Skip to content

Commit

Permalink
merged
Browse files Browse the repository at this point in the history
  • Loading branch information
@jvanburen
jvanburen committed Dec 20, 2024
2 parents 03f9cc8 + e356f62 commit f7e4a58
Show file tree
Hide file tree
Showing 101 changed files with 5,447 additions and 2,090 deletions.
4 changes: 2 additions & 2 deletions .github/workflows/build.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -109,8 +109,8 @@ jobs:
- name: gi
config: --enable-middle-end=flambda2
os: ubuntu-latest
build_ocamlparam: '_,w=-46,regalloc=gi,regalloc-param=SPLIT_LIVE_RANGES:on,regalloc-param=GI_PRIORITY_HEURISTICS:interval-length,regalloc-param=GI_SELECTION_HEURISTICS:first-available,regalloc-param=GI_SPILLING_HEURISTICS:flat-uses,regalloc-validate=1,cfg-cse-optimize=1'
ocamlparam: '_,w=-46,regalloc=gi,regalloc-param=SPLIT_LIVE_RANGES:on,regalloc-param=GI_PRIORITY_HEURISTICS:interval-length,regalloc-param=GI_SELECTION_HEURISTICS:first-available,regalloc-param=GI_SPILLING_HEURISTICS:flat-uses,regalloc-validate=1,cfg-cse-optimize=1'
build_ocamlparam: '_,w=-46,regalloc=gi,regalloc-param=SPLIT_LIVE_RANGES:on,regalloc-param=GI_PRIORITY_HEURISTICS:interval-length,regalloc-param=GI_SELECTION_HEURISTICS:first-available,regalloc-param=GI_SPILLING_HEURISTICS:flat-uses,regalloc-validate=1,cfg-cse-optimize=1,vectorize=1'
ocamlparam: '_,w=-46,regalloc=gi,regalloc-param=SPLIT_LIVE_RANGES:on,regalloc-param=GI_PRIORITY_HEURISTICS:interval-length,regalloc-param=GI_SELECTION_HEURISTICS:first-available,regalloc-param=GI_SPILLING_HEURISTICS:flat-uses,regalloc-validate=1,cfg-cse-optimize=1,vectorize=1'
check_arch: true

- name: cfg-selection
Expand Down
5 changes: 5 additions & 0 deletions Makefile.common-jst
View file
Original file line number Diff line number Diff line change
Expand Up @@ -78,6 +78,11 @@ runtime-stdlib: boot-compiler
# Dune does not believe the compiler can make .cmxs unless the following file exists.
@touch _build/runtime_stdlib_install/lib/ocaml_runtime_stdlib/dynlink.cmxa

# This target is a polling version of "make runtime-stdlib"
.PHONY: runtime-stdlib-hacking
runtime-stdlib-hacking: boot-compiler
RUNTIME_DIR=$(RUNTIME_DIR) $(dune) build -w $(ws_runstd) --only-package=ocaml_runtime_stdlib @install

compiler: runtime-stdlib
RUNTIME_DIR=$(RUNTIME_DIR) SYSTEM=$(SYSTEM) MODEL=$(MODEL) \
ASPP="$(ASPP)" ASPPFLAGS="$(ASPPFLAGS)" \
Expand Down
6 changes: 6 additions & 0 deletions backend/.ocamlformat-enable
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3,15 +3,21 @@ amd64/selection.ml
amd64/selection_utils.ml
amd64/simd*.ml
amd64/stack_check.ml
amd64/vectorize_specific.ml
arm64/cfg_selection.ml
arm64/selection.ml
arm64/selection_utils.ml
arm64/simd*.ml
arm64/stack_check.ml
arm64/vectorize_specific.ml
asm_targets/**/*.ml
asm_targets/**/*.mli
cfg/**/*.ml
cfg/**/*.mli
cfg/vectorize.ml
cfg/vectorize.mli
vectorize_utils.ml
vectorize_utils.mli
cfg_selectgen.ml
cfg_selectgen.mli
cfg_selection.mli
Expand Down
163 changes: 94 additions & 69 deletions backend/amd64/arch.ml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -288,8 +288,9 @@ let win64 =
| "win64" | "mingw64" | "cygwin" -> true
| _ -> false

(* Specific operations that are pure *)

(* Specific operations that are pure *)
(* Keep in sync with [Vectorize_specific] *)
let operation_is_pure = function
| Ilea _ | Ibswap _ | Isextend32 | Izextend32
| Ifloatarithmem _ -> true
Expand All @@ -300,7 +301,7 @@ let operation_is_pure = function
| Isimd op -> Simd.is_pure op

(* Specific operations that can raise *)

(* Keep in sync with [Vectorize_specific] *)
let operation_can_raise = function
| Ilea _ | Ibswap _ | Isextend32 | Izextend32
| Ifloatarithmem _
Expand All @@ -309,6 +310,7 @@ let operation_can_raise = function
| Istore_int (_, _, _) | Ioffset_loc (_, _)
| Icldemote _ | Iprefetch _ -> false

(* Keep in sync with [Vectorize_specific] *)
let operation_allocates = function
| Ilea _ | Ibswap _ | Isextend32 | Izextend32
| Ifloatarithmem _
Expand Down Expand Up @@ -410,84 +412,107 @@ let equal_specific_operation left right =

(* addressing mode functions *)

let compare_addressing_mode_without_displ (addressing_mode_1: addressing_mode) (addressing_mode_2 : addressing_mode) =
(* Ignores displ when comparing to show that it is possible to calculate the offset *)
let equal_addressing_mode_without_displ (addressing_mode_1: addressing_mode) (addressing_mode_2 : addressing_mode) =
(* Ignores [displ] when comparing to show that it is possible to calculate the offset,
see [addressing_offset_in_bytes]. *)
match addressing_mode_1, addressing_mode_2 with
| Ibased (symbol1, global1, _), Ibased (symbol2, global2, _) -> (
match global1, global2 with
| Global, Global | Local, Local ->
String.compare symbol1 symbol2
| Global, Local -> -1
| Local, Global -> 1)
| Ibased _, _ -> -1
| _, Ibased _ -> 1
| Iindexed _, Iindexed _ -> 0
| Iindexed _, _ -> -1
| _, Iindexed _ -> 1
| Iindexed2 _, Iindexed2 _ -> 0
| Iindexed2 _, _ -> -1
| _, Iindexed2 _ -> 1
| Iscaled (scale1, _), Iscaled (scale2, _) -> Int.compare scale1 scale2
| Iscaled _, _ -> -1
| _, Iscaled _ -> 1
String.equal symbol1 symbol2
| (Global | Local), _ -> false)
| Iindexed _, Iindexed _ -> true
| Iindexed2 _, Iindexed2 _ -> true
| Iscaled (scale1, _), Iscaled (scale2, _) -> Int.equal scale1 scale2
| Iindexed2scaled (scale1, _), Iindexed2scaled (scale2, _) ->
Int.compare scale1 scale2

let compare_addressing_mode_displ (addressing_mode_1: addressing_mode) (addressing_mode_2 : addressing_mode) =
match addressing_mode_1, addressing_mode_2 with
| Ibased (symbol1, global1, n1), Ibased (symbol2, global2, n2) -> (
match global1, global2 with
| Global, Global | Local, Local ->
if symbol1 = symbol2 then Some (Int.compare n1 n2) else None
| Global, Local | Local, Global -> None)
| Iindexed n1, Iindexed n2 -> Some (Int.compare n1 n2)
| Iindexed2 n1, Iindexed2 n2 -> Some (Int.compare n1 n2)
| Iscaled (scale1, n1), Iscaled (scale2, n2) ->
let scale_compare = scale1 - scale2 in
if scale_compare = 0 then Some (Int.compare n1 n2) else None
| Iindexed2scaled (scale1, n1), Iindexed2scaled (scale2, n2) ->
let scale_compare = scale1 - scale2 in
if scale_compare = 0 then Some (Int.compare n1 n2) else None
| Ibased _, _ -> None
| Iindexed _, _ -> None
| Iindexed2 _, _ -> None
| Iscaled _, _ -> None
| Iindexed2scaled _, _ -> None

let addressing_offset_in_bytes (addressing_mode_1: addressing_mode) (addressing_mode_2 : addressing_mode) =
Int.equal scale1 scale2
| (Ibased _ | Iindexed _ | Iindexed2 _ | Iscaled _ | Iindexed2scaled _), _ -> false

let addressing_offset_in_bytes
(addressing_mode_1: addressing_mode)
(addressing_mode_2 : addressing_mode)
~arg_offset_in_bytes
args_1
args_2
=
let address_arg_offset_in_bytes index =
arg_offset_in_bytes args_1.(index) args_2.(index)
in
match addressing_mode_1, addressing_mode_2 with
| Ibased (symbol1, global1, n1), Ibased (symbol2, global2, n2) -> (
match global1, global2 with
| Global, Global | Local, Local ->
if symbol1 = symbol2 then Some (n2 - n1) else None
| Global, Local | Local, Global -> None)
| Iindexed n1, Iindexed n2 -> Some (n2 - n1)
| Iindexed2 n1, Iindexed2 n2 -> Some (n2 - n1)
| Ibased (symbol1, global1, n1), Ibased (symbol2, global2, n2) ->
(* symbol + displ *)
(match global1, global2 with
| Global, Global | Local, Local ->
if String.equal symbol1 symbol2 then Some (n2 - n1) else None
| Global, Local | Local, Global -> None)
| Iindexed n1, Iindexed n2 ->
(* reg + displ *)
(match address_arg_offset_in_bytes 0 with
| Some base_off -> Some (base_off + (n2 - n1))
| None -> None)
| Iindexed2 n1, Iindexed2 n2 ->
(* reg + reg + displ *)
(match address_arg_offset_in_bytes 0, address_arg_offset_in_bytes 1 with
| Some arg0_offset, Some arg1_offset ->
Some (arg0_offset + arg1_offset + (n2 - n1))
| (None, _|Some _, _) -> None)
| Iscaled (scale1, n1), Iscaled (scale2, n2) ->
let scale_compare = scale1 - scale2 in
if scale_compare = 0 then Some (n2 - n1) else None
(* reg * scale + displ *)
if not (Int.compare scale1 scale2 = 0) then None
else
(match address_arg_offset_in_bytes 0 with
| Some offset -> Some ((offset * scale1) + (n2 - n1))
| None -> None)
| Iindexed2scaled (scale1, n1), Iindexed2scaled (scale2, n2) ->
let scale_compare = scale1 - scale2 in
if scale_compare = 0 then Some (n2 - n1) else None
(* reg + reg * scale + displ *)
if not (Int.compare scale1 scale2 = 0) then None else
(match address_arg_offset_in_bytes 0, address_arg_offset_in_bytes 1 with
| Some arg0_offset, Some arg1_offset ->
Some (arg0_offset + (arg1_offset*scale1) + (n2 - n1))
| (None, _|Some _, _) -> None)
| Ibased _, _ -> None
| Iindexed _, _ -> None
| Iindexed2 _, _ -> None
| Iscaled _, _ -> None
| Iindexed2scaled _, _ -> None

let can_cross_loads_or_stores (specific_operation : specific_operation) =
match specific_operation with
| Ilea _ | Istore_int _ | Ioffset_loc _ | Ifloatarithmem _ | Isimd _ | Icldemote _
| Iprefetch _ ->
false
| Ibswap _ | Isextend32 | Izextend32 | Irdtsc | Irdpmc | Ilfence | Isfence | Imfence
| Ipause ->
true

let may_break_alloc_freshness (specific_operation : specific_operation) =
match specific_operation with
| Isimd _ -> true
| Ilea _ | Istore_int _ | Ioffset_loc _ | Ifloatarithmem _ | Ibswap _ | Isextend32
| Izextend32 | Irdtsc | Irdpmc | Ilfence | Isfence | Imfence | Ipause | Icldemote _
| Iprefetch _ ->
false
let isomorphic_specific_operation op1 op2 =
match op1, op2 with
| Ilea a1, Ilea a2 -> equal_addressing_mode_without_displ a1 a2
| Istore_int (_n1, a1, is_assign1), Istore_int (_n2, a2, is_assign2) ->
equal_addressing_mode_without_displ a1 a2 && Bool.equal is_assign1 is_assign2
| Ioffset_loc (_n1, a1), Ioffset_loc (_n2, a2) ->
equal_addressing_mode_without_displ a1 a2
| Ifloatarithmem (w1, o1, a1), Ifloatarithmem (w2, o2, a2) ->
Cmm.equal_float_width w1 w2 &&
equal_float_operation o1 o2 &&
equal_addressing_mode_without_displ a1 a2
| Ibswap { bitwidth = left }, Ibswap { bitwidth = right } ->
Int.equal (int_of_bswap_bitwidth left) (int_of_bswap_bitwidth right)
| Isextend32, Isextend32 ->
true
| Izextend32, Izextend32 ->
true
| Irdtsc, Irdtsc ->
true
| Irdpmc, Irdpmc ->
true
| Ilfence, Ilfence ->
true
| Isfence, Isfence ->
true
| Imfence, Imfence ->
true
| Ipause, Ipause -> true
| Icldemote x, Icldemote x' -> equal_addressing_mode_without_displ x x'
| Iprefetch { is_write = left_is_write; locality = left_locality; addr = left_addr; },
Iprefetch { is_write = right_is_write; locality = right_locality; addr = right_addr; } ->
Bool.equal left_is_write right_is_write
&& equal_prefetch_temporal_locality_hint left_locality right_locality
&& equal_addressing_mode_without_displ left_addr right_addr
| Isimd l, Isimd r ->
Simd.equal_operation l r
| (Ilea _ | Istore_int _ | Ioffset_loc _ | Ifloatarithmem _ | Ibswap _ |
Isextend32 | Izextend32 | Irdtsc | Irdpmc | Ilfence | Isfence | Imfence |
Ipause | Isimd _ | Icldemote _ | Iprefetch _), _ ->
false
17 changes: 9 additions & 8 deletions backend/amd64/arch.mli
View file
Original file line number Diff line number Diff line change
Expand Up @@ -140,14 +140,15 @@ val operation_allocates : specific_operation -> bool
val float_cond_and_need_swap
: Lambda.float_comparison -> X86_ast.float_condition * bool

val isomorphic_specific_operation : specific_operation -> specific_operation -> bool
(* addressing mode functions *)

val compare_addressing_mode_without_displ : addressing_mode -> addressing_mode -> int
val equal_addressing_mode_without_displ : addressing_mode -> addressing_mode -> bool

val compare_addressing_mode_displ : addressing_mode -> addressing_mode -> int option

val addressing_offset_in_bytes : addressing_mode -> addressing_mode -> int option

val can_cross_loads_or_stores : specific_operation -> bool

val may_break_alloc_freshness : specific_operation -> bool
val addressing_offset_in_bytes
: addressing_mode
-> addressing_mode
-> arg_offset_in_bytes:('a -> 'a -> int option)
-> 'a array
-> 'a array
-> int option
Loading

0 comments on commit f7e4a58

Please sign in to comment.