October 22, 2026
\nIn Why Lean is Faster Than Rust, the author shares an impressive result: a DEFLATE compression library written in Lean 4, a functional programming language and proof assistant, outperforming an equivalent Rust package that crates.io says has been downloaded over a billion times. From the article:
\n\n
lean-zipbeats every single compression levelminiz_oxideprovides, i.e. we're capable of providing the same compression at higher throughput. At its fastest setting we're 15% quicker and the output is 8.8% smaller.
However, the article also notes that while they managed to surpass Rust's miniz_oxide, they were no match for the end boss of DEFLATE performance: libdeflate, written in C.
\n
lean-zipis certainly not the best here:libdeflateunsurprisingly blows it out of the water (unsurprisingly because this is a very carefully tuned implementation using architecture-specific SIMD, that we can't touch in Lean).
Enter Roc, a functional programming language (but not a theorem prover) with an explicit focus on runtime performance.
\nRoc's tagline is "Fast, Friendly, Functional," and Fast comes first because both running fast and compiling fast have always been central goals. Roc is not a mature language (it's still weeks away from its first numbered release, 0.1.0), and its nascent package ecosystem has had almost no optimization work done on it. Still, between its language design, longstanding optimizations, and brand-new compiler implementation, I figured Roc already had enough of a foundation in place to take on the champ: libdeflate.
\nSo I spent a bunch of time grinding through performance optimizations on roc-deflate, a package Niclas Ahden wrote (and uses in paid production work) for DEFLATE compression. I didn't expect to beat libdeflate's performance across the board, but I wanted to see how close it could get on the non-streaming APIs (as roc-deflate hasn't implemented streaming yet), using the classic Silesia corpus for benchmarking and verifying that it produced identical outputs.
\nHere's where the benchmarks stand today, running on the 2021 M1 MacBook Pro where I did all the optimization work, as well as on an AMD Ryzen 9 3900X Ubuntu machine for an x64 comparison.
\nroc-deflate ran these 144 benchmarks slower on average than libdeflate, in the median case 8.9% slower on the Ryzen 3900X, and 2.6% on the M1 MacBook Pro.
\nM1 MacBook Pro\nRyzen 3900X Ubuntu
\nMB/s
\n700\n600\n500\n400\n300\n200\n100\n0
\nfile that ran fastest
\nfile that ran slowest
\nL1
\nL2
\nL3
\nL4
\nL5
\nL6
\nL7
\nL8
\nL9
\nL10
\nL11
\nL12
\nlibdeflate\nroc-deflate
\nlibdeflate at L1 compression
\n230.57 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n248.31 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n312.26 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n601.03 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n169.98 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n284.01 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n281.14 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n344.00 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n160.21 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n268.58 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n145.93 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L1 compression
\n476.83 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n202.92 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n225.79 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n277.57 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n529.49 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n153.99 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n257.14 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n245.80 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n309.79 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n145.28 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n239.18 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n134.66 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L1 compression
\n419.07 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n161.68 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n184.55 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n204.86 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n453.14 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n128.05 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n203.10 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n207.66 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n249.57 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n112.78 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n198.86 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n117.18 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L2 compression
\n352.59 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n143.60 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n168.37 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n182.07 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n409.82 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n118.08 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n185.41 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n183.69 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n228.02 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n105.68 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n181.07 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n107.00 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L2 compression
\n323.17 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n130.85 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n169.26 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n171.11 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n415.29 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n118.92 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n183.10 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n169.42 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n223.27 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n100.60 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n173.36 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n116.62 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L3 compression
\n324.65 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n122.04 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n158.29 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n159.07 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n382.21 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n111.14 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n171.27 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n157.19 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n209.86 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n95.51 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n162.75 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n105.93 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L3 compression
\n300.31 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n119.94 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n158.07 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n156.88 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n376.70 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n113.68 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n164.59 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n152.16 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n207.97 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n94.48 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n158.64 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n116.63 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L4 compression
\n294.31 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n113.71 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n148.77 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n149.00 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n356.47 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n107.59 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n156.41 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n142.45 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n197.17 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n89.37 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n151.15 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n106.32 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L4 compression
\n278.53 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n95.14 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n133.06 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n129.45 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n287.09 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n95.87 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n140.08 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n120.63 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n170.75 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n83.22 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n125.60 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n102.48 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L5 compression
\n229.70 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n90.27 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n125.87 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n122.34 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n311.35 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n89.73 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n133.37 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n115.52 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n167.01 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n76.99 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n123.12 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n90.40 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L5 compression
\n237.29 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n70.60 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n107.04 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n96.85 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n221.32 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n82.91 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n122.89 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n79.91 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n129.19 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n67.48 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n96.76 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n102.50 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L6 compression
\n178.60 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n72.83 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n107.54 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n98.92 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n246.59 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n81.43 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n122.02 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n84.85 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n134.11 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n66.36 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n101.16 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n90.67 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L6 compression
\n194.35 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n54.23 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n75.96 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n66.46 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n157.22 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n70.52 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n117.53 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n46.77 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n91.00 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n59.49 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n71.34 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n102.43 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L7 compression
\n128.20 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n56.93 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n79.18 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n70.19 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n174.91 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n70.27 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n115.95 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n51.20 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n96.46 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n59.40 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n75.99 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n90.26 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L7 compression
\n141.60 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n37.50 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n39.17 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n43.21 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n83.68 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n52.45 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n106.76 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n24.87 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n55.69 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n51.86 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n45.55 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n91.30 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L8 compression
\n80.95 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n39.56 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n43.00 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n45.70 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n91.28 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n52.69 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n101.66 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n27.81 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n60.04 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n51.65 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n48.78 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n81.01 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L8 compression
\n86.27 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n34.82 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n27.51 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n38.60 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n58.64 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n47.26 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n106.10 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n20.65 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n43.95 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n51.18 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n40.57 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n91.69 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L9 compression
\n68.13 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n36.73 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n30.48 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n40.95 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n64.32 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n48.05 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n101.46 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n22.85 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n47.64 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n50.68 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n43.36 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n80.68 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L9 compression
\n73.29 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n12.07 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n14.87 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n16.07 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n7.51 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n18.09 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n16.97 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n8.64 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n14.74 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n24.08 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n10.10 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n34.15 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L10 compression
\n11.46 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n12.97 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n14.67 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n15.47 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n7.19 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n18.02 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n17.23 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n9.63 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n14.59 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n22.93 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n11.18 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n32.15 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L10 compression
\n10.73 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n8.85 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n10.27 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n8.76 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n4.82 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n13.56 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n12.91 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n6.49 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n8.46 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n19.43 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n6.77 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n26.05 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L11 compression
\n6.60 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n10.10 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n10.53 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n8.19 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n4.74 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n14.03 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n13.62 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n7.71 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n8.68 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n18.87 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n8.07 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n24.93 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L11 compression
\n6.51 MB/s compressing xml
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n7.35 MB/s compressing dickens
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n7.97 MB/s compressing mozilla
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n4.26 MB/s compressing mr
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n3.51 MB/s compressing nci
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n11.45 MB/s compressing ooffice
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n11.40 MB/s compressing osdb
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n5.59 MB/s compressing reymont
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n5.92 MB/s compressing samba
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n18.10 MB/s compressing sao
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n5.82 MB/s compressing webster
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n23.50 MB/s compressing x-ray
\n(median across 15 runs)
\nlibdeflate at L12 compression
\n4.19 MB/s compressing xml
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n8.59 MB/s compressing dickens
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n8.22 MB/s compressing mozilla
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n3.83 MB/s compressing mr
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n3.48 MB/s compressing nci
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n11.99 MB/s compressing ooffice
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n12.21 MB/s compressing osdb
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n6.87 MB/s compressing reymont
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n6.03 MB/s compressing samba
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n17.74 MB/s compressing sao
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n7.11 MB/s compressing webster
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n22.60 MB/s compressing x-ray
\n(median across 15 runs)
\nroc-deflate at L12 compression
\n4.18 MB/s compressing xml
\n(median across 15 runs)
\nCompression Level
\nMB/s
\n700\n600\n500\n400\n300\n200\n100\n0
\nfile that ran fastest\nfile that ran slowest
\nL1
\nL2
\nL3
\nL4
\nL5
\nL6
\nL7
\nL8
\nL9
\nL10
\nL11
\nL12
\nlibdeflate\nroc-deflate
\nlibdeflate at L1 compression
\n191.10 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n235.50 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n261.70 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n387.30 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n167.90 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n254.70 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n229.60 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n291.00 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n157.40 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n207.00 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n142.20 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L1 compression
\n411.30 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n182.30 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n204.50 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n228.90 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n428.60 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n144.60 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n222.30 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n217.30 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n270.30 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n138.50 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n213.90 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n130.80 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L1 compression
\n353.80 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n138.50 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n162.80 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n153.40 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n286.50 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n120.40 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n175.30 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n156.80 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n210.50 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n104.20 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n163.50 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n111.60 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L2 compression
\n258.40 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n121.20 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n144.50 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n145.30 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n326.40 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n105.10 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n155.00 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n146.00 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n188.40 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n94.10 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n151.60 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n95.50 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L2 compression
\n248.60 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n112.90 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n152.80 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n132.10 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n304.30 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n113.00 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n164.00 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n128.00 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n191.40 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n94.50 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n143.70 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n112.20 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L3 compression
\n209.90 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n100.80 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n132.50 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n125.30 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n301.90 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n97.90 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n148.80 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n120.80 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n169.40 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n84.80 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n133.90 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n95.10 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L3 compression
\n231.70 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n100.80 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n144.50 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n123.40 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n254.00 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n109.70 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n151.90 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n118.90 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n178.70 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n89.10 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n129.70 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n111.30 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L4 compression
\n201.50 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n91.90 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n126.60 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n117.60 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n282.80 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n95.50 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n138.00 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n106.80 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n159.20 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n81.50 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n125.00 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n95.60 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L4 compression
\n215.50 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n82.40 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n120.00 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n103.10 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n250.40 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n89.30 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n128.80 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n94.60 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n152.10 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n74.80 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n112.20 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n93.70 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L5 compression
\n198.40 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n73.50 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n105.70 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n96.60 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n247.90 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n77.50 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n116.60 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n86.20 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n133.00 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n68.10 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n100.10 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n82.10 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L5 compression
\n182.10 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n65.40 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n102.90 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n81.60 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n203.10 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n78.50 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n118.80 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n71.00 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n120.10 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n64.40 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n91.80 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n94.80 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L6 compression
\n168.30 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n57.80 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n89.40 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n74.10 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n188.20 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n70.60 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n107.80 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n60.90 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n105.40 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n59.20 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n81.30 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n83.20 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L6 compression
\n149.50 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n49.60 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n76.10 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n56.30 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n150.20 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n69.10 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n113.80 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n39.90 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n85.90 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n53.10 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n66.10 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n94.20 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L7 compression
\n119.90 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n45.20 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n67.20 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n52.40 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n141.70 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n61.10 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n106.10 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n36.20 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n76.50 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n52.80 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n59.40 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n81.60 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L7 compression
\n108.40 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n32.10 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n41.40 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n35.80 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n80.70 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n50.40 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n97.80 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n20.60 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n53.40 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n46.60 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n40.00 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n78.30 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L8 compression
\n72.60 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n30.00 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n36.80 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n33.90 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n75.50 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n45.60 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n90.20 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n19.50 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n49.00 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n44.50 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n37.10 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n72.10 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L8 compression
\n66.40 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n29.00 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n29.10 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n32.50 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n58.50 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n45.00 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n95.30 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n16.90 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n41.60 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n45.60 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n33.90 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n78.60 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L9 compression
\n57.90 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n27.10 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n26.30 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n30.60 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n54.40 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n41.20 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n90.80 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n15.90 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n38.70 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n43.30 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n31.70 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n72.70 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L9 compression
\n54.80 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n12.10 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n14.80 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n16.20 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n7.30 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n18.20 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n16.10 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n8.70 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n14.40 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n23.30 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n10.00 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n34.70 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L10 compression
\n10.80 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n12.30 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n13.70 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n14.70 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n6.80 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n16.80 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n15.50 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n9.00 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n13.60 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n20.60 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n10.20 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n29.40 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L10 compression
\n9.80 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n8.80 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n10.30 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n9.30 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n4.70 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n13.70 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n12.30 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n6.50 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n8.40 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n18.90 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n6.70 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n26.80 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L11 compression
\n6.20 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n9.50 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n9.90 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n8.30 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n4.50 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n13.00 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n12.30 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n7.10 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n8.30 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n17.00 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n7.30 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n23.00 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L11 compression
\n6.00 MB/s compressing xml
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n7.40 MB/s compressing dickens
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n8.00 MB/s compressing mozilla
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n4.70 MB/s compressing mr
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n3.40 MB/s compressing nci
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n11.50 MB/s compressing ooffice
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n10.90 MB/s compressing osdb
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n5.60 MB/s compressing reymont
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n5.90 MB/s compressing samba
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n17.70 MB/s compressing sao
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n5.80 MB/s compressing webster
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n24.60 MB/s compressing x-ray
\n(median across 7 runs)
\nlibdeflate at L12 compression
\n4.00 MB/s compressing xml
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n8.10 MB/s compressing dickens
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n7.80 MB/s compressing mozilla
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n4.10 MB/s compressing mr
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n3.30 MB/s compressing nci
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n11.20 MB/s compressing ooffice
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n11.10 MB/s compressing osdb
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n6.30 MB/s compressing reymont
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n5.80 MB/s compressing samba
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n15.80 MB/s compressing sao
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n6.40 MB/s compressing webster
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n21.00 MB/s compressing x-ray
\n(median across 7 runs)
\nroc-deflate at L12 compression
\n3.90 MB/s compressing xml
\n(median across 7 runs)
\nCompression Level
\nSo while we didn't outperform the champ on average, a median slowdown percentage in the single digits compared to such a carefully-tuned SIMD-powered C library is wild. It's frankly not a result I knew was achievable going in, since Roc is a functional language with automatic memory management racing against code that doesn't have to do bounds checks. (Bounds checks have a cost, but then again, the last time a skipped bounds check in C compression code led to a major Remote Code Execution vulnerability was yesterday.)
\nThe graph shows more detail than just the median gap. On the worst-case inputs, roc-deflate was 18% slower than libdeflate, but on the best-case inputs, roc-deflate was slightly over 18% faster! That's not because Roc has some magical optimization that C can't access; it's just that the two implementations are different, and although on average roc-deflate is slower, for certain inputs it happens to be faster. So while it's possible to get a speedup by switching from libdeflate to roc-deflate, you shouldn't expect it on average.
\nTo be clear, C's performance ceiling is higher than Roc's. For any input where roc-deflate currently outperforms libdeflate, C code can close that gap. The reverse is not true; Roc code is memory-safe, does automatic reference counting, checks bounds, and supports no Rust-style unsafe to opt out of any of that. These properties give Roc code stronger safety guarantees than code written in C, Zig, Rust, and other languages that allow full access to the hardware—and the price it pays for those guarantees is a lower performance ceiling.
Obviously we can't extrapolate from this one result about how close Roc can get to C in the general case, but now something has been demonstrated that nobody knew was achievable before. As far as I know, this is the most optimization effort that has been put into any Roc package to date, so how close others can get to this result is a complete unknown right now. So while I'm thrilled by this result, I'm fully aware that we won't know how well it generalizes until we've optimized a wider sample of real-world Roc code bases!
\nAn obvious question is "Why settle for any amount slower?" If you could get C levels of performance, why not use libdeflate itself and close that remaining gap?
\nA pure-Roc implementation has several advantages over a pure-C one:
\nBool -> Bool, whereas effectful functions use a thick arrow, like Bool => Bool. (As with memory safety, Roc enforces this and supports no way to opt out.) So when you see that roc-deflate's functions have only thin arrows in their annotations, you know they really are pure functions.arrayref package two months ago.If any of these guarantees don't hold in practice, it can only be due to a defect in Roc's compiler, the platform the application is running on, or the underlying hardware. So when you see in roc-deflate's docs that the entire package is 100% pure functions, you can be confident you are getting all of the above properties from everything the package has to offer. The type system enforces it!
\nThe reason we've been working so hard on all this might become clearer if you look at the proposition in reverse: would you give up guaranteed memory safety, guaranteed lack of side effects, guaranteed deterministic behavior across all targets, and the security benefits of having no exploitable build scripts in packages...for an average speedup percentage in the single digits?
\nThat answer can legitimately vary by use case—Roc's own compiler, for example, wants the maximum performance the hardware can deliver—but I bet a lot of people reading this can see why we're so excited to have achieved this result.
\nMost of the optimizations that improved roc-deflate's numbers came from general optimization techniques that are widely beneficial across languages. (I always recommend Casey Muratori's Performance-Aware Programming Series, which I paid for and learned a lot from.) Some of the optimization techniques were Roc-specific, though, and I wanted to mention them here as a reference for others. Even if you're not writing Roc code, this can also give you an idea of how Roc achieves its performance.
\nOne of Roc's most effective optimizations is opportunistic mutation (which Koka calls "Functional but In-Place"). You can learn how it works in Roc's docs on opportunistic mutation, but from a performance optimization perspective, it's critically important to avoid unnecessary sharing so that values remain unique and therefore eligible for in-place mutation.
\nYou can't opt out of bounds checks (e.g. getting a value out of a dynamically-sized collection) in Roc like you can in C, Zig, and Rust, but Roc's compiler can optimize them away if it knows they'd be guaranteed to pass. Bounds checks are most costly inside loops, so you can get a lot of mileage out of adding something like if collection.len() <= highest_index_used { return 0 } before the loop starts. This adds an additional check that looks redundant, but it can allow the optimizer to safely delete costlier checks inside the loop, once it can tell that failures would not be reachable after that early return.
Roc's compile-time evaluation can be used to assemble lookup tables at compile time as long as nothing depends on runtime inputs. Even better, since their lengths are known at compile time, you can do things like Ok(entry) = lookup_table.get(5) to do an exhaustive pattern match that Roc's compiler will know is valid thanks to compile-time evaluation. This avoids runtime bounds checks that non-constant tables might have to pay for.
In Roc, a + b desugars to a.plus(b). For an integer like U64, U64.plus includes an overflow check, whereas plus_wrap does not. Although plus is the safer default, overflow checks have a runtime cost, as CPUs can do wrapping integer arithmetic in one instruction. When Roc's compiler can prove that overflow won't occur, both plus and plus_wrap will optimize to the same instructions, but when it can't, you may get a speedup by choosing plus_wrap.
I'm sure there are still opportunities to improve roc-deflate's performance—especially on x64 CPUs—as well as the optimization pipeline in Roc's compiler.
\nAt this point I feel reasonably confident that all the low-hanging fruit has been picked, but if anyone wants to try it out for themselves, here's the script I ran to set up and run the benchmark suite on Linux. (Note that Claude Opus 5.5 generated this script, and I haven't vetted it for security; run it outside a sandbox at your own risk!) It pins both Roc compiler revisions and roc-deflate versions, so you'll want to adjust those commits if you're trying out potential improvements.
\nFor my part, I'll be focusing on the upcoming Roc 0.1.0 release from now until it's out. It's been eight years in the making, so I'll be very happy for us to finally have a numbered release! If you'd like to learn more about Roc, check out roc-lang.org or come say hi on Roc Zulip!
","headings":[{"level":1,"text":"How close can pure FP get to C?","id":"how-close-can-pure-fp-get-to-c"},{"level":2,"text":"Under 10% Slower","id":"under-10-slower"},{"level":3,"text":"Compression on Silesia corpus","id":"compression-on-silesia-corpus"},{"level":2,"text":"What the Performance Gap Buys","id":"what-the-performance-gap-buys"},{"level":2,"text":"Roc-Specific Optimizations","id":"roc-specific-optimizations"},{"level":3,"text":"Avoid Unnecessary Sharing","id":"avoid-unnecessary-sharing"},{"level":3,"text":"Hoist Bounds Checks","id":"hoist-bounds-checks"},{"level":3,"text":"Build Lookup Tables at Compile Time","id":"build-lookup-tables-at-compile-time"},{"level":3,"text":"Consider Wrapping Arithmetic","id":"consider-wrapping-arithmetic"},{"level":2,"text":"Room for Improvement","id":"room-for-improvement"}]}}