## Mu's instructions and their table-driven translation See http://akkartik.name/akkartik-convivial-20200607.pdf for the complete story. In brief: Mu is a statement-oriented language. Blocks consist of flat lists of instructions. Instructions can have inputs after the operation, and outputs to the left of a '<-'. Inputs and outputs must be variables. They can't include nested expressions. Variables can be literals ('n'), or live in a register ('var/reg') or in memory ('var') at some 'stack-offset' from the 'ebp' register. Outputs must be registers. To modify a variable in memory, pass it in by reference as an input. (Inputs are more precisely called 'inouts'.) Conversely, registers that are just read from must not be passed as outputs. The following chart shows all the instruction forms supported by Mu, along with the SubX instruction they're translated to. ## Integer instructions These instructions use the general-purpose registers. var/eax <- increment => "40/increment-eax" var/ecx <- increment => "41/increment-ecx" var/edx <- increment => "42/increment-edx" var/ebx <- increment => "43/increment-ebx" var/esi <- increment => "46/increment-esi" var/edi <- increment => "47/increment-edi" increment var => "ff 0/subop/increment *(ebp+" var.stack-offset ")" increment *var/reg => "ff 0/subop/increment *" reg var/eax <- decrement => "48/decrement-eax" var/ecx <- decrement => "49/decrement-ecx" var/edx <- decrement => "4a/decrement-edx" var/ebx <- decrement => "4b/decrement-ebx" var/esi <- decrement => "4e/decrement-esi" var/edi <- decrement => "4f/decrement-edi" decrement var => "ff 1/subop/decrement *(ebp+" var.stack-offset ")" decrement *var/reg => "ff 1/subop/decrement *" reg var/reg <- add var2/reg2 => "01/add-to %" reg " " reg2 "/r32" var/reg <- add var2 => "03/add *(ebp+" var2.stack-offset ") " reg "/r32" var/reg <- add *var2/reg2 => "03/add *" reg2 " " reg "/r32" add-to var1, var2/reg => "01/add-to *(ebp+" var1.stack-offset ") " reg "/r32" add-to *var1/reg1, var2/reg2 => "01/add-to *" reg1 " " reg2 "/r32" var/eax <- add n => "05/add-to-eax " n "/imm32" var/reg <- add n => "81 0/subop/add %" reg " " n "/imm32" add-to var, n => "81 0/subop/add *(ebp+" var.stack-offset ") " n "/imm32" add-to *var/reg, n => "81 0/subop/add *" reg " " n "/imm32" var/reg <- subtract var2/reg2 => "29/subtract-from %" reg " " reg2 "/r32" var/reg <- subtract var2 => "2b/subtract *(ebp+" var2.stack-offset ") " reg "/r32" var/reg <- subtract *var2/reg2 => "2b/subtract *" reg2 " " reg1 "/r32" subtract-from var1, var2/reg2 => "29/subtract-from *(ebp+" var1.stack-offset ") " reg2 "/r32" subtract-from *var1/reg1, var2/reg2 => "29/subtract-from *" reg1 " " reg2 "/r32" var/eax <- subtract n => "2d/subtract-from-eax " n "/imm32" var/reg <- subtract n => "81 5/subop/subtract %" reg " " n "/imm32" subtract-from var, n => "81 5/subop/subtract *(ebp+" var.stack-offset ") " n "/imm32" subtract-from *var/reg, n => "81 5/subop/subtract *" reg " " n "/imm32" var/reg <- and var2/reg2 => "21/and-with %" reg " " reg2 "/r32" var/reg <- and var2 => "23/and *(ebp+" var2.stack-offset " " reg "/r32" var/reg <- and *var2/reg2 => "23/and *" reg2 " " reg "/r32" and-with var1, var2/reg => "21/and-with *(ebp+" var1.stack-offset ") " reg "/r32" and-with *var1/reg1, var2/reg2 => "21/and-with *" reg1 " " reg2 "/r32" var/eax <- and n => "25/and-with-eax " n "/imm32" var/reg <- and n => "81 4/subop/and %" reg " " n "/imm32" and-with var, n => "81 4/subop/and *(ebp+" var.stack-offset ") " n "/imm32" and-with *var/reg, n => "81 4/subop/and *" reg " " n "/imm32" var/reg <- or var2/reg2 => "09/or-with %" reg " " reg2 "/r32" var/reg <- or var2 => "0b/or *(ebp+" var2.stack-offset ") " reg "/r32" var/reg <- or *var2/reg2 => "0b/or *" reg2 " " reg "/r32" or-with var1, var2/reg2 => "09/or-with *(ebp+" var1.stack-offset " " reg2 "/r32" or-with *var1/reg1, var2/reg2 => "09/or-with *" reg1 " " reg2 "/r32" var/eax <- or n => "0d/or-with-eax " n "/imm32" var/reg <- or n => "81 1/subop/or %" reg " " n "/imm32" or-with var, n => "81 1/subop/or *(ebp+" var.stack-offset ") " n "/imm32" or-with *var/reg, n => "81 1/subop/or *" reg " " n "/imm32" var/reg <- xor var2/reg2 => "31/xor-with %" reg " " reg2 "/r32" var/reg <- xor var2 => "33/xor *(ebp+" var2.stack-offset ") " reg "/r32" var/reg <- xor *var2/reg2 => "33/xor *" reg2 " " reg "/r32" xor-with var1, var2/reg => "31/xor-with *(ebp+" var1.stack-offset ") " reg "/r32" xor-with *var1/reg1, var2/reg2 => "31/xor-with *" reg1 " " reg2 "/r32" var/eax <- xor n => "35/xor-with-eax " n "/imm32" var/reg <- xor n => "81 6/subop/xor %" reg " " n "/i
#!/bin/bash
# Helper to change the numerical prefixes across the repo, say if you want to
# create room between 023 and 024, and so on.
#
# Assumes there's only ever one file with any numeric prefix. If you move
# 003trace.test.cc you might need to do some manual patch-up.

set -e

if [[ $# -eq 0 && `git diff HEAD |wc -l` -gt 0 ]]
then
  echo "Uncommitted changes"
  exit
fi

if [[ $# -gt 0 ]] # dry run
then
  git() {
    echo $*
  }
fi

#

index=0
ls [0-9]* |grep -v "trace.test" |sort -n |
  while read file
  do
    while [[ $file != `printf "%03d" $index`* ]]
    do
      echo
      index=$(($index+1))
    done
    echo $file
    index=$(($index+1))
  done > .layout

vim -c "set nu" .layout

#

root() {
  echo $1 |sed 's/^[0-9]*//'
}

index=0
cat .layout |
  while read file
  do
    if [ ! -z $file ]
    then
      newfile=`printf "%03d" $index``root $file`
      if [[ $newfile != $file ]]
      then
        echo git mv $file $newfile
        git mv $file $newfile
      fi
    fi
    index=$(($index+1))
  done

rm .layout

# Scenarios considered:
#   Don't redo the layout if Vim exits with error.
ons only update a subset of flags. ## Returns The `return` instruction cleans up variable declarations just like an unconditional `jump` to end of function, but also emits a series of copies before the final `jump`, copying each argument of `return` to the register appropriate to the respective function output. This doesn't work if a function output register contains a later `return` argument (e.g. if the registers for two outputs are swapped in `return`), so you can't do that. return => "c3/return" --- In the following instructions types are provided for clarity even if they must be provided in an earlier 'var' declaration. # Address operations var/reg: (addr T) <- address var2: T => "8d/copy-address *(ebp+" var2.stack-offset ") " reg "/r32" # Array operations var/reg: (addr T) <- index arr/rega: (addr array T), idx/regi: int | if size-of(T) is 1, 2, 4 or 8 => "(__check-mu-array-bounds *" rega " %" regi " " size-of(T) ")" "8d/copy-address *(" rega "+" regi "<<" log2(size-of(T)) "+4) " reg "/r32" var/reg: (addr T) <- index arr: (array T len), idx/regi: int => "(__check-mu-array-bounds *(ebp+" arr.stack-offset ") %" regi " " size-of(T) ")" "8d/copy-address *(ebp+" regi "<<" log2(size-of(T)) "+" (arr.stack-offset + 4) ") " reg "/r32" var/reg: (addr T) <- index arr/rega: (addr array T), n => "(__check-mu-array-bounds *" rega " " n " " size-of(T) ")" "8d/copy-address *(" rega "+" (n*size-of(T)+4) ") " reg "/r32" var/reg: (addr T) <- index arr: (array T len), n => "(__check-mu-array-bounds *(ebp+" arr.stack-offset ") " n " " size-of(T) ")" "8d/copy-address *(ebp+" (arr.stack-offset+4+n*size-of(T)) ") " reg "/r32" var/reg: (offset T) <- compute-offset arr: (addr array T), idx/regi: int # arr can be in reg or mem => "69/multiply %" regi " " size-of(T) "/imm32 " reg "/r32" var/reg: (offset T) <- compute-offset arr: (addr array T), idx: int # arr can be in reg or mem => "69/multiply *(ebp+" idx.stack-offset ") " size-of(T) "/imm32 " reg "/r32" var/reg: (addr T) <- index arr/rega: (addr array T), o/rego: (offset T) => "(__check-mu-array-bounds %" rega " %" rego " 1 \"" function-name "\")" "8d/copy-address *(" rega "+" rego "+4) " reg "/r32" Computing the length of an array is complex. var/reg: int <- length arr/reg2: (addr array T) | if T is byte (TODO) => "8b/-> *" reg2 " " reg "/r32" | if size-of(T) is 4 or 8 or 16 or 32 or 64 or 128 => "8b/-> *" reg2 " " reg "/r32" "c1/shift 5/subop/logic-right %" reg " " log2(size-of(T)) "/imm8" | otherwise x86 has no instruction to divide by a literal, so we need up to 3 extra registers! eax/edx for division and say ecx => if reg is not eax "50/push-eax" if reg is not ecx "51/push-ecx" if reg is not edx "52/push-edx" "8b/-> *" reg2 " eax/r32" "31/xor %edx 2/r32/edx" # sign-extend, but array size can't be negative "b9/copy-to-ecx " size-of(T) "/imm32" "f7 7/subop/idiv-eax-edx-by %ecx" if reg is not eax "89/<- %" reg " 0/r32/eax" if reg is not edx "5a/pop-to-edx" if reg is not ecx "59/pop-to-ecx" if reg is not eax "58/pop-to-eax" # User-defined types If a record (product) type T was defined to have elements a, b, c, ... of types T_a, T_b, T_c, ..., then accessing one of those elements f of type T_f: var/reg: (addr T_f) <- get var2/reg2: (addr T), f => "8d/copy-address *(" reg2 "+" offset(f) ") " reg "/r32" var/reg: (addr T_f) <- get var2: T, f => "8d/copy-address *(ebp+" var2.stack-offset "+" offset(f) ") " reg "/r32" # Allocating memory allocate in: (addr handle T) => "(allocate Heap " size-of(T) " " in ")" populate in: (addr handle array T), num # can be literal or variable on stack or register => "(allocate-array2 Heap " size-of(T) " " num " " in ")" populate-stream in: (addr handle stream T), num # can be literal or variable on stack or register => "(new-stream Heap " size-of(T) " " num " " in ")" # Some miscellaneous helpers to avoid error-prone size computations clear x: (addr T) => "(zero-out " s " " size-of(T) ")" read-from-stream s: (addr stream T), out: (addr T) => "(read-from-stream " s " " out " " size-of(T) ")" write-to-stream s: (addr stream T), in: (addr T) => "(write-to-stream " s " " in " " size-of(T) ")" vim:ft=mu:nowrap:textwidth=0