|
|
## Mu's instructions and their table-driven translation
See http://akkartik.name/akkartik-convivial-20200315.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 inputs.
The following chart shows all the instruction forms supported by Mu, along with
the SubX instruction they're translated to.
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"
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"
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"
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"
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"
var/eax <- xor n => "35/xor-with-eax " n "/imm32"
var/reg <- xor n => "81 6/subop/xor %" reg " " n "/imm32"
xor-with var, n => "81 6/subop/xor *(ebp+" var.stack-offset ") " n "/imm32"
xor-with *var/reg, n => "81 6/subop/xor *" reg " " n "/imm32"
var/eax <- copy n => "b8/copy-to-eax " n "/imm32"
var/ecx <- copy n => "b9/copy-to-ecx " n "/imm32"
var/edx <- copy n => "ba/copy-to-edx " n "/imm32"
var/ebx <- copy n => "bb/copy-to-ebx " n "/imm32"
var/esi <- copy n => "be/copy-to-esi " n "/imm32"
var/edi <- copy n => "bf/copy-to-edi " n "/imm32"
var/reg <- copy var2/reg2 => "89/<- %" reg " " reg2 "/r32"
copy-to var1, var2/reg => "89/<- *(ebp+" var1.stack-offset ") " reg "/r32"
var/reg <- copy var2 => "8b/-> *(ebp+" var2.stack-offset ") " reg "/r32"
var/reg <- copy *var2/reg2 => "8b/-> *" reg2 " " reg "/r32"
var/reg <- copy n => "c7 0/subop/copy %" reg " " n "/imm32"
copy-to var, n => "c7 0/subop/copy *(ebp+" var.stack-offset ") " n "/imm32"
copy-to *var/reg, n => "c7 0/subop/copy *" reg " " n "/imm32"
compare var1, var2/reg2 => "39/compare *(ebp+" var1.stack-offset ") " reg2 "/r32"
compare *var1/reg1, var2/reg2 => "39/compare *" reg1 " " reg2 "/r32"
compare var1/reg1, var2 => "3b/compare<- *(ebp+" var2.stack-offset ") " reg1 "/r32"
compare var/reg, *var2/reg2 => "3b/compare<- *" reg " " n "/imm32"
compare var/eax, n => "3d/compare-eax-with " n "/imm32"
compare var, n => "81 7/subop/compare *(ebp+" var.stack-offset ") " n "/imm32"
compare *var/reg, n => "81 7/subop/compare *" reg " " n "/imm32"
var/reg <- multiply var2 => "0f af/multiply *(ebp+" var2.stack-offset ") " reg "/r32"
var/reg <- multiply *var2/reg2 => "0f af/multiply *" reg2 " " reg "/r32"
break => "e9/jump break/disp32"
break label => "e9/jump " label ":break/disp32"
loop => "e9/jump loop/disp32"
loop label => "e9/jump " label ":loop/disp32"
break-if-= => "0f 84/jump-if-= break/disp32"
break-if-= label => "0f 84/jump-if-= " label ":break/disp32"
loop-if-= => "0f 84/jump-if-= loop/disp32"
loop-if-= label => "0f 84/jump-if-= " label ":loop/disp32"
break-if-!= => "0f 85/jump-if-!= break/disp32"
break-if-!= label => "0f 85/jump-if-!= " label ":break/disp32"
loop-if-!= => "0f 85/jump-if-!= loop/disp32"
loop-if-!= label => "0f 85/jump-if-!= " label ":loop/disp32"
break-if-< => "0f 8c/jump-if-< break/disp32"
break-if-< label => "0f 8c/jump-if-< " label ":break/disp32"
loop-if-< => "0f 8c/jump-if-< loop/disp32"
loop-if-< label => "0f 8c/jump-if-< " label ":loop/disp32"
break-if-> => "0f 8f/jump-if-> break/disp32"
break-if-> label => "0f 8f/jump-if-> " label ":break/disp32"
loop-if-> => "0f 8f/jump-if-> loop/disp32"
loop-if-> label => "0f 8f/jump-if-> " label ":loop/disp32"
break-if-<= => "0f 8e/jump-if-<= break/disp32"
break-if-<= label => "0f 8e/jump-if-<= " label ":break/disp32"
loop-if-<= => "0f 8e/jump-if-<= loop/disp32"
loop-if-<= label => "0f 8e/jump-if-<= " label ":loop/disp32"
break-if->= => "0f 8d/jump-if->= break/disp32"
break-if->= label => "0f 8d/jump-if->= " label ":break/disp32"
loop-if->= => "0f 8d/jump-if->= loop/disp32"
loop-if->= label => "0f 8d/jump-if->= " label ":loop/disp32"
break-if-addr< => "0f 82/jump-if-addr< break/disp32"
break-if-addr< label => "0f 82/jump-if-addr< " label ":break/disp32"
loop-if-addr< => "0f 82/jump-if-addr< loop/disp32"
loop-if-addr< label => "0f 82/jump-if-addr< " label ":loop/disp32"
break-if-addr> => "0f 87/jump-if-addr> break/disp32"
break-if-addr> label => "0f 87/jump-if-addr> " label ":break/disp32"
loop-if-addr> => "0f 87/jump-if-addr> loop/disp32"
loop-if-addr> label => "0f 87/jump-if-addr> " label ":loop/disp32"
break-if-addr<= => "0f 86/jump-if-addr<= break/disp32"
break-if-addr<= label => "0f 86/jump-if-addr<= " label ":break/disp32"
loop-if-addr<= => "0f 86/jump-if-addr<= loop/disp32"
loop-if-addr<= label => "0f 86/jump-if-addr<= " label ":loop/disp32"
break-if-addr>= => "0f 83/jump-if-addr>= break/disp32"
break-if-addr>= label => "0f 83/jump-if-addr>= " label ":break/disp32"
loop-if-addr>= => "0f 83/jump-if-addr>= loop/disp32"
loop-if-addr>= label => "0f 83/jump-if-addr>= " label ":loop/disp32"
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 (TODO: bounds-checking)
var/reg <- index arr/rega: (addr array T), idx/regi: int
| if size-of(T) is 4 or 8
=> "8d/copy-address *(" rega "+" regi "<<" log2(size-of(T)) "+4) " reg "/r32"
var/reg <- index arr: (array T sz), idx/regi: int
=> "8d/copy-address *(ebp+" regi "<<" log2(size-of(T)) "+" (arr.stack-offset + 4) ") " reg "/r32"
var/reg <- index arr/rega: (addr array T), n
=> "8d/copy-address *(" rega "+" (n*size-of(T)+4) ") " reg "/r32"
var/reg <- index arr: (array T sz), n
=> "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 <- index arr/rega: (addr array T), o/rego: offset
=> "8d/copy-address *(" rega "+" rego "+4) " reg "/r32"
Computing the length of an array is complex.
var/reg <- 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 F), f
=> "8d/copy-address *(" reg2 "+" offset(f) ") " reg "/r32"
var/reg: (addr T_f) <- get var2: (addr F), f
=> "8d/copy-address *(ebp+" var2.stack-offset "+" offset(f) ") " reg "/r32"
# Handles for safe access to the heap
copy-handle-to dest: (handle T), src: (handle T)
=> "50/push-eax"
"8b/-> *(ebp+" src.stack-offset ") 0/r32/eax"
"89/<- *(ebp+" dest.stack-offset ") 0/r32/eax"
"8b/-> *(ebp+" src.stack-offset+4 ") 0/r32/eax"
"89/<- *(ebp+" dest.stack-offset+4 ") 0/r32/eax"
"58/pop-to-eax"
copy-handle-to *dest/reg: (addr handle T), src: (handle T)
=> "50/push-eax"
"8b/-> *(ebp+" src.stack-offset ") 0/r32/eax"
"89/<- *" reg " 0/r32/eax"
"8b/-> *(ebp+" src.stack-offset+4 ") 0/r32/eax"
"89/<- *(" reg "+4) 0/r32/eax"
"58/pop-to-eax"
out/reg: (addr T) <- lookup in: (handle T)
=> # payload_allocid = in->address->allocid
"8b/-> *(epb+" (in.stack-offset+4) ") " reg "/r32"
"8b/-> *" reg " " reg "/r32"
# if (payload_allocid != handle->allocid) abort
"39/compare *(ebp+" in.stack-offset ") " reg "/r32"
"0f 85/jump-if-!= $lookup:abort/disp32"
# return payload
"8b/-> *(epb+" (in.stack-offset+4) ") " reg "/r32"
"81 0/subop/add %" reg " 4/imm32" # skip payload->allocid
vim:ft=mu:nowrap:textwidth=0
|