Merge branch 'backports-release-1.11' into ib/regex_finalizer

Author: IanButterworth

THIRDPARTY.md

@@ -6,7 +6,6 @@ for exceptions.
 - [crc32c.c](https://stackoverflow.com/questions/17645167/implementing-sse-4-2s-crc32c-in-software) (CRC-32c checksum code by Mark Adler) [[ZLib](https://opensource.org/licenses/Zlib)].
 - [LDC](https://github.com/ldc-developers/ldc/blob/master/LICENSE) (for ccall/cfunction ABI definitions) [BSD-3]. The portion of code that Julia uses from LDC is [BSD-3] licensed.
 - [LLVM](https://releases.llvm.org/3.9.0/LICENSE.TXT) (for parts of src/disasm.cpp) [UIUC]
-- [MINGW](https://sourceforge.net/p/mingw/mingw-org-wsl/ci/legacy/tree/mingwrt/mingwex/dirname.c) (for dirname implementation on Windows) [MIT]
 - [NetBSD](https://www.netbsd.org/about/redistribution.html) (for setjmp, longjmp, and strptime implementations on Windows) [BSD-3]
 - [Python](https://docs.python.org/3/license.html) (for strtod implementation on Windows) [PSF]
 - [FEMTOLISP](https://github.com/JeffBezanson/femtolisp) [BSD-3]

base/abstractarray.jl

@@ -3654,7 +3654,31 @@ function _keepat!(a::AbstractVector, m::AbstractVector{Bool})
     deleteat!(a, j:lastindex(a))
 end
 
-## 1-d circshift ##
+"""
+    circshift!(a::AbstractVector, shift::Integer)
+
+Circularly shift, or rotate, the data in vector `a` by `shift` positions.
+
+# Examples
+
+```jldoctest
+julia> circshift!([1, 2, 3, 4, 5], 2)
+5-element Vector{Int64}:
+ 4
+ 5
+ 1
+ 2
+ 3
+
+julia> circshift!([1, 2, 3, 4, 5], -2)
+5-element Vector{Int64}:
+ 3
+ 4
+ 5
+ 1
+ 2
+```
+"""
 function circshift!(a::AbstractVector, shift::Integer)
     n = length(a)
     n == 0 && return a

base/math.jl

@@ -1140,6 +1140,10 @@ function modf(x::T) where T<:IEEEFloat
     return (rx, ix)
 end
 
+@inline function use_power_by_squaring(n::Integer)
+    -2^12 <= n <= 3 * 2^13
+end
+
 # @constprop aggressive to help the compiler see the switch between the integer and float
 # variants for callers with constant `y`
 @constprop :aggressive function ^(x::Float64, y::Float64)
@@ -1152,24 +1156,33 @@ end
         y = sign(y)*0x1.8p62
     end
     yint = unsafe_trunc(Int64, y) # This is actually safe since julia freezes the result
-    y == yint && return @noinline x^yint
-    2*xu==0 && return abs(y)*Inf*(!(y>0)) # if x==0
-    x<0 && throw_exp_domainerror(x) # |y| is small enough that y isn't an integer
-    !isfinite(x) && return x*(y>0 || isnan(x))           # x is inf or NaN
+    yisint = y == yint
+    if yisint
+        yint == 0 && return 1.0
+        use_power_by_squaring(yint) && return @noinline pow_body(x, yint)
+    end
+    2*xu==0 && return abs(y)*Inf*(!(y>0)) # if x === +0.0 or -0.0 (Inf * false === 0.0)
+    s = 1
+    if x < 0
+        !yisint && throw_exp_domainerror(x) # y isn't an integer
+        s = ifelse(isodd(yint), -1, 1)
+    end
+    !isfinite(x) && return copysign(x,s)*(y>0 || isnan(x))           # x is inf or NaN
+    return copysign(pow_body(abs(x), y), s)
+end
+
+@assume_effects :foldable @noinline function pow_body(x::Float64, y::Float64)
+    xu = reinterpret(UInt64, x)
     if xu < (UInt64(1)<<52) # x is subnormal
         xu = reinterpret(UInt64, x * 0x1p52) # normalize x
         xu &= ~sign_mask(Float64)
         xu -= UInt64(52) << 52 # mess with the exponent
     end
-    return pow_body(xu, y)
-end
-
-@inline function pow_body(xu::UInt64, y::Float64)
     logxhi,logxlo = _log_ext(xu)
     xyhi, xylo = two_mul(logxhi,y)
     xylo = muladd(logxlo, y, xylo)
     hi = xyhi+xylo
-    return Base.Math.exp_impl(hi, xylo-(hi-xyhi), Val(:ℯ))
+    return @inline Base.Math.exp_impl(hi, xylo-(hi-xyhi), Val(:ℯ))
 end
 
 @constprop :aggressive function ^(x::T, y::T) where T <: Union{Float16, Float32}
@@ -1193,12 +1206,29 @@ end
     return T(exp2(log2(abs(widen(x))) * y))
 end
 
-# compensated power by squaring
 @constprop :aggressive @inline function ^(x::Float64, n::Integer)
+    x^clamp(n, Int64)
+end
+@constprop :aggressive @inline function ^(x::Float64, n::Int64)
     n == 0 && return one(x)
-    return pow_body(x, n)
+    if use_power_by_squaring(n)
+        return pow_body(x, n)
+    else
+        s = ifelse(x < 0 && isodd(n), -1.0, 1.0)
+        x = abs(x)
+        y = float(n)
+        if y == n
+            return copysign(pow_body(x, y), s)
+        else
+            n2 = n % 1024
+            y = float(n - n2)
+            return pow_body(x, y) * copysign(pow_body(x, n2), s)
+        end
+    end
 end
 
+# compensated power by squaring
+# this method is only reliable for -2^20 < n < 2^20 (cf. #53881 #53886)
 @assume_effects :terminates_locally @noinline function pow_body(x::Float64, n::Integer)
     y = 1.0
     xnlo = ynlo = 0.0

base/precompilation.jl

@@ -141,15 +141,16 @@ function ExplicitEnv(envpath::String=Base.active_project())
 
             # Extensions
             deps_pkg = get(Dict{String, Any}, pkg_info, "extensions")::Dict{String, Any}
+            deps_pkg_concrete = Dict{String, Vector{String}}()
             for (ext, triggers) in deps_pkg
                 if triggers isa String
                     triggers = [triggers]
                 else
                     triggers = triggers::Vector{String}
                 end
-                deps_pkg[ext] = triggers
+                deps_pkg_concrete[ext] = triggers
             end
-            extensions[m_uuid] = deps_pkg
+            extensions[m_uuid] = deps_pkg_concrete
 
             # Determine strategy to find package
             lookup_strat = begin

base/special/exp.jl

@@ -252,7 +252,7 @@ end
             twopk = (k + UInt64(53)) << 52
             return reinterpret(T, twopk + reinterpret(UInt64, small_part))*0x1p-53
         end
-        #k == 1024 && return (small_part * 2.0) * 2.0^1023
+        k == 1024 && return (small_part * 2.0) * 2.0^1023
     end
     twopk = Int64(k) << 52
     return reinterpret(T, twopk + reinterpret(Int64, small_part))

base/strings/basic.jl

@@ -512,11 +512,11 @@ prevind(s::AbstractString, i::Int)                 = prevind(s, i, 1)
 
 function prevind(s::AbstractString, i::Int, n::Int)
     n < 0 && throw(ArgumentError("n cannot be negative: $n"))
-    z = ncodeunits(s) + 1
+    z = ncodeunits(s)::Int + 1
     @boundscheck 0 < i ≤ z || throw(BoundsError(s, i))
-    n == 0 && return thisind(s, i) == i ? i : string_index_err(s, i)
+    n == 0 && return thisind(s, i)::Int == i ? i : string_index_err(s, i)
     while n > 0 && 1 < i
-        @inbounds n -= isvalid(s, i -= 1)
+        @inbounds n -= isvalid(s, i -= 1)::Bool
     end
     return i - n
 end
@@ -571,11 +571,11 @@ nextind(s::AbstractString, i::Int)                 = nextind(s, i, 1)
 
 function nextind(s::AbstractString, i::Int, n::Int)
     n < 0 && throw(ArgumentError("n cannot be negative: $n"))
-    z = ncodeunits(s)
+    z = ncodeunits(s)::Int
     @boundscheck 0 ≤ i ≤ z || throw(BoundsError(s, i))
-    n == 0 && return thisind(s, i) == i ? i : string_index_err(s, i)
+    n == 0 && return thisind(s, i)::Int == i ? i : string_index_err(s, i)
     while n > 0 && i < z
-        @inbounds n -= isvalid(s, i += 1)
+        @inbounds n -= isvalid(s, i += 1)::Bool
     end
     return i + n
 end

base/summarysize.jl

@@ -149,13 +149,8 @@ function (ss::SummarySize)(obj::GenericMemory)
     datakey = unsafe_convert(Ptr{Cvoid}, obj)
     if !haskey(ss.seen, datakey)
         ss.seen[datakey] = true
-        dsize = sizeof(obj)
+        size += sizeof(obj)
         T = eltype(obj)
-        if isbitsunion(T)
-            # add 1 union selector byte for each element
-            dsize += length(obj)
-        end
-        size += dsize
         if !isempty(obj) && T !== Symbol && (!Base.allocatedinline(T) || (T isa DataType && !Base.datatype_pointerfree(T)))
             push!(ss.frontier_x, obj)
             push!(ss.frontier_i, 1)

doc/src/manual/calling-c-and-fortran-code.md

@@ -547,15 +547,14 @@ is not valid, since the type layout of `T` is not known statically.
 
 ### SIMD Values
 
-Note: This feature is currently implemented on 64-bit x86 and AArch64 platforms only.
-
 If a C/C++ routine has an argument or return value that is a native SIMD type, the corresponding
 Julia type is a homogeneous tuple of `VecElement` that naturally maps to the SIMD type. Specifically:
 
->   * The tuple must be the same size as the SIMD type. For example, a tuple representing an `__m128`
->     on x86 must have a size of 16 bytes.
->   * The element type of the tuple must be an instance of `VecElement{T}` where `T` is a primitive type that
->     is 1, 2, 4 or 8 bytes.
+>   * The tuple must be the same size and elements as the SIMD type. For example, a tuple
+>     representing an `__m128` on x86 must have a size of 16 bytes and Float32 elements.
+>   * The element type of the tuple must be an instance of `VecElement{T}` where `T` is a
+>     primitive type with a power-of-two number of bytes (e.g. 1, 2, 4, 8, 16, etc) such as
+>     Int8 or Float64.
 
 For instance, consider this C routine that uses AVX intrinsics:
 
@@ -628,6 +627,10 @@ For translating a C argument list to Julia:
 
       * `T`, where `T` is a Julia leaf type
       * argument value will be copied (passed by value)
+  * `vector T` (or `__attribute__ vector_size`, or a typedef such as `__m128`)
+
+      * `NTuple{N, VecElement{T}}`, where `T` is a primitive Julia type of the correct size
+        and N is the number of elements in the vector (equal to `vector_size / sizeof T`).
   * `void*`
 
       * depends on how this parameter is used, first translate this to the intended pointer type, then
@@ -674,13 +677,16 @@ For translating a C return type to Julia:
   * `T`, where `T` is one of the primitive types: `char`, `int`, `long`, `short`, `float`, `double`,
     `complex`, `enum` or any of their `typedef` equivalents
 
-      * `T`, where `T` is an equivalent Julia Bits Type (per the table above)
-      * if `T` is an `enum`, the argument type should be equivalent to `Cint` or `Cuint`
+      * same as C argument list
       * argument value will be copied (returned by-value)
   * `struct T` (including typedef to a struct)
 
-      * `T`, where `T` is a Julia Leaf Type
+      * same as C argument list
       * argument value will be copied (returned by-value)
+
+  * `vector T`
+
+      * same as C argument list
   * `void*`
 
       * depends on how this parameter is used, first translate this to the intended pointer type, then

src/cgutils.cpp

@@ -1971,7 +1971,7 @@ static jl_cgval_t typed_load(jl_codectx_t &ctx, Value *ptr, Value *idx_0based, j
     else if (!alignment)
         alignment = julia_alignment(jltype);
     if (intcast && Order == AtomicOrdering::NotAtomic) {
-        emit_memcpy(ctx, intcast, jl_aliasinfo_t::fromTBAA(ctx, ctx.tbaa().tbaa_stack), data, jl_aliasinfo_t::fromTBAA(ctx, tbaa), nb, Align(alignment), intcast->getAlign());
+        emit_memcpy(ctx, intcast, jl_aliasinfo_t::fromTBAA(ctx, ctx.tbaa().tbaa_stack), data, jl_aliasinfo_t::fromTBAA(ctx, tbaa), nb, intcast->getAlign(), Align(alignment));
     }
     else {
         if (!isboxed && jl_is_genericmemoryref_type(jltype)) {
@@ -3214,7 +3214,7 @@ static void union_alloca_type(jl_uniontype_t *ut,
             [&](unsigned idx, jl_datatype_t *jt) {
                 if (!jl_is_datatype_singleton(jt)) {
                     size_t nb1 = jl_datatype_size(jt);
-                    size_t align1 = jl_datatype_align(jt);
+                    size_t align1 = julia_alignment((jl_value_t*)jt);
                     if (nb1 > nbytes)
                         nbytes = nb1;
                     if (align1 > align)
@@ -3796,9 +3796,10 @@ static jl_cgval_t emit_new_struct(jl_codectx_t &ctx, jl_value_t *ty, size_t narg
 
             // whether we should perform the initialization with the struct as a IR value
             // or instead initialize the stack buffer with stores
+            // although we do the former if it is a vector or could be a vector element
             auto tracked = CountTrackedPointers(lt);
             bool init_as_value = false;
-            if (lt->isVectorTy() || jl_is_vecelement_type(ty)) { // maybe also check the size ?
+            if (lt->isVectorTy() || jl_special_vector_alignment(1, ty) != 0) {
                 init_as_value = true;
             }
             else if (tracked.count) {

src/codegen.cpp

@@ -8552,6 +8552,8 @@ static jl_llvm_functions_t
             Type *RT = Arg->getParamStructRetType();
             TypeSize sz = DL.getTypeAllocSize(RT);
             Align al = DL.getPrefTypeAlign(RT);
+            if (al > MAX_ALIGN)
+                al = Align(MAX_ALIGN);
             param.addAttribute(Attribute::NonNull);
             // The `dereferenceable` below does not imply `nonnull` for non addrspace(0) pointers.
             param.addDereferenceableAttr(sz);

src/datatype.c

@@ -298,9 +298,10 @@ static jl_datatype_layout_t *jl_get_layout(uint32_t sz,
 }
 
 // Determine if homogeneous tuple with fields of type t will have
-// a special alignment beyond normal Julia rules.
+// a special alignment and vector-ABI beyond normal rules for aggregates.
 // Return special alignment if one exists, 0 if normal alignment rules hold.
 // A non-zero result *must* match the LLVM rules for a vector type <nfields x t>.
+// Matching the compiler's `__attribute__ vector_size` behavior.
 // For sake of Ahead-Of-Time (AOT) compilation, this routine has to work
 // without LLVM being available.
 unsigned jl_special_vector_alignment(size_t nfields, jl_value_t *t)
@@ -315,8 +316,12 @@ unsigned jl_special_vector_alignment(size_t nfields, jl_value_t *t)
         // motivating use case comes up for Julia, we reject pointers.
         return 0;
     size_t elsz = jl_datatype_size(ty);
-    if (elsz != 1 && elsz != 2 && elsz != 4 && elsz != 8)
-        // Only handle power-of-two-sized elements (for now)
+    if (next_power_of_two(elsz) != elsz)
+        // Only handle power-of-two-sized elements (for now), since other
+        // lengths may be packed into very complicated arrangements (llvm pads
+        // extra bits on most platforms when computing alignment but not when
+        // computing type size, but adds no extra bytes for each element, so
+        // their effect on offsets are never what you may naturally expect).
         return 0;
     size_t size = nfields * elsz;
     // Use natural alignment for this vector: this matches LLVM and clang.
@@ -707,9 +712,9 @@ void jl_compute_field_offsets(jl_datatype_t *st)
             }
             else {
                 fsz = sizeof(void*);
-                if (fsz > MAX_ALIGN)
-                    fsz = MAX_ALIGN;
                 al = fsz;
+                if (al > MAX_ALIGN)
+                    al = MAX_ALIGN;
                 desc[i].isptr = 1;
                 zeroinit = 1;
                 npointers++;
@@ -929,6 +934,18 @@ JL_DLLEXPORT jl_datatype_t *jl_new_primitivetype(jl_value_t *name, jl_module_t *
                                         jl_emptysvec, jl_emptysvec, jl_emptysvec, 0, 0, 0);
     uint32_t nbytes = (nbits + 7) / 8;
     uint32_t alignm = next_power_of_two(nbytes);
+# if defined(_CPU_X86_) && !defined(_OS_WINDOWS_)
+    // datalayout strings are often weird: on 64-bit they usually follow fairly simple rules,
+    // but on x86 32 bit platforms, sometimes 5 to 8 byte types are
+    // 32-bit aligned even though the MAX_ALIGN (for types 9+ bytes) is 16
+    // (except for f80 which is align 4 on Mingw, Linux, and BSDs--but align 16 on MSVC and Darwin)
+    // https://llvm.org/doxygen/ARMTargetMachine_8cpp.html#adb29b487708f0dc2a940345b68649270
+    // https://llvm.org/doxygen/AArch64TargetMachine_8cpp.html#a003a58caf135efbf7273c5ed84e700d7
+    // https://llvm.org/doxygen/X86TargetMachine_8cpp.html#aefdbcd6131ef195da070cef7fdaf0532
+    // 32-bit alignment is weird
+    if (alignm == 8)
+        alignm = 4;
+# endif
     if (alignm > MAX_ALIGN)
         alignm = MAX_ALIGN;
     // memoize isprimitivetype, since it is much easier than checking

src/llvm-late-gc-lowering.cpp

@@ -2331,8 +2331,10 @@ bool LateLowerGCFrame::CleanupIR(Function &F, State *S, bool *CFGModified) {
             // strip all constant alias information, as it might depend on the gc having
             // preserved a gc root, which stops being true after this pass (#32215)
             // similar to RewriteStatepointsForGC::stripNonValidData, but less aggressive
-            if (I->getMetadata(LLVMContext::MD_invariant_load))
-                I->setMetadata(LLVMContext::MD_invariant_load, NULL);
+            if (auto *LI = dyn_cast<LoadInst>(I)){
+                if (isSpecialPtr(LI->getPointerOperand()->getType()) && LI->getMetadata(LLVMContext::MD_invariant_load))
+                    LI->setMetadata(LLVMContext::MD_invariant_load, NULL);
+            }
             if (MDNode *TBAA = I->getMetadata(LLVMContext::MD_tbaa)) {
                 if (TBAA->getNumOperands() == 4 && isTBAA(TBAA, {"jtbaa_const", "jtbaa_memoryptr", "jtbaa_memorylen", "tbaa_memoryown"})) {
                     MDNode *MutableTBAA = createMutableTBAAAccessTag(TBAA);