File: codegen.jl

package info (click to toggle)
julia 1.0.3%2Bdfsg-4
  • links: PTS, VCS
  • area: main
  • in suites: buster
  • size: 49,452 kB
  • sloc: lisp: 236,453; ansic: 55,579; cpp: 25,603; makefile: 1,685; pascal: 1,130; sh: 956; asm: 86; xml: 76
file content (336 lines) | stat: -rw-r--r-- 10,586 bytes parent folder | download
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
# This file is a part of Julia. License is MIT: https://julialang.org/license

# tests for codegen and optimizations

using Random
using InteractiveUtils

const opt_level = Base.JLOptions().opt_level
const coverage = (Base.JLOptions().code_coverage > 0) || (Base.JLOptions().malloc_log > 0)
const Iptr = sizeof(Int) == 8 ? "i64" : "i32"

# `_dump_function` might be more efficient but it doesn't really matter here...
get_llvm(@nospecialize(f), @nospecialize(t), strip_ir_metadata=true, dump_module=false) =
    sprint(code_llvm, f, t, strip_ir_metadata, dump_module)

get_llvm_noopt(@nospecialize(f), @nospecialize(t), strip_ir_metadata=true, dump_module=false) =
    InteractiveUtils._dump_function(f, t,
                #=native=# false, #=wrapper=# false, #=strip=# strip_ir_metadata,
                #=dump_module=# dump_module, #=syntax=#:att, #=optimize=#false)


if opt_level > 0
    # Make sure getptls call is removed at IR level with optimization on
    @test !occursin(" call ", get_llvm(identity, Tuple{String}))
end

jl_string_ptr(s::String) = ccall(:jl_string_ptr, Ptr{UInt8}, (Any,), s)
core_sizeof(o) = Core.sizeof(o)
function test_loads_no_call(ir, load_types)
    in_function = false
    load_idx = 1
    for line in eachline(IOBuffer(ir))
        if !in_function
            if startswith(line, "define ")
                in_function = true
            end
            continue
        end
        @test !occursin(" call ", line)
        load_split = split(line, " load ", limit=2)
        if !coverage && length(load_split) >= 2
            @test load_idx <= length(load_types)
            if load_idx <= length(load_types)
                @test startswith(load_split[2], "$(load_types[load_idx]),")
            end
            load_idx += 1
        end
        if startswith(line, "}")
            break
        end
    end
    if !coverage
        @test load_idx == length(load_types) + 1
    end
end

# This function tests if functions are output when compiled if jl_dump_compiles is enabled.
# Have to go through pains with recursive function (eval probably not required) to make sure
# that inlining won't happen.
function test_jl_dump_compiles()
    tfile = tempname()
    io = open(tfile, "w")
    @eval(test_jl_dump_compiles_internal(x) = x)
    ccall(:jl_dump_compiles, Cvoid, (Ptr{Cvoid},), io.handle)
    @eval test_jl_dump_compiles_internal(1)
    ccall(:jl_dump_compiles, Cvoid, (Ptr{Cvoid},), C_NULL)
    close(io)
    tstats = stat(tfile)
    tempty = tstats.size == 0
    rm(tfile)
    @test tempty == false
end

# This function tests if a toplevel thunk is output if jl_dump_compiles is enabled.
# The eval statement creates the toplevel thunk.
function test_jl_dump_compiles_toplevel_thunks()
    tfile = tempname()
    io = open(tfile, "w")
    # Make sure to cause compilation of the eval function
    # before calling it below.
    Core.eval(Main, Any[:(nothing)][1])
    topthunk = Meta.lower(Main, :(for i in 1:10; end))
    ccall(:jl_dump_compiles, Cvoid, (Ptr{Cvoid},), io.handle)
    Core.eval(Main, topthunk)
    ccall(:jl_dump_compiles, Cvoid, (Ptr{Cvoid},), C_NULL)
    close(io)
    tstats = stat(tfile)
    tempty = tstats.size == 0
    rm(tfile)
    @test tempty == true
end

if opt_level > 0
    # Make sure `jl_string_ptr` is inlined
    @test !occursin(" call ", get_llvm(jl_string_ptr, Tuple{String}))
    s = "aaa"
    @test jl_string_ptr(s) == pointer_from_objref(s) + sizeof(Int)
    # String
    test_loads_no_call(get_llvm(core_sizeof, Tuple{String}), [Iptr])
    # String
    test_loads_no_call(get_llvm(core_sizeof, Tuple{Core.SimpleVector}), [Iptr])
    # Array
    test_loads_no_call(get_llvm(core_sizeof, Tuple{Vector{Int}}), [Iptr])
    # As long as the eltype is known we don't need to load the elsize
    test_loads_no_call(get_llvm(core_sizeof, Tuple{Array{Any}}), [Iptr])
    # Check that we load the elsize
    test_loads_no_call(get_llvm(core_sizeof, Tuple{Vector}), [Iptr, "i16"])

    test_jl_dump_compiles()
    test_jl_dump_compiles_toplevel_thunks()
end

# Make sure we will not elide the allocation
@noinline create_ref1() = Ref(1)
function pointer_not_safepoint()
    a = create_ref1()
    unsafe_store!(Ptr{Int}(pointer_from_objref(a)), 3)
    return a[]
end
@test pointer_not_safepoint() == 3

# The current memcmp threshold is 512bytes, make sure this struct has the same size on
# 32bits and 64bits
struct LargeStruct
    x::NTuple{1024,Int8}
    LargeStruct() = new()
end

const large_struct = LargeStruct()
@noinline create_ref_struct() = Ref(large_struct)
function compare_large_struct(a)
    b = create_ref_struct()
    if a[] === b[]
        b[].x[1]
    else
        a[].x[2]
    end
end

mutable struct MutableStruct
    a::Int
    MutableStruct() = new()
end

breakpoint_mutable(a::MutableStruct) = ccall(:jl_breakpoint, Cvoid, (Ref{MutableStruct},), a)

# Allocation with uninitialized field as gcroot
mutable struct BadRef
    x::MutableStruct
    y::MutableStruct
    BadRef(x) = new(x)
end
Base.cconvert(::Type{Ptr{BadRef}}, a::MutableStruct) = BadRef(a)
Base.unsafe_convert(::Type{Ptr{BadRef}}, ar::BadRef) = Ptr{BadRef}(pointer_from_objref(ar.x))

breakpoint_badref(a::MutableStruct) = ccall(:jl_breakpoint, Cvoid, (Ptr{BadRef},), a)

struct PtrStruct
    a::Ptr{Cvoid}
    b::Int
end

mutable struct RealStruct
    a::Float64
    b::Int
end

function Base.cconvert(::Type{Ref{PtrStruct}}, a::RealStruct)
    (a, Ref(PtrStruct(pointer_from_objref(a), a.b)))
end
Base.unsafe_convert(::Type{Ref{PtrStruct}}, at::Tuple) =
    Base.unsafe_convert(Ref{PtrStruct}, at[2])

breakpoint_ptrstruct(a::RealStruct) =
    ccall(:jl_breakpoint, Cvoid, (Ref{PtrStruct},), a)

@noinline r_typeassert(c) = c ? (1,1) : nothing
function f_typeassert(c)
    r_typeassert(c)::Tuple
end
@test !occursin("jl_subtype", get_llvm(f_typeassert, Tuple{Bool}))

if opt_level > 0
    @test !occursin("%gcframe", get_llvm(pointer_not_safepoint, Tuple{}))
    compare_large_struct_ir = get_llvm(compare_large_struct, Tuple{typeof(create_ref_struct())})
    @test occursin("call i32 @memcmp", compare_large_struct_ir)
    @test !occursin("%gcframe", compare_large_struct_ir)

    @test occursin("jl_gc_pool_alloc", get_llvm(MutableStruct, Tuple{}))
    breakpoint_mutable_ir = get_llvm(breakpoint_mutable, Tuple{MutableStruct})
    @test !occursin("%gcframe", breakpoint_mutable_ir)
    @test !occursin("jl_gc_pool_alloc", breakpoint_mutable_ir)

    breakpoint_badref_ir = get_llvm(breakpoint_badref, Tuple{MutableStruct})
    @test !occursin("%gcframe", breakpoint_badref_ir)
    @test !occursin("jl_gc_pool_alloc", breakpoint_badref_ir)

    breakpoint_ptrstruct_ir = get_llvm(breakpoint_ptrstruct, Tuple{RealStruct})
    @test !occursin("%gcframe", breakpoint_ptrstruct_ir)
    @test !occursin("jl_gc_pool_alloc", breakpoint_ptrstruct_ir)
end

function two_breakpoint(a::Float64)
    ccall(:jl_breakpoint, Cvoid, (Ref{Float64},), a)
    ccall(:jl_breakpoint, Cvoid, (Ref{Float64},), a)
end

function load_dummy_ref(x::Int)
    r = Ref{Int}(x)
    GC.@preserve r begin
        unsafe_load(Ptr{Int}(pointer_from_objref(r)))
    end
end

if opt_level > 0
    breakpoint_f64_ir = get_llvm((a)->ccall(:jl_breakpoint, Cvoid, (Ref{Float64},), a),
                                 Tuple{Float64})
    @test !occursin("jl_gc_pool_alloc", breakpoint_f64_ir)
    breakpoint_any_ir = get_llvm((a)->ccall(:jl_breakpoint, Cvoid, (Ref{Any},), a),
                                 Tuple{Float64})
    @test occursin("jl_gc_pool_alloc", breakpoint_any_ir)
    two_breakpoint_ir = get_llvm(two_breakpoint, Tuple{Float64})
    @test !occursin("jl_gc_pool_alloc", two_breakpoint_ir)
    @test occursin("llvm.lifetime.end", two_breakpoint_ir)

    @test load_dummy_ref(1234) === 1234
    load_dummy_ref_ir = get_llvm(load_dummy_ref, Tuple{Int})
    @test !occursin("jl_gc_pool_alloc", load_dummy_ref_ir)
    # Hopefully this is reliable enough. LLVM should be able to optimize this to a direct return.
    @test occursin("ret $Iptr %0", load_dummy_ref_ir)
end

# Issue 22770
let was_gced = false
    @noinline make_tuple(x) = tuple(x)
    @noinline use(x) = ccall(:jl_breakpoint, Cvoid, ())
    @noinline assert_not_gced() = @test !was_gced

    function foo22770()
        b = Ref(2)
        finalizer(x -> was_gced = true, b)
        y = make_tuple(b)
        x = y[1]
        a = Ref(1)
        use(x); use(a); use(y)
        c = Ref(3)
        GC.gc()
        assert_not_gced()
        use(x)
        use(c)
    end
    foo22770()
    GC.gc()
    @test was_gced
end

function egal_svecs()
    a = Core.svec(:a, :b)
    b = Core.svec(:a, :b)
    a === b
end
@test egal_svecs()
@test Core.svec(:a, :b) === Core.svec(:a, :b)

# issue #22582
function issue22582!(a::AbstractArray, b)
    len = length(a)
    if b
        ccall(:jl_array_grow_end, Cvoid, (Any, Csize_t), a, 1)
    end
    return len
end
let c = [1,2,3]
    len1 = length(c)
    len2 = issue22582!(c, true)
    @test len1 == len2
end

# PR #23595
@generated f23595(g, args...) = Expr(:call, :g, Expr(:(...), :args))
x23595 = rand(1)
@test f23595(Core.arrayref, true, x23595, 1) == x23595[]

# Issue #22421
@noinline f22421_1(x) = x[] + 1
@noinline f22421_2(x) = x[] + 2
@noinline f22421_3(x, y, z, v) = x[] + y[] + z[] + v
function g22421_1(x, y, b)
    # Most likely generates a branch with phi node
    if b
        z = x
        v = f22421_1(y)
    else
        z = y
        v = f22421_2(x)
    end
    return f22421_3(x, y, z, v)
end
function g22421_2(x, y, b)
    # Most likely generates a select
    return f22421_3(x, y, b ? x : y, 1)
end

struct A24108
    x::Vector{Int}
end
struct B24108
    x::A24108
end
@noinline f24108(x) = length(x)
# Test no gcframe is allocated for `x.x.x` even though `x.x` isn't live at the call site
g24108(x::B24108) = f24108(x.x.x)

@test g22421_1(Ref(1), Ref(2), true) === 7
@test g22421_1(Ref(3), Ref(4), false) === 16
@test g22421_2(Ref(5), Ref(6), true) === 17
@test g22421_2(Ref(7), Ref(8), false) === 24

if opt_level > 0
    @test !occursin("%gcframe",
                    get_llvm(g22421_1, Tuple{Base.RefValue{Int},Base.RefValue{Int},Bool}))
    @test !occursin("%gcframe",
                    get_llvm(g22421_2, Tuple{Base.RefValue{Int},Base.RefValue{Int},Bool}))
    @test !occursin("%gcframe", get_llvm(g24108, Tuple{B24108}))
end

str_22330 = """
Base.convert(::Type{Array{T,n}}, a::Array) where {T<:Number,n} =
             copyto!(Array{T,n}(undef, size(a)), a)

empty(Dict(),  Pair{Union{},Union{}})
"""
f_22330 = tempname()
write(f_22330, str_22330)
@test success(`$(Base.julia_cmd()) --startup-file=no $f_22330`)