Retrying CL 222782, with a fix that will hopefully stop the random crashing.
The issue with the previous CL is that it does pointer arithmetic
in a way that may briefly generate an out-of-bounds pointer. If an
interrupt happens to occur in that state, the referenced object may
be collected incorrectly.
Suppose there was code that did s[x+c]. The previous CL had a rule
to the effect of ptr + (x + c) -> c + (ptr + x). But ptr+x is not
guaranteed to point to the same object as ptr. In contrast,
ptr+(x+c) is guaranteed to point to the same object as ptr, because
we would have already checked that x+c is in bounds.
For example, strconv.trim used to have this code:
MOVZX -0x1(BX)(DX*1), BP
CMPL $0x30, AL
After CL 222782, it had this code:
LEAL 0(BX)(DX*1), BP
CMPB $0x30, -0x1(BP)
An interrupt between those last two instructions could see BP pointing
outside the backing store of the slice involved.
It's really hard to actually demonstrate a bug. First, you need to
have an interrupt occur at exactly the right time. Then, there must
be no other pointers to the object in question. Since the interrupted
frame will be scanned conservatively, there can't even be a dead
pointer in another register or on the stack. (In the example above,
a bug can't happen because BX still holds the original pointer.)
Then, the object in question needs to be collected (or at least
scanned?) before the interrupted code continues.
This CL needs to handle load combining somewhat differently than CL 222782
because of the new restriction on arithmetic. That's the only real
difference (other than removing the bad rules) from that old CL.
This bug is also present in the amd64 rewrite rules, and we haven't
seen any crashing as a result. I will fix up that code similarly to
this one in a separate CL.
Update #37881
Change-Id: I5f0d584d9bef4696bfe89a61ef0a27c8d507329f
Reviewed-on: https://go-review.googlesource.com/c/go/+/225798
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
The load and test instructions compare the given value
against zero and will produce a condition code indicating
one of the following scenarios:
0: Result is zero
1: Result is less than zero
2: Result is greater than zero
3: Result is not a number (NaN)
The instruction can be used to simplify floating point comparisons
against zero, which can enable further optimizations.
This CL also reduces the size of .text section of math.test binary by around
0.7 KB (in hexadecimal, from 1358f0 to 135620).
Change-Id: I33cb714f0c6feebac7a1c46dfcc735e7daceff9c
Reviewed-on: https://go-review.googlesource.com/c/go/+/209159
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This reverts commit CL 222782.
Reason for revert: Reverting to see if 386 errors go away
Update #37881
Change-Id: I74f287404c52414db1b6ff1649effa4ed9e5cc0c
Reviewed-on: https://go-review.googlesource.com/c/go/+/225218
Reviewed-by: Bryan C. Mills <bcmills@google.com>
Rolling back portions of CL 222782 to see if that helps
issue #37881 any.
Update #37881
Change-Id: I9cc3ff8c469fa5e4b22daec715d04148033f46f7
Reviewed-on: https://go-review.googlesource.com/c/go/+/224837
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Bryan C. Mills <bcmills@google.com>
This commit extends the -spectre flag to cmd/asm and adds
a new Spectre mitigation mode "ret", which enables the use
of retpolines.
Retpolines prevent speculation about the target of an indirect
jump or call and are described in more detail here:
https://support.google.com/faqs/answer/7625886
Change-Id: I4f2cb982fa94e44d91e49bd98974fd125619c93a
Reviewed-on: https://go-review.googlesource.com/c/go/+/222661
Reviewed-by: Keith Randall <khr@golang.org>
This commit adds a new cmd/compile flag -spectre,
which accepts a comma-separated list of possible
Spectre mitigations to apply, or the empty string (none),
or "all". The only known mitigation right now is "index",
which uses conditional moves to ensure that x86-64 CPUs
do not speculate past index bounds checks.
Speculating past index bounds checks may be problematic
on systems running privileged servers that accept requests
from untrusted users who can execute their own programs
on the same machine. (And some more constraints that
make it even more unlikely in practice.)
The cases this protects against are analogous to the ones
Microsoft explains in the "Array out of bounds load/store feeding ..."
sections here:
https://docs.microsoft.com/en-us/cpp/security/developer-guidance-speculative-execution?view=vs-2019#array-out-of-bounds-load-feeding-an-indirect-branch
Change-Id: Ib7532d7e12466b17e04c4e2075c2a456dc98f610
Reviewed-on: https://go-review.googlesource.com/c/go/+/222660
Reviewed-by: Keith Randall <khr@golang.org>
Use a separate compiler pass to introduce complicated x86 addressing
modes. Loads in the normal architecture rules (for x86 and all other
platforms) can have constant offsets (AuxInt values) and symbols (Aux
values), but no more.
The complex addressing modes (x+y, x+2*y, etc.) are introduced in a
separate pass that combines loads with LEAQx ops.
Organizing rewrites this way simplifies the number of rewrites
required, as there are lots of different rule orderings that have to
be specified to ensure these complex addressing modes are always found
if they are possible.
Update #36468
Change-Id: I5b4bf7b03a1e731d6dfeb9ef19b376175f3b4b44
Reviewed-on: https://go-review.googlesource.com/c/go/+/217097
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
There are still two places in src/runtime/string.go that use
staticbytes, so we cannot delete it just yet.
There is a new codegen test to verify that the index calculation
is constant-folded, at least on amd64. ppc64, mips[64] and s390x
cannot currently do that.
There is also a new runtime benchmark to ensure that this does not
slow down performance (tested against parent commit):
name old time/op new time/op delta
ConvT2EByteSized/bool-4 1.07ns ± 1% 1.07ns ± 1% ~ (p=0.060 n=14+15)
ConvT2EByteSized/uint8-4 1.06ns ± 1% 1.07ns ± 1% ~ (p=0.095 n=14+15)
Updates #37612
Change-Id: I5ec30738edaa48cda78dfab4a78e24a32fa7fd6a
Reviewed-on: https://go-review.googlesource.com/c/go/+/221957
Run-TryBot: Ian Lance Taylor <iant@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Trying this CL again, with a fixed test that allows platforms
to disagree on the exact behavior of converting NaNs.
We store 32-bit floating point constants in a 64-bit field, by
converting that 32-bit float to 64-bit float to store it, and convert
it back to use it.
That works for *almost* all floating-point constants. The exception is
signaling NaNs. The round trip described above means we can't represent
a 32-bit signaling NaN, because conversions strip the signaling bit.
To fix this issue, just forbid NaNs as floating-point constants in SSA
form. This shouldn't affect any real-world code, as people seldom
constant-propagate NaNs (except in test code).
Additionally, NaNs are somewhat underspecified (which of the many NaNs
do you get when dividing 0/0?), so when cross-compiling there's a
danger of using the compiler machine's NaN regime for some math, and
the target machine's NaN regime for other math. Better to use the
target machine's NaN regime always.
Update #36400
Change-Id: Idf203b688a15abceabbd66ba290d4e9f63619ecb
Reviewed-on: https://go-review.googlesource.com/c/go/+/221790
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
Ensure that any comparison between two values has the same argument
order. This helps ensure that they can be eliminated during the
lowered CSE pass which will be particularly important if we eliminate
the Greater and Geq ops (see #37316).
Example:
CMP R0, R1
BLT L1
CMP R1, R0 // different order, cannot eliminate
BEQ L2
CMP R0, R1
BLT L1
CMP R0, R1 // same order, can eliminate
BEQ L2
This does have some drawbacks. Notably comparisons might 'flip'
direction in the assembly output after even small changes to the
code or compiler. It should help make optimizations more reliable
however.
compilecmp master -> HEAD
master (218f4572f5): text/template: make reflect.Value indirections more robust
HEAD (f1661fef3e): cmd/compile: canonicalize comparison argument order
platform: linux/amd64
file before after Δ %
api 6063927 6068023 +4096 +0.068%
asm 5191757 5183565 -8192 -0.158%
cgo 4893518 4901710 +8192 +0.167%
cover 5330345 5326249 -4096 -0.077%
fix 3417778 3421874 +4096 +0.120%
pprof 14889456 14885360 -4096 -0.028%
test2json 2848138 2844042 -4096 -0.144%
trace 11746239 11733951 -12288 -0.105%
total 132739173 132722789 -16384 -0.012%
Change-Id: I11736b3fe2a4553f6fc65018f475e88217fa22f9
Reviewed-on: https://go-review.googlesource.com/c/go/+/220425
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
This reverts CL 213477.
Reason for revert: tests are failing on linux-mips*-rtrk builders.
Change-Id: I8168f7450890233f1bd7e53930b73693c26d4dc0
Reviewed-on: https://go-review.googlesource.com/c/go/+/220897
Run-TryBot: Bryan C. Mills <bcmills@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
We store 32-bit floating point constants in a 64-bit field, by
converting that 32-bit float to 64-bit float to store it, and convert
it back to use it.
That works for *almost* all floating-point constants. The exception is
signaling NaNs. The round trip described above means we can't represent
a 32-bit signaling NaN, because conversions strip the signaling bit.
To fix this issue, just forbid NaNs as floating-point constants in SSA
form. This shouldn't affect any real-world code, as people seldom
constant-propagate NaNs (except in test code).
Additionally, NaNs are somewhat underspecified (which of the many NaNs
do you get when dividing 0/0?), so when cross-compiling there's a
danger of using the compiler machine's NaN regime for some math, and
the target machine's NaN regime for other math. Better to use the
target machine's NaN regime always.
This has been a bug since 1.10, and there's an easy workaround
(declare a global varaible containing the signaling NaN pattern, and
use that as the argument to math.Float32frombits) so we'll fix it in
1.15.
Fixes#36400
Update #36399
Change-Id: Icf155e743281560eda2eed953d19a829552ccfda
Reviewed-on: https://go-review.googlesource.com/c/go/+/213477
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Josh Bleecher Snyder <josharian@gmail.com>
The address calculations in the example end up doing x << 4 + y + 0.
Before this CL we use a SHLQ+LEAQ. Since the constant offset is 0,
we can use SHLQ+ADDQ instead.
Change-Id: Ia048c4fdbb3a42121c7e1ab707961062e8247fca
Reviewed-on: https://go-review.googlesource.com/c/go/+/209959
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This API was added for #25819, where it was discussed as math.FMA.
The commit adding it used math.Fma, presumably for consistency
with the rest of the unusual names in package math
(Sincos, Acosh, Erfcinv, Float32bits, etc).
I believe that using an idiomatic Go name is more important here
than consistency with these other names, most of which are historical
baggage from C's standard library.
Early additions like Float32frombits happened before "uppercase for export"
(so they were originally like "float32frombits") and they were not properly
reconsidered when we uppercased the symbols to export them.
That's a mistake we live with.
The names of functions we have added since then, and even a few
that were legacy, are more properly Go-cased, such as IsNaN, IsInf,
and RoundToEven, rather than Isnan, Isinf, and Roundtoeven.
And also constants like MaxFloat32.
For new API, we should keep using proper Go-cased symbols
instead of minimally-upper-cased-C symbols.
So math.FMA, not math.Fma.
This API has not yet been released, so this change does not break
the compatibility promise.
This CL also modifies cmd/compile, since the compiler knows
the name of the function. I could have stopped at changing the
string constants, but it seemed to make more sense to use a
consistent casing everywhere.
Change-Id: I0f6f3407f41e99bfa8239467345c33945088896e
Reviewed-on: https://go-review.googlesource.com/c/go/+/205317
Run-TryBot: Russ Cox <rsc@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
This change introduces an arm intrinsic that generates the FMULAD
instruction for the fused-multiply-add operation on systems that
support it. System support is detected via cpu.ARM.HasVFPv4. A rewrite
rule translates the generic intrinsic to FMULAD.
Updates #25819.
Change-Id: I8459e5dd1cdbdca35f88a78dbeb7d387f1e20efa
Reviewed-on: https://go-review.googlesource.com/c/go/+/142117
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
To permit ssa-level optimization, this change introduces an amd64 intrinsic
that generates the VFMADD231SD instruction for the fused-multiply-add
operation on systems that support it. System support is detected via
cpu.X86.HasFMA. A rewrite rule can then translate the generic ssa intrinsic
("Fma") to VFMADD231SD.
The benchmark compares the software implementation (old) with the intrinsic
(new).
name old time/op new time/op delta
Fma-4 27.2ns ± 1% 1.0ns ± 9% -96.48% (p=0.008 n=5+5)
Updates #25819.
Change-Id: I966655e5f96817a5d06dff5942418a3915b09584
Reviewed-on: https://go-review.googlesource.com/c/go/+/137156
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
In order to make math.FMA a compiler intrinsic for ISAs like ARM64,
PPC64[le], and S390X, a generic 3-argument opcode "Fma" is provided and
rewritten as
ARM64: (Fma x y z) -> (FMADDD z x y)
PPC64: (Fma x y z) -> (FMADD x y z)
S390X: (Fma x y z) -> (FMADD z x y)
Updates #25819.
Change-Id: Ie5bc628311e6feeb28ddf9adaa6e702c8c291efa
Reviewed-on: https://go-review.googlesource.com/c/go/+/131959
Run-TryBot: Akhil Indurti <aindurti@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
When a subsequent load/store of a ptr makes the nil check of that pointer
unnecessary, if their lines differ, change the line of the load/store
to that of the nilcheck, and attempt to rehome the load/store position
instead.
This fix makes profiling less accurate in order to make panics more
informative.
Fixes#33724
Change-Id: Ib9afaac12fe0d0320aea1bf493617facc34034b3
Reviewed-on: https://go-review.googlesource.com/c/go/+/200197
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
For commuting ops, check whether the second argument is dead before
checking if the first argument is rematerializeable. Reusing the register
holding a dead value is always best.
Fixes#33580
Change-Id: I7372cfc03d514e6774d2d9cc727a3e6bf6ce2657
Reviewed-on: https://go-review.googlesource.com/c/go/+/199559
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: David Chase <drchase@google.com>
Add a bunch of extra tests and benchmarks for defer, in preparation for new
low-cost (open-coded) implementation of defers (see #34481),
- New file defer_test.go that tests a bunch more unusual defer scenarios,
including things that might have problems for open-coded defers.
- Additions to callers_test.go actually verifying what the stack trace looks like
for various panic or panic-recover scenarios.
- Additions to crash_test.go testing several more crash scenarios involving
recursive panics.
- New benchmark in runtime_test.go measuring speed of panic-recover
- New CGo benchmark in cgo_test.go calling from Go to C back to Go that
shows defer overhead
Updates #34481
Change-Id: I423523f3e05fc0229d4277dd00073289a5526188
Reviewed-on: https://go-review.googlesource.com/c/go/+/197017
Run-TryBot: Dan Scales <danscales@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
On modern 64bit CPUs a SHR, SHL or AND instruction take 1 cycle to execute.
A pair of shifts that operate on the same register will take 2 cycles
and needs to wait for the input register value to be available.
Large constants used to mask the high bits of a register with an AND
instruction can not be encoded as an immediate in the AND instruction
on amd64 and therefore need to be loaded into a register with a MOV
instruction.
However that MOV instruction is not dependent on the output register and
on many CPUs does not compete with the AND or shift instructions for
execution ports.
Using a pair of shifts to mask high bits instead of an AND to mask high
bits of a register has a shorter encoding and uses one less general
purpose register but is slower due to taking one clock cycle longer
if there is no register pressure that would make the AND variant need to
generate a spill.
For example the instructions emitted for (x & 1 << 63) before this CL are:
48c1ea3f SHRQ $0x3f, DX
48c1e23f SHLQ $0x3f, DX
after this CL the instructions are the same as GCC and LLVM use:
48b80000000000000080 MOVQ $0x8000000000000000, AX
4821d0 ANDQ DX, AX
Some platforms such as arm64 already have SSA optimization rules to fuse
two shift instructions back into an AND.
Removing the general rule to rewrite AND to SHR+SHL speeds up this benchmark:
var GlobalU uint
func BenchmarkAndHighBits(b *testing.B) {
x := uint(0)
for i := 0; i < b.N; i++ {
x &= 1 << 63
}
GlobalU = x
}
amd64/darwin on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz:
name old time/op new time/op delta
AndHighBits-4 0.61ns ± 6% 0.42ns ± 6% -31.42% (p=0.000 n=25+25):
'go run run.go -all_codegen -v codegen' passes with following adjustments:
ARM64: The BFXIL pattern ((x << lc) >> rc | y & ac) needed adjustment
since ORshiftRL generation fusing '>> rc' and '|' interferes
with matching ((x << lc) >> rc) to generate UBFX. Previously
ORshiftLL was created first using the shifts generated for (y & ac).
S390X: Add rules for abs and copysign to match use of AND instead of SHIFTs.
Updates #33826
Updates #32781
Change-Id: I5a59f6239660d53c029cd22dfb44ddf39f93a56c
Reviewed-on: https://go-review.googlesource.com/c/go/+/196810
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
On modern 64bit CPUs a SHR, SHL or AND instruction take 1 cycle to execute.
A pair of shifts that operate on the same register will take 2 cycles
and needs to wait for the input register value to be available.
Large constants used to mask the high bits of a register with an AND
instruction can not be encoded as an immediate in the AND instruction
on amd64 and therefore need to be loaded into a register with a MOV
instruction.
However that MOV instruction is not dependent on the output register and
on many CPUs does not compete with the AND or shift instructions for
execution ports.
Using a pair of shifts to mask high bits instead of an AND to mask high
bits of a register has a shorter encoding and uses one less general
purpose register but is slower due to taking one clock cycle longer
if there is no register pressure that would make the AND variant need to
generate a spill.
For example the instructions emitted for (x & 1 << 63) before this CL are:
48c1ea3f SHRQ $0x3f, DX
48c1e23f SHLQ $0x3f, DX
after this CL the instructions are the same as GCC and LLVM use:
48b80000000000000080 MOVQ $0x8000000000000000, AX
4821d0 ANDQ DX, AX
Some platforms such as arm64 already have SSA optimization rules to fuse
two shift instructions back into an AND.
Removing the general rule to rewrite AND to SHR+SHL speeds up this benchmark:
var GlobalU uint
func BenchmarkAndHighBits(b *testing.B) {
x := uint(0)
for i := 0; i < b.N; i++ {
x &= 1 << 63
}
GlobalU = x
}
amd64/darwin on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz:
name old time/op new time/op delta
AndHighBits-4 0.61ns ± 6% 0.42ns ± 6% -31.42% (p=0.000 n=25+25):
'go run run.go -all_codegen -v codegen' passes with following adjustments:
ARM64: The BFXIL pattern ((x << lc) >> rc | y & ac) needed adjustment
since ORshiftRL generation fusing '>> rc' and '|' interferes
with matching ((x << lc) >> rc) to generate UBFX. Previously
ORshiftLL was created first using the shifts generated for (y & ac).
S390X: Add rules for abs and copysign to match use of AND instead of SHIFTs.
Updates #33826
Updates #32781
Change-Id: I43227da76b625de03fbc51117162b23b9c678cdb
Reviewed-on: https://go-review.googlesource.com/c/go/+/194297
Run-TryBot: Martin Möhrmann <martisch@uos.de>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
i32.eqz instructions don't appear unless needed in if conditions anymore
after CL 195204. I forgot to run the codegen tests while submitting the CL.
Thanks to @martisch for catching it.
Fixes#34442
Change-Id: I177b064b389be48e39d564849714d7a8839be13e
Reviewed-on: https://go-review.googlesource.com/c/go/+/196580
Run-TryBot: Agniva De Sarker <agniva.quicksilver@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
When compiling expression switches, we try to optimize runs of
constants into binary searches. The ordering used isn't visible to the
application, so it's unimportant as long as we're consistent between
sorting and searching.
For strings, it's much cheaper to compare string lengths than strings
themselves, so instead of ordering strings by "si <= sj", we currently
order them by "len(si) < len(sj) || len(si) == len(sj) && si <= sj"
(i.e., the lexicographical ordering on the 2-tuple (len(s), s)).
However, it's also somewhat cheaper to compare strings for equality
(i.e., ==) than for ordering (i.e., <=). And if there were two or
three string constants of the same length in a switch statement, we
might unnecessarily emit ordering comparisons.
For example, given:
switch s {
case "", "1", "2", "3": // ordered by length then content
goto L
}
we currently compile this as:
if len(s) < 1 || len(s) == 1 && s <= "1" {
if s == "" { goto L }
else if s == "1" { goto L }
} else {
if s == "2" { goto L }
else if s == "3" { goto L }
}
This CL switches to using a 2-level binary search---first on len(s),
then on s itself---so that string ordering comparisons are only needed
when there are 4 or more strings of the same length. (4 being the
cut-off for when using binary search is actually worthwhile.)
So the above switch instead now compiles to:
if len(s) == 0 {
if s == "" { goto L }
} else if len(s) == 1 {
if s == "1" { goto L }
else if s == "2" { goto L }
else if s == "3" { goto L }
}
which is better optimized by walk and SSA. (Notably, because there are
only two distinct lengths and no more than three strings of any
particular length, this example ends up falling back to simply using
linear search.)
Test case by khr@ from CL 195138.
Fixes#33934.
Change-Id: I8eeebcaf7e26343223be5f443d6a97a0daf84f07
Reviewed-on: https://go-review.googlesource.com/c/go/+/195340
Run-TryBot: Matthew Dempsky <mdempsky@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
golang.org/cl/109517 optimized the compiler to avoid the allocation for make in
append(x, make([]T, y)...). This was only implemented for the case that y has type int.
This change extends the optimization to trigger for all integer types where the value
is known at compile time to fit into an int.
name old time/op new time/op delta
ExtendInt-12 106ns ± 4% 106ns ± 0% ~ (p=0.351 n=10+6)
ExtendUint64-12 1.03µs ± 5% 0.10µs ± 4% -90.01% (p=0.000 n=9+10)
name old alloc/op new alloc/op delta
ExtendInt-12 0.00B 0.00B ~ (all equal)
ExtendUint64-12 13.6kB ± 0% 0.0kB -100.00% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
ExtendInt-12 0.00 0.00 ~ (all equal)
ExtendUint64-12 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10)
Updates #29785
Change-Id: Ief7760097c285abd591712da98c5b02bc3961fcd
Reviewed-on: https://go-review.googlesource.com/c/go/+/182559
Run-TryBot: Cuong Manh Le <cuong.manhle.vn@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
This reverts CL 192101.
Reason for revert: The same paragraph was added 2 weeks ago
(look a few lines above)
Change-Id: I05efb2631d7b4966f66493f178f2a649c715a3cc
Reviewed-on: https://go-review.googlesource.com/c/go/+/195637
Reviewed-by: Cherry Zhang <cherryyz@google.com>
It is useful to know about the -all_codegen option for running
codegen tests for all platforms. I was puzzling that some codegen
test was not failing on my local machine or on trybot, until I
found this option.
Change-Id: I062cf4d73f6a6c9ebc2258195779d2dab21bc36d
Reviewed-on: https://go-review.googlesource.com/c/go/+/192101
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
Run-TryBot: Daniel Martí <mvdan@mvdan.cc>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This change adds an intrinsic for Mul64 on s390x. To achieve that,
a new assembly instruction, MLGR, is introduced in s390x/asmz.go. This assembly
instruction directly uses an existing instruction on Z and supports multiplication
of two 64 bit unsigned integer and stores the result in two separate registers.
In this case, we require the multiplcand to be stored in register R3 and
the output result (the high and low 64 bit of the product) to be stored in
R2 and R3 respectively.
A test case is also added.
Benchmark:
name old time/op new time/op delta
Mul-18 11.1ns ± 0% 1.4ns ± 0% -87.39% (p=0.002 n=8+10)
Mul32-18 2.07ns ± 0% 2.07ns ± 0% ~ (all equal)
Mul64-18 11.1ns ± 1% 1.4ns ± 0% -87.42% (p=0.000 n=10+10)
Change-Id: Ieca6ad1f61fff9a48a31d50bbd3f3c6d9e6675c1
Reviewed-on: https://go-review.googlesource.com/c/go/+/194572
Reviewed-by: Michael Munday <mike.munday@ibm.com>
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
This CL gets rid of the MOVDreg and MOVDnop SSA operations on
s390x. They were originally inserted to help avoid situations
where a sign/zero extension was elided but a spill invalidated
the optimization. It's not really clear we need to do this though
(amd64 doesn't have these ops for example) so long as we are
careful when removing sign/zero extensions. Also, the MOVDreg
technique doesn't work if the register is spilled before the
MOVDreg op (I haven't seen that in practice).
Removing these ops reduces the complexity of the rules and also
allows us to unblock optimizations. For example, the compiler can
now merge the loads in binary.{Big,Little}Endian.PutUint16 which
it wasn't able to do before. This CL reduces the size of the .text
section in the go tool by about 4.7KB (0.09%).
Change-Id: Icaddae7f2e4f9b2debb6fabae845adb3f73b41db
Reviewed-on: https://go-review.googlesource.com/c/go/+/173897
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
On modern 64bit CPUs a SHR, SHL or AND instruction take 1 cycle to execute.
A pair of shifts that operate on the same register will take 2 cycles
and needs to wait for the input register value to be available.
Large constants used to mask the high bits of a register with an AND
instruction can not be encoded as an immediate in the AND instruction
on amd64 and therefore need to be loaded into a register with a MOV
instruction.
However that MOV instruction is not dependent on the output register and
on many CPUs does not compete with the AND or shift instructions for
execution ports.
Using a pair of shifts to mask high bits instead of an AND to mask high
bits of a register has a shorter encoding and uses one less general
purpose register but is slower due to taking one clock cycle longer
if there is no register pressure that would make the AND variant need to
generate a spill.
For example the instructions emitted for (x & 1 << 63) before this CL are:
48c1ea3f SHRQ $0x3f, DX
48c1e23f SHLQ $0x3f, DX
after this CL the instructions are the same as GCC and LLVM use:
48b80000000000000080 MOVQ $0x8000000000000000, AX
4821d0 ANDQ DX, AX
Some platforms such as arm64 already have SSA optimization rules to fuse
two shift instructions back into an AND.
Removing the general rule to rewrite AND to SHR+SHL speeds up this benchmark:
var GlobalU uint
func BenchmarkAndHighBits(b *testing.B) {
x := uint(0)
for i := 0; i < b.N; i++ {
x &= 1 << 63
}
GlobalU = x
}
amd64/darwin on Intel(R) Core(TM) i7-3520M CPU @ 2.90GHz:
name old time/op new time/op delta
AndHighBits-4 0.61ns ± 6% 0.42ns ± 6% -31.42% (p=0.000 n=25+25):
Updates #33826
Updates #32781
Change-Id: I862d3587446410c447b9a7265196b57f85358633
Reviewed-on: https://go-review.googlesource.com/c/go/+/191780
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
For performance reasons (avoiding costly cross-compilations) CL 177577
changed the codegen test harness to only run the tests for the
machine's GOARCH by default.
This change updates the codegen README accordingly, explaining what
all.bash does run by default and how to perform the tests for all
architectures.
Fixes#33924
Change-Id: I43328d878c3e449ebfda46f7e69963a44a511d40
Reviewed-on: https://go-review.googlesource.com/c/go/+/192619
Reviewed-by: Daniel Martí <mvdan@mvdan.cc>
This CL eliminates unnecessary pairs of I32WrapI64 and
I64ExtendI32U generated by the WASM backend for IF
statements. And it makes the total size of pkg/js_wasm/
decreases about 490KB.
Change-Id: I16b0abb686c4e30d5624323166ec2d0ec57dbe2d
Reviewed-on: https://go-review.googlesource.com/c/go/+/191758
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Richard Musiol <neelance@gmail.com>
This CL reverts CL 192097 and fixes the issue in CL 189277.
Change-Id: Icd271262e1f5019a8e01c91f91c12c1261eeb02b
Reviewed-on: https://go-review.googlesource.com/c/go/+/192519
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
The syntax of a shifted operation does not have a "$" sign for
the shift amount. Remove it.
Change-Id: I50782fe942b640076f48c2fafea4d3175be8ff99
Reviewed-on: https://go-review.googlesource.com/c/go/+/192100
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
