| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
This patch is a revision to CL 185518, which added code to perform
finalization of methods on types created by the importer and not
directly reachable until inlining is done.
The original fix invoked the finalization code at the end of
Import::read_types(), but it turns out this doesn't handle the case
where a type with methods is read in due to a reference from something
later in the export data (a function or variable). The fix is to move
the import finalization call to the end of Import::import().
Testcase for this bug is in CL 187057.
Fixes golang/go#33219.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/187058
From-SVN: r273713
|
|
The current Mark_inline_candidates helper looks only at budget when
deciding to mark a function or method as inline (with the proviso that
IR constructs not yet supported by the inliner are given artificially
high cost). This patch changes the helper to also look at whether a
function has the "go:noinline" pragma; if it does have the pragma
there is no point putting it into the export data (it will just make
the export data bigger).
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/186923
From-SVN: r273611
|
|
This patch fixes a buglet in the function body importer. Add hooks for
keeping a stack of blocks corresponding to the block nesting in the
imported function. This ensures that local variables and temps wind up
correctly scoped and don't introduce collisions.
New test case for this problem in CL 186717.
Fixes golang/go#33158.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/186757
From-SVN: r273577
|
|
In CL 183850 a change was made to combine tracking/discovery of
exported types and imported packages during export data generation. As
a result of this refactoring a bug was introduced: the new code can
potentially insert items into the exports set (an unordered_set) while
iterating through the same set, which is illegal according to the spec
for std::unordered_set.
This patch fixes the problem by changing the type discovery phase to
iterate through a separate list of sorted exports, as opposed to
iterating through the main unordered set. Also included is a change
to fix the code that looks for variables that are referenced from
inlined routine bodies (this code wasn't scanning all of the function
that it needed to scan).
New test case for this problem in CL 186697.
Updates golang/go#33020.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/185977
From-SVN: r273564
|
|
Currently, getg is implemented in C, which loads the thread-local
g variable. The g variable is declared static in C.
This CL exposes the g variable, so it can be accessed from the Go
side. This allows the Go compiler to inline getg calls to direct
access of g.
Currently, the actual inlining is only implemented in the gollvm
compiler. The g variable is thread-local and the compiler backend
may choose to cache the TLS address in a register or on stack. If
a thread switch happens the cache may become invalid. I don't
know how to disable the TLS address cache in gccgo, therefore
the inlining of getg is not implemented. In the future gccgo may
gain this if we know how to do it safely.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/186238
From-SVN: r273499
|
|
The type hash and equality functions are generated after the
order_evaluations pass. They may contain shortcut operators and
Set_and_use_temporary_expressions (e.g. from lowering a
Binary_exprssion) that need to be ordered. Run order_evaluations
and remove_shortcuts on these functions. (The hash functions
may be fine, but to be on the safe side we run on them anyway.
We do need to run on the equality functions.)
A Set_and_use_temporary_expression is effectively an assignment,
so it needs to be ordered. Otherwise if we insert a temporary
statement before it, we may get wrong evaluation order.
A test case is CL 185818.
Fixes golang/go#33062.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/185817
From-SVN: r273425
|
|
This patch changes the compiler to be more aggressive about finalizing
methods on imported types, to avoid problems with interface types that
are imported but remain unreachable until a later stage in the compilation.
The normal pattern prior to this change was that the import process would
leave imported interface types alone, and rely on Gogo::finalize_methods
to locate and finalize all interface types at a later point. This way
of doing things was not working in all cases due to the fact that we can
import an interface type that is only reachable from the body of an
inlinable function, meaning that we can't "find" the type during
the methods finalize phase.
The importer's Import::read_types() now makes a pass over all imported
types to finalize methods on any newly imported type, which takes care
of the issue.
New test case for this problem in CL 185517.
Fixes golang/go#33013
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/185518
From-SVN: r273364
|
|
CL 184998 added optimizations for one- and two-case select
statements. But it didn't handle break statement in the select
case correctly. Specifically, it didn't add the label definition,
so it could result in a dangling goto. This CL fixes this, by
adding the label definition.
A test case is CL 185520.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/185519
From-SVN: r273359
|
|
For a select statement with zero-, one-, or two-case with a
default case, we can generate simpler code instead of calling the
generic selectgo. A zero-case select is just blocking the
execution. A one-case select is mostly just executing the case. A
two-case select with a default case is a non-blocking send or
receive. We add these special cases for lowering a select
statement.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184998
From-SVN: r273034
|
|
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184997
From-SVN: r273032
|
|
Fix compiler buglet: varargs lowering happens before inlinable
function bodies are written out to export data, so set the "varargs
lowered" flag on call expressions that we import.
Fixes golang/go#32922
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184919
From-SVN: r273026
|
|
In CL 179598, we were using Gogo::packages_, when compiling the
main package, as the list of packages of which we need to
register the type descriptors. This is not complete. It only
includes main's direct import and one-level indirect imports. It
does not include all the packages transitively imported.
To fix that, we need to track all the transitive imports. We
have almost already done that, for init functions. However, there
may be packages that don't need init functions but do need to
register type descriptors. For them, we add a dummy init function
to its export data. So when we compile the main package we will
see all the transitive imports. The dummy init functions are not
real functions and are not called.
Fixes golang/go#32901.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184717
From-SVN: r273009
|
|
Revamps the way the exporter tracks exported types and imported
packages that need to be mentioned in the export data.
The previous implementation wasn't properly handling the case where an
exported non-inlinable function refers to an imported type whose
method set includes an inlinable function whose body makes a call to a
function in another package that's not directly used in the original
package.
This patch integrates together two existing traversal helper classes,
"Collect_references_from_inline" and "Find_types_to_prepare" into a
single helper "Collect_export_references", so as to have common/shared
code that looks for indirectly imported packages.
Fixes golang/go#32778
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/183850
From-SVN: r272955
|
|
For zeroing a range of memory that doesn't contain pointer, we
can use builtin memset directly.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184438
* go-gcc.cc (Gcc_backend::Gcc_backend): Define __builtin_memset.
From-SVN: r272944
|
|
Convert the Export::type_refs map from a static object to a field
contained (indirectly, via an impl class) in Export itself, for better
encapsulation and to be able to reclaim its memory when exporting is
done. No change in compiler functionality.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/184170
From-SVN: r272919
|
|
On AIX, a function has two symbols, a text symbol (with a leading dot)
and a data one (without it).
As the tests must be run only once, only the data symbol can be used to
retrieve the final go symbol. Therefore, all symbols beginning with a dot
are ignored by symtogo.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/177837
From-SVN: r272666
|
|
The first call of ar must not show its output in order to avoid useless
error messages about D flag.
The corresponding Go toolchain patch is CL 182077.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/183817
From-SVN: r272661
|
|
They are wrappers of libc functions that use no stack. Mark them
nosplit so the linker won't patch it to call __morestack_non_split.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/183629
From-SVN: r272633
|
|
Open code string equality with builtin memcmp. This allows
further optimizations in the backend.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/183538
From-SVN: r272624
|
|
Instead of going through a C function __go_memcmp, we can just
use __builtin_memcmp directly. This allows more optimizations in
the compiler backend.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/183537
From-SVN: r272620
|
|
GCC recently introduced -Wformat-diag to scrutinize GCC error messages.
It reports a number of warnings about gofrontend code, such as:
../../trunk/gcc/go/gofrontend/import.cc: In member function ‘Type* Import::type_for_index(int, const string&, size_t, bool*)’:
../../trunk/gcc/go/gofrontend/import.cc:1129:48: warning: unquoted operator ‘>=’ in format [-Wformat-diag]
1129 | "error in %s at %lu: bad type index %d >= %d",
| ^~
../../trunk/gcc/go/gofrontend/ast-dump.cc: In member function ‘void Ast_dump_context::dump(Gogo*, const char*)’:
../../trunk/gcc/go/gofrontend/ast-dump.cc:203:25: warning: unquoted option name ‘-fgo-dump-ast’ in format [-Wformat-diag]
203 | "cannot open %s:%m, -fgo-dump-ast ignored", dumpname.c_str());
| ^~~~~~~~~~~~~
../../trunk/gcc/go/gofrontend/expressions.cc: In static member function ‘static Bexpression* Func_expression::get_code_pointer(Gogo*, Named_object*, Location)’:
../../trunk/gcc/go/gofrontend/expressions.cc:1350:29: warning: misspelled term ‘builtin function’ in format; use ‘built-in function’ instead [-Wformat-diag]
1350 | "invalid use of special builtin function %qs; must be called",
| ^~~~~~~~~~~~~~~~
../../trunk/gcc/go/gofrontend/gogo.cc: In member function ‘void Gogo::add_linkname(const string&, bool, const string&, Location)’:
../../trunk/gcc/go/gofrontend/gogo.cc:2527:4: warning: unquoted sequence of 2 consecutive punctuation characters ‘//’ in format [-Wformat-diag]
2527 | ("%s is not a function; "
| ~^~~~~~~~~~~~~~~~~~~~~~~~
2528 | "//go:linkname is only supported for functions"),
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
This CL edits error messages to avoid these warnings.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/183497
* go.test/test/blank1.go: Update for diagnostic message changes.
From-SVN: r272608
|
|
GCC recently added a new warning -Wformat-diag which does a lot of
rigorous checks on GCC diagnostic messages. This produces a number of
unnecessary diagnostics on gofrontend diagnostic output, such as
../../trunk/gcc/go/gofrontend/escape.cc: In member function ‘virtual int Escape_analysis_assign::statement(Block*, size_t*, Statement*)’:
../../trunk/gcc/go/gofrontend/escape.cc:1336:33: warning: spurious leading punctuation sequence ‘[’ in format [-Wformat-diag]
1336 | go_inform(s->location(), "[%d] %s esc: %s",
| ^
../../trunk/gcc/go/gofrontend/escape.cc: In member function ‘void Escape_analysis_assign::call(Call_expression*)’:
../../trunk/gcc/go/gofrontend/escape.cc:1964:17: warning: unquoted operator ‘::’ in format [-Wformat-diag]
1964 | "esccall:: indirect call <- %s, untracked",
| ^~
../../trunk/gcc/go/gofrontend/escape.cc:1964:34: warning: unbalanced punctuation character ‘<’ in format [-Wformat-diag]
1964 | "esccall:: indirect call <- %s, untracked",
| ^
Avoid these messages by adding a new function go_debug that uses only
printf formatting, not GCC diagnostic formatting, and change all the
optimization debugging messages to use it. None of the debugging
messages used the GCC diagnostic formatting specifiers anyhow.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/183437
From-SVN: r272607
|
|
Let the Go frontend recognize some math/bits functions and turn
them into intrinsics.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/183266
* go-gcc.cc (Gcc_backend::Gcc_backend): Define math/bits
builtins.
From-SVN: r272579
|
|
Now that type equality is just a pointer equality, write it
inlined and remove the eqtype function.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/182978
From-SVN: r272578
|
|
Now that type equality is just simple pointer equality, we can
open code some type assertions instead of making runtime calls.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/182977
From-SVN: r272577
|
|
Assignments to local variables don't need a write barrier. But
currently the compiler inserts a write barrier if the LHS is a
local variable with type converted, as *(convert(&local)). Let
the compiler recognize this pattern and omit the write barrier.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/182541
From-SVN: r272550
|
|
Currently a string slice expression is implemented with a runtime
call __go_string_slice. Change it to open code it, which is more
efficient, and allows the backend to further optimize it.
Also omit the write barrier for length-only update (i.e.
s = s[:n]).
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/182540
From-SVN: r272549
|
|
runtime.concatstring{2,3,4,5} are just wrappers of concatstrings.
These wrappers don't provide any benefit, at least in the C
calling convention we use, where passing arrays by value isn't an
efficient thing. Change it to always use concatstrings.
Also, the cap field of the slice passed to concatstrings is not
necessary. So change it to pass a pointer and a length directly,
which is more efficient than passing a slice header by value.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/182539
From-SVN: r272476
|
|
For string concatenation, string to/from byte or rune slice
conversion, and int to string conversion, if the result does not
escape, we can allocate a small (32-element, or 4-byte for int to
string) buffer on stack, and pass it to the runtime function. If
the result fits in the buffer, it doesn't need to do a heap
allocation.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/182538
From-SVN: r272468
|
|
If a string([]byte) conversion is used immediately in a string
concatenation, we don't need to copy the backing store of the
byte slice, as the runtime function doesn't hold any reference
to it.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/182437
* go.dg/concatstring.go: New test.
From-SVN: r272460
|
|
Also move the determine_types pass on an inlined function body to one
place, rather than doing it ad hoc as needed.
This adds 79 new inlinable functions in the standard library, such as
bytes.HasPrefix and bytes.LastIndexByte.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181261
From-SVN: r272133
|
|
Heap_expression::do_get_backend emits an unconditional write
barrier if the type has pointers and it is not a stack allocation.
This CL changes it to use a write barrier for the assignment only
when write barriers are enabled. While here, also change it to
call gcWriteBarrier instead of typedmemmove for pointer-shaped
types.
For this to work, Function::build needs to be adjusted so that
Heap_expression::do_get_backend is called when there is a parent
block.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181540
From-SVN: r272132
|
|
This permits inlining functions with for loops and some switches, as
they are lowered to if and goto statements before exporting them.
This by itself only adds three new inlinable functions in the standard
library: sort.Search, context.(*emptyCtx).String, and
cmd/go/internal/work.(*Builder).disableBuildID.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181197
From-SVN: r272131
|
|
If a struct/array is pointer-shaped (i.e. having a single field
that is pointer-shaped), we can use gcWriteBarrier instead of
typedmemmove for the write barrier.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181539
From-SVN: r272130
|
|
The escape analysis was written before we import inlineable
function bodies, and in some places it skipped functions that are
not in the local package. Now that there are imported function
bodies, make the escape analysis work with them.
Note that it is necessary for the escape analysis to run on
imported function bodies, even if they are already tagged. The
tags only have the information of the parameters (receiver,
results), but not the internal nodes, e.g. local variables. We
still need to do the analysis to get all the information. (In the
future maybe we could export/import escape info for internal
nodes also, then we don't need to redo the analysis.)
Also add assertions to ensure that if we analyze the same
function in multiple places, they'd better agree with each other.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181537
From-SVN: r272124
|
|
For string, slice, interface values, do assignments field by
field instead of using typedmemmove.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181297
From-SVN: r272055
|
|
This increases the number of inlinable functions from 455 to 500.
An example of a newly inlinable function is strings.Compare.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181137
From-SVN: r272045
|
|
Normally the backend will do deadcode elimination and this is
sufficient. However, the escape analysis operates on the AST that
may have deadcode, and may cause things to escape that otherwise
do not.
This CL adds a simple deadcode elimination, run before the escape
analysis.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/181080
From-SVN: r272043
|
|
Due to inlining, we can now see unexported functions and variables,
and functions and variables imported from different packages.
Ignore them rather than reporting them from this package.
Handle $hash and $equal functions consistently, so that we discard the
inline body if there is one.
Ignore names created for result parameters for inlining purposes.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/180758
From-SVN: r272023
|
|
This increases the number of inlinable functions from 439 to 455.
An example is math/bits.Mul32, which uses temporaries to handle the
tuple assignment.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/180837
From-SVN: r272022
|
|
In the runtime there are specialized fast map routines for
certain kep types. This CL lets the compiler make use of these
functions, instead of always using the generic ones.
As we now generate multiple versions of map delete calls, to make
things easier we delay the expansion of the built-in delete
function to flatten phase.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/180858
From-SVN: r271983
|
|
Scan inlinable methods for references to global variables and
functions (forgot to do that earlier).
Track all packages mentioned by exports (that should have been done
earlier too).
Record assembler name in export data, so that we can inline calls to
non-Go functions. Modify gccgoimporter code to skip assembler name.
This increases the number of inlinable functions in the standard
library from 215 to 439.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/180677
From-SVN: r271976
|
|
When converting a constant to interface, such as interface{}(42)
or interface{}("hello"), if the interface escapes, we currently
generate a heap allocation to hold the constant value.
This CL changes it to generate a static allocation instead, as
the gc compiler does. This reduces allocations in such cases.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/180277
From-SVN: r271945
|
|
Currently, the compiler already generates common symbols for type
descriptors, so the type descriptors are unique. However, when a
type is created through reflection, it is not deduplicated with
compiler-generated types. As a consequence, we cannot assume type
descriptors are unique, and cannot use pointer equality to
compare them. Also, when constructing a reflect.Type, it has to
go through a canonicalization map, which introduces overhead to
reflect.TypeOf, and lock contentions in concurrent programs.
In order for the reflect package to deduplicate types with
compiler-created types, we register all the compiler-created type
descriptors at startup time. The reflect package, when it needs
to create a type, looks up the registry of compiler-created types
before creates a new one. There is no lock contention since the
registry is read-only after initialization.
This lets us get rid of the canonicalization map, and also makes
it possible to compare type descriptors with pointer equality.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179598
From-SVN: r271894
|
|
When the runtime collects a stack trace to associate it with some
profiling event (mem alloc, mutex, etc) there is a skip count passed
to runtime.Callers (or equivalent) to skip some known count of frames
in order to get to the "interesting" frame corresponding to the
profile event. Now that the profiling mechanism uses lazy fixup (when
removing compiler artifacts like thunks, morestack calls etc), we also
need to move the frame skipping logic after the fixup, so as to insure
that the skip count isn't thrown off by these artifacts.
Fixes golang/go#32290.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179740
From-SVN: r271892
|
|
This requires tracking all references to unexported variables, so that
we can make them global symbols in the object file, and can export
them so that other compilations can see the right definition for their
own inline bodies.
This introduces a syntax for referencing names defined in other
packages: a <pNN> prefix, where NN is the package index. This will
need to be added to gccgoimporter, but I didn't do it yet since it
isn't yet possible to create an object for which gccgoimporter will
see a <pNN> prefix.
This increases the number of inlinable functions in the standard
library from 181 to 215, adding functions like context.Background.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/177920
From-SVN: r271891
|
|
These functions were needed during the transition of the runtime from
C to Go, but are no longer necessary.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179879
From-SVN: r271890
|
|
Some assembler doesn't accept ULL suffix. In fact the suffix
is not really necessary. Drop it.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/180217
From-SVN: r271883
|
|
In particular, drop __go_type_descriptors_equal, which is no longer
used, and will be made obsolete by CL 179598.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179858
From-SVN: r271823
|
|
The gc compiler recognizes append(s, make([]T, n)...), and
generates code to directly zero the tail instead of allocating a
new slice and copying. This CL lets the Go frontend do basically
the same.
The difficulty is that at the point we handle append, there may
already be temporaries introduced (e.g. in order_evaluations),
which makes it hard to find the append-of-make pattern. The
compiler could "see through" the value of a temporary, but it is
only safe to do if the temporary is not assigned multiple times.
For this, we add tracking of assignments and uses for temporaries.
This also helps in optimizing non-escape slice make. We already
optimize non-escape slice make with constant len/cap to stack
allocation. But it failed to handle things like f(make([]T, n))
(where the slice doesn't escape and n is constant), because of
the temporary. With tracking of temporary assignments and uses,
it can handle this now as well.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179597
From-SVN: r271822
|
