cannot allocate an audomatic temp
while real registers are allocated.
there is a chance that the automatic
will be allocated to one of the
allocated registers. the fix is to
not registerize such variables.
R=rsc
CC=golang-dev
https://golang.org/cl/1202042
Time to draw.Draw a 200x200 image fell from 18.4ms (and 1 malloc) to
5.6ms (and 0 mallocs). It's still relatively slow since it assumes
nothing about the src or mask images, but it does remove the malloc.
There are existing faster, more specialized paths for copies, fills
and image glyph masks.
Also added a "compare to a slow but obviously correct implementation"
check to draw_test.go.
R=rsc, r
CC=golang-dev
https://golang.org/cl/1223044
- implemented setWord, use it where setUint64 is wrong
- divLarge: use fast mulWW, divWW; implemented mulWW, divWW
- better assembly code for addMulVVW
R=rsc
CC=golang-dev
https://golang.org/cl/1258042
Previously the compiler would just emit "internal compiler error" when
trying to compute the remainder of floats or complex types.
R=rsc
CC=golang-dev
https://golang.org/cl/1243041
- support for binary prefix 0b (to match fmt.Format)
- renamed nat.new -> nat.setUint64 for consistency
- more tests
R=r
CC=golang-dev
https://golang.org/cl/1233041
Also update range of Phase and Polar due to signed zero.
[Phase(cmplx(-1, +0)) = pi and Phase(cmplx(-1, -0)) = -pi]
R=rsc, r
CC=golang-dev
https://golang.org/cl/1235041
Timings (as for change 1122043) go from 49ms to 48ms ish. It's
mostly lost in the noise, but it probably doesn't hurt.
R=r
CC=golang-dev
https://golang.org/cl/1179041
To draw.Draw a 32x32 image.Alpha 10000 times,
Before: 633ms with 10000 mallocs
After: 49ms with 0 mallocs
These times are just blitting an image.Alpha, and do not include
rasterizing a glyph's vector contours to an image.Alpha.
The "generic" test case in draw_test.go tests this fast path.
R=rsc
CC=golang-dev
https://golang.org/cl/1122043
- removed last argument (n) from all core arithmetic routines;
instead, use the length of the result
- simplified nat.make implementation and chose a better capacity
for new values, removed a TODO in the process
Changing the constant e from 1 (old) to 4 (new) improved
pidigits -s -n 10000 by ~9% (on a 3.06GHz Intel Core 2 Duo):
user 0m3.882s (old)
user 0m3.549s (new)
R=rsc
CC=golang-dev
https://golang.org/cl/1133043
This permits cgo callbacks to work when run in init code.
Otherwise cgocallback switches to the wrong stack address.
R=rsc
CC=golang-dev
https://golang.org/cl/1123043
- no need to make copies in cases of aliases
- removed deprecated internal shift functions
- minor unrelated simplifications
This change improves pidigits -s -n10000 by almost 20%:
user 0m6.156s (old)
user 0m4.999s (new)
(pidigits -s -n20000 goes from ~25s to ~19s)
R=rsc
CC=golang-dev
https://golang.org/cl/1149041
Because maps are mostly a hidden type, they must be
implemented using reflection values and will not be as
efficient as arrays and slices.
R=rsc
CC=golang-dev
https://golang.org/cl/1127041
- renamed Len -> BitLen, simplified implementation
- renamed old Div, Mod, DivMod -> Que, Rem, QuoRem
- implemented Div, Mod, DivMod (Euclidian definition, more
useful in a mathematical context)
- fixed a bug in Exp (-0 was possible)
- added extra tests to check normalized results everywhere
- uniformly set Int.neg flag at the end of computations
- minor cosmetic cleanups
- ran all tests
R=rsc
CC=golang-dev
https://golang.org/cl/1091041
Import _mulv from Inferno again, change R9 to R2.
Not sure what the other differences were for, but
they weren't working.
TBR=kaib
CC=golang-dev
https://golang.org/cl/1079041
built. Note that they are only built if a //export comment
appears, so other targets should only use them if there is
such a comment.
Fixes#723.
R=rsc
CC=golang-dev
https://golang.org/cl/1073041
When trying to regenerate src/pkg/runtime/darwin/386/defs.h
on a 64 bit capable Snow Leopard (OS X 10.6.3) system I
needed to add -f -m32 to godefs, as this OS and hardware
combination defaults to 64 bit compilation.
For safety, make the same change to the 32 bit FreeBSD
instructions in .../freebsd/defs.c. (Tested OK and no
problems introduced.)
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/1052042
- fixed a couple of bugs in the process
(shift right was incorrect for negative numbers)
- added more tests and made some tests more robust
- changed pidigits back to using shifts to multiply
by 2 instead of add
This improves pidigit -s -n 10000 by approx. 5%:
user 0m6.496s (old)
user 0m6.156s (new)
R=rsc
CC=golang-dev
https://golang.org/cl/963044
This results in an improvement of > 35% for the existing Mul benchmark
using the same karatsuba threshold, and an improvement of > 50% with
a slightly higher threshold (32 instead of 30):
big.BenchmarkMul 500 6731846 ns/op (old alg.)
big.BenchmarkMul 500 4351122 ns/op (new alg.)
big.BenchmarkMul 500 3133782 ns/op (new alg., new theshold)
Also:
- tweaked calibrate.go, use same benchmark as for Mul benchmark
R=rsc
CC=golang-dev
https://golang.org/cl/1037041
Plus:
- calibration "test" - include in tests with gotest -calibrate
- basic Mul benchmark
- extra multiplication tests
- various cleanups
This change improves multiplication speed of numbers >= 30 words
in length (current threshold; found empirically with calibrate):
The multiplication benchmark (multiplication of a variety of long numbers)
improves by ~35%, individual multiplies can be significantly faster.
gotest -benchmarks=Mul
big.BenchmarkMul 500 6829290 ns/op (w/ Karatsuba)
big.BenchmarkMul 100 10600760 ns/op
There's no impact on pidigits for -n=10000 or -n=20000
because the operands are are too small.
R=rsc
CC=golang-dev
https://golang.org/cl/1004042
Rather than confuse people by linking to a spec we
don't fully support, let them just use the package docs.
R=r, rsc
CC=golang-dev
https://golang.org/cl/1016041
This matches the old JSON package behavior.
All lowercase names are not as standard as I believed,
and it seems less surprising to need to write
type T struct { Field string "field" }
to get lower case (behavior after this CL) than it does to need
to write
type T struct { Field string "Field" }
to preserve the case (behavior before this CL).
Also test and fix unmarshal into non-nil interface
value or pointer.
Fixes#744.
R=r
CC=golang-dev
https://golang.org/cl/1013041
Adds two more methods, Timeout and Temporary.
Implemented by os.Errno too. The intent is to make
the checks for os.EAGAIN a little less clunky.
It should also let us clean up a bug that Mike Solomon
pointed out: if a network server gets an "out of file descriptors"
error from Accept, the listener should not stop.
It will be able to check this because that error would
have Temporary() == true.
Also clean up some underscore names.
Fixes#442.
R=r
CC=golang-dev, msolo
https://golang.org/cl/957045
There was a bug in makeN that caused lots of unnecessary
allocations. Fixing this exposed a few bugs in other
functions which worked when makeN allocated a new slice, but
failed when it simply resized an existing slice. The result
is a pretty big performance improvement. When running
pidigits, here were the numbers I got on amd64:
Before this change:
pidigits 10000
gcc -O2 pidigits.c -lgmp 2.09u 0.02s 2.11r
gc pidigits 12.68u 0.04s 12.72r
gc_B pidigits 12.71u 0.03s 12.75r
After:
pidigits 10000
gcc -O2 pidigits.c -lgmp 2.09u 0.00s 2.10r
gc pidigits 6.82u 0.00s 6.85r
gc_B pidigits 6.55u 0.01s 6.59r
R=rsc, gri
CC=golang-dev
https://golang.org/cl/953042
Due to page boundary rounding, the header would have
been loaded as part of the text segment already, but this
change placates the "paxctl" tool on so-called hardened
Linux distributions (as if normal distributions weren't already
hard enough to use).
R=r
CC=golang-dev
https://golang.org/cl/954041
SETSID does return an errno - any reason why it has been done this
way in zsyscall_linux_* ? Otherwise it should be the same as darwin.
From SETSID(2) on my Linux box:
ERRORS
On error, -1 is returned, and errno is set.
Fixes#730
R=rsc
CC=golang-dev
https://golang.org/cl/878047
This is the first of probably four separate CLs
for the new implementation of the json package.
The scanner is the core of the new implementation.
The other CLs would be the new decoder,
the new encoder, and support for JSON streams.
R=r
CC=golang-dev
https://golang.org/cl/802051
$ godoc xml | grep Copy\(\)
func (c CharData) Copy() CharData
func (c Comment) Copy() Comment
func (d Directive) Copy() Directive
func (p ProcInst) Copy() ProcInst
func (e StartElement) Copy() StartElement
--------------------------------------------
$ godoc -src xml | grep Copy\(\)
func (c CharData) Copy() CharData
--------------------------------------------
$ godoc -src xml Copy
func (c CharData) Copy() CharData { return CharData(makeCopy(c)) }
--------------------------------------------
The command "godoc -src pkg_name" should output the interface of the named package, but it excludes all duplicate entries. Also the command "godoc -src pkg_name method_name" will output the source code only for one method even if there are more of them with the same name in the same package. This patch set fixes this issue.
R=gri
CC=golang-dev
https://golang.org/cl/883051
fmt.Printf("%b", int8(-1)) prints 64 ones instead of 8.
This happens only for signed integers (int8, in16 and int32). I guess it's because of the way the conversion between integer types works. From go spec: "Conversions between integer types. If the value is a signed quantity, it is sign extended to implicit infinite precision ....". And there are several conversions to int64 and uint64 in the fmt package. This pathch solves only half of the problem. On a 32 bit system, an fmt.Printf("%b", int(-1)) should still print 64 ones.
R=golang-dev, r
CC=golang-dev
https://golang.org/cl/891049
This is required to make cgo export work on Darwin. Note that
this corrects the stack alignment when calling initcgo to that
required by gcc on amd64.
R=rsc
CC=golang-dev
https://golang.org/cl/907041
The new //export comment marks a Go function as callable from
C. The syntax is "//export NAME" where NAME is the name of
the function as seen from C. If such a comment is seen, cgo
will generate two new files: _cgo_export.h and _cgo_export.c.
The _cgo_export.h file provides declarations which C code may
use to call Go functions. The _cgo_export.c file contains
wrappers, and is to be compiled with gcc.
The changes to Make.pkg support using this from a Go Makefile,
though it could probably be more convenient.
R=rsc
CC=golang-dev
https://golang.org/cl/853042
These functions are used to call from a C function back to a
Go function. This only includes 386 support.
R=rsc
CC=golang-dev
https://golang.org/cl/834045
tested on linux/amd64, linux/386, linux/arm, darwin/amd64, darwin/386.
freebsd untested; will finish in a separate CL.
for now all the panics are errorStrings.
richer structures can be added as necessary
once the mechanism is shaked out.
R=r
CC=golang-dev
https://golang.org/cl/906041
Could not take a signal on threads other than the main thread.
If you look at the spinning binary with dtrace, you can see a
fault happening over and over:
$ dtrace -n '
fbt::user_trap:entry /execname=="boot32" && self->count < 10/
{
self->count++;
printf("%s %x %x %x %x", probefunc, arg1, arg2, arg3, arg4);
stack();
tracemem(arg4, 256);
}'
dtrace: description 'fbt::user_trap:entry ' matched 1 probe
CPU ID FUNCTION:NAME
1 17015 user_trap:entry user_trap 0 10 79af0a0 79af0a0
mach_kernel`lo_alltraps+0x12a
0 1 2 3 4 5 6 7 8 9 a b c d e f 0123456789abcdef
0: 0e 00 00 00 37 00 00 00 00 00 00 00 1f 00 00 00 ....7...........
10: 1f 00 00 00 a8 33 00 00 00 00 00 01 00 00 00 00 .....3..........
20: 98 ba dc fe 07 09 00 00 00 00 00 00 98 ba dc fe ................
30: 06 00 00 00 0d 00 00 00 34 00 00 00 9e 1c 00 00 ........4.......
40: 17 00 00 00 00 02 00 00 ac 30 00 00 1f 00 00 00 .........0......
50: 00 00 00 00 00 00 00 00 0d 00 00 00 e0 e6 29 00 ..............).
60: 34 00 00 00 00 00 00 00 9e 1c 00 00 00 00 00 00 4...............
70: 17 00 00 00 00 00 00 00 00 02 00 00 00 00 00 00 ................
80: ac 30 00 00 00 00 00 00 1f 00 00 00 00 00 00 00 .0..............
90: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
a0: 48 00 00 00 10 00 00 00 85 00 00 00 a0 f2 29 00 H.............).
b0: 69 01 00 02 00 00 00 00 e6 93 04 82 ff 7f 00 00 i...............
c0: 2f 00 00 00 00 00 00 00 06 02 00 00 00 00 00 00 /...............
d0: 78 ee 42 01 01 00 00 00 1f 00 00 00 00 00 00 00 x.B.............
e0: 00 ed 9a 07 00 00 00 00 00 00 00 00 00 00 00 00 ................
f0: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
...
The memory dump shows a 32-bit exception frame:
x86_saved_state32
gs = 0x37
fs = 0
es = 0x1f
ds = 0x1f
edi = 0x33a8
esi = 0x01000000
ebp = 0
cr2 = 0xfedcba98
ebx = 0x0907
edx = 0
ecx = 0xfedcba98
eax = 0x06
trapno = 0x0d
err = 0x34
eip = 0x1c9e
cs = 0x17
efl = 0x0200
uesp = 0x30ac
ss = 0x1f
The cr2 of 0xfedcba98 is the address that the new thread read
to cause the fault, but note that the trap is now a GP fault with
error code 0x34, meaning it's moved past the cr2 problem and on
to an invaild segment selector. The 0x34 is suspiciously similar
to the 0x37 in gs, and sure enough, OS X forces gs to have
that value in the signal handler, and if your thread hasn't set
up that segment (known as USER_CTHREAD), you'll fault on the IRET
into the signal handler and never be able to handle a signal.
The kernel bug is that it forces segment 0x37 without making sure
it is a valid segment. Leopard also forced 0x37 but had the courtesy
to set it up first.
Since OS X requires us to set up that segment (using the
thread_fast_set_cthread_self system call), we might as well
use it instead of the more complicated i386_set_ldt call to
set up our per-OS thread storage.
Also add some more zeros to bsdthread_register for new arguments
in Snow Leopard (apparently unnecessary, but being careful).
Fixes#510.
R=r
CC=golang-dev
https://golang.org/cl/824046
data just read from the channel.
this will make it easier to
recognize when to garbage
collect and finalize.
R=rsc
CC=golang-dev
https://golang.org/cl/882043
The cycle is *netFD -> cw chanl *netFD in struct ->
same *netFD in channel read buffer.
Because channels are finalized, the cycle makes them
uncollectable. A better fix is to make channels not
finalized anymore, and that will happen, but this is
an easy, reasonable workaround until then.
Another good fix would be to zero the channel receive
buffer entry after the receive. That too will happen.
R=r
CC=golang-dev
https://golang.org/cl/875043
1. Fix bug in GOMAXPROCS when trying to cut number of procs
Race could happen on any system but was
manifesting only on Xen hosted Linux.
2. Fix recover on ARM, where FP != caller SP.
R=r
CC=golang-dev
https://golang.org/cl/880043
Instead of calling the package scope Lex function,
Parse now takes an argument which is used to
do the lexing.
I reverted to having the generated switch
code inside Parse rather than a separate function because
the function needs 7 arguments or a context structure,
which seems unnecessary.
I used yyrun(), not the original $A so that
it's possible to run the backquoted code through gofmt.
R=rsc, ken2, ken3
CC=golang-dev
https://golang.org/cl/879041
This is a documentation enhancement only, without any code
change.
The rationale for documenting this precisely is that Unix
programmers who "know" that errno's value is undefined after
a successful system call may be surprised otherwise and
search to be sure that a zero errno may be relied upon after
successful calls.
R=r, rsc1, rsc
CC=golang-dev
https://golang.org/cl/812044
log: add test of Caller
New regexp in log test is picky and will require some
maintenance, but it catches off-by-one mistakes too.
Fixes#710.
R=gri
CC=esko.luontola, golang-dev
https://golang.org/cl/887043
