The entry for LEAL/LEAQ in these optabs was listed as having
two data bytes in the y array. In fact they had and expect no data
bytes. However, the general loop expects to be able to look at at
least one data byte, to make sure it is not 0x0f. So give them each
a single data byte set to 0 (not 0x0f).
Since the MOV instructions have the largest optab cases, this
requires growing the size of the data array.
Clang found this bug because the general o->op[z] == 0x0f
test was using z == 22, which was out of bounds.
In practice the next byte in memory was probably not 0x0f
so it wasn't truly broken. But might as well be clean.
Update #5764
R=ken2
CC=golang-dev
https://golang.org/cl/13241050
The portable code in cmd/ld already knows how to process it,
we just have to ignore it during code generation.
R=ken2
CC=golang-dev
https://golang.org/cl/11363043
Design at http://golang.org/s/go12symtab.
This enables some cleanup of the garbage collector metadata
that will be done in future CLs.
This CL does not move the old symtab and pclntab back into
an unmapped section of the file. That's a bit tricky and will be
done separately.
Fixes#4020.
R=golang-dev, dave, cshapiro, iant, r
CC=golang-dev, nigeltao
https://golang.org/cl/11085043
With this change the compiler emits a bitmap for each function
covering its stack frame arguments area. If an argument word
is known to contain a pointer, a bit is set. The garbage
collector reads this information when scanning the stack by
frames and uses it to ignores locations known to not contain a
pointer.
R=golang-dev, bradfitz, daniel.morsing, dvyukov, khr, khr, iant, cshapiro
CC=golang-dev
https://golang.org/cl/9223046
XOR key into data 128 bits at a time instead of 64 bits
and pipeline half of state loads. Rotate loop to allow
single-register indexing for state[i].
On a MacBookPro10,2 (Core i5):
benchmark old ns/op new ns/op delta
BenchmarkRC4_128 412 224 -45.63%
BenchmarkRC4_1K 3179 1613 -49.26%
BenchmarkRC4_8K 25223 12545 -50.26%
benchmark old MB/s new MB/s speedup
BenchmarkRC4_128 310.51 570.42 1.84x
BenchmarkRC4_1K 322.09 634.48 1.97x
BenchmarkRC4_8K 320.97 645.32 2.01x
For comparison, on the same machine, openssl 0.9.8r reports
its rc4 speed as somewhat under 350 MB/s for both 1K and 8K
(it is operating 64 bits at a time).
On an Intel Xeon E5520:
benchmark old ns/op new ns/op delta
BenchmarkRC4_128 418 259 -38.04%
BenchmarkRC4_1K 3200 1884 -41.12%
BenchmarkRC4_8K 25173 14529 -42.28%
benchmark old MB/s new MB/s speedup
BenchmarkRC4_128 306.04 492.48 1.61x
BenchmarkRC4_1K 319.93 543.26 1.70x
BenchmarkRC4_8K 321.61 557.20 1.73x
For comparison, on the same machine, openssl 1.0.1
reports its rc4 speed as 587 MB/s for 1K and 601 MB/s for 8K.
R=agl
CC=golang-dev
https://golang.org/cl/7865046
The type information is (and for years has been) included
as an extra field in the address chunk of an instruction.
Unfortunately, suppose there is a string at a+24(FP) and
we have an instruction reading its length. It will say:
MOVQ x+32(FP), AX
and the type of *that* argument is int (not slice), because
it is the length being read. This confuses the picture seen
by debuggers and now, worse, by the garbage collector.
Instead of attaching the type information to all uses,
emit an explicit list of TYPE instructions with the information.
The TYPE instructions are no-ops whose only role is to
provide an address to attach type information to.
For example, this function:
func f(x, y, z int) (a, b string) {
return
}
now compiles into:
--- prog list "f" ---
0000 (/Users/rsc/x.go:3) TEXT f+0(SB),$0-56
0001 (/Users/rsc/x.go:3) LOCALS ,
0002 (/Users/rsc/x.go:3) TYPE x+0(FP){int},$8
0003 (/Users/rsc/x.go:3) TYPE y+8(FP){int},$8
0004 (/Users/rsc/x.go:3) TYPE z+16(FP){int},$8
0005 (/Users/rsc/x.go:3) TYPE a+24(FP){string},$16
0006 (/Users/rsc/x.go:3) TYPE b+40(FP){string},$16
0007 (/Users/rsc/x.go:3) MOVQ $0,b+40(FP)
0008 (/Users/rsc/x.go:3) MOVQ $0,b+48(FP)
0009 (/Users/rsc/x.go:3) MOVQ $0,a+24(FP)
0010 (/Users/rsc/x.go:3) MOVQ $0,a+32(FP)
0011 (/Users/rsc/x.go:4) RET ,
The { } show the formerly hidden type information.
The { } syntax is used when printing from within the gc compiler.
It is not accepted by the assemblers.
The same type information is now included on global variables:
0055 (/Users/rsc/x.go:15) GLOBL slice+0(SB){[]string},$24(AL*0)
This more accurate type information fixes a bug in the
garbage collector's precise heap collection.
The linker only cares about globals right now, but having the
local information should make things a little nicer for Carl
in the future.
Fixes#4907.
R=ken2
CC=golang-dev
https://golang.org/cl/7395056
The new src argument is ignored during linking
(that is, CALL r1, r2 is identical to CALL r2 for linking),
but it serves as a hint to the 5g/6g/8g optimizer
that the src register is live on entry to the called
function and must be preserved.
It is possible to avoid exposing this fact to the rest of
the toolchain, keeping it entirely within 5g/6g/8g,
but I think it will help to be able to look in object files
and assembly listings and linker -a / -W output to
see CALL instructions are "Go func value" calls and
which are "C function pointer" calls.
R=ken2
CC=golang-dev
https://golang.org/cl/7364045
Previously, the func structure contained an inaccurate value for
the args member and a 0 value for the locals member.
This change populates the func structure with args and locals
values computed by the compiler. The number of args was
already available in the ATEXT instruction. The number of
locals is now passed through in the new ALOCALS instruction.
This change also switches the unit of args and locals to be
bytes, just like the frame member, instead of 32-bit words.
R=golang-dev, bradfitz, cshapiro, dave, rsc
CC=golang-dev
https://golang.org/cl/7399045
This is an experiment in static analysis of Go programs
to understand which struct fields a program might use.
It is not part of the Go language specification, it must
be enabled explicitly when building the toolchain,
and it may be removed at any time.
After building the toolchain with GOEXPERIMENT=fieldtrack,
a specific field can be marked for tracking by including
`go:"track"` in the field tag:
package pkg
type T struct {
F int `go:"track"`
G int // untracked
}
To simplify usage, only named struct types can have
tracked fields, and only exported fields can be tracked.
The implementation works by making each function begin
with a sequence of no-op USEFIELD instructions declaring
which tracked fields are accessed by a specific function.
After the linker's dead code elimination removes unused
functions, the fields referred to by the remaining
USEFIELD instructions are the ones reported as used by
the binary.
The -k option to the linker specifies the fully qualified
symbol name (such as my/pkg.list) of a string variable that
should be initialized with the field tracking information
for the program. The field tracking string is a sequence
of lines, each terminated by a \n and describing a single
tracked field referred to by the program. Each line is made
up of one or more tab-separated fields. The first field is
the name of the tracked field, fully qualified, as in
"my/pkg.T.F". Subsequent fields give a shortest path of
reverse references from that field to a global variable or
function, corresponding to one way in which the program
might reach that field.
A common source of false positives in field tracking is
types with large method sets, because a reference to the
type descriptor carries with it references to all methods.
To address this problem, the CL also introduces a comment
annotation
//go:nointerface
that marks an upcoming method declaration as unavailable
for use in satisfying interfaces, both statically and
dynamically. Such a method is also invisible to package
reflect.
Again, all of this is disabled by default. It only turns on
if you have GOEXPERIMENT=fieldtrack set during make.bash.
R=iant, ken
CC=golang-dev
https://golang.org/cl/6749064
This CL adds support for the these 7 new instructions to 6a/6l in
preparation of the upcoming CL for AES-NI accelerated crypto/aes:
AESENC, AESENCLAST, AESDEC, AESDECLAST, AESIMC, AESKEYGENASSIST,
and PSHUFD.
R=golang-dev, rsc
CC=golang-dev
https://golang.org/cl/5970055
Broke tests on 386.
««« original CL description
6l/8l: emit correct opcodes to F(SUB|DIV)R?D.
When the destination was not F0, 6l and 8l swapped FSUBD/FSUBRD and
FDIVD/FDIVRD.
R=golang-dev, dave, rsc
CC=golang-dev
https://golang.org/cl/6498092
»»»
R=golang-dev
CC=golang-dev
https://golang.org/cl/6492100
On 6l and 8l, this is a real instruction, guaranteed to
cause an 'undefined instruction' exception.
On 5l, we simulate it as BL to address 0.
The plan is to use it as a signal to the linker that this
point in the instruction stream cannot be reached
(hence the changes to nofollow). This will help the
compiler explain that panicindex and friends do not
return without having to put a list of these functions
in the linker.
R=ken2
CC=golang-dev
https://golang.org/cl/6255064
The old code generated for a bounds check was
CMP
JLT ok
CALL panicindex
ok:
...
The new code is (once the linker finishes with it):
CMP
JGE panic
...
panic:
CALL panicindex
which moves the calls out of line, putting more useful
code in each cache line. This matters especially in tight
loops, such as in Fannkuch. The benefit is more modest
elsewhere, but real.
From test/bench/go1, amd64:
benchmark old ns/op new ns/op delta
BenchmarkBinaryTree17 6096092000 6088808000 -0.12%
BenchmarkFannkuch11 6151404000 4020463000 -34.64%
BenchmarkGobDecode 28990050 28894630 -0.33%
BenchmarkGobEncode 12406310 12136730 -2.17%
BenchmarkGzip 179923 179903 -0.01%
BenchmarkGunzip 11219 11130 -0.79%
BenchmarkJSONEncode 86429350 86515900 +0.10%
BenchmarkJSONDecode 334593800 315728400 -5.64%
BenchmarkRevcomp25M 1219763000 1180767000 -3.20%
BenchmarkTemplate 492947600 483646800 -1.89%
And 386:
benchmark old ns/op new ns/op delta
BenchmarkBinaryTree17 6354902000 6243000000 -1.76%
BenchmarkFannkuch11 8043769000 7326965000 -8.91%
BenchmarkGobDecode 19010800 18941230 -0.37%
BenchmarkGobEncode 14077500 13792460 -2.02%
BenchmarkGzip 194087 193619 -0.24%
BenchmarkGunzip 12495 12457 -0.30%
BenchmarkJSONEncode 125636400 125451400 -0.15%
BenchmarkJSONDecode 696648600 685032800 -1.67%
BenchmarkRevcomp25M 2058088000 2052545000 -0.27%
BenchmarkTemplate 602140000 589876800 -2.04%
To implement this, two new instruction forms:
JLT target // same as always
JLT $0, target // branch expected not taken
JLT $1, target // branch expected taken
The linker could also emit the prediction prefixes, but it
does not: expected taken branches are reversed so that the
expected case is not taken (as in example above), and
the default expectaton for such a jump is not taken
already.
R=golang-dev, gri, r, dave
CC=golang-dev
https://golang.org/cl/6248049
Without an explicit signal for a truncation, copy propagation
will sometimes propagate a 32-bit truncation and end up
overwriting uses of the original 64-bit value.
The case that arose in practice is in C but I believe
that the same could plausibly happen in Go.
The main reason we didn't run into the same in Go
is that I (perhaps incorrectly?) drop MOVL AX, AX
during gins, so the truncation was never generated, so
it didn't confuse the optimizer.
Fixes#1315.
Fixes#3488.
R=ken2
CC=golang-dev
https://golang.org/cl/6002043
Although Intel considers the three-argument form of IMUL to be a
variant of IMUL, I couldn't make 6l able to differentiate it without
huge changes, so I called it IMUL3.
R=rsc
CC=golang-dev
https://golang.org/cl/5686055
(Here, quoted strings are the official AMD names.)
The amd64 "movsxd" instruction, when invoked
with a 64-bit REX prefix, moves and sign extends
a 32-bit value from register or memory into a
64-bit register. 6.out.h spells this MOVLQSX.
6.out.h also includes MOVLQZX, the zero extending
version, which it implements as "movsxd" without
the REX prefix. Without the REX prefix it's only sign
extending 32 bits to 32 bits (i.e., not doing anything
to the bits) and then storing in a 32-bit register.
Any write to a 32-bit register zeros the top half of the
corresponding 64-bit register, giving the advertised effect.
This particular implementation of the functionality is
non-standard, because an ordinary 32-bit "mov" would
do the same thing.
Because it is non-standard, it is often mishandled or
not handled by binary translation tools like valgrind.
Switching to the standard "mov" makes the binaries
work better with those tools.
It's probably useful in 6c and 6g to have an explicit
instruction, though, so that the intent of the size
change is clear. Thus we leave the concept of MOVLQZX
and just implement it by the standard "mov" instead of
the non-standard 32-bit "movsxd".
Fixes#896.
R=ken2
CC=golang-dev
https://golang.org/cl/1733046
Russ Cox
Adam Langley
Carl Shapiro
Keith Randall
Shenghou Ma
Michał Derkacz
Jaroslavas Počepko
Dmitriy Vyukov
Evan Shaw
Ken Thompson
Rob Pike