There may be further savings if convT2I can avoid the function call
if the cache is good and T is uintptr-shaped, a la convT2E, but that
will be a follow-up CL.
src/pkg/runtime:
benchmark old ns/op new ns/op delta
BenchmarkConvT2ISmall 43 15 -64.01%
BenchmarkConvT2IUintptr 45 14 -67.48%
BenchmarkConvT2ILarge 130 101 -22.31%
test/bench/go1:
benchmark old ns/op new ns/op delta
BenchmarkBinaryTree17 8588997000 8499058000 -1.05%
BenchmarkFannkuch11 5300392000 5358093000 +1.09%
BenchmarkGobDecode 30295580 31040190 +2.46%
BenchmarkGobEncode 18102070 17675650 -2.36%
BenchmarkGzip 774191400 771591400 -0.34%
BenchmarkGunzip 245915100 247464100 +0.63%
BenchmarkJSONEncode 123577000 121423050 -1.74%
BenchmarkJSONDecode 451969800 596256200 +31.92%
BenchmarkMandelbrot200 10060050 10072880 +0.13%
BenchmarkParse 10989840 11037710 +0.44%
BenchmarkRevcomp 1782666000 1716864000 -3.69%
BenchmarkTemplate 798286600 723234400 -9.40%
R=rsc, bradfitz, go.peter.90, daniel.morsing, dave, uriel
CC=golang-dev
https://golang.org/cl/6337058
CL 4313064 fixed its test case but did not address a
general enough problem:
type T1 struct { F *T2 }
type T2 T1
type T3 T2
could still end up copying the definition of T1 for T2
before T1 was done being evaluated, or T3 before T2
was done.
In order to propagate the updates correctly,
record a copy of an incomplete type for re-execution
once the type is completed. Roll back CL 4313064.
Fixes#3709.
R=ken2
CC=golang-dev, lstoakes
https://golang.org/cl/6301059
Drop expecttaken function in favor of extra argument
to gbranch and bgen. Mark loop condition as likely to
be true, so that loops are generated inline.
The main benefit here is contiguous code when trying
to read the generated assembly. It has only minor effects
on the timing, and they mostly cancel the minor effects
that aligning function entry points had. One exception:
both changes made Fannkuch faster.
Compared to before CL 6244066 (before aligned functions)
benchmark old ns/op new ns/op delta
BenchmarkBinaryTree17 4222117400 4201958800 -0.48%
BenchmarkFannkuch11 3462631800 3215908600 -7.13%
BenchmarkGobDecode 20887622 20899164 +0.06%
BenchmarkGobEncode 9548772 9439083 -1.15%
BenchmarkGzip 151687 152060 +0.25%
BenchmarkGunzip 8742 8711 -0.35%
BenchmarkJSONEncode 62730560 62686700 -0.07%
BenchmarkJSONDecode 252569180 252368960 -0.08%
BenchmarkMandelbrot200 5267599 5252531 -0.29%
BenchmarkRevcomp25M 980813500 985248400 +0.45%
BenchmarkTemplate 361259100 357414680 -1.06%
Compared to tip (aligned functions):
benchmark old ns/op new ns/op delta
BenchmarkBinaryTree17 4140739800 4201958800 +1.48%
BenchmarkFannkuch11 3259914400 3215908600 -1.35%
BenchmarkGobDecode 20620222 20899164 +1.35%
BenchmarkGobEncode 9384886 9439083 +0.58%
BenchmarkGzip 150333 152060 +1.15%
BenchmarkGunzip 8741 8711 -0.34%
BenchmarkJSONEncode 65210990 62686700 -3.87%
BenchmarkJSONDecode 249394860 252368960 +1.19%
BenchmarkMandelbrot200 5273394 5252531 -0.40%
BenchmarkRevcomp25M 996013800 985248400 -1.08%
BenchmarkTemplate 360620840 357414680 -0.89%
R=ken2
CC=golang-dev
https://golang.org/cl/6245069
* Eliminate bounds check on known small shifts.
* Rewrite x<<s | x>>(32-s) as a rotate (constant s).
* More aggressive (but still minimal) range analysis.
R=ken, dave, iant
CC=golang-dev
https://golang.org/cl/6209077
I tried before to make relative imports work by simply
invoking the compiler in the right directory, so that
an import of ./foo could be resolved by ./foo.a.
This required creating a separate tree of package binaries
that included the full path to the source directory, so that
/home/gopher/bar.go would be compiled in
tmpdir/work/local/home/gopher and perhaps find
a ./foo.a in that directory.
This model breaks on Windows because : appears in path
names but cannot be used in subdirectory names, and I
missed one or two places where it needed to be removed.
The model breaks more fundamentally when compiling
a test of a package that lives outside the Go path, because
we effectively use a ./ import in the generated testmain,
but there we want to be able to resolve the ./ import
of the test package to one directory and all the other ./
imports to a different directory. Piggybacking on the compiler's
current working directory is then no longer possible.
Instead, introduce a new compiler option -D prefix that
makes the compiler turn a ./ import into prefix+that,
so that import "./foo" with -D a/b/c turns into import
"a/b/c/foo". Then we can invent a package hierarchy
"_/" with subdirectories named for file system paths:
import "./foo" in the directory /home/gopher becomes
import "_/home/gopher/foo", and since that final path
is just an ordinary import now, all the ordinary processing
works, without special cases.
We will have to change the name of the hierarchy if we
ever decide to introduce a standard package with import
path "_", but that seems unlikely, and the detail is known
only in temporary packages that get thrown away at the
end of a build.
Fixes#3169.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/5732045
Also allow multiple invalid import statements in a
single file.
Fixes#3021. The changes to go/parser and the
language specifcation have already been committed.
R=rsc, gri
CC=golang-dev
https://golang.org/cl/5672084
This is a manual undo of CL 5674098.
It does not implement the even less strict spec
that we just agreed on, but it gets us back where
we were at the last weekly.
R=ken2
CC=golang-dev
https://golang.org/cl/5683069
unsafe: delete Typeof, Reflect, Unreflect, New, NewArray
Part of issue 2955 and issue 2968.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/5650069
If the values being compared have different concrete types,
then they're clearly unequal without needing to invoke the
actual interface compare routine. This speeds tests for
specific values, like if err == io.EOF, by about 3x.
benchmark old ns/op new ns/op delta
BenchmarkIfaceCmp100 843 287 -65.95%
BenchmarkIfaceCmpNil100 184 182 -1.09%
Fixes#2591.
R=ken2
CC=golang-dev
https://golang.org/cl/5651073
This can happen on Plan 9 if we we're building
with the 32-bit and 64-bit host compilers, one
after the other.
R=rsc
CC=golang-dev
https://golang.org/cl/5599053
Pulling function calls out to happen before the
expression being evaluated was causing illegal
reorderings even without inlining; with inlining
it got worse. This CL adds a separate ordering pass
to move things with a fixed order out of expressions
and into the statement sequence, where they will
not be reordered by walk.
Replaces lvd's CL 5534079.
Fixes#2740.
R=lvd
CC=golang-dev
https://golang.org/cl/5569062
flag -l means: inlining on, -ll inline with early typecheck
-l lazily typechecks imports on use and re-export, nicer for debugging
-lm produces output suitable for errchk tests, repeated -mm... increases inl.c's verbosity
export processed constants, instead of originals
outparams get ->inlvar too, and initialized to zero
fix shared rlist bug, that lead to typecheck messing up the patched tree
properly handle non-method calls to methods T.meth(t, a...)
removed embryonic code to handle closures in inlined bodies
also inline calls inside closures (todo: move from phase 6b to 4)
Fixes#2579.
R=rsc
CC=golang-dev
https://golang.org/cl/5489106
Cross- and intra package inlining of single assignments or return <expression>.
Minus some hairy cases, currently including other calls, expressions with closures and ... arguments.
R=rsc, rogpeppe, adg, gri
CC=golang-dev
https://golang.org/cl/5400043
To allow these types as map keys, we must fill in
equal and hash functions in their algorithm tables.
Structs or arrays that are "just memory", like [2]int,
can and do continue to use the AMEM algorithm.
Structs or arrays that contain special values like
strings or interface values use generated functions
for both equal and hash.
The runtime helper func runtime.equal(t, x, y) bool handles
the general equality case for x == y and calls out to
the equal implementation in the algorithm table.
For short values (<= 4 struct fields or array elements),
the sequence of elementwise comparisons is inlined
instead of calling runtime.equal.
R=ken, mpimenov
CC=golang-dev
https://golang.org/cl/5451105
Equality on structs will require arbitrary code for type equality,
so change algorithm in type data from uint8 to table pointer.
In the process, trim top-level map structure from
104/80 bytes (64-bit/32-bit) to 24/12.
Equality on structs will require being able to call code generated
by the Go compiler, and C code has no way to access Go return
values, so change the hash and equal algorithm functions to take
a pointer to a result instead of returning the result.
R=ken
CC=golang-dev
https://golang.org/cl/5453043
I've modified Plan 9's yacc to work with
the grammar in go.y. These are the only
changes necessary on the Go side.
R=rsc
CC=golang-dev
https://golang.org/cl/5375104
An experiment: allow structs to be copied even if they
contain unexported fields. This gives packages the
ability to return opaque values in their APIs, like reflect
does for reflect.Value but without the kludgy hacks reflect
resorts to.
In general, we trust programmers not to do silly things
like *x = *y on a package's struct pointers, just as we trust
programmers not to do unicode.Letter = unicode.Digit,
but packages that want a harder guarantee can introduce
an extra level of indirection, like in the changes to os.File
in this CL or by using an interface type.
All in one CL so that it can be rolled back more easily if
we decide this is a bad idea.
Originally discussed in March 2011.
https://groups.google.com/group/golang-dev/t/3f5d30938c7c45ef
R=golang-dev, adg, dvyukov, r, bradfitz, jan.mercl, gri
CC=golang-dev
https://golang.org/cl/5372095
I promised it wouldn't last very long.
People who really need this can sync to 6a5647d82728.
««« original CL description
gc: add GOEXPERIMENT=os.Error
This won't last long, I promise.
R=ken2
CC=golang-dev
https://golang.org/cl/5330066
»»»
R=ken2
CC=golang-dev
https://golang.org/cl/5333053
Got rid of all the magic mystery globals. Now
for %N, %T, and %S, the flags +,- and # set a sticky
debug, sym and export mode, only visible in the new fmt.c.
Default is error mode. Handle h and l flags consistently with
the least side effects, so we can now change
things without worrying about unrelated things
breaking.
fixes#2361
R=rsc
CC=golang-dev
https://golang.org/cl/5316043
There is no semantic change here, just better errors.
If a function says it takes a byte, and you pass it an int,
the compiler error now says that you need a byte, not
that you need a uint8.
Groundwork for rune.
R=ken2
CC=golang-dev
https://golang.org/cl/5300042
The old m[x] = 0, false syntax will be deleted
in a month or so, once people have had time to
change their code (there is a gofix in a separate CL).
R=ken2
CC=golang-dev
https://golang.org/cl/5265048
