d2e7c647d0
setting registers in the allocator
...
unfortunately the reg instances are spread across instructions
this causes problems when setting them
2020-03-22 14:31:43 +02:00
c890e8402b
change in register_names protocol
...
move to returning the attribute names
getting and setting can then be automated in the base class
2020-03-22 14:31:43 +02:00
0137056b89
change platform to return register names
...
not just the number of them
also adds protocol to map registers
(like message to r0 , or syscalls)
2020-03-22 14:31:43 +02:00
f13e6dcf57
fix releasing in allocator
...
fell into hash new trap, which reuses the object you give it. not good for mutable objects like the array.
also previous logic was broken in terms of machine vs ssa names
2020-03-22 14:31:43 +02:00
3f131a4018
start with register assigning
...
have to find a way to set them next
2020-03-22 14:31:43 +02:00
4f290ee246
basic liveliness for allocator
2020-03-22 14:31:43 +02:00
d0b734c57c
adding register_names to instruction protocol
2020-03-22 14:31:43 +02:00
8df1b8126f
add register names to allocator
...
make platform instantiate it
basic plumbing
2020-03-22 14:31:43 +02:00
500df01425
rename allocator to standard_allocator
...
planning to have platform dish it out
2020-03-22 14:31:43 +02:00
2e109a16dc
starting on risc allocation
...
inserting allocator stage in method translation
2020-03-22 14:31:43 +02:00
e4a79aac50
reset flags in interpreter
...
fix at least part of the problem why comparisons dont work
2020-03-22 14:31:43 +02:00
3ce6ed5263
fix putstring and puts
...
using syscall regs
2020-03-22 14:31:43 +02:00
17a7f29b0c
define and use syscall_XX registers
...
rather than use hardcoded r0 etc use syscall_X
change the syscalls and interpreter to use them
later use platform to map from syscall_X to actually used register (like r0 in arm)
2020-03-22 14:31:43 +02:00
7232c28ecd
operator also fell into ssa trap
...
relying on register identity
in fact the whole operator concept was geared towards this, using 2 regs instead of one to avoid the whole issue
better now
2020-03-22 14:31:43 +02:00
5b0c1195e4
Fix interpreter and resolve method
...
Interpreter was not handling ssa correctly (ie overwriting)
Resolve was assuming wrong registers (also non ssa)
return value still broken
2020-03-22 14:31:43 +02:00
4db71c1c03
fix all uses of operators now they are ssa
2020-03-22 14:31:43 +02:00
0cde7c1d0a
debugging to find operator not being ssa
2020-03-22 14:31:43 +02:00
61fc8a3991
make operator_instruction single ass
...
create result register automatically
usually not used, but register allocation will sort that
2020-03-22 14:31:43 +02:00
1378745907
first interpreter tests
2020-03-22 14:31:43 +02:00
c16ed5ab3a
fix interpreter
...
by currently not checking for old register pattern
feels like wip, but passes (all but one)
2020-03-22 14:31:43 +02:00
f2250bc206
fix resolve and introduce method to chop strings to max
...
strings are 20 bytes. I want to implement the extension idea, just not now
2020-03-22 14:31:43 +02:00
2656bfacb2
create a reduce_int on RegisterSlot
...
that just passes through to RegisterValue by calling to_reg
2020-03-22 14:31:43 +02:00
8abcaa330b
fix allocate in builder
...
which accessed unknown types.
also moved assert_allocate to support
2020-03-22 14:31:43 +02:00
c9fedec230
add a way to bend the type for register_value
...
specifically for factories, where we know the type of next_object even it is not specified
2020-03-22 14:31:43 +02:00
6267bf3ad0
fix slot_to_reg to allow register indexes
...
we mostly use pre-calculated indexes, ie integers
but registers are allowed (in arm/risc), so we try to check the registers type at least is right.
The return is really a machine word, but we call it Object (yes, more work that way)
2020-03-22 14:31:43 +02:00
fb4fa598f2
fix get_internal_byte
...
improved operators and tests
some logic errors still
2020-03-22 14:31:43 +02:00
12fe0b9a10
add constants automatically
2020-03-22 14:31:43 +02:00
22d513d895
fix div4 and tests
...
also bug in reduce_int, compiler not carried through
2020-03-22 14:31:43 +02:00
eed9ba082f
Fix div10 and test
...
fix and use load_data (similar to load_constant)
and integrate into load_object when appropriate (ie for integers)
2020-03-22 14:31:43 +02:00
bd02f69824
reduce_int was overwriting register
...
No more. But the type question is open, ie what type does the resulting register have
2020-03-22 14:31:43 +02:00
53eb28fff4
load constant to create register names with class
...
Just the id_ did give no clue to the contents, just took care of the uniqueness.
Better for debugging
2020-03-22 14:31:43 +02:00
3ec5557ddb
fix builder logic error
...
and tests
2020-03-22 14:31:43 +02:00
9c5d17a3bb
Fix div10 without method_missing
...
but reanimate infer_type to auto create the needed regsiters
also some helpers
2020-03-22 14:31:43 +02:00
9a5e0f15cd
reannimate infer_type
...
now with a _purpose_
2020-03-22 14:31:43 +02:00
3688c967d3
Fix comparison macro
...
which leaves a definite need for instruction level testing
2020-03-22 14:31:43 +02:00
fd43fc9e5c
finally fixes the builder
2020-03-22 14:31:43 +02:00
db5a59f735
Unify instruction namings also dirs
...
Was getting confused myself, where it was instruction or instructions, when if the base class was inside or out of dir.
Now dirs are plural, and base class is inside.
2020-03-22 14:31:43 +02:00
f3d299208e
fix message_setup with new builder
2020-03-22 14:31:43 +02:00
ece1e8c87b
fix some more of builder
...
but still not all. removed some and fixed the register allocation in allocate_int
2020-03-22 14:31:43 +02:00
ff49ff50c0
Convert SimpleCall to new regs
...
Also fix bug in RegsiterValue/Slot with chain, where logic was dodgy and compiler not set
2020-03-22 14:31:43 +02:00
4b303977a7
Fix first Slot instruction that uses builder
...
now without method_missing and names
but still with instance_eval, hmm. Tried without, makes code much less readable
2020-03-22 14:31:43 +02:00
4888b3b6db
Starting to rework slot instructions that create risc
...
have to go through all and all macros and all thems tests. What did the wise man say: one step at a time
2020-03-22 14:31:43 +02:00
4643be0ae6
codong RegisterSlot with reg and slot
2020-03-22 14:31:43 +02:00
64d860b2bf
create a load on the compiler
...
thus removing the need for << with objects on RegisterValue
2020-03-22 14:31:43 +02:00
95f3eec043
repurpose RValue as RegisterSlot
...
with the idea of the better name came also the one about not needing the builder anymore
2020-03-22 14:31:43 +02:00
d22da1ab97
SA for slot_to_reg
2020-03-22 14:31:43 +02:00
77003eed06
remove use_reg on compiler and SA for load
2020-03-22 14:31:43 +02:00
0ce14bdfd1
moving to SA register names (wip)
...
starting to implement register allocation by first creating SA
Single Assignment means a register is only every assigned a value once. Hence for any operation involving another register, a new register is created.
We do this with a naming scheme for the registers in dot notation (as it would be in c) which means 2 registers with the same name, should have the same contents. This does not apply to temporaries, but that is another day.
Starting WIP now, and will create many red commits before merging when green.
2020-03-22 14:31:43 +02:00
393f0d9a60
fix argument transfer (to be logical)
2020-02-27 18:19:27 +02:00
685022a6e0
Externalise register allocation into own class
...
On the way to the register allocation
2020-02-27 11:57:18 +02:00
