-
Notifications
You must be signed in to change notification settings - Fork 17.7k
CompilerOptimizations
This page lists optimizations done by the compilers. Note that these are not guaranteed by the language specification.
Putting a zero-width type in an interface value doesn't allocate.
- gc: 1.0+
- gccgo: ?
Putting a word-sized-or-less non-pointer type in an interface value doesn't allocate.
- gc: 1.0-1.3, but not in 1.4+
- gccgo: never
For a map m
of type map[string]T
and []byte b
, m[string(b)]
doesn't allocate. (the temporary string copy of the byte slice isn't made)
- gc: 1.4+
- gccgo: ?
No allocation when converting a string
into a []byte
for ranging over the bytes:
s := "foo"
for i, c := range []byte(s) {
// ...
}
No allocation done when converting a []byte
into a string
for comparison purposes
var b1 string
var b2 []byte
var x = string(b1) == string(b2) // memeq
var y = string(b1) < string(b2) // lexicographical comparison
- gc: 1.5+ (CL 3790)
- gccgo: ?
Use -gcflags -m
to observe the result of escape analysis and inlining
decisions for the gc toolchain.
(TODO: explain the output of -gcflags -m
).
Gc compiler does global escape analysis across function and package boundaries. However, there are lots of cases where it gives up. For example, anything assigned to any kind of indirection (*p = ...
) is considered escaped. Other things that can inhibit analysis are: function calls, package boundaries, slice literals, subslicing and indexing, etc. Full rules are too complex to describe, so check the -m
output.
- gc: 1.0+
- gccgo: 8.0+.
Only short and simple functions are inlined. To be inlined a function must conform to the rules:
- function should be simple enough, the number of AST nodes must less than the budget (80);
- function doesn't contain complex things like closures, defer, recover, select, etc;
- function isn't prefixed by go:noinline;
- function isn't prefixed by go:uintptrescapes, since the escape information will be lost during inlining;
- function has body;
- etc.
- gc: 1.0+
- gccgo: -O1 and above.
For a slice or array s, loops of the form
for i := range s {
s[i] = <zero value for element of s>
}
are converted into efficient runtime memclr calls. Issue and commit.
- gc: 1.5+
- gccgo: ?
Garbage collector does not scan underlying buffers of slices, channels and maps when element type does not contain pointers (both key and value for maps). This allows to hold large data sets in memory without paying high price during garbage collection. For example, the following map won't visibly affect GC time:
type Key [64]byte // SHA-512 hash
type Value struct {
Name [32]byte
Balance uint64
Timestamp int64
}
m := make(map[Key]Value, 1e8)
- gc: 1.5+
- gccgo: ?