Support & downloads

Mathematical Acceleration Subsystem
Features and benefits
System requirements
Support

Software > Rational >

Performance information for the MASS libraries for AIX

Product documentation

Abstract

This document provides performance data for the MASS scalar and vector libraries for AIX, in comparison with the standard system math libraries.

Content

Performance information for the MASS libraries for AIX

The following tables provide approximate performance data for the MASS scalar and vector libraries running under AIX on various POWER machines.

The columns labelled libm and mass list the results obtained with the libm.a system library and the libmass.a library, respectively. This data was obtained by timing many repetitions of a loop over 1000 random arguments and includes all overheads. Timing in this way brings the input and output vectors into the on-chip cache (because the loop is short enough for the vectors to fit in the cache). Performance may deteriorate significantly when the input and output vectors are not in the cache. Performance may also deteriorate for arguments at or near the end-points of the valid argument ranges.

The columns labeled massv, vp4, vp5, and vp6 list the results obtained with the libmassv.a, libmassvp4.a, libmassvp5.a, and libmassvp6.a libraries, respectively. They give estimates of the number of cycles per evaluation of a vector element. The estimates used vectors of length 1000 so that the caches contain all the vectors.

Although the vector names (e.g. vacos) are used in the Function column, the libm and mass columns refer to the corresponding scalar function (e.g. acos).

The system library measurements were made with the versions of the library available on the test systems. They may vary from the versions timed for previous versions of MASS. Users may experience performance that differs from that found in these tables.

Results will vary with vector length. Entries in the table where the library function does not exist, or the measurement was not done, are left blank.

The range key at the end of this section applies to all the tables.

POWER6 Performance

POWER6 libmassvp6.a Performance -- double precision functions (cycles per evaluation, length 1000 loop)
libm/ libm/ Function Range libm mass vp6 mass vp6 ================================================== vacos B 207.45 209.89 27.79 0.99 7.46 vacosh G 909.74 315.96 46.82 2.88 19.43 vasin B 190.48 231.28 27.72 0.82 6.87 vasinh D 707.13 316.86 30.28 2.23 23.35 vatan B 199.42 122.24 1.63 vatan D 209.35 163.78 1.28 vatan2 D 1402.90 174.20 37.15 8.05 37.76 vatanh B 515.45 249.08 26.48 2.07 19.47 vcbrt D 465.86 191.78 16.57 2.43 28.11 vcopysign D 81.70 103.11 0.79 vcos B 147.89 93.09 15.87 1.59 9.32 vcos D 201.36 160.31 16.94 1.26 11.89 vcosh D 451.98 148.04 14.54 3.05 31.09 vcosisin B 314.20 153.68 57.21 2.04 5.49 vcosisin D 427.27 219.00 110.59 1.95 3.86 vdint D 15.68 vdiv D 7.88 vdnint D 333.75 106.77 16.70 3.13 19.99 verf C 174.88 129.52 1.35 verfc C 238.96 189.26 1.26 vexp D 308.19 110.07 11.74 2.80 26.25 vexpm1 D 319.71 123.91 13.46 2.58 23.75 vhypot D 791.58 231.96 3.41 vlgamma H 667.95 387.43 1.72 vlog C 362.77 163.75 12.34 2.22 29.40 vlog10 C 468.10 142.26 12.34 3.29 37.93 vlog1p H 346.44 168.21 18.47 2.06 18.76 vpow C 739.87 297.85 36.44 2.48 20.30 vqdrt C 10.26 vrcbrt D 16.82 vrec D 7.85 vrqdrt C 10.89 vrsqrt C 214.11 148.20 11.17 1.44 19.17 vsin B 149.26 93.14 15.95 1.60 9.36 vsin D 206.34 160.70 16.86 1.28 12.24 vsincos B 276.75 133.54 103.31 2.07 2.68 vsincos D 400.12 169.55 110.48 2.36 3.62 vsinh D 552.82 126.78 15.58 4.36 35.48 vsqrt C 176.05 146.32 11.14 1.20 15.80 vtan D 358.63 133.32 61.13 2.69 5.87 vtanh F 584.75 148.94 22.43 3.93 26.07

*libm routine uses hardware instruction

POWER6 libmassvp6.a Performance -- single precision functions (cycles per evaluation, length 1000 loop)
libm/ libm/ Function Range libm mass vp6 mass vp6 ================================================== vsacos B 236.16 190.27 23.49 1.24 10.05 vsacosh G 936.49 289.47 39.96 3.24 23.44 vsasin B 218.83 191.08 22.49 1.15 9.73 vsasinh D 720.21 302.50 25.79 2.38 27.93 vsatan B 229.88 131.22 1.75 vsatan D 243.57 204.41 1.19 vsatan2 D 1431.90 218.79 22.43 6.54 63.84 vsatanh B 526.16 227.70 24.66 2.31 21.34 vscbrt D 481.13 179.58 12.35 2.68 38.96 vscopysign D 109.55 113.27 0.97 vscos B 171.00 113.53 14.23 1.51 12.02 vscos D 230.56 177.10 14.74 1.30 15.64 vscosh D 579.94 161.61 15.58 3.59 37.22 vscosisin B 54.42 vscosisin D 104.33 vsdiv D 7.83 vserf C 200.58 105.43 1.90 vserfc C 278.30 125.64 2.22 vsexp D 388.71 127.43 10.78 3.05 36.06 vsexpm1 D 416.46 143.63 12.32 2.90 33.80 vshypot D 934.26 206.52 4.52 vslgamma H 703.24 358.59 1.96 vslog C 400.68 148.25 10.75 2.70 37.27 vslog10 C 497.09 142.86 10.69 3.48 46.50 vslog1p H 389.21 143.08 13.49 2.72 28.85 vspow C 1043.39 230.82 23.55 4.52 44.31 vsqdrt C 10.15 vsrcbrt D 11.37 vsrec D 5.61 vsrint D 211.34 97.18 2.17 vsrqdrt C 9.94 vsrsqrt C 10.09 vssin B 168.95 118.95 13.64 1.42 12.39 vssin D 232.15 179.46 14.60 1.29 15.90 vssincos B 54.67 vssincos D 104.31 vssinh D 679.49 170.26 16.82 3.99 40.40 vssqrt C 8.85 vstan D 353.55 195.37 58.09 1.81 6.09 vstanh F 581.74 153.27 21.28 3.80 27.34
POWER5 Performance

POWER5 libmassvp5.a Performance -- double precision functions (cycles per evaluation, length 1000 loop)
libm/ libm/ Function Range libm mass vp5 mass vp5 ================================================== vacos B 122.40 107.72 22.95 1.14 5.33 vacosh G 591.30 175.51 30.66 3.37 19.29 vasin B 121.08 151.69 22.94 0.80 5.28 vasinh D 487.30 180.82 21.74 2.69 22.41 vatan B 70.78 49.48 1.43 vatan D 118.35 78.45 1.51 vatan2 D 787.62 94.98 24.56 8.29 32.07 vatanh B 310.43 156.44 18.78 1.98 16.53 vcbrt D 330.95 100.39 11.43 3.30 28.95 vcopysign D 37.96 42.07 0.90 vcos B 68.22 33.08 13.15 2.06 5.19 vcos D 87.14 69.09 14.66 1.26 5.94 vcosh D 282.10 56.25 12.72 5.02 22.18 vcosisin B 157.57 71.85 9.30 2.19 16.94 vcosisin D 200.15 129.25 17.79 1.55 11.25 vdint D 6.36 vdiv D 5.38 vdnint D 118.36 48.09 7.72 2.46 15.33 verf C 82.31 57.42 1.43 verfc C 111.44 111.71 1.00 vexp D 172.40 46.87 10.61 3.68 16.25 vexpm1 D 238.04 49.44 11.47 4.81 20.75 vhypot D 643.93 108.88 5.91 vlgamma H 391.16 236.46 1.65 vlog C 215.88 86.74 9.36 2.49 23.06 vlog10 C 325.08 79.44 9.44 4.09 34.44 vlog1p H 220.76 71.64 12.09 3.08 18.26 vpow C 463.26 149.60 29.58 3.10 15.66 vqdrt C 7.56 vrcbrt D 11.32 vrec D 5.00 vrqdrt C 6.84 vrsqrt C 182.82 74.92 6.31 2.44 28.97 vsin B 69.99 34.57 13.19 2.02 5.31 vsin D 88.63 69.09 14.68 1.28 6.04 vsincos B 131.02 46.69 15.66 2.81 8.37 vsincos D 171.33 92.53 17.78 1.85 9.64 vsinh D 357.66 63.43 13.18 5.64 27.14 vsqrt C 139.37 75.50 6.27 1.85 22.23 vtan D 222.71 73.50 17.42 3.03 12.78

*libm routine uses hardware instruction

POWER5 libmassvp5.a Performance -- single precision functions (cycles per evaluation, length 1000 loop)
libm/ libm/ Function Range libm mass vp5 mass vp5 ================================================== vsacos B 143.39 88.92 13.96 1.61 10.27 vsacosh G 610.35 160.14 22.59 3.81 27.02 vsasin B 136.26 82.44 13.75 1.65 9.91 vsasinh D 496.50 172.10 17.31 2.88 28.68 vsatan B 81.85 48.10 1.70 vsatan D 131.21 98.11 1.34 vsatan2 D 797.35 147.51 15.50 5.41 51.44 vsatanh B 320.59 109.71 16.57 2.92 19.35 vscbrt D 340.06 89.94 7.80 3.78 43.60 vscopysign D 52.84 87.04 0.61 vscos B 80.61 34.02 10.33 2.37 7.80 vscos D 111.43 71.87 11.89 1.55 9.37 vscosh D 323.33 68.11 12.40 4.75 26.08 vscosisin B 7.25 vscosisin D 14.54 vsdiv D 4.68 vserf C 106.99 38.87 2.75 vserfc C 119.67 64.13 1.87 vsexp D 228.64 58.80 10.18 3.89 22.46 vsexpm1 D 281.13 72.35 10.86 3.89 25.89 vshypot D 697.12 107.04 6.51 vslgamma H 431.48 246.31 1.75 vslog C 213.10 57.08 6.83 3.73 31.20 vslog10 C 345.52 58.30 6.83 5.93 50.59 vslog1p H 234.36 70.87 8.81 3.31 26.60 vspow C 542.71 128.08 19.99 4.24 27.15 vsqdrt C 6.86 vsrcbrt D 7.71 vsrec D 3.71 vsrint D 50.09 64.42 0.78 vsrqdrt C 6.18 vsrsqrt C 5.30 vssin B 82.56 34.51 10.35 2.39 7.98 vssin D 116.39 69.77 11.84 1.67 9.83 vssincos B 7.24 vssincos D 14.56 vssinh D 423.50 68.59 13.06 6.17 32.43 vssqrt C 5.46 vstan D 218.19 83.15 15.39 2.62 14.18 vstanh F 362.76 67.85 15.10 5.35 24.02 vtanh F 363.40 77.29 18.06 4.70 20.12

POWER4+ Performance

POWER4+ libmassvp4.a Performance -- double precision functions (cycles per evaluation, length 1000 loop)
libm/ libm/ Function Range libm mass vp4 mass vp4 ================================================== vacos B 115.06 101.62 24.09 1.13 4.78 vacosh G 536.48 175.32 32.30 3.06 16.61 vasin B 109.83 143.17 23.98 0.77 4.58 vasinh D 462.63 182.09 23.39 2.54 19.78 vatan B 65.07 60.10 1.08 vatan D 100.50 89.50 1.12 vatan2 D 715.44 104.95 24.11 6.82 29.67 vatanh B 294.31 133.58 20.42 2.20 14.41 vcbrt D 333.19 102.93 10.63 3.24 31.34 vcopysign D 37.09 36.33 1.02 vcos B 65.48 36.61 13.60 1.79 4.81 vcos D 91.77 66.66 14.89 1.38 6.16 vcosh D 272.80 55.58 13.93 4.91 19.58 vcosisin B 159.73 71.67 10.21 2.23 15.64 vcosisin D 209.38 144.78 17.62 1.45 11.88 vdint D 6.58 vdiv D 5.79 vdnint D 112.80 42.19 8.81 2.67 12.80 verf C 78.75 62.42 1.26 verfc C 105.09 108.70 0.97 vexp D 162.75 46.41 10.81 3.51 15.06 vexpm1 D 216.95 51.41 11.95 4.22 18.15 vhypot D 550.15 101.81 5.40 vlgamma H 368.00 209.47 1.76 vlog C 203.36 85.63 9.85 2.37 20.65 vlog10 C 284.48 82.07 10.15 3.47 28.03 vlog1p H 201.22 78.70 12.66 2.56 15.89 vpow C 449.94 146.70 29.78 3.07 15.11 vqdrt C 7.85 vrcbrt D 10.45 vrec D 5.13 vrqdrt C 7.66 vrsqrt C 163.48 74.76 6.95 2.19 23.52 vsin B 62.60 37.30 13.61 1.68 4.60 vsin D 91.27 66.58 14.84 1.37 6.15 vsincos B 119.41 49.25 16.24 2.42 7.35 vsincos D 175.32 88.17 17.53 1.99 10.00 vsinh D 349.25 58.65 14.76 5.95 23.66 vsqrt C 124.84 74.12 7.12 1.68 17.53 vtan D 200.03 73.33 18.59 2.73 10.76 vtanh F 380.90 72.80 18.57 5.23 20.51

*libm routine uses hardware instruction

POWER4+ libmassvp4.a Performance -- single precision functions (cycles per evaluation, length 1000 loop)
libm/ libm/ Function Range libm mass vp4 mass vp4 ================================================== vsacos B 130.37 79.18 14.19 1.65 9.19 vsacosh G 564.96 164.27 24.25 3.44 23.30 vsasin B 123.76 94.03 14.16 1.32 8.74 vsasinh D 482.55 167.74 18.69 2.88 25.82 vsatan B 71.95 41.49 1.73 vsatan D 111.71 89.64 1.25 vsatan2 D 734.10 129.02 14.51 5.69 50.59 vsatanh B 323.16 107.91 17.50 2.99 18.47 vscbrt D 359.67 80.08 7.58 4.49 47.45 vscopysign D 45.04 79.13 0.57 vscos B 67.81 29.57 10.88 2.29 6.23 vscos D 91.88 67.73 12.28 1.36 7.48 vscosh D 364.40 64.55 13.34 5.65 27.32 vscosisin B 8.20 vscosisin D 14.56 vsdiv D 5.24 vserf C 91.93 33.42 2.75 vserfc C 114.37 61.17 1.87 vsexp D 212.16 57.98 10.36 3.66 20.48 vsexpm1 D 278.57 65.83 10.97 4.23 25.39 vshypot D 709.54 101.54 6.99 vslgamma H 497.11 185.17 2.68 vslog C 210.54 55.16 6.73 3.82 31.28 vslog10 C 289.76 54.95 6.81 5.27 42.55 vslog1p H 206.38 71.02 9.27 2.91 22.26 vspow C 554.52 122.42 19.14 4.53 28.97 vsqdrt C 7.35 vsrcbrt D 7.33 vsrec D 4.77 vsrint D 56.16 39.69 1.41 vsrqdrt C 6.76 vsrsqrt C 5.58 vssin B 66.98 35.75 10.92 1.87 6.13 vssin D 92.40 67.51 12.26 1.37 7.54 vssincos B 8.18 vssincos D 14.55 vssinh D 431.54 69.83 14.22 6.18 30.35 vssqrt C 5.70 vstan D 197.23 76.56 14.54 2.58 13.56 vstanh F 378.52 61.20 16.74 6.18 22.61
PowerPC 604e Performance

PowerPC 604e libmassv.a Performance -- double precision functions (cycles per evaluation, length 1000 loop)
libm/ libm/ Function Range libm mass massv mass massv ================================================== vacos B 103.14 111.17 40.58 0.93 2.54 vacosh G 350.27 135.58 46.76 2.58 7.49 vasin B 95.86 123.42 40.46 0.78 2.37 vasinh D 364.40 154.43 44.25 2.36 8.24 vatan B 68.89 64.03 1.08 vatan D 106.62 94.78 1.12 vatan2 D 597.78 103.53 40.33 5.77 14.82 vatanh B 251.65 140.03 41.23 1.80 6.10 vcbrt D 295.03 90.84 19.73 3.25 14.95 vcopysign D 33.03 40.03 0.83 vcos B 67.61 41.32 9.28 1.64 7.29 vcos D 89.70 59.95 26.64 1.50 3.37 vcosh D 201.22 58.67 24.93 3.43 8.07 vcosisin B 78.04 19.00 vcosisin D 109.94 28.75 vdint D 6.26 vdiv D 11.46 vdnint D 44.02 6.77 verf C 72.04 52.09 1.38 verfc C 53.70 93.58 0.57 vexp D 129.80 50.09 16.06 2.59 8.08 vexpm1 D 145.63 53.37 19.74 2.73 7.38 vhypot D 366.89 98.45 3.73 vlgamma H 301.86 194.67 1.55 vlog C 145.65 66.72 19.29 2.18 7.55 vlog10 C 212.18 77.03 19.43 2.75 10.92 vlog1p H 169.01 72.04 25.81 2.35 6.55 vpow C 336.49 124.39 50.96 2.71 6.60 vqdrt C 14.81 vrcbrt D 19.44 vrec D 10.03 vrqdrt C 13.94 vrsqrt C 114.92 63.80 16.03 1.80 7.17 vsin B 66.01 41.32 10.48 1.60 6.30 vsin D 92.43 59.95 26.60 1.54 3.47 vsincos B 128.97 66.04 19.37 1.95 6.66 vsincos D 177.44 94.67 29.19 1.87 6.08 vsinh D 260.16 62.91 26.55 4.14 9.80 vsqrt C 80.94 61.82 16.01 1.31 5.06 vtan D 205.24 87.72 31.34 2.34 6.55 vtanh F 308.29 79.59 34.38 3.87 8.97

*libm routine uses hardware instruction

PowerPC 604e libmassv.a Performance -- single precision functions (cycles per evaluation, length 1000 loop)
libm/ libm/ Function Range libm mass massv mass massv ================================================== vsacos B 154.65 110.51 26.72 1.40 5.79 vsacosh G 151.88 37.53 vsasin B 145.87 110.00 27.02 1.33 5.40 vsasinh D 165.78 36.68 vsatan B 120.23 84.02 1.43 vsatan D 156.87 127.21 1.23 vsatan2 D 655.98 157.48 39.96 4.17 16.42 vsatanh B 142.04 35.51 vscbrt D 101.72 14.10 vscopysign D 71.53 vscos B 111.62 66.02 7.24 1.69 15.42 vscos D 133.71 98.39 20.11 1.36 6.65 vscosh D 252.05 90.71 21.06 2.78 11.97 vscosisin B 15.00 vscosisin D 23.12 vsdiv D 9.83 vserf C 67.88 vserfc C 81.33 vsexp D 188.01 82.18 13.15 2.29 14.30 vsexpm1 D 82.80 16.26 vshypot D 123.85 vslgamma H 208.64 vslog C 190.66 89.03 15.76 2.14 12.10 vslog10 C 257.18 89.03 15.71 2.89 16.37 vslog1p H 99.03 20.28 vspow C 399.93 129.85 29.93 3.08 13.36 vsqdrt C 13.20 vsrcbrt D 14.10 vsrec D 7.83 vsrint D 48.19 vsrqdrt C 12.44 vsrsqrt C 9.08 vssin B 110.01 68.02 8.37 1.62 13.14 vssin D 136.43 92.56 20.12 1.47 6.78 vssincos B 14.75 vssincos D 23.62 vssinh D 325.72 86.63 23.77 3.76 13.70 vssqrt C 9.67 vstan D 207.83 123.59 25.89 1.68 8.03 vstanh F 311.24 84.75 28.02 3.67 11.11
Range Key

A 0,1
B -1,1
C 0,100
D -100,100
E -10,10
F -20,20
G 1,100
H -1,100
I 0,10

Original publication date

2005/12/5

Cross Reference information

Segment	Product	Component	Platform	Version
pSeries Unix Servers (incl pSeries Intellistation)	pSeries - Enterprise Servers
RS/6000 (Servers and Workstations)	RS/6000 Enterprise Servers
Operating Systems	AIX family		AIX	All Versions
Software Development	VisualAge C++		AIX	All versions
Software Development	XL Fortran		AIX, AIXL	All versions
Software Development	XL C Enterprise Edition for AIX	Not Applicable	AIX	All Versions
Software Development	XL C/C++		AIX, AIX5L, AIXL	All Versions

IBM, the IBM logo and ibm.com are trademarks of International Business Machines Corp., registered in many jurisdictions worldwide. Other product and service names might be trademarks of IBM or other companies. A current list of IBM trademarks is available on the Web at "Copyright and trademark information" at www.ibm.com/legal/copytrade.shtml.

Document information


Product categories:

	Software

	Software Development

	Traditional Progamming Language & Compilers

	Mathematical Acceleration Subsystem

	Libraries


Operating system(s):
	AIX, AIX5L, AIXL

Software version:
	4.4

Reference #:
	7005374

IBM Group:
	Software Group

Modified date:
	2007-06-25

Translate My Page

Rate this page

Help us improve this page. Your response will be used to improve our document content. Requests for assistance, if applicable, should be submitted through your normal support channel as we cannot respond from this site.