Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Decompression of level 2 frames

Decoder overview · Benchmark index

Recorded run: long-20261109 (2026-11-09). CSV · Environment · Methodology and limits.

Level belongs to the C encoder that prepared the frame; decoders have no level setting. Both decoders restore the same bytes.

Median speed across 8 datasets

Median of per-dataset C mean time / decoder mean time ratios, with equal weight. Higher is faster; 1× matches C. No confidence interval is inferred for this across-dataset median.

DecoderMedian speed / C
zstd C1.000×
mzstd1.611×

Datasets are ranked by mzstd speed relative to C, highest first. Throughput counts restored bytes; empty input has latency only. Whiskers and intervals describe sampling uncertainty, not all host variation. Cold APIs include allocation and disposal. C reads the frame’s content size and decodes into a buffer of exactly that size; mzstd returns its own vector.

empty

Empty input; measures API overhead.

9 compressed → 0 restored bytes; compressed / original: undefined.

DecoderMean µs95% mean interval µsDecoded MiB/sSpeed / C
zstd C2.0822.022–2.148—1.000×
mzstd0.044310.04299–0.04565—46.989×

tiny-text

44-byte quick-brown-fox sentence.

53 compressed → 44 restored bytes; compressed / original: 120.455%.

DecoderMean µs95% mean interval µsDecoded MiB/sSpeed / C
zstd C1.9451.932–1.9621.571.000×
mzstd0.053490.05118–0.056784.4436.360×

random-64k

64 KiB prefix of the deterministic pseudorandom corpus.

65,546 compressed → 65,536 restored bytes; compressed / original: 100.015%.

DecoderMean µs95% mean interval µsDecoded MiB/sSpeed / C
zstd C3.3633.283–3.44318,584.971.000×
mzstd0.97580.9592–0.992764,049.713.446×

binary-64k

64 KiB of structured binary bytes: ((i / 16) XOR i) modulo 256.

2,425 compressed → 65,536 restored bytes; compressed / original: 3.700%.

DecoderMean µs95% mean interval µsDecoded MiB/sSpeed / C
zstd C5.1044.896–5.32512,245.561.000×
mzstd3.0982.989–3.21420,173.511.647×

repeated-1m

1 MiB of repeated a bytes.

50 compressed → 1,048,576 restored bytes; compressed / original: 0.005%.

DecoderMean µs95% mean interval µsDecoded MiB/sSpeed / C
zstd C14.4814.15–14.8569,074.211.000×
mzstd9.1889.079–9.313108,831.921.576×

random-1m

1 MiB of deterministic xorshift64 pseudorandom bytes.

1,048,609 compressed → 1,048,576 restored bytes; compressed / original: 100.003%.

DecoderMean µs95% mean interval µsDecoded MiB/sSpeed / C
zstd C19.4619.24–19.751,379.361.000×
mzstd16.8816.74–17.0559,231.561.153×

alice29

Alice in Wonderland text from the Canterbury corpus (testdata/alice29.txt).

57,348 compressed → 152,089 restored bytes; compressed / original: 37.707%.

DecoderMean µs95% mean interval µsDecoded MiB/sSpeed / C
zstd C93.9492.21–96.141,543.921.000×
mzstd105.3101–110.61,377.900.892×

text-1m

Alice text repeated cyclically to 1 MiB.

58,340 compressed → 1,048,576 restored bytes; compressed / original: 5.564%.

DecoderMean µs95% mean interval µsDecoded MiB/sSpeed / C
zstd C113.5109.8–117.68,812.281.000×
mzstd134.7128.5–141.37,423.310.842×