1
0
mirror of https://github.com/golang/go synced 2024-11-19 13:34:45 -07:00
go/src/math/log_s390x.s
Bill O'Farrell 88672de7af math: use SIMD to accelerate additional scalar math functions on s390x
As necessary, math functions were structured to use stubs, so that they can
be accelerated with assembly on any platform.

Technique used was minimax polynomial approximation using tables of
polynomial coefficients, with argument range reduction.

Benchmark         New     Old     Speedup
BenchmarkAcos     12.2    47.5    3.89
BenchmarkAcosh    18.5    56.2    3.04
BenchmarkAsin     13.1    40.6    3.10
BenchmarkAsinh    19.4    62.8    3.24
BenchmarkAtan     10.1    23      2.28
BenchmarkAtanh    19.1    53.2    2.79
BenchmarkAtan2    16.5    33.9    2.05
BenchmarkCbrt     14.8    58      3.92
BenchmarkErf      10.8    20.1    1.86
BenchmarkErfc     11.2    23.5    2.10
BenchmarkExp      8.77    53.8    6.13
BenchmarkExpm1    10.1    38.3    3.79
BenchmarkLog      13.1    40.1    3.06
BenchmarkLog1p    12.7    38.3    3.02
BenchmarkPowInt   31.7    40.5    1.28
BenchmarkPowFrac  33.1    141     4.26
BenchmarkTan      11.5    30      2.61

Accuracy was tested against a high precision
reference function to determine maximum error.
Note: ulperr is error in "units in the last place"

       max
      ulperr
Acos  1.15
Acosh 1.07
Asin  2.22
Asinh 1.72
Atan  1.41
Atanh 3.00
Atan2 1.45
Cbrt  1.18
Erf   1.29
Erfc  4.82
Exp   1.00
Expm1 2.26
Log   0.94
Log1p 2.39
Tan   3.14

Pow will have 99.99% correctly rounded results with reasonable inputs
producing numeric (non Inf or NaN) results

Change-Id: I850e8cf7b70426e8b54ec49d74acd4cddc8c6cb2
Reviewed-on: https://go-review.googlesource.com/38585
Reviewed-by: Michael Munday <munday@ca.ibm.com>
Run-TryBot: Michael Munday <munday@ca.ibm.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
2017-05-08 19:52:30 +00:00

181 lines
4.8 KiB
ArmAsm

// Copyright 2017 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.
#include "textflag.h"
// Minimax polynomial approximations
DATA ·logrodataL21<> + 0(SB)/8, $-.499999999999999778E+00
DATA ·logrodataL21<> + 8(SB)/8, $0.333333333333343751E+00
DATA ·logrodataL21<> + 16(SB)/8, $-.250000000001606881E+00
DATA ·logrodataL21<> + 24(SB)/8, $0.199999999971603032E+00
DATA ·logrodataL21<> + 32(SB)/8, $-.166666663114122038E+00
DATA ·logrodataL21<> + 40(SB)/8, $-.125002923782692399E+00
DATA ·logrodataL21<> + 48(SB)/8, $0.111142014580396256E+00
DATA ·logrodataL21<> + 56(SB)/8, $0.759438932618934220E-01
DATA ·logrodataL21<> + 64(SB)/8, $0.142857144267212549E+00
DATA ·logrodataL21<> + 72(SB)/8, $-.993038938793590759E-01
DATA ·logrodataL21<> + 80(SB)/8, $-1.0
GLOBL ·logrodataL21<> + 0(SB), RODATA, $88
// Constants
DATA ·logxminf<> + 0(SB)/8, $0xfff0000000000000
GLOBL ·logxminf<> + 0(SB), RODATA, $8
DATA ·logxnan<> + 0(SB)/8, $0x7ff8000000000000
GLOBL ·logxnan<> + 0(SB), RODATA, $8
DATA ·logx43f<> + 0(SB)/8, $0x43f0000000000000
GLOBL ·logx43f<> + 0(SB), RODATA, $8
DATA ·logxl2<> + 0(SB)/8, $0x3fda7aecbeba4e46
GLOBL ·logxl2<> + 0(SB), RODATA, $8
DATA ·logxl1<> + 0(SB)/8, $0x3ffacde700000000
GLOBL ·logxl1<> + 0(SB), RODATA, $8
/* Input transform scale and add constants */
DATA ·logxm<> + 0(SB)/8, $0x3fc77604e63c84b1
DATA ·logxm<> + 8(SB)/8, $0x40fb39456ab53250
DATA ·logxm<> + 16(SB)/8, $0x3fc9ee358b945f3f
DATA ·logxm<> + 24(SB)/8, $0x40fb39418bf3b137
DATA ·logxm<> + 32(SB)/8, $0x3fccfb2e1304f4b6
DATA ·logxm<> + 40(SB)/8, $0x40fb393d3eda3022
DATA ·logxm<> + 48(SB)/8, $0x3fd0000000000000
DATA ·logxm<> + 56(SB)/8, $0x40fb393969e70000
DATA ·logxm<> + 64(SB)/8, $0x3fd11117aafbfe04
DATA ·logxm<> + 72(SB)/8, $0x40fb3936eaefafcf
DATA ·logxm<> + 80(SB)/8, $0x3fd2492af5e658b2
DATA ·logxm<> + 88(SB)/8, $0x40fb39343ff01715
DATA ·logxm<> + 96(SB)/8, $0x3fd3b50c622a43dd
DATA ·logxm<> + 104(SB)/8, $0x40fb39315adae2f3
DATA ·logxm<> + 112(SB)/8, $0x3fd56bbeea918777
DATA ·logxm<> + 120(SB)/8, $0x40fb392e21698552
GLOBL ·logxm<> + 0(SB), RODATA, $128
// Log returns the natural logarithm of the argument.
//
// Special cases are:
// Log(+Inf) = +Inf
// Log(0) = -Inf
// Log(x < 0) = NaN
// Log(NaN) = NaN
// The algorithm used is minimax polynomial approximation using a table of
// polynomial coefficients determined with a Remez exchange algorithm.
TEXT ·logAsm(SB), NOSPLIT, $0-16
FMOVD x+0(FP), F0
MOVD $·logrodataL21<>+0(SB), R9
MOVH $0x8006, R4
WORD $0xB3CD0010 //lgdr %r1,%f0
MOVD $0x3FF0000000000000, R6
SRAD $48, R1, R1
MOVD $0x40F03E8000000000, R8
SUBW R1, R4
WORD $0xEC2420BB //risbg %r2,%r4,32,128+59,0
BYTE $0x00
BYTE $0x55
WORD $0xEC62000F //risbgn %r6,%r2,64-64+0,64-64+0+16-1,64-0-16
BYTE $0x30
BYTE $0x59
WORD $0xEC82101F //risbgn %r8,%r2,64-64+16,64-64+16+16-1,64-16-16
BYTE $0x20
BYTE $0x59
MOVW R1, R7
CMPBGT R7, $22, L17
WORD $0xB3120000 //ltdbr %f0,%f0
MOVD $·logx43f<>+0(SB), R1
FMOVD 0(R1), F2
BLEU L3
MOVH $0x8005, R12
MOVH $0x8405, R0
BR L15
L7:
WORD $0xB3120000 //ltdbr %f0,%f0
BLEU L3
L15:
FMUL F2, F0
WORD $0xB3CD0010 //lgdr %r1,%f0
SRAD $48, R1, R1
SUBW R1, R0, R2
SUBW R1, R12, R3
BYTE $0x18 //lr %r4,%r2
BYTE $0x42
ANDW $0xFFFFFFF0, R3
ANDW $0xFFFFFFF0, R2
BYTE $0x18 //lr %r5,%r1
BYTE $0x51
MOVW R1, R7
CMPBLE R7, $22, L7
WORD $0xEC63000F //risbgn %r6,%r3,64-64+0,64-64+0+16-1,64-0-16
BYTE $0x30
BYTE $0x59
WORD $0xEC82101F //risbgn %r8,%r2,64-64+16,64-64+16+16-1,64-16-16
BYTE $0x20
BYTE $0x59
L2:
MOVH R5, R5
MOVH $0x7FEF, R1
CMPW R5, R1
BGT L1
WORD $0xB3C10026 //ldgr %f2,%r6
FMUL F2, F0
WORD $0xEC4439BB //risbg %r4,%r4,57,128+59,3
BYTE $0x03
BYTE $0x55
FMOVD 80(R9), F2
MOVD $·logxm<>+0(SB), R7
ADD R7, R4
FMOVD 72(R9), F4
WORD $0xED004000 //madb %f2,%f0,0(%r4)
BYTE $0x20
BYTE $0x1E
FMOVD 64(R9), F1
FMOVD F2, F0
FMOVD 56(R9), F2
WFMADB V0, V2, V4, V2
WFMDB V0, V0, V6
FMOVD 48(R9), F4
WFMADB V0, V2, V4, V2
FMOVD 40(R9), F4
WFMADB V2, V6, V1, V2
FMOVD 32(R9), F1
WFMADB V6, V4, V1, V4
FMOVD 24(R9), F1
WFMADB V6, V2, V1, V2
FMOVD 16(R9), F1
WFMADB V6, V4, V1, V4
MOVD $·logxl1<>+0(SB), R1
FMOVD 8(R9), F1
WFMADB V6, V2, V1, V2
FMOVD 0(R9), F1
WFMADB V6, V4, V1, V4
FMOVD 8(R4), F1
WFMADB V0, V2, V4, V2
WORD $0xB3C10048 //ldgr %f4,%r8
WFMADB V6, V2, V0, V2
WORD $0xED401000 //msdb %f1,%f4,0(%r1)
BYTE $0x10
BYTE $0x1F
MOVD ·logxl2<>+0(SB), R1
WORD $0xB3130001 //lcdbr %f0,%f1
WORD $0xB3C10041 //ldgr %f4,%r1
WFMADB V0, V4, V2, V0
L1:
FMOVD F0, ret+8(FP)
RET
L3:
WORD $0xB3120000 //ltdbr %f0,%f0
BEQ L20
BGE L1
BVS L1
MOVD $·logxnan<>+0(SB), R1
FMOVD 0(R1), F0
BR L1
L20:
MOVD $·logxminf<>+0(SB), R1
FMOVD 0(R1), F0
FMOVD F0, ret+8(FP)
RET
L17:
BYTE $0x18 //lr %r5,%r1
BYTE $0x51
BR L2