Evidence · FSL

fslmaths

fslmaths is FSL’s image calculator: it reads an image, applies a chain of operations from left to right and writes the result. fMRIPrep calls it in every BOLD reference’s brain mask, and fourteen times in its optional goodvoxels mask. The replica (larmorx-nc) gives FSL 6.0.7.23’s files, messages and exit status byte for byte in 360 of its 361 compared cases, out of 422, fMRIPrep’s calls among them; the other is one unit in the last place where glibc’s tan is not correctly rounded. The clean room (larmorx) is within its record’s thresholds in 386 of its 391 compared cases, out of 454, and gives FSL’s values in every file in 370 of them, fMRIPrep’s calls among them; the other 5 are expected divergences, each with its reason. On a real series it differs from FSL in one call of fifteen, a large kernel FSL convolves by its single-precision FFT, where its own sums lie closer to the exact result. On the heavy operations of a real BOLD series and T1w it runs 1.8 to 28 times as fast as FSL on one thread, up to 268 times on 14.

Ports
  • replica larmorx-nc FSL Licence: non-commercial use only: translated from FSL 6.0.7.23's source (fsl-avwutils 2209.6, with fsl-newimage 2601.0's image functions, fsl-miscmaths 2412.6 and fsl-cprob 2111.0); it runs as a separate program
  • clean room larmorx Apache-2.0: crate larmorx-fsl, written from specs/fslmaths.md, FSL's usage text and documentation and runs of FSL 6.0.7.23, without FSL's source (2026-11-03 and 04)
Status
replica experimental; clean room experimental
Records
the replica's: standard comparison (FSL 6.0.7.23) · the clean room's: standard comparison (FSL 6.0.7.23) · benchmark fsl-maths
Reference
fslmaths · API page fsl-maths · its companion: fslstats

What is compared

fslmaths <input> [operations] <output> reads an image, applies each operation in turn to the current image and writes the result. Its operations:

  • arithmetic with a number or another image: -add, -sub, -mul, -div, -rem, -mas, -max, -min;
  • thresholds by value or by percentage of the robust range: -thr, -thrp, -thrP and their upper forms;
  • voxelwise functions: -exp, -log, the trigonometric functions, -sqrt, -abs, -bin, -nan, -index, -grid, seeded noise;
  • kernels and the operations that use them: dilations and erosions, mean and median filters, Gaussian smoothing, convolutions by direct sums or by the FFT, subsampling, edges, hole filling;
  • mean normalisation, and reductions along a dimension: -Tmean, -Tstd, -Tmax, -Tmedian, -Tperc, -Tar1 and their X, Y and Z forms;
  • p-values, ranks and z ↔ p conversions, regions;
  • band-pass temporal filtering (-bptf), the threshold-free cluster enhancement (-tfce, -tfceS), ROC curves (-roc) and the diffusion tensor’s decomposition.

-dt chooses the type it computes in (float by default, or double), -odt the output’s.

fMRIPrep calls it in niworkflows’ BOLD reference workflows (combine_masks: fslmaths <mask> -mul <mask> <out>, the reference’s brain mask as the product of two masks). With --project-goodvoxels it calls it fourteen times in its goodvoxels mask, with four fslstats calls between them.

The two ports run beside FSL 6.0.7.23’s fslmaths on the same files with the same arguments. fMRIPrep’s own outputs are the files it wrote in its runs, with fslmaths in fMRIPrep 25.2.6’s image.

bash
larmorx fsl fslmaths mask_a.nii.gz -mul mask_b.nii.gz combined.nii.gz
larmorx fsl fslmaths bold.nii.gz -Tstd -div mean.nii.gz cov.nii.gz

The first multiplies two masks, as fMRIPrep does. The second divides a series’ standard deviation over time by its mean.

python
import larmorx as lx

combined = lx.fsl.fslmaths("mask_a.nii.gz", "-mul", "mask_b.nii.gz")   # the command line's operations
smooth = lx.fsl.smooth("mean.nii.gz", sigma=3.0)                       # one typed wrapper per family

The calls above leave implementation at "auto". The API page describes the generic function and the typed wrapper of every operation family. How to read what follows: how to read the evidence.

Its two ports

FSL comes under the FSL Licence, which allows non-commercial use only. So LARMORX has two ports of its FSL programs. The replica is translated from FSL’s source and lives in its own package, larmorx-nc (pip install larmorx-nc). The clean room is in larmorx itself. It is written without FSL’s code, from fslmaths’ behavioural specification (specs/fslmaths.md), FSL’s usage text and documentation and runs of FSL 6.0.7.23 as a black box.

implementation="auto" | "replica" | "clean-room" chooses between them ("original" in 0.1.0rc1). On the command line, larmorx --implementation clean-room fsl fslmaths ... does the same from the next release; in 0.1.0rc1 it is the environment variable LARMORX_IMPLEMENTATION. auto, and larmorx fsl fslmaths, take the replica when larmorx-nc is installed, else the clean room.

The replica follows FSL 6.0.7.23. It needs no FSL installation. What fslmaths needs from FSL’s libraries is translated once into larmorx-nc’s shared core, beside what mcflirt, BET, FLIRT, FAST and topup use: newimage’s image functions, kernels, morphology, convolutions, connected components and TFCE, the Cephes probability functions of fsl-cprob, Armadillo’s Kiss FFT and symmetric eigen-decomposition. BET’s own fslmaths and fslstats calls run the same code.

The clean room has every operation of the usage text and the hidden ones, every data type in and out, every message, exit status and failure case, and the defects with their switches. By default it computes exp, log, the trigonometric functions and pow correctly rounded on every platform; in reproduce it computes them as glibc 2.35 does, but for tan, asin, acos and atan (below). It is built for speed, and its result is the same bits at any thread count (speed):

  • the separable filters and the direct sums run a row of voxels at a time, -bptf in blocks of voxels, the reductions per column;
  • every parallel loop sums in a fixed order;
  • it holds one copy of the image: the input’s values are moved into the computation, the output’s to the writer;
  • where FSL convolves a large kernel by its FFT, the clean room takes the direct sum too, in the orientation of FSL’s direct sum (FSL-192). It is faster than FSL’s FFT on every kernel measured (speed).

Where the specification leaves a convention open, the clean room follows FSL’s as black-box runs measured it:

  • the Gaussian’s exponent, i² (d d) per axis;
  • the geometry of a new grid;
  • -roc’s candidates;
  • TFCE’s checks, in FSL’s order.

The C library’s last digits (since 2026-11-04). fslmaths’ double-precision -sin, -cos, -exp, -log and -pow come from the C library, glibc on Linux, and so do the weights of its Gaussian filters (-s, -kernel gauss, -dog_edge), of -bptf, -tfce and -roc. glibc’s results are now and then one unit in the last place away from the exact value rounded once. LARMORX computes them correctly rounded, the same on every platform.

  • Until 2026-11-04 the clean room did so in fix and in reproduce. Since then its reproduce computes them exactly as glibc 2.35 does, with LARMORX’s versions of glibc’s functions (larmorx_core::math).
  • tan, asin, acos and atan stay correctly rounded in fix and in reproduce (as decided on 2026-11-04).
  • Ten cases of its own show what that changes: nine orig/glibc-*, their inputs chosen where glibc’s last bit differs, and orig/edge-double, -edge on a double image, whose squares FSL computes as products, black-box runs showed. Each now gives FSL’s values bit for bit. No other number of the record moved, as a float result rounds such a last place away.
  • Its default, fix, names GLIBC-07 exactly where a voxelwise function’s value differs from glibc’s. The filters name nothing: a filter would have to run twice to know (GLIBC-07 below).

The verdicts at a glance

The replica is a translation of FSL’s source, so the question for it is whether it gives FSL’s bytes: every file fslmaths writes (each image’s header and data), its standard output and error, and its exit status.

Agreement, replica against the original: FSL fslmaths 6.0.7.23 (standard comparison, 422 cases of the record), by category and verdict
fMRIPrep's calls18
the command line24
reading images5
data types20
defects of FSL82
arithmetic with numbers and images25
thresholds12
voxelwise functions37
morphology24
-dilall4
kernels and filters29
subsampling7
-edge6
-dog_edge5
-fillh5
intensity normalisation6
-tensor_decomp4
reductions along a dimension38
p-values, ranks and z16
-roi7
-bptf12
-tfce15
-tfceS8
-roc8
random testing5

Select a segment to list its cases below.

The replica’s record runs 422 cases against FSL 6.0.7.23 (the record, made again on 2026-11-04). 360 give FSL’s files byte for byte. In 61 FSL refuses the input or stops, and the replica does the same, with the same message and exit status (rejected by both). 1 differs by one unit in the last place of a few values, where FSL’s result comes from glibc 2.35’s tan, which is not correctly rounded, and the replica’s is (GLIBC-07, expected divergence, below).

The cases cover:

  • fMRIPrep’s calls: the four combine_masks calls of three fMRIPrep runs and the goodvoxels mask’s fourteen;
  • every operation and option of the command line, in every data type;
  • the odd headers and sizes, and FSL’s own failure cases;
  • the defects LARMORX corrects by default. Each case also runs the replica’s default, fix, which names every correction where it changes the result (below).

Since 2026-11-04 the replica’s reproduce computes glibc’s double sin and cos, and the twiddle factors of fslmaths’ FFT, exactly as glibc 2.35 does. It does so with LARMORX’s own version of glibc’s sin, cos and sincos, written from a behavioural specification of glibc’s code (specs/glibc-dbl.md). Its default stays correctly rounded and names GLIBC-07 wherever a voxelwise value differs. The FFT case of random testing, an expected divergence until then, is FSL’s bit for bit since.

The same day two mistakes of the replica’s in the sign of a NaN were fixed, each kept by the record:

  • -ztop of a NaN voxel. The replica’s expx2 was written as plain arithmetic, whose NaN signs the optimiser does not keep; it now takes x86-64’s rules explicitly. Four cases of the record had found it.
  • -dilall of an image holding NaN and infinities. FSL’s build passes on the neighbour’s NaN, the replica passed on the sum’s. Random testing found it: the permanent case difftest/dilall-nan-order.

The clean room is written without FSL’s source, so its record holds it to the thresholds of specs/fslmaths.md §21 instead:

  • FSL’s exit status, messages, file names and headers identical;
  • the values bit for bit, the double-precision functions within one unit in the last place per function;
  • the sums within 10−5 of FSL’s largest value, with 99 % of the voxels within 4 units in the last place;
  • TFCE, -tensor_decomp and -roc by their own measures.

It also shows how close the clean room comes: each case’s verdict is read from what the record measures.

Agreement, clean room against the original: FSL fslmaths 6.0.7.23 (standard comparison, 454 cases of the record), by category and verdict
fMRIPrep's calls18
the command line24
reading images5
data types20
defects of FSL82
arithmetic with numbers and images25
thresholds12
voxelwise functions37
morphology24
-dilall4
kernels and filters29
subsampling7
-edge6
-dog_edge5
-fillh5
intensity normalisation6
-tensor_decomp4
reductions along a dimension38
p-values, ranks and z16
-roi7
-bptf12
-tfce15
-tfceS8
-roc8
the clean room's own cases18
random testing19

Select a segment to list its cases below.

The clean room’s record runs 454 cases against FSL 6.0.7.23 (the record, made again on 2026-11-07): the replica suite’s cases but its five random-testing cases, 18 of its own (orig/, ten of them since 2026-11-04) and 19 permanent cases of random testing. Two of those came on 2026-11-07: a series whose time step is 0 or negative, which FSL’s image layer reads as its magnitude, and as 1 for a time step of 0, and which every output keeps. The record counts 386 within its thresholds, 63 rejected by both alike and 5 expected divergences. By what each case measures:

  • Bit-identical, 370. FSL’s values in every file, with FSL’s headers, messages and exit status. fMRIPrep’s four combine_masks calls and the goodvoxels mask’s fourteen are among them, and so is the -s 5 at 3 × 3 × 3.3 mm.
  • Open, 15. They are within the thresholds without being bit for bit, each with its reason in the case explorer:
    • FSL’s FFT path on four kernels;
    • -subsamp2 and -subsamp2offc under the floating types (FSL’s arithmetic there is a question for the specification);
    • the double-precision -ztop and -ranknorm (the last bits of Φ and Φ−1);
    • -tensor_decomp’s eigenvectors, the same up to their sign: FSL’s signs are LAPACK’s, the clean room’s make each one’s largest component positive;
    • one -roc area, 10−7 from FSL’s.
  • Expected divergences, 6. FSL’s FFT path on 3 kernels and -subsamp2offc’s means on 2 are within the sums’ bound but not its share of voxels within 4 units in the last place: the record’s 5. The sixth is -dt double -tan (GLIBC-07, below), which the record holds within its threshold of one unit per function.

On the FFT cases the clean room lies as close to the exact result as FSL’s single-precision FFT, or closer (findings 31). Before its double sums of 2026-11-04 two more of them were expected divergences, up to 2.6 × 10−6 of FSL’s largest value from FSL.

In fMRIPrep: a BOLD reference’s brain mask

fMRIPrep’s combine_masks node of ds005040’s head-motion reference: the mask of BET’s first pass on the reference, times the template’s brain mask brought onto it. It was given to FSL 6.0.7.23, to the replica in reproduce and in fix, and to the clean room in reproduce at 14 threads and at one and in fix. Each mask is drawn over the reference it was computed from. Below the viewer: every combine_masks call of both records, on ds000005, ds005040 (both kinds of reference) and ds003592. Each gives the same bytes as FSL’s and as fMRIPrep’s own output, in both ports.

Worked example

fMRIPrep's brain mask of a BOLD reference: two masks combined

What this shows: On fMRIPrep’s own call, both ports give FSL’s mask, every value, and the same values as fMRIPrep’s own output; the clean room the same at 1 and 14 threads. So do the records’ four combine_masks calls on three fMRIPrep runs (the table below the viewer). No correction by default changes this call, in either port.

The combine_masks node fMRIPrep 25.2.6 ran on ds005040’s head-motion reference: niworkflows multiplies the mask of BET’s first pass by the template’s brain mask brought onto the reference, and the product is the reference’s brain mask. Given to FSL 6.0.7.23, to the replica (larmorx-nc) in reproduce and in fix, and to LARMORX’s clean room in reproduce at 14 threads and at one and in fix, its default; each mask is drawn over the reference it was computed from. Below the viewer, every combine_masks call of both records, three fMRIPrep runs.

Data
OpenNeuro ds005040, sub-002 (CC0-1.0): fMRIPrep’s BOLD reference and its two masks from the run’s working directory; shown inside the union of the masks, dilated by 2 voxels, so no face shows
Node
fMRIPrep 25.2.6, work/fmriprep_25_2_wf/sub_002_wf/bold_task_clip_run_1_wf/bold_fit_wf/enhance_and_skullstrip_bold_wf/combine_masks
Command
fslmaths <BET's first-pass mask> -mul <the template's mask> <out>
Programs
FSL 6.0.7.23 (fsl-avwutils 2209.6, fsl-newimage 2601.0; ~/fsl-6.0.7.23, never on the PATH) (served from the cache of the original’s runs, as the record ran it); larmorx-nc 0.0.1 (the replica), from LARMORX at commit 73451f78a400; LARMORX 0.0.1 (the clean rooms, crate larmorx-fsl, Apache-2.0), from LARMORX at commit 80cc208fa4a3
Grid
64 × 64 × 40 voxels of 3 × 3 × 3.3 mm; shown: 55 × 63 × 40 around the brain
InputMask A: BET's first passskullstrip_first_pass (bet -f 0.2 -m), float32
InputMask B: the template's brain maskmap_brainmask: the template's mask on the reference (fixhdr_skullstrip2), uint8
OriginalFSL fslmaths 6.0.7.23the conda installation, as built (through the cache of the original's runs)
Replicareplicalarmorx-nc, FSL Licence · reproduce · against FSL fslmaths 6.0.7.23
Clean roomclean roomlarmorx, Apache-2.0 · reproduce, 14 threads · against FSL fslmaths 6.0.7.23

Bit-identical163,840 of 163,840 voxels equal

the replica against the originalreplica − FSL fslmaths 6.0.7.23

Bit-identical163,840 of 163,840 voxels equal

the clean room against the originalclean room − FSL fslmaths 6.0.7.23

Axial slice 19 of 40; each mask drawn over the BOLD reference it was computed from. The interactive viewer needs JavaScript; the metrics below are complete without it.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The replica against the original and fMRIPrep's output
replica − FSL fslmaths 6.0.7.23Bit-identicalThe same values, every one, in every output fslmaths writes (the record compares every file, the messages and the exit status too).identical: 0 of 163,840 values differ
replica − fMRIPrep's outputBit-identicalfMRIPrep’s own file (fslmaths in fMRIPrep 25.2.6’s image) holds the same values.identical: 0 of 163,840 values differ
The replica against itself: fix and reproduce
replica, fix − reproduceBit-identicalNo correction of the register applies to these calls: fix and reproduce give the same values.identical: 0 of 163,840 values differ
The clean room against the original and fMRIPrep's output
clean room − FSL fslmaths 6.0.7.23Bit-identicalThe same values, every one, in every output fslmaths writes (the record compares every file, the header, the messages and the exit status too).identical: 0 of 163,840 values differ
clean room − fMRIPrep's outputBit-identicalfMRIPrep’s own file (fslmaths in fMRIPrep 25.2.6’s image) holds the same values.identical: 0 of 163,840 values differ
The clean room against itself: thread counts, fix and reproduce
clean room, 1 thread − 14 threadsBit-identicalThe thread count changes nothing: the clean room’s parallel loops (per voxel, column, row or block, every sum in a fixed order) give its one-thread result bit for bit at 14 threads.identical: 0 of 163,840 values differ
clean room, fix − reproduceBit-identicalNo correction of the register applies to these calls: fix and reproduce give the same values.identical: 0 of 163,840 values differ

Δ is A − B, computed in float64 over all 163,840 voxels of the outputs, not only those shown. The replica’s record’s threshold: every file fslmaths writes bit-identical (images decompressed, descrip left out), with FSL’s exit status and messages. The clean room’s record (fsl-maths) holds it to specs/fslmaths.md §21: FSL’s exit status, messages, file names and headers identical, the values bit for bit (double-precision functions within one unit in the last place per function, GLIBC-07), the sums (-s, -fmean, -fmeanu, the means and standard deviations, -bptf, …) within 1e-5 of FSL’s largest value with 99 % of the voxels within 4 units in the last place.

fMRIPrep’s combine_masks calls, both ports against the original: FSL fslmaths 6.0.7.23 (the records’ four calls)

RunGridMask A voxelsMask B voxelsProduct voxelsreplica against FSL fslmaths 6.0.7.23clean room against FSL fslmaths 6.0.7.23fMRIPrep's own output
ds000005-sub-01HMC reference (BET’s first pass × the template’s mask)64 × 64 × 3436,91732,10232,069identicalidenticalthe same values
ds003592-sub-280HMC reference (BET’s first pass × the template’s mask)64 × 64 × 4348,55442,48642,464identicalidenticalthe same values
ds005040-sub-002HMC reference (BET’s first pass × the template’s mask)64 × 64 × 4053,50849,30849,253identicalidenticalthe same values
ds005040-sub-002coregistration reference (BET’s two passes)64 × 64 × 4053,36352,45651,891identicalidenticalthe same values

The record’s four calls on three fMRIPrep runs: ds000005 (single-band 3 T), ds005040 (both kinds: the head-motion reference’s mask, BET’s first pass times the template’s brain mask on the reference, and the coregistration reference’s, BET’s first pass times its second) and ds003592 (multi-echo). Voxels counted where the mask is not 0. Identical: every file fslmaths writes byte for byte (decompressed, descrip aside), with FSL’s exit status and messages (the clean room’s record: its values bit for bit and its header byte for byte); fMRIPrep’s own output is the node’s file in the run’s working directory (fslmaths in fMRIPrep 25.2.6’s image).

Provenance of this example
  • Data: OpenNeuro ds005040, sub-002, CC0-1.0
  • Inputs: BET’s first-pass mask (sha256 da8a28916643c2c6…); the template’s mask, on the reference (sha256 e07b0a17f73c28dc…); the BOLD reference after N4 (sha256 54b587437b15bd28…)
  • Runs: figures cache entry fsl-maths/fmriprep-combine-masks, computed 2026-11-04T20:40:54+00:00 with larmorx 73451f78a400; FSL 6.0.7.23 (fsl-avwutils 2209.6, fsl-newimage 2601.0; ~/fsl-6.0.7.23, never on the PATH); larmorx-nc 0.0.1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • Runs: figures cache entry fsl-maths/original-fmriprep-combine-masks, computed 2026-11-07T02:06:27+00:00 with larmorx 80cc208fa4a3; FSL 6.0.7.23 (fsl-avwutils 2209.6, fsl-newimage 2601.0; ~/fsl-6.0.7.23, never on the PATH); “larmorx 0.0.1 (the clean-room originals, crate larmorx-fsl, Apache-2.0)”, as the cache entry records it; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in fsl-maths.json, exported by python -m larmorx_validation evidence fsl-maths.

In fMRIPrep: the goodvoxels mask

With --project-goodvoxels, fMRIPrep leaves out of the surface projection the voxels whose temporal coefficient of variation is high compared with their neighbourhood in the cortical ribbon. It takes fourteen fslmaths calls and four fslstats calls (init_goodvoxels_bold_mask_wf). The fMRIPrep runs do not use the option, so the workflow is built from fMRIPrep’s code on ds005040’s BOLD series in T1w space. Each call has its arguments as the workflow makes them, and its inputs are FSL’s outputs of the calls before it. The runs have no FreeSurfer reconstruction, so the cortical ribbon is a stand-in: the anatomical segmentation’s grey matter at each BOLD voxel.

Every call’s output is FSL’s bit for bit, in the replica and in the clean room at 14 threads and at one. So are the four numbers fslstats prints between them (fslstats’ page).

Worked example

fMRIPrep's goodvoxels mask: fourteen fslmaths calls

What this shows: Each of the fourteen calls of fMRIPrep’s goodvoxels mask gives FSL’s output, every value, in both ports, and the clean room the same at 1 and 14 threads; the four numbers fslstats prints between them are FSL’s text in both (the table below the viewer). No correction by default changes these calls, in either port.

fMRIPrep 25.2.6’s goodvoxels workflow (--project-goodvoxels, used before projecting a BOLD series onto the cortical surface; not in the fMRIPrep runs) built from its code on ds005040’s BOLD series in T1w space: each fslmaths node with its arguments as the workflow makes them, in its order, its inputs FSL’s outputs of the nodes before it, the numbers between them printed by fslstats. The runs have no FreeSurfer ribbon, so a stand-in takes its place: the anatomical segmentation’s grey matter. Choose a node above the panels; each was given to FSL 6.0.7.23, to the replica (larmorx-nc) in reproduce and in fix, and to LARMORX’s clean room in reproduce at 14 threads and at one and in fix, its default.

Data
OpenNeuro ds005040, sub-002 (CC0-1.0): fMRIPrep 25.2.6’s preprocessed BOLD series in T1w space (out/); shown inside fMRIPrep’s brain mask of the series, dilated by 2 voxels, so no face shows
Workflow
fMRIPrep 25.2.6, init_goodvoxels_bold_mask_wf (fmriprep/workflows/bold/resampling.py)
Ribbon
a stand-in: the segmentation’s grey matter (label 1) at each BOLD voxel’s centre; the no-reconall fMRIPrep runs have no FreeSurfer ribbon, so the mask differs from the one fMRIPrep would make with recon-all
Numbers
the ribbon’s mean CoV 0.010577, the modulated CoV’s mean 0.95208 and standard deviation 3.6842: the threshold 2.7941800000 (fslstats -M and -S, as the table below the viewer shows)
Programs
FSL 6.0.7.23 (fsl-avwutils 2209.6, fsl-newimage 2601.0; ~/fsl-6.0.7.23, never on the PATH) (served from the cache of the original’s runs, as the record ran it); larmorx-nc 0.0.1 (the replica), from LARMORX at commit 73451f78a400; LARMORX 0.0.1 (the clean rooms, crate larmorx-fsl, Apache-2.0), from LARMORX at commit 80cc208fa4a3
Grid
50 × 62 × 44 × 63 voxels of 3 × 3 × 3.3 mm; shown: 50 × 62 × 43 around the brain
Inputits mean over timeFSL's -Tmean (the node mean_volume): the image the masks are drawn over
Inputthe ribbon's stand-inthe segmentation's grey matter at each BOLD voxel (uint8): the no-reconall fMRIPrep runs have no FreeSurfer ribbon
OriginalFSL fslmaths 6.0.7.23the conda installation, as built (through the cache of the original's runs)
Replicareplicalarmorx-nc, FSL Licence · reproduce · against FSL fslmaths 6.0.7.23
Clean roomclean roomlarmorx, Apache-2.0 · reproduce, 14 threads · against FSL fslmaths 6.0.7.23

Bit-identical136,400 of 136,400 voxels equal

the replica against the originalreplica − FSL fslmaths 6.0.7.23

Bit-identical136,400 of 136,400 voxels equal

the clean room against the originalclean room − FSL fslmaths 6.0.7.23

Axial slice 24 of 43; the final mask, good voxels inside the ribbon, drawn over the series' mean. The interactive viewer needs JavaScript; the metrics below are complete without it.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonOutputmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The replica against the original
replica − FSL fslmaths 6.0.7.23Bit-identicalThe same values, every one, in every output fslmaths writes (the record compares every file, the messages and the exit status too).every outputidentical: 0 of 136,400 values differ
The replica against itself: fix and reproduce
replica, fix − reproduceBit-identicalNo correction of the register applies to these calls: fix and reproduce give the same values.every outputidentical: 0 of 136,400 values differ
The clean room against the original
clean room − FSL fslmaths 6.0.7.23Bit-identicalThe same values, every one, in every output of the fourteen calls (the record compares every file, the header, the messages and the exit status too).every outputidentical: 0 of 136,400 values differ
The clean room against itself: thread counts, fix and reproduce
clean room, 1 thread − 14 threadsBit-identicalThe thread count changes nothing: the clean room’s parallel loops (per voxel, column, row or block, every sum in a fixed order) give its one-thread result bit for bit at 14 threads.every outputidentical: 0 of 136,400 values differ
clean room, fix − reproduceBit-identicalNo correction of the register applies to these calls: fix and reproduce give the same values.every outputidentical: 0 of 136,400 values differ

Δ is A − B, computed in float64 over all 136,400 voxels of the outputs, not only those shown. The replica’s record’s threshold: every file fslmaths writes bit-identical (images decompressed, descrip left out), with FSL’s exit status and messages. The clean room’s record (fsl-maths) holds it to specs/fslmaths.md §21: FSL’s exit status, messages, file names and headers identical, the values bit for bit (double-precision functions within one unit in the last place per function, GLIBC-07), the sums (-s, -fmean, -fmeanu, the means and standard deviations, -bptf, …) within 1e-5 of FSL’s largest value with 99 % of the voxels within 4 units in the last place.

The numbers between the steps, both ports against the original: FSL fslstats 6.0.7.23 (its four calls)

NodeFSL fslstats 6.0.7.23 printsThe replica printsThe clean room printsAs read
cov_ribbon_meanfslstats <cov_ribbon> -M0.0105770.0105770.0105770.010577
cov_ribbon_stdfslstats <cov_ribbon> -S0.9990390.9990390.9990390.999039
mod_ribbon_meanfslstats <cov_norm_modulate_ribbon> -M0.9520800.9520800.9520800.95208
mod_ribbon_stdfslstats <cov_norm_modulate_ribbon> -S3.6842003.6842003.6842003.6842
upper_threshmod_ribbon_mean + mod_ribbon_std × 0.5 (fMRIPrep’s Python), given to -thr2.79418000002.79418000002.79418000002.79418

Each fslstats call, by FSL on its output of the node before it (through the cache of the original’s runs), and by the replica (larmorx-nc) and the clean room (reproduce, 14 threads) on the same file: the printed text compared byte for byte. fMRIPrep reads each number as Python’s float, divides by the means and passes the threshold to fslmaths as nipype formats them (%.8f, %.10f); fslstats’ evidence page has the four calls in its record.

Provenance of this example
  • Data: OpenNeuro ds005040, sub-002, CC0-1.0
  • Inputs: BOLD series in T1w space (sha256 ebfbc7b1648ca221…); the ribbon’s stand-in (sha256 788750569ea12c93…)
  • Runs: figures cache entry fsl-maths/fmriprep-goodvoxels, computed 2026-11-04T20:40:56+00:00 with larmorx 73451f78a400; FSL 6.0.7.23 (fsl-avwutils 2209.6, fsl-newimage 2601.0; ~/fsl-6.0.7.23, never on the PATH); larmorx-nc 0.0.1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • Runs: figures cache entry fsl-maths/original-fmriprep-goodvoxels, computed 2026-11-07T02:06:29+00:00 with larmorx 80cc208fa4a3; FSL 6.0.7.23 (fsl-avwutils 2209.6, fsl-newimage 2601.0; ~/fsl-6.0.7.23, never on the PATH); “larmorx 0.0.1 (the clean-room originals, crate larmorx-fsl, Apache-2.0)”, as the cache entry records it; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in fsl-maths.json, exported by python -m larmorx_validation evidence fsl-maths.

One call per operation family

fslmaths’ operation families on the same real series, one call each: arithmetic with images (the temporal SNR), a threshold by the robust range, a voxelwise function, reductions along time, morphology, the three kinds of filters, edges, normalisation, temporal filtering, a region and the threshold-free cluster enhancement. The filters are Gaussian smoothing by separable sums, a mean filter whose large spherical kernel fslmaths convolves by the FFT, and a median filter. Every output of the replica is FSL’s bit for bit, in fix and in reproduce.

The clean room gives FSL’s values in fourteen of the fifteen calls, at 14 threads and at one, in fix and in reproduce. The fifteenth is the mean filter with the 15 × 15 × 13 spherical kernel (1,313 voxels):

  • FSL convolves it by its FFT in single precision. The clean room takes a direct sum: it adds the products in double precision and rounds the sum once. (Where FSL sums directly, the clean room keeps FSL’s single-precision sum, bit for bit.)
  • The two differ by up to 2 × 10−7 of FSL’s largest value, within §21’s bound of 10−5, with 75.7 % of the voxels within 4 units in the last place, below §21’s 99 % (expected divergence, as the record’s FFT cases).
  • Against the exact result, computed here in float64 from the specification’s definition, FSL’s FFT lies within 1.5 × 10−7 of the largest value and the clean room within 7.9 × 10−8. FSL’s FFT rounds relative to the image’s largest values, so its small values sit many units in their last place from the exact result, where the clean room’s sum is closer (findings 31).
  • Until 2026-11-04 the clean room added each product to a single-precision sum there too, as FSL does on its direct path. Over 1,313 terms that sum drifted, 1.4 × 10−5 from the exact result and from FSL, beyond §21’s bound. The double sum costs the benchmark’s two FFT-path calls 1.3 and 1.5 times the time.

Choose filters: the FFT convolution above the panels to see where they differ. The second table below the viewer gives each call’s numbers.

Worked example

One call per operation family, on a real series

What this shows: The replica gives FSL’s output in all 15 calls, every value. The clean room gives FSL’s values in 14 of them, at 1 and 14 threads. The other is the mean filter with the 15 × 15 × 13 spherical kernel (1,313 voxels), which FSL convolves by its FFT in single precision and the clean room by a direct sum in double precision: they differ by up to 2 × 10⁻⁷ of FSL’s largest value, within the bound of 10⁻⁵, with 75.7 % of the voxels within 4 units in the last place, below the 99 % asked (an expected divergence). Against the exact result FSL’s FFT lies within 1.5 × 10⁻⁷ of the largest value and the clean room within 7.9 × 10⁻⁸. No correction by default changes these calls, in either port.

fslmaths’ operation families, one call each, on ds005040’s BOLD series in T1w space (the goodvoxels workflow’s input), FSL’s mean and standard deviation of it, and fMRIPrep’s brain mask: arithmetic with images, the thresholds, the voxelwise functions, the reductions along time, morphology, the convolutions (separable, by the FFT, the median), edges, normalisation, temporal filtering, regions and the threshold-free cluster enhancement. Choose a call above the panels; each was given to FSL 6.0.7.23, to the replica (larmorx-nc) in reproduce and in fix, and to LARMORX’s clean room in reproduce at 14 threads and at one and in fix, its default.

Data
OpenNeuro ds005040, sub-002 (CC0-1.0): fMRIPrep 25.2.6’s preprocessed BOLD series in T1w space and its brain mask (out/); shown inside that mask, dilated by 2 voxels, so no face shows
Programs
FSL 6.0.7.23 (fsl-avwutils 2209.6, fsl-newimage 2601.0; ~/fsl-6.0.7.23, never on the PATH) (served from the cache of the original’s runs, as the record ran it); larmorx-nc 0.0.1 (the replica), from LARMORX at commit 73451f78a400; LARMORX 0.0.1 (the clean rooms, crate larmorx-fsl, Apache-2.0), from LARMORX at commit 80cc208fa4a3
Grid
50 × 62 × 44 voxels of 3 × 3 × 3.3 mm; shown: 50 × 62 × 43 around the brain
Inputds005040 BOLD series' mean over timeFSL's -Tmean of fMRIPrep's series in T1w space (float32); the series itself is the input of the calls along time
OriginalFSL fslmaths 6.0.7.23the conda installation, as built (through the cache of the original's runs)
Replicareplicalarmorx-nc, FSL Licence · reproduce · against FSL fslmaths 6.0.7.23
Replicareplicalarmorx-nc, FSL Licence · fix (the default)
Clean roomclean roomlarmorx, Apache-2.0 · reproduce, 14 threads · against FSL fslmaths 6.0.7.23

Bit-identical136,400 of 136,400 voxels equal

the replica against the originalreplica − FSL fslmaths 6.0.7.23

Bit-identical136,400 of 136,400 voxels equal

the clean room against the originalclean room − FSL fslmaths 6.0.7.23

Axial slice 24 of 43; choose another call above the panels. The interactive viewer needs JavaScript; the metrics below are complete without it.

Every comparison, computed in float64 over every voxel of the outputs

Comparison (A − B), verdict and reasonOutputmax |Δ|Values that differRMS of Δmax |Δ| ÷ range1 − NCC
The replica against the original
replica − FSL fslmaths 6.0.7.23Bit-identicalThe same values, every one, in every output fslmaths writes (the record compares every file, the messages and the exit status too).every outputidentical: 0 of 136,400 values differ
The replica against itself: fix and reproduce
replica, fix − reproduceBit-identicalNo correction of the register applies to these calls: fix and reproduce give the same values.every outputidentical: 0 of 136,400 values differ
The clean room against the original
clean room − FSL fslmaths 6.0.7.23Expected divergence14 of 15 outputs FSL’s bit for bit; mean.nii.gz -kernel sphere 21 -fmean: max |Δ| 2e-07 of FSL’s largest value (§21: at most 1e-5), 75.7% of the voxels within 4 units in the last place (§21: at least 99 %): within §21’s bound and below its share of voxels within 4 units in the last place, an expected divergence as the record’s on FSL’s FFT path: FSL convolves this kernel by its FFT in single precision, whose rounding is relative to the image’s largest values; the clean room by a direct sum whose products it adds in double precision, the sum rounded once to single precision (findings 8 and 31; spec question 14); against the exact result (computed in float64 from §10.2’s definition; 15 × 15 × 13 kernel of 1313 voxels) FSL’s FFT lies within 1.5e-07 of the largest value (median 1.6e-08), the clean room’s direct sum (its products in double precision, rounded once) within 7.9e-08 (median 3.4e-09): the clean room is the closer of the two (findings 31). findingsbinary operations with images: temporal SNR: mean ÷ std, masked00 of 136,400000
thresholds (the robust range): -thrP 30 -bin: above 30 % of the robust range00 of 136,400000
voxelwise functions: -abs -sqrt00 of 136,400000
reductions along time: -Tmedian00 of 136,400000
reductions along time: -Tar1: lag-1 autocorrelation00 of 136,400000
morphology: the mask eroded (-kernel sphere 6 -ero)00 of 136,400000
morphology: mean dilation of the masked mean (-kernel 3D -dilM)00 of 136,400000
filters: the separable convolution: Gaussian smoothing, sigma 3 mm (-s 3)00 of 136,400000
filters: the FFT convolution: -kernel sphere 21 -fmean: a 15 × 15 × 13 kernel, by the FFT0.0156107,214 of 136,400 (78.6 %)0.003631.96e−74.94e−15
filters: order statistics: median filter (-kernel 3D -fmedian)00 of 136,400000
edges: -edge: the gradient's magnitude00 of 136,400000
intensity normalisation: masked, its mean scaled to 10,000 (-inm)00 of 136,400000
temporal filtering: high-pass filtered (-bptf 25 -1), then -Tstd00 of 136,400000
regions: -roi: zero outside a box00 of 136,400000
threshold-free cluster enhancement: TFCE of the temporal SNR (-tfce 2 0.5 6)00 of 136,400000
The clean room against itself: thread counts, fix and reproduce
clean room, 1 thread − 14 threadsBit-identicalThe thread count changes nothing: the clean room’s parallel loops (per voxel, column, row or block, every sum in a fixed order) give its one-thread result bit for bit at 14 threads.every outputidentical: 0 of 136,400 values differ
clean room, fix − reproduceBit-identicalNo correction of the register applies to these calls: fix and reproduce give the same values.every outputidentical: 0 of 136,400 values differ

Δ is A − B, computed in float64 over all 136,400 voxels of the outputs, not only those shown. The replica’s record’s threshold: every file fslmaths writes bit-identical (images decompressed, descrip left out), with FSL’s exit status and messages. The clean room’s record (fsl-maths) holds it to specs/fslmaths.md §21: FSL’s exit status, messages, file names and headers identical, the values bit for bit (double-precision functions within one unit in the last place per function, GLIBC-07), the sums (-s, -fmean, -fmeanu, the means and standard deviations, -bptf, …) within 1e-5 of FSL’s largest value with 99 % of the voxels within 4 units in the last place.

One call per family, the replica against the original: FSL fslmaths 6.0.7.23

FamilyVoxelsValues that differLargest |Δ|
binary operations with imagesfslmaths mean.nii.gz -div std.nii.gz -mas mask.nii.gz tsnr.nii.gz136,4000–
thresholds (the robust range)fslmaths mean.nii.gz -thrP 30 -bin thrp.nii.gz136,4000–
voxelwise functionsfslmaths mean.nii.gz -abs -sqrt sqrt.nii.gz136,4000–
reductions along timefslmaths bold.nii.gz -Tmedian tmedian.nii.gz136,4000–
reductions along timefslmaths bold.nii.gz -Tar1 tar1.nii.gz136,4000–
morphologyfslmaths mask.nii.gz -kernel sphere 6 -ero ero.nii.gz136,4000–
morphologyfslmaths mean.nii.gz -mas mask.nii.gz -kernel 3D -dilM dilm.nii.gz136,4000–
filters: the separable convolutionfslmaths mean.nii.gz -s 3 smooth.nii.gz136,4000–
filters: the FFT convolutionfslmaths mean.nii.gz -kernel sphere 21 -fmean fmean-fft.nii.gz136,4000–
filters: order statisticsfslmaths mean.nii.gz -kernel 3D -fmedian fmedian.nii.gz136,4000–
edgesfslmaths mean.nii.gz -edge edge.nii.gz136,4000–
intensity normalisationfslmaths mean.nii.gz -mas mask.nii.gz -inm 10000 inm.nii.gz136,4000–
temporal filteringfslmaths bold.nii.gz -bptf 25 -1 -Tstd bptf.nii.gz136,4000–
regionsfslmaths mean.nii.gz -roi 10 30 10 40 5 30 0 1 roi.nii.gz136,4000–
threshold-free cluster enhancementfslmaths mean.nii.gz -div std.nii.gz -mas mask.nii.gz -tfce 2 0.5 6 tfce.nii.gz136,4000–

Each call’s output by the replica (reproduce) against FSL’s, every voxel of the files. bold.nii.gz is fMRIPrep’s series, mean.nii.gz and std.nii.gz FSL’s -Tmean and -Tstd of it, mask.nii.gz fMRIPrep’s brain mask; with -kernel sphere 21 (15 × 15 × 13 voxels on this grid) fslmaths convolves by the FFT, with -s by separable sums.

One call per family, the clean room against the original: FSL fslmaths 6.0.7.23 (§21’s metric of the sums)

FamilyVoxelsValues that differLargest |Δ|Largest |Δ| ÷ FSL's largest value (§21: ≤ 1e-5)Within 4 units in the last place (§21: ≥ 99 %)
binary operations with imagesfslmaths mean.nii.gz -div std.nii.gz -mas mask.nii.gz tsnr.nii.gz136,4000–––
thresholds (the robust range)fslmaths mean.nii.gz -thrP 30 -bin thrp.nii.gz136,4000–––
voxelwise functionsfslmaths mean.nii.gz -abs -sqrt sqrt.nii.gz136,4000–––
reductions along timefslmaths bold.nii.gz -Tmedian tmedian.nii.gz136,4000–––
reductions along timefslmaths bold.nii.gz -Tar1 tar1.nii.gz136,4000–––
morphologyfslmaths mask.nii.gz -kernel sphere 6 -ero ero.nii.gz136,4000–––
morphologyfslmaths mean.nii.gz -mas mask.nii.gz -kernel 3D -dilM dilm.nii.gz136,4000–––
filters: the separable convolutionfslmaths mean.nii.gz -s 3 smooth.nii.gz136,4000–––
filters: the FFT convolutionfslmaths mean.nii.gz -kernel sphere 21 -fmean fmean-fft.nii.gz136,400107,2140.01561.96e−775.7%
filters: order statisticsfslmaths mean.nii.gz -kernel 3D -fmedian fmedian.nii.gz136,4000–––
edgesfslmaths mean.nii.gz -edge edge.nii.gz136,4000–––
intensity normalisationfslmaths mean.nii.gz -mas mask.nii.gz -inm 10000 inm.nii.gz136,4000–––
temporal filteringfslmaths bold.nii.gz -bptf 25 -1 -Tstd bptf.nii.gz136,4000–––
regionsfslmaths mean.nii.gz -roi 10 30 10 40 5 30 0 1 roi.nii.gz136,4000–––
threshold-free cluster enhancementfslmaths mean.nii.gz -div std.nii.gz -mas mask.nii.gz -tfce 2 0.5 6 tfce.nii.gz136,4000–––

Each call’s output by the clean room (reproduce, 14 threads; the same bits at one thread) against FSL’s, every voxel of the files; where a value differs, specs/fslmaths.md §21’s metric of the sums over every voxel (the non-finite voxels identical). FSL convolves the 15 × 15 × 13 spherical kernel by its FFT in single precision, the clean room by a direct sum whose products it adds in double precision and rounds once; against the exact result (computed in float64 from §10.2’s definition; 15 × 15 × 13 kernel of 1313 voxels) FSL’s FFT lies within 1.5e-07 of the largest value (median 1.6e-08), the clean room’s direct sum (its products in double precision, rounded once) within 7.9e-08 (median 3.4e-09): the clean room is the closer of the two (findings 31).

Provenance of this example
  • Data: OpenNeuro ds005040, sub-002, CC0-1.0
  • Inputs: bold (sha256 ebfbc7b1648ca221…); mean (sha256 1470ac106d7dc916…); std (sha256 c02ef2250ee40f00…); mask (sha256 07f29647b79803b1…)
  • Runs: figures cache entry fsl-maths/families, computed 2026-11-04T20:41:05+00:00 with larmorx 73451f78a400; FSL 6.0.7.23 (fsl-avwutils 2209.6, fsl-newimage 2601.0; ~/fsl-6.0.7.23, never on the PATH); larmorx-nc 0.0.1; every command, its environment, threads and exit status are in its RUN_INFO.json
  • Runs: figures cache entry fsl-maths/original-families, computed 2026-11-07T02:06:32+00:00 with larmorx 80cc208fa4a3; FSL 6.0.7.23 (fsl-avwutils 2209.6, fsl-newimage 2601.0; ~/fsl-6.0.7.23, never on the PATH); “larmorx 0.0.1 (the clean-room originals, crate larmorx-fsl, Apache-2.0)”, as the cache entry records it; every command, its environment, threads and exit status are in its RUN_INFO.json

Every number above is in fsl-maths.json, exported by python -m larmorx_validation evidence fsl-maths.

GLIBC-07: glibc’s double-precision functions

FSL’s build takes its mathematical functions from glibc 2.35. Its double-precision sin, cos, tan, asin, acos, atan, exp, log and pow are not always correctly rounded: for a small share of arguments the result is one unit in the last place from the exact value. LARMORX computes them correctly rounded, the same on every platform (CORE-MATH). In fslmaths’ default type, float, the double result is rounded to float and no value differs; with -dt double the last place shows.

The table runs each function on a real image, FSL’s mean of ds005040’s series, scaled into the function’s range. In the replica:

  • for exp and log, reproduce gives glibc’s own bits, and fix corrects them with a notice;
  • glibc’s sin, cos, tan, asin, acos and atan are IBM’s Accurate Mathematical Library (LGPL, never translated). For sin and cos the replica’s reproduce gives glibc’s bits since 2026-11-04, with LARMORX’s own version of glibc’s sin, cos and sincos, written from a behavioural specification of glibc’s code; its fix corrects them with a notice;
  • tan, asin, acos and atan stay correctly rounded in fix and in reproduce (as decided on 2026-11-04), and differ from FSL on those values (expected divergence).

glibc chooses the build of each of these functions at run time: an FMA build on every x86-64 CPU since 2013, as on the machine that ran FSL. FSL’s own double results therefore depend on the CPU, and “glibc’s bits” here are the FMA build’s.

glibc’s double functions on a real image, the replica against the original: FSL fslmaths 6.0.7.23 (GLIBC-07)

FunctionValuesreproduce: values that differ from FSL'sLargest distance (units in the last place)fix: values that differ from FSL'sNamed by fixThe replica's maths
sinfslmaths -dt double mean.nii.gz -mul 1e-4 -sin sin.nii.gz -odt double136,40000211GLIBC-07glibc’s bits in reproduce (LARMORX’s own version of glibc’s sin and cos, since 2026-11-04), correctly rounded by default (GLIBC-07 named)
cosfslmaths -dt double mean.nii.gz -mul 1e-4 -cos cos.nii.gz -odt double136,4000084GLIBC-07glibc’s bits in reproduce (LARMORX’s own version of glibc’s sin and cos, since 2026-11-04), correctly rounded by default (GLIBC-07 named)
tanfslmaths -dt double mean.nii.gz -mul 1e-4 -tan tan.nii.gz -odt double136,4003131313–correctly rounded in fix and in reproduce (IBM’s code in glibc, LGPL, never translated)
asinfslmaths -dt double mean.nii.gz -mul 5e-6 -asin asin.nii.gz -odt double136,4002091209–correctly rounded in fix and in reproduce (IBM’s code in glibc, LGPL, never translated)
acosfslmaths -dt double mean.nii.gz -mul 5e-6 -acos acos.nii.gz -odt double136,40020120–correctly rounded in fix and in reproduce (IBM’s code in glibc, LGPL, never translated)
atanfslmaths -dt double mean.nii.gz -mul 1e-4 -atan atan.nii.gz -odt double136,4002001200–correctly rounded in fix and in reproduce (IBM’s code in glibc, LGPL, never translated)
expfslmaths -dt double mean.nii.gz -mul 1e-4 -exp exp.nii.gz -odt double136,40000110GLIBC-07glibc’s bits in reproduce, correctly rounded by default (GLIBC-07 named)
logfslmaths -dt double mean.nii.gz -abs -add 1 -log log.nii.gz -odt double136,4000015GLIBC-07glibc’s bits in reproduce, correctly rounded by default (GLIBC-07 named)

fslmaths with -dt double and -odt double on FSL’s mean of ds005040’s BOLD series in T1w space, scaled into each function’s range; every finite value of the output compared bit for bit with FSL 6.0.7.23’s (through the cache of the original’s runs: glibc 2.35’s functions as FSL’s build calls them, the FMA builds this CPU selects). Where glibc’s result is not correctly rounded and the replica’s is, they differ by one unit in the last place: GLIBC-07, an expected divergence for tan, asin, acos and atan (IBM’s code in glibc: no permissive code gives their bits, and they stay correctly rounded in fix and in reproduce, as decided on 2026-11-04), a correction by default for exp, log, sin and cos, whose glibc bits reproduce gives (sin and cos since 2026-11-04, from LARMORX’s own version of glibc’s code, written from its behavioural specification). Float results (fslmaths’ default type) are the same: no float differs after rounding.

Since 2026-11-04 the clean room’s reproduce gives glibc’s bits for sin, cos, exp, log and pow, as the specification asks of it. It uses LARMORX’s versions of glibc’s functions: exp, log and pow translated from Arm’s Optimized Routines as glibc 2.35 ships them (under the MIT licence), sin and cos written from a behavioural specification of glibc’s code and checked against the system’s glibc on 11.8 billion arguments each. tan, asin, acos and atan stay correctly rounded in fix and in reproduce (as decided on 2026-11-04): one unit in the last place from FSL’s where glibc’s are not correctly rounded, within its record’s threshold of one unit per function. Its default computes every one correctly rounded, the replica’s default’s bits, and names GLIBC-07 exactly where a value differs from glibc’s.

glibc’s double functions on a real image, the clean room against the original: FSL fslmaths 6.0.7.23 (GLIBC-07)

FunctionValuesThe clean room's functionreproduce: values that differ from FSL'sLargest distance (units in the last place)fix: values that differ from FSL'sNamed by fixfix against the replica's fix (correctly rounded)1 thread against 14
sinfslmaths -dt double mean.nii.gz -mul 1e-4 -sin sin.nii.gz -odt double136,400glibc’s bits in reproduce, correctly rounded by default (GLIBC-07 named)00211GLIBC-07identicalidentical
cosfslmaths -dt double mean.nii.gz -mul 1e-4 -cos cos.nii.gz -odt double136,400glibc’s bits in reproduce, correctly rounded by default (GLIBC-07 named)0084GLIBC-07identicalidentical
tanfslmaths -dt double mean.nii.gz -mul 1e-4 -tan tan.nii.gz -odt double136,400correctly rounded in fix and in reproduce3131313–identicalidentical
asinfslmaths -dt double mean.nii.gz -mul 5e-6 -asin asin.nii.gz -odt double136,400correctly rounded in fix and in reproduce2091209–identicalidentical
acosfslmaths -dt double mean.nii.gz -mul 5e-6 -acos acos.nii.gz -odt double136,400correctly rounded in fix and in reproduce20120–identicalidentical
atanfslmaths -dt double mean.nii.gz -mul 1e-4 -atan atan.nii.gz -odt double136,400correctly rounded in fix and in reproduce2001200–identicalidentical
expfslmaths -dt double mean.nii.gz -mul 1e-4 -exp exp.nii.gz -odt double136,400glibc’s bits in reproduce, correctly rounded by default (GLIBC-07 named)00110GLIBC-07identicalidentical
logfslmaths -dt double mean.nii.gz -abs -add 1 -log log.nii.gz -odt double136,400glibc’s bits in reproduce, correctly rounded by default (GLIBC-07 named)0015GLIBC-07identicalidentical

The same calls by LARMORX’s clean room (14 threads, and at one). Since 2026-11-04 its reproduce gives glibc 2.35’s bits for sin, cos, exp, log and pow (larmorx_core::math: exp, log and pow from Arm’s Optimized Routines as glibc ships them, MIT; sin and cos written from a behavioural specification of glibc’s code, specs/glibc-dbl.md, and checked against the system’s glibc on 11.8 billion arguments each), as specs/fslmaths.md §19 asks; tan, asin, acos and atan stay correctly rounded in fix and in reproduce (as decided on 2026-11-04). Its default computes every one correctly rounded (CORE-MATH’s, the replica’s default’s bits) and names GLIBC-07 exactly where a value differs from glibc’s. Its record holds the double functions to one unit in the last place per function of the chain (§21); every value that differs here is one unit away (GLIBC-07, an expected divergence). Float results are FSL’s.

The replica’s record has one expected divergence, GLIBC-07: fn/double-tan, -dt double -tan, 3 of 210 values one unit in the last place.

Until 2026-11-04 there was a second, from random testing: difftest/fft-twiddles-glibc-sincos, a mean filter of a 30 × 1 × 3 float image that fslmaths convolves by the FFT, 1 of 90 voxels one unit in the last place.

  • fslmaths’ FFT (Armadillo’s Kiss FFT) takes each twiddle factor from glibc’s sincos, which is not correctly rounded on 0.148 % of the angles: of the transform lengths 1 to 256, 80 hold a factor one unit in the last place from the exact one. The replica computed the factors correctly rounded.
  • Its reproduce now takes them from LARMORX’s own version of glibc’s sincos: Armadillo’s 65,792 factors and 197,376 transform values of the lengths 1 to 256 bit for bit, and this case FSL’s output bit for bit (the probe).
  • Its default keeps the correctly rounded factors and names nothing: a filter would have to run twice to know.

It is rare. A float result usually rounds it away (the FFT convolution of the real series above is FSL’s bit for bit in fix and in reproduce), and fMRIPrep’s calls convolve by separable sums.

Random testing

The replica’s random-testing run (2026-11-03), 24 minutes long, drew random chains of operations on random images. Each was run by FSL 6.0.7.23 (through the cache of the original’s runs) and by the replica, and every file was compared.

Random testing, the replica against the original: FSL fslmaths 6.0.7.23 (114,906 random calls, 2026-11-03)

Cases
Random cases of fslmaths, each compared with FSL 6.0.7.23114,906
bit-identical (every file, the messages, the exit status)114,905
of them, a notice of the default, fix, without a visible change595
differing in the run (each explained in the note)1
distinct options drawn114
chains with -fmean or -fmeanu (convolutions, the FFT’s among them)5,305
-dt double13,337
length (minutes; 13 jobs at one thread each, both programs)24

A random-testing run of the replica, fslmaths and fslstats at once, 24 minutes long (seed fsl-maths-trial, larmorx 0427fad149ac): chains of one to four operations drawn for the current image (every family), on six kinds of images (scan-like, series, masks, z and p maps, tensors) on grids of 6 to 32 voxels per axis, with -dt, -odt, the output’s name and FSLOUTPUTTYPE. Every case in a memory box of 1.5 GiB (peaks below 1.05 GiB), FSL through the cache of the original’s runs (use mode). Its one difference (a convolution by the FFT of a 30 × 1 × 3 image: 1 of 90 voxels one unit in the last place) is GLIBC-07, the twiddle factors FSL takes from glibc’s sincos: a permanent case of the record, an expected divergence until 2026-11-04 and FSL’s bit for bit since (the replica’s reproduce takes them from LARMORX’s own version of glibc’s sincos, written from a behavioural specification of glibc’s code). The notices without a visible change name a correction of an intermediate result that a later operation removes (a threshold, -bin, -restart, a reduction). The check made before the run found three mistakes of the replica’s, fixed with permanent cases.

The clean room’s random-testing run (2026-11-03) drew the replica’s run’s chains again and more, so that run’s calls of FSL were served by the cache of the original’s runs: 150,000 random chains of one to four operations in 20 minutes, each compared as the record compares. 143,191 agreed within the thresholds, and 804 differed only by a correction its default names. Each of the 6,005 that differed was run again with the part’s final build and put in one category below:

  • most were the clean room’s own mistakes, fixed with permanent cases of the record: the geometry of a new grid along x (2,022), -roc’s false-positive candidates (1,477), the morphological filters’ messages with a kernel image, a subsampled image’s quaternion, TFCE’s checks in FSL’s order, the integer arithmetic of -edge and the subsampling, and four defects now named wherever they apply (FSL-75, FSL-154, FSL-157, FSL-165);
  • 1,178 are explained: degenerate tensors whose eigenvectors are not defined, the header’s offsets after a grid change (at most 6 units in the last place), the last bits of -subsamp2 under the floating types, of Φ and Φ−1 and of sums that cancel to near 0, -dog_edge’s crossings at exact ties, signed zeros;
  • 33 are -subsamp2offc’s means, as the record’s expected divergences;
  • 7 are FSL-75, FSL’s heap corruption, which ends FSL before it writes, while the clean room finishes and names it;
  • 10 stay open, with their reproducers: 8 display ranges after -range, in their last bits, and 2 -roc curve numbers 2.4 × 10−7 and 5 × 10−7 from FSL’s.

Random testing, the clean room against the original: FSL fslmaths 6.0.7.23 (150,000 random calls, 2026-11-03)

Cases
Random cases of fslmaths, each compared with FSL 6.0.7.23150,000
within the record’s thresholds, every exact check passed143,191
differing only by a correction the default, fix, names804
differing in the run (each one triaged below)6,005
distinct options and settings drawn114
chains with -s, -fmean or -fmeanu (the convolutions)29,306
-dt double, -dt input17,424 / 8,649
length (minutes; 13 jobs at one thread each, both programs)19.8

A random-testing run of the clean rooms, fslmaths and fslstats at once, 20 minutes long (seed fsl-maths-trial, larmorx a09928ecb344): the streams of the replica’s run drawn again, so that run’s calls of FSL were served by the cache of the original’s runs and the rest stored there; realistic cases only, each compared as the record compares. Every case in a memory box of 1.5 GiB (peaks below 0.92 GiB). The differing cases were re-run with the part’s final build and each put in one category (below).

The clean room’s differences from the original in the run of 2026-11-03, by category

CategoryCasesReproducer
a LARMORX mistake, fixed: the geometry of a new grid along x (-subsamp2, the reductions along X of an image FSL holds mirrored or without matrices)2,022fsl-maths-trial-00015-original
a LARMORX mistake, fixed: -roc’s false-positive candidates (the mask’s voxels below the true voxels’ bound)1,477fsl-maths-trial-00025-original
explained: -tensor_decomp of degenerate tensors (eigenvalues at the rounding level: their eigenvectors are not defined)444fsl-maths-trial-00575-original
explained: the header’s offsets after a grid change, at most 6 units in the last place of one offset (spec question 31)423fsl-maths-trial-00510-original
a LARMORX mistake, fixed: the morphological filters’ messages with a kernel image304fsl-maths-trial-00563-original
a LARMORX mistake, fixed: a subsampled image’s qform quaternion282fsl-maths-trial-00610-original
a LARMORX mistake, fixed: TFCE’s checks in FSL’s order191fsl-maths-trial-00681-original
a LARMORX mistake, fixed: FSL-157 named for every int8 image written with -odt input144fsl-maths-trial-00237-original
explained: -subsamp2’s last bits under the floating types (spec questions 13 and 38)125fsl-maths-trial-00763-original
a LARMORX mistake, fixed: FSL-165 for -fillh of every image with an axis of size 190fsl-maths-trial-00909-original
explained: -ztop’s and -ptoz’s last bits (Φ, Φ⁻¹), within the sums’ bound (spec question 35)59fsl-maths-trial-09852-original
fixed: a chain that one of the fixes above reaches58fsl-maths-trial-06324-original
the suite’s own mistake, fixed: its tensor check on single slices, NIfTI-2 quaternions51fsl-maths-trial-00792-original
explained: -dog_edge’s zero crossings at exact ties, within the sums’ bound50fsl-maths-trial-01697-original
explained: sums beyond 4 units in the last place where they cancel to near 0 (spec question 14)40fsl-maths-trial-11289-original
a LARMORX mistake, fixed: TFCE of a double image in single precision38fsl-maths-trial-02984-original
the suite’s own mistake, fixed: its tensor check (compressed names, single slices)35fsl-maths-trial-05690-original
expected: -subsamp2offc’s means (as the record’s two expected divergences)33fsl-maths-trial-00518-original
a LARMORX mistake, fixed: FSL-154 named for every uint8 image computed unsigned32fsl-maths-trial-09917-original
explained: -randn in double precision: Φ⁻¹’s last bits (spec question 36)31fsl-maths-trial-01935-original
a LARMORX mistake, fixed: -ptoz’s and -ztop’s NaN29fsl-maths-trial-07071-original
a LARMORX mistake, fixed: the subsampling’s sums under the integer types11fsl-maths-trial-35728-original
the suite’s own mistake, fixed: §21’s TFCE metric10fsl-maths-trial-00413-original
open: the display range (cal_min, cal_max) after -range, in its last bits, or -2147483648 in FSL against NaN8fsl-maths-trial-41014-original
FSL-75: FSL’s heap corruption ends it before it writes; the clean room finishes and names it, correcting it by default7fsl-maths-trial-21013-original
explained: TFCE’s last bits through a discontinuous operation (-rem)3fsl-maths-trial-47605-original
explained: the median’s and the mode’s signed zero (§18: ±0’s order)3fsl-maths-trial-67729-original
a LARMORX mistake, fixed: -edge’s differences under the integer types2fsl-maths-trial-39749-original
open: -roc’s curve numbers 2.4e-7 and 5e-7 from FSL’s, beyond half a unit of the 5th significant digit2fsl-maths-trial-54988-original
a LARMORX mistake, fixed: FSL-75 named in -subsamp21fsl-maths-trial-00010-original

Every differing case of the run re-run with the part’s final build and put in one category by its recipe and its first failed check (validation/investigations/cleanroom-maths/triage_counts.py, its output validation/investigations/cleanroom-maths/trial_triage.json), each category’s cause established by probes and replays (docs/findings/fsl-maths-original.md, findings 24 to 30); LARMORX’s mistakes and the suite’s are fixed, with 17 permanent cases of fslmaths and 4 of fslstats in the records; the reproducer is a finding of the run (difftest --replay).

Every case

Each case of the two records, the replica’s and the clean room’s, both against FSL 6.0.7.23 (a column each), with what it tests, its command line, each port’s verdict and the reason. The 37 cases only the clean room’s record holds are its own: orig/ and the permanent cases random testing gave it. Search, filter by category or verdict (or select a segment of a chart above), sort by a column; #case=<name> links to one case.

459 cases

CaseCategoryWhat it testsreplicaclean roomChecks
fmriprep/ds000005-sub-01-noreconall/sub_01_wf/enhance_and_skullstrip_bold_wffMRIPrep's callsfMRIPrep's combine_masks call (fslmaths <mask> -mul <mask> <out>)
command linefslmaths <BET's first-pass mask> -mul <the second mask> <out>
Bit-identical

every file written (1/1) byte for byte, with FSL’s exit status and messages; fMRIPrep’s own mask: the same values

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
fmriprep/ds003592-sub-280-noreconall/sub_280_ses_1_wf/enhance_and_skullstrip_bold_wffMRIPrep's callsfMRIPrep's combine_masks call (fslmaths <mask> -mul <mask> <out>)
command linefslmaths <BET's first-pass mask> -mul <the second mask> <out>
Bit-identical

every file written (1/1) byte for byte, with FSL’s exit status and messages; fMRIPrep’s own mask: the same values

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
fmriprep/ds005040-sub-002-noreconall/sub_002_wf/enhance_and_skullstrip_bold_wf worked examplefMRIPrep's callsfMRIPrep's combine_masks call (fslmaths <mask> -mul <mask> <out>)
command linefslmaths <BET's first-pass mask> -mul <the second mask> <out>
Bit-identical

every file written (1/1) byte for byte, with FSL’s exit status and messages; fMRIPrep’s own mask: the same values

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
fmriprep/ds005040-sub-002-noreconall/sub_002_wf/skullstrip_bold_wffMRIPrep's callsfMRIPrep's combine_masks call (fslmaths <mask> -mul <mask> <out>)
command linefslmaths <BET's first-pass mask> -mul <the second mask> <out>
Bit-identical

every file written (1/1) byte for byte, with FSL’s exit status and messages; fMRIPrep’s own mask: the same values

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/stdev_volume worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node stdev_volume (--project-goodvoxels)
command linefslmaths <bold> -Tstd stdev_volume.nii.gz
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/mean_volume worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node mean_volume (--project-goodvoxels)
command linefslmaths <bold> -Tmean mean_volume.nii.gz
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/cov_volume worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node cov_volume (--project-goodvoxels)
command linefslmaths <stdev_volume> -div <mean_volume> cov_volume.nii.gz
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/cov_ribbon worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node cov_ribbon (--project-goodvoxels)
command linefslmaths <cov_volume> -mas <ribbon> cov_ribbon.nii.gz
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/cov_ribbon_norm worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node cov_ribbon_norm (--project-goodvoxels)
command linefslmaths <cov_ribbon> -div <cov_ribbon_mean> cov_ribbon_norm.nii.gz
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/smooth_norm worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node smooth_norm (--project-goodvoxels)
command linefslmaths <cov_ribbon_norm> -bin -s 5 smooth_norm.nii.gz
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/cov_ribbon_norm_smooth worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node cov_ribbon_norm_smooth (--project-goodvoxels)
command linefslmaths <cov_ribbon_norm> -s 5 -div <smooth_norm> -dilD cov_ribbon_norm_smooth.nii.gz
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/cov_norm worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node cov_norm (--project-goodvoxels)
command linefslmaths <cov_volume> -div <cov_ribbon_mean> cov_norm.nii.gz
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/cov_norm_modulate worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node cov_norm_modulate (--project-goodvoxels)
command linefslmaths <cov_norm> -div <cov_ribbon_norm_smooth> cov_norm_modulate.nii.gz
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/cov_norm_modulate_ribbon worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node cov_norm_modulate_ribbon (--project-goodvoxels)
command linefslmaths <cov_norm_modulate> -mas <ribbon> cov_norm_modulate_ribbon.nii.gz
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/bin_mean_volume worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node bin_mean_volume (--project-goodvoxels)
command linefslmaths <mean_volume> -bin bin_mean_volume.nii.gz
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/goodvoxels_thr worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node goodvoxels_thr (--project-goodvoxels)
command linefslmaths <cov_norm_modulate> -thr <upper_thresh> goodvoxels_thr.nii.gz
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/goodvoxels_mask worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node goodvoxels_mask (--project-goodvoxels)
command linefslmaths <goodvoxels_thr> -bin -sub <bin_mean_volume> -mul -1 goodvoxels_mask.nii.gz
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fmriprep-goodvoxels/ds005040-sub-002-noreconall/goodvoxels_ribbon_mask worked examplefMRIPrep's callsfMRIPrep's goodvoxels mask, node goodvoxels_ribbon_mask (--project-goodvoxels)
command linefslmaths <goodvoxels_mask> -mas <ribbon> goodvoxels_ribbon_mask.nii.gz
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/usagethe command lineno arguments: the usage (exit 1)
command linefslmaths
Rejected by both

fslmaths exits 1; the same from the replica.

Rejected by both

fslmaths exits 1; the same from the clean room.

5/5 · 4/4
cli/helpthe command line-h: the usage (exit 0)
command linefslmaths -h
Bit-identical

every file written (0/0) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 5/5
cli/help-longthe command line--help
command linefslmaths --help
Bit-identical

every file written (0/0) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 5/5
cli/missing-inputthe command linean input that does not exist
command linefslmaths nothere out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #63 :: No image files match: nothere terminate called after throwing an instance of ‘std::runtime_error’ what(): No image files match: nothere

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #63 :: No image files match: nothere terminate called after throwing an instance of ‘std::runtime_error’ what(): No image files match: nothere

5/5 · 4/4
cli/no-outputthe command lineno output name
command linefslmaths <f3: a synthetic input>
Rejected by both

fslmaths exits 1; the same from the replica. Error: no output filename specified!

Rejected by both

fslmaths exits 1; the same from the clean room. Error: no output filename specified!

5/5 · 4/4
cli/copythe command lineno operation: a copy (the display range kept)
command linefslmaths <f3: a synthetic input> out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/unknown-opthe command linean unknown operation: the usage
command linefslmaths <f3: a synthetic input> -foo out
Rejected by both

fslmaths exits 1; the same from the replica.

Rejected by both

fslmaths exits 1; the same from the clean room.

5/5 · 4/4
cli/odt-unknownthe command line-odt with an unknown type
command linefslmaths <f3: a synthetic input> out -odt bogus
Rejected by both

fslmaths exits 1; the same from the replica.

Rejected by both

fslmaths exits 1; the same from the clean room.

5/5 · 4/4
cli/dt-unknownthe command line-dt with an unknown type
command linefslmaths -dt bogus <f3: a synthetic input> out
Rejected by both

fslmaths exits 1; the same from the replica.

Rejected by both

fslmaths exits 1; the same from the clean room.

5/5 · 4/4
cli/dt-no-inputthe command line-dt float without an input (FSL-158)
command linefslmaths -dt float
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): basic_string: construction from null is not valid The replica’s default, fix, ends with exit status 1, naming FSL-158.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): basic_string: construction from null is not valid The clean room’s default, fix, ends with exit status 1, naming FSL-158.

5/5 · 4/4
cli/op-as-outputthe command linean operation in the output's place
command linefslmaths <f3: a synthetic input> -add
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/add-output-imagethe command line-add reading the output's name as an image
command linefslmaths <f3: a synthetic input> -add out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #63 :: No image files match: out Image Exception : #22 :: Failed to read volume out Error : No image files match: out terminate called after throwing an instance of 'std::runtime_err

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #63 :: No image files match: out Image Exception : #22 :: Failed to read volume out Error : No image files match: out terminate called after t

5/5 · 4/4
cli/chainthe command linea chain of operations
command linefslmaths <f3: a synthetic input> -mul 2 -add <f3b: a synthetic input> -abs -sqrt out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/datatype-longthe command line-datatype and -output_datatype
command linefslmaths -datatype double <f3: a synthetic input> -div 3 out -output_datatype int
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/save-restartthe command line-save mid-chain and -restart
command linefslmaths <f3: a synthetic input> -mul 3 -save mid -restart <f3b: a synthetic input> -add 1 out
Bit-identical

every file written (4/4) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 9/9
cli/rangethe command line-range: the display range of the result
command linefslmaths <f3: a synthetic input> -mul 2 -range out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/niftithe command lineFSLOUTPUTTYPE=NIFTI
command lineFSLOUTPUTTYPE=NIFTI fslmaths <f3: a synthetic input> -add 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/pairthe command lineFSLOUTPUTTYPE=NIFTI_PAIR, a pair as input
command lineFSLOUTPUTTYPE=NIFTI_PAIR fslmaths <pair.hdr: a synthetic input> -add 1 out
Bit-identical

every file written (4/4) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/nifti2the command lineFSLOUTPUTTYPE=NIFTI2_GZ
command lineFSLOUTPUTTYPE=NIFTI2_GZ fslmaths <f3: a synthetic input> -add 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/analyzethe command lineFSLOUTPUTTYPE=ANALYZE
command lineFSLOUTPUTTYPE=ANALYZE fslmaths <f3: a synthetic input> -add 1 out
Rejected by both

fslmaths exits 1; the same from the replica. ERROR:: Unrecognised value (ANALYZE) of environment variable FSLOUTPUTTYPE Legal values are: NIFTI, NIFTI_PAIR, NIFTI2, NIFTI2_PAIR, NIFTI_GZ, NIFTI_PAIR_GZ, NIFTI2_GZ, NIFTI2_PAIR_GZ, NIFTI_ZST, NIFT

Rejected by both

fslmaths exits 1; the same from the clean room. ERROR:: Unrecognised value (ANALYZE) of environment variable FSLOUTPUTTYPE Legal values are: NIFTI, NIFTI_PAIR, NIFTI2, NIFTI2_PAIR, NIFTI_GZ, NIFTI_PAIR_GZ, NI

5/5 · 4/4
cli/outtype-badthe command lineFSLOUTPUTTYPE unknown
command lineFSLOUTPUTTYPE=BOGUS fslmaths <f3: a synthetic input> -add 1 out
Rejected by both

fslmaths exits 1; the same from the replica. ERROR:: Unrecognised value (BOGUS) of environment variable FSLOUTPUTTYPE Legal values are: NIFTI, NIFTI_PAIR, NIFTI2, NIFTI2_PAIR, NIFTI_GZ, NIFTI_PAIR_GZ, NIFTI2_GZ, NIFTI2_PAIR_GZ, NIFTI_ZST, NIFTI_

Rejected by both

fslmaths exits 1; the same from the clean room. ERROR:: Unrecognised value (BOGUS) of environment variable FSLOUTPUTTYPE Legal values are: NIFTI, NIFTI_PAIR, NIFTI2, NIFTI2_PAIR, NIFTI_GZ, NIFTI_PAIR_GZ, NIFT

5/5 · 4/4
cli/nii-inputthe command linean uncompressed input
command linefslmaths <f3.nii: a synthetic input> -add 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
io/truncated-gzreading imagesa gzip stream cut inside the data (FSL's short read)
command linefslmaths <trunc.nii.gz: a synthetic input> -add 1 out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #22 :: Failed to read volume trunc.nii.gz Error : Error: short read, file may be truncated terminate called after throwing an instance of ‘std::runtime_error’ what(): Failed to read

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #22 :: Failed to read volume trunc.nii.gz Error : Error: short read, file may be truncated terminate called after throwing an instance of 'std

5/5 · 4/4
io/truncated-rawreading imagesan uncompressed file cut inside the data
command linefslmaths <truncraw.nii: a synthetic input> -add 1 out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #22 :: Failed to read volume truncraw.nii Error : Error: short read, file may be truncated terminate called after throwing an instance of ‘std::runtime_error’ what(): Failed to read

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #22 :: Failed to read volume truncraw.nii Error : Error: short read, file may be truncated terminate called after throwing an instance of 'std

5/5 · 4/4
io/huge-headerreading imagesa header claiming 1000 x 1000 x 1000 voxels, its file 64 bytes of data
command linefslmaths <hugehdr.nii: a synthetic input> -add 1 out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #22 :: Failed to read volume hugehdr.nii Error : Error: short read, file may be truncated terminate called after throwing an instance of ‘std::runtime_error’ what(): Failed to read v

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #22 :: Failed to read volume hugehdr.nii Error : Error: short read, file may be truncated terminate called after throwing an instance of 'std:

5/5 · 4/4
io/gap-rawreading imagesthe data after a gap (vox_offset 1024), uncompressed
command linefslmaths <gap.nii: a synthetic input> -mul 2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
io/gap-gzreading imagesthe data after a gap (vox_offset 1024), gzip-compressed
command linefslmaths <gap.nii.gz: a synthetic input> -mul 2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/neurothe command linea neurological input (swapped in memory, back on output)
command linefslmaths <neuro: a synthetic input> -mul <f3b: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
cli/noformthe command linean input without sform and qform
command linefslmaths <noform: a synthetic input> -add <f3b: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/chardata types-dt char on a uint8 image
command linefslmaths -dt char <u8: a synthetic input> -add 100 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-154 and names it: 97 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-154 and names it: 97 values change.

6/6 · 7/7
dt/shortdata types-dt short (wrap-around)
command linefslmaths -dt short <s3: a synthetic input> -mul 20 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/intdata types-dt int
command linefslmaths -dt int <s3: a synthetic input> -div 7 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/doubledata types-dt double
command linefslmaths -dt double <f3: a synthetic input> -div 3 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/floatdata types-dt float
command linefslmaths -dt float <d3: a synthetic input> -mul 3 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/input-shortdata types-dt input on int16
command linefslmaths -dt input <s3: a synthetic input> -div 7 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/input-chardata types-dt input on uint8
command linefslmaths -dt input <u8: a synthetic input> -bin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-152, FSL-154 and names it: 106 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-154, FSL-152 and names it: 106 values change.

6/6 · 7/7
dt/input-uint16data types-dt input on uint16 (FSL-155)
command linefslmaths -dt input <u16: a synthetic input> -div 3 out
Bit-identical

every file written (1/1) byte for byte, with FSL’s exit status and messages; its default, fix, names FSL-155, with the same values

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, names FSL-155, with the same values.

6/6 · 5/5
dt/input-int8data types-dt input on int8 (FSL-155)
command linefslmaths -dt input <i8: a synthetic input> -add 1 out
Bit-identical

every file written (1/1) byte for byte, with FSL’s exit status and messages; its default, fix, names FSL-155, with the same values

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, names FSL-155, with the same values.

6/6 · 5/5
dt/double-defaultdata typesa double input: computed and written in double
command linefslmaths <d3: a synthetic input> -mul 3 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/scaleddata typesa scaled int16 input (read as float)
command linefslmaths <scaled: a synthetic input> -add 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/odt-chardata types-odt char (rounded; values out of range: FSL-151)
command linefslmaths <f3: a synthetic input> out -odt char
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-151 and names it: 88 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-151 and names it: 88 values change.

6/6 · 7/7
dt/odt-shortdata types-odt short
command linefslmaths <f3: a synthetic input> -mul 100 out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/odt-intdata types-odt int
command linefslmaths <f3: a synthetic input> -mul 1e8 out -odt int
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-151 and names it: 68 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-151 and names it: 68 values change.

6/6 · 7/7
dt/odt-doubledata types-odt double
command linefslmaths <f3: a synthetic input> out -odt double
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/odt-inputdata types-odt input on int16
command linefslmaths <s3: a synthetic input> -div 3 out -odt input
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dt/odt-input-uint16data types-odt input on uint16
command linefslmaths <u16: a synthetic input> -div 3 out -odt input
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-157 and names it: 61 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-157 and names it: 61 values change.

6/6 · 7/7
d13/char-uint8-thrdefects of FSLFSL-154: -dt char on a uint8 image, -thr 100 (values 128-255 read as negative)
command linefslmaths -dt char <u8: a synthetic input> -thr 100 out -odt char
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-152, FSL-154 and names it: 106 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-154, FSL-152 and names it: 106 values change.

6/6 · 7/7
d13/input-uint8-divdefects of FSLFSL-154: -dt input on a uint8 image, -div 2, written as float
command linefslmaths -dt input <u8: a synthetic input> -div 2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-154 and names it: 106 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-154 and names it: 106 values change.

6/6 · 7/7
d13/odt-input-int8defects of FSLFSL-157: -odt input on an int8 image (truncated, written as uint8)
command linefslmaths <i8: a synthetic input> -mul 0.75 out -odt input
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-157 and names it: 210 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-157 and names it: 210 values change.

6/6 · 7/7
d13/odt-input-uint16-rounddefects of FSLFSL-157: -odt input on a uint16 image, -mul 0.5 (truncated, not rounded)
command linefslmaths <u16: a synthetic input> -mul 0.5 out -odt input
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-157 and names it: 118 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-157 and names it: 118 values change.

6/6 · 7/7
dt/odt-nandata types-odt short with NaN and infinite voxels (FSL-151)
command linefslmaths <nan3: a synthetic input> out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-151 and names it: 1 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-151 and names it: 1 values change.

6/6 · 7/7
dt/int-overflowdata types-dt int with a number beyond int (FSL-151)
command linefslmaths -dt int <s3: a synthetic input> -add 1e10 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-151 and names it: 2 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-151 and names it: 2 values change.

6/6 · 7/7
dt/int-expdata types-dt short -exp (results beyond short: FSL-151)
command linefslmaths -dt short <s3: a synthetic input> -exp out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-151 and names it: 194 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-151 and names it: 194 values change.

6/6 · 7/7
op/add-numberarithmetic with numbers and images-add a number
command linefslmaths <f3: a synthetic input> -add 2.5 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/sub-numberarithmetic with numbers and images-sub a number
command linefslmaths <f3: a synthetic input> -sub -1e3 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/mul-numberarithmetic with numbers and images-mul a number
command linefslmaths <f3: a synthetic input> -mul 0.1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/div-numberarithmetic with numbers and images-div a number
command linefslmaths <f3: a synthetic input> -div 7 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/div-zeroarithmetic with numbers and images-div 0: nothing done
command linefslmaths <f3: a synthetic input> -div 0 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/add-imagearithmetic with numbers and images-add an image
command linefslmaths <f3: a synthetic input> -add <f3b: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/sub-imagearithmetic with numbers and images-sub an image
command linefslmaths <f3: a synthetic input> -sub <f3b: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/mul-imagearithmetic with numbers and images-mul an image
command linefslmaths <f3: a synthetic input> -mul <f3b: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/div-imagearithmetic with numbers and images-div an image with zeros (0 there)
command linefslmaths <f3: a synthetic input> -div <f3b: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/3d-by-4darithmetic with numbers and imagesa 3D image times a 4D one: cloned over time
command linefslmaths <f3: a synthetic input> -mul <f4: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/4d-by-3darithmetic with numbers and imagesa 4D image times a 3D one
command linefslmaths <f4: a synthetic input> -sub <f3: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/4d-by-4d-other-tarithmetic with numbers and images4D images of 3 and 5 volumes: refused
command linefslmaths <f4: a synthetic input> -add <f45: a synthetic input> out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #3 :: Attempted to add images of different sizes terminate called after throwing an instance of ‘std::runtime_error’ what(): Attempted to add images of different sizes

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #3 :: Attempted to add images of different sizes terminate called after throwing an instance of ‘std::runtime_error’ what(): Attempted to add

5/5 · 4/4
op/size-mismatcharithmetic with numbers and images-add an image of another size: refused
command linefslmaths <f3: a synthetic input> -add <other: a synthetic input> out
Rejected by both

fslmaths exits -6; the same from the replica. WARNING:: Inconsistent orientations for individual images in pipeline! Will use voxel-based orientation which is probably incorrect - PLEASE CHECK! Image Exception : #3 :: Attempted to add images of

Rejected by both

fslmaths exits -6; the same from the clean room. WARNING:: Inconsistent orientations for individual images in pipeline! Will use voxel-based orientation which is probably incorrect - PLEASE CHECK! Image Exce

5/5 · 4/4
op/rowarithmetic with numbers and images-add a 7x1x1 image (newimage compares the first dimension only)
command linefslmaths <f3: a synthetic input> -add <row: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/orientationarithmetic with numbers and images-add an image of another orientation: the warning
command linefslmaths <f3: a synthetic input> -add <rot: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/masarithmetic with numbers and images-mas a uint8 mask
command linefslmaths <f3: a synthetic input> -mas <m3: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/mas-4darithmetic with numbers and images-mas a 4D mask on a 3D image
command linefslmaths <f3: a synthetic input> -mas <m4: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/mas-infarithmetic with numbers and images-mas: infinite voxels outside the mask
command linefslmaths <inf3: a synthetic input> -mas <m3: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/rem-numberarithmetic with numbers and images-rem a number
command linefslmaths <f3: a synthetic input> -rem 7 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/rem-imagearithmetic with numbers and images-rem an image (0 voxels skipped)
command linefslmaths <s3: a synthetic input> -rem <rem3: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/max-numberarithmetic with numbers and images-max a number
command linefslmaths <f3: a synthetic input> -max 0 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/min-numberarithmetic with numbers and images-min a number
command linefslmaths <f3: a synthetic input> -min 5 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/max-imagearithmetic with numbers and images-max an image
command linefslmaths <f3: a synthetic input> -max <f3b: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/min-image-4darithmetic with numbers and images-min a 4D image
command linefslmaths <f3: a synthetic input> -min <f4: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
op/nan-numberarithmetic with numbers and images-add nan
command linefslmaths <nan3: a synthetic input> -add nan out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
thr/thrthresholds-thr
command linefslmaths <f3: a synthetic input> -thr 5 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
thr/uthrthresholds-uthr
command linefslmaths <f3: a synthetic input> -uthr 5 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
thr/thrpthresholds-thrp: the robust range
command linefslmaths <f3: a synthetic input> -thrp 40 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
thr/thrPthresholds-thrP: the positive voxels' robust range
command linefslmaths <f3: a synthetic input> -thrP 40 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
thr/uthrpthresholds-uthrp
command linefslmaths <f3: a synthetic input> -uthrp 60 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
thr/uthrPthresholds-uthrP
command linefslmaths <f3: a synthetic input> -uthrP 60 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
thr/thrP-emptythresholds-thrP of an image without positive voxels: the warnings
command linefslmaths <neg3: a synthetic input> -thrP 20 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
d13/thrp-fardefects of FSLFSL-187: -thrp 50 of 1000 values on [0, 1) and one at 1e12 (the robust range)
command linefslmaths <far3: a synthetic input> -thrp 50 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-187 and names it: 1 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-187 and names it: 1 values change.

6/6 · 7/7
d13/uthrP-fardefects of FSLFSL-187: -uthrP 98 of the same image (the positive voxels' robust range)
command linefslmaths <far3: a synthetic input> -uthrP 98 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-187 and names it: 1 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-187 and names it: 1 values change.

6/6 · 7/7
thr/thr-wordthresholds-thr without a number
command linefslmaths <f3: a synthetic input> -thr five out
Rejected by both

fslmaths exits 1; the same from the replica. Error: -thr requires a number

Rejected by both

fslmaths exits 1; the same from the clean room. Error: -thr requires a number

5/5 · 4/4
thr/uthrP-wordthresholds-uthrP without a number
command linefslmaths <f3: a synthetic input> -uthrP x out
Rejected by both

fslmaths exits 1; the same from the replica. Error: -uthrP requires a number

Rejected by both

fslmaths exits 1; the same from the clean room. Error: -uthrP requires a number

5/5 · 4/4
thr/thr-shortthresholds-thr under -dt short
command linefslmaths -dt short <s3: a synthetic input> -thr 1000 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
thr/thr-4dthresholds-thrp of a 4D image
command linefslmaths <f4: a synthetic input> -thrp 50 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
thr/thr-infthresholds-thr with infinite voxels
command linefslmaths <inf3: a synthetic input> -thr 0 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/expvoxelwise functions-exp
command linefslmaths <f3: a synthetic input> -div 10 -exp out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/logvoxelwise functions-log
command linefslmaths <f3: a synthetic input> -log out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/sinvoxelwise functions-sin
command linefslmaths <f3: a synthetic input> -sin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/cosvoxelwise functions-cos
command linefslmaths <f3: a synthetic input> -cos out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/tanvoxelwise functions-tan
command linefslmaths <f3: a synthetic input> -tan out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/asinvoxelwise functions-asin (NaN outside [-1, 1])
command linefslmaths <unit3: a synthetic input> -asin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/acosvoxelwise functions-acos
command linefslmaths <unit3: a synthetic input> -acos out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/atanvoxelwise functions-atan
command linefslmaths <f3: a synthetic input> -atan out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/sqrvoxelwise functions-sqr
command linefslmaths <f3: a synthetic input> -sqr out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/sqrtvoxelwise functions-sqrt (0 for non-positive values)
command linefslmaths <f3: a synthetic input> -sqrt out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/recipvoxelwise functions-recip
command linefslmaths <f3: a synthetic input> -recip out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/absvoxelwise functions-abs
command linefslmaths <nan3: a synthetic input> -abs out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/sin-infvoxelwise functions-sin of infinite and NaN voxels
command linefslmaths <nan3: a synthetic input> -sin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/tan-infvoxelwise functions-tan of infinite and NaN voxels
command linefslmaths <nan3: a synthetic input> -tan out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/exp-infvoxelwise functions-exp of infinite and NaN voxels
command linefslmaths <inf3: a synthetic input> -exp out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/log-infvoxelwise functions-log of infinite voxels
command linefslmaths <inf3: a synthetic input> -log out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/double-sinvoxelwise functions-sin in double
command linefslmaths <d3: a synthetic input> -sin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/double-cosvoxelwise functions-cos in double
command linefslmaths <d3: a synthetic input> -cos out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/double-tanvoxelwise functions-tan in double
command linefslmaths <d3: a synthetic input> -tan out
Expected divergence

GLIBC-07: with -dt double, glibc 2.35’s tan (IBM’s code in glibc, LGPL) is not always correctly rounded and the replica’s is, in fix and in reproduce: here 3 of 210 double values one unit in the last place from FSL’s (tan differs on 0.21 % of image intensities). No permissive code gives glibc’s tan, and tan, asin, acos and atan stay correctly rounded in fix and in reproduce (as decided on 2026-11-04).

Expected divergence

GLIBC-07: with -dt double, glibc 2.35’s tan (IBM’s code in glibc, LGPL) is not always correctly rounded, and the clean room’s is (CORE-MATH’s, in fix and in reproduce): one unit in the last place on a few of the 210 double values, within §21’s one unit per function. Since 2026-11-04 reproduce gives glibc’s bits for sin, cos, exp, log and pow (LARMORX’s glibc functions; sin and cos written from a behavioural specification of glibc’s code); tan, asin, acos and atan stay correctly rounded in fix and in reproduce (as decided on 2026-11-04).

4/5 · 7/7
fn/double-atanvoxelwise functions-atan in double
command linefslmaths <d3: a synthetic input> -atan out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/double-asinvoxelwise functions-asin and -acos in double
command linefslmaths <d3: a synthetic input> -div 50 -asin -mul 0.5 -acos out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/double-exp-logvoxelwise functions-exp, -log and -pow in double
command linefslmaths <d3: a synthetic input> -div 10 -exp -log -pow 1.7 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/binvoxelwise functions-bin
command linefslmaths <f3: a synthetic input> -bin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/binvvoxelwise functions-binv
command linefslmaths <f3: a synthetic input> -binv out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/nanvoxelwise functions-nan
command linefslmaths <nan3: a synthetic input> -nan out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/nanmvoxelwise functions-nanm
command linefslmaths <nan3: a synthetic input> -nanm out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/indexvoxelwise functions-index
command linefslmaths <f3: a synthetic input> -index out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/gridvoxelwise functions-grid
command linefslmaths <f3: a synthetic input> -grid 100 3 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/randvoxelwise functions-rand (fixed initial state)
command linefslmaths <f3: a synthetic input> -rand out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/randnvoxelwise functions-randn
command linefslmaths <f3: a synthetic input> -randn out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/seed-randvoxelwise functions-seed then -rand
command linefslmaths <f3: a synthetic input> -seed 42 -rand out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/seed-randnvoxelwise functions-seed then -randn
command linefslmaths <f4: a synthetic input> -seed -7 -randn out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/powvoxelwise functions-pow (undocumented)
command linefslmaths <p3: a synthetic input> -pow 2.5 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/invbinvoxelwise functions-invbin (undocumented)
command linefslmaths <m3: a synthetic input> -invbin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/int-recipvoxelwise functions-recip under -dt int
command linefslmaths -dt int <s3: a synthetic input> -recip out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/int-sqrtvoxelwise functions-sqrt under -dt short
command linefslmaths -dt short <s3: a synthetic input> -sqrt out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fn/char-indexvoxelwise functions-index under -dt char (wraps)
command linefslmaths -dt char <u8: a synthetic input> -index out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-154 and names it: 209 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-154 and names it: 209 values change.

6/6 · 7/7
d13/bin-largedefects of FSLFSL-152: -bin of values from 2^24
command linefslmaths <big3: a synthetic input> -bin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-152 and names it: 1 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-152 and names it: 1 values change.

6/6 · 7/7
d13/bin-chardefects of FSLFSL-152: -bin under -dt char (max 127 + 1 wraps)
command linefslmaths -dt char <i8: a synthetic input> -bin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-152 and names it: 107 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-152 and names it: 107 values change.

6/6 · 7/7
d13/thr-infdefects of FSLFSL-152: -thr when the maximum is infinite
command linefslmaths <inf3: a synthetic input> -bin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-152 and names it: 1 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-152 and names it: 1 values change.

6/6 · 7/7
d13/uthr-chardefects of FSLFSL-152: -uthr under -dt char (min - 1 wraps)
command linefslmaths -dt char <i8: a synthetic input> -uthr 50 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-152 and names it: 148 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-152 and names it: 148 values change.

6/6 · 7/7
d13/nan-thrdefects of FSLFSL-153: -thr with a NaN first voxel
command linefslmaths <nan3: a synthetic input> -thr 0 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-153 and names it: 120 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-153 and names it: 120 values change.

6/6 · 7/7
d13/nan-bindefects of FSLFSL-153: -bin with a NaN first voxel
command linefslmaths <nan3: a synthetic input> -bin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-153 and names it: 120 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-153 and names it: 120 values change.

6/6 · 7/7
d13/nan-masdefects of FSLFSL-153: -mas by a mask with a NaN first voxel
command linefslmaths <f3: a synthetic input> -mas <nan3: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-153 and names it: 120 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-153 and names it: 120 values change.

6/6 · 7/7
d13/nan-rangedefects of FSLFSL-153: -range with a NaN first voxel
command linefslmaths <nan3: a synthetic input> -range out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, names FSL-153, with the same values

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, names FSL-153, with the same values.

6/6 · 7/7
d13/nan-thrpdefects of FSLFSL-153: -thrp with NaN voxels
command linefslmaths <nan3: a synthetic input> -thrp 30 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
d13/rem-zerodefects of FSLFSL-156: -rem 0 (FSL crashes)
command linefslmaths <f3: a synthetic input> -rem 0 out
Rejected by both

fslmaths exits -8; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-156.

Rejected by both

fslmaths exits -8; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-156.

5/5 · 4/4
d13/rem-halfdefects of FSLFSL-156: -rem 0.5 (FSL crashes)
command linefslmaths <f3: a synthetic input> -rem 0.5 out
Rejected by both

fslmaths exits -8; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-156.

Rejected by both

fslmaths exits -8; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-156.

5/5 · 4/4
d13/rem-image-halfdefects of FSLFSL-156: -rem by an image with a voxel 0.5
command linefslmaths <f3: a synthetic input> -rem <frac3: a synthetic input> out
Rejected by both

fslmaths exits -8; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-156.

Rejected by both

fslmaths exits -8; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-156.

5/5 · 4/4
d13/div-int-halfdefects of FSLFSL-156: -dt int -div 0.5 (FSL crashes)
command linefslmaths -dt int <s3: a synthetic input> -div 0.5 out
Rejected by both

fslmaths exits -8; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-156.

Rejected by both

fslmaths exits -8; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-156.

5/5 · 4/4
d13/grid-zerodefects of FSLFSL-156: -grid with spacing 0
command linefslmaths <f3: a synthetic input> -grid 1 0 out
Rejected by both

fslmaths exits -8; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-156.

Rejected by both

fslmaths exits -8; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-156.

5/5 · 4/4
d13/grid-nulldefects of FSLFSL-158: -grid without its spacing (reads the null pointer)
command linefslmaths <f3: a synthetic input> -grid out
Rejected by both

fslmaths exits -11; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-158.

Rejected by both

fslmaths exits -11; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-158.

5/5 · 4/4
d13/div-otherdefects of FSLFSL-159: -div by an image of another size
command linefslmaths <f3: a synthetic input> -div <other: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-159

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-159.

6/6 · 7/7
d13/rem-otherdefects of FSLFSL-159: -rem by a smaller image
command linefslmaths <f3: a synthetic input> -rem <smallint: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-159

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-159.

6/6 · 7/7
d13/max-otherdefects of FSLFSL-159: -max with a larger image
command linefslmaths <f3: a synthetic input> -max <other: a synthetic input> out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-159

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-159.

6/6 · 7/7
d13/min-smalldefects of FSLFSL-159: -min with a smaller image (FSL reads past its end)
command linefslmaths <f3: a synthetic input> -min <small: a synthetic input> out
Bit-identical

every file written byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-159

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-159.

2/2
morph/dilFmorphology-dilF (maximum filter, default kernel)
command linefslmaths <f3: a synthetic input> -dilF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/dilF-nanmorphology-dilF with NaN voxels (left out)
command linefslmaths <nan3: a synthetic input> -dilF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/dilMmorphology-dilM (mean dilation)
command linefslmaths <sparse3: a synthetic input> -dilM out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/dilM-nanmorphology-dilM with NaN neighbours
command linefslmaths <sparsenan3: a synthetic input> -dilM out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/dilDmorphology-dilD (modal dilation)
command linefslmaths <sparse3: a synthetic input> -dilD out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/dilD-shortmorphology-dilD of labels under -dt short
command linefslmaths -dt short <lab3: a synthetic input> -dilD out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/eromorphology-ero (zeroing next to zeros)
command linefslmaths <blob3: a synthetic input> -ero out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/eroFmorphology-eroF (minimum filter)
command linefslmaths <f3: a synthetic input> -eroF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/eroF-nanmorphology-eroF with NaN voxels
command linefslmaths <nan3: a synthetic input> -eroF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/fmedianmorphology-fmedian
command linefslmaths <f3: a synthetic input> -fmedian out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/fmedian-2Dmorphology-kernel 2D -fmedian
command linefslmaths <f3: a synthetic input> -kernel 2D -fmedian out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/fmedian-charmorphology-fmedian under -dt char
command linefslmaths -dt char <u8: a synthetic input> -fmedian out -odt char
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-151, FSL-154 and names it: 210 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-154 and names it: 210 values change.

6/6 · 7/7
morph/eroF-doublemorphology-eroF of a double image
command linefslmaths <d3: a synthetic input> -eroF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/sphere-4dmorphology-kernel sphere 4 -dilF on each volume of a 4D image
command linefslmaths <f4: a synthetic input> -kernel sphere 4 -dilF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/boxv-evenmorphology-kernel boxv 2: the off-centre warning
command linefslmaths <f3: a synthetic input> -kernel boxv 2 -dilF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/boxv3morphology-kernel boxv3 3 1 5 -fmedian
command linefslmaths <f3: a synthetic input> -kernel boxv3 3 1 5 -fmedian out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/box-mmmorphology-kernel box 4 (mm, anisotropic voxels) -eroF
command linefslmaths <odd3: a synthetic input> -kernel box 4 -eroF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/filemorphology-kernel file (values around 0.5) -dilF
command linefslmaths <f3: a synthetic input> -kernel file <kern3: a synthetic input> -dilF out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/file-evenmorphology-kernel file of even size -dilM
command linefslmaths <sparse3: a synthetic input> -kernel file <kern_even: a synthetic input> -dilM out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/file-4dmorphology-kernel file of a 4D image (the first volume, a warning)
command linefslmaths <f3: a synthetic input> -kernel file <kern4d: a synthetic input> -eroF out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/file-infmorphology-kernel file with an infinite value (Armadillo's bounds check)
command linefslmaths <f3: a synthetic input> -kernel file <kern_inf: a synthetic input> -dilF out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/2d-eromorphology-kernel 2D -ero of a 2D image
command linefslmaths <two2d: a synthetic input> -kernel 2D -ero out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
morph/2d-dilDmorphology-dilD of a 2D image
command linefslmaths <two2d: a synthetic input> -dilD out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dilall/3d-dilall-dilall
command linefslmaths <sparse3: a synthetic input> -dilall out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dilall/4d-dilall-thr 10 -dilall of a 4D image
command linefslmaths <f4: a synthetic input> -thr 10 -dilall out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dilall/int-dilall-dilall under -dt int
command linefslmaths -dt int <lab3: a synthetic input> -dilall out -odt int
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dilall/nan-dilall-dilall with NaN voxels
command linefslmaths <sparsenan3: a synthetic input> -dilall out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/skernels and filters-s 2 (separable Gaussian)
command linefslmaths <f3: a synthetic input> -s 2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/s-oddkernels and filters-s 1.5 of an odd, anisotropic image
command linefslmaths <odd3: a synthetic input> -s 1.5 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/s-4dkernels and filters-s 1 of a 4D image
command linefslmaths <f4: a synthetic input> -s 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/s-2dkernels and filters-s 2 of a 2D image
command linefslmaths <two2d: a synthetic input> -s 2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/s-nankernels and filters-s 1 with NaN voxels
command linefslmaths <nan3: a synthetic input> -s 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmean-defaultkernels and filters-fmean with the default kernel (not separable)
command linefslmaths <f3: a synthetic input> -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmean-boxkernels and filters-kernel box 5 -fmean (separable)
command linefslmaths <odd3: a synthetic input> -kernel box 5 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmean-boxv-evenkernels and filters-kernel boxv 4 -fmean (separable, even: warnings)
command linefslmaths <f3: a synthetic input> -kernel boxv 4 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmean-spherekernels and filters-kernel sphere 3 -fmean (direct sum)
command linefslmaths <odd3: a synthetic input> -kernel sphere 3 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmean-sphere-nankernels and filters-kernel sphere 4 -fmean with NaN voxels
command linefslmaths <nan3: a synthetic input> -kernel sphere 4 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmean-doublekernels and filters-kernel sphere 4 -fmean of a double image
command linefslmaths <d3: a synthetic input> -kernel sphere 4 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmean-filekernels and filters-kernel file -fmean (not separable)
command linefslmaths <f3: a synthetic input> -kernel file <kern3: a synthetic input> -fmean out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmean-file-evenkernels and filters-kernel file of even size -fmean (warnings)
command linefslmaths <f3: a synthetic input> -kernel file <kern_even: a synthetic input> -fmean out
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmean-2Dkernels and filters-kernel 2D -fmean
command linefslmaths <f3: a synthetic input> -kernel 2D -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/separable-keptkernels and filters-kernel gauss 1 -kernel 3D -fmean (separable kept from gauss)
command linefslmaths <f3: a synthetic input> -kernel gauss 1 -kernel 3D -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmeanukernels and filters-kernel sphere 5 -fmeanu
command linefslmaths <f3: a synthetic input> -kernel sphere 5 -fmeanu out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmeanu-gausskernels and filters-kernel gauss 2 -fmeanu (not separable)
command linefslmaths <f3: a synthetic input> -kernel gauss 2 -fmeanu out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fmeanu-charkernels and filters-kernel sphere 3 -fmeanu under -dt char
command linefslmaths -dt char <u8: a synthetic input> -kernel sphere 3 -fmeanu out -odt char
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-151, FSL-154 and names it: 203 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-154 and names it: 203 values change.

6/6 · 7/7
conv/fmean-intkernels and filters-kernel boxv 3 -fmean under -dt int (separable)
command linefslmaths -dt int <s3: a synthetic input> -kernel boxv 3 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fft-spherekernels and filters-kernel sphere 12 -fmean (the FFT path: radices 2, 3, 11, 13)
command linefslmaths <f3: a synthetic input> -kernel sphere 12 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: values within the sums’ bound: 9.57e-08): FSL convolves this kernel by its FFT in single precision, whose rounding is relative to the image’s largest values; the clean room by a direct sum whose products it adds in double precision, the sum rounded once to single precision (findings 8 and 31; spec question 14).

6/6 · 8/8
conv/fft-boxkernels and filters-kernel box 24 -fmeanu (the FFT path)
command linefslmaths <f3: a synthetic input> -kernel box 24 -fmeanu out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: values within the sums’ bound: 4.86e-08): FSL convolves this kernel by its FFT in single precision, whose rounding is relative to the image’s largest values; the clean room by a direct sum whose products it adds in double precision, the sum rounded once to single precision (findings 8 and 31; spec question 14).

6/6 · 8/8
conv/fft-evenkernels and filters-kernel boxv 12 -fmeanu (the FFT path, an even kernel: radix 4)
command linefslmaths <f3: a synthetic input> -kernel boxv 12 -fmeanu out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: values within the sums’ bound: 8.08e-08): FSL convolves this kernel by its FFT in single precision, whose rounding is relative to the image’s largest values; the clean room by a direct sum whose products it adds in double precision, the sum rounded once to single precision (findings 8 and 31; spec question 14); in fix, its default, the clean room corrects FSL-192 and names it: 60 values change.

6/6 · 8/8
conv/fft-oddkernels and filters-kernel sphere 10 -fmean of an odd, anisotropic image (the FFT path: radix 5)
command linefslmaths <odd3: a synthetic input> -kernel sphere 10 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: values within the sums’ bound: 2.04e-07): FSL convolves this kernel by its FFT in single precision, whose rounding is relative to the image’s largest values; the clean room by a direct sum whose products it adds in double precision, the sum rounded once to single precision (findings 8 and 31; spec question 14).

6/6 · 8/8
conv/fft-doublekernels and filters-kernel sphere 12 -fmean of a double image (the FFT path)
command linefslmaths <d3: a synthetic input> -kernel sphere 12 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Expected divergence

FSL’s single-precision FFT convolution against the clean room’s direct sum (where FSL takes its FFT, its products added in double precision and rounded once): the FFT’s rounding is relative to the image’s largest values, not to each voxel’s own last place; against the exact result (float64) FSL’s FFT lies within 6.9e-8 to 1.0e-7 of the largest value on these cases, the clean room within 4.1e-8 to 6.3e-8 (findings 8 and 31; spec question 14). Inside the bound: max |diff| 1.13e-07 of max |FSL| (bound 1e-05). Outside the threshold: out.nii.gz: the sums’ share … Findings 31: against the exact result FSL’s FFT lies within 8e-08 of the largest value, the clean room’s direct sum (its products in double precision, rounded once) within 6.3e-08: the difference is FSL’s single-precision FFT’s.

6/6 · 7/8
conv/fft-4dkernels and filters-kernel sphere 12 -fmeanu of a 4D image (the FFT path)
command linefslmaths <f4: a synthetic input> -kernel sphere 12 -fmeanu out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Expected divergence

FSL’s single-precision FFT convolution against the clean room’s direct sum (where FSL takes its FFT, its products added in double precision and rounded once): the FFT’s rounding is relative to the image’s largest values, not to each voxel’s own last place; against the exact result (float64) FSL’s FFT lies within 6.9e-8 to 1.0e-7 of the largest value on these cases, the clean room within 4.1e-8 to 6.3e-8 (findings 8 and 31; spec question 14). Inside the bound: max |diff| 8.6e-08 of max |FSL| (bound 1e-05). Outside the threshold: out.nii.gz: the sums’ share o… Findings 31: against the exact result FSL’s FFT lies within 6.9e-08 of the largest value, the clean room’s direct sum (its products in double precision, rounded once) within 4.1e-08: the difference is FSL’s single-precision FFT’s.

6/6 · 7/8
conv/fft-intkernels and filters-kernel sphere 12 -fmeanu under -dt int (the FFT path)
command linefslmaths -dt int <s3: a synthetic input> -kernel sphere 12 -fmeanu out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/fft-nankernels and filters-kernel sphere 12 -fmeanu with NaN voxels (the FFT path)
command linefslmaths <nan3: a synthetic input> -kernel sphere 12 -fmeanu out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
conv/kernel-no-outputkernels and filters-kernel box 3 without an output name
command linefslmaths <f3: a synthetic input> -kernel box 3
Rejected by both

fslmaths exits 1; the same from the replica. Error: no output filename specified!

Rejected by both

fslmaths exits 1; the same from the clean room. Error: no output filename specified!

5/5 · 4/4
conv/kernel-missing-sizekernels and filters-kernel box with the output name as its size: no output name left
command linefslmaths <f3: a synthetic input> -kernel box out
Rejected by both

fslmaths exits 1; the same from the replica. Error: no output filename specified!

Rejected by both

fslmaths exits 1; the same from the clean room. Error: no output filename specified!

5/5 · 4/4
morph/file-inf2morphology-kernel file with an infinite value and none at 0.5: more values than the vector holds
command linefslmaths <f3: a synthetic input> -kernel file <kern_inf2: a synthetic input> -dilF out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds

5/5 · 4/4
sub/subsamp2subsampling-subsamp2
command linefslmaths <f3: a synthetic input> -subsamp2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: values within the sums’ bound: 4.3e-08): FSL’s subsampling arithmetic under the floating types is not known: no order of a single or double sum of the voxels gives FSL’s last bits (findings 13 and 29; spec questions 13 and 38).

6/6 · 8/8
sub/subsamp2offcsubsampling-subsamp2offc
command linefslmaths <f3: a synthetic input> -subsamp2offc out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Expected divergence

FSL’s arithmetic of -subsamp2offc’s 2 x 2 x 2 means, which no order of a single or double sum of the eight values gives (findings 13 and 29; spec question 13), against the spec’s mean: the voxels beyond 4 ulp are those whose value cancels to near 0. Inside th…

6/6 · 7/8
sub/subsamp2-shortsubsampling-subsamp2 under -dt short
command linefslmaths -dt short <s3: a synthetic input> -subsamp2 out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
sub/subsamp2offc-intsubsampling-subsamp2offc under -dt int
command linefslmaths -dt int <s3: a synthetic input> -subsamp2offc out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
sub/subsamp2offc-doublesubsampling-subsamp2offc of a double image
command linefslmaths <d3: a synthetic input> -subsamp2offc out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Expected divergence

FSL’s arithmetic of -subsamp2offc’s 2 x 2 x 2 means, which no order of a single or double sum of the eight values gives (findings 13 and 29; spec question 13), against the spec’s mean: the voxels beyond 4 ulp are those whose value cancels to near 0. Inside th…

6/6 · 7/8
sub/subsamp2-noformsubsampling-subsamp2 without sform or qform
command linefslmaths <noform: a synthetic input> -subsamp2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: values within the sums’ bound: 4.3e-08): as sub/subsamp2: FSL’s subsampling arithmetic under the floating types (spec questions 13 and 38).

6/6 · 8/8
sub/subsamp2-nansubsampling-subsamp2 with NaN voxels
command linefslmaths <nan3: a synthetic input> -subsamp2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: values within the sums’ bound: 5.17e-08): as sub/subsamp2: FSL’s subsampling arithmetic under the floating types (spec questions 13 and 38).

6/6 · 8/8
edge/3d-edge-edge
command linefslmaths <f3: a synthetic input> -edge out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
edge/4d-edge-edge of a 4D image
command linefslmaths <f4: a synthetic input> -edge out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
edge/2d-edge-edge of a 2D image
command linefslmaths <two2d: a synthetic input> -edge out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
edge/short-edge-edge under -dt short
command linefslmaths -dt short <s3: a synthetic input> -edge out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
edge/nan-edge-edge with NaN voxels
command linefslmaths <nan3: a synthetic input> -edge out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
edge/aniso-edge-edge with anisotropic voxels
command linefslmaths <odd3: a synthetic input> -edge out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dog/3d-dog_edge-dog_edge 1 1.6
command linefslmaths <f3: a synthetic input> -dog_edge 1 1.6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dog/aniso-dog_edge-dog_edge 0.5 2 of an odd image
command linefslmaths <odd3: a synthetic input> -dog_edge 0.5 2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dog/double-dog_edge-dog_edge of a double image
command linefslmaths <d3: a synthetic input> -dog_edge 1 1.6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dog/int-dog_edge-dog_edge under -dt int
command linefslmaths -dt int <s3: a synthetic input> -dog_edge 1 1.6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dog/radius0-dog_edge-dog_edge 0.1 0.2 (radius 0)
command linefslmaths <f3: a synthetic input> -dog_edge 0.1 0.2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fillh/6-fillh-fillh
command linefslmaths <holes3: a synthetic input> -fillh out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fillh/26-fillh-fillh26
command linefslmaths <holes3: a synthetic input> -fillh26 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fillh/blob-fillh-fillh of a blob at the edge
command linefslmaths <blob3: a synthetic input> -fillh out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fillh/short-fillh-fillh under -dt short
command linefslmaths -dt short <holes3: a synthetic input> -fillh out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
fillh/4d-fillh-fillh26 of a 4D mask
command linefslmaths <m4: a synthetic input> -fillh26 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
norm/inmintensity normalisation-inm 100 of a 4D image
command linefslmaths <f4: a synthetic input> -inm 100 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
norm/ingintensity normalisation-ing 100 of a 4D image
command linefslmaths <f4: a synthetic input> -ing 100 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
norm/inm-sparseintensity normalisation-inm 100 of a sparse image
command linefslmaths <sparse3: a synthetic input> -inm 100 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
norm/inm-emptyintensity normalisation-inm of an image of zeros (an empty mask: NaN)
command linefslmaths <sparse3: a synthetic input> -mul 0 -inm 100 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
norm/ing-intintensity normalisation-ing under -dt int
command linefslmaths -dt int <s3: a synthetic input> -ing 1000 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
norm/inm-nanintensity normalisation-inm with NaN voxels
command linefslmaths <nan3: a synthetic input> -inm 100 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-153 and names it: 208 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-153 and names it: 208 values change.

6/6 · 7/7
tensor/decomp-tensor_decomp-tensor_decomp
command linefslmaths <tensor: a synthetic input> -tensor_decomp out
Bit-identical

every file written (11/11) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out_V1.nii.gz: eigenvectors up to their sign: 1.000000; out_V2.nii.gz: eigenvectors up to their sign: 1.000000; out_V3.nii.gz: eigenvectors up to their sign: 1.000000): the eigenvalues, FA, MD and mode are FSL’s bit for bit; the eigenvectors are the same up to their sign (|cos| = 1.000000): FSL’s signs are LAPACK’s, which the specification does not define (spec question 20), the clean room’s each one’s largest component positive.

6/6 · 25/25
tensor/chain-tensor_decomp-tensor_decomp, then the FA image multiplied by 2
command linefslmaths <tensor: a synthetic input> -tensor_decomp -mul 2 out
Bit-identical

every file written (11/11) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out_V1.nii.gz: eigenvectors up to their sign: 1.000000; out_V2.nii.gz: eigenvectors up to their sign: 1.000000; out_V3.nii.gz: eigenvectors up to their sign: 1.000000): as tensor/decomp: the eigenvectors’ signs (spec question 20).

6/6 · 25/25
tensor/double-tensor_decomp-tensor_decomp under -dt double, an output name with its extension
command linefslmaths -dt double <tensor: a synthetic input> -tensor_decomp out.nii.gz
Bit-identical

every file written (11/11) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz_V1.nii.gz: eigenvectors up to their sign: 1.000000; out.nii.gz_V2.nii.gz: eigenvectors up to their sign: 1.000000; out.nii.gz_V3.nii.gz: eigenvectors up to their sign: 1.000000): as tensor/decomp: the eigenvectors’ signs (spec question 20).

6/6 · 25/25
d13/tensor-nandefects of FSLFSL-166: -tensor_decomp with a NaN tensor
command linefslmaths <tensornan: a synthetic input> -tensor_decomp out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds The replica’s default, fix, finishes instead (exit status 0), naming FSL-166.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds The clean room’s default, fix, finishes instead (exit status 0), naming FSL-166.

5/5 · 4/4
tensor/not-a-tensor-tensor_decomp-tensor_decomp of a 3-volume image
command linefslmaths <f4: a synthetic input> -tensor_decomp out
Rejected by both

fslmaths exits 1; the same from the replica.

Rejected by both

fslmaths exits 1; the same from the clean room.

5/5 · 4/4
d13/kernel-boxv3-nulldefects of FSLFSL-158: -kernel boxv3 without its sizes (reads the null pointer)
command linefslmaths <f3: a synthetic input> -kernel boxv3 3 out
Rejected by both

fslmaths exits -11; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-158.

Rejected by both

fslmaths exits -11; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-158.

5/5 · 4/4
d13/subsamp2-4ddefects of FSLFSL-163: -subsamp2 of a 4D image
command linefslmaths <f4: a synthetic input> -subsamp2 out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #75 :: 3D only method called by higher-dimensional volume. terminate called after throwing an instance of ‘std::runtime_error’ what(): 3D only method called by higher-dimensional vol The replica’s default, fix, finishes instead (exit status 0), naming FSL-163.

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #75 :: 3D only method called by higher-dimensional volume. terminate called after throwing an instance of ‘std::runtime_error’ what(): 3D only The clean room’s default, fix, finishes instead (exit status 0), naming FSL-163.

5/5 · 4/4
d13/subsamp2offc-4d-odddefects of FSLFSL-163: -subsamp2offc of a 4D image of odd sizes
command linefslmaths <f4: a synthetic input> -subsamp2offc out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #75 :: 3D only method called by higher-dimensional volume. terminate called after throwing an instance of ‘std::runtime_error’ what(): 3D only method called by higher-dimensional vol The replica’s default, fix, finishes instead (exit status 0), naming FSL-163.

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #75 :: 3D only method called by higher-dimensional volume. terminate called after throwing an instance of ‘std::runtime_error’ what(): 3D only The clean room’s default, fix, finishes instead (exit status 0), naming FSL-163.

5/5 · 4/4
d13/subsamp2offc-4d-evendefects of FSLFSL-163: -subsamp2offc of a 4D image of even sizes (the first volume)
command linefslmaths <even4: a synthetic input> -subsamp2offc out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-163 and names it: 24 values change

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: values within the sums’ bound: 4.58e-08): FSL’s arithmetic of -subsamp2offc’s 2 × 2 × 2 means (findings 13 and 29; spec question 13); in fix, its default, the clean room corrects FSL-163 and names it: 24 values change.

6/6 · 8/8
d13/dog-4ddefects of FSLFSL-164: -dog_edge of a 4D image
command linefslmaths <f4: a synthetic input> -dog_edge 1 1.6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-164 and names it: 30 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-164 and names it: 30 values change.

6/6 · 7/7
d13/dog-negativedefects of FSLFSL-161: -dog_edge -1 -2 (a kernel without voxels)
command linefslmaths <f3: a synthetic input> -dog_edge -1 -2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161.

6/6 · 7/7
d13/fillh-2ddefects of FSLFSL-165: -fillh of a 2D image
command linefslmaths <blob2d: a synthetic input> -fillh out
Rejected by both

fslmaths exits -8; the same from the replica. The replica’s default, fix, finishes instead (exit status 0), naming FSL-165.

Rejected by both

fslmaths exits -8; the same from the clean room. The clean room’s default, fix, finishes instead (exit status 0), naming FSL-165.

5/5 · 4/4
d13/s-zerodefects of FSLFSL-160: -s 0
command linefslmaths <f3: a synthetic input> -s 0 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-160 and names it: 210 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-160 and names it: 210 values change.

6/6 · 7/7
d13/gauss-zerodefects of FSLFSL-160: -kernel gauss 0 -fmean
command linefslmaths <f3: a synthetic input> -kernel gauss 0 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-160 and names it: 210 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-160 and names it: 210 values change.

6/6 · 7/7
d13/gauss-zero-dilFdefects of FSLFSL-160: -kernel gauss 0 -dilF (no change: no notice)
command linefslmaths <f3: a synthetic input> -kernel gauss 0 -dilF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
d13/kernel-boxv0defects of FSLFSL-161: -kernel boxv 0 -dilF
command linefslmaths <f3: a synthetic input> -kernel boxv 0 -dilF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161.

6/6 · 7/7
d13/kernel-negativedefects of FSLFSL-161: -kernel boxv -3 -dilF
command linefslmaths <f3: a synthetic input> -kernel boxv -3 -dilF out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161.

6/6 · 7/7
d13/kernel-unknowndefects of FSLFSL-161: -kernel foo 3 -fmean
command linefslmaths <f3: a synthetic input> -kernel foo 3 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161.

6/6 · 7/7
d13/kernel-unknown-dilMdefects of FSLFSL-161: -kernel foo 3 -dilM
command linefslmaths <f3: a synthetic input> -kernel foo 3 -dilM out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161.

6/6 · 7/7
d13/gauss-negativedefects of FSLFSL-161: -kernel gauss -1 -fmean
command linefslmaths <f3: a synthetic input> -kernel gauss -1 -fmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161.

6/6 · 7/7
d13/s-negativedefects of FSLFSL-161: -s -1
command linefslmaths <f3: a synthetic input> -s -1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-161.

6/6 · 7/7
d13/fmedian-nandefects of FSLFSL-162: -fmedian with NaN voxels
command linefslmaths <nan3: a synthetic input> -fmedian out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The replica’s default, fix, finishes instead (exit status 0), naming FSL-162.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The clean room’s default, fix, finishes instead (exit status 0), naming FSL-162.

5/5 · 4/4
d13/dilD-nandefects of FSLFSL-162: -dilD with a NaN next to a zero voxel
command linefslmaths <sparsenan3: a synthetic input> -dilD out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The replica’s default, fix, finishes instead (exit status 0), naming FSL-162.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The clean room’s default, fix, finishes instead (exit status 0), naming FSL-162.

5/5 · 4/4
d13/fmean-int-trapdefects of FSLFSL-156: -kernel sphere 5 -fmean under -dt int (an integer division by zero)
command linefslmaths -dt int <s3: a synthetic input> -kernel sphere 5 -fmean out
Rejected by both

fslmaths exits -8; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-156.

Rejected by both

fslmaths exits -8; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-156.

5/5 · 4/4
dim/Tmeanreductions along a dimension-Tmean
command linefslmaths <tsc: a synthetic input> -Tmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tstdreductions along a dimension-Tstd
command linefslmaths <tsc: a synthetic input> -Tstd out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tmaxreductions along a dimension-Tmax
command linefslmaths <tsc: a synthetic input> -Tmax out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tminreductions along a dimension-Tmin
command linefslmaths <tsc: a synthetic input> -Tmin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tmaxnreductions along a dimension-Tmaxn (ties: the first index)
command linefslmaths <tsc: a synthetic input> -Tmaxn out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tmedianreductions along a dimension-Tmedian (ties, signed zeros)
command linefslmaths <tsc: a synthetic input> -Tmedian out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tpercreductions along a dimension-Tperc 30
command linefslmaths <tsc: a synthetic input> -Tperc 30 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tperc-0reductions along a dimension-Tperc 0
command linefslmaths <tsc: a synthetic input> -Tperc 0 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tperc-100reductions along a dimension-Tperc 100
command linefslmaths <tsc: a synthetic input> -Tperc 100 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tar1reductions along a dimension-Tar1
command linefslmaths <tsc: a synthetic input> -Tar1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tmean-nanreductions along a dimension-Tmean of series holding NaN
command linefslmaths <ts: a synthetic input> -Tmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tar1-nanreductions along a dimension-Tar1 of series holding NaN
command linefslmaths <ts: a synthetic input> -Tar1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Xmeanreductions along a dimension-Xmean
command linefslmaths <tsc: a synthetic input> -Xmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Ymaxreductions along a dimension-Ymax
command linefslmaths <tsc: a synthetic input> -Ymax out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Zstdreductions along a dimension-Zstd
command linefslmaths <tsc: a synthetic input> -Zstd out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Xmaxnreductions along a dimension-Xmaxn
command linefslmaths <tsc: a synthetic input> -Xmaxn out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Ymedianreductions along a dimension-Ymedian
command linefslmaths <tsc: a synthetic input> -Ymedian out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Zpercreductions along a dimension-Zperc 75
command linefslmaths <tsc: a synthetic input> -Zperc 75 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Xar1reductions along a dimension-Xar1
command linefslmaths <tsc: a synthetic input> -Xar1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Yminreductions along a dimension-Ymin
command linefslmaths <tsc: a synthetic input> -Ymin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Smeanreductions along a dimension-Smean (a capital that reduces nothing)
command linefslmaths <tsc: a synthetic input> -Smean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/chainreductions along a dimension-Tmean then -Xmax then -add
command linefslmaths <tsc: a synthetic input> -Tmean -Xmax -add 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/unknownreductions along a dimension-Tfoo: an unknown reduction
command linefslmaths <tsc: a synthetic input> -Tfoo out
Rejected by both

fslmaths exits 1; the same from the replica. Error unknown Dimensionality operation: -Tfoo

Rejected by both

fslmaths exits 1; the same from the clean room. Error unknown Dimensionality operation: -Tfoo

5/5 · 4/4
dim/emptyreductions along a dimension-T alone
command linefslmaths <tsc: a synthetic input> -T out
Rejected by both

fslmaths exits 1; the same from the replica. Error unknown Dimensionality operation: -T

Rejected by both

fslmaths exits 1; the same from the clean room. Error unknown Dimensionality operation: -T

5/5 · 4/4
dim/perc-rangereductions along a dimension-Tperc 150 (newimage's exception)
command linefslmaths <tsc: a synthetic input> -Tperc 150 out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #4 :: Percentiles must be in the range [0.0,1.0] terminate called after throwing an instance of ‘std::runtime_error’ what(): Percentiles must be in the range [0.0,1.0]

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #4 :: Percentiles must be in the range [0.0,1.0] terminate called after throwing an instance of ‘std::runtime_error’ what(): Percentiles must

5/5 · 4/4
dim/perc-negativereductions along a dimension-Tperc -5
command linefslmaths <tsc: a synthetic input> -Tperc -5 out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #4 :: Percentiles must be in the range [0.0,1.0] terminate called after throwing an instance of ‘std::runtime_error’ what(): Percentiles must be in the range [0.0,1.0]

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #4 :: Percentiles must be in the range [0.0,1.0] terminate called after throwing an instance of ‘std::runtime_error’ what(): Percentiles must

5/5 · 4/4
dim/perc-nanreductions along a dimension-Tperc nan
command linefslmaths <tsc: a synthetic input> -Tperc nan out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/perc-no-argumentreductions along a dimension-Tperc with the output name as its argument
command linefslmaths <tsc: a synthetic input> -Tperc out
Rejected by both

fslmaths exits 1; the same from the replica. Error: no output filename specified!

Rejected by both

fslmaths exits 1; the same from the clean room. Error: no output filename specified!

5/5 · 4/4
dim/3d-meanreductions along a dimension-Tmean of a 3D image
command linefslmaths <f3: a synthetic input> -Tmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tmean-shortreductions along a dimension-Tmean under -dt short
command linefslmaths -dt short <tss: a synthetic input> -Tmean out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tstd-intreductions along a dimension-Tstd under -dt int
command linefslmaths -dt int <tss: a synthetic input> -Tstd out -odt int
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tar1-shortreductions along a dimension-Tar1 under -dt short
command linefslmaths -dt short <tss: a synthetic input> -Tar1 out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tar1-int-largereductions along a dimension-Tar1 under -dt int with products beyond int's range
command linefslmaths -dt int <tsi: a synthetic input> -Tar1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tmedian-intreductions along a dimension-Tmedian under -dt int
command linefslmaths -dt int <tsi: a synthetic input> -Tmedian out -odt int
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tmaxn-shortreductions along a dimension-Tmaxn under -dt short
command linefslmaths -dt short <tss: a synthetic input> -Tmaxn out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tstd-doublereductions along a dimension-Tstd of a double image
command linefslmaths <tsd: a synthetic input> -Tstd out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-167 and names it: 1 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-167 and names it: 1 values change.

6/6 · 7/7
dim/Tmedian-doublereductions along a dimension-Tmedian of a double image
command linefslmaths <tsd: a synthetic input> -Tmedian out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
dim/Tmean-longreductions along a dimension-Tmean of 48 time points
command linefslmaths <tsvar: a synthetic input> -Tmean out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/pvalp-values, ranks and z-pval
command linefslmaths <stat: a synthetic input> -pval out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/pval0p-values, ranks and z-pval0 (zeros left out)
command linefslmaths <stat0: a synthetic input> -pval0 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/cpvalp-values, ranks and z-cpval
command linefslmaths <stat: a synthetic input> -cpval out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/cpval-doublep-values, ranks and z-cpval of a double image
command linefslmaths <statd: a synthetic input> -cpval out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/cpval-intp-values, ranks and z-cpval under -dt int
command linefslmaths -dt int <stats16: a synthetic input> -cpval out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/pval-shortp-values, ranks and z-pval under -dt short
command linefslmaths -dt short <stats16: a synthetic input> -pval out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/pval-3dp-values, ranks and z-pval of a 3D image
command linefslmaths <f3: a synthetic input> -pval out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/ztopp-values, ranks and z-ztop (with +-40 and NaN)
command linefslmaths <zmap: a synthetic input> -ztop out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/ztop-doublep-values, ranks and z-ztop in double
command linefslmaths -dt double <zmap: a synthetic input> -ztop out -odt double
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: the sums’ share of voxels within 4 ulp: 99.5%; out.nii.gz: values within the sums’ bound: 1.11e-16): the last bits of Φ in double precision (the clean room’s Φ is the specification’s form with libm’s erf and erfc; spec question 35).

6/6 · 8/8
stat/ptozp-values, ranks and z-ptoz (with 0, 1, outside [0, 1] and NaN)
command linefslmaths <pmap: a synthetic input> -ptoz out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/ztop-ptozp-values, ranks and z-ztop then -ptoz
command linefslmaths <zmap: a synthetic input> -ztop -ptoz out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/rankp-values, ranks and z-rank (ties, signed zeros)
command linefslmaths <tsc: a synthetic input> -rank out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/ranknormp-values, ranks and z-ranknorm
command linefslmaths <tsc: a synthetic input> -ranknorm out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/rank-shortp-values, ranks and z-rank under -dt short
command linefslmaths -dt short <tss: a synthetic input> -rank out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/rank-3dp-values, ranks and z-rank of a 3D image
command linefslmaths <f3: a synthetic input> -rank out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
stat/ranknorm-doublep-values, ranks and z-ranknorm of a double image
command linefslmaths <tsd: a synthetic input> -ranknorm out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-167 and names it: 9 values change

Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: the sums’ share of voxels within 4 ulp: 99.9%; out.nii.gz: values within the sums’ bound: 1.67e-16): the last bits of Φ⁻¹ in double precision (spec questions 35 and 36); in fix, its default, the clean room corrects FSL-167 and names it: 9 values change.

6/6 · 8/8
roi/box-roi-roi of a box
command linefslmaths <tsc: a synthetic input> -roi 1 3 0 2 1 2 2 4 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
roi/full-roi-roi with every size -1
command linefslmaths <tsc: a synthetic input> -roi 0 -1 0 -1 0 -1 0 -1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
roi/neuro-roi-roi of a neurological image (x mapped to newimage's order)
command linefslmaths <neuro: a synthetic input> -roi 1 3 2 2 0 -1 0 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
roi/outside-roi-roi of a box outside the image
command linefslmaths <tsc: a synthetic input> -roi 10 2 0 1 0 1 0 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
roi/negative-size-roi-roi with a size of -3
command linefslmaths <tsc: a synthetic input> -roi 4 -3 0 -1 0 -1 0 -1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
roi/into-output-roi-roi with the output name as its last argument
command linefslmaths <f3: a synthetic input> -roi 0 1 0 1 0 1 0 out
Rejected by both

fslmaths exits 1; the same from the replica. Error: no output filename specified!

Rejected by both

fslmaths exits 1; the same from the clean room. Error: no output filename specified!

5/5 · 4/4
roi/null-roiFSL-158: -roi without its last sizes (reads the null pointer)
command linefslmaths <f3: a synthetic input> -roi 1 2 out
Rejected by both

fslmaths exits -11; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-158.

Rejected by both

fslmaths exits -11; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-158.

5/5 · 4/4
bptf/hp-bptf-bptf 2 -1 (high-pass)
command linefslmaths <tsc: a synthetic input> -bptf 2 -1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/lp-bptf-bptf -1 1.5 (low-pass)
command linefslmaths <tsc: a synthetic input> -bptf -1 1.5 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/both-bptf-bptf 3 0.8
command linefslmaths <tsc: a synthetic input> -bptf 3 0.8 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/none-bptf-bptf 0 0
command linefslmaths <tsc: a synthetic input> -bptf 0 0 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/nan-bptf-bptf nan nan (both skipped)
command linefslmaths <tsc: a synthetic input> -bptf nan nan out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/wide-bptf-bptf 100 50 (wider than the series)
command linefslmaths <tsc: a synthetic input> -bptf 100 50 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/3d-bptf-bptf of a 3D image
command linefslmaths <f3: a synthetic input> -bptf 2 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/long-bptf-bptf 5 2 of 48 time points
command linefslmaths <tsvar: a synthetic input> -bptf 5 2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/short-bptf-bptf under -dt short
command linefslmaths -dt short <tss: a synthetic input> -bptf 2 1 out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/double-bptf-bptf of a double image
command linefslmaths <tsd: a synthetic input> -bptf 2.5 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/nan-values-bptf-bptf of series holding NaN
command linefslmaths <ts: a synthetic input> -bptf 2 1 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
bptf/into-output-bptf-bptf with the output name as its second sigma
command linefslmaths <tsc: a synthetic input> -bptf 2 out
Rejected by both

fslmaths exits 1; the same from the replica. Error: no output filename specified!

Rejected by both

fslmaths exits 1; the same from the clean room. Error: no output filename specified!

5/5 · 4/4
d13/Tmax-nandefects of FSLFSL-153: -Tmax with a NaN first value
command linefslmaths <ts: a synthetic input> -Tmax out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-153 and names it: 1 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-153 and names it: 1 values change.

6/6 · 7/7
d13/Tmin-nandefects of FSLFSL-153: -Tmin with a NaN first value
command linefslmaths <ts: a synthetic input> -Tmin out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-153 and names it: 1 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-153 and names it: 1 values change.

6/6 · 7/7
d13/Tmaxn-nandefects of FSLFSL-153: -Tmaxn with a NaN first value
command linefslmaths <ts: a synthetic input> -Tmaxn out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-153 and names it: 1 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-153 and names it: 1 values change.

6/6 · 7/7
d13/Tmedian-nandefects of FSLFSL-168: -Tmedian of values holding NaN
command linefslmaths <ts: a synthetic input> -Tmedian out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-168 and names it: 3 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-168 and names it: 3 values change.

6/6 · 7/7
d13/Tperc-nandefects of FSLFSL-168: -Tperc 80 of values holding NaN
command linefslmaths <ts: a synthetic input> -Tperc 80 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-168 and names it: 2 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-168 and names it: 2 values change.

6/6 · 7/7
d13/Tstd-3ddefects of FSLFSL-167: -Tstd of a 3D image
command linefslmaths <f3: a synthetic input> -Tstd out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-167 and names it: 210 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-167 and names it: 210 values change.

6/6 · 7/7
d13/Tstd-negativedefects of FSLFSL-167: -Tstd of series whose variance rounds below 0
command linefslmaths <tsvar: a synthetic input> -Tstd out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-167 and names it: 2 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-167 and names it: 2 values change.

6/6 · 7/7
d13/ranknorm-negativedefects of FSLFSL-167: -ranknorm of series whose variance rounds below 0
command linefslmaths <tsvar: a synthetic input> -ranknorm out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-167 and names it: 96 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-167 and names it: 96 values change.

6/6 · 7/7
d13/ranknorm-3ddefects of FSLFSL-167: -ranknorm of a 3D image
command linefslmaths <f3: a synthetic input> -ranknorm out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-167 and names it: 210 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-167 and names it: 210 values change.

6/6 · 7/7
d13/rank-nandefects of FSLFSL-162: -rank of series holding NaN
command linefslmaths <ts: a synthetic input> -rank out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The replica’s default, fix, finishes instead (exit status 0), naming FSL-162.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The clean room’s default, fix, finishes instead (exit status 0), naming FSL-162.

5/5 · 4/4
d13/ranknorm-nandefects of FSLFSL-162: -ranknorm of series holding NaN
command linefslmaths <ts: a synthetic input> -ranknorm out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The replica’s default, fix, finishes instead (exit status 0), naming FSL-162.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The clean room’s default, fix, finishes instead (exit status 0), naming FSL-162.

5/5 · 4/4
d13/pval-nandefects of FSLFSL-169: -pval with NaN statistics
command linefslmaths <statnan: a synthetic input> -pval out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-169 and names it: 3 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-169 and names it: 3 values change.

6/6 · 7/7
d13/pval0-nandefects of FSLFSL-169: -pval0 with NaN statistics
command linefslmaths <statnan: a synthetic input> -pval0 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-169 and names it: 3 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-169 and names it: 3 values change.

6/6 · 7/7
d13/cpval-nandefects of FSLFSL-169: -cpval with NaN statistics
command linefslmaths <statnan: a synthetic input> -cpval out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-169 and names it: 57 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-169 and names it: 57 values change.

6/6 · 7/7
d13/ztop-infdefects of FSLFSL-170: -ztop of infinite z
command linefslmaths <zinf: a synthetic input> -ztop out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-170 and names it: 2 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-170 and names it: 2 values change.

6/6 · 7/7
d13/ptoz-tinydefects of FSLFSL-172: -ptoz of p-values whose 1 - p rounds to 1
command linefslmaths <ptiny: a synthetic input> -ptoz out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-172 and names it: 3 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-172 and names it: 3 values change.

6/6 · 7/7
d13/bptf-lp-hugedefects of FSLFSL-171: -bptf -1 2e8 (FSL crashes)
command linefslmaths <tsc: a synthetic input> -bptf -1 2e8 out
Rejected by both

fslmaths exits -11; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-171.

Rejected by both

fslmaths exits -11; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-171.

5/5 · 4/4
d13/bptf-hp-lengthdefects of FSLFSL-171: -bptf 4e8 -1 (std::bad_array_new_length)
command linefslmaths <tsc: a synthetic input> -bptf 4e8 -1 out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::bad_array_new_length’ what(): std::bad_array_new_length The replica’s default, fix, ends with exit status 1, naming FSL-171.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::bad_array_new_length’ what(): std::bad_array_new_length The clean room’s default, fix, ends with exit status 1, naming FSL-171.

5/5 · 4/4
d13/bptf-lp-lengthdefects of FSLFSL-171: -bptf -1 6e7 (std::bad_array_new_length)
command linefslmaths <tsc: a synthetic input> -bptf -1 6e7 out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::bad_array_new_length’ what(): std::bad_array_new_length The replica’s default, fix, ends with exit status 1, naming FSL-171.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::bad_array_new_length’ what(): std::bad_array_new_length The clean room’s default, fix, ends with exit status 1, naming FSL-171.

5/5 · 4/4
d13/bptf-hp-infdefects of FSLFSL-171: -bptf inf -1 (FSL writes outside its array)
command linefslmaths <tsc: a synthetic input> -bptf inf -1 out
Bit-identical

every file written byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-171

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-171.

2/2
tfce/basic-tfce-tfce 2 0.5 6
command linefslmaths <tfz: a synthetic input> -tfce 2 0.5 6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/conn18-tfce-tfce 2 0.5 18
command linefslmaths <tfz: a synthetic input> -tfce 2 0.5 18 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/conn26-tfce-tfce 2 0.5 26
command linefslmaths <tfz: a synthetic input> -tfce 2 0.5 26 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/minT-tfce-tfce_minT 1 -tfce 2 0.5 6
command linefslmaths <tfz: a synthetic input> -tfce_minT 1 -tfce 2 0.5 6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/delta-tfce-tfce_delta 0.3 -tfce 1.5 0.66 26
command linefslmaths <tfz: a synthetic input> -tfce_delta 0.3 -tfce 1.5 0.66 26 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/negative-minT-tfce-tfce_minT -1 -tfce 0.5 0.5 6 (powf of negative heights: NaN)
command linefslmaths <tfz: a synthetic input> -tfce_minT -1 -tfce 0.5 0.5 6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/many-steps-tfce-tfce_delta 0.0004: the warning of many integral steps
command linefslmaths <tfz: a synthetic input> -tfce_delta 0.0004 -tfce 2 0.5 6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/4d-tfce-tfce of each volume of a 4D image
command linefslmaths <tfz4: a synthetic input> -tfce 2 1 6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/double-tfce-tfce of a double image
command linefslmaths <tfd: a synthetic input> -tfce 2 0.5 6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/short-tfce-tfce under -dt short
command linefslmaths -dt short <tfs: a synthetic input> -tfce 1 0.5 6 out -odt short
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/inf-tfce-tfce with an infinite maximum
command linefslmaths <tfinf: a synthetic input> -tfce 2 0.5 6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfce/small-tfce-tfce of an image 2 voxels wide (tfce's error)
command linefslmaths <tfsmall: a synthetic input> -tfce 2 0.5 6 out
Rejected by both

fslmaths exits 1; the same from the replica. ERROR: TFCE failed, please check your file for valid sizes and voxel values.Error: tfce currently requires an input with at least 3 voxels extent into each dimension. Exiting

Rejected by both

fslmaths exits 1; the same from the clean room. ERROR: TFCE failed, please check your file for valid sizes and voxel values.Error: tfce currently requires an input with at least 3 voxels extent into each dime

5/5 · 4/4
tfce/negative-delta-tfce-tfce_delta -1 (tfce's error)
command linefslmaths <tfz: a synthetic input> -tfce_delta -1 -tfce 2 0.5 6 out
Rejected by both

fslmaths exits 1; the same from the replica. ERROR: TFCE failed, please check your file for valid sizes and voxel values.Error: tfce requires a positive deltaT input. Exiting

Rejected by both

fslmaths exits 1; the same from the clean room. ERROR: TFCE failed, please check your file for valid sizes and voxel values.Error: tfce requires a positive deltaT input. Exiting

5/5 · 4/4
tfce/into-output-tfce-tfce with the output name as its connectivity
command linefslmaths <tfz: a synthetic input> -tfce 2 0.5 out
Rejected by both

fslmaths exits 1; the same from the replica. Error: no output filename specified! The replica’s default, fix, ends with exit status 1, naming FSL-176.

Rejected by both

fslmaths exits 1; the same from the clean room. Error: no output filename specified! The clean room’s default, fix, ends with exit status 1, naming FSL-176.

5/5 · 4/4
tfce/chain-tfce-tfce then -mul (the options kept between operations)
command linefslmaths <tfz: a synthetic input> -tfce_delta 0.25 -tfce 2 0.5 6 -mul 2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfceS/basic-tfceS-tfceS 2 0.5 6 3 3 3 30
command linefslmaths <tfz: a synthetic input> -tfceS 2 0.5 6 3 3 3 30 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfceS/conn26-tfceS-tfceS 2 0.5 26 7 6 5 20
command linefslmaths <tfz: a synthetic input> -tfceS 2 0.5 26 7 6 5 20 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfceS/not-reached-tfceS-tfceS with a threshold its sum never reaches (the message)
command linefslmaths <tfz: a synthetic input> -tfceS 2 0.5 6 3 3 3 1e9 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfceS/delta-tfceS-tfce_delta 0.35 -tfceS (the arrays sized from the step)
command linefslmaths <tfz: a synthetic input> -tfce_delta 0.35 -tfceS 2 0.5 6 3 3 3 15 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfceS/4d-tfceS-tfceS of each volume of a 4D image
command linefslmaths <tfz4: a synthetic input> -tfceS 2 0.5 6 3 3 3 10 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfceS/double-tfceS-tfceS of a double image
command linefslmaths <tfd: a synthetic input> -tfceS 1 1 18 7 6 5 2 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfceS/low-voxel-tfceS-tfceS at a voxel without positive value
command linefslmaths <tfz: a synthetic input> -tfceS 2 0.5 6 0 8 0 5 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 7/7
tfceS/inf-tfceS-tfceS at an infinite voxel (in no cluster: std::out_of_range)
command linefslmaths <tfinf: a synthetic input> -tfceS 2 0.5 6 3 3 3 5 out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds

5/5 · 4/4
d13/tfce-zerodefects of FSLFSL-175: -tfce of an image of zeros
command linefslmaths <tfzero: a synthetic input> -tfce 2 0.5 6 out
Rejected by both

fslmaths exits 1; the same from the replica. ERROR: TFCE failed, please check your file for valid sizes and voxel values.Error: tfce requires a positive deltaT input. Exiting The replica’s default, fix, finishes instead (exit status 0), naming FSL-175.

Rejected by both

fslmaths exits 1; the same from the clean room. ERROR: TFCE failed, please check your file for valid sizes and voxel values.Error: tfce requires a positive deltaT input. Exiting The clean room’s default, fix, finishes instead (exit status 0), naming FSL-175.

5/5 · 4/4
d13/tfce-negativedefects of FSLFSL-175: -tfce of an image without positive values
command linefslmaths <tfneg: a synthetic input> -tfce 2 0.5 6 out
Rejected by both

fslmaths exits 1; the same from the replica. ERROR: TFCE failed, please check your file for valid sizes and voxel values.Error: tfce requires a positive deltaT input. Exiting The replica’s default, fix, finishes instead (exit status 0), naming FSL-175.

Rejected by both

fslmaths exits 1; the same from the clean room. ERROR: TFCE failed, please check your file for valid sizes and voxel values.Error: tfce requires a positive deltaT input. Exiting The clean room’s default, fix, finishes instead (exit status 0), naming FSL-175.

5/5 · 4/4
d13/tfce-nandefects of FSLFSL-153: -tfce with a NaN first voxel
command linefslmaths <tfnan: a synthetic input> -tfce 2 0.5 6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-153 and names it: 201 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-153 and names it: 201 values change.

6/6 · 7/7
d13/tfce-conn7defects of FSLFSL-176: -tfce 2 0.5 7
command linefslmaths <tfz: a synthetic input> -tfce 2 0.5 7 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-176

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-176.

6/6 · 7/7
d13/tfceS-conn7defects of FSLFSL-176: -tfceS with connectivity 7
command linefslmaths <tfz: a synthetic input> -tfceS 2 0.5 7 3 3 3 30 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-176

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-176.

6/6 · 7/7
d13/tfce-delta-nandefects of FSLFSL-174: -tfce_delta nan (FSL: zeros)
command linefslmaths <tfz: a synthetic input> -tfce_delta nan -tfce 2 0.5 6 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-174

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-174.

6/6 · 7/7
d13/tfce-stalldefects of FSLFSL-174: -tfce_minT -1e9 (the thresholds never advance: FSL never ends)
command linefslmaths <tfz: a synthetic input> -tfce_minT -1e9 -tfce 2 0.5 6 out
Bit-identical

every file written byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-174

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-174.

2/2
d13/tfceS-stalldefects of FSLFSL-174: -tfceS with -tfce_minT -1e9 (std::out_of_range)
command linefslmaths <tfz: a synthetic input> -tfce_minT -1e9 -tfceS 2 0.5 6 3 3 3 30 out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds The replica’s default, fix, ends with exit status 1, naming FSL-174.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds The clean room’s default, fix, ends with exit status 1, naming FSL-174.

5/5 · 4/4
d13/tfceS-negative-deltadefects of FSLFSL-174: -tfceS with -tfce_delta -0.5
command linefslmaths <tfz: a synthetic input> -tfce_delta -0.5 -tfceS 2 0.5 6 3 3 3 30 out
Rejected by both

fslmaths exits -8; the same from the replica. The replica’s default, fix, ends with exit status 1, naming FSL-174.

Rejected by both

fslmaths exits -8; the same from the clean room. The clean room’s default, fix, ends with exit status 1, naming FSL-174.

5/5 · 4/4
d13/tfceS-minTdefects of FSLFSL-177: -tfceS with -tfce_minT -2 (more thresholds than its arrays)
command linefslmaths <tfz: a synthetic input> -tfce_minT -2 -tfceS 2 0.5 6 3 3 3 30 out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds The replica’s default, fix, finishes instead (exit status 0), naming FSL-177.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::out_of_range’ what(): Mat::operator(): index out of bounds The clean room’s default, fix, finishes instead (exit status 0), naming FSL-177.

5/5 · 4/4
d13/tfceS-outsidedefects of FSLFSL-178: -tfceS at x = 10 of 10 (FSL reads the next row's first voxel)
command linefslmaths <tfz: a synthetic input> -tfceS 2 0.5 6 10 2 3 30 out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-178

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-178.

6/6 · 7/7
d13/tfceS-fardefects of FSLFSL-178: -tfceS at a voxel far outside (FSL reads outside the image)
command linefslmaths <tfz: a synthetic input> -tfceS 2 0.5 6 4000 2 3 30 out
Bit-identical

every file written byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-178

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-178.

2/2
roc/noise-roc-roc 0.05 with a noise image (the FWE rate)
command linefslmaths <rocim: a synthetic input> -roc 0.05 roc.txt <rocnoise: a synthetic input> <roctruth: a synthetic input> out
Bit-identical

every file written (5/5) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 8/8
roc/truth-roc-roc -0.05: the FP rate from the truth's zero voxels (negative ones left out)
command linefslmaths <rocim: a synthetic input> -roc -0.05 roc.txt <roctruth: a synthetic input> out
Bit-identical

every file written (4/4) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 8/8
roc/never-reached-roc-roc 2 (the FP rate never reaches it)
command linefslmaths <rocim: a synthetic input> -roc 2 roc.txt <rocnoise: a synthetic input> <roctruth: a synthetic input> out
Bit-identical

every file written (5/5) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 8/8
roc/3d-roc-roc of a 3D image
command linefslmaths <rocim3: a synthetic input> -roc -0.1 roc.txt <roctruth: a synthetic input> out
Bit-identical

every file written (4/4) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 8/8
roc/double-roc-roc of a double image (log in double)
command linefslmaths <rocimd: a synthetic input> -roc 0.05 roc.txt <rocnoise: a synthetic input> <roctruth: a synthetic input> out
Bit-identical

every file written (5/5) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 8/8
roc/short-roc-roc under -dt short
command linefslmaths -dt short <rocims: a synthetic input> -roc -0.05 roc.txt <roctruth: a synthetic input> out
Bit-identical

every file written (4/4) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 8/8
roc/zero-roc-roc 0 (0 / 0 printed)
command linefslmaths <rocim: a synthetic input> -roc 0 roc.txt <rocnoise: a synthetic input> <roctruth: a synthetic input> out
Bit-identical

every file written (5/5) byte for byte, with FSL’s exit status and messages

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

6/6 · 8/8
roc/truth-size-roc-roc with a truth image of another size (newimage's exception)
command linefslmaths <rocim: a synthetic input> -roc 0.05 roc.txt <rocnoise: a synthetic input> <roctruthsmall: a synthetic input> out
Rejected by both

fslmaths exits -6; the same from the replica. Image Exception : #3 :: Attempted to multiply images of different sizes terminate called after throwing an instance of ‘std::runtime_error’ what(): Attempted to multiply images of different sizes

Rejected by both

fslmaths exits -6; the same from the clean room. Image Exception : #3 :: Attempted to multiply images of different sizes terminate called after throwing an instance of ‘std::runtime_error’ what(): Attempted to

5/5
d13/roc-nandefects of FSLFSL-153: -roc of an image whose first voxel is NaN
command linefslmaths <rocnan: a synthetic input> -roc 0.05 roc.txt <rocnoise: a synthetic input> <roctruth: a synthetic input> out
Bit-identical

every file written (5/5) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-153 and names it: 300 values change

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, names FSL-153, with the same values.

6/6 · 8/8
d13/roc-noise-largerdefects of FSLFSL-179: -roc with a noise image of 4 volumes for 3
command linefslmaths <rocim: a synthetic input> -roc 0.05 roc.txt <rocnoise4: a synthetic input> <roctruth: a synthetic input> out
Bit-identical

every file written (5/5) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-179 and names it: 4,855 values change

Open difference

Within the record’s thresholds, not bit for bit (standard output (within 1e-05 absolute): 0.01): -roc’s printed area 1e-7 from FSL’s (0.01 of §21’s allowance of 1e-5), in its last printed digit; the curve file is FSL’s; the cause is not established; its default, fix, refuses the call (exit 1), naming FSL-179.

6/6 · 8/8
d13/roc-noise-smallerdefects of FSLFSL-179: -roc with a noise image of 2 volumes for 3 (FSL writes outside it)
command linefslmaths <rocim: a synthetic input> -roc 0.05 roc.txt <rocnoise2: a synthetic input> <roctruth: a synthetic input> out
Bit-identical

every file written byte for byte, with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-179

Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-179.

2/2
d13/roc-no-truedefects of FSLFSL-180: -roc with a truth image without true voxels
command linefslmaths <rocim: a synthetic input> -roc 0.05 roc.txt <rocnoise: a synthetic input> <roctruthneg: a synthetic input> out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The replica’s default, fix, ends with exit status 1, naming FSL-180.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The clean room’s default, fix, ends with exit status 1, naming FSL-180.

5/5
d13/roc-no-noisedefects of FSLFSL-180: -roc -0.05 with a truth image without zero voxels
command linefslmaths <rocim: a synthetic input> -roc -0.05 roc.txt <roctruthones: a synthetic input> out
Rejected by both

fslmaths exits -6; the same from the replica. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The replica’s default, fix, ends with exit status 1, naming FSL-180.

Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The clean room’s default, fix, ends with exit status 1, naming FSL-180.

5/5
difftest/dilall-nan-orderrandom testingrandom testing (2026-11-04): -dilall of a 7 x 13 x 3 float image holding NaN, inf and -inf, written as double: 42 of 273 voxels NaN of the other sign; FSL's build adds each neighbour with the neighbour's value as the destination register, so of a NaN neighbour and the default NaN of an earlier inf + -inf it passes on the neighbour's; the replica added the sum first; fixed (filters::dilall takes the neighbour first)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths <random testing's input case/fsl-mat> -dilall out -odt double
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

–6/6
difftest/dt-input-uint8-dog-edgerandom testingrandom testing (2026-11-03): -dt input on a uint8 image, -mul and -dog_edge: the default's unsigned template (FSL-154) holds no negative difference of Gaussians where FSL's signed char does; FSL-154 is now named for every image the unsigned template reads
command lineFSLOUTPUTTYPE=NIFTI fslmaths -dt input <random testing's input case/fsl-mat> -mul <random testing's input case/fsl-mat> -dog_edge 0.5 0.8 out -odt input
Bit-identical

every file written (3/3) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects FSL-154 and names it: 2 values change

–6/6
difftest/edge-double-squarerandom testingrandom testing (2026-11-03): -edge of a double series: FSL's build squares with x * x (GCC folds pow(x, 2.0)), not glibc's pow, which differed in the last place on 2 of 4608 voxels; fixed in the replica
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths <random testing's input case/fsl-mat> -edge out.nii
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages

–6/6
difftest/fft-twiddles-glibc-sincosrandom testingrandom testing (2026-11-03): -recip, -kernel gauss 2.25 and -fmeanu of a 30 x 1 x 3 float image (an FFT convolution: a 60-point transform along x); 1 of 90 voxels one unit in the last place from FSL's then: two of the 60-point transform's twiddle factors are glibc's sincos values, one bit from the correctly rounded ones (GLIBC-07). Since 2026-11-04 the replica's reproduce takes glibc's sincos (larmorx_core::math::sincos_glibc): FSL's output bit for bit; the default keeps the correctly rounded factors and names nothing (a filter)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths <random testing's input case/fsl-mat> -recip -kernel gauss 2.25 -fmeanu out
Bit-identical

every file written (2/2) byte for byte, with FSL’s exit status and messages; in fix, its default, the replica corrects a register entry and names it: 1 values change

–6/6
difftest/subsamp2-three-widerandom testingrandom testing (2026-11-03): -subsamp2 of a 3 x 1 x 2 image (a 6-voxel axis halved twice): the background value's buffer is too small (FSL-75), fslmaths writes past it and aborts (double free or corruption); the replica names FSL-75 (the value is never used)
command lineFSLOUTPUTTYPE=NIFTI2_GZ fslmaths <random testing's input case/fsl-mat> -subsamp2 out
Rejected by both

fslmaths exits -6; the same from the replica. double free or corruption (out) The replica’s default, fix, finishes instead (exit status 0), naming FSL-75.

–2/2
orig/fft-asymthe clean room's own casesFSL-192: -kernel file (asymmetric, 13 x 11 x 9) -fmeanu of a 16 x 13 x 11 image (FSL's FFT path: a true convolution, the kernel mirrored; the default keeps the direct sum's orientation)
command linefslmaths <ramp16: a synthetic input of the clean room's suite> -kernel file <asym: a synthetic input of the clean room's suite> -fmeanu out
–Expected divergence

FSL’s single-precision FFT convolution against the clean room’s direct sum (where FSL takes its FFT, its products added in double precision and rounded once): the FFT’s rounding is relative to the image’s largest values, not to each voxel’s own last place; against the exact result (float64) FSL’s FFT lies within 6.9e-8 to 1.0e-7 of the largest value on these cases, the clean room within 4.1e-8 to 6.3e-8 (findings 8 and 31; spec question 14). Inside the bound: max |diff| 1e-07 of max |FSL| (bound 1e-05). Outside the threshold: out.nii.gz: the sums’ share of … Findings 31: against the exact result FSL’s FFT lies within 1e-07 of the largest value, the clean room’s direct sum (its products in double precision, rounded once) within 5e-08: the difference is FSL’s single-precision FFT’s.

7/8
orig/direct-asymthe clean room's own casesthe same kernel -fmeanu of a 25 x 25 x 25 image (FSL's direct sum: a correlation, the kernel as stored; FSL-192's reference orientation)
command linefslmaths <ramp25: a synthetic input of the clean room's suite> -kernel file <asym: a synthetic input of the clean room's suite> -fmeanu out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
orig/fft-nan-localthe clean room's own casesFSL-191: -kernel boxv 11 -fmeanu (FSL's FFT path) of an image with one NaN voxel beyond the kernel's reach of most voxels
command linefslmaths <rampnan: a synthetic input of the clean room's suite> -kernel boxv 11 -fmeanu out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-191 and names it: 2,072 values change.

8/8
orig/roc-rampthe clean room's own cases-roc -2 without a noise image: the rate never reaches 2 and the last true-positive rate is below 1 (the image's minimum is a true voxel: the area's last rectangle)
command linefslmaths <rocramp: a synthetic input of the clean room's suite> -roc -2 roc.txt <roctruthx: a synthetic input of the clean room's suite> out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

8/8
orig/roc-ramp-noisethe clean room's own cases-roc 2 with the same image as its noise: the FWE rate never reaches 2, the last true-positive rate below 1
command linefslmaths <rocramp: a synthetic input of the clean room's suite> -roc 2 roc.txt <rocramp: a synthetic input of the clean room's suite> <roctruthx: a synthetic input of the clean room's suite> out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

8/8
orig/roc-below-zerothe clean room's own cases-roc -0.5 with truth voxels of -0.0005 (inside the mask, below 0: false-positive candidates)
command linefslmaths <rocramp: a synthetic input of the clean room's suite> -roc -0.5 roc.txt <roctruthneg5: a synthetic input of the clean room's suite> out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

8/8
orig/roc-truth4dthe clean room's own cases-roc with a 4D truth image (its first volume, with the image library's warning)
command linefslmaths <rocim: a synthetic input> -roc 0.05 roc.txt <rocnoise: a synthetic input> <rocim: a synthetic input> out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

8/8
orig/glibc-double-sinthe clean room's own cases-sin of a 32³ double image in double (-dt double -odt double): glibc's sin where it is not correctly rounded (GLIBC-07)
command linefslmaths -dt double <dbl32: a synthetic input of the clean room's suite> -sin out -odt double
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects GLIBC-07 and names it: 49 values change.

7/7
orig/glibc-double-costhe clean room's own cases-cos of a 32³ double image in double (-dt double -odt double): glibc's cos where it is not correctly rounded (GLIBC-07)
command linefslmaths -dt double <dbl32: a synthetic input of the clean room's suite> -cos out -odt double
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects GLIBC-07 and names it: 40 values change.

7/7
orig/glibc-double-expthe clean room's own cases-exp of a 32³ double image in double (-dt double -odt double): glibc's exp where it is not correctly rounded (GLIBC-07)
command linefslmaths -dt double <dbl32: a synthetic input of the clean room's suite> -exp out -odt double
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects GLIBC-07 and names it: 34 values change.

7/7
orig/glibc-double-logthe clean room's own cases-log of a 32³ double image in double (-dt double -odt double): glibc's log where it is not correctly rounded (GLIBC-07)
command linefslmaths -dt double <dblpos32: a synthetic input of the clean room's suite> -log out -odt double
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects GLIBC-07 and names it: 3 values change.

7/7
orig/glibc-double-powthe clean room's own cases-pow 1.7 of a 32³ double image in double (-dt double -odt double): glibc's pow where it is not correctly rounded (GLIBC-07)
command linefslmaths -dt double <dblpos32: a synthetic input of the clean room's suite> -pow 1.7 out -odt double
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects GLIBC-07 and names it: 29 values change.

7/7
orig/edge-doublethe clean room's own cases-edge of a 32³ double image in double (-dt double -odt double): FSL's squares are exact products (§10.4's pow(x, 2), as FSL is compiled: no glibc pow) and its terms are added from z to x (observed)
command linefslmaths -dt double <dbl32: a synthetic input of the clean room's suite> -edge out -odt double
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
orig/glibc-sthe clean room's own cases-s 3.95 of a 24³ float image: a Gaussian tap whose expf differs between glibc's and the correctly rounded one (GLIBC-07; the default names nothing)
command linefslmaths <flt24: a synthetic input of the clean room's suite> -s 3.95 out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, 1,217 values change, from the filters’ expf, exp, pow, powf and logf (-s and -kernel gauss, -bptf, -tfce, -roc, -dog_edge) computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

7/7
orig/glibc-kernel-gaussthe clean room's own cases-kernel gauss 4.8 -fmean: the same for -kernel gauss (another tap)
command linefslmaths <flt24: a synthetic input of the clean room's suite> -kernel gauss 4.8 -fmean out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, 32 values change, from the filters’ expf, exp, pow, powf and logf (-s and -kernel gauss, -bptf, -tfce, -roc, -dog_edge) computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

7/7
orig/glibc-bptfthe clean room's own cases-bptf 1.73 -1 of a double 4D image (-dt double -odt double): a high-pass weight exp(-1/(2·1.73²)) whose glibc exp differs (GLIBC-07; the default names nothing)
command linefslmaths -dt double <dbl4d: a synthetic input of the clean room's suite> -bptf 1.73 -1 out -odt double
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, 11,785 values change, from the filters’ expf, exp, pow, powf and logf (-s and -kernel gauss, -bptf, -tfce, -roc, -dog_edge) computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

7/7
orig/glibc-tfcethe clean room's own cases-tfce_delta 0.01 -tfce 1.7 0.5 6: about 1,170 thresholds, among them powers (powf(h, 1.7), pow(s, 0.5)) where glibc's differ (GLIBC-07; the default names nothing)
command linefslmaths <tstat: a synthetic input of the clean room's suite> -tfce_delta 0.01 -tfce 1.7 0.5 6 out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, 2 values change, from the filters’ expf, exp, pow, powf and logf (-s and -kernel gauss, -bptf, -tfce, -roc, -dog_edge) computed correctly rounded where reproduce computes them as glibc 2.35 does (GLIBC-07, not named in the iterative programs, as decided on 2026-11-04).

7/7
orig/roc-nodirthe clean room's own casesFSL-193: -roc with a curve file in a folder that does not exist (FSL writes no curve and says nothing; the default stops with an error)
command linefslmaths <rocim: a synthetic input> -roc -0.05 nodir/roc.txt <roctruth: a synthetic input> out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, refuses the call (exit 1), naming FSL-193.

7/7
difftest/edge-char-int-differencesrandom testing-edge under -dt char: the differences as C's int (findings 29)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths -dt char <the case's img> -edge out.nii.gz -odt short
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-151 and names it: 57,937 values change.

7/7
difftest/fsl154-unsigned-wraprandom testing-dt char on a uint8 image of zeros, -sub 1: FSL-154 named (the unsigned type wraps at 0)
command lineFSLOUTPUTTYPE=NIFTI fslmaths -dt char <the case's img> -sub 1 out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-154 and names it: 1,771 values change.

7/7
difftest/fsl157-int8-odt-inputrandom testing-odt input on an int8 image: FSL-157 named for every int8 image (findings 29)
command lineFSLOUTPUTTYPE=NIFTI fslmaths <the case's img> out -odt input
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; in fix, its default, the clean room corrects FSL-157 and names it: 1 values change.

7/7
difftest/fsl165-no-holesrandom testing-fillh26 of a single slice without holes: FSL-165 (findings 27)
command lineFSLOUTPUTTYPE=NIFTI fslmaths -dt input <the case's img> -Tstd -restart <the case's i1> -fillh26 out
–Rejected by both

fslmaths exits -8; the same from the clean room. The clean room’s default, fix, finishes instead (exit status 0), naming FSL-165.

4/4
difftest/fsl75-subsamp2offc-heaprandom testing-subsamp2offc of a 3 x 6 x 1 image: FSL-75's unused background value overruns FSL's buffer (FSL aborts after writing); named in fix and in reproduce (findings 24)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths <the case's img> -subsamp2offc out -odt short
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, names FSL-75, with the same values.

7/7
difftest/geometry-subsamp2offc-mirroredrandom testing-subsamp2offc of an image FSL holds mirrored: the output's matrices through the held frame (findings 24)
command lineFSLOUTPUTTYPE=NIFTI_PAIR_GZ fslmaths <the case's img> -subsamp2offc out.nii.gz
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
difftest/geometry-xmax-mirroredrandom testing-Xmax of an image FSL holds mirrored: the output's offset moved to the held frame's first voxel (findings 24)
command lineFSLOUTPUTTYPE=NIFTI fslmaths <the case's img> -Xmax out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
difftest/kernel-file-below-halfrandom testing-kernel file with values below 0.5: no empty-mask message (findings 25)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths <the case's img> -kernel file <the case's i1> -dilM -ztop out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
difftest/kernel-file-odd-sizesrandom testing-kernel file of odd sizes: no off-centre warning (findings 25)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths -dt double <the case's img> -add <the case's i1> -kernel file <the case's i2> -fmedian out.nii.gz
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
difftest/ptoz-nan-signrandom testing-ptoz of the default NaN: the positive NaN (findings 30)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths <the case's img> -bin -inm 100 -ptoz out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
difftest/roc-candidates-below-boundrandom testing-roc without a noise image: the false-positive candidates are the voxels below the true voxels' bound (findings 29)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths <the case's img> -binv -roc -0.05 roc.txt <the case's i1> -thr 5 out.nii
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

8/8
difftest/roc-truth-zerosrandom testing-roc with a truth of zeros: no candidate, FSL-180 (findings 29)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths <the case's img> -roc -1 roc.txt <the case's i1> out
–Rejected by both

fslmaths exits -6; the same from the clean room. terminate called after throwing an instance of ‘std::logic_error’ what(): sort(): detected NaN The clean room’s default, fix, ends with exit status 1, naming FSL-180.

5/5
difftest/subsamp2-char-sumsrandom testing-subsamp2 under the char type: the sums not wrapped (findings 29)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths -dt input <the case's img> -Zmean -subsamp2 -Ystd out -odt short
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages; its default, fix, names FSL-154, with the same values.

7/7
difftest/subsamp2-qform-quaternionrandom testing-subsamp2 of an image with qfac -1: the qform's quaternion from FSL's product of matrices (findings 24)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths -dt input <the case's img> -subsamp2 out
–Open difference

Within the record’s thresholds, not bit for bit (out.nii.gz: values within the sums’ bound: 7.9e-08): FSL’s subsampling arithmetic under the floating types (spec questions 13 and 38).

8/8
difftest/tfce-double-imagerandom testing-tfce of a float64 image: FSL's single-precision result (findings 26)
command lineFSLOUTPUTTYPE=NIFTI_ZST fslmaths -dt double <the case's img> -tfce 2 0.5 6 out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
difftest/tfce-step-before-sizerandom testing-tfce of a thin image without positive values: the step's message first (findings 26)
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths -dt float <the case's img> -tfce 1 0.5 18 out
–Rejected by both

fslmaths exits 1; the same from the clean room. ERROR: TFCE failed, please check your file for valid sizes and voxel values.Error: tfce requires a positive deltaT input. Exiting The clean room’s default, fix, finishes instead (exit status 0), naming FSL-175.

4/4
difftest/time-step-negative-seriesrandom testingfslmaths on a series of time step -2: FSL writes the time step as its image layer reads it (its magnitude, 1 for a series of 0); the clean rooms wrote the stored value
command lineFSLOUTPUTTYPE=NIFTI_GZ fslmaths <the case's img> -bin out
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
difftest/time-step-zero-seriesrandom testingfslmaths on a series of time step 0: FSL writes the time step as its image layer reads it (its magnitude, 1 for a series of 0); the clean rooms wrote the stored value
command lineFSLOUTPUTTYPE=NIFTI_PAIR fslmaths <the case's img> -thr 50 -sub <the case's i1> -s 2.5 out.nii
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7
difftest/ztop-nan-signrandom testing-ztop of the default NaN: the positive NaN (findings 30)
command lineFSLOUTPUTTYPE=NIFTI2_PAIR_BZ2 fslmaths <the case's img> -binv -inm 1000 -ztop out.nii.gz
–Bit-identical

Every file fslmaths writes with FSL’s values (the record measures every value), with FSL’s exit status and messages.

7/7

Speed

fslmaths runs on one thread; the clean room runs on as many as it is given, with the same result.

  • On fMRIPrep’s goodvoxels chain (its fourteen fslmaths calls and four fslstats calls, in order) the clean room takes 0.38 s against FSL’s 1.70 s, with half FSL’s memory.
  • On the heavy operations of a real BOLD series and T1w it is 1.8 to 28 times as fast as FSL on one thread, and up to 268 times on 14.
  • The largest factors are on FSL’s FFT path, whose padded transforms cost FSL 28.7 s and 228 s (and 4 GiB). The clean room’s direct sum, in double precision there, takes 3.9 s and 8.1 s on one thread.
  • The replica, a translation, is faster than FSL on most calls, and slower on -bptf.

These timings come from the report. They are informal: one run each, on a machine running other work beside them. combine_masks takes milliseconds in every port, below the report’s resolution, so it is left out of the chart; the table holds it.

Speed, both ports against the original: FSL fslmaths 6.0.7.23

Seconds for the whole call (reading the images, computing, writing every output), one run each, on a log scale. The factor is how many times faster the clean room is than FSL, both on one thread.

  • FSL fslmaths 6.0.7.23, 1 thread
  • replica, 1 thread
  • clean room, 1 thread
  • clean room, 14 threads
  1. goodvoxels: its 14 fslmaths and 4 fslstats calls in orderFSL fslmaths 6.0.7.23, 1 thread 1.7 sreplica, 1 thread 0.73 sclean room, 1 thread 0.38 sclean room, 14 threads 0.37 s4.5×
  2. BOLD -TmeanFSL fslmaths 6.0.7.23, 1 thread 0.26 sreplica, 1 thread 0.12 sclean room, 1 thread 0.12 sclean room, 14 threads 0.1 s2.2×
  3. BOLD -Tstd (63 volumes)FSL fslmaths 6.0.7.23, 1 thread 0.3 sreplica, 1 thread 0.16 sclean room, 1 thread 0.14 sclean room, 14 threads 0.17 s2.1×
  4. BOLD -TmedianFSL fslmaths 6.0.7.23, 1 thread 0.49 sreplica, 1 thread 0.29 sclean room, 1 thread 0.27 sclean room, 14 threads 0.11 s1.8×
  5. BOLD -s 5 (goodvoxels' smoothing, on the series)FSL fslmaths 6.0.7.23, 1 thread 2.79 sreplica, 1 thread 1.71 sclean room, 1 thread 0.85 sclean room, 14 threads 0.21 s3.3×
  6. BOLD -fmedianFSL fslmaths 6.0.7.23, 1 thread 5.32 sreplica, 1 thread 4.19 sclean room, 1 thread 2.22 sclean room, 14 threads 0.39 s2.4×
  7. BOLD -bptf 25 2FSL fslmaths 6.0.7.23, 1 thread 1.82 sreplica, 1 thread 5.26 sclean room, 1 thread 0.79 sclean room, 14 threads 0.26 s2.3×
  8. BOLD -kernel boxv 13 -fmeanu (FSL's FFT path)FSL fslmaths 6.0.7.23, 1 thread 28.7 sreplica, 1 thread 16.1 sclean room, 1 thread 3.86 sclean room, 14 threads 0.51 s7.4×
  9. T1w -s 5FSL fslmaths 6.0.7.23, 1 thread 11.2 sreplica, 1 thread 8.62 sclean room, 1 thread 2.28 sclean room, 14 threads 0.35 s4.9×
  10. T1w -dilDFSL fslmaths 6.0.7.23, 1 thread 3.8 sreplica, 1 thread 1.72 sclean room, 1 thread 1.05 sclean room, 14 threads 0.24 s3.6×
  11. T1w -kernel sphere 6 -fmean (FSL's FFT path)FSL fslmaths 6.0.7.23, 1 thread 228 sreplica, 1 thread 196 sclean room, 1 thread 8.14 sclean room, 14 threads 0.85 s28×
0.1 s1 s10 s100 s1000 s

From the benchmark report docs/benchmarks/fsl-maths.md (the main machine: AMD, 16 cores and 32 threads, 124 GiB, Ubuntu 22.04; informal, one run each, while other work ran beside it). FSL and the replica are single-threaded; the clean room’s outputs are the same bits at 1 and 14 threads.

Wall-clock seconds and peak memory, both ports against the original: FSL fslmaths 6.0.7.23 (the benchmark report)

CallFSL 6.0.7.23replica (1 thread)clean room (1 thread)clean room (14 threads)The clean room's checks
combine_masks: -mul 0.01 s (16 MiB)0.00 s (11 MiB)0.00 s (11 MiB)0.01 s (11 MiB)1 = 14 threads; identical to FSL’s
goodvoxels: its 14 fslmaths and 4 fslstats calls in order1.70 s (79 MiB)0.73 s (38 MiB)0.38 s (41 MiB)0.37 s (40 MiB)1 = 14 threads; 34 of 34 images identical to FSL’s
BOLD -Tmean0.26 s (79 MiB)0.12 s (38 MiB)0.12 s (40 MiB)0.10 s (40 MiB)1 = 14 threads; identical
BOLD -Tstd (63 volumes)0.30 s (79 MiB)0.16 s (37 MiB)0.14 s (41 MiB)0.17 s (40 MiB)1 = 14 threads; identical
BOLD -Tmedian0.49 s (79 MiB)0.29 s (38 MiB)0.27 s (41 MiB)0.11 s (40 MiB)1 = 14 threads; identical
BOLD -s 5 (goodvoxels’ smoothing, on the series)2.79 s (80 MiB)1.71 s (70 MiB)0.85 s (42 MiB)0.21 s (88 MiB)1 = 14 threads; identical
BOLD -fmedian5.32 s (80 MiB)4.19 s (70 MiB)2.22 s (42 MiB)0.39 s (81 MiB)1 = 14 threads; identical
BOLD -bptf 25 21.82 s (80 MiB)5.26 s (70 MiB)0.79 s (48 MiB)0.26 s (90 MiB)1 = 14 threads; identical
BOLD -kernel boxv 13 -fmeanu (FSL’s FFT path)28.73 s (113 MiB)16.12 s (95 MiB)3.86 s (42 MiB)0.51 s (80 MiB)1 = 14 threads; max |d| 1.0e-7 of FSL’s largest value since the double sums of 2026-11-04 (before: 2.98 s and 0.43 s, 2.8e-6)
T1w -s 511.15 s (196 MiB)8.62 s (186 MiB)2.28 s (128 MiB)0.35 s (127 MiB)1 = 14 threads; identical since the Gaussian kernel’s correction of findings 23 (before: max |d| 2.8e-7 of FSL’s largest value, 98.6 % within 4 ulp)
T1w -dilD3.80 s (195 MiB)1.72 s (185 MiB)1.05 s (128 MiB)0.24 s (128 MiB)1 = 14 threads; identical
T1w -kernel sphere 6 -fmean (FSL’s FFT path)227.71 s (4063 MiB)195.74 s (3159 MiB)8.14 s (188 MiB)0.85 s (187 MiB)1 = 14 threads; max |d| 1.9e-7 of FSL’s largest value since the double sums of 2026-11-04 (before: 5.41 s and 0.64 s, 8.4e-6)

The benchmark report docs/benchmarks/fsl-maths.md: wall-clock seconds of the whole program and its peak resident memory (MiB), one run each on the main machine with other work beside it (informal). The memory rule (at most 10 % above FSL’s peak plus 8 MiB) is checked at one thread by tests/parity/test_memory_peak.py; on 14 threads a .nii.gz output adds about 3 MiB per worker (the shared parallel gzip writer).

Corrections by default

The replica corrects FSL’s deterministic defects by default, fix, and warns. upstream_defects="reproduce" (on the command line --upstream-defects reproduce from the next release, LARMORX_UPSTREAM_DEFECTS=reproduce in 0.1.0rc1) gives FSL’s behaviour byte for byte, where FSL’s behaviour is defined (register). Where FSL crashes or reads memory it never wrote, the replica refuses in fix and in reproduce, or gives a defined result.

  • Values converted to an integer type outside its range, or NaN, are clamped (NaN to 0) instead of wrapping around (FSL-151).
  • The thresholds, -bin, -binv and -mas bound their range without wrapping or rounding back to the maximum (FSL-152).
  • A NaN first voxel no longer makes the image’s minimum and maximum NaN, and NaN voxels stay out of the robust range and of the reductions’ maxima and minima (FSL-153).
  • -dt char and -dt input on a uint8 image compute in unsigned bytes (FSL-154).
  • -dt input on int8, uint16, uint32, int64, uint64 or float128 images, where FSL writes nothing and exits with status 0, computes in the closest type and writes the result (FSL-155).
  • Integer divisions by zero are refused instead of crashing (FSL-156).
  • -odt input on int8 and uint16 images rounds as the other integer outputs do, and writes int8 as int8 (FSL-157).
  • An operation missing its arguments is refused instead of reading past the command line (FSL-158), and an image of another size than the input for -div, -rem, -max and -min is refused (FSL-159).
  • A Gaussian kernel of zero width leaves the image as it is instead of making it NaN (FSL-160); a kernel without voxels, an unknown kernel type and a negative width are refused (FSL-161).
  • -fmedian, -dilD, -rank and -ranknorm leave NaN values out instead of stopping (FSL-162).
  • -subsamp2 and -subsamp2offc of a 4D image subsample every volume (FSL-163); -dog_edge of a 4D image filters every volume (FSL-164); -fillh of a single slice fills it (FSL-165).
  • -tensor_decomp leaves a voxel whose tensor holds NaN or infinity at 0 (FSL-166).
  • The standard deviation along a dimension takes the clamped variance, 0 for a single value (FSL-167); the percentiles leave NaN out (FSL-168), and so do -pval, -pval0 and -cpval (FSL-169).
  • -ztop of an infinite z gives 0 or 1 (FSL-170); -bptf with a width beyond an int is refused (FSL-171); -ptoz of a p-value below about 1.1 × 10−16 gives its z (FSL-172).
  • -tfce and -tfceS refuse a step that cannot advance (FSL-174), connectivities other than 6, 18 and 26 (FSL-176) and a voxel outside the image (FSL-178); -tfce of an image without positive values gives zeros (FSL-175); -tfceS sizes its arrays for every threshold (FSL-177).
  • -roc refuses a noise image of another size (FSL-179) and a truth image without true or false voxels (FSL-180).
  • A value more than about two billion bin widths above a histogram’s range counts in its last bin (FSL-187: the robust range of -thrp, -thrP, -uthrp, -uthrP, shared with fslstats, BET and FLIRT).
  • -subsamp2 of an image 3 voxels wide computes its unused background value without writing past it (FSL-75, FLIRT’s).
  • glibc’s double exp, log and pow are correctly rounded (GLIBC-07, above).

The clean room makes the same corrections by default, with the same notices. Its reproduce follows FSL where FSL’s behaviour is defined and known from the specification and black-box runs. For FSL-75 no output voxel depends on the corrupted value, so the clean room names it in fix and in reproduce and writes its output. It adds three corrections that its black-box runs found, registered with the same switches:

  • with a kernel large enough for FSL’s FFT, one NaN or infinite voxel makes FSL’s whole volume NaN, where its direct sum for smaller kernels leaves NaN only within the kernel’s reach; the default keeps that local result (FSL-191);
  • on the same path FSL applies the kernel mirrored along every axis, where its direct sum applies it as stored, so the result of a kernel that is not point-symmetric depends on the kernel’s and the image’s sizes; the default keeps the direct sum’s orientation at every size (FSL-192);
  • -roc with a curve file it cannot write (a folder that does not exist) writes no curve and says nothing; the default stops with an error naming the file (FSL-193).

Its double-precision functions are correctly rounded by default (above).

Known limits

Both ports are experimental. Each record’s standard comparison holds fMRIPrep’s calls on three fMRIPrep runs. The replica’s full comparison, the data panel and the long random-testing runs come in the final validation phase.

The clean room is held to thresholds, not to FSL’s bytes:

  • where FSL convolves a large kernel by its FFT, FSL’s single-precision FFT and the clean room’s double sums differ by up to 2 × 10−7 of FSL’s largest value: within §21’s bound, but below its share of voxels within 4 units in the last place on some images (expected divergences, with the clean room as close to the exact result or closer);
  • FSL’s arithmetic of -subsamp2 under the floating types, the last bits of Φ and Φ−1 in double precision and the eigenvectors’ signs are questions for its specification;
  • erf and erfc (in -ztop) come from the libm crate until LARMORX’s own correctly rounded ones exist;
  • random testing left 10 differences open (above).

tan, asin, acos and atan in double precision are correctly rounded in both ports, in fix and in reproduce (as decided on 2026-11-04); since 2026-11-04 sin, cos and the FFT’s twiddle factors (glibc’s sincos) are glibc’s bits in the replica’s reproduce, and sin, cos, exp, log and pow in the clean room’s (above).

The goodvoxels calls use a stand-in for the cortical ribbon, the segmentation’s grey matter. They are to be checked again with a real ribbon once an fMRIPrep run with FreeSurfer’s reconstruction exists.

Images compressed with zstd or bzip2 are decompressed whole before they are read; gzip and uncompressed images are read as a stream.

The cross-platform checks (how they work) hold the other five platforms to LARMORX’s own Linux output. The clean room’s cover the command line in three cases and the typed wrappers in four, each family with a path that makes NaN from numbers, so x86-64’s NaN bits are checked on ARM too. The benchmark is informal: one run each, on a shared machine.

History

  • 2026-11-03: the replica, translated from FSL 6.0.7.23’s source, and its random-testing run, 114,906 fslmaths calls in 24 minutes beside fslstats’. The clean room, written from specs/fslmaths.md, FSL’s usage text and documentation and runs of FSL, and its random-testing run, 150,000 chains in 20 minutes beside fslstats’.
  • 2026-11-04: the clean room complete: every operation and option (the roadmap). The replica’s reproduce computes glibc’s sin, cos and the FFT’s twiddle factors as glibc 2.35 does; two of its mistakes in the sign of a NaN fixed; its record made again, 422 cases. The clean room’s reproduce computes glibc’s sin, cos, exp, log, pow and the filters’ functions as glibc 2.35 does, with ten cases of its own; on FSL’s FFT path it adds its products in double precision.
  • 2026-11-07: the clean room’s record made again, 454 cases, with two permanent cases of random testing: series whose time step is 0 or negative.

Data credits

The data this page shows or names: each dataset with its version, DOI and licence, from LARMORX's test-data catalog. Public-domain dedications (CC0, PDDL) ask nothing; the datasets are cited all the same.