[PATCH] Tunable guile@3.0.9t package

Hi,

The attached patch adds a guile@3.0.9t package that enables CPU

architecture tuning for Guile and for a specialized libgc.

It creates new packages to avoid causing rebuilds.

The results of a benchmark are still not fully conclusive: regular Guile

seems to be slightly slower in the Geometric mean of all tests, but

there are benchmarks where a tuned Guile is slower.

Note that this is tuning for znver3 while I actually have znver4, but

support for the correct compile-target will only arrive with GCC 13.

To replicate the benchmarks:

echo guix pull --disable-authentication --allow-downgrades -u … # get the new packages

guix shell mercurial -- hg clone https://hg.sr.ht/~arnebab/wisp

guix shell git -- git clone https://github.com/ecraven/r7rs-benchmarks

cd r7rs-benchmarks

for N in {1..3}; do

guix shell --tune guile@3.0.9t -- ./bench guile all && mv results.Guile{,-tuned${N}}

guix shell guile@3.0.9 -- ./bench guile all && mv results.Guile{,-regular${N}}

sed -i 's/\+guile/+guiletuned/' results.Guile-tuned${N}

sed -i 's/\+guile/+guileregular/' results.Guile-regular${N}

done

grep -a -h '+!CSVLINE' results.Guile-{regular,tuned}* | sed 's/+!CSVLINE!+//' > all.csv

guix shell guile-wisp guile -- ../wisp/examples/evaluate-r7rs-benchmark.w ~/eigenes/Programme/r7rs-benchmarks/all.csv guiletuned

guix shell guile-wisp guile -- ../wisp/examples/evaluate-r7rs-benchmark.w ~/eigenes/Programme/r7rs-benchmarks/all.csv guileregular

My results with N in {1..2}, the Geometric mean:

Guile tuned has a slowdown of 1.03512801239189

Guile regular has a slowdown of 1.0504373210343294

This suggests that the regular Guile is 1.48% slower than tuned Guile.

Since I have no errorbars, I cannot say whether this is significant.

=== Guiletuned slowdown ===

(("bv2string:1000:1000:100" . 1.0)

("equal" . #f)

("mperm:20:10:2:1" . 1.146669471586088)

("gcbench:20:1" . 1.0197445340007967)

("sboyer:5:1" . 1.0398502157522214)

("nboyer:5:1" . 1.050828554948064)

("chudnovsky:50:500:50:1000"

.

1.0236854348323314)

("slatex:500" . 1.0063944149759714)

("scheme:100000" . 1.0689169431069399)

("quicksort:10000:2500" . 1.003088477529797)

("primes:1000:10000" . 1.041839283639136)

("peval:2000" . 1.0560148767561295)

("parsing:2500" . 1.0003564615141731)

("paraffins:23:10" . 2.0039145601078086)

("nqueens:13:10" . 1.0)

("mazefun:11:11:10000" . 1.034627134692567)

("maze:20:7:10000" . 1.001234149986604)

("matrix:5:5:2500" . 1.047237971262676)

("lattice:44:10" . 1.003284878892407)

("graphs:7:3" . 1.101724767552233)

("earley:1" . 1.0507663201957202)

("dynamic:500" . 1.0806476338778386)

("conform:500" . 1.0)

("compiler:2000" . 1.0973709486022507)

("read1:2500" . 1.0)

("wc:inputs/bib:50" . 1.0252716381981584)

("tail:50" . 1.070013760352806)

("cat:50" . 1.1006490143447183)

("sum1:25" . 1.0)

("string:500000:100" . 1.005448377967908)

("array1:1000000:500" . 1.017108880809611)

("ack:3:12:2" . 1.0343152722188402)

("simplex:1000000" . 1.0213054034897673)

("ray:50" . 1.0090968950973398)

("pnpoly:1000000" . 1.0091804774033606)

("pi:50:500:50:100" . 1.0)

("nucleic:50" . 1.0173803271423734)

("mbrotZ:75:1000" . 1.022665128035874)

("mbrot:75:1000" . 1.0)

("fft:65536:100" . 1.0)

("sumfp:1000000.0:500" . 1.0)

("sum:10000:200000" . 1.0028978038917624)

("fibfp:35.0:10" . 1.029590065926432)

("fibc:30:10" . 1.0072040559671238)

("fib:40:5" . 1.0023503962392322)

("ctak:27:16:8:1" . 1.0)

("cpstak:40:20:11:1" . 1.0)

("ntakl:40:20:12:1" . 1.046107228813898)

("takl:40:20:12:1" . 1.0327425092626041)

("tak:40:20:11:1" . 1.0)

("triangl:22:1:50" . 1.0143942222863194)

("puzzle:1000" . 1.2229189054254013)

("divrec:1000:1000000" . 1.0074179348861916)

("diviter:1000:1000000" . 1.043931078583493)

("destruc:600:50:4000" . 1.0240507947888393)

("deriv:10000000" . 1.0319784815172544)

("browse:2000" . 1.1070660842252487)

("bv2string:1000:1000:100" . 1.0338536673222782)

("equal" . #f)

("mperm:20:10:2:1" . 1.1881732405136267)

("gcbench:20:1" . 1.0)

("sboyer:5:1" . 1.0631408998092902)

("nboyer:5:1" . 1.0217971951625024)

("chudnovsky:50:500:50:1000" . 1.0)

("slatex:500" . 1.0)

("scheme:100000" . 1.0)

("quicksort:10000:2500" . 1.0543263325575605)

("primes:1000:10000" . 1.0067412573033894)

("peval:2000" . 1.0)

("parsing:2500" . 1.0)

("paraffins:23:10" . 1.0)

("nqueens:13:10" . 1.0176605905015021)

("mazefun:11:11:10000" . 1.037746391910457)

("maze:20:7:10000" . 1.0)

("matrix:5:5:2500" . 1.0200605013910409)

("lattice:44:10" . 1.0102542438570286)

("graphs:7:3" . 1.0384572473968816)

("earley:1" . 1.0)

("dynamic:500" . 1.0)

("conform:500" . 1.0091118513899833)

("compiler:2000" . 1.0)

("read1:2500" . 1.0599058441532945)

("wc:inputs/bib:50" . 1.0746142263219225)

("tail:50" . 1.0632740266035832)

("cat:50" . 1.0929665813153533)

("sum1:25" . 1.036061672363642)

("string:500000:100" . 1.0162155078942512)

("array1:1000000:500" . 1.0515376902103308)

("ack:3:12:2" . 1.0118706763775067)

("simplex:1000000" . 1.0164588830600216)

("ray:50" . 1.0)

("pnpoly:1000000" . 1.0)

("pi:50:500:50:100" . 1.045459192674658)

("nucleic:50" . 1.0017484672084827)

("mbrotZ:75:1000" . 1.0)

("mbrot:75:1000" . 1.0311423294990856)

("fft:65536:100" . 1.0504466238403414)

("sumfp:1000000.0:500" . 1.0091469248322233)

("sum:10000:200000" . 1.0)

("fibfp:35.0:10" . 1.0)

("fibc:30:10" . 1.1257119803649067)

("fib:40:5" . 1.0)

("ctak:27:16:8:1" . 1.0148864364373957)

("cpstak:40:20:11:1" . 1.011283388260497)

("ntakl:40:20:12:1" . 1.0)

("takl:40:20:12:1" . 1.007232975497366)

("tak:40:20:11:1" . 1.0362220583744437)

("triangl:22:1:50" . 1.022212297184504)

("puzzle:1000" . 1.2350889918747268)

("divrec:1000:1000000" . 1.0064081034755894)

("diviter:1000:1000000" . 1.0)

("destruc:600:50:4000" . 1.0)

("deriv:10000000" . 1.0061506030191765)

("browse:2000" . 1.0065279614595284))

=== Guiletuned Geometric Mean slowdown (successful tests / total tests) ===

1.03512801239189 (112 / 114)

=== Guileregular slowdown ===

(("bv2string:1000:1000:100" . 1.0674803047626884)

("equal" . #f)

("mperm:20:10:2:1" . 1.094739035540671)

("gcbench:20:1" . 1.0172414374269374)

("sboyer:5:1" . 1.0)

("nboyer:5:1" . 1.0)

("chudnovsky:50:500:50:1000"

.

1.1734845659670659)

("slatex:500" . 1.0239374916619306)

("scheme:100000" . 1.122700507430036)

("quicksort:10000:2500" . 1.0)

("primes:1000:10000" . 1.0)

("peval:2000" . 1.0088858652748092)

("parsing:2500" . 1.0418903162390507)

("paraffins:23:10" . 2.042048452193248)

("nqueens:13:10" . 1.0100153261373226)

("mazefun:11:11:10000" . 1.0)

("maze:20:7:10000" . 1.048959091743095)

("matrix:5:5:2500" . 1.0)

("lattice:44:10" . 1.0079437861216478)

("graphs:7:3" . 1.198135373192713)

("earley:1" . 1.0959681146976517)

("dynamic:500" . 1.16155735839947)

("conform:500" . 1.0119217370442324)

("compiler:2000" . 1.0306165278415973)

("read1:2500" . 1.0460552396933456)

("wc:inputs/bib:50" . 1.0)

("tail:50" . 1.0319054479289398)

("cat:50" . 1.0)

("sum1:25" . 1.034137468078866)

("string:500000:100" . 1.0)

("array1:1000000:500" . 1.0)

("ack:3:12:2" . 1.0057561961050057)

("simplex:1000000" . 1.0)

("ray:50" . 1.0438647708135202)

("pnpoly:1000000" . 1.0874091135942872)

("pi:50:500:50:100" . 1.0215934390378087)

("nucleic:50" . 1.0)

("mbrotZ:75:1000" . 1.052938347579755)

("mbrot:75:1000" . 1.0329378591712912)

("fft:65536:100" . 1.0189087888402792)

("sumfp:1000000.0:500" . 1.0482534106366277)

("sum:10000:200000" . 1.0289912283371638)

("fibfp:35.0:10" . 1.033671691949834)

("fibc:30:10" . 1.178850020182939)

("fib:40:5" . 1.024466514102094)

("ctak:27:16:8:1" . 1.1039955311250749)

("cpstak:40:20:11:1" . 1.0371479505513828)

("ntakl:40:20:12:1" . 1.0213756257162367)

("takl:40:20:12:1" . 1.0)

("tak:40:20:11:1" . 1.0126417896363957)

("triangl:22:1:50" . 1.0)

("puzzle:1000" . 1.0)

("divrec:1000:1000000" . 1.0450154469330961)

("diviter:1000:1000000" . 1.0591207445325488)

("destruc:600:50:4000" . 1.0838399255052662)

("deriv:10000000" . 1.0)

("browse:2000" . 1.018717561314545)

("bv2string:1000:1000:100" . 1.036089236974368)

("equal" . #f)

("mperm:20:10:2:1" . 1.0)

("gcbench:20:1" . 1.0243606915521992)

("sboyer:5:1" . 1.00368112089037)

("nboyer:5:1" . 1.0500604616494615)

("chudnovsky:50:500:50:1000" . 1.115409983742645)

("slatex:500" . 1.0002570334324394)

("scheme:100000" . 1.0390531892971318)

("quicksort:10000:2500" . 1.0098428078516584)

("primes:1000:10000" . 1.06437827292514)

("peval:2000" . 1.024359358481343)

("parsing:2500" . 1.0004232513670308)

("paraffins:23:10" . 2.024687502614597)

("nqueens:13:10" . 1.0058665553469563)

("mazefun:11:11:10000" . 1.056194215282642)

("maze:20:7:10000" . 1.020543843425558)

("matrix:5:5:2500" . 1.0028500406404608)

("lattice:44:10" . 1.0)

("graphs:7:3" . 1.0)

("earley:1" . 1.0256148071738138)

("dynamic:500" . 1.008662479565268)

("conform:500" . 1.0332430727462598)

("compiler:2000" . 1.006956809282123)

("read1:2500" . 1.0564058966125223)

("wc:inputs/bib:50" . 1.0335552278476583)

("tail:50" . 1.0)

("cat:50" . 1.0263527061380673)

("sum1:25" . 1.0638438694063255)

("string:500000:100" . 1.1084001200578586)

("array1:1000000:500" . 1.0547700539059255)

("ack:3:12:2" . 1.0)

("simplex:1000000" . 1.021845418562125)

("ray:50" . 1.0558263145762616)

("pnpoly:1000000" . 1.0598933662254997)

("pi:50:500:50:100" . 1.0211985475879175)

("nucleic:50" . 1.0313190698159154)

("mbrotZ:75:1000" . 1.0191939711368345)

("mbrot:75:1000" . 1.0325883765897468)

("fft:65536:100" . 1.1023483002887104)

("sumfp:1000000.0:500" . 1.0794299746553448)

("sum:10000:200000" . 1.063704305941153)

("fibfp:35.0:10" . 1.1160079648284265)

("fibc:30:10" . 1.0)

("fib:40:5" . 1.0533185743228644)

("ctak:27:16:8:1" . 1.2839324786349615)

("cpstak:40:20:11:1" . 1.0290465857280968)

("ntakl:40:20:12:1" . 1.017152038958873)

("takl:40:20:12:1" . 1.0008854776027978)

("tak:40:20:11:1" . 1.0077102329273817)

("triangl:22:1:50" . 1.0435679627887324)

("puzzle:1000" . 1.0006869727744467)

("divrec:1000:1000000" . 1.0)

("diviter:1000:1000000" . 1.0496228639946887)

("destruc:600:50:4000" . 1.0446502508269544)

("deriv:10000000" . 1.0901526807172006)

("browse:2000" . 1.0))

=== Guileregular Geometric Mean slowdown (successful tests / total tests) ===

1.0504373210343294 (112 / 114)

Actual full results from two runs:

=== Guiletuned times ===

(("ack:3:12:2" "2.85671086")

("ack:3:12:2" "2.920076389")

("array1:1000000:500" "2.719338434")

("array1:1000000:500" "2.630303504")

("browse:2000" "4.653660319")

("browse:2000" "5.118496161")

("bv2string:1000:1000:100" "1.433438423")

("bv2string:1000:1000:100" "1.386500303")

("cat:50" "10.334967974")

("cat:50" "10.407612189")

("chudnovsky:50:500:50:1000" "0.362723212")

("chudnovsky:50:500:50:1000" "0.371314469")

("compiler:2000" "2.14676404")

("compiler:2000" "2.355796491")

("conform:500" "4.878185941")

("conform:500" "4.834137994")

("cpstak:40:20:11:1" "17.407945755")

("cpstak:40:20:11:1" "17.213716706")

("ctak:27:16:8:1" "13.296419091")

("ctak:27:16:8:1" "13.101386139")

("deriv:10000000" "6.006358226")

("deriv:10000000" "6.160541397")

("destruc:600:50:4000" "2.524561809")

("destruc:600:50:4000" "2.585279527")

("diviter:1000:1000000" "4.459944493")

("diviter:1000:1000000" "4.655874665")

("divrec:1000:1000000" "6.71596458")

("divrec:1000:1000000" "6.722703389")

("dynamic:500" "3.525123929")

("dynamic:500" "3.809416833")

("earley:1" "4.054948659")

("earley:1" "4.260803481")

("equal" "ULIMITKILLED")

("equal" "ULIMITKILLED")

("fft:65536:100" "2.635766227")

("fft:65536:100" "2.509186252")

("fib:40:5" "4.193038108")

("fib:40:5" "4.202893409")

("fibc:30:10" "40.788333173")

("fibc:30:10" "36.494392282")

("fibfp:35.0:10" "9.395625366")

("fibfp:35.0:10" "9.67364254")

("gcbench:20:1" "0.99502713")

("gcbench:20:1" "1.014673477")

("graphs:7:3" "12.370817417")

("graphs:7:3" "13.124503659")

("lattice:44:10" "6.197259938")

("lattice:44:10" "6.154507367")

("matrix:5:5:2500" "3.641334999")

("matrix:5:5:2500" "3.73835108")

("maze:20:7:10000" "1.613933494")

("maze:20:7:10000" "1.61592533")

("mazefun:11:11:10000" "3.749008984")

("mazefun:11:11:10000" "3.737740216")

("mbrot:75:1000" "19.750189296")

("mbrot:75:1000" "19.153698506")

("mbrotZ:75:1000" "20.612462381")

("mbrotZ:75:1000" "21.07964648")

("mperm:20:10:2:1" "9.617642771")

("mperm:20:10:2:1" "9.281691405")

("nboyer:5:1" "1.872344958")

("nboyer:5:1" "1.925542129")

("nqueens:13:10" "6.834513689")

("nqueens:13:10" "6.715906809")

("ntakl:40:20:12:1" "3.226853095")

("ntakl:40:20:12:1" "3.375634349")

("nucleic:50" "4.703444216")

("nucleic:50" "4.776839468")

("paraffins:23:10" "2.986478347")

("paraffins:23:10" "5.984647443")

("parsing:2500" "3.588179787")

("parsing:2500" "3.589458835")

("peval:2000" "4.217629667")

("peval:2000" "4.453879673")

("pi:50:500:50:100" "0.526672392")

("pi:50:500:50:100" "0.503771353")

("pnpoly:1000000" "12.105856386")

("pnpoly:1000000" "12.216993927")

("primes:1000:10000" "2.763279203")

("primes:1000:10000" "2.859615422")

("puzzle:1000" "7.432281644")

("puzzle:1000" "7.359046832")

("quicksort:10000:2500" "4.919145241")

("quicksort:10000:2500" "4.680086002")

("ray:50" "6.232801675")

("ray:50" "6.289500818")

("read1:2500" "3.322952744")

("read1:2500" "3.135139562")

("sboyer:5:1" "1.585267198")

("sboyer:5:1" "1.550538069")

("scheme:100000" "5.899088913")

("scheme:100000" "6.305636088")

("simplex:1000000" "4.74155611")

("simplex:1000000" "4.764164057")

("slatex:500" "5.5026227")

("slatex:500" "5.537808753")

("string:500000:100" "0.68799508")

("string:500000:100" "0.680705551")

("sum1:25" "3.378466719")

("sum1:25" "3.260874144")

("sum:10000:200000" "1.715693396")

("sum:10000:200000" "1.720665139")

("sumfp:1000000.0:500" "17.278236223")

("sumfp:1000000.0:500" "17.121625997")

("tail:50" "4.837888687")

("tail:50" "4.868554424")

("tak:40:20:11:1" "2.207812177")

("tak:40:20:11:1" "2.13063615")

("takl:40:20:12:1" "3.273725333")

("takl:40:20:12:1" "3.356636843")

("triangl:22:1:50" "3.13643409")

("triangl:22:1:50" "3.112446043")

("wc:inputs/bib:50" "6.354178924")

("wc:inputs/bib:50" "6.062416889"))

=== Guileregular times ===

(("ack:3:12:2" "2.823197595")

("ack:3:12:2" "2.839448474")

("array1:1000000:500" "2.727697517")

("array1:1000000:500" "2.58605893")

("browse:2000" "4.62347843")

("browse:2000" "4.710018671")

("bv2string:1000:1000:100" "1.436538041")

("bv2string:1000:1000:100" "1.480061766")

("cat:50" "9.705074729")

("cat:50" "9.45588653")

("chudnovsky:50:500:50:1000" "0.404585092")

("chudnovsky:50:500:50:1000" "0.425650091")

("compiler:2000" "2.161698668")

("compiler:2000" "2.212490501")

("conform:500" "4.994839595")

("conform:500" "4.891769316")

("cpstak:40:20:11:1" "17.713716404")

("cpstak:40:20:11:1" "17.853171003")

("ctak:27:16:8:1" "16.821295179")

("ctak:27:16:8:1" "14.463871749")

("deriv:10000000" "6.507820501")

("deriv:10000000" "5.969641332")

("destruc:600:50:4000" "2.637284127")

("destruc:600:50:4000" "2.736220883")

("diviter:1000:1000000" "4.681259712")

("diviter:1000:1000000" "4.723619732")

("divrec:1000:1000000" "6.673202011")

("divrec:1000:1000000" "6.973599182")

("dynamic:500" "3.555660243")

("dynamic:500" "4.094633639")

("earley:1" "4.158815387")

("earley:1" "4.444094437")

("equal" "ULIMITKILLED")

("equal" "ULIMITKILLED")

("fft:65536:100" "2.7659972")

("fft:65536:100" "2.556631925")

("fib:40:5" "4.416604922")

("fib:40:5" "4.295627134")

("fibc:30:10" "36.233365092")

("fibc:30:10" "42.71370317")

("fibfp:35.0:10" "10.485592743")

("fibfp:35.0:10" "9.711991969")

("gcbench:20:1" "1.019266679")

("gcbench:20:1" "1.012182828")

("graphs:7:3" "11.912688219")

("graphs:7:3" "14.273013145")

("lattice:44:10" "6.134356748")

("lattice:44:10" "6.183086766")

("matrix:5:5:2500" "3.579898395")

("matrix:5:5:2500" "3.569724535")

("maze:20:7:10000" "1.647089891")

("maze:20:7:10000" "1.692950212")

("mazefun:11:11:10000" "3.815654415")

("mazefun:11:11:10000" "3.612644682")

("mbrot:75:1000" "19.777886446")

("mbrot:75:1000" "19.78458033")

("mbrotZ:75:1000" "21.008097389")

("mbrotZ:75:1000" "21.703652079")

("mperm:20:10:2:1" "8.094478518")

("mperm:20:10:2:1" "8.861341606")

("nboyer:5:1" "1.924134672")

("nboyer:5:1" "1.832403697")

("nqueens:13:10" "6.755306048")

("nqueens:13:10" "6.783168806")

("ntakl:40:20:12:1" "3.282200205")

("ntakl:40:20:12:1" "3.295829099")

("nucleic:50" "4.842285137")

("nucleic:50" "4.695234752")

("paraffins:23:10" "6.046685386")

("paraffins:23:10" "6.098533486")

("parsing:2500" "3.589698489")

("parsing:2500" "3.738489773")

("peval:2000" "4.32036842")

("peval:2000" "4.255106956")

("pi:50:500:50:100" "0.514450574")

("pi:50:500:50:100" "0.514649509")

("pnpoly:1000000" "12.830916876")

("pnpoly:1000000" "13.164018562")

("primes:1000:10000" "2.921479898")

("primes:1000:10000" "2.744775962")

("puzzle:1000" "6.021742132")

("puzzle:1000" "6.017608199")

("quicksort:10000:2500" "4.71159952")

("quicksort:10000:2500" "4.665676166")

("ray:50" "6.580756022")

("ray:50" "6.506202092")

("read1:2500" "3.31197992")

("read1:2500" "3.279529166")

("sboyer:5:1" "1.496605726")

("sboyer:5:1" "1.491116745")

("scheme:100000" "6.129467149")

("scheme:100000" "6.622910116")

("simplex:1000000" "4.766683108")

("simplex:1000000" "4.664779057")

("slatex:500" "5.504037058")

("slatex:500" "5.634341685")

("string:500000:100" "0.750405621")

("string:500000:100" "0.677016907")

("sum1:25" "3.469060967")

("sum1:25" "3.372192131")

("sum:10000:200000" "1.824990453")

("sum:10000:200000" "1.765433455")

("sumfp:1000000.0:500" "18.481596316")

("sumfp:1000000.0:500" "17.947802847")

("tail:50" "4.54999235")

("tail:50" "4.695161894")

("tak:40:20:11:1" "2.147063851")

("tak:40:20:11:1" "2.157571204")

("takl:40:20:12:1" "3.25309459")

("takl:40:20:12:1" "3.250216596")

("triangl:22:1:50" "3.201959263")

("triangl:22:1:50" "3.068280531")

("wc:inputs/bib:50" "6.111397639")

("wc:inputs/bib:50" "5.912986045"))

Best wishes,

Arne

--

Unpolitisch sein

heißt politisch sein,

ohne es zu merken.

draketo.de

Hi,

"Dr. Arne Babenhauserheide" <arne_bab@web.de> skribis:

Marking a package as tunable does not involve any rebuilds, by design:

https://hpc.guix.info/blog/2022/01/tuning-packages-for-a-cpu-micro-architecture/

Thus there’s no need to create package variants.

[...]

As far as I know, there’s no reason why Guile’s C code would benefit

from ISA extensions (we’re mostly talking about vector instructions like

those of AVX). The only part in libguile that’s performance-critical is

the VM engine—but I don’t see how that code could benefit from vector

instructions. Likewise, I doubt libgc build with ‘-march=znver3’ has

any chance of being any faster; most likely, the generated code is close

to the baseline.

For these reasons, I’m reluctant to this patch. Tuning should be done

only on packages where it makes sense, typically linear algebra code,

image/video processing, and so on.

WDYT?

Ludo’.