For the SubFromLen64 codegen test case to work as intended, we need
to fold c-(-(x-d)) into x+(c-d).
Still, some instances of LeadingZeros are not optimized into single
CLZ instructions right now (actually, the LeadingZeros micro-benchmarks
are currently still compiled with redundant adds/subs of 64, due to
interference of loop optimizations before lowering), but perf numbers
indicate it's not that bad after all.
Micro-benchmark results on Loongson 3A5000 and 3A6000:
goos: linux
goarch: loong64
pkg: math/bits
cpu: Loongson-3A5000 @ 2500.00MHz
| bench.old | bench.new |
| sec/op | sec/op vs base |
LeadingZeros 3.660n ± 0% 1.348n ± 0% -63.17% (p=0.000 n=20)
LeadingZeros8 1.777n ± 0% 1.767n ± 0% -0.56% (p=0.000 n=20)
LeadingZeros16 2.816n ± 0% 1.770n ± 0% -37.14% (p=0.000 n=20)
LeadingZeros32 5.293n ± 1% 1.683n ± 0% -68.21% (p=0.000 n=20)
LeadingZeros64 3.622n ± 0% 1.349n ± 0% -62.76% (p=0.000 n=20)
geomean 3.229n 1.571n -51.35%
goos: linux
goarch: loong64
pkg: math/bits
cpu: Loongson-3A6000 @ 2500.00MHz
| bench.old | bench.new |
| sec/op | sec/op vs base |
LeadingZeros 2.410n ± 0% 1.103n ± 1% -54.23% (p=0.000 n=20)
LeadingZeros8 1.236n ± 0% 1.501n ± 0% +21.44% (p=0.000 n=20)
LeadingZeros16 2.106n ± 0% 1.501n ± 0% -28.73% (p=0.000 n=20)
LeadingZeros32 2.860n ± 0% 1.324n ± 0% -53.72% (p=0.000 n=20)
LeadingZeros64 2.6135n ± 0% 0.9509n ± 0% -63.62% (p=0.000 n=20)
geomean 2.159n 1.256n -41.81%
Updates #59120
This patch is a copy of CL 483356.
Co-authored-by: WANG Xuerui <git@xen0n.name>
Change-Id: Iee81a17f7da06d77a427e73dfcc016f2b15ae556
Reviewed-on: https://go-review.googlesource.com/c/go/+/624575
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: David Chase <drchase@google.com>
Reviewed-by: Carlos Amedee <carlos@golang.org>
Reviewed-by: abner chenc <chenguoqi@loongson.cn>
Change-Id: I7d996b8d46fbeef933943f806052a30f1f8d50c3
Reviewed-on: https://go-review.googlesource.com/c/go/+/588836
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Joel Sing <joel@sing.id.au>
Reviewed-by: Tim King <taking@google.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Change-Id: I51e9832317e5dee1e3fe0772e7592b3dae95a625
Reviewed-on: https://go-review.googlesource.com/c/go/+/586797
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Auto-Submit: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
This usage shows up in quite a few places, and helps reduce
register pressure in several complex cryto functions by
removing a MOVD $0,... instruction.
Change-Id: I9444ea8f9d19bfd68fb71ea8dc34e109681b3802
Reviewed-on: https://go-review.googlesource.com/c/go/+/571055
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Run-TryBot: Paul Murphy <murp@ibm.com>
This optab matching rule was used to match signed 16 bit values shifted
left by 16 bits. Unsigned 16 bit values greater than 0x7FFF<<16 were
classified as C_U32CON which led to larger than necessary codegen.
Instead, rewrite logical/arithmetic operations in the preprocessor pass
to use the 16 bit shifted immediate operation (e.g ADDIS vs ADD). This
simplifies the optab matching rules, while also minimizing codegen size
for large unsigned values.
Note, ADDIS sign-extends the constant argument, all others do not.
For matching opcodes, this means:
MOVD $is<<16,Rx becomes ADDIS $is,Rx or ORIS $is,Rx
MOVW $is<<16,Rx becomes ADDIS $is,Rx
ADD $is<<16,[Rx,]Ry becomes ADDIS $is[Rx,]Ry
OR $is<<16,[Rx,]Ry becomes ORIS $is[Rx,]Ry
XOR $is<<16,[Rx,]Ry becomes XORIS $is[Rx,]Ry
Change-Id: I1a988d9f52517a04bb8dc2e41d7caf3d5fff867c
Reviewed-on: https://go-review.googlesource.com/c/go/+/536735
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
Run-TryBot: Paul Murphy <murp@ibm.com>
Reviewed-by: Heschi Kreinick <heschi@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Stop using BTSconst and friends when ORLconst can be used instead.
OR can be issued by more function units than BTS can, so it could
lead to better IPC. OR might take a few more bytes to encode, but
not a lot more.
Still use BTSconst for cases where the constant otherwise wouldn't
fit and would require a separate movabs instruction to materialize
the constant. This happens when setting bits 31-63 of 64-bit targets.
Add BTS-to-memory operations so we don't need to load/bts/store.
Fixes#61694
Change-Id: I00379608df8fb0167cb01466e97d11dec7c1596c
Reviewed-on: https://go-review.googlesource.com/c/go/+/515755
Reviewed-by: Keith Randall <khr@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
This CL add support for instrinsifying the TrialingZeros{8,32,64}
functions for 386 architecture. We need handle the case when the input
is 0, which could lead to undefined output from the BSFL instruction.
Next CL will remove the assembly code in runtime/internal/sys package.
Change-Id: Ic168edf68e81bf69a536102100fdd3f56f0f4a1b
Reviewed-on: https://go-review.googlesource.com/c/go/+/475735
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Run-TryBot: Wayne Zuo <wdvxdr@golangcn.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
The BSWAPL instruction is supported in i486 and newer.
https://github.com/golang/go/wiki/MinimumRequirements#386 says we
support "All Pentium MMX or later". The Pentium is also referred to as
i586, so that we are safe with these instructions.
Change-Id: I6dea1f9d864a45bb07c8f8f35a81cfe16cca216c
Reviewed-on: https://go-review.googlesource.com/c/go/+/465515
Run-TryBot: Wayne Zuo <wdvxdr@golangcn.org>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Keith Randall <khr@google.com>
This change intrinsifies ReverseBytes{16|32|64} by generating the
corresponding new instructions in Power10: brh, brd and brw and
adds a verification test for the same.
On Power 9 and 8, the .go code performs optimally as it is.
Performance improvement seen on Power10:
ReverseBytes32 1.38ns ± 0% 1.18ns ± 0% -14.2
ReverseBytes64 1.52ns ± 0% 1.11ns ± 0% -26.87
ReverseBytes16 1.41ns ± 1% 1.18ns ± 0% -16.47
Change-Id: I88f127f3ab9ba24a772becc21ad90acfba324b37
Reviewed-on: https://go-review.googlesource.com/c/go/+/446675
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
Manually consolidate the remaining ppc64/ppc64le test which
are not so trivial to automatically merge.
The remaining ppc64le tests are limited to cases where load/stores are
merged (this only happens on ppc64le) and the race detector (only
supported on ppc64le).
Change-Id: I1f9c0f3d3ddbb7fbbd8c81fbbd6537394fba63ce
Reviewed-on: https://go-review.googlesource.com/c/go/+/463217
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
Run-TryBot: Paul Murphy <murp@ibm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Use a small python script to consolidate duplicate
ppc64/ppc64le tests into a single ppc64x codegen test.
This makes small assumption that anytime two tests with
for different arch/variant combos exists, those tests
can be combined into a single ppc64x test.
E.x:
// ppc64le: foo
// ppc64le/power9: foo
into
// ppc64x: foo
or
// ppc64: foo
// ppc64le: foo
into
// ppc64x: foo
import glob
import re
files = glob.glob("codegen/*.go")
for file in files:
with open(file) as f:
text = [l for l in f]
i = 0
while i < len(text):
first = re.match("\s*// ?ppc64(le)?(/power[89])?:(.*)", text[i])
if first:
j = i+1
while j < len(text):
second = re.match("\s*// ?ppc64(le)?(/power[89])?:(.*)", text[j])
if not second:
break
if (not first.group(2) or first.group(2) == second.group(2)) and first.group(3) == second.group(3):
text[i] = re.sub(" ?ppc64(le|x)?"," ppc64x",text[i])
text=text[:j] + (text[j+1:])
else:
j += 1
i+=1
with open(file, 'w') as f:
f.write("".join(text))
Change-Id: Ic6b009b54eacaadc5a23db9c5a3bf7331b595821
Reviewed-on: https://go-review.googlesource.com/c/go/+/463220
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Bryan Mills <bcmills@google.com>
Run-TryBot: Paul Murphy <murp@ibm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
This needs to be as low as possible while not breaking priority
assumptions of other scores to correctly schedule carry chains.
Prior to the arm64 changes, it was set below ReadTuple. At the time,
this prevented the MulHiLo implementation on PPC64 from occluding
the scheduling of a full carry chain.
Memory scores can also prevent better scheduling, as can be observed
with crypto/internal/edwards25519/field.feMulGeneric.
Fixes#56497
Change-Id: Ia4b54e6dffcce584faf46b1b8d7cea18a3913887
Reviewed-on: https://go-review.googlesource.com/c/go/+/447435
Reviewed-by: Cherry Mui <cherryyz@google.com>
Reviewed-by: Keith Randall <khr@google.com>
Run-TryBot: Paul Murphy <murp@ibm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Bryan Mills <bcmills@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
This CL optimizes RotateLeft8/16 on arm64.
For 16 bits, we form a 32 bits register by duplicating two 16 bits
registers, then use RORW instruction to do the rotate shift.
For 8 bits, we just use LSR and LSL instead of RORW because the code is
simpler.
Benchmark Old ThisCL delta
RotateLeft8-46 2.16 ns/op 1.73 ns/op -19.70%
RotateLeft16-46 2.16 ns/op 1.54 ns/op -28.53%
Change-Id: I09cde4383d12e31876a57f8cdfd3bb4f324fadb0
Reviewed-on: https://go-review.googlesource.com/c/go/+/420976
Reviewed-by: Keith Randall <khr@google.com>
Auto-Submit: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Heschi Kreinick <heschi@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
This is a follow up of CL 425101 on RISCV64.
According to RISCV Volume 1, Unprivileged Spec v. 20191213 Chapter 7.1:
If both the high and low bits of the same product are required, then the
recommended code sequence is: MULH[[S]U] rdh, rs1, rs2; MUL rdl, rs1, rs2
(source register specifiers must be in same order and rdh cannot be the
same as rs1 or rs2). Microarchitectures can then fuse these into a single
multiply operation instead of performing two separate multiplies.
So we should not split Muluhilo to separate instructions.
Updates #54607
Change-Id: If47461f3aaaf00e27cd583a9990e144fb8bcdb17
Reviewed-on: https://go-review.googlesource.com/c/go/+/425203
Auto-Submit: Keith Randall <khr@golang.org>
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Wayne Zuo <wdvxdr@golangcn.org>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Cherry Mui <cherryyz@google.com>
On ARM64 we use two separate instructions to compute the hi and lo
results of a 64x64->128 multiplication. Lower to two separate ops
so if only one result is needed we can deadcode the other.
Fixes#54607.
Change-Id: Ib023e77eb2b2b0bcf467b45471cb8a294bce6f90
Reviewed-on: https://go-review.googlesource.com/c/go/+/425101
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Keith Randall <khr@google.com>
Updated multiple tests in test/codegen: math.go, mathbits.go, shift.go
and slices.go to verify on ppc64/ppc64le as well
Change-Id: Id88dd41569b7097819fb4d451b615f69cf7f7a94
Reviewed-on: https://go-review.googlesource.com/c/go/+/412115
TryBot-Result: Gopher Robot <gobot@golang.org>
Run-TryBot: Archana Ravindar <aravind5@in.ibm.com>
Reviewed-by: Than McIntosh <thanm@google.com>
Reviewed-by: Paul Murphy <murp@ibm.com>
Reviewed-by: Ian Lance Taylor <iant@google.com>
math/bits.Add64 and math/bits.Sub64 now lower and optimize
directly in SSA form.
The optimization of carry chains focuses around eliding
XER<->GPR transfers of the CA bit when used exclusively as an
input to a single carry operations, or when the CA value is
known.
This also adds support for handling XER spills in the assembler
which could happen if carry chains contain inter-dependencies
on each other (which seems very unlikely with practical usage),
or a clobber happens (SRAW/SRAD/SUBFC operations clobber CA).
With PPC64 Add64/Sub64 lowering into SSA and this patch, the net
performance difference in crypto/elliptic benchmarks on P9/ppc64le
are:
name old time/op new time/op delta
ScalarBaseMult/P256 46.3µs ± 0% 46.9µs ± 0% +1.34%
ScalarBaseMult/P224 356µs ± 0% 209µs ± 0% -41.14%
ScalarBaseMult/P384 1.20ms ± 0% 0.57ms ± 0% -52.14%
ScalarBaseMult/P521 3.38ms ± 0% 1.44ms ± 0% -57.27%
ScalarMult/P256 199µs ± 0% 199µs ± 0% -0.17%
ScalarMult/P224 357µs ± 0% 212µs ± 0% -40.56%
ScalarMult/P384 1.20ms ± 0% 0.58ms ± 0% -51.86%
ScalarMult/P521 3.37ms ± 0% 1.44ms ± 0% -57.32%
MarshalUnmarshal/P256/Uncompressed 2.59µs ± 0% 2.52µs ± 0% -2.63%
MarshalUnmarshal/P256/Compressed 2.58µs ± 0% 2.52µs ± 0% -2.06%
MarshalUnmarshal/P224/Uncompressed 1.54µs ± 0% 1.40µs ± 0% -9.42%
MarshalUnmarshal/P224/Compressed 1.54µs ± 0% 1.39µs ± 0% -9.87%
MarshalUnmarshal/P384/Uncompressed 2.40µs ± 0% 1.80µs ± 0% -24.93%
MarshalUnmarshal/P384/Compressed 2.35µs ± 0% 1.81µs ± 0% -23.03%
MarshalUnmarshal/P521/Uncompressed 3.79µs ± 0% 2.58µs ± 0% -31.81%
MarshalUnmarshal/P521/Compressed 3.80µs ± 0% 2.60µs ± 0% -31.67%
Note, P256 uses an asm implementation, thus, little variation is expected.
Change-Id: I88a24f6bf0f4f285c649e40243b1ab69cc452b71
Reviewed-on: https://go-review.googlesource.com/c/go/+/346870
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Dmitri Shuralyov <dmitshur@google.com>
Run-TryBot: Paul Murphy <murp@ibm.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Ian Lance Taylor <iant@google.com>
LZCNT is similar to BSR, but BSR(x) is undefined when x == 0, so using
LZCNT can avoid a special case for zero input. Except that case,
LZCNTQ(x) == 63-BSRQ(x) and LZCNTL(x) == 31-BSRL(x).
And according to https://www.agner.org/optimize/instruction_tables.pdf,
LZCNT instructions are much faster than BSR on AMD CPU.
name old time/op new time/op delta
LeadingZeros-8 0.91ns ± 1% 0.80ns ± 7% -11.68% (p=0.000 n=9+9)
LeadingZeros8-8 0.98ns ±15% 0.91ns ± 1% -7.34% (p=0.000 n=9+9)
LeadingZeros16-8 0.94ns ± 3% 0.92ns ± 2% -2.36% (p=0.001 n=10+10)
LeadingZeros32-8 0.89ns ± 1% 0.78ns ± 2% -12.49% (p=0.000 n=10+10)
LeadingZeros64-8 0.92ns ± 1% 0.78ns ± 1% -14.48% (p=0.000 n=10+10)
Change-Id: I125147fe3d6994a4cfe558432780408e9a27557a
Reviewed-on: https://go-review.googlesource.com/c/go/+/396794
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Emmanuel Odeke <emmanuel@orijtech.com>
Run-TryBot: Emmanuel Odeke <emmanuel@orijtech.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
It was noticed through some other investigation that BitLen32
was not generating the best code and found that it wasn't recognized
as an intrinsic. This corrects that and enables the test for PPC64.
Change-Id: Iab496a8830c8552f507b7292649b1b660f3848b5
Reviewed-on: https://go-review.googlesource.com/c/go/+/355872
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Trust: Lynn Boger <laboger@linux.vnet.ibm.com>
Reviewed-by: Cherry Mui <cherryyz@google.com>
In case of amd64 the compiler issues checks if extensions are
available on a platform. With GOAMD64 microarchitecture levels
provided, some of the checks could be eliminated.
Change-Id: If15c178bcae273b2ce7d3673415cb8849292e087
Reviewed-on: https://go-review.googlesource.com/c/go/+/352010
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Go Bot <gobot@golang.org>
Fixes this failure:
go test cmd/compile/internal/ssa -run TestStmtLines -v
=== RUN TestStmtLines
stmtlines_test.go:115: Saw too many (amd64, > 1%) lines without
statement marks, total=88263, nostmt=1930
('-run TestStmtLines -v' lists failing lines)
The failure has two causes.
One is that the first-line adjuster in code generation was relocating
"first lines" to instructions that would either not have any code generated,
or would have the statment marker removed by a different believed-good heuristic.
The other was that statement boundaries were getting attached to register
values (that with the old ABI were loads from the stack, hence real instructions).
The register values disappear at code generation.
The fixes are to (1) note that certain instructions are not good choices for
"first value" and skip them, and (2) in an expandCalls post-pass, look for
register valued instructions and under appropriate conditions move their
statement marker to a compatible use.
Also updates TestStmtLines to always log the score, for easier comparison of
minor compiler changes.
Updates #40724.
Change-Id: I485573ce900e292d7c44574adb7629cdb4695c3f
Reviewed-on: https://go-review.googlesource.com/c/go/+/309649
Trust: David Chase <drchase@google.com>
Run-TryBot: David Chase <drchase@google.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Optimize combinations of left and right shifts by a constant value
into a 'rotate then insert selected bits [into zero]' instruction.
Use the same instruction for contiguous masks since it has some
benefits over 'and immediate' (not restricted to 32-bits, does not
overwrite source register).
To keep the complexity of this change under control I've only
implemented 64 bit operations for now.
There are a lot more optimizations that can be done with this
instruction family. However, since their function overlaps with other
instructions we need to be somewhat careful not to break existing
optimization rules by creating optimization dead ends. This is
particularly true of the load/store merging rules which contain lots
of zero extensions and shifts.
This CL does interfere with the store merging rules when an operand
is shifted left before it is stored:
binary.BigEndian.PutUint64(b, x << 1)
This is unfortunate but it's not critical and somewhat complex so
I plan to fix that in a follow up CL.
file before after Δ %
addr2line 4117446 4117282 -164 -0.004%
api 4945184 4942752 -2432 -0.049%
asm 4998079 4991891 -6188 -0.124%
buildid 2685158 2684074 -1084 -0.040%
cgo 4553732 4553394 -338 -0.007%
compile 19294446 19245070 -49376 -0.256%
cover 4897105 4891319 -5786 -0.118%
dist 3544389 3542785 -1604 -0.045%
doc 3926795 3927617 +822 +0.021%
fix 3302958 3293868 -9090 -0.275%
link 6546274 6543456 -2818 -0.043%
nm 4102021 4100825 -1196 -0.029%
objdump 4542431 4548483 +6052 +0.133%
pack 2482465 2416389 -66076 -2.662%
pprof 13366541 13363915 -2626 -0.020%
test2json 2829007 2761515 -67492 -2.386%
trace 10216164 10219684 +3520 +0.034%
vet 6773956 6773572 -384 -0.006%
total 107124151 106917891 -206260 -0.193%
Change-Id: I7591cce41e06867ba10a745daae9333513062746
Reviewed-on: https://go-review.googlesource.com/c/go/+/233317
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Go Bot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Trust: Michael Munday <mike.munday@ibm.com>
This merges an lis + subf into subfic, and for 32b constants
lwa + subf into oris + ori + subf.
The carry bit is no longer used in code generation, therefore
I think we can clobber it as needed. Note, lowered borrow/carry
arithmetic is self-contained and thus is not affected.
A few extra rules are added to ensure early transformations to
SUBFCconst don't trip up earlier rules, fold constant operations,
or otherwise simplify lowering. Likewise, tests are added to
ensure all rules are hit. Generic constant folding catches
trivial cases, however some lowering rules insert arithmetic
which can introduce new opportunities (e.g BitLen or Slicemask).
I couldn't find a specific benchmark to demonstrate noteworthy
improvements, but this is generating subfic in many of the default
bent test binaries, so we are at least saving a little code space.
Change-Id: Iad7c6e5767eaa9dc24dc1c989bd1c8cfe1982012
Reviewed-on: https://go-review.googlesource.com/c/go/+/249461
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
This CL optimizes code that uses a carry from a function such as
bits.Add64 as the condition in an if statement. For example:
x, c := bits.Add64(a, b, 0)
if c != 0 {
panic("overflow")
}
Rather than converting the carry into a 0 or a 1 value and using
that as an input to a comparison instruction the carry flag is now
used as the input to a conditional branch directly. This typically
removes an ADD LOGICAL WITH CARRY instruction when user code is
doing overflow detection and is closer to the code that a user
would expect to generate.
Change-Id: I950431270955ab72f1b5c6db873b6abe769be0da
Reviewed-on: https://go-review.googlesource.com/c/go/+/219757
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Before using some CPU instructions, we must check for their presence.
We use global variables in the runtime package to record features.
Prior to this CL, we issued a regular memory load for these features.
The downside to this is that, because it is a regular memory load,
it cannot be hoisted out of loops or otherwise reordered with other loads.
This CL introduces a new intrinsic just for checking cpu features.
It still ends up resulting in a memory load, but that memory load can
now be floated to the entry block and rematerialized as needed.
One downside is that the regular load could be combined with the comparison
into a CMPBconstload+NE. This new intrinsic cannot; it generates MOVB+TESTB+NE.
(It is possible that MOVBQZX+TESTQ+NE would be better.)
This CL does only amd64. It is easy to extend to other architectures.
For the benchmark in #36196, on my machine, this offers a mild speedup.
name old time/op new time/op delta
FMA-8 1.39ns ± 6% 1.29ns ± 9% -7.19% (p=0.000 n=97+96)
NonFMA-8 2.03ns ±11% 2.04ns ±12% ~ (p=0.618 n=99+98)
Updates #15808
Updates #36196
Change-Id: I75e2fcfcf5a6df1bdb80657a7143bed69fca6deb
Reviewed-on: https://go-review.googlesource.com/c/go/+/212360
Run-TryBot: Josh Bleecher Snyder <josharian@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
Reviewed-by: Giovanni Bajo <rasky@develer.com>
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 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>
This CL optimizes math.bits.TrailingZeros16 on 386 with
a pair of BSFL and ORL instrcutions.
The case TrailingZeros16-4 of the benchmark test in
math/bits shows big improvement.
name old time/op new time/op delta
TrailingZeros16-4 1.55ns ± 1% 0.87ns ± 1% -43.87% (p=0.000 n=50+49)
Change-Id: Ia899975b0e46f45dcd20223b713ed632bc32740b
Reviewed-on: https://go-review.googlesource.com/c/go/+/189277
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
This CL adds intrinsics for the 64-bit addition and subtraction
functions in math/bits. These intrinsics use the condition code
to propagate the carry or borrow bit.
To make the carry chains more efficient I've removed the
'clobberFlags' property from most of the load and store
operations. Originally these ops did clobber flags when using
offsets that didn't fit in a signed 20-bit integer, however
that is no longer true.
As with other platforms the intrinsics are faster when executed
in a chain rather than a loop because currently we need to spill
and restore the carry bit between each loop iteration. We may
be able to reduce the need to do this on s390x (e.g. by using
compare-and-branch instructions that do not clobber flags) in the
future.
name old time/op new time/op delta
Add64 1.21ns ± 2% 2.03ns ± 2% +67.18% (p=0.000 n=7+10)
Add64multiple 2.98ns ± 3% 1.03ns ± 0% -65.39% (p=0.000 n=10+9)
Sub64 1.23ns ± 4% 2.03ns ± 1% +64.85% (p=0.000 n=10+10)
Sub64multiple 3.73ns ± 4% 1.04ns ± 1% -72.28% (p=0.000 n=10+8)
Change-Id: I913bbd5e19e6b95bef52f5bc4f14d6fe40119083
Reviewed-on: https://go-review.googlesource.com/c/go/+/174303
Run-TryBot: Michael Munday <mike.munday@ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
This change creates an intrinsic for Add64 for ppc64x and adds a
testcase for it.
name old time/op new time/op delta
Add64-160 1.90ns ±40% 2.29ns ± 0% ~ (p=0.119 n=5+5)
Add64multiple-160 6.69ns ± 2% 2.45ns ± 4% -63.47% (p=0.016 n=4+5)
Change-Id: I9abe6fb023fdf62eea3c9b46a1820f60bb0a7f97
Reviewed-on: https://go-review.googlesource.com/c/go/+/173758
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
Run-TryBot: Carlos Eduardo Seo <cseo@linux.vnet.ibm.com>
This CL instrinsifies Add64 with arm64 instruction sequence ADDS, ADCS
and ADC, and optimzes the case of carry chains.The CL also changes the
test code so that the intrinsic implementation can be tested.
Benchmarks:
name old time/op new time/op delta
Add-224 2.500000ns +- 0% 2.090000ns +- 4% -16.40% (p=0.000 n=9+10)
Add32-224 2.500000ns +- 0% 2.500000ns +- 0% ~ (all equal)
Add64-224 2.500000ns +- 0% 1.577778ns +- 2% -36.89% (p=0.000 n=10+9)
Add64multiple-224 6.000000ns +- 0% 2.000000ns +- 0% -66.67% (p=0.000 n=10+10)
Change-Id: I6ee91c9a85c16cc72ade5fd94868c579f16c7615
Reviewed-on: https://go-review.googlesource.com/c/go/+/159017
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
