qbit/go
1
0
Fork 0
mirror of https://github.com/golang/go synced 2024-11-07 14:36:17 -07:00
Code Releases Activity
41,403 Commits 58 Branches 402 Tags 522 MiB
cc21e4d130
Commit Graph

17 Commits

Author SHA1 Message Date
Martin Möhrmann
f41451e7eb compile: prefer an AND instead of SHR+SHL instructions 
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>
 2019-09-24 20:30:59 +00:00
Bryan C. Mills
34fe8295c5 Revert "compile: prefer an AND instead of SHR+SHL instructions" 
This reverts CL 194297.

Reason for revert: introduced register allocation failures on PPC64LE builders.

Updates #33826
Updates #32781
Updates #34468

Change-Id: I7d0b55df8cdf8e7d2277f1814299b083c2692e48
Reviewed-on: https://go-review.googlesource.com/c/go/+/196957
Run-TryBot: Bryan C. Mills <bcmills@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Dmitri Shuralyov <dmitshur@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Reviewed-by: Martin Möhrmann <moehrmann@google.com>
 2019-09-23 15:20:12 +00:00
Martin Möhrmann
4e2b84ffc5 compile: prefer an AND instead of SHR+SHL instructions 
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>
 2019-09-21 18:00:13 +00:00
Martin Möhrmann
5bb59b6d16 Revert "compile: prefer an AND instead of SHR+SHL instructions" 
This reverts commit 9ec7074a94.

Reason for revert: broke s390x (copysign, abs) and arm64 (bitfield) tests.

Change-Id: I16c1b389c062e8c4aa5de079f1d46c9b25b0db52
Reviewed-on: https://go-review.googlesource.com/c/go/+/193850
Run-TryBot: Martin Möhrmann <moehrmann@google.com>
Reviewed-by: Agniva De Sarker <agniva.quicksilver@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
 2019-09-09 07:33:25 +00:00
Martin Möhrmann
9ec7074a94 compile: prefer an AND instead of SHR+SHL instructions 
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>
 2019-09-09 06:49:17 +00:00
Ben Shi
c683ab8128 cmd/compile: optimize ARM's math.Abs 
This CL optimizes math.Abs to an inline ABSD instruction on ARM.

The benchmark results of src/math/ show big improvements.
name                   old time/op  new time/op  delta
Acos-4                  181ns ± 0%   182ns ± 0%   +0.30%  (p=0.000 n=40+40)
Acosh-4                 202ns ± 0%   202ns ± 0%     ~     (all equal)
Asin-4                  163ns ± 0%   163ns ± 0%     ~     (all equal)
Asinh-4                 242ns ± 0%   242ns ± 0%     ~     (all equal)
Atan-4                  120ns ± 0%   121ns ± 0%   +0.83%  (p=0.000 n=40+40)
Atanh-4                 202ns ± 0%   202ns ± 0%     ~     (all equal)
Atan2-4                 173ns ± 0%   173ns ± 0%     ~     (all equal)
Cbrt-4                 1.06µs ± 0%  1.06µs ± 0%   +0.09%  (p=0.000 n=39+37)
Ceil-4                 72.9ns ± 0%  72.8ns ± 0%     ~     (p=0.237 n=40+40)
Copysign-4             13.2ns ± 0%  13.2ns ± 0%     ~     (all equal)
Cos-4                   193ns ± 0%   183ns ± 0%   -5.18%  (p=0.000 n=40+40)
Cosh-4                  254ns ± 0%   239ns ± 0%   -5.91%  (p=0.000 n=40+40)
Erf-4                   112ns ± 0%   112ns ± 0%     ~     (all equal)
Erfc-4                  117ns ± 0%   117ns ± 0%     ~     (all equal)
Erfinv-4                127ns ± 0%   127ns ± 1%     ~     (p=0.492 n=40+40)
Erfcinv-4               128ns ± 0%   128ns ± 0%     ~     (all equal)
Exp-4                   212ns ± 0%   206ns ± 0%   -3.05%  (p=0.000 n=40+40)
ExpGo-4                 216ns ± 0%   209ns ± 0%   -3.24%  (p=0.000 n=40+40)
Expm1-4                 142ns ± 0%   142ns ± 0%     ~     (all equal)
Exp2-4                  191ns ± 0%   184ns ± 0%   -3.45%  (p=0.000 n=40+40)
Exp2Go-4                194ns ± 0%   187ns ± 0%   -3.61%  (p=0.000 n=40+40)
Abs-4                  14.4ns ± 0%   6.3ns ± 0%  -56.39%  (p=0.000 n=38+39)
Dim-4                  12.6ns ± 0%  12.6ns ± 0%     ~     (all equal)
Floor-4                49.6ns ± 0%  49.6ns ± 0%     ~     (all equal)
Max-4                  27.6ns ± 0%  27.6ns ± 0%     ~     (all equal)
Min-4                  27.0ns ± 0%  27.0ns ± 0%     ~     (all equal)
Mod-4                   349ns ± 0%   305ns ± 1%  -12.55%  (p=0.000 n=33+40)
Frexp-4                54.0ns ± 0%  47.1ns ± 0%  -12.78%  (p=0.000 n=38+38)
Gamma-4                 242ns ± 0%   234ns ± 0%   -3.16%  (p=0.000 n=36+40)
Hypot-4                84.8ns ± 0%  67.8ns ± 0%  -20.05%  (p=0.000 n=31+35)
HypotGo-4              88.5ns ± 0%  71.6ns ± 0%  -19.12%  (p=0.000 n=40+38)
Ilogb-4                45.8ns ± 0%  38.9ns ± 0%  -15.12%  (p=0.000 n=40+32)
J0-4                    821ns ± 0%   802ns ± 0%   -2.33%  (p=0.000 n=33+40)
J1-4                    816ns ± 0%   807ns ± 0%   -1.05%  (p=0.000 n=40+29)
Jn-4                   1.67µs ± 0%  1.65µs ± 0%   -1.45%  (p=0.000 n=40+39)
Ldexp-4                61.5ns ± 0%  54.6ns ± 0%  -11.27%  (p=0.000 n=40+32)
Lgamma-4                188ns ± 0%   188ns ± 0%     ~     (all equal)
Log-4                   154ns ± 0%   147ns ± 0%   -4.78%  (p=0.000 n=40+40)
Logb-4                 50.9ns ± 0%  42.7ns ± 0%  -16.11%  (p=0.000 n=34+39)
Log1p-4                 160ns ± 0%   159ns ± 0%     ~     (p=0.828 n=40+40)
Log10-4                 173ns ± 0%   166ns ± 0%   -4.05%  (p=0.000 n=40+40)
Log2-4                 65.3ns ± 0%  58.4ns ± 0%  -10.57%  (p=0.000 n=37+37)
Modf-4                 36.4ns ± 0%  36.4ns ± 0%     ~     (all equal)
Nextafter32-4          36.4ns ± 0%  36.4ns ± 0%     ~     (all equal)
Nextafter64-4          32.7ns ± 0%  32.6ns ± 0%     ~     (p=0.375 n=40+40)
PowInt-4                300ns ± 0%   277ns ± 0%   -7.78%  (p=0.000 n=40+40)
PowFrac-4               676ns ± 0%   635ns ± 0%   -6.00%  (p=0.000 n=40+35)
Pow10Pos-4             17.6ns ± 0%  17.6ns ± 0%     ~     (all equal)
Pow10Neg-4             22.0ns ± 0%  22.0ns ± 0%     ~     (all equal)
Round-4                30.1ns ± 0%  30.1ns ± 0%     ~     (all equal)
RoundToEven-4          38.9ns ± 0%  38.9ns ± 0%     ~     (all equal)
Remainder-4             291ns ± 0%   263ns ± 0%   -9.62%  (p=0.000 n=40+40)
Signbit-4              11.3ns ± 0%  11.3ns ± 0%     ~     (all equal)
Sin-4                   185ns ± 0%   185ns ± 0%     ~     (all equal)
Sincos-4                230ns ± 0%   230ns ± 0%     ~     (all equal)
Sinh-4                  253ns ± 0%   246ns ± 0%   -2.77%  (p=0.000 n=39+39)
SqrtIndirect-4         41.4ns ± 0%  41.4ns ± 0%     ~     (all equal)
SqrtLatency-4          13.8ns ± 0%  13.8ns ± 0%     ~     (all equal)
SqrtIndirectLatency-4  37.0ns ± 0%  37.0ns ± 0%     ~     (p=0.632 n=40+40)
SqrtGoLatency-4         911ns ± 0%   911ns ± 0%   +0.08%  (p=0.000 n=40+40)
SqrtPrime-4            13.2µs ± 0%  13.2µs ± 0%   +0.01%  (p=0.038 n=38+40)
Tan-4                   205ns ± 0%   205ns ± 0%     ~     (all equal)
Tanh-4                  264ns ± 0%   247ns ± 0%   -6.44%  (p=0.000 n=39+32)
Trunc-4                45.2ns ± 0%  45.2ns ± 0%     ~     (all equal)
Y0-4                    796ns ± 0%   792ns ± 0%   -0.55%  (p=0.000 n=35+40)
Y1-4                    804ns ± 0%   797ns ± 0%   -0.82%  (p=0.000 n=24+40)
Yn-4                   1.64µs ± 0%  1.62µs ± 0%   -1.27%  (p=0.000 n=40+39)
Float64bits-4          8.16ns ± 0%  8.16ns ± 0%   +0.04%  (p=0.000 n=35+40)
Float64frombits-4      10.7ns ± 0%  10.7ns ± 0%     ~     (all equal)
Float32bits-4          7.53ns ± 0%  7.53ns ± 0%     ~     (p=0.760 n=40+40)
Float32frombits-4      6.91ns ± 0%  6.91ns ± 0%   -0.04%  (p=0.002 n=32+38)
[Geo mean]              111ns        106ns        -3.98%

Change-Id: I54f4fd7f5160db020b430b556bde59cc0fdb996d
Reviewed-on: https://go-review.googlesource.com/c/go/+/188678
Run-TryBot: Ben Shi <powerman1st@163.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
 2019-08-28 15:41:28 +00:00
Richard Musiol
5ee1b84959 math, math/bits: add intrinsics for wasm 
This commit adds compiler intrinsics for the packages math and
math/bits on the wasm architecture for better performance.

benchmark                        old ns/op     new ns/op     delta
BenchmarkCeil                    8.31          3.21          -61.37%
BenchmarkCopysign                5.24          3.88          -25.95%
BenchmarkAbs                     5.42          3.34          -38.38%
BenchmarkFloor                   8.29          3.18          -61.64%
BenchmarkRoundToEven             9.76          3.26          -66.60%
BenchmarkSqrtLatency             8.13          4.88          -39.98%
BenchmarkSqrtPrime               5246          3535          -32.62%
BenchmarkTrunc                   8.29          3.15          -62.00%
BenchmarkLeadingZeros            13.0          4.23          -67.46%
BenchmarkLeadingZeros8           4.65          4.42          -4.95%
BenchmarkLeadingZeros16          7.60          4.38          -42.37%
BenchmarkLeadingZeros32          10.7          4.48          -58.13%
BenchmarkLeadingZeros64          12.9          4.31          -66.59%
BenchmarkTrailingZeros           6.52          4.04          -38.04%
BenchmarkTrailingZeros8          4.57          4.14          -9.41%
BenchmarkTrailingZeros16         6.69          4.16          -37.82%
BenchmarkTrailingZeros32         6.97          4.23          -39.31%
BenchmarkTrailingZeros64         6.59          4.00          -39.30%
BenchmarkOnesCount               7.93          3.30          -58.39%
BenchmarkOnesCount8              3.56          3.19          -10.39%
BenchmarkOnesCount16             4.85          3.19          -34.23%
BenchmarkOnesCount32             7.27          3.19          -56.12%
BenchmarkOnesCount64             8.08          3.28          -59.41%
BenchmarkRotateLeft              4.88          3.80          -22.13%
BenchmarkRotateLeft64            5.03          3.63          -27.83%

Change-Id: Ic1e0c2984878be8defb6eb7eb6ee63765c793222
Reviewed-on: https://go-review.googlesource.com/c/go/+/165177
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
 2019-03-14 19:46:19 +00:00
Lynn Boger
39fa301bdc test/codegen: enable more tests for ppc64/ppc64le 
Adding cases for ppc64,ppc64le to the codegen tests
where appropriate.

Change-Id: Idf8cbe88a4ab4406a4ef1ea777bd15a58b68f3ed
Reviewed-on: https://go-review.googlesource.com/c/142557
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
 2018-10-16 19:00:53 +00:00
fanzha02
a19a83c8ef cmd/compile: optimize math.Float64(32)bits and math.Float64(32)frombits on arm64 
Use float <-> int register moves without conversion instead of stores
and loads to move float <-> int values.

Math package benchmark results.
name                 old time/op  new time/op  delta
Acosh                 153ns ± 0%   147ns ± 0%   -3.92%  (p=0.000 n=10+10)
Asinh                 183ns ± 0%   177ns ± 0%   -3.28%  (p=0.000 n=10+10)
Atanh                 157ns ± 0%   155ns ± 0%   -1.27%  (p=0.000 n=10+10)
Atan2                 118ns ± 0%   117ns ± 1%   -0.59%  (p=0.003 n=10+10)
Cbrt                  119ns ± 0%   114ns ± 0%   -4.20%  (p=0.000 n=10+10)
Copysign             7.51ns ± 0%  6.51ns ± 0%  -13.32%  (p=0.000 n=9+10)
Cos                  73.1ns ± 0%  70.6ns ± 0%   -3.42%  (p=0.000 n=10+10)
Cosh                  119ns ± 0%   121ns ± 0%   +1.68%  (p=0.000 n=10+9)
ExpGo                 154ns ± 0%   149ns ± 0%   -3.05%  (p=0.000 n=9+10)
Expm1                 101ns ± 0%    99ns ± 0%   -1.88%  (p=0.000 n=10+10)
Exp2Go                150ns ± 0%   146ns ± 0%   -2.67%  (p=0.000 n=10+10)
Abs                  7.01ns ± 0%  6.01ns ± 0%  -14.27%  (p=0.000 n=10+9)
Mod                   234ns ± 0%   212ns ± 0%   -9.40%  (p=0.000 n=9+10)
Frexp                34.5ns ± 0%  30.0ns ± 0%  -13.04%  (p=0.000 n=10+10)
Gamma                 112ns ± 0%   111ns ± 0%   -0.89%  (p=0.000 n=10+10)
Hypot                73.6ns ± 0%  68.6ns ± 0%   -6.79%  (p=0.000 n=10+10)
HypotGo              77.1ns ± 0%  72.1ns ± 0%   -6.49%  (p=0.000 n=10+10)
Ilogb                31.0ns ± 0%  28.0ns ± 0%   -9.68%  (p=0.000 n=10+10)
J0                    437ns ± 0%   434ns ± 0%   -0.62%  (p=0.000 n=10+10)
J1                    433ns ± 0%   431ns ± 0%   -0.46%  (p=0.000 n=10+10)
Jn                    927ns ± 0%   922ns ± 0%   -0.54%  (p=0.000 n=10+10)
Ldexp                41.5ns ± 0%  37.0ns ± 0%  -10.84%  (p=0.000 n=9+10)
Log                   124ns ± 0%   118ns ± 0%   -4.84%  (p=0.000 n=10+9)
Logb                 34.0ns ± 0%  32.0ns ± 0%   -5.88%  (p=0.000 n=10+10)
Log1p                 110ns ± 0%   108ns ± 0%   -1.82%  (p=0.000 n=10+10)
Log10                 136ns ± 0%   132ns ± 0%   -2.94%  (p=0.000 n=10+10)
Log2                 51.6ns ± 0%  47.1ns ± 0%   -8.72%  (p=0.000 n=10+10)
Nextafter32          33.0ns ± 0%  30.5ns ± 0%   -7.58%  (p=0.000 n=10+10)
Nextafter64          29.0ns ± 0%  26.5ns ± 0%   -8.62%  (p=0.000 n=10+10)
PowInt                169ns ± 0%   160ns ± 0%   -5.33%  (p=0.000 n=10+10)
PowFrac               375ns ± 0%   361ns ± 0%   -3.73%  (p=0.000 n=10+10)
RoundToEven          14.0ns ± 0%  12.5ns ± 0%  -10.71%  (p=0.000 n=10+10)
Remainder             206ns ± 0%   192ns ± 0%   -6.80%  (p=0.000 n=10+9)
Signbit              6.01ns ± 0%  5.51ns ± 0%   -8.32%  (p=0.000 n=10+9)
Sin                  70.1ns ± 0%  69.6ns ± 0%   -0.71%  (p=0.000 n=10+10)
Sincos               99.1ns ± 0%  99.6ns ± 0%   +0.50%  (p=0.000 n=9+10)
SqrtGoLatency         178ns ± 0%   146ns ± 0%  -17.70%  (p=0.000 n=8+10)
SqrtPrime            9.19µs ± 0%  9.20µs ± 0%   +0.01%  (p=0.000 n=9+9)
Tanh                  125ns ± 1%   127ns ± 0%   +1.36%  (p=0.000 n=10+10)
Y0                    428ns ± 0%   426ns ± 0%   -0.47%  (p=0.000 n=10+10)
Y1                    431ns ± 0%   429ns ± 0%   -0.46%  (p=0.000 n=10+9)
Yn                    906ns ± 0%   901ns ± 0%   -0.55%  (p=0.000 n=10+10)
Float64bits          4.50ns ± 0%  3.50ns ± 0%  -22.22%  (p=0.000 n=10+10)
Float64frombits      4.00ns ± 0%  3.50ns ± 0%  -12.50%  (p=0.000 n=10+9)
Float32bits          4.50ns ± 0%  3.50ns ± 0%  -22.22%  (p=0.002 n=8+10)
Float32frombits      4.00ns ± 0%  3.50ns ± 0%  -12.50%  (p=0.000 n=10+10)

Change-Id: Iba829e15d5624962fe0c699139ea783efeefabc2
Reviewed-on: https://go-review.googlesource.com/129715
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
 2018-09-17 20:49:04 +00:00
erifan01
8149db4f64 cmd/compile: intrinsify math.RoundToEven and math.Abs on arm64 
math.RoundToEven can be done by one arm64 instruction FRINTND, intrinsify it to improve performance.
The current pure Go implementation of the function Abs is translated into five instructions on arm64:
str, ldr, and, str, ldr. The intrinsic implementation requires only one instruction, so in terms of
performance, intrinsify it is worthwhile.

Benchmarks:
name           old time/op  new time/op  delta
Abs-8          3.50ns ± 0%  1.50ns ± 0%  -57.14%  (p=0.000 n=10+10)
RoundToEven-8  9.26ns ± 0%  1.50ns ± 0%  -83.80%  (p=0.000 n=10+10)

Change-Id: I9456b26ab282b544dfac0154fc86f17aed96ac3d
Reviewed-on: https://go-review.googlesource.com/116535
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
 2018-09-13 14:52:51 +00:00
fanzha02
d5377c2026 test: fix the wrong test of math.Copysign(c, -1) for arm64 
The CL 132915 added the wrong codegen test for math.Copysign(c, -1),
it should test that AND is not emitted. This CL fixes this error.

Change-Id: Ida1d3d54ebfc7f238abccbc1f70f914e1b5bfd91
Reviewed-on: https://go-review.googlesource.com/134815
Reviewed-by: Giovanni Bajo <rasky@develer.com>
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
 2018-09-12 15:34:20 +00:00
fanzha02
2e5c32518c cmd/compile: optimize math.Copysign on arm64 
Add rewrite rules to optimize math.Copysign() when the second
argument is negative floating point constant.

For example, math.Copysign(c, -2): The previous compile output is
"AND $9223372036854775807, R0, R0; ORR $-9223372036854775808, R0, R0".
The optimized compile output is "ORR $-9223372036854775808, R0, R0"

Math package benchmark results.
name                   old time/op  new time/op  delta
Copysign-8             2.61ns ± 2%  2.49ns ± 0%  -4.55%  (p=0.000 n=10+10)
Cos-8                  43.0ns ± 0%  41.5ns ± 0%  -3.49%  (p=0.000 n=10+10)
Cosh-8                 98.6ns ± 0%  98.1ns ± 0%  -0.51%  (p=0.000 n=10+10)
ExpGo-8                 107ns ± 0%   105ns ± 0%  -1.87%  (p=0.000 n=10+10)
Exp2Go-8                100ns ± 0%   100ns ± 0%  +0.39%  (p=0.000 n=10+8)
Max-8                  6.56ns ± 2%  6.45ns ± 1%  -1.63%  (p=0.002 n=10+10)
Min-8                  6.66ns ± 3%  6.47ns ± 2%  -2.82%  (p=0.006 n=10+10)
Mod-8                   107ns ± 1%   104ns ± 1%  -2.72%  (p=0.000 n=10+10)
Frexp-8                11.5ns ± 1%  11.0ns ± 0%  -4.56%  (p=0.000 n=8+10)
HypotGo-8              19.4ns ± 0%  19.4ns ± 0%  +0.36%  (p=0.019 n=10+10)
Ilogb-8                8.63ns ± 0%  8.51ns ± 0%  -1.36%  (p=0.000 n=10+10)
Jn-8                    584ns ± 0%   585ns ± 0%  +0.17%  (p=0.000 n=7+8)
Ldexp-8                13.8ns ± 0%  13.5ns ± 0%  -2.17%  (p=0.002 n=8+10)
Logb-8                 10.2ns ± 0%   9.9ns ± 0%  -2.65%  (p=0.000 n=10+7)
Nextafter64-8          7.54ns ± 0%  7.51ns ± 0%  -0.37%  (p=0.000 n=10+10)
Remainder-8            73.5ns ± 1%  70.4ns ± 1%  -4.27%  (p=0.000 n=10+10)
SqrtGoLatency-8        79.6ns ± 0%  76.2ns ± 0%  -4.30%  (p=0.000 n=9+10)
Yn-8                    582ns ± 0%   579ns ± 0%  -0.52%  (p=0.000 n=10+10)

Change-Id: I0c9cd1ea87435e7b8bab94b4e79e6e29785f25b1
Reviewed-on: https://go-review.googlesource.com/132915
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
 2018-09-06 19:57:25 +00:00
Milan Knezevic
2959128dc5 cmd/compile: add softfloat support to mips64{,le} 
mips64 softfloat support is based on mips implementation and introduces
new enviroment variable GOMIPS64.

GOMIPS64 is a GOARCH=mips64{,le} specific option, for a choice between
hard-float and soft-float. Valid values are 'hardfloat' (default) and
'softfloat'. It is passed to the assembler as
'GOMIPS64_{hardfloat,softfloat}'.

Change-Id: I7f73078627f7cb37c588a38fb5c997fe09c56134
Reviewed-on: https://go-review.googlesource.com/108475
Reviewed-by: Cherry Zhang <cherryyz@google.com>
Run-TryBot: Cherry Zhang <cherryyz@google.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
 2018-04-27 14:50:17 +00:00
Carlos Eduardo Seo
ebb67d993a cmd/compile, cmd/internal/obj/ppc64: make math.Round an intrinsic on ppc64x 
This change implements math.Round as an intrinsic on ppc64x so it can be
done using a single instruction.

benchmark                   old ns/op     new ns/op     delta
BenchmarkRound-16           2.60          0.69          -73.46%

Change-Id: I9408363e96201abdfc73ced7bcd5f0c29db006a8
Reviewed-on: https://go-review.googlesource.com/109395
Run-TryBot: Lynn Boger <laboger@linux.vnet.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Lynn Boger <laboger@linux.vnet.ibm.com>
 2018-04-26 14:12:09 +00:00
Giovanni Bajo
284ba47b49 test: run codegen tests on all supported architecture variants 
This CL makes the codegen testsuite automatically test all
architecture variants for architecture specified in tests. For
instance, if a test file specifies a "arm" test, it will be
automatically run on all GOARM variants (5,6,7), to increase
the coverage.

The CL also introduces a syntax to specify only a specific
variant (eg: "arm/7") in case the test makes sense only there.
The same syntax also allows to specify the operating system
in case it matters (eg: "plan9/386/sse2").

Fixes #24658

Change-Id: I2eba8b918f51bb6a77a8431a309f8b71af07ea22
Reviewed-on: https://go-review.googlesource.com/107315
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
 2018-04-15 20:02:43 +00:00
Giovanni Bajo
79112707bb cmd/compile: add patterns for bit set/clear/complement on amd64 
This patch completes implementation of BT(Q|L), and adds support
for BT(S|R|C)(Q|L).

Example of code changes from time.(*Time).addSec:

        if t.wall&hasMonotonic != 0 {
  0x1073465               488b08                  MOVQ 0(AX), CX
  0x1073468               4889ca                  MOVQ CX, DX
  0x107346b               48c1e93f                SHRQ $0x3f, CX
  0x107346f               48c1e13f                SHLQ $0x3f, CX
  0x1073473               48f7c1ffffffff          TESTQ $-0x1, CX
  0x107347a               746b                    JE 0x10734e7

        if t.wall&hasMonotonic != 0 {
  0x1073435               488b08                  MOVQ 0(AX), CX
  0x1073438               480fbae13f              BTQ $0x3f, CX
  0x107343d               7363                    JAE 0x10734a2

Another example:

                        t.wall = t.wall&nsecMask | uint64(dsec)<<nsecShift | hasMonotonic
  0x10734c8               4881e1ffffff3f          ANDQ $0x3fffffff, CX
  0x10734cf               48c1e61e                SHLQ $0x1e, SI
  0x10734d3               4809ce                  ORQ CX, SI
  0x10734d6               48b90000000000000080    MOVQ $0x8000000000000000, CX
  0x10734e0               4809f1                  ORQ SI, CX
  0x10734e3               488908                  MOVQ CX, 0(AX)

                        t.wall = t.wall&nsecMask | uint64(dsec)<<nsecShift | hasMonotonic
  0x107348b		4881e2ffffff3f		ANDQ $0x3fffffff, DX
  0x1073492		48c1e61e		SHLQ $0x1e, SI
  0x1073496		4809f2			ORQ SI, DX
  0x1073499		480fbaea3f		BTSQ $0x3f, DX
  0x107349e		488910			MOVQ DX, 0(AX)

Go1 benchmarks seem unaffected, and I would be surprised
otherwise:

name                     old time/op    new time/op     delta
BinaryTree17-4              2.64s ± 4%      2.56s ± 9%  -2.92%  (p=0.008 n=9+9)
Fannkuch11-4                2.90s ± 1%      2.95s ± 3%  +1.76%  (p=0.010 n=10+9)
FmtFprintfEmpty-4          35.3ns ± 1%     34.5ns ± 2%  -2.34%  (p=0.004 n=9+8)
FmtFprintfString-4         57.0ns ± 1%     58.4ns ± 5%  +2.52%  (p=0.029 n=9+10)
FmtFprintfInt-4            59.8ns ± 3%     59.8ns ± 6%    ~     (p=0.565 n=10+10)
FmtFprintfIntInt-4         93.9ns ± 3%     91.2ns ± 5%  -2.94%  (p=0.014 n=10+9)
FmtFprintfPrefixedInt-4     107ns ± 6%      104ns ± 6%    ~     (p=0.099 n=10+10)
FmtFprintfFloat-4           187ns ± 3%      188ns ± 3%    ~     (p=0.505 n=10+9)
FmtManyArgs-4               410ns ± 1%      415ns ± 6%    ~     (p=0.649 n=8+10)
GobDecode-4                5.30ms ± 3%     5.27ms ± 3%    ~     (p=0.436 n=10+10)
GobEncode-4                4.62ms ± 5%     4.47ms ± 2%  -3.24%  (p=0.001 n=9+10)
Gzip-4                      197ms ± 4%      193ms ± 3%    ~     (p=0.123 n=10+10)
Gunzip-4                   30.4ms ± 3%     30.1ms ± 3%    ~     (p=0.481 n=10+10)
HTTPClientServer-4         76.3µs ± 1%     76.0µs ± 1%    ~     (p=0.236 n=8+9)
JSONEncode-4               10.5ms ± 9%     10.3ms ± 3%    ~     (p=0.280 n=10+10)
JSONDecode-4               42.3ms ±10%     41.3ms ± 2%    ~     (p=0.053 n=9+10)
Mandelbrot200-4            3.80ms ± 2%     3.72ms ± 2%  -2.15%  (p=0.001 n=9+10)
GoParse-4                  2.88ms ±10%     2.81ms ± 2%    ~     (p=0.247 n=10+10)
RegexpMatchEasy0_32-4      69.5ns ± 4%     68.6ns ± 2%    ~     (p=0.171 n=10+10)
RegexpMatchEasy0_1K-4       165ns ± 3%      162ns ± 3%    ~     (p=0.137 n=10+10)
RegexpMatchEasy1_32-4      65.7ns ± 6%     64.4ns ± 2%  -2.02%  (p=0.037 n=10+10)
RegexpMatchEasy1_1K-4       278ns ± 2%      279ns ± 3%    ~     (p=0.991 n=8+9)
RegexpMatchMedium_32-4     99.3ns ± 3%     98.5ns ± 4%    ~     (p=0.457 n=10+9)
RegexpMatchMedium_1K-4     30.1µs ± 1%     30.4µs ± 2%    ~     (p=0.173 n=8+10)
RegexpMatchHard_32-4       1.40µs ± 2%     1.41µs ± 4%    ~     (p=0.565 n=10+10)
RegexpMatchHard_1K-4       42.5µs ± 1%     41.5µs ± 3%  -2.13%  (p=0.002 n=8+9)
Revcomp-4                   332ms ± 4%      328ms ± 5%    ~     (p=0.720 n=9+10)
Template-4                 48.3ms ± 2%     49.6ms ± 3%  +2.56%  (p=0.002 n=8+10)
TimeParse-4                 252ns ± 2%      249ns ± 3%    ~     (p=0.116 n=9+10)
TimeFormat-4                262ns ± 4%      252ns ± 3%  -4.01%  (p=0.000 n=9+10)

name                     old speed      new speed       delta
GobDecode-4               145MB/s ± 3%    146MB/s ± 3%    ~     (p=0.436 n=10+10)
GobEncode-4               166MB/s ± 5%    172MB/s ± 2%  +3.28%  (p=0.001 n=9+10)
Gzip-4                   98.6MB/s ± 4%  100.4MB/s ± 3%    ~     (p=0.123 n=10+10)
Gunzip-4                  639MB/s ± 3%    645MB/s ± 3%    ~     (p=0.481 n=10+10)
JSONEncode-4              185MB/s ± 8%    189MB/s ± 3%    ~     (p=0.280 n=10+10)
JSONDecode-4             46.0MB/s ± 9%   47.0MB/s ± 2%  +2.21%  (p=0.046 n=9+10)
GoParse-4                20.1MB/s ± 9%   20.6MB/s ± 2%    ~     (p=0.239 n=10+10)
RegexpMatchEasy0_32-4     460MB/s ± 4%    467MB/s ± 2%    ~     (p=0.165 n=10+10)
RegexpMatchEasy0_1K-4    6.19GB/s ± 3%   6.28GB/s ± 3%    ~     (p=0.165 n=10+10)
RegexpMatchEasy1_32-4     487MB/s ± 5%    497MB/s ± 2%  +2.00%  (p=0.043 n=10+10)
RegexpMatchEasy1_1K-4    3.67GB/s ± 2%   3.67GB/s ± 3%    ~     (p=0.963 n=8+9)
RegexpMatchMedium_32-4   10.1MB/s ± 3%   10.1MB/s ± 4%    ~     (p=0.435 n=10+9)
RegexpMatchMedium_1K-4   34.0MB/s ± 1%   33.7MB/s ± 2%    ~     (p=0.173 n=8+10)
RegexpMatchHard_32-4     22.9MB/s ± 2%   22.7MB/s ± 4%    ~     (p=0.565 n=10+10)
RegexpMatchHard_1K-4     24.0MB/s ± 3%   24.7MB/s ± 3%  +2.64%  (p=0.001 n=9+9)
Revcomp-4                 766MB/s ± 4%    775MB/s ± 5%    ~     (p=0.720 n=9+10)
Template-4               40.2MB/s ± 2%   39.2MB/s ± 3%  -2.47%  (p=0.002 n=8+10)

The rules match ~1800 times during all.bash.

Fixes #18943

Change-Id: I64be1ada34e89c486dfd935bf429b35652117ed4
Reviewed-on: https://go-review.googlesource.com/94766
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
 2018-03-24 02:38:50 +00:00
Giovanni Bajo
89ae7045f3 test: convert all math-related tests from asm_test 
Change-Id: If542f0b5c5754e6eb2f9b302fe5a148ba9a57338
Reviewed-on: https://go-review.googlesource.com/98443
Run-TryBot: Giovanni Bajo <rasky@develer.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>
 2018-03-04 16:52:33 +00:00