GCC 8.0 vs. Clang 6.0 AMD EPYC Tuning Comparison

Tests for a future article on Phoronix.

HTML result view exported from: https://openbenchmarking.org/result/1801022-AL-CLANGCC8680.

ProcessorMotherboardChipsetMemoryDiskGraphicsMonitorNetworkOSKernelDesktopDisplay DriverOpenCLCompilerFile-SystemScreen ResolutionSystem LayerGCC 8.0Clang 6.0 -march=x86-64 -march=znver1 -march=x86-64 -march=znver1AMD EPYC 7601 32-Core @ 2.20GHz (64 Cores)TYAN B8026T70AE24HRAMD Device 1450126976MB280GB INTEL SSDPE21D280GAASPEED ASPEED FamilyVE228Broadcom Limited NetXtreme BCM5720 Gigabit PCIeUbuntu 17.104.13.0-21-generic (x86_64)GNOME Shell 3.26.1modesetting 1.19.5OpenCL 1.2 pocl 1.0 LLVM 5.0.0GCC 8.0.0 20171231 + clang (GCC) 8.0.0 20171231 (experimental) + LLVM 5.0.0ext41920x1080vm-other Xen 4.9.0 HypervisorClang 6.0.0 (SVN 321623) + LLVM 6.0.0svnOpenBenchmarking.orgCompiler Details- GCC 8.0: -march=x86-64: --disable-multilib --enable-checking=release- GCC 8.0: -march=znver1: --disable-multilib --enable-checking=release- Clang 6.0: -march=x86-64: Optimized build; Default target: x86_64-unknown-linux-gnu; Host CPU: znver1- Clang 6.0: -march=znver1: Optimized build; Default target: x86_64-unknown-linux-gnu; Host CPU: znver1Disk Details- NONE / data=ordered,errors=remount-ro,relatime,rw

sqlite: Default Test Directorypolybench-c: 3 Matrix Multiplicationsfftw: Stock - 2D FFT Size 4096fftw: Float + SSE - 2D FFT Size 4096hmmer: Pfam Database Searchscimark2: Compositescimark2: Monte Carloscimark2: Fast Fourier Transformscimark2: Sparse Matrix Multiplyscimark2: Dense LU Matrix Factorizationscimark2: Jacobi Successive Over-Relaxationtscp: AI Chess Performancegraphics-magick: Blurgraphics-magick: Sharpengraphics-magick: HWB Color Spacegraphics-magick: Local Adaptive Thresholdinghimeno: Poisson Pressure Solverebizzy: c-ray: Total Timebullet: Raytestsbullet: 3000 Fallbullet: 1000 Stackbullet: 1000 Convexbullet: 136 Ragdollsbullet: Prim Trimeshbullet: Convex Trimeshencode-flac: WAV To FLACencode-mp3: WAV To MP3apache: Static Web Page ServingGCC 8.0Clang 6.0 -march=x86-64 -march=znver1 -march=x86-64 -march=znver17.6160.684959.7313.651579.48561.03233.892263.873513.111423.1487425111615717792949.1911260323.933.125.346.185.443.261.101.347.1211.109841.307.1665.455627.831363012.401680.45555.76231.092259.953678.861676.6287508512316518695935.6411011763.373.065.275.935.283.191.101.307.4510.819791.237.5362.984660.831364912.851479.53531.38179.292190.103190.431110.65917658101131150971032.7110766484.533.225.486.305.433.281.101.337.9411.339531.437.4862.755031.601248111.091699.32552.19226.682258.644034.891424.21918269104136155981052.4711454054.483.185.346.085.313.231.091.326.6312.819663.93OpenBenchmarking.org

SQLite

Test Target: Default Test Directory

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterSQLite 3.8.10.2Test Target: Default Test Directory-march=x86-64-march=znver1246810SE +/- 0.12, N = 3SE +/- 0.12, N = 6SE +/- 0.06, N = 3SE +/- 0.15, N = 67.617.167.537.481. (CC) gcc options: -O3 -ldl -lpthread

PolyBench-C

Test: 3 Matrix Multiplications

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterPolyBench-C 3.2Test: 3 Matrix Multiplications-march=x86-64-march=znver11530456075SE +/- 0.03, N = 3SE +/- 0.21, N = 3SE +/- 0.24, N = 3SE +/- 0.26, N = 360.6865.4562.9862.751. (CC) gcc options: -O3 -march=native

FFTW

Build: Stock - Size: 2D FFT Size 4096

GCC 8.0Clang 6.0OpenBenchmarking.orgMflops, More Is BetterFFTW 3.3.6Build: Stock - Size: 2D FFT Size 4096-march=x86-64-march=znver112002400360048006000SE +/- 1.16, N = 3SE +/- 10.37, N = 3SE +/- 57.25, N = 3SE +/- 17.93, N = 34959.735627.834660.835031.601. (CC) gcc options: -pthread -O3 -lm

FFTW

Build: Float + SSE - Size: 2D FFT Size 4096

GCC 8.0Clang 6.0OpenBenchmarking.orgMflops, More Is BetterFFTW 3.3.6Build: Float + SSE - Size: 2D FFT Size 4096-march=znver1-march=x86-643K6K9K12K15KSE +/- 15.38, N = 3SE +/- 100.95, N = 3SE +/- 76.70, N = 31363013649124811. (CC) gcc options: -pthread -O3 -lm

Timed HMMer Search

Pfam Database Search

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterTimed HMMer Search 2.3.2Pfam Database Search-march=x86-64-march=znver148121620SE +/- 1.74, N = 6SE +/- 0.04, N = 3SE +/- 1.28, N = 6SE +/- 0.12, N = 313.6512.4012.8511.091. (CC) gcc options: -O3 -pthread -lhmmer -lsquid -lm

SciMark

Computational Test: Composite

GCC 8.0Clang 6.0OpenBenchmarking.orgMflops, More Is BetterSciMark 2.0Computational Test: Composite-march=x86-64-march=znver1400800120016002000SE +/- 20.51, N = 7SE +/- 20.36, N = 4SE +/- 36.18, N = 8SE +/- 25.95, N = 41579.481680.451479.531699.321. (CC) gcc options: -O3 -lm

SciMark

Computational Test: Monte Carlo

GCC 8.0Clang 6.0OpenBenchmarking.orgMflops, More Is BetterSciMark 2.0Computational Test: Monte Carlo-march=x86-64-march=znver1120240360480600SE +/- 0.04, N = 4SE +/- 0.01, N = 4SE +/- 0.05, N = 4SE +/- 0.05, N = 4561.03555.76531.38552.191. (CC) gcc options: -O3 -lm

SciMark

Computational Test: Fast Fourier Transform

GCC 8.0Clang 6.0OpenBenchmarking.orgMflops, More Is BetterSciMark 2.0Computational Test: Fast Fourier Transform-march=x86-64-march=znver150100150200250SE +/- 0.35, N = 4SE +/- 0.14, N = 4SE +/- 45.12, N = 4SE +/- 0.38, N = 4233.89231.09179.29226.681. (CC) gcc options: -O3 -lm

SciMark

Computational Test: Sparse Matrix Multiply

GCC 8.0Clang 6.0OpenBenchmarking.orgMflops, More Is BetterSciMark 2.0Computational Test: Sparse Matrix Multiply-march=x86-64-march=znver15001000150020002500SE +/- 6.72, N = 4SE +/- 9.55, N = 4SE +/- 11.89, N = 4SE +/- 29.07, N = 42263.872259.952190.102258.641. (CC) gcc options: -O3 -lm

SciMark

Computational Test: Dense LU Matrix Factorization

GCC 8.0Clang 6.0OpenBenchmarking.orgMflops, More Is BetterSciMark 2.0Computational Test: Dense LU Matrix Factorization-march=x86-64-march=znver19001800270036004500SE +/- 177.13, N = 4SE +/- 110.38, N = 4SE +/- 17.92, N = 4SE +/- 135.63, N = 43513.113678.863190.434034.891. (CC) gcc options: -O3 -lm

SciMark

Computational Test: Jacobi Successive Over-Relaxation

GCC 8.0Clang 6.0OpenBenchmarking.orgMflops, More Is BetterSciMark 2.0Computational Test: Jacobi Successive Over-Relaxation-march=x86-64-march=znver1400800120016002000SE +/- 0.28, N = 4SE +/- 0.77, N = 4SE +/- 310.16, N = 4SE +/- 0.49, N = 41423.141676.621110.651424.211. (CC) gcc options: -O3 -lm

TSCP

AI Chess Performance

GCC 8.0Clang 6.0OpenBenchmarking.orgNodes Per Second, More Is BetterTSCP 1.81AI Chess Performance-march=x86-64-march=znver1200K400K600K800K1000KSE +/- 438.77, N = 5SE +/- 556.88, N = 5SE +/- 572.37, N = 5SE +/- 306.40, N = 58742518750859176589182691. (CC) gcc options: -O3 -march=native

GraphicsMagick

Operation: Blur

GCC 8.0Clang 6.0OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.19Operation: Blur-march=x86-64-march=znver1306090120150SE +/- 0.33, N = 3SE +/- 0.33, N = 31161231011041. (CC) gcc options: -fopenmp -O3 -pthread -ljbig -lwebp -ltiff -ljpeg -lXext -lSM -lICE -lX11 -llzma -lbz2 -lz -lm -lgomp -lpthread

GraphicsMagick

Operation: Sharpen

GCC 8.0Clang 6.0OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.19Operation: Sharpen-march=x86-64-march=znver14080120160200SE +/- 0.33, N = 3SE +/- 0.67, N = 31571651311361. (CC) gcc options: -fopenmp -O3 -pthread -ljbig -lwebp -ltiff -ljpeg -lXext -lSM -lICE -lX11 -llzma -lbz2 -lz -lm -lgomp -lpthread

GraphicsMagick

Operation: HWB Color Space

GCC 8.0Clang 6.0OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.19Operation: HWB Color Space-march=x86-64-march=znver14080120160200SE +/- 1.20, N = 3SE +/- 0.88, N = 31771861501551. (CC) gcc options: -fopenmp -O3 -pthread -ljbig -lwebp -ltiff -ljpeg -lXext -lSM -lICE -lX11 -llzma -lbz2 -lz -lm -lgomp -lpthread

GraphicsMagick

Operation: Local Adaptive Thresholding

GCC 8.0Clang 6.0OpenBenchmarking.orgIterations Per Minute, More Is BetterGraphicsMagick 1.3.19Operation: Local Adaptive Thresholding-march=x86-64-march=znver120406080100SE +/- 0.33, N = 3SE +/- 0.33, N = 3929597981. (CC) gcc options: -fopenmp -O3 -pthread -ljbig -lwebp -ltiff -ljpeg -lXext -lSM -lICE -lX11 -llzma -lbz2 -lz -lm -lgomp -lpthread

Himeno Benchmark

Poisson Pressure Solver

GCC 8.0Clang 6.0OpenBenchmarking.orgMFLOPS, More Is BetterHimeno Benchmark 3.0Poisson Pressure Solver-march=x86-64-march=znver12004006008001000SE +/- 0.43, N = 3SE +/- 0.79, N = 3SE +/- 1.46, N = 3SE +/- 1.50, N = 3949.19935.641032.711052.471. (CC) gcc options: -O3 -mavx2

ebizzy

GCC 8.0Clang 6.0OpenBenchmarking.orgRecords/s, More Is Betterebizzy 0.3-march=x86-64-march=znver1200K400K600K800K1000KSE +/- 20747.16, N = 6SE +/- 19461.21, N = 3SE +/- 12350.97, N = 3SE +/- 17141.58, N = 611260321101176107664811454051. (CC) gcc options: -pthread -lpthread -O3 -march=native

C-Ray

Total Time

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterC-Ray 1.1Total Time-march=x86-64-march=znver11.01932.03863.05794.07725.0965SE +/- 0.00, N = 3SE +/- 0.02, N = 3SE +/- 0.02, N = 3SE +/- 0.01, N = 33.933.374.534.481. (CC) gcc options: -lm -lpthread -O3

Bullet Physics Engine

Test: Raytests

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterBullet Physics Engine 2.81Test: Raytests-march=x86-64-march=znver10.72451.4492.17352.8983.6225SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 33.123.063.223.181. (CXX) g++ options: -O3 -rdynamic

Bullet Physics Engine

Test: 3000 Fall

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterBullet Physics Engine 2.81Test: 3000 Fall-march=x86-64-march=znver11.2332.4663.6994.9326.165SE +/- 0.01, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 35.345.275.485.341. (CXX) g++ options: -O3 -rdynamic

Bullet Physics Engine

Test: 1000 Stack

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterBullet Physics Engine 2.81Test: 1000 Stack-march=x86-64-march=znver1246810SE +/- 0.02, N = 3SE +/- 0.01, N = 3SE +/- 0.03, N = 3SE +/- 0.01, N = 36.185.936.306.081. (CXX) g++ options: -O3 -rdynamic

Bullet Physics Engine

Test: 1000 Convex

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterBullet Physics Engine 2.81Test: 1000 Convex-march=x86-64-march=znver11.2242.4483.6724.8966.12SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 3SE +/- 0.01, N = 35.445.285.435.311. (CXX) g++ options: -O3 -rdynamic

Bullet Physics Engine

Test: 136 Ragdolls

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterBullet Physics Engine 2.81Test: 136 Ragdolls-march=x86-64-march=znver10.7381.4762.2142.9523.69SE +/- 0.00, N = 3SE +/- 0.01, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 33.263.193.283.231. (CXX) g++ options: -O3 -rdynamic

Bullet Physics Engine

Test: Prim Trimesh

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterBullet Physics Engine 2.81Test: Prim Trimesh-march=x86-64-march=znver10.24750.4950.74250.991.2375SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 31.101.101.101.091. (CXX) g++ options: -O3 -rdynamic

Bullet Physics Engine

Test: Convex Trimesh

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterBullet Physics Engine 2.81Test: Convex Trimesh-march=x86-64-march=znver10.30150.6030.90451.2061.5075SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 3SE +/- 0.00, N = 31.341.301.331.321. (CXX) g++ options: -O3 -rdynamic

FLAC Audio Encoding

WAV To FLAC

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterFLAC Audio Encoding 1.3.1WAV To FLAC-march=x86-64-march=znver1246810SE +/- 0.01, N = 5SE +/- 0.01, N = 5SE +/- 0.01, N = 5SE +/- 0.00, N = 57.127.457.946.631. (CXX) g++ options: -O3 -logg -lm

LAME MP3 Encoding

WAV To MP3

GCC 8.0Clang 6.0OpenBenchmarking.orgSeconds, Fewer Is BetterLAME MP3 Encoding 3.99.5WAV To MP3-march=x86-64-march=znver13691215SE +/- 0.02, N = 5SE +/- 0.01, N = 5SE +/- 0.00, N = 5SE +/- 0.01, N = 511.1010.8111.3312.811. (CC) gcc options: -O3 -ffast-math -funroll-loops -fschedule-insns2 -fbranch-count-reg -fforce-addr -pipe -lm

Apache Benchmark

Static Web Page Serving

GCC 8.0Clang 6.0OpenBenchmarking.orgRequests Per Second, More Is BetterApache Benchmark 2.4.7Static Web Page Serving-march=x86-64-march=znver12K4K6K8K10KSE +/- 22.88, N = 3SE +/- 35.54, N = 3SE +/- 161.49, N = 3SE +/- 121.20, N = 39841.309791.239531.439663.931. (CC) gcc options: -shared -fPIC -pthread -O3

Phoronix Test Suite v10.8.4