2020-02-01 02:39:27 +00:00
|
|
|
## Mu's instructions and their table-driven translation
|
|
|
|
|
|
|
|
Mu is a statement-oriented language. Blocks consist of flat lists of instructions.
|
|
|
|
The chart at the bottom of this page shows all the instruction forms supported
|
|
|
|
by Mu, one to a line. Each line shows an example of the instruction on the
|
|
|
|
left. Past the first column, everything inside the {..} is a summary of the
|
|
|
|
data structure the Mu compiler uses (`Primitives` in apps/mu.subx) to translate
|
|
|
|
it.
|
|
|
|
|
|
|
|
The syntax of the data structure is intended to be similar to C++'s aggregate
|
|
|
|
initializers (https://en.cppreference.com/w/cpp/language/aggregate_initialization)
|
|
|
|
However, there are differences:
|
|
|
|
- We use adjacency for string concatenation.
|
|
|
|
- We use [] for array literals.
|
|
|
|
- The objects inside [] are not fully described. They include various
|
|
|
|
metadata about the variable in the instruction. For our purposes, assume
|
|
|
|
that variables on the stack have a stack offset computed for them, and
|
|
|
|
register variables evaluate to their register.
|
|
|
|
- registers may be specified by name: /eax /ecx /edx /ebx /esi /edi
|
|
|
|
- registers may be specified as a wildcard: /reg
|
|
|
|
- integer literals are always called 'n'
|
|
|
|
- any other variable names that don't specify a register are assumed to be on the stack
|
|
|
|
|
|
|
|
There are no checks for types yet, because Mu programs only have `int` types so far.
|
|
|
|
|
|
|
|
Example 1 (use the widest screen you can for this page):
|
|
|
|
-- instruction form -- | -------------------------- data structure ----------------------------
|
|
|
|
|<------------- pattern matching ---------->|<--- code generation ------------------->
|
6158 - standardize opcode names
At the lowest level, SubX without syntax sugar uses names without prepositions.
For example, 01 and 03 are both called 'add', irrespective of source and
destination operand. Horizontal space is at a premium, and we rely on the
comments at the end of each line to fully describe what is happening.
Above that, however, we standardize on a slightly different naming convention
across:
a) SubX with syntax sugar,
b) Mu, and
c) the SubX code that the Mu compiler emits.
Conventions, in brief:
- by default, the source is on the left and destination on the right.
e.g. add %eax, 1/r32/ecx ("add eax to ecx")
- prepositions reverse the direction.
e.g. add-to %eax, 1/r32/ecx ("add ecx to eax")
subtract-from %eax, 1/r32/ecx ("subtract ecx from eax")
- by default, comparisons are left to right while 'compare<-' reverses.
Before, I was sometimes swapping args to make the operation more obvious,
but that would complicate the code-generation of the Mu compiler, and it's
nice to be able to read the output of the compiler just like hand-written
code.
One place where SubX differs from Mu: copy opcodes are called '<-' and
'->'. Hopefully that fits with the spirit of Mu rather than the letter
of the 'copy' and 'copy-to' instructions.
2020-03-21 22:32:30 +00:00
|
|
|
var/reg <- add var2/reg {.name="add", .inouts=[reg], .outputs=[reg], .subx-name="01/add-to", .rm32=outputs[0], .r32=inouts[0]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
|
|
|
Read this as:
|
|
|
|
if an instruction's name is "add"
|
|
|
|
and it has one inout that's in a register
|
|
|
|
and it has one output that's in a register,
|
|
|
|
then emit the following on a single line
|
6158 - standardize opcode names
At the lowest level, SubX without syntax sugar uses names without prepositions.
For example, 01 and 03 are both called 'add', irrespective of source and
destination operand. Horizontal space is at a premium, and we rely on the
comments at the end of each line to fully describe what is happening.
Above that, however, we standardize on a slightly different naming convention
across:
a) SubX with syntax sugar,
b) Mu, and
c) the SubX code that the Mu compiler emits.
Conventions, in brief:
- by default, the source is on the left and destination on the right.
e.g. add %eax, 1/r32/ecx ("add eax to ecx")
- prepositions reverse the direction.
e.g. add-to %eax, 1/r32/ecx ("add ecx to eax")
subtract-from %eax, 1/r32/ecx ("subtract ecx from eax")
- by default, comparisons are left to right while 'compare<-' reverses.
Before, I was sometimes swapping args to make the operation more obvious,
but that would complicate the code-generation of the Mu compiler, and it's
nice to be able to read the output of the compiler just like hand-written
code.
One place where SubX differs from Mu: copy opcodes are called '<-' and
'->'. Hopefully that fits with the spirit of Mu rather than the letter
of the 'copy' and 'copy-to' instructions.
2020-03-21 22:32:30 +00:00
|
|
|
"01/add-to" (the opcode or subx-name)
|
2020-02-01 02:39:27 +00:00
|
|
|
"%{reg}", interpolating the output's register
|
|
|
|
"{reg}/r32", interpolating the inout's register code.
|
|
|
|
|
|
|
|
Example 2:
|
|
|
|
-- instruction form -- | -------------------------- data structure ----------------------------
|
|
|
|
|<------- pattern matching ------>|<--- code generation ------------------->
|
|
|
|
add-to var, n {.name="add-to", .inouts=[var, n], .subx-name="81 0/subop/add", .rm32="*(ebp+" inouts[0].stack-offset ")", .imm32=inouts[1]}
|
|
|
|
|
|
|
|
Read this as:
|
|
|
|
if an instruction's name is "add-to"
|
|
|
|
and it has two inouts
|
|
|
|
the first on the stack
|
|
|
|
and the second a literal,
|
|
|
|
then emit the following on a single line
|
|
|
|
"81 0/subop/add" (the opcode or subx-name)
|
|
|
|
"*(ebp+{stack})", interpolating the first inout's stack offset
|
|
|
|
"{n}/imm32", interpolating the second inout's contents
|
|
|
|
|
|
|
|
Ok, here's the complete chart.
|
|
|
|
|
|
|
|
-- instruction form -- | -------------------------- data structure ----------------------------
|
|
|
|
|<------------------- pattern matching ------------------->|<---- code generation ------------------->
|
|
|
|
var/eax <- increment {.name="increment", .outputs=[eax], .subx-name="40/increment-eax"}
|
|
|
|
var/ecx <- increment {.name="increment", .outputs=[ecx], .subx-name="41/increment-ecx"}
|
|
|
|
var/edx <- increment {.name="increment", .outputs=[edx], .subx-name="42/increment-edx"}
|
|
|
|
var/ebx <- increment {.name="increment", .outputs=[ebx], .subx-name="43/increment-ebx"}
|
|
|
|
var/esi <- increment {.name="increment", .outputs=[esi], .subx-name="46/increment-esi"}
|
|
|
|
var/edi <- increment {.name="increment", .outputs=[edi], .subx-name="47/increment-edi"}
|
|
|
|
increment var {.name="increment", .inouts=[var], .subx-name="ff 0/subop/increment", .rm32="*(ebp+" inouts[0].stack-offset ")"}
|
2020-02-21 07:56:11 +00:00
|
|
|
increment *var/reg {.name="increment", .inouts=[reg], .subx-name="ff 0/subop/increment", .rm32="*" inouts[0]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
|
|
|
var/eax <- decrement {.name="decrement", .outputs=[eax], .subx-name="48/decrement-eax"}
|
|
|
|
var/ecx <- decrement {.name="decrement", .outputs=[ecx], .subx-name="49/decrement-ecx"}
|
|
|
|
var/edx <- decrement {.name="decrement", .outputs=[edx], .subx-name="4a/decrement-edx"}
|
|
|
|
var/ebx <- decrement {.name="decrement", .outputs=[ebx], .subx-name="4b/decrement-ebx"}
|
|
|
|
var/esi <- decrement {.name="decrement", .outputs=[esi], .subx-name="4e/decrement-esi"}
|
|
|
|
var/edi <- decrement {.name="decrement", .outputs=[edi], .subx-name="4f/decrement-edi"}
|
|
|
|
decrement var {.name="decrement", .inouts=[var], .subx-name="ff 1/subop/decrement", .rm32="*(ebp+" inouts[0].stack-offset ")"}
|
2020-02-21 07:56:11 +00:00
|
|
|
decrement *var/reg {.name="decrement", .inouts=[reg], .subx-name="ff 1/subop/decrement", .rm32="*" inouts[0]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
6158 - standardize opcode names
At the lowest level, SubX without syntax sugar uses names without prepositions.
For example, 01 and 03 are both called 'add', irrespective of source and
destination operand. Horizontal space is at a premium, and we rely on the
comments at the end of each line to fully describe what is happening.
Above that, however, we standardize on a slightly different naming convention
across:
a) SubX with syntax sugar,
b) Mu, and
c) the SubX code that the Mu compiler emits.
Conventions, in brief:
- by default, the source is on the left and destination on the right.
e.g. add %eax, 1/r32/ecx ("add eax to ecx")
- prepositions reverse the direction.
e.g. add-to %eax, 1/r32/ecx ("add ecx to eax")
subtract-from %eax, 1/r32/ecx ("subtract ecx from eax")
- by default, comparisons are left to right while 'compare<-' reverses.
Before, I was sometimes swapping args to make the operation more obvious,
but that would complicate the code-generation of the Mu compiler, and it's
nice to be able to read the output of the compiler just like hand-written
code.
One place where SubX differs from Mu: copy opcodes are called '<-' and
'->'. Hopefully that fits with the spirit of Mu rather than the letter
of the 'copy' and 'copy-to' instructions.
2020-03-21 22:32:30 +00:00
|
|
|
var1/reg1 <- add var2/reg2 {.name="add", .inouts=[reg2], .outputs=[reg1], .subx-name="01/add-to", .rm32=outputs[0], .r32=inouts[0]}
|
|
|
|
var/reg <- add var2 {.name="add", .inouts=[var2], .outputs=[reg], .subx-name="03/add", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=outputs[0]}
|
|
|
|
var/reg <- add *var2/reg2 {.name="add", .inouts=[reg2], .outputs=[reg], .subx-name="03/add", .rm32="*" inouts[0], .r32=outputs[0]}
|
|
|
|
add-to var1, var2/reg {.name="add-to", .inouts=[var1, var2], .subx-name="01/add-to", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
var/eax <- add n {.name="add", .inouts=[n], .outputs=[eax], .subx-name="05/add-to-eax", .imm32=inouts[0]}
|
|
|
|
var/reg <- add n {.name="add", .inouts=[n], .outputs=[reg], .subx-name="81 0/subop/add", .rm32=outputs[0], .imm32=inouts[0]}
|
|
|
|
add-to var, n {.name="add-to", .inouts=[var, n], .subx-name="81 0/subop/add", .rm32="*(ebp+" inouts[0].stack-offset ")", .imm32=inouts[1]}
|
2020-02-21 07:56:11 +00:00
|
|
|
add-to *var/reg, n {.name="add-to", .inouts=[reg, n], .subx-name="81 0/subop/add", .rm32="*" inouts[0], .imm32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
6158 - standardize opcode names
At the lowest level, SubX without syntax sugar uses names without prepositions.
For example, 01 and 03 are both called 'add', irrespective of source and
destination operand. Horizontal space is at a premium, and we rely on the
comments at the end of each line to fully describe what is happening.
Above that, however, we standardize on a slightly different naming convention
across:
a) SubX with syntax sugar,
b) Mu, and
c) the SubX code that the Mu compiler emits.
Conventions, in brief:
- by default, the source is on the left and destination on the right.
e.g. add %eax, 1/r32/ecx ("add eax to ecx")
- prepositions reverse the direction.
e.g. add-to %eax, 1/r32/ecx ("add ecx to eax")
subtract-from %eax, 1/r32/ecx ("subtract ecx from eax")
- by default, comparisons are left to right while 'compare<-' reverses.
Before, I was sometimes swapping args to make the operation more obvious,
but that would complicate the code-generation of the Mu compiler, and it's
nice to be able to read the output of the compiler just like hand-written
code.
One place where SubX differs from Mu: copy opcodes are called '<-' and
'->'. Hopefully that fits with the spirit of Mu rather than the letter
of the 'copy' and 'copy-to' instructions.
2020-03-21 22:32:30 +00:00
|
|
|
var1/reg1 <- sub var2/reg2 {.name="sub", .inouts=[reg2], .outputs=[reg1], .subx-name="29/subtract-from", .rm32=outputs[0], .r32=inouts[0]}
|
|
|
|
var/reg <- sub var2 {.name="sub", .inouts=[var2], .outputs=[reg], .subx-name="2b/subtract", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=outputs[0]}
|
|
|
|
var/reg <- sub *var2/reg2 {.name="sub", .inouts=[reg2], .outputs=[reg], .subx-name="2b/subtract", .rm32="*" inouts[0], .r32=outputs[0]}
|
|
|
|
sub-from var1, var2/reg {.name="sub-from", .inouts=[var1, var2], .subx-name="29/subtract-from", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=inouts[1]}
|
|
|
|
var/eax <- sub n {.name="sub", .inouts=[n], .outputs=[eax], .subx-name="2d/subtract-from-eax", .imm32=inouts[0]}
|
2020-02-01 02:39:27 +00:00
|
|
|
var/reg <- sub n {.name="sub", .inouts=[n], .outputs=[reg], .subx-name="81 5/subop/subtract", .rm32=outputs[0], .imm32=inouts[0]}
|
|
|
|
sub-from var, n {.name="sub-from", .inouts=[var, n], .subx-name="81 5/subop/subtract", .rm32="*(ebp+" inouts[0].stack-offset ")", .imm32=inouts[1]}
|
2020-02-21 07:56:11 +00:00
|
|
|
sub-from *var/reg, n {.name="sub-from", .inouts=[reg, n], .subx-name="81 5/subop/subtract", .rm32="*" inouts[0], .imm32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
6158 - standardize opcode names
At the lowest level, SubX without syntax sugar uses names without prepositions.
For example, 01 and 03 are both called 'add', irrespective of source and
destination operand. Horizontal space is at a premium, and we rely on the
comments at the end of each line to fully describe what is happening.
Above that, however, we standardize on a slightly different naming convention
across:
a) SubX with syntax sugar,
b) Mu, and
c) the SubX code that the Mu compiler emits.
Conventions, in brief:
- by default, the source is on the left and destination on the right.
e.g. add %eax, 1/r32/ecx ("add eax to ecx")
- prepositions reverse the direction.
e.g. add-to %eax, 1/r32/ecx ("add ecx to eax")
subtract-from %eax, 1/r32/ecx ("subtract ecx from eax")
- by default, comparisons are left to right while 'compare<-' reverses.
Before, I was sometimes swapping args to make the operation more obvious,
but that would complicate the code-generation of the Mu compiler, and it's
nice to be able to read the output of the compiler just like hand-written
code.
One place where SubX differs from Mu: copy opcodes are called '<-' and
'->'. Hopefully that fits with the spirit of Mu rather than the letter
of the 'copy' and 'copy-to' instructions.
2020-03-21 22:32:30 +00:00
|
|
|
var1/reg1 <- and var2/reg2 {.name="and", .inouts=[reg2], .outputs=[reg1], .subx-name="21/and-with", .rm32=outputs[0], .r32=inouts[0]}
|
|
|
|
var/reg <- and var2 {.name="and", .inouts=[var2], .outputs=[reg], .subx-name="23/and", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=outputs[0]}
|
|
|
|
var/reg <- and *var2/reg2 {.name="and", .inouts=[reg2], .outputs=[reg], .subx-name="23/and", .rm32="*" inouts[0], .r32=outputs[0]}
|
|
|
|
and-with var1, var2/reg {.name="and-with", .inouts=[var1, reg], .subx-name="21/and-with", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
var/eax <- and n {.name="and", .inouts=[n], .outputs=[eax], .subx-name="25/and-with-eax", .imm32=inouts[0]}
|
|
|
|
var/reg <- and n {.name="and", .inouts=[n], .outputs=[reg], .subx-name="81 4/subop/and", .rm32=outputs[0], .imm32=inouts[0]}
|
|
|
|
and-with var, n {.name="and-with", .inouts=[var, n], .subx-name="81 4/subop/and", .rm32="*(ebp+" inouts[0].stack-offset ")", .imm32=inouts[1]}
|
2020-02-21 07:56:11 +00:00
|
|
|
and-with *var/reg, n {.name="and-with", .inouts=[reg, n], .subx-name="81 4/subop/and", .rm32="*" inouts[0], .imm32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
6158 - standardize opcode names
At the lowest level, SubX without syntax sugar uses names without prepositions.
For example, 01 and 03 are both called 'add', irrespective of source and
destination operand. Horizontal space is at a premium, and we rely on the
comments at the end of each line to fully describe what is happening.
Above that, however, we standardize on a slightly different naming convention
across:
a) SubX with syntax sugar,
b) Mu, and
c) the SubX code that the Mu compiler emits.
Conventions, in brief:
- by default, the source is on the left and destination on the right.
e.g. add %eax, 1/r32/ecx ("add eax to ecx")
- prepositions reverse the direction.
e.g. add-to %eax, 1/r32/ecx ("add ecx to eax")
subtract-from %eax, 1/r32/ecx ("subtract ecx from eax")
- by default, comparisons are left to right while 'compare<-' reverses.
Before, I was sometimes swapping args to make the operation more obvious,
but that would complicate the code-generation of the Mu compiler, and it's
nice to be able to read the output of the compiler just like hand-written
code.
One place where SubX differs from Mu: copy opcodes are called '<-' and
'->'. Hopefully that fits with the spirit of Mu rather than the letter
of the 'copy' and 'copy-to' instructions.
2020-03-21 22:32:30 +00:00
|
|
|
var1/reg1 <- or var2/reg2 {.name="or", .inouts=[reg2], .outputs=[reg1], .subx-name="09/or-with", .rm32=outputs[0], .r32=inouts[0]}
|
|
|
|
var/reg <- or var2 {.name="or", .inouts=[var2], .outputs=[reg], .subx-name="0b/or", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=outputs[0]}
|
|
|
|
var/reg <- or *var2/reg2 {.name="or", .inouts=[reg2], .outputs=[reg], .subx-name="0b/or", .rm32="*" inouts[0], .r32=outputs[0]}
|
|
|
|
or-with var1, var2/reg {.name="or-with", .inouts=[var1, reg], .subx-name="09/or-with", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
var/eax <- or n {.name="or", .inouts=[n], .outputs=[eax], .subx-name="0d/or-with-eax", .imm32=inouts[0]}
|
|
|
|
var/reg <- or n {.name="or", .inouts=[n], .outputs=[reg], .subx-name="81 1/subop/or", .rm32=outputs[0], .imm32=inouts[0]}
|
|
|
|
or-with var, n {.name="or-with", .inouts=[var, n], .subx-name="81 1/subop/or", .rm32="*(ebp+" inouts[0].stack-offset ")", .imm32=inouts[1]}
|
2020-02-21 07:56:11 +00:00
|
|
|
or-with *var/reg, n {.name="or-with", .inouts=[reg, n], .subx-name="81 1/subop/or", .rm32="*" inouts[0], .imm32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
6158 - standardize opcode names
At the lowest level, SubX without syntax sugar uses names without prepositions.
For example, 01 and 03 are both called 'add', irrespective of source and
destination operand. Horizontal space is at a premium, and we rely on the
comments at the end of each line to fully describe what is happening.
Above that, however, we standardize on a slightly different naming convention
across:
a) SubX with syntax sugar,
b) Mu, and
c) the SubX code that the Mu compiler emits.
Conventions, in brief:
- by default, the source is on the left and destination on the right.
e.g. add %eax, 1/r32/ecx ("add eax to ecx")
- prepositions reverse the direction.
e.g. add-to %eax, 1/r32/ecx ("add ecx to eax")
subtract-from %eax, 1/r32/ecx ("subtract ecx from eax")
- by default, comparisons are left to right while 'compare<-' reverses.
Before, I was sometimes swapping args to make the operation more obvious,
but that would complicate the code-generation of the Mu compiler, and it's
nice to be able to read the output of the compiler just like hand-written
code.
One place where SubX differs from Mu: copy opcodes are called '<-' and
'->'. Hopefully that fits with the spirit of Mu rather than the letter
of the 'copy' and 'copy-to' instructions.
2020-03-21 22:32:30 +00:00
|
|
|
var1/reg1 <- xor var2/reg2 {.name="xor", .inouts=[reg2], .outputs=[reg1], .subx-name="31/xor-with", .rm32=outputs[0], .r32=inouts[0]}
|
|
|
|
var/reg <- xor var2 {.name="xor", .inouts=[var2], .outputs=[reg], .subx-name="33/xor", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=outputs[0]}
|
|
|
|
var/reg <- xor *var2/reg2 {.name="xor", .inouts=[reg2], .outputs=[reg], .subx-name="33/xor", .rm32="*" inouts[0], .r32=outputs[0]}
|
|
|
|
xor-with var1, var2/reg {.name="xor-with", .inouts=[var1, reg], .subx-name="31/xor-with", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
var/eax <- xor n {.name="xor", .inouts=[n], .outputs=[eax], .subx-name="35/xor-with-eax", .imm32=inouts[0]}
|
|
|
|
var/reg <- xor n {.name="xor", .inouts=[n], .outputs=[reg], .subx-name="81 6/subop/xor", .rm32=outputs[0], .imm32=inouts[0]}
|
|
|
|
xor-with var, n {.name="xor-with", .inouts=[var, n], .subx-name="81 6/subop/xor", .rm32="*(ebp+" inouts[0].stack-offset ")", .imm32=inouts[1]}
|
2020-02-21 07:56:11 +00:00
|
|
|
xor-with *var/reg, n {.name="xor-with", .inouts=[reg, n], .subx-name="81 6/subop/xor", .rm32="*" inouts[0], .imm32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
|
|
|
var/eax <- copy n {.name="copy", .inouts=[n], .outputs=[eax], .subx-name="b8/copy-to-eax", .imm32=inouts[0]}
|
|
|
|
var/ecx <- copy n {.name="copy", .inouts=[n], .outputs=[ecx], .subx-name="b9/copy-to-ecx", .imm32=inouts[0]}
|
|
|
|
var/edx <- copy n {.name="copy", .inouts=[n], .outputs=[edx], .subx-name="ba/copy-to-edx", .imm32=inouts[0]}
|
|
|
|
var/ebx <- copy n {.name="copy", .inouts=[n], .outputs=[ebx], .subx-name="bb/copy-to-ebx", .imm32=inouts[0]}
|
|
|
|
var/esi <- copy n {.name="copy", .inouts=[n], .outputs=[esi], .subx-name="be/copy-to-esi", .imm32=inouts[0]}
|
|
|
|
var/edi <- copy n {.name="copy", .inouts=[n], .outputs=[edi], .subx-name="bf/copy-to-edi", .imm32=inouts[0]}
|
6158 - standardize opcode names
At the lowest level, SubX without syntax sugar uses names without prepositions.
For example, 01 and 03 are both called 'add', irrespective of source and
destination operand. Horizontal space is at a premium, and we rely on the
comments at the end of each line to fully describe what is happening.
Above that, however, we standardize on a slightly different naming convention
across:
a) SubX with syntax sugar,
b) Mu, and
c) the SubX code that the Mu compiler emits.
Conventions, in brief:
- by default, the source is on the left and destination on the right.
e.g. add %eax, 1/r32/ecx ("add eax to ecx")
- prepositions reverse the direction.
e.g. add-to %eax, 1/r32/ecx ("add ecx to eax")
subtract-from %eax, 1/r32/ecx ("subtract ecx from eax")
- by default, comparisons are left to right while 'compare<-' reverses.
Before, I was sometimes swapping args to make the operation more obvious,
but that would complicate the code-generation of the Mu compiler, and it's
nice to be able to read the output of the compiler just like hand-written
code.
One place where SubX differs from Mu: copy opcodes are called '<-' and
'->'. Hopefully that fits with the spirit of Mu rather than the letter
of the 'copy' and 'copy-to' instructions.
2020-03-21 22:32:30 +00:00
|
|
|
var1/reg1 <- copy var2/reg2 {.name="copy", .inouts=[reg2], .outputs=[reg1], .subx-name="89/<-", .rm32=outputs[0], .r32=inouts[0]}
|
|
|
|
copy-to var1, var2/reg {.name="copy-to", .inouts=[var1, var2], .subx-name="89/<-", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=inouts[1]}
|
|
|
|
var/reg <- copy var2 {.name="copy", .inouts=[var2], .outputs=[reg], .subx-name="8b/->", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=outputs[0]}
|
|
|
|
var1/reg <- copy *var2/reg2 {.name="copy", .inouts=[reg2], .outputs=[reg], .subx-name="8b/->", .rm32="*" inouts[0], .r32=outputs[0]}
|
2020-02-01 02:39:27 +00:00
|
|
|
var/reg <- copy n {.name="copy", .inouts=[n], .outputs=[reg], .subx-name="c7 0/subop/copy", .rm32=outputs[0], .imm32=inouts[0]}
|
|
|
|
copy-to var, n {.name="copy-to", .inouts=[var, n], .subx-name="c7 0/subop/copy", .rm32="*(ebp+" inouts[0].stack-offset ")", .imm32=inouts[1]}
|
2020-02-21 07:56:11 +00:00
|
|
|
copy-to *var/reg, n {.name="copy-to", .inouts=[reg, n], .subx-name="c7 0/subop/copy", .rm32="*" inouts[0], .imm32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
6158 - standardize opcode names
At the lowest level, SubX without syntax sugar uses names without prepositions.
For example, 01 and 03 are both called 'add', irrespective of source and
destination operand. Horizontal space is at a premium, and we rely on the
comments at the end of each line to fully describe what is happening.
Above that, however, we standardize on a slightly different naming convention
across:
a) SubX with syntax sugar,
b) Mu, and
c) the SubX code that the Mu compiler emits.
Conventions, in brief:
- by default, the source is on the left and destination on the right.
e.g. add %eax, 1/r32/ecx ("add eax to ecx")
- prepositions reverse the direction.
e.g. add-to %eax, 1/r32/ecx ("add ecx to eax")
subtract-from %eax, 1/r32/ecx ("subtract ecx from eax")
- by default, comparisons are left to right while 'compare<-' reverses.
Before, I was sometimes swapping args to make the operation more obvious,
but that would complicate the code-generation of the Mu compiler, and it's
nice to be able to read the output of the compiler just like hand-written
code.
One place where SubX differs from Mu: copy opcodes are called '<-' and
'->'. Hopefully that fits with the spirit of Mu rather than the letter
of the 'copy' and 'copy-to' instructions.
2020-03-21 22:32:30 +00:00
|
|
|
compare var1, var2/reg {.name="compare", .inouts=[var1, reg], .subx-name="39/compare", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=inouts[1]}
|
|
|
|
compare *var1/reg1, var/reg {.name="compare", .inouts=[reg1, reg], .subx-name="39/compare", .rm32="*" inouts[0], .r32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
compare var1/reg, var2 {.name="compare", .inouts=[reg, var2], .subx-name="3b/compare<-", .rm32="*(ebp+" inouts[1].stack-offset ")", .r32=inouts[0]}
|
2020-02-21 07:56:11 +00:00
|
|
|
compare var/reg, *var2/reg2 {.name="compare", .inouts=[reg, reg2], .subx-name="3b/compare<-", .rm32="*" inouts[1], .r32=inouts[0]}
|
2020-02-01 02:39:27 +00:00
|
|
|
compare var/eax, n {.name="compare", .inouts=[eax, n], .subx-name="3d/compare-eax-with", .imm32=inouts[1]}
|
|
|
|
compare var, n {.name="compare", .inouts=[var, n], .subx-name="81 7/subop/compare", .rm32="*(ebp+" inouts[0].stack-offset ")", .imm32=inouts[1]}
|
2020-02-21 07:56:11 +00:00
|
|
|
compare *var/reg, n {.name="compare", .inouts=[reg, n], .subx-name="81 7/subop/compare", .rm32="*" inouts[0], .imm32=inouts[1]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
|
|
|
var/reg <- multiply var2 {.name="multiply", .inouts=[var2], .outputs=[reg], .subx-name="0f af/multiply", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=outputs[0]}
|
2020-02-21 07:56:11 +00:00
|
|
|
var/reg <- multiply *var2/reg2 { .name="multiply", .inouts=[reg2], .outputs=[reg], .subx-name="0f af/multiply", .rm32="*" inouts[0], .r32=outputs[0]}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
|
|
|
Jumps have a slightly simpler format. Most of the time they take no inouts or
|
2020-02-01 20:14:12 +00:00
|
|
|
outputs. Occasionally you give them a label for a containing block to jump to
|
2020-02-28 01:03:08 +00:00
|
|
|
the start or end of. (Conditional branches read different combinations of CPU
|
|
|
|
flags.)
|
|
|
|
|
|
|
|
break {.name="break", .subx-name="e9/jump break/disp32"}
|
|
|
|
break label {.name="break", .inouts=[label], .subx-name="e9/jump", .disp32=inouts[0] ":break"}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
|
|
|
break-if-= {.name="break-if-=", .subx-name="0f 84/jump-if-= break/disp32"}
|
|
|
|
break-if-= label {.name="break-if-=", .inouts=[label], .subx-name="0f 84/jump-if-=", .disp32=inouts[0] ":break"}
|
|
|
|
break-if-!= {.name="break-if-!=", .subx-name="0f 85/jump-if-!= break/disp32"}
|
|
|
|
break-if-!= label {.name="break-if-!=", .inouts=[label], .subx-name="0f 85/jump-if-!=", .disp32=inouts[0] ":break"}
|
|
|
|
|
2020-02-01 20:14:12 +00:00
|
|
|
break-if-< {.name="break-if-<", .subx-name="0f 8c/jump-if-< break/disp32"}
|
|
|
|
break-if-< label {.name="break-if-<", .inouts=[label], .subx-name="0f 8c/jump-if-<", .disp32=inouts[0] ":break"}
|
|
|
|
break-if-> {.name="break-if->", .subx-name="0f 8f/jump-if-> break/disp32"}
|
|
|
|
break-if-> label {.name="break-if->", .inouts=[label], .subx-name="0f 8f/jump-if->", .disp32=inouts[0] ":break"}
|
|
|
|
break-if-<= {.name="break-if-<=", .subx-name="0f 8e/jump-if-<= break/disp32"}
|
|
|
|
break-if-<= label {.name="break-if-<=", .inouts=[label], .subx-name="0f 8e/jump-if-<=", .disp32=inouts[0] ":break"}
|
|
|
|
break-if->= {.name="break-if->=", .subx-name="0f 8d/jump-if->= break/disp32"}
|
|
|
|
break-if->= label {.name="break-if->=", .inouts=[label], .subx-name="0f 8d/jump-if->=", .disp32=inouts[0] ":break"}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
|
|
|
break-if-addr< {.name="break-if-addr<", .subx-name="0f 82/jump-if-addr< break/disp32"}
|
|
|
|
break-if-addr< label {.name="break-if-addr<", .inouts=[label], .subx-name="0f 82/jump-if-addr<", .disp32=inouts[0] ":break"}
|
|
|
|
break-if-addr> {.name="break-if-addr>", .subx-name="0f 87/jump-if-addr> break/disp32"}
|
|
|
|
break-if-addr> label {.name="break-if-addr>", .inouts=[label], .subx-name="0f 87/jump-if-addr>", .disp32=inouts[0] ":break"}
|
|
|
|
break-if-addr<= {.name="break-if-addr<=", .subx-name="0f 86/jump-if-addr<= break/disp32"}
|
|
|
|
break-if-addr<= label {.name="break-if-addr<=", .inouts=[label], .subx-name="0f 86/jump-if-addr<=", .disp32=inouts[0] ":break"}
|
|
|
|
break-if-addr>= {.name="break-if-addr>=", .subx-name="0f 83/jump-if-addr>= break/disp32"}
|
|
|
|
break-if-addr>= label {.name="break-if-addr>=", .inouts=[label], .subx-name="0f 83/jump-if-addr>=", .disp32=inouts[0] ":break"}
|
|
|
|
|
2020-02-28 01:03:08 +00:00
|
|
|
We repeat all the 'break' variants almost identically for 'loop' instructions.
|
|
|
|
This works because the compiler inserts ':loop' labels at the start of such
|
|
|
|
named blocks, and ':break' labels at the end.
|
|
|
|
|
|
|
|
loop {.name="loop", .subx-name="e9/jump loop/disp32"}
|
|
|
|
loop label {.name="loop", .inouts=[label], .subx-name="e9/jump", .disp32=inouts[0] ":loop"}
|
2020-02-01 02:39:27 +00:00
|
|
|
|
|
|
|
loop-if-= {.name="loop-if-=", .subx-name="0f 84/jump-if-= loop/disp32"}
|
|
|
|
loop-if-= label {.name="loop-if-=", .inouts=[label], .subx-name="0f 84/jump-if-=", .disp32=inouts[0] ":loop"}
|
|
|
|
loop-if-!= {.name="loop-if-!=", .subx-name="0f 85/jump-if-!= loop/disp32"}
|
|
|
|
loop-if-!= label {.name="loop-if-!=", .inouts=[label], .subx-name="0f 85/jump-if-!=", .disp32=inouts[0] ":loop"}
|
|
|
|
|
|
|
|
loop-if-< {.name="loop-if-<", .subx-name="0f 8c/jump-if-< loop/disp32"}
|
|
|
|
loop-if-< label {.name="loop-if-<", .inouts=[label], .subx-name="0f 8c/jump-if-<", .disp32=inouts[0] ":loop"}
|
|
|
|
loop-if-> {.name="loop-if->", .subx-name="0f 8f/jump-if-> loop/disp32"}
|
|
|
|
loop-if-> label {.name="loop-if->", .inouts=[label], .subx-name="0f 8f/jump-if->", .disp32=inouts[0] ":loop"}
|
|
|
|
loop-if-<= {.name="loop-if-<=", .subx-name="0f 8e/jump-if-<= loop/disp32"}
|
|
|
|
loop-if-<= label {.name="loop-if-<=", .inouts=[label], .subx-name="0f 8e/jump-if-<=", .disp32=inouts[0] ":loop"}
|
|
|
|
loop-if->= {.name="loop-if->=", .subx-name="0f 8d/jump-if->= loop/disp32"}
|
|
|
|
loop-if->= label {.name="loop-if->=", .inouts=[label], .subx-name="0f 8d/jump-if->=", .disp32=inouts[0] ":loop"}
|
|
|
|
|
2020-02-01 20:14:12 +00:00
|
|
|
loop-if-addr< {.name="loop-if-addr<", .subx-name="0f 82/jump-if-addr< loop/disp32"}
|
|
|
|
loop-if-addr< label {.name="loop-if-addr<", .inouts=[label], .subx-name="0f 82/jump-if-addr<", .disp32=inouts[0] ":loop"}
|
|
|
|
loop-if-addr> {.name="loop-if-addr>", .subx-name="0f 87/jump-if-addr> loop/disp32"}
|
|
|
|
loop-if-addr> label {.name="loop-if-addr>", .inouts=[label], .subx-name="0f 87/jump-if-addr>", .disp32=inouts[0] ":loop"}
|
|
|
|
loop-if-addr<= {.name="loop-if-addr<=", .subx-name="0f 86/jump-if-addr<= loop/disp32"}
|
|
|
|
loop-if-addr<= label {.name="loop-if-addr<=", .inouts=[label], .subx-name="0f 86/jump-if-addr<=", .disp32=inouts[0] ":loop"}
|
|
|
|
loop-if-addr>= {.name="loop-if-addr>=", .subx-name="0f 83/jump-if-addr>= loop/disp32"}
|
|
|
|
loop-if-addr>= label {.name="loop-if-addr>=", .inouts=[label], .subx-name="0f 83/jump-if-addr>=", .disp32=inouts[0] ":loop"}
|
|
|
|
|
2020-03-14 21:46:45 +00:00
|
|
|
Address operations
|
|
|
|
|
|
|
|
var/reg: (addr T) <- address var: T
|
|
|
|
{.name="address", .inouts=[var], .outputs=[reg], .subx-name="8d/copy-address", .rm32="*(ebp+" inouts[0].stack-offset ")", .r32=outputs[0]}
|
|
|
|
|
2020-02-21 18:05:04 +00:00
|
|
|
Array operations
|
|
|
|
|
2020-03-08 01:32:39 +00:00
|
|
|
var/reg <- length arr/reg2: (addr array T)
|
2020-02-21 18:05:04 +00:00
|
|
|
{.name="length", .inouts=[reg2], .outputs=[reg1], .subx-name="8b/copy-from", .rm32="*" inouts[0], .r32=outputs[0]}
|
|
|
|
var/reg <- index arr/rega: (addr array T), idx/regi: int
|
2020-03-12 07:17:35 +00:00
|
|
|
{.name="index", .inouts=[rega, regi], .outputs=[reg], .subx-name="8d/copy-address", .rm32="*(" inouts[0] "+" inouts[1] "<<" log2(sizeof(T)) "+4)", .r32=outputs[0]}
|
|
|
|
var/reg <- index arr: (array T sz), idx/regi: int
|
|
|
|
{.name="index", .inouts=[arr, regi], .outputs=[reg], .subx-name="8d/copy-address", .rm32="*(ebp+" inouts[1] "<<" log2(sizeof(T)) " + " inouts[0].stack-offset+4 ")", .r32=outputs[0]}
|
2020-02-21 18:05:04 +00:00
|
|
|
var/reg <- index arr/rega: (addr array T), n
|
2020-03-08 01:32:39 +00:00
|
|
|
{.name="index", .inouts=[rega, n], .outputs=[reg], .subx-name="8d/copy-address", .rm32="*(" inouts[0] "+" inouts[1]*size(T) ")", .r32=outputs[0]}
|
2020-03-12 07:17:35 +00:00
|
|
|
var/reg <- index arr: (array T sz), n
|
|
|
|
{.name="index", .inouts=[arr, n], .outputs=[reg], .subx-name="8d/copy-address", .rm32="*(ebp+" inouts[0].stack-offset+4 + inouts[1]*size(T) ")", .r32=outputs[0]}
|
2020-03-08 01:32:39 +00:00
|
|
|
|
|
|
|
var/reg: (offset T) <- compute-offset arr: (addr array T), idx/regi: int # arr can be in reg or mem
|
|
|
|
{.name="compute-offset", .inouts=[arr, regi], .outputs=[reg], .subx-name="69/multiply", .rm32=inouts[1], .r32=outputs[0], .imm32=sizeof(T)}
|
|
|
|
var/reg: (offset T) <- compute-offset arr: (addr array T), idx: int # arr can be in reg or mem
|
|
|
|
{.name="compute-offset", .inouts=[arr, regi], .outputs=[reg], .subx-name="69/multiply", .rm32="*(ebp+" inouts[1].stack-offset ")", .r32=outputs[0], .imm32=sizeof(T)}
|
|
|
|
var/reg <- index arr/rega: (addr array T), o/rego: offset
|
|
|
|
{.name="index", .inouts=[rega, rego], .outputs=[reg], .subx-name="8d/copy-address", .rm32="*(" inouts[0] "+" inouts[1] "+" "4)", .r32=outputs[0]}
|
2020-02-21 18:05:04 +00:00
|
|
|
|
2020-02-28 00:58:16 +00:00
|
|
|
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: field name => 8d/copy-addr *(reg2+offset(f))
|
2020-03-12 03:56:27 +00:00
|
|
|
{.name="get", .inouts=[reg2, f], .subx-name="8d/copy-address", .rm32="*(" inouts[0] "+" offset(f) ")", .r32=outputs[0]}
|
|
|
|
var/reg: (addr T_f) <- get var2: (addr F), f: field name
|
|
|
|
{.name="get", .inouts=[reg2, f], .subx-name="8d/copy-address", .rm32="*(ebp+" inouts[0].stack-offset "+" offset(f) ")", .r32=outputs[0]}
|
2020-02-28 00:58:16 +00:00
|
|
|
|
2020-02-01 02:39:27 +00:00
|
|
|
vim:ft=c:nowrap
|