ocaml-flambda
diff --git a/‎.github/workflows/build.yml
+10-10 b/‎.github/workflows/build.yml
+10-10
diff --git a/‎.github/workflows/coverage.yml
+2-2 b/‎.github/workflows/coverage.yml
+2-2
diff --git a/‎.github/workflows/ocamlformat.yml
+1-1 b/‎.github/workflows/ocamlformat.yml
+1-1
diff --git a/‎Makefile
+1-1 b/‎Makefile
+1-1
diff --git a/‎asmcomp/asmlink.ml
+12 b/‎asmcomp/asmlink.ml
+12
diff --git a/‎backend/amd64/CSE.ml
+12-7 b/‎backend/amd64/CSE.ml
+12-7
diff --git a/‎backend/amd64/arch.ml
+14-5 b/‎backend/amd64/arch.ml
+14-5
diff --git a/‎backend/amd64/arch.mli
+3-1 b/‎backend/amd64/arch.mli
+3-1
diff --git a/‎backend/amd64/cfg_selection.ml
+16-3 b/‎backend/amd64/cfg_selection.ml
+16-3
@@ -78,14 +78,14 @@ jobs:
             config: --enable-middle-end=flambda2 --disable-warn-error
             os: macos-latest
 
-          - name: flambda2_macos_arm64_irc
-            config: --enable-middle-end=flambda2 --disable-warn-error
+          - name: flambda2_macos_arm64_runtime5_irc
+            config: --enable-middle-end=flambda2 --enable-runtime5 --disable-warn-error
             os: macos-latest
             build_ocamlparam: '_,w=-46,regalloc=irc'
             ocamlparam: '_,w=-46,regalloc=irc'
 
-          - name: flambda2_macos_arm64_ls
-            config: --enable-middle-end=flambda2 --disable-warn-error
+          - name: flambda2_macos_arm64_runtime5_ls
+            config: --enable-middle-end=flambda2 --enable-runtime5 --disable-warn-error
             os: macos-latest
             build_ocamlparam: '_,w=-46,regalloc=ls'
             ocamlparam: '_,w=-46,regalloc=ls'
@@ -282,22 +282,22 @@ jobs:
       run: |
         PATH=$GITHUB_WORKSPACE/ocaml-414/_install/bin:$PATH make check_all_arches
 
-    - uses: actions/upload-artifact@v3
+    - uses: actions/upload-artifact@v4
       if: ${{ failure() }} && matrix.os != 'macos-latest'
       with:
-        name: cores
+        name: cores-${{ github.sha }}
         path: /cores
 
-    - uses: actions/upload-artifact@v3
+    - uses: actions/upload-artifact@v4
       if: ${{ failure() }} && matrix.os != 'macos-latest'
       with:
-        name: _build
+        name: _build-${{ github.sha }}
         path: $GITHUB_WORKSPACE/_build
 
-    - uses: actions/upload-artifact@v3
+    - uses: actions/upload-artifact@v4
       if: ${{ failure() }} && matrix.os != 'macos-latest'
       with:
-        name: _runtest
+        name: _runtest-${{ github.sha }}
         path: $GITHUB_WORKSPACE/_runtest
 
 concurrency:
 
@@ -89,8 +89,8 @@ jobs:
 #         BUILD_OCAMLPARAM: ${{ matrix.ocamlparam }}
 #
 #     - name: Publish coverage report
-#       uses: actions/upload-artifact@v3
+#       uses: actions/upload-artifact@v4
 #       with:
-#         name: coverage
+#         name: coverage-${{ github.sha }}
 #         path: flambda_backend/_coverage/**
 #
@@ -21,7 +21,7 @@ jobs:
         path: 'flambda_backend'
 
     - name: Setup OCaml ${{ matrix.ocaml-compiler }}
-      uses: ocaml/setup-ocaml@v2
+      uses: ocaml/setup-ocaml@v3
       with:
         ocaml-compiler: ${{ matrix.ocaml-compiler }}
 
 
@@ -87,7 +87,7 @@ promote:
 
 .PHONY: fmt
 fmt:
-	ocamlformat -i $$(find . \( -name "*.ml" -or -name "*.mli" \))
+	find . \( -name "*.ml" -or -name "*.mli" \) | xargs -P $$(nproc 2>/dev/null || echo 1) -n 20 ocamlformat -i
 
 .PHONY: check-fmt
 check-fmt:
 
@@ -349,6 +349,16 @@ let sourcefile_for_dwarf ~named_startup_file filename =
   if named_startup_file then filename
   else ".startup"
 
+let emit_ocamlrunparam ~ppf_dump =
+  Asmgen.compile_phrase ~ppf_dump
+    (Cmm.Cdata [
+      Cmm.Cdefine_symbol {
+        sym_name = "caml_ocamlrunparam";
+        sym_global = Global
+      };
+      Cmm.Cstring (!Clflags.ocamlrunparam ^ "\000")
+    ])
+
 let make_startup_file unix ~ppf_dump ~sourcefile_for_dwarf genfns units cached_gen =
   Location.input_name := "caml_startup"; (* set name of "current" input *)
   let startup_comp_unit =
@@ -361,6 +371,7 @@ let make_startup_file unix ~ppf_dump ~sourcefile_for_dwarf genfns units cached_g
   let compile_phrase p = Asmgen.compile_phrase ~ppf_dump p in
   let name_list =
     List.flatten (List.map (fun u -> u.defines) units) in
+  emit_ocamlrunparam ~ppf_dump;
   List.iter compile_phrase (Cmm_helpers.entry_point name_list);
   List.iter compile_phrase
     (* Emit the GC roots table, for dynlink. *)
@@ -414,6 +425,7 @@ let make_shared_startup_file unix ~ppf_dump ~sourcefile_for_dwarf genfns units =
   Emitaux.Dwarf_helpers.init ~disable_dwarf:(not !Dwarf_flags.dwarf_for_startup_file)
     ~sourcefile:sourcefile_for_dwarf;
   Emit.begin_assembly unix;
+  emit_ocamlrunparam ~ppf_dump;
   List.iter compile_phrase
     (Cmm_helpers.emit_gc_roots_table ~symbols:[]
       (Generic_fns.compile ~shared:true genfns));
 
@@ -1,4 +1,3 @@
-# 2 "backend/amd64/CSE.ml"
 (**************************************************************************)
 (*                                                                        *)
 (*                                 OCaml                                  *)
@@ -21,6 +20,12 @@ open Arch
 open Mach
 open CSE_utils
 
+let of_simd_class (cl : Simd.operation_class)  =
+  match cl with
+  | Pure -> Op_pure
+  | Load { is_mutable = true } -> Op_load Mutable
+  | Load { is_mutable = false } -> Op_load Immutable
+
 class cse = object
 
 inherit CSEgen.cse_generic as super
@@ -37,9 +42,9 @@ method! class_of_operation op =
     | Irdtsc | Irdpmc
     | Ilfence | Isfence | Imfence -> Op_other
     | Isimd op ->
-      begin match Simd.class_of_operation op with
-      | Pure -> Op_pure
-      end
+      of_simd_class (Simd.class_of_operation op)
+    | Isimd_mem (op,_addr) ->
+      of_simd_class (Simd.Mem.class_of_operation op)
     | Ipause
     | Icldemote _
     | Iprefetch _ -> Op_other
@@ -81,9 +86,9 @@ class cfg_cse = object
     | Irdtsc | Irdpmc
     | Ilfence | Isfence | Imfence -> Op_other
     | Isimd op ->
-      begin match Simd.class_of_operation op with
-      | Pure -> Op_pure
-      end
+      of_simd_class (Simd.class_of_operation op)
+    | Isimd_mem (op,_addr) ->
+      of_simd_class (Simd.Mem.class_of_operation op)
     | Ipause
     | Icldemote _
     | Iprefetch _ -> Op_other
 
@@ -1,4 +1,3 @@
-# 2 "backend/amd64/arch.ml"
 (**************************************************************************)
 (*                                                                        *)
 (*                                 OCaml                                  *)
@@ -153,6 +152,9 @@ type specific_operation =
   | Imfence                            (* memory fence *)
   | Ipause                             (* hint for spin-wait loops *)
   | Isimd of Simd.operation            (* SIMD instruction set operations *)
+  | Isimd_mem of Simd.Mem.operation * addressing_mode
+                                       (* SIMD instruction set operations
+                                          with memory args *)
   | Icldemote of addressing_mode       (* hint to demote a cacheline to L3 *)
   | Iprefetch of                       (* memory prefetching hint *)
       { is_write: bool;
@@ -273,6 +275,8 @@ let print_specific_operation printreg op ppf arg =
       fprintf ppf "rdpmc %a" printreg arg.(0)
   | Isimd simd ->
       Simd.print_operation printreg simd ppf arg
+  | Isimd_mem (simd, addr) ->
+      Simd.Mem.print_operation printreg (print_addressing printreg addr) simd ppf arg
   | Ipause ->
       fprintf ppf "pause"
   | Icldemote _ ->
@@ -299,13 +303,14 @@ let operation_is_pure = function
   | Istore_int (_, _, _) | Ioffset_loc (_, _)
   | Icldemote _ | Iprefetch _ -> false
   | Isimd op -> Simd.is_pure op
+  | Isimd_mem (op, _addr) -> Simd.Mem.is_pure op
 
 (* Specific operations that can raise *)
 (* Keep in sync with [Vectorize_specific] *)
 let operation_can_raise = function
   | Ilea _ | Ibswap _ | Isextend32 | Izextend32
   | Ifloatarithmem _
-  | Irdtsc | Irdpmc | Ipause | Isimd _
+  | Irdtsc | Irdpmc | Ipause | Isimd _ | Isimd_mem _
   | Ilfence | Isfence | Imfence
   | Istore_int (_, _, _) | Ioffset_loc (_, _)
   | Icldemote _ | Iprefetch _ -> false
@@ -314,7 +319,7 @@ let operation_can_raise = function
 let operation_allocates = function
   | Ilea _ | Ibswap _ | Isextend32 | Izextend32
   | Ifloatarithmem _
-  | Irdtsc | Irdpmc | Ipause | Isimd _
+  | Irdtsc | Irdpmc | Ipause | Isimd _ | Isimd_mem _
   | Ilfence | Isfence | Imfence
   | Istore_int (_, _, _) | Ioffset_loc (_, _)
   | Icldemote _ | Iprefetch _ -> false
@@ -405,9 +410,11 @@ let equal_specific_operation left right =
     && equal_addressing_mode left_addr right_addr
   | Isimd l, Isimd r ->
     Simd.equal_operation l r
+  | Isimd_mem (l,al), Isimd_mem (r,ar) ->
+    Simd.Mem.equal_operation l r && equal_addressing_mode al ar
   | (Ilea _ | Istore_int _ | Ioffset_loc _ | Ifloatarithmem _ | Ibswap _ |
      Isextend32 | Izextend32 | Irdtsc | Irdpmc | Ilfence | Isfence | Imfence |
-     Ipause | Isimd _ | Icldemote _ | Iprefetch _), _ ->
+     Ipause | Isimd _ | Isimd_mem _ | Icldemote _ | Iprefetch _), _ ->
     false
 
 (* addressing mode functions *)
@@ -512,7 +519,9 @@ let isomorphic_specific_operation op1 op2 =
     && equal_addressing_mode_without_displ left_addr right_addr
   | Isimd l, Isimd r ->
     Simd.equal_operation l r
+  | Isimd_mem (l,al), Isimd_mem (r,ar) ->
+    Simd.Mem.equal_operation l r && equal_addressing_mode_without_displ al ar
   | (Ilea _ | Istore_int _ | Ioffset_loc _ | Ifloatarithmem _ | Ibswap _ |
      Isextend32 | Izextend32 | Irdtsc | Irdpmc | Ilfence | Isfence | Imfence |
-     Ipause | Isimd _ | Icldemote _ | Iprefetch _), _ ->
+     Ipause | Isimd _ | Isimd_mem _ | Icldemote _ | Iprefetch _), _ ->
     false
@@ -1,4 +1,3 @@
-# 2 "asmcomp/amd64/arch.mli"
 (**************************************************************************)
 (*                                                                        *)
 (*                                 OCaml                                  *)
@@ -86,6 +85,9 @@ type specific_operation =
   | Imfence                            (* memory fence *)
   | Ipause                             (* hint for spin-wait loops *)
   | Isimd of Simd.operation            (* SIMD instruction set operations *)
+  | Isimd_mem of Simd.Mem.operation * addressing_mode
+                                       (* SIMD instruction set operations
+                                          with memory args *)
   | Icldemote of addressing_mode       (* hint to demote a cacheline to L3 *)
   | Iprefetch of                       (* memory prefetching hint *)
       { is_write: bool;
 
@@ -32,12 +32,17 @@ let pseudoregs_for_operation op arg res =
   | Intop (Iadd | Isub | Imul | Iand | Ior | Ixor)
   | Floatop ((Float32 | Float64), (Iaddf | Isubf | Imulf | Idivf)) ->
     [| res.(0); arg.(1) |], res
-  | Intop_atomic { op = Compare_and_swap; size = _; addr = _ } ->
+  | Intop_atomic { op = Compare_set; size = _; addr = _ } ->
     (* first arg must be rax *)
     let arg = Array.copy arg in
     arg.(0) <- rax;
     arg, res
-  | Intop_atomic { op = Fetch_and_add; size = _; addr = _ } ->
+  | Intop_atomic { op = Compare_exchange; size = _; addr = _ } ->
+    (* first arg must be rax, res.(0) must be rax. *)
+    let arg = Array.copy arg in
+    arg.(0) <- rax;
+    arg, [| rax |]
+  | Intop_atomic { op = Exchange | Fetch_and_add; size = _; addr = _ } ->
     (* first arg must be the same as res.(0) *)
     let arg = Array.copy arg in
     arg.(0) <- res.(0);
@@ -86,14 +91,22 @@ let pseudoregs_for_operation op arg res =
        edx (high) and eax (low). Make it simple and force the argument in rcx,
        and rax and rdx clobbered *)
     [| rcx |], res
-  | Specific (Isimd op) -> Simd_selection.pseudoregs_for_operation op arg res
+  | Specific (Isimd op) ->
+    Simd_selection.pseudoregs_for_operation
+      (Simd_proc.register_behavior op)
+      arg res
+  | Specific (Isimd_mem (op, _addr)) ->
+    Simd_selection.pseudoregs_for_operation
+      (Simd_proc.Mem.register_behavior op)
+      arg res
   | Csel _ ->
     (* last arg must be the same as res.(0) *)
     let len = Array.length arg in
     let arg = Array.copy arg in
     arg.(len - 1) <- res.(0);
     arg, res
   (* Other instructions are regular *)
+  | Intop_atomic { op = Add | Sub | Land | Lor | Lxor; _ }
   | Intop (Ipopcnt | Iclz _ | Ictz _ | Icomp _)
   | Intop_imm ((Imulh _ | Idiv | Imod | Icomp _ | Ipopcnt | Iclz _ | Ictz _), _)
   | Specific
Original file line number	Diff line number	Diff line change
`@@ -89,8 +89,8 @@ jobs:`
`89`	`89`	`# BUILD_OCAMLPARAM: ${{ matrix.ocamlparam }}`
`90`	`90`	`#`
`91`	`91`	`# - name: Publish coverage report`
`92`		`-# uses: actions/upload-artifact@v3`
	`92`	`+# uses: actions/upload-artifact@v4`
`93`	`93`	`# with:`
`94`		`-# name: coverage`
	`94`	`+# name: coverage-${{ github.sha }}`
`95`	`95`	`# path: flambda_backend/_coverage/**`
`96`	`96`	`#`