Merge branch 'upstream-ready/immix' into add-makefile-support

Compiler/src/tfuncs.jl

@@ -2017,6 +2017,12 @@ function tuple_tfunc(𝕃::AbstractLattice, argtypes::Vector{Any})
     return anyinfo ? PartialStruct(𝕃, typ, argtypes) : typ
 end
 
+@nospecs function memorynew_tfunc(𝕃::AbstractLattice, memtype, m)
+    hasintersect(widenconst(m), Int) || return Bottom
+    return tmeet(𝕃, instanceof_tfunc(memtype, true)[1], GenericMemory)
+end
+add_tfunc(Core.memorynew, 2, 2, memorynew_tfunc, 10)
+
 @nospecs function memoryrefget_tfunc(𝕃::AbstractLattice, mem, order, boundscheck)
     memoryref_builtin_common_errorcheck(mem, order, boundscheck) || return Bottom
     return memoryref_elemtype(mem)
@@ -2238,13 +2244,31 @@ end
     return boundscheck ⊑ Bool && memtype ⊑ GenericMemoryRef && order ⊑ Symbol
 end
 
+@nospecs function memorynew_nothrow(argtypes::Vector{Any})
+    if !(argtypes[1] isa Const && argtypes[2] isa Const)
+        return false
+    end
+    MemT = argtypes[1].val
+    if !(isconcretetype(MemT) && MemT <: GenericMemory)
+        return false
+    end
+    len = argtypes[2].val
+    if !(len isa Int && 0 <= len < typemax(Int))
+        return false
+    end
+    elsz = datatype_layoutsize(MemT)
+    overflows = checked_smul_int(len, elsz)[2]
+    return !overflows
+end
 # Query whether the given builtin is guaranteed not to throw given the `argtypes`.
 # `argtypes` can be assumed not to contain varargs.
 function _builtin_nothrow(𝕃::AbstractLattice, @nospecialize(f::Builtin), argtypes::Vector{Any},
                           @nospecialize(rt))
     ⊑ = partialorder(𝕃)
     na = length(argtypes)
-    if f === memoryrefnew
+    if f === Core.memorynew
+        return memorynew_nothrow(argtypes)
+    elseif f === memoryrefnew
         return memoryref_builtin_common_nothrow(argtypes)
     elseif f === memoryrefoffset
         length(argtypes) == 1 || return false
@@ -2347,6 +2371,7 @@ const _EFFECT_FREE_BUILTINS = [
     isa,
     UnionAll,
     getfield,
+    Core.memorynew,
     memoryrefnew,
     memoryrefoffset,
     memoryrefget,
@@ -2381,6 +2406,7 @@ const _INACCESSIBLEMEM_BUILTINS = Any[
     compilerbarrier,
     Core._typevar,
     donotdelete,
+    Core.memorynew,
 ]
 
 const _ARGMEM_BUILTINS = Any[
@@ -2543,7 +2569,7 @@ function builtin_effects(𝕃::AbstractLattice, @nospecialize(f::Builtin), argty
             consistent = ALWAYS_TRUE
         elseif f === memoryrefget || f === memoryrefset! || f === memoryref_isassigned
             consistent = CONSISTENT_IF_INACCESSIBLEMEMONLY
-        elseif f === Core._typevar
+        elseif f === Core._typevar || f === Core.memorynew
             consistent = CONSISTENT_IF_NOTRETURNED
         else
             consistent = ALWAYS_FALSE

Compiler/test/irpasses.jl

@@ -1083,12 +1083,13 @@ end
 # test `flags_for_effects` and DCE
 # ================================
 
-let # effect-freeness computation for array allocation
+@testset "effect-freeness computation for array allocation" begin
 
     # should eliminate dead allocations
     good_dims = [1, 2, 3, 4, 10]
     Ns = [1, 2, 3, 4, 10]
-    for dim = good_dims, N = Ns
+    Ts = Any[Int, Union{Missing,Nothing}, Nothing, Any]
+    @testset "$dim, $N" for dim in good_dims, N in Ns
         Int64(dim)^N > typemax(Int) && continue
         dims = ntuple(i->dim, N)
         @test @eval fully_eliminated() do
@@ -1099,7 +1100,7 @@ let # effect-freeness computation for array allocation
 
     # shouldn't eliminate erroneous dead allocations
     bad_dims = [-1, typemax(Int)]
-    for dim in bad_dims, N in [1, 2, 3, 4, 10], T in Any[Int, Union{Missing,Nothing}, Nothing, Any]
+    @testset "$dim, $N, $T" for dim in bad_dims, N in Ns, T in Ts
         dims = ntuple(i->dim, N)
         @test @eval !fully_eliminated() do
             Array{$T,$N}(undef, $(dims...))

base/boot.jl

@@ -575,12 +575,7 @@ struct UndefInitializer end
 const undef = UndefInitializer()
 
 # type and dimensionality specified
-(self::Type{GenericMemory{kind,T,addrspace}})(::UndefInitializer, m::Int) where {T,addrspace,kind} =
-    if isdefined(self, :instance) && m === 0
-        self.instance
-    else
-        ccall(:jl_alloc_genericmemory, Ref{GenericMemory{kind,T,addrspace}}, (Any, Int), self, m)
-    end
+(self::Type{GenericMemory{kind,T,addrspace}})(::UndefInitializer, m::Int) where {T,addrspace,kind} = memorynew(self, m)
 (self::Type{GenericMemory{kind,T,addrspace}})(::UndefInitializer, d::NTuple{1,Int}) where {T,kind,addrspace} = self(undef, getfield(d,1))
 # empty vector constructor
 (self::Type{GenericMemory{kind,T,addrspace}})() where {T,kind,addrspace} = self(undef, 0)

deps/checksums/LazyArtifacts-a719c0e3d68a95c6f3dc9571459428ca8761fa2c.tar.gz/md5

@@ -0,0 +1 @@
+ce3754f1d195e1533fc1f3c51e2a2d4c

deps/checksums/LazyArtifacts-a719c0e3d68a95c6f3dc9571459428ca8761fa2c.tar.gz/sha512

@@ -0,0 +1 @@
+1e1beb6202f089b82f06c1f1fcb129a31d2b438df8a3a49c7367d3df4587d8b9b8fbcf74d3f5ff3c25c241314c71e5dfd72014338b822f01252890e2a14831d1

doc/src/manual/performance-tips.md

@@ -1058,12 +1058,12 @@ the output. As a trivial example, compare
 
 ```jldoctest prealloc
 julia> function xinc(x)
-           return [x, x+1, x+2]
+           return [x + i for i  in 1:3000]
        end;
 
 julia> function loopinc()
            y = 0
-           for i = 1:10^7
+           for i = 1:10^5
                ret = xinc(i)
                y += ret[2]
            end
@@ -1075,16 +1075,16 @@ with
 
 ```jldoctest prealloc
 julia> function xinc!(ret::AbstractVector{T}, x::T) where T
-           ret[1] = x
-           ret[2] = x+1
-           ret[3] = x+2
+           for i in 1:3000
+               ret[i] = x+i
+           end
            nothing
        end;
 
 julia> function loopinc_prealloc()
-           ret = Vector{Int}(undef, 3)
+           ret = Vector{Int}(undef, 3000)
            y = 0
-           for i = 1:10^7
+           for i = 1:10^5
                xinc!(ret, i)
                y += ret[2]
            end
@@ -1096,12 +1096,12 @@ Timing results:
 
 ```jldoctest prealloc; filter = r"[0-9\.]+ seconds \(.*?\)"
 julia> @time loopinc()
-  0.529894 seconds (40.00 M allocations: 1.490 GiB, 12.14% gc time)
-50000015000000
+  0.297454 seconds (200.00 k allocations: 2.239 GiB, 39.80% gc time)
+5000250000
 
 julia> @time loopinc_prealloc()
-  0.030850 seconds (6 allocations: 288 bytes)
-50000015000000
+  0.009410 seconds (2 allocations: 23.477 KiB)
+5000250000
 ```
 
 Preallocation has other advantages, for example by allowing the caller to control the "output"

src/array.c

@@ -16,12 +16,6 @@
 extern "C" {
 #endif
 
-#if defined(_P64) && defined(UINT128MAX)
-typedef __uint128_t wideint_t;
-#else
-typedef uint64_t wideint_t;
-#endif
-
 #define MAXINTVAL (((size_t)-1)>>1)
 
 JL_DLLEXPORT int jl_array_validate_dims(size_t *nel, uint32_t ndims, size_t *dims)
@@ -30,10 +24,9 @@ JL_DLLEXPORT int jl_array_validate_dims(size_t *nel, uint32_t ndims, size_t *dim
     size_t _nel = 1;
     for (i = 0; i < ndims; i++) {
         size_t di = dims[i];
-        wideint_t prod = (wideint_t)_nel * (wideint_t)di;
-        if (prod >= (wideint_t) MAXINTVAL || di >= MAXINTVAL)
+        int overflow = __builtin_mul_overflow(_nel, di, &_nel);
+        if (overflow || di >= MAXINTVAL)
             return 1;
-        _nel = prod;
     }
     *nel = _nel;
     return 0;
@@ -204,7 +197,7 @@ JL_DLLEXPORT void jl_array_grow_end(jl_array_t *a, size_t inc)
     int isbitsunion = jl_genericmemory_isbitsunion(a->ref.mem);
     size_t newnrows = n + inc;
     if (!isbitsunion && elsz == 0) {
-        jl_genericmemory_t *newmem = jl_alloc_genericmemory(mtype, MAXINTVAL - 1);
+        jl_genericmemory_t *newmem = jl_alloc_genericmemory(mtype, MAXINTVAL - 2);
         a->ref.mem = newmem;
         jl_gc_wb(a, newmem);
         a->dimsize[0] = newnrows;

src/builtin_proto.h

@@ -46,6 +46,7 @@ DECLARE_BUILTIN(is);
 DECLARE_BUILTIN(isa);
 DECLARE_BUILTIN(isdefined);
 DECLARE_BUILTIN(issubtype);
+DECLARE_BUILTIN(memorynew);
 DECLARE_BUILTIN(memoryref);
 DECLARE_BUILTIN(memoryref_isassigned);
 DECLARE_BUILTIN(memoryrefget);

src/builtins.c

@@ -1675,6 +1675,15 @@ JL_CALLABLE(jl_f__typevar)
 }
 
 // genericmemory ---------------------------------------------------------------------
+JL_CALLABLE(jl_f_memorynew)
+{
+    JL_NARGS(memorynew, 2, 2);
+    jl_datatype_t *jl_genericmemory_type_type = jl_datatype_type;
+    JL_TYPECHK(memorynew, genericmemory_type, args[0]);
+    JL_TYPECHK(memorynew, long, args[1]);
+    size_t nel = jl_unbox_long(args[1]);
+    return (jl_value_t*)jl_alloc_genericmemory(args[0], nel);
+}
 
 JL_CALLABLE(jl_f_memoryref)
 {
@@ -2441,6 +2450,7 @@ void jl_init_primitives(void) JL_GC_DISABLED
     jl_builtin_setglobalonce = add_builtin_func("setglobalonce!", jl_f_setglobalonce);
 
     // memory primitives
+    jl_builtin_memorynew = add_builtin_func("memorynew", jl_f_memorynew);
     jl_builtin_memoryref = add_builtin_func("memoryrefnew", jl_f_memoryref);
     jl_builtin_memoryrefoffset = add_builtin_func("memoryrefoffset", jl_f_memoryrefoffset);
     jl_builtin_memoryrefget = add_builtin_func("memoryrefget", jl_f_memoryrefget);

src/ccall.cpp

@@ -1877,33 +1877,6 @@ static jl_cgval_t emit_ccall(jl_codectx_t &ctx, jl_value_t **args, size_t nargs)
         JL_GC_POP();
         return mark_julia_type(ctx, obj, true, jl_any_type);
     }
-    else if (is_libjulia_func(jl_alloc_genericmemory)) {
-        ++CCALL_STAT(jl_alloc_genericmemory);
-        assert(lrt == ctx.types().T_prjlvalue);
-        assert(!isVa && !llvmcall && nccallargs == 2);
-        const jl_cgval_t &typ = argv[0];
-        const jl_cgval_t &nel = argv[1];
-        auto arg_typename = [&] JL_NOTSAFEPOINT {
-            auto istyp = argv[0].constant;
-            std::string type_str;
-            if (istyp && jl_is_datatype(istyp) && jl_is_genericmemory_type(istyp)){
-                auto eltype = jl_tparam1(istyp);
-                if (jl_is_datatype(eltype))
-                    type_str = jl_symbol_name(((jl_datatype_t*)eltype)->name->name);
-                else if (jl_is_uniontype(eltype))
-                    type_str = "Union";
-                else
-                    type_str = "<unknown type>";
-            }
-            else
-                type_str = "<unknown type>";
-            return "Memory{" + type_str + "}[]";
-            };
-        auto alloc = ctx.builder.CreateCall(prepare_call(jl_allocgenericmemory), { boxed(ctx,typ), emit_unbox(ctx, ctx.types().T_size, nel, (jl_value_t*)jl_ulong_type)});
-        setName(ctx.emission_context, alloc, arg_typename);
-        JL_GC_POP();
-        return mark_julia_type(ctx, alloc, true, jl_any_type);
-    }
     else if (is_libjulia_func(memcpy) && (rt == (jl_value_t*)jl_nothing_type || jl_is_cpointer_type(rt))) {
         ++CCALL_STAT(memcpy);
         const jl_cgval_t &dst = argv[0];