW Rd) or 3 (Wr Rd Rd) arguments!".into(), self.start_pos.clone()))
+            }
+        } else {
+            Ok(None)
         }
+    }
+
+
+    /// Parse unary instruction operands (i.e. complement) with bit masks.
+    /// Accepts (Wr, Rd) and (RdWr)
+    pub fn parse_masked_wr_rd(&mut self, keyword: &str, prefix: &'static str) -> Result<Option<(Wr, Rd, BitMask)>, CrsnError> {
+        if let Some(s) = keyword.strip_prefix(prefix) {
+            let width = if s.is_empty() {
+                (std::mem::size_of::<Value>() as u32) * 8
+            } else {
+                s.parse().err_pos(self.start_pos)?
+            };
+            if self.len() == 1 {
+                let (rw, dst_pos) = self.next_rdwr_offset()?;
 
-        Ok(RdWr::new(wr.0))
+                let mask = BitMask {
+                    width,
+                    dst_pos,
+                    src_pos: dst_pos,
+                    src2_pos: 0,
+                };
+                mask.validate(self.start_pos)?;
+
+                Ok(Some((rw.wr(), rw.rd(), mask)))
+            } else if self.len() == 2 {
+                let (wr, dst_pos) = self.next_wr_offset()?;
+                let (rd, src_pos) = self.next_rd_offset()?;
+
+                let mask = BitMask {
+                    width,
+                    dst_pos,
+                    src_pos,
+                    src2_pos: 0,
+                };
+                mask.validate(self.start_pos)?;
+
+                Ok(Some((wr, rd, mask)))
+            } else {
+                Err(CrsnError::Parse("Instruction needs 1 (RW) or 2 (Wr Rd) arguments!".into(), self.start_pos.clone()))
+            }
+        } else {
+            Ok(None)
+        }
+    }
+
+    /// Parse a swap-type binary instruction operands (i.e. exchange) with bit masks
+    /// Accepts (RdWr, RdWr)
+    pub fn parse_masked_rdwr_rdwr(&mut self, keyword: &str, prefix: &str) -> Result<Option<(RdWr, RdWr, BitMask)>, CrsnError> {
+        if let Some(s) = keyword.strip_prefix(prefix) {
+            let width = if s.is_empty() {
+                (std::mem::size_of::<Value>() as u32) * 8
+            } else {
+                s.parse().err_pos(self.start_pos)?
+            };
+            if self.len() == 2 {
+                let (wr1, dst_pos) = self.next_rdwr_offset()?;
+                let (wr2, src_pos) = self.next_rdwr_offset()?;
+
+                let mask = BitMask {
+                    width,
+                    dst_pos,
+                    src_pos,
+                    src2_pos: 0,
+                };
+                mask.validate(self.start_pos)?;
+
+                Ok(Some((wr1, wr2, mask)))
+            } else {
+                Err(CrsnError::Parse("Instruction needs 2 (RW RW) arguments!".into(), self.start_pos.clone()))
+            }
+        } else {
+            Ok(None)
+        }
+    }
+
+    /// Parse combining binary instruction operands (i.e. add) without masks
+    pub fn parse_wr_rd_rd(&mut self) -> Result<(Wr, Rd, Rd), CrsnError> {
+        if self.len() == 2 {
+            let rw = self.next_rdwr()?;
+            let rd = self.next_rd()?;
+            Ok((rw.wr(), rw.rd(), rd))
+        } else if self.len() == 3 {
+            let wr = self.next_wr()?;
+            let rd1 = self.next_rd()?;
+            let rd2 = self.next_rd()?;
+            Ok((wr, rd1, rd2))
+        } else {
+            Err(CrsnError::Parse("Instruction needs 2 (RW Rd) or 3 (Wr Rd Rd) arguments!".into(), self.start_pos.clone()))
+        }
+    }
+
+    /// Parse unary instruction operands (i.e. complement) without masks
+    pub fn parse_wr_rd(&mut self) -> Result<(Wr, Rd), CrsnError> {
+        if self.len() == 1 {
+            let rw = self.next_rdwr()?;
+            Ok((rw.wr(), rw.rd()))
+        } else if self.len() == 2 {
+            let wr = self.next_wr()?;
+            let rd = self.next_rd()?;
+            Ok((wr, rd))
+        } else {
+            Err(CrsnError::Parse("Instruction needs 1 (RW) or 2 (Wr Rd) arguments!".into(), self.start_pos.clone()))
+        }
+    }
+
+    /// Parse a swap-type binary instruction operands (i.e. exchange) without masks
+    pub fn parse_rdwr_rdwr(&mut self, _keyword: &str, _prefix: &str, _pos: &SourcePosition) -> Result<(RdWr, RdWr), CrsnError> {
+        if self.len() == 2 {
+            let rw1 = self.next_rdwr()?;
+            let rw2 = self.next_rdwr()?;
+            Ok((rw1, rw2))
+        } else {
+            Err(CrsnError::Parse("Instruction needs 2 (RW RW) arguments!".into(), self.start_pos.clone()))
+        }
     }
 }
diff --git a/crsn/src/asm/parse/parse_data.rs b/crsn/src/asm/parse/parse_data.rs
index b522f39..27e5152 100644
--- a/crsn/src/asm/parse/parse_data.rs
+++ b/crsn/src/asm/parse/parse_data.rs
@@ -3,7 +3,7 @@ use std::convert::TryFrom;
 
 use sexp::{Atom, Sexp, SourcePosition};
 
-use crate::asm::data::{DataDisp, Rd, RdData, reg, Wr, WrData};
+use crate::asm::data::{DataDisp, Rd, RdData, reg, Wr, WrData, RdWr};
 use crate::asm::data::literal::{ConstantName, Label, RegisterAlias, Value};
 use crate::asm::error::CrsnError;
 use crate::asm::parse::ParserContext;
@@ -71,9 +71,6 @@ pub fn parse_label_str(name: &str, pos: &SourcePosition) -> Result<Label, CrsnEr
 /// Parse data disposition (address/value, without the read/write restriction)
 pub fn parse_data_disp(tok: Sexp, pcx: &ParserContext) -> Result<DataDisp, CrsnError> {
     // trace!("parse data: {:?}", tok);
-
-    // TODO implement masks
-
     match tok {
         Sexp::Atom(Atom::I(val), _pos) => {
             Ok(DataDisp::Immediate(unsafe { std::mem::transmute(val) }))
@@ -94,69 +91,94 @@ pub fn parse_data_disp(tok: Sexp, pcx: &ParserContext) -> Result<DataDisp, CrsnE
             Err(CrsnError::Parse("List not expected here".into(), pos))
         }
         Sexp::Atom(Atom::S(s), pos) => {
-            if s == "_" {
-                return Ok(DataDisp::Discard);
-            }
+            parse_data_disp_from_str(&s, &pos, pcx)
+        }
+    }
+}
 
-            if let Some(reference) = s.strip_prefix('@') {
-                /* extension constants (pre-defined handles) */
-                for p in pcx.parsers {
-                    if let Some(val) = p.get_constant_value(reference) {
-                        return Ok(DataDisp::ImmObject(val));
-                    }
-                }
+/// Parse data disp from a string token (a &str is used so the string can be preprocessed)
+pub fn parse_data_disp_from_str(s: &str, pos: &SourcePosition, pcx: &ParserContext) -> Result<DataDisp, CrsnError> {
+    if s == "_" {
+        return Ok(DataDisp::Discard);
+    }
 
-                /* register aliases */
-                let pstate = pcx.state.borrow();
-                if let Some(reg) = pstate.reg_aliases.get(reference) {
-                    return Ok(DataDisp::RegObject(*reg))
-                } else if let Some(reg) = pstate.global_reg_aliases.get(reference) {
-                    return Ok(DataDisp::RegObject(*reg))
-                }
+    if let Some(reference) = s.strip_prefix('@') {
+        /* extension constants (pre-defined handles) */
+        for p in pcx.parsers {
+            if let Some(val) = p.get_constant_value(reference) {
+                return Ok(DataDisp::ImmObject(val));
+            }
+        }
 
-                let reg = reg::parse_reg(reference, &pos)?;
+        /* register aliases */
+        let pstate = pcx.state.borrow();
+        if let Some(reg) = pstate.reg_aliases.get(reference) {
+            return Ok(DataDisp::RegObject(*reg))
+        } else if let Some(reg) = pstate.global_reg_aliases.get(reference) {
+            return Ok(DataDisp::RegObject(*reg))
+        }
 
-                if pstate.reg_aliases.values().find(|v| **v == reg).is_some() {
-                    Err(CrsnError::Parse(format!("Sym exists for register {}, direct access denied. Unsym it if needed.", reg).into(), pos))
-                } else if pstate.global_reg_aliases.values().find(|v| **v == reg).is_some() {
-                    Err(CrsnError::Parse(format!("Global sym exists for register {}, direct access denied. Unsym it if needed.", reg).into(), pos))
-                } else {
-                    Ok(DataDisp::RegObject(reg))
-                }
-            } else {
-                /* extension constants */
-                for p in pcx.parsers {
-                    if let Some(val) = p.get_constant_value(&s) {
-                        return Ok(DataDisp::Immediate(val));
-                    }
-                }
+        let reg = reg::parse_reg(reference, pos)?;
 
-                /* program constants */
-                let pstate = pcx.state.borrow();
-                if let Some(val) = pstate.constants.get(&s) {
-                    return Ok(DataDisp::Immediate(*val));
-                }
+        if pstate.reg_aliases.values().find(|v| **v == reg).is_some() {
+            Err(CrsnError::Parse(format!("Sym exists for register {}, direct access denied. Unsym it if needed.", reg).into(), pos.clone()))
+        } else if pstate.global_reg_aliases.values().find(|v| **v == reg).is_some() {
+            Err(CrsnError::Parse(format!("Global sym exists for register {}, direct access denied. Unsym it if needed.", reg).into(), pos.clone()))
+        } else {
+            Ok(DataDisp::RegObject(reg))
+        }
+    } else {
+        /* extension constants */
+        for p in pcx.parsers {
+            if let Some(val) = p.get_constant_value(&s) {
+                return Ok(DataDisp::Immediate(val));
+            }
+        }
 
-                /* register aliases */
-                if let Some(val) = pstate.reg_aliases.get(&s) {
-                    return Ok(DataDisp::Register(*val));
-                } else if let Some(val) = pstate.global_reg_aliases.get(&s) {
-                    return Ok(DataDisp::Register(*val));
-                }
+        /* program constants */
+        let pstate = pcx.state.borrow();
+        if let Some(val) = pstate.constants.get(s) {
+            return Ok(DataDisp::Immediate(*val));
+        }
 
-                /* register */
-                let reg = reg::parse_reg(&s, &pos)?;
+        /* register aliases */
+        if let Some(val) = pstate.reg_aliases.get(s) {
+            return Ok(DataDisp::Register(*val));
+        } else if let Some(val) = pstate.global_reg_aliases.get(s) {
+            return Ok(DataDisp::Register(*val));
+        }
 
-                let pstate = pcx.state.borrow();
-                if pstate.reg_aliases.values().find(|v| **v == reg).is_some() {
-                    Err(CrsnError::Parse(format!("Sym exists for register {}, direct access denied. Unsym it if needed.", reg).into(), pos))
-                } else if pstate.global_reg_aliases.values().find(|v| **v == reg).is_some() {
-                    Err(CrsnError::Parse(format!("Global sym exists for register {}, direct access denied. Unsym it if needed.", reg).into(), pos))
-                } else {
-                    Ok(DataDisp::Register(reg))
+        /* It can also be a plain old number */
+
+        if s.starts_with('#')
+            || s.starts_with('-')
+            || s.starts_with('.')
+            || s.starts_with("0x")
+            || s.starts_with("0b")
+            || s.starts_with(|c : char| c.is_ascii_digit())
+        {
+            match sexp::atom_of_string(s.to_string()) {
+                Atom::C(v) => return Ok(DataDisp::Immediate(v as Value)),
+                Atom::I(v) => return Ok(DataDisp::Immediate(unsafe { std::mem::transmute::<_, u64>(v) } as Value)),
+                Atom::U(v) => return Ok(DataDisp::Immediate(v as Value)),
+                Atom::F(v) => return Ok(DataDisp::Immediate(unsafe { std::mem::transmute::<_, u64>(v) } as Value)),
+                Atom::S(_) | Atom::QS(_) => {
+                    // this will probably fail validation down the line, but it is definitely not a number
                 }
             }
         }
+
+        /* register */
+        let reg = reg::parse_reg(&s, &pos)?;
+
+        let pstate = pcx.state.borrow();
+        if pstate.reg_aliases.values().find(|v| **v == reg).is_some() {
+            Err(CrsnError::Parse(format!("Sym exists for register {}, direct access denied. Unsym it if needed.", reg).into(), pos.clone()))
+        } else if pstate.global_reg_aliases.values().find(|v| **v == reg).is_some() {
+            Err(CrsnError::Parse(format!("Global sym exists for register {}, direct access denied. Unsym it if needed.", reg).into(), pos.clone()))
+        } else {
+            Ok(DataDisp::Register(reg))
+        }
     }
 }
 
@@ -225,3 +247,59 @@ pub fn parse_wr(tok: Sexp, pcx: &ParserContext) -> Result<Wr, CrsnError> {
     let pos = tok.pos().clone();
     Ok(Wr::new(WrData::try_from(parse_data_disp(tok, pcx)?).err_pos(&pos)?))
 }
+
+pub fn parse_rdwr(tok: Sexp, pcx: &ParserContext) -> Result<RdWr, CrsnError> {
+    let pos = tok.pos().clone();
+    let wr = parse_wr(tok, pcx)?;
+    if !wr.is_readable() {
+        return Err(CrsnError::Parse("Argument is not readable!".into(), pos));
+    }
+    Ok(RdWr::new(wr.0))
+}
+
+pub fn parse_wr_offset(tok: Sexp, pcx: &ParserContext) -> Result<(Wr, u32), CrsnError> {
+    match tok {
+        Sexp::Atom(Atom::S(s), pos) => {
+            let (s, offset) = parse_disp_offs(&s, &pos)?;
+            Ok((Wr::new(WrData::try_from(parse_data_disp_from_str(s, &pos, pcx)?).err_pos(&pos)?), offset))
+        }
+        _ => Ok((parse_wr(tok, pcx)?, 0))
+    }
+}
+
+pub fn parse_rd_offset(tok: Sexp, pcx: &ParserContext) -> Result<(Rd, u32), CrsnError> {
+    match tok {
+        Sexp::Atom(Atom::S(s), pos) => {
+            let (s, offset) = parse_disp_offs(&s, &pos)?;
+            Ok((Rd::new(RdData::try_from(parse_data_disp_from_str(s, &pos, pcx)?).err_pos(&pos)?), offset))
+        }
+        _ => Ok((parse_rd(tok, pcx)?, 0))
+    }
+}
+
+pub fn parse_rdwr_offset(tok: Sexp, pcx: &ParserContext) -> Result<(RdWr, u32), CrsnError> {
+    match tok {
+        Sexp::Atom(Atom::S(s), pos) => {
+            let (s, offset) = parse_disp_offs(&s, &pos)?;
+            let wr = WrData::try_from(parse_data_disp_from_str(s, &pos, pcx)?).err_pos(&pos)?;
+
+            if !wr.is_readable() {
+                return Err(CrsnError::Parse("Argument is not readable!".into(), pos));
+            }
+            Ok((RdWr::new(wr), offset))
+        }
+        _ => Ok((parse_rdwr(tok, pcx)?, 0))
+    }
+}
+
+fn parse_disp_offs<'a>(s : &'a str, pos: &SourcePosition) -> Result<(&'a str, u32), CrsnError> {
+    if s.contains(':') {
+        let mut split = s.split(':');
+
+        let disp = split.next().unwrap();
+        let num : u32 = split.next().unwrap().parse().err_pos(pos)?;
+        Ok((disp, num))
+    } else {
+        Ok((&s, 0))
+    }
+}
diff --git a/crsn/src/builtin/defs.rs b/crsn/src/builtin/defs.rs
index 3bf1b65..4681131 100644
--- a/crsn/src/builtin/defs.rs
+++ b/crsn/src/builtin/defs.rs
@@ -1,13 +1,13 @@
 use sexp::SourcePosition;
 
 use crate::asm::data::{Rd, RdObj, Wr, RdWr};
-use crate::asm::data::literal::{DebugMsg, Label, RoutineName};
+use crate::asm::data::literal::{DebugMsg, Label, RoutineName, Value};
 use crate::asm::instr::Op;
 use crate::asm::instr::op::OpKind;
 use std::fmt::{Display, Formatter};
 use std::fmt;
 use crate::asm::error::CrsnError;
-use crate::asm::patches::ErrWithPos;
+
 
 #[derive(Debug)]
 pub enum Barrier {
@@ -41,14 +41,20 @@ pub enum LdsValue {
     Chars(String),
 }
 
+/// Instruction's bit mask
 #[derive(Clone, Copy, Debug, Eq, PartialEq)]
-pub struct BitSlice {
+pub struct BitMask {
+    /// Operation bit width
     pub width: u32,
+    /// Destination (or first operand's) offset
+    pub dst_pos: u32,
+    /// Source (or second operand's) offset
     pub src_pos: u32,
-    pub dst_pos: u32
+    /// Source 2 offset (if used)
+    pub src2_pos: u32,
 }
 
-impl Display for BitSlice {
+impl Display for BitMask {
     fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
         if self.is_full() {
             return Ok(());
@@ -65,94 +71,42 @@ impl Display for BitSlice {
     }
 }
 
-impl Default for BitSlice {
+impl Default for BitMask {
     fn default() -> Self {
         Self {
             width: 64,
+            dst_pos: 0,
             src_pos: 0,
-            dst_pos: 0
+            src2_pos: 0,
         }
     }
 }
 
-impl BitSlice {
+impl BitMask {
     pub fn full() -> Self {
         Self::default()
     }
 
-    /// Parse LEN (no SRC and DST)
-    pub fn parse1(src: &str, pos: &SourcePosition) -> Result<Option<Self>, CrsnError> {
-        match Self::parse(src, pos) {
-            Ok(Some(slice)) => {
-                if slice.src_pos != 0 || slice.dst_pos != 0 {
-                    return Err(CrsnError::Parse("Excess bit slice modifiers".into(), pos.clone()));
-                }
-
-                Ok(Some(BitSlice {
-                    width: slice.width,
-                    src_pos: 0,
-                    dst_pos: 0
-                }))
-            }
-            other => other
-        }
-    }
-
-    /// Parse LEN/SRC
-    pub fn parse2(src: &str, pos: &SourcePosition) -> Result<Option<Self>, CrsnError> {
-        match Self::parse(src, pos) {
-            Ok(Some(slice)) => {
-                if slice.src_pos != 0 {
-                    return Err(CrsnError::Parse("Excess bit slice modifiers".into(), pos.clone()));
-                }
-
-                Ok(Some(BitSlice {
-                    width: slice.width,
-                    src_pos: slice.dst_pos,
-                    dst_pos: 0
-                }))
-            }
-            other => other
-        }
-    }
-
-    /// Parse a bit slice LEN/DST/SRC
-    /// - String not starting with a digit is rejected as `Ok(None)`
-    /// - Empty string is parsed as a full slice
-    pub fn parse(src: &str, pos: &SourcePosition) -> Result<Option<Self>, CrsnError> {
-        if src.is_empty() {
-            return Ok(Some(BitSlice::full()));
-        }
+    pub fn validate(&self, pos: &SourcePosition) -> Result<(), CrsnError> {
+        #[allow(non_snake_case)]
+        let MAXWIDTH = (std::mem::size_of::<Value>() as u32)*8;
 
-        if !src.starts_with(|c: char| c.is_ascii_digit()) {
-            return Ok(None);
-        }
-
-        // We have a opXX, opXX/YY, or opXX/YY/ZZ
-        let mut numbers : Vec<u32> = vec![];
-        for p in src.split('/') {
-            numbers.push(p.parse().err_pos(pos)?);
-        }
-
-        let slice = BitSlice {
-            width: numbers[0],
-            src_pos: numbers.get(2).copied().unwrap_or(0),
-            dst_pos: numbers.get(1).copied().unwrap_or(0)
-        };
-
-        if slice.width == 0 || slice.width > 64 {
-            return Err(CrsnError::Parse("Bit slice width must be 1-64".into(), pos.clone()));
+        if self.width == 0 || self.width > MAXWIDTH {
+            return Err(CrsnError::Parse(format!("Bit width must be 1-{}", MAXWIDTH).into(), pos.clone()));
         }
 
         // Validation
-        if slice.src_pos + slice.width > 64 {
-            return Err(CrsnError::Parse("Invalid source bit slice".into(), pos.clone()));
+        if self.src_pos + self.width > MAXWIDTH {
+            return Err(CrsnError::Parse("Invalid source bit mask".into(), pos.clone()));
+        }
+        if self.src2_pos + self.width > MAXWIDTH {
+            return Err(CrsnError::Parse("Invalid source 2 bit mask".into(), pos.clone()));
         }
-        if slice.dst_pos + slice.width > 64 {
-            return Err(CrsnError::Parse("Invalid destination bit slice".into(), pos.clone()));
+        if self.dst_pos + self.width > MAXWIDTH {
+            return Err(CrsnError::Parse("Invalid destination bit mask".into(), pos.clone()));
         }
 
-        Ok(Some(slice))
+        Ok(())
     }
 
     pub fn is_full(self) -> bool {
@@ -203,13 +157,13 @@ pub enum BuiltinOp {
     /// Move a value
     Load { dst: Wr, src: Rd },
     /// Move bits of a value
-    LoadBits { dst: RdWr, src: Rd, slice: BitSlice },
+    LoadBits { dst: Wr, a: Rd, b: Rd, mask: BitMask },
     /// Move N values
     LoadMultiple { dst: Wr, src: Rd, count: Rd },
     /// Move values from a string or integer list
     LoadSequence { dst: Wr, value: LdsValue },
     /// Swap two registers
-    Exchange { a: RdWr, b: RdWr, slice : BitSlice },
+    Exchange { a: RdWr, b: RdWr, mask: BitMask },
     /// Store runtime status to a register
     StoreFlags { dst: Wr },
     /// Load runtime status from a register
diff --git a/crsn/src/builtin/exec.rs b/crsn/src/builtin/exec.rs
index bad3117..80eaab8 100644
--- a/crsn/src/builtin/exec.rs
+++ b/crsn/src/builtin/exec.rs
@@ -119,12 +119,12 @@ impl OpTrait for BuiltinOp {
                 }
                 state.update_status(last);
             }
-            BuiltinOp::LoadBits { dst, src, slice, } => {
+            BuiltinOp::LoadBits { dst, a, b, mask: slice, } => {
                 state.clear_status();
-                let new = state.read(src)?;
-                let old = state.read(dst)?;
+                let new = state.read(b)?;
+                let old = state.read(a)?;
                 let ones : u64 = (1 << slice.width) - 1;
-                let val = (old & !(ones << slice.dst_pos)) | (((new & (ones << slice.src_pos)) >> slice.src_pos) << slice.dst_pos);
+                let val = (old & !(ones << slice.src_pos)) | (((new & (ones << slice.src2_pos)) >> slice.src2_pos) << slice.src_pos);
                 state.update_status(val);
                 state.write(dst, val)?;
             }
@@ -158,7 +158,7 @@ impl OpTrait for BuiltinOp {
                     }
                 }
             }
-            BuiltinOp::Exchange { a, b, slice } => {
+            BuiltinOp::Exchange { a, b, mask: slice } => {
                 let aa = state.read(a)?;
                 let bb = state.read(b)?;
 
diff --git a/crsn/src/builtin/parse.rs b/crsn/src/builtin/parse.rs
index 9342ec3..1e9fd18 100644
--- a/crsn/src/builtin/parse.rs
+++ b/crsn/src/builtin/parse.rs
@@ -8,10 +8,10 @@ use crate::asm::parse::arg_parser::TokenParser;
 use crate::asm::parse::parse_data::{parse_constant_name, parse_label, parse_label_str, parse_rd, parse_reg_alias, parse_value};
 use crate::asm::parse::sexp_expect::{expect_any_string_atom};
 use crate::asm::patches::ErrWithPos;
-use crate::builtin::defs::{Barrier, BuiltinOp, SleepUnit, LdsValue, BitSlice};
+use crate::builtin::defs::{Barrier, BuiltinOp, SleepUnit, LdsValue};
 use crate::module::ParseRes;
-use crate::utils::A;
-use crate::asm::data::{Rd, RdData, RdObj};
+use crate::utils::{A, AM};
+use crate::asm::data::{Rd, RdData, RdObj, DataDispEquals};
 
 pub(crate) fn parse_op<'a>(op_pos: &SourcePosition, keyword: &str, mut args: TokenParser<'a>) -> Result<ParseRes<'a, OpKind>, CrsnError> {
     let pcx = args.pcx;
@@ -238,13 +238,13 @@ pub(crate) fn parse_op<'a>(op_pos: &SourcePosition, keyword: &str, mut args: Tok
             BuiltinOp::LoadSequence { dst, value }
         }
 
-        "xch" => {
-            BuiltinOp::Exchange {
-                a: args.next_rdwr()?,
-                b: args.next_rdwr()?,
-                slice: BitSlice::full(),
-            }
-        }
+        // "xch" => {
+        //     BuiltinOp::Exchange {
+        //         a: args.next_rdwr()?,
+        //         b: args.next_rdwr()?,
+        //         mask: BitMask::full(),
+        //     }
+        // }
 
         "stf" => {
             BuiltinOp::StoreFlags {
@@ -276,24 +276,17 @@ pub(crate) fn parse_op<'a>(op_pos: &SourcePosition, keyword: &str, mut args: Tok
         }
 
         other => {
-            if let Some(s) = other.strip_prefix("ld") {
-                if let Some(slice) = BitSlice::parse(s, op_pos)? {
-                    return Ok(ParseRes::builtin(BuiltinOp::LoadBits {
-                        dst: args.next_rdwr()?,
-                        src: args.next_rd()?,
-                        slice,
-                    }));
+            if let Some((dst, a, b, mask)) = args.parse_masked_wr_rd_rd(other, "ld")? {
+                if !dst.disp_equals(a) && mask.dst_pos != 0 {
+                    return Err(CrsnError::Parse(
+                        "Invalid offsets - permissible formats are (RW:dst_offs Rd:src_offs) and (Wr Rd:dst_offs, Rd:src_offs) ".into(),
+                        op_pos.clone()));
                 }
+                return Ok(ParseRes::builtin(BuiltinOp::LoadBits { dst, a, b, mask }));
             }
 
-            if let Some(s) = other.strip_prefix("xch") {
-                if let Some(slice) = BitSlice::parse(s, op_pos)? {
-                    return Ok(ParseRes::builtin(BuiltinOp::Exchange {
-                        a: args.next_rdwr()?,
-                        b: args.next_rdwr()?,
-                        slice,
-                    }));
-                }
+            if let Some((a, b, mask)) = args.parse_masked_rdwr_rdwr(other, "xch")? {
+                return Ok(ParseRes::builtin(BuiltinOp::Exchange { a, b, mask }));
             }
 
             if other.starts_with(':') {
@@ -381,11 +374,21 @@ pub(crate) fn to_sexp(op: &BuiltinOp) -> Sexp {
         BuiltinOp::Load { dst, src, } => {
             sexp::list(&[A("ld"), A(dst), A(src)])
         },
-        BuiltinOp::LoadBits { dst, src, slice, } => {
-            sexp::list(&[A(format!("ld{}", slice)), A(dst), A(src)])
+        BuiltinOp::LoadBits { dst, a, b, mask, } => {
+            if dst.disp_equals(a) {
+                sexp::list(&[A(format!("ld{}", mask.width)), AM(dst, mask.dst_pos), AM(b, mask.src2_pos)])
+            } else {
+                sexp::list(&[A(format!("ld{}", mask.width)), AM(dst, mask.dst_pos), AM(a, mask.src_pos), AM(b, mask.src2_pos)])
+            }
         },
         BuiltinOp::LoadMultiple { dst, src, count } => sexp::list(&[A("ldn"), A(dst), A(src), A(count)]),
-        BuiltinOp::Exchange { a, b, slice } => sexp::list(&[A(format!("xch{}", slice)), A(a), A(b)]),
+        BuiltinOp::Exchange { a, b, mask } => {
+            if mask.is_full() {
+                sexp::list(&[A("xch"), A(a), A(b)])
+            } else {
+                sexp::list(&[A(format!("xch{}", mask.width)), AM(a, mask.dst_pos), AM(b, mask.src_pos)])
+            }
+        },
         BuiltinOp::StoreFlags { dst } => sexp::list(&[A("stf"), A(dst)]),
         BuiltinOp::LoadFlags { src } => sexp::list(&[A("ldf"), A(src)]),
         BuiltinOp::LoadSequence { dst, value } => {
@@ -470,14 +473,15 @@ mod test {
             ("(fault \"do pr*ele\")", "(fault \"do pr*ele\")"),
             ("(xch r0 r1)", "(xch r0 r1)"),
             ("(xch32 r0 r1)", "(xch32 r0 r1)"),
-            ("(xch32/8/16 r0 r1)", "(xch32/8/16 r0 r1)"),
+            ("(xch32 r0:8 r1:16)", "(xch32 r0:8 r1:16)"),
             ("(ld r0 r0)", "(ld r0 r0)"),
             ("(ld8 r0 r1)", "(ld8 r0 r1)"),
             ("(ld16 r0 r1)", "(ld16 r0 r1)"),
             ("(ld32 r0 r1)", "(ld32 r0 r1)"),
-            ("(ld32/32 r0 r1)", "(ld32/32 r0 r1)"),
-            ("(ld32/32/5 r0 r1)", "(ld32/32/5 r0 r1)"),
-            ("(ld32/0/16 r0 r1)", "(ld32/0/16 r0 r1)"),
+            ("(ld32 r0:32 r1)", "(ld32 r0:32 r1)"),
+            ("(ld32 r0:32 r1:5)", "(ld32 r0:32 r1:5)"),
+            ("(ld32 r0 r1:16)", "(ld32 r0 r1:16)"),
+            ("(ld32 r7 r0:3 r1:16)", "(ld32 r7 r0:3 r1:16)"),
             ("(ld r0 156)", "(ld r0 156)"),
             ("(ld _ -32767)", "(ld _ -32767)"),
             ("(ldn _ @r0 7)", "(ldn _ @r0 7)"),
@@ -512,7 +516,7 @@ mod test {
                 .expect("flatten").remove(0);
 
             let exported = parsed.to_sexp().to_string();
-            println!("Parsed: {}", exported);
+            println!("Parsed: {}, dbg {:?}", exported, parsed);
 
             assert_eq!(expected, exported);
 
diff --git a/crsn/src/runtime/run_thread.rs b/crsn/src/runtime/run_thread.rs
index f34a946..cb04b63 100644
--- a/crsn/src/runtime/run_thread.rs
+++ b/crsn/src/runtime/run_thread.rs
@@ -87,6 +87,7 @@ impl RunThread {
                 }
                 Err(e) => {
                     error!("Fault: {:?}", e);
+                    error!("Core dump: {:?}", self.state);
                     break 'run;
                 }
             }
diff --git a/crsn/src/runtime/run_thread/state.rs b/crsn/src/runtime/run_thread/state.rs
index 7e286f0..3022e82 100644
--- a/crsn/src/runtime/run_thread/state.rs
+++ b/crsn/src/runtime/run_thread/state.rs
@@ -10,6 +10,8 @@ use std::sync::Arc;
 use crate::runtime::run_thread::ThreadInfo;
 use nudge::{likely};
 use crate::asm::instr::cond::Flag;
+use std::fmt::{Debug, Formatter};
+use std::fmt;
 
 pub struct RunState {
     pub thread_info: Arc<ThreadInfo>,
@@ -23,6 +25,17 @@ pub struct RunState {
     pub ext_data: ExtensionDataStore,
 }
 
+impl Debug for RunState {
+    fn fmt(&self, f: &mut Formatter<'_>) -> fmt::Result {
+        f.debug_struct("RunState")
+            .field("frame", &self.frame)
+            .field("call_stack", &self.call_stack)
+            .field("global_regs", &self.global_regs)
+            //.field("ext_data", &self.ext_data)
+            .finish()
+    }
+}
+
 #[derive(Debug,Default)]
 pub struct ExtensionDataStore {
     store: HashMap<TypeId, Box<dyn Any + Send>>
diff --git a/crsn/src/utils/mod.rs b/crsn/src/utils/mod.rs
index 95765c8..ec95cf1 100644
--- a/crsn/src/utils/mod.rs
+++ b/crsn/src/utils/mod.rs
@@ -4,9 +4,10 @@ use std::str::FromStr;
 pub use option_ext::UncheckedOptionExt;
 use sexp::{Atom, Sexp};
 
+
 mod option_ext;
 
-/// Convert a string token to Sexp
+/// Convert a value to Sexp
 #[allow(non_snake_case)]
 pub fn A(s: impl Display) -> Sexp {
     let s = s.to_string();
@@ -21,9 +22,9 @@ pub fn A(s: impl Display) -> Sexp {
         return Sexp::Atom(Atom::I(x), Default::default());
     }
 
-    let y: Result<f64, _> = FromStr::from_str(&s);
-    if let Ok(y) = y {
-        return Sexp::Atom(Atom::F(y), Default::default());
+    let x: Result<f64, _> = FromStr::from_str(&s);
+    if let Ok(x) = x {
+        return Sexp::Atom(Atom::F(x), Default::default());
     }
 
     if s.contains(|c: char| " \t\"\\\n\t\r".contains(c)) {
@@ -32,3 +33,33 @@ pub fn A(s: impl Display) -> Sexp {
         Sexp::Atom(Atom::S(s), Default::default())
     }
 }
+/// Convert a value to Sexp, with a bit mask
+#[allow(non_snake_case)]
+pub fn AM(s: impl Display, offs : u32) -> Sexp {
+    if offs == 0 {
+        return A(s);
+    }
+
+    let s = s.to_string();
+
+    let x: Result<u64, _> = FromStr::from_str(&s);
+    if let Ok(x) = x {
+        return Sexp::Atom(Atom::S(format!("{}:{}", x, offs)), Default::default());
+    }
+
+    let x: Result<i64, _> = FromStr::from_str(&s);
+    if let Ok(x) = x {
+        return Sexp::Atom(Atom::S(format!("{}:{}", x, offs)), Default::default());
+    }
+
+    let x: Result<f64, _> = FromStr::from_str(&s);
+    if let Ok(x) = x {
+        return Sexp::Atom(Atom::S(format!("{}:{}", x, offs)), Default::default());
+    }
+
+    if s.contains(|c: char| " \t\"\\\n\t\r".contains(c)) {
+        panic!("Quoted string with a bit mask!");
+    } else {
+        Sexp::Atom(Atom::S(format!("{}:{}", s, offs)), Default::default())
+    }
+}
diff --git a/crsn_arith/src/defs.rs b/crsn_arith/src/defs.rs
index e15ecce..3e96dbb 100644
--- a/crsn_arith/src/defs.rs
+++ b/crsn_arith/src/defs.rs
@@ -1,5 +1,5 @@
 use crsn::asm::data::{Rd, Wr};
-use crsn::builtin::defs::BitSlice;
+use crsn::builtin::defs::BitMask;
 
 /// A low level instruction
 #[derive(Clone, Debug, Eq, PartialEq)]
@@ -20,11 +20,11 @@ pub enum ArithOp {
     /// Reverse bits
     Rbit { dst: Wr, src: Rd },
     /// Count leading zeros in a slice
-    Clz { dst: Wr, src: Rd, slice: BitSlice },
+    Clz { dst: Wr, src: Rd, mask: BitMask },
     /// Count leading ones in a slice
-    Clo { dst: Wr, src: Rd, slice: BitSlice },
+    Clo { dst: Wr, src: Rd, mask: BitMask },
     /// Sign extend a slice (zero aligned)
-    SignExtend { dst: Wr, src: Rd, slice: BitSlice },
+    SignExtend { dst: Wr, src: Rd, mask: BitMask },
 
     Add { dst: Wr, a: Rd, b: Rd },
     Sub { dst: Wr, a: Rd, b: Rd },
diff --git a/crsn_arith/src/exec.rs b/crsn_arith/src/exec.rs
index 967812a..c2f3966 100644
--- a/crsn_arith/src/exec.rs
+++ b/crsn_arith/src/exec.rs
@@ -250,19 +250,19 @@ impl OpTrait for ArithOp {
                 let res = val.reverse_bits();
                 state.write(dst, res)?;
             }
-            ArithOp::Clz { dst, src, slice } => {
+            ArithOp::Clz { dst, src, mask: slice } => {
                 state.clear_status();
-                let mut val = ((state.read(src)? >> slice.src_pos) << (64 - slice.width));
+                let mut val = (state.read(src)? >> slice.src_pos) << (64 - slice.width);
                 if !slice.is_full() {
-                    val |= ((1 << slice.width) - 1);
+                    val |= (1 << slice.width) - 1;
                 }
                 let res = val.leading_zeros() as u64;
                 state.update_status(res);
                 state.write(dst, res)?;
             }
-            ArithOp::Clo { dst, src, slice } => {
+            ArithOp::Clo { dst, src, mask: slice } => {
                 state.clear_status();
-                let mut val = ((state.read(src)? >> slice.src_pos) << (64 - slice.width));
+                let mut val = (state.read(src)? >> slice.src_pos) << (64 - slice.width);
                 if !slice.is_full() {
                     val &= ((1 << slice.width) - 1) << slice.src_pos;
                 }
@@ -270,7 +270,7 @@ impl OpTrait for ArithOp {
                 state.update_status(res);
                 state.write(dst, res)?;
             }
-            ArithOp::SignExtend { dst, src, slice } => {
+            ArithOp::SignExtend { dst, src, mask: slice } => {
                 state.clear_status();
                 let val = state.read(src)?;
 
@@ -330,9 +330,9 @@ impl OpTrait for ArithOp {
             ArithOp::Sw8 { dst, src } => to_sexp_1_or_2("sw8", dst, src),
             ArithOp::Rev { dst, src } => to_sexp_1_or_2("rev", dst, src),
             ArithOp::Rbit { dst, src } => to_sexp_1_or_2("rbit", dst, src),
-            ArithOp::Clz { dst, src, slice } => to_sexp_1_or_2(&format!("clz{}", slice), dst, src),
-            ArithOp::Clo { dst, src, slice } => to_sexp_1_or_2(&format!("clo{}", slice), dst, src),
-            ArithOp::SignExtend { dst, src, slice } => to_sexp_1_or_2(&format!("se{}", slice), dst, src),
+            ArithOp::Clz { dst, src, mask: slice } => to_sexp_1_or_2(&format!("clz{}", slice), dst, src),
+            ArithOp::Clo { dst, src, mask: slice } => to_sexp_1_or_2(&format!("clo{}", slice), dst, src),
+            ArithOp::SignExtend { dst, src, mask: slice } => to_sexp_1_or_2(&format!("se{}", slice), dst, src),
         }
     }
 }
diff --git a/crsn_arith/src/parse.rs b/crsn_arith/src/parse.rs
index 8501dfe..99d9d2a 100644
--- a/crsn_arith/src/parse.rs
+++ b/crsn_arith/src/parse.rs
@@ -1,4 +1,4 @@
-use crsn::asm::data::{Rd, Wr};
+use crsn::asm::data::{Rd, Wr, DataDispEquals};
 use crsn::asm::error::CrsnError;
 use crsn::asm::instr::op::OpKind;
 use crsn::asm::parse::arg_parser::TokenParser;
@@ -6,7 +6,7 @@ use crsn::module::ParseRes;
 use crsn::sexp::SourcePosition;
 
 use crate::defs::ArithOp;
-use crsn::builtin::defs::BitSlice;
+
 
 pub(crate) fn parse<'a>(op_pos: &SourcePosition, keyword: &str, mut args: TokenParser<'a>) -> Result<ParseRes<'a, OpKind>, CrsnError> {
     Ok(ParseRes::ext(match keyword {
@@ -574,58 +574,43 @@ pub(crate) fn parse<'a>(op_pos: &SourcePosition, keyword: &str, mut args: TokenP
         }
 
         other => {
-            if let Some(s) = other.strip_prefix("clz") {
-                if let Some(slice) = BitSlice::parse2(s, op_pos)? {
-                    return Ok(ParseRes::ext(match args.len() {
-                        2 => {
-                            ArithOp::Clz { dst: args.next_wr()?, src: args.next_rd()?, slice }
-                        }
-                        1 => {
-                            let dst = args.next_rdwr()?;
-                            ArithOp::Clz { dst: dst.wr(), src: dst.rd(), slice }
-                        }
-                        _ => {
-                            return Err(CrsnError::Parse("Clz requires 1 or 2 arguments".into(), op_pos.clone()));
-                        }
-                    }));
-                }
+            if let Some((dst, src, mask)) = args.parse_masked_wr_rd(other, "clz")? {
+                if !dst.disp_equals(src) && mask.dst_pos != 0 {
+                    return Err(CrsnError::Parse(
+                        "Invalid offsets - permissible formats are (RW:src_offs) and (Wr Rd:src_offs) ".into(),
+                        op_pos.clone()));
+                }
+                return Ok(ParseRes::ext(ArithOp::Clz {
+                    dst,
+                    src,
+                    mask
+                }));
             }
 
-            if let Some(s) = other.strip_prefix("clo") {
-                if let Some(slice) = BitSlice::parse2(s, op_pos)? {
-                    return Ok(ParseRes::ext(match args.len() {
-                        2 => {
-                            ArithOp::Clo { dst: args.next_wr()?, src: args.next_rd()?, slice }
-                        }
-                        1 => {
-                            let dst = args.next_rdwr()?;
-                            ArithOp::Clo { dst: dst.wr(), src: dst.rd(), slice }
-                        }
-                        _ => {
-                            return Err(CrsnError::Parse("Clz requires 1 or 2 arguments".into(), op_pos.clone()));
-                        }
-                    }));
-                }
+            if let Some((dst, src, mask)) = args.parse_masked_wr_rd(other, "clo")? {
+                if !dst.disp_equals(src) && mask.dst_pos != 0 {
+                    return Err(CrsnError::Parse(
+                        "Invalid offsets - permissible formats are (RW:src_offs) and (Wr Rd:src_offs)".into(),
+                        op_pos.clone()));
+                }
+                return Ok(ParseRes::ext(ArithOp::Clo {
+                    dst,
+                    src,
+                    mask
+                }));
             }
 
-            if let Some(s) = other.strip_prefix("se") {
-                if let Some(slice) = BitSlice::parse1(s, op_pos)? {
-                    if slice.is_full() {
-                        return Err(CrsnError::Parse("Sign extend requires a bit size (< 64)".into(), op_pos.clone()));
-                    }
-                    return Ok(ParseRes::ext(match args.len() {
-                        2 => {
-                            ArithOp::SignExtend { dst: args.next_wr()?, src: args.next_rd()?, slice }
-                        }
-                        1 => {
-                            let dst = args.next_rdwr()?;
-                            ArithOp::SignExtend { dst: dst.wr(), src: dst.rd(), slice }
-                        }
-                        _ => {
-                            return Err(CrsnError::Parse("Sign extend requires 1 or 2 arguments".into(), op_pos.clone()));
-                        }
-                    }));
-                }
+            if let Some((dst, src, mask)) = args.parse_masked_wr_rd(other, "se")? {
+                if mask.src_pos != 0 || mask.dst_pos != 0 {
+                    return Err(CrsnError::Parse(
+                        "Offsets do not make sense for SignExtend".into(),
+                        op_pos.clone()));
+                }
+                return Ok(ParseRes::ext(ArithOp::SignExtend {
+                    dst,
+                    src,
+                    mask
+                }));
             }
 
             return Ok(ParseRes::Unknown(args));
diff --git a/examples/bmp-parser.csn b/examples/bmp-parser.csn
index 2d173fe..0f09bbe 100644
--- a/examples/bmp-parser.csn
+++ b/examples/bmp-parser.csn
@@ -104,16 +104,16 @@
     (proc rd4
         ; Read 4 bytes as little endian
         (ld8 r0 @cin_r)
-        (ld8/8 r0 @cin_r)
-        (ld8/16 r0 @cin_r)
-        (ld8/24 r0 @cin_r)
+        (ld8 r0:8 @cin_r)
+        (ld8 r0:16 @cin_r)
+        (ld8 r0:24 @cin_r)
         (ret r0)
     )
     
     (proc rd2
         ; Read 2 bytes as little endian
         (ld8 r0 @cin_r)
-        (ld8/8 r0 @cin_r)
+        (ld8 r0:8 @cin_r)
         (ret r0)
     )
 )
diff --git a/examples/test_clz_clo_se.csn b/examples/test_clz_clo_se.csn
index 8f336a3..6465b6d 100644
--- a/examples/test_clz_clo_se.csn
+++ b/examples/test_clz_clo_se.csn
@@ -26,10 +26,10 @@
     (clz32 r0 0xFF)
     (cmp r0 24 (ne? (fault "9")))
 
-    (clz32/16 r0 0x00ff_0000_0000)
+    (clz32 r0 0x00ff_0000_0000:16)
     (cmp r0 8 (ne? (fault "10")))
 
-    (clz32/16 r0 0x0000_0000_0000)
+    (clz32 r0 0x0000_0000_0000:16)
     (cmp r0 32 (ne? (fault "11")))
 
     (clo r0 0)
diff --git a/examples/test_ldbits.csn b/examples/test_ldbits_xchbits.csn
similarity index 75%
rename from examples/test_ldbits.csn
rename to examples/test_ldbits_xchbits.csn
index 67e9425..e5ba239 100644
--- a/examples/test_ldbits.csn
+++ b/examples/test_ldbits_xchbits.csn
@@ -12,11 +12,15 @@
     (cmp r0 0x11223344_5566778d (ne? (fault "3")))
 
     (ld r0 0x11223344_55667788)
-    (ld32/32 r0 0xaabbccdd)
+    (ld32 r0:32 0xaabbccdd)
     (cmp r0 0xaabbccdd_55667788 (ne? (fault "4")))
 
+    ; ldX supports number literals, and can take 3 arguments to combine 2 and place the result in the 3rd
+    (ld24 r0 0xf0000000_0000abcd:32 0x77aabbcc)
+    (cmp  r0 0xf0aabbcc_0000abcd (ne? (fault "4b")))
+
     (ld r0 0x11223344_55667788)
-    (ld32/16/32 r0 0xaabbccdd_00000000)
+    (ld32 r0:16 0xaabbccdd_00000000:32)
     (cmp r0 0x1122aabb_ccdd7788 (ne? (fault "5")))
 
     (ld r0 0x11223344_55667788)
@@ -27,13 +31,13 @@
 
     (ld r0 0x11223344_55667788)
     (ld r1 0xaabbccdd_eeff9900)
-    (xch16/32 r0 r1)
+    (xch16 r0:32 r1)
     (cmp r0 0x11229900_55667788 (ne? (fault "7")))
     (cmp r1 0xaabbccdd_eeff3344 (ne? (fault "7")))
 
     (ld r0 0x11223344_55667788)
     (ld r1 0xaabbccdd_eeff9900)
-    (xch16/32/16 r0 r1)
+    (xch16 r0:32 r1:16)
     (cmp r0 0x1122eeff_55667788 (ne? (fault "7")))
     (cmp r1 0xaabbccdd_33449900 (ne? (fault "7")))
 )
diff --git a/launcher/src/main.rs b/launcher/src/main.rs
index ea0a278..c8e8fa0 100644
--- a/launcher/src/main.rs
+++ b/launcher/src/main.rs
@@ -34,6 +34,8 @@ struct Config {
     program_file: String,
     #[serde(skip)]
     assemble_only: bool,
+    #[serde(skip)]
+    assemble_debug: bool,
     #[serde(with = "serde_duration_millis")]
     cycle_time: Duration,
 }
@@ -47,6 +49,7 @@ impl Default for Config {
             },
             program_file: "".to_string(),
             assemble_only: false,
+            assemble_debug: false,
             cycle_time: Duration::default(),
         }
     }
@@ -88,6 +91,12 @@ impl AppConfig for Config {
                     .long("asm")
                     .help("Only assemble, do not run."),
             )
+            .arg(
+                clap::Arg::with_name("asm-debug")
+                    .short("D")
+                    .long("asm-debug")
+                    .help("Only assemble, do not run. Print the result in debug format."),
+            )
             .arg(
                 clap::Arg::with_name("cycle")
                     .long("cycle")
@@ -100,7 +109,8 @@ impl AppConfig for Config {
 
     fn configure(mut self, clap: &ArgMatches) -> anyhow::Result<Self> {
         self.program_file = clap.value_of("input").unwrap().to_string();
-        self.assemble_only = clap.is_present("asm-only");
+        self.assemble_debug = clap.is_present("asm-debug");
+        self.assemble_only = self.assemble_debug || clap.is_present("asm-only");
         if let Some(t) = clap.value_of("cycle") {
             let (t, mul) = if t.ends_with("us") {
                 (&t[..(t.len()-2)], 1)
@@ -142,7 +152,11 @@ fn main() -> anyhow::Result<()> {
 
     if config.assemble_only {
         for (n, op) in parsed.ops.iter().enumerate() {
-            println!("{:04} : {}", n, op.to_sexp());
+            if config.assemble_debug {
+                println!("{:04} : {:?}", n, op);
+            } else {
+                println!("{:04} : {}", n, op.to_sexp());
+            }
         }
         return Ok(());
     } else {