Files
swift-mirror/test/SILOptimizer/globalopt_global_propagation.swift
Andrew Trick a17dbc7c74 Enable run-time exclusivity checking in release mode.
This change could impact Swift programs that previously appeared
well-behaved, but weren't fully tested in debug mode. Now, when running
in release mode, they may trap with the message "error: overlapping
accesses...".

Recent optimizations have brought performance where I think it needs
to be for adoption. More optimizations are planned, and some
benchmarks should be further improved, but at this point we're ready
to begin receiving bug reports. That will help prioritize the
remaining work for Swift 5.

Of the 656 public microbenchmarks in the Swift repository, there are
still several regressions larger than 10%:

TEST                    OLD      NEW      DELTA      RATIO
ClassArrayGetter2       139      1307     +840.3%    **0.11x**
HashTest                631      1233     +95.4%     **0.51x**
NopDeinit               21269    32389    +52.3%     **0.66x**
Hanoi                   1478     2166     +46.5%     **0.68x**
Calculator              127      158      +24.4%     **0.80x**
Dictionary3OfObjects    391      455      +16.4%     **0.86x**
CSVParsingAltIndices2   526      604      +14.8%     **0.87x**
Prims                   549      626      +14.0%     **0.88x**
CSVParsingAlt2          1252     1411     +12.7%     **0.89x**
Dictionary4OfObjects    206      232      +12.6%     **0.89x**
ArrayInClass            46       51       +10.9%     **0.90x**

The common pattern in these benchmarks is to define an array of data
as a class property and to repeatedly access that array through the
class reference. Each of those class property accesses now incurs a
runtime call. Naturally, introducing a runtime call in a loop that
otherwise does almost no work incurs substantial overhead. This is
similar to the issue caused by automatic reference counting. In some
cases, more sophistacated optimization will be able to determine the
same object is repeatedly accessed. Furthermore, the overhead of the
runtime call itself can be improved. But regardless of how well we
optimize, there will always a class of microbenchmarks in which the
runtime check has a noticeable impact.

As a general guideline, avoid performing class property access within
the most performance critical loops, particularly on different objects
in each loop iteration. If that isn't possible, it may help if the
visibility of those class properties is private or internal.
2018-11-02 16:54:31 -07:00

230 lines
7.1 KiB
Swift

// RUN: %target-swift-frontend -O -emit-sil -enforce-exclusivity=unchecked %s | %FileCheck %s
// RUN: %target-swift-frontend -O -wmo -emit-sil -enforce-exclusivity=unchecked %s | %FileCheck -check-prefix=CHECK-WMO %s
// Check that values of internal and private global variables, which are provably assigned only
// once, are propagated into their uses and enable further optimizations like constant
// propagation, simplifications, etc.
// Define some global variables.
public var VD = 3.1415
public var VI = 100
private var PVD = 3.1415
private var PVI = 100
private var PVIAssignTwice = 1
private var PVITakenAddress = 1
internal var IVD = 3.1415
internal var IVI = 100
internal var IVIAssignTwice = 1
internal var IVITakenAddress = 1
// Taking the address of a global should prevent from performing the propagation of its value.
@inline(never)
@_optimize(none)
public func takeInout<T>(_ x: inout T) {
}
// Compiler should detect that we assign a global here as well and prevent a global optimization.
public func assignSecondTime() {
PVIAssignTwice = 2
IVIAssignTwice = 2
}
// Having multiple assignments to a global should prevent from performing the propagation of its value.
// Loads from private global variables can be removed,
// because they cannot be changed outside of this source file.
// CHECK-LABEL: sil [noinline] @$s28globalopt_global_propagation013test_private_B11_var_doubleSdyF
// CHECK: bb0:
// CHECK-NOT: global_addr
// CHECK: float_literal
// CHECK: struct
// CHECK: return
@inline(never)
public func test_private_global_var_double() -> Double {
return PVD + 1.0
}
// Loads from private global variables can be removed,
// because they cannot be changed outside of this source file.
// CHECK-LABEL: sil [noinline] @$s28globalopt_global_propagation013test_private_B8_var_intSiyF
// CHECK: bb0:
// CHECK-NOT: global_addr
// CHECK: integer_literal
// CHECK: struct
// CHECK: return
@inline(never)
public func test_private_global_var_int() -> Int {
return PVI + 1
}
// Loads from internal global variables can be removed if this is a WMO compilation, because
// they cannot be changed outside of this module.
// CHECK-WMO-LABEL: sil [noinline] @$s28globalopt_global_propagation014test_internal_B11_var_doubleSdyF
// CHECK-WMO: bb0:
// CHECK-WMO-NOT: global_addr
// CHECK-WMO: float_literal
// CHECK-WMO: struct
// CHECK-WMO: return
@inline(never)
public func test_internal_global_var_double() -> Double {
return IVD + 1.0
}
// Loads from internal global variables can be removed if this is a WMO compilation, because
// they cannot be changed outside of this module.
// CHECK-WMO-LABEL: sil [noinline] @$s28globalopt_global_propagation014test_internal_B8_var_intSiyF
// CHECK_WMO: bb0:
// CHECK-WMO-NOT: global_addr
// CHECK-WMO: integer_literal
// CHECK-WMO: struct
// CHECK_WMO: return
@inline(never)
public func test_internal_global_var_int() -> Int {
return IVI + 1
}
// Loads from public global variables cannot be removed, because their values could be changed elsewhere.
// CHECK-WMO-LABEL: sil [noinline] @$s28globalopt_global_propagation012test_public_B11_var_doubleSdyF
// CHECK-WMO: bb0:
// CHECK-WMO-NEXT: global_addr
// CHECK-WMO-NEXT: struct_element_addr
// CHECK-WMO-NEXT: load
@inline(never)
public func test_public_global_var_double() -> Double {
return VD + 1.0
}
// Loads from public global variables cannot be removed, because their values could be changed elsewhere.
// CHECK-LABEL: sil [noinline] @$s28globalopt_global_propagation012test_public_B8_var_intSiyF
// CHECK: bb0:
// CHECK-NEXT: global_addr
// CHECK-NEXT: struct_element_addr
// CHECK-NEXT: load
@inline(never)
public func test_public_global_var_int() -> Int {
return VI + 1
}
// Values of globals cannot be propagated as there are multiple assignments to it.
// CHECK-WMO-LABEL: sil [noinline] @$s28globalopt_global_propagation026test_internal_and_private_B25_var_with_two_assignmentsSiyF
// CHECK-WMO: bb0:
// CHECK-WMO: global_addr
// CHECK-WMO: global_addr
// CHECK-WMO: struct_element_addr
// CHECK-WMO: load
// CHECK-WMO: struct_element_addr
// CHECK-WMO: load
// CHECK-WMO: return
@inline(never)
public func test_internal_and_private_global_var_with_two_assignments() -> Int {
return IVIAssignTwice + PVIAssignTwice
}
// Values of globals cannot be propagated as their address was taken and
// therefore their value could have been changed elsewhere.
// CHECK-WMO-LABEL: sil @$s28globalopt_global_propagation05test_B13_take_addressSiyF
// CHECK-WMO: bb0:
// CHECK-WMO: global_addr
// CHECK-WMO: global_addr
// CHECK-WMO: struct_element_addr
// CHECK-WMO: load
// CHECK-WMO: struct_element_addr
// CHECK-WMO: load
// CHECK-WMO: return
public func test_global_take_address() -> Int {
takeInout(&PVITakenAddress)
takeInout(&IVITakenAddress)
return IVITakenAddress + PVITakenAddress
}
struct IntWrapper1 {
let val: Int
}
struct IntWrapper2 {
let val: IntWrapper1
}
struct IntWrapper3 {
let val: IntWrapper2
}
struct IntWrapper4 {
let val: IntWrapper2
let val2: IntWrapper1
}
let IW3 = IntWrapper3(val: IntWrapper2(val: IntWrapper1(val: 10)))
let IW4 = IntWrapper4(val: IntWrapper2(val: IntWrapper1(val: 10)), val2: IntWrapper1(val: 100))
// Test accessing single Int wrapped into multiple structs, where each struct has only one field.
// CHECK-LABEL: sil [noinline] @$s28globalopt_global_propagation34test_let_struct_wrapped_single_intSiyF
// CHECK: bb0:
// CHECK-NOT: global_addr
// CHECK: integer_literal
// CHECK: struct
// CHECK: return
// CHECK-WMO-LABEL: sil [noinline] @$s28globalopt_global_propagation34test_let_struct_wrapped_single_intSiyF
// CHECK-WMO: bb0:
// CHECK-WMO-NOT: global_addr
// CHECK-WMO: integer_literal
// CHECK-WMO: struct
// CHECK-WMO: return
@inline(never)
public func test_let_struct_wrapped_single_int() -> Int {
return IW3.val.val.val + 1
}
// Test accessing multiple Int fields wrapped into multiple structs, where each struct may have
// multiple fields.
// CHECK-LABEL: sil [noinline] @$s28globalopt_global_propagation37test_let_struct_wrapped_multiple_intsSiyF
// CHECK: bb0:
// CHECK-NOT: global_addr
// CHECK: integer_literal
// CHECK: struct
// CHECK: return
// CHECK-WMO-LABEL: sil [noinline] @$s28globalopt_global_propagation37test_let_struct_wrapped_multiple_intsSiyF
// CHECK-WMO: bb0:
// CHECK-WMO-NOT: global_addr
// CHECK-WMO: integer_literal
// CHECK-WMO: struct
// CHECK-WMO: return
@inline(never)
public func test_let_struct_wrapped_multiple_ints() -> Int {
return IW4.val.val.val + IW4.val2.val + 1
}
let IT1 = ((10, 20), 30, 40)
let IT2 = (100, 200, 300)
// Test accessing multiple Int fields wrapped into multiple tuples, where each tuple may have
// multiple fields.
// CHECK-LABEL: sil [noinline] @$s28globalopt_global_propagation27test_let_tuple_wrapped_intsSiyF
// CHECK: bb0:
// CHECK-NOT: global_addr
// CHECK: integer_literal
// CHECK: struct
// CHECK: return
// CHECK-WMO-LABEL: sil [noinline] @$s28globalopt_global_propagation27test_let_tuple_wrapped_intsSiyF
// CHECK-WMO: bb0:
// CHECK-WMO-NOT: global_addr
// CHECK-WMO: integer_literal
// CHECK-WMO: struct
// CHECK-WMO: return
@inline(never)
public func test_let_tuple_wrapped_ints() -> Int {
return IT1.0.0 + IT2.1
}