PROGRESS

mdxfind documentation

Progress Output

While mdxfind runs, the ReportStats() thread writes a one-line status update to stderr every 15 seconds. Each line is a comfort message: it confirms the run is alive and reports where it is. Hash matches are still emitted to stdout as they occur; the progress line never carries match data.

Line Formats

The format selected depends on (a) whether a wordlist total is available — set explicitly with -W <count> or computed in the background by -W auto — and (b) whether at least one GPU job slot is active (gpujob_available()).

1. GPU + wordlist total available

Working on rockyou.txt, w=39, gq=129/0, 4.5Gh/71.7Gh (6.3%), Found=7284, 298.54Mh/s, 401.71Mc/s, ETA 3m45s

2. CPU only + wordlist total available

Working on rockyou.txt, w=39, 4.5Gh/71.7Gh (6.3%), Found=7284, 298.54Mh/s, 401.71Mc/s, ETA 3m45s

3. GPU + no wordlist total

Working on rockyou.txt, w=39, gq=129/0, line 1234567, Found=7284, 298.54Mh/s, 401.71Mc/s

4. CPU only + no wordlist total

Working on rockyou.txt, w=39, line 1234567, Found=7284, 298.54Mh/s, 401.71Mc/s

5. Brute-force mode

Brute ?l?l?l?l?l?l?l?l?l, 3.9T/5.4T (72.7%), Found=224732, 5.37Gh/s, 'sxegdtkfb', ETA 4m36s

Brute-force mode is selected positionally — when exactly one non-option argument is supplied and stat(2) on it fails, mdxfind parses it as a mask and runs against the full keyspace. The progress line carries:

• Brute literal prefix (replaces Working on).
• The mask string itself in the Curfile slot (Curfile = argv[0] in this mode — there is no real input file).
• Keyspace progress dprog/dtotal (pct%), where the denominator is BruteForceTotal = MaskTotal and the numerator is Tothash divided by the active multiplier (algorithm count × salt count).
• A sample candidate at the current progress index — see the field reference below.

The brute-force line has no gq= field and no w= field. Multi-GPU coordination in brute-force mode uses an atomic mask cursor rather than the work-queue model used in wordlist mode, so the queue-depth diagnostics that gate gq= and w= do not apply. The CPU/GPU split that produces the four wordlist variants above does not apply either: there is one brute-force format. See BRUTE_FORCE.md for activation rules and dispatch details.

Field Reference

Field	Meaning
Working on FILE	Current input wordlist (Curfile).
w=N	Depth of the WorkWaiting queue — line-buffers produced by the reader and not yet claimed by a worker thread. High values mean workers are saturated; values near zero mean the reader is the bottleneck. Same definition in all four formats.
gq=PENDING/FREE	Present only when at least one GPU job slot is available. PENDING is gpujob_queue_depth() (jobs queued for GPU dispatch); FREE is gpujob_free_count() (unused jobg buffers). High PENDING + low FREE = GPU saturated; low PENDING + high FREE = CPU producer is starving the GPU.
Xh/Yh (P%)	Progress as Tothash / expected_total, where expected_total = total_lines × salt_mult × rules × iterations. The h suffix is hashes computed; the SI prefix (K/M/G/T) is autoscaled by format_rate().
line N	Lowest line position currently being processed (Lowline + LowSkip), reported when no wordlist total is known.
Found=N	Total successful hash matches across all algorithms in the run (Totfound), cumulative since start.
X.YY[K\|M\|G\|T]h/s	Hash rate. Tothash / wtime — total hashes computed divided by elapsed wallclock seconds since the run started. Long-run average, not an instantaneous reading.
X.YY[K\|M\|G\|T]c/s	Candidate rate over the most recent 15-second window. When rules are loaded this is (Totrules - lastline)/15; otherwise it is (TotLines - lastline)/15. A 15s sliding window, not a long-run average — short-term changes show up here first.
ETA TIME or ETA~TIME	Estimated time remaining, formatted by format_eta(). See ETA prefix below.
Brute MASK	Brute-force-mode prefix. Replaces Working on FILE. The mask string itself is shown in place of Curfile because there is no input file in this mode.
'sample'	Brute-force-mode only. A concrete candidate string drawn from the current keyspace progress index via mask_expand_into(). It is the candidate at the completed-progress position, not the most recently tried candidate — a fast GPU may already be hashing values past the position counted into Tothash. The two are very close in practice.

ETA prefix: space vs tilde

ETA 3m45s     <- space prefix: total line count is final
ETA~3m45s     <- tilde prefix: total line count is still being computed

The tilde appears when -W auto was requested but the background line-count thread has not yet finished. The estimate uses whatever lines have been counted so far and is refined each tick. Once counting completes, the prefix becomes a space and the figure stabilizes.

ETA finishing

... ETA finishing

Emitted when remaining <= 0 (progress fraction has hit 1.0 or saturated) but the work queue still has entries (wq > 0). Indicates the producer has read all input and the workers — or queued GPU jobs — are draining.

ETA scale

format_eta() autoscales:

Range	Format
< 60 s	45s
< 1 h	3m45s
< 1 d	1h22m
< 30 d	4d3h
< 1 y	~2mo15d
≥ 1 y	~1y3mo

The leading ~ on month/year forms is part of the unit display (months and years are approximate calendar units), independent of the line-count tilde described above.

Cadence and Silence

The reporting loop is for (x = 0; x < 15; x++) { sleep(1); ... } — exactly 15 one-second naps per tick. There is no per-progress filter: a Working on line is emitted on every tick during the running state, even if Tothash did not change. The only cases where a tick does not produce a Working on line:

• The run has finished (HashWaiting lock signals exit). ReportStats() returns.
• MDXpause is set. Pause-state messages (pausing..., paused at ..., paused for N minutes) are emitted in place of the progress line. On resume, a resumed (paused d:hh:mm:ss) line is emitted before normal ticks resume.
• Brute-force mode emits a Brute ... line instead. See BRUTE_FORCE.md and the format-5 entry above.

A single hit (match) printed to stdout never suppresses the next stderr tick — they are independent streams.

Interpretation Tips

• h/s falling steadily over a long run with c/s flat or rising — the CPU producer is no longer keeping the workers fed (e.g., a slow rule expansion or a regression in the hot path). Rev 1.339 surfaced this pattern as a regression signal.
• h/s ≈ c/s — no mask expansion in flight; one candidate produces one hash.
• h/s much greater than c/s — masks are multiplying each candidate. With -n '?d?d', expect roughly h/s ≈ 100 × c/s per active algorithm.
• gq=0/N persistently with GPU enabled — the GPU is starved. The producer (CPU rule/mask expansion or wordlist read) is the bottleneck, not the GPU.
• gq=N/0 persistently — the GPU job queue is saturated and the free-buffer pool is empty. Workers will block waiting for a GPU job slot until the GPU drains.
• w=0 for long stretches — the reader has run out of input or is blocked on I/O. With multiple wordlists, this is normal at file boundaries.
• ETA~ not converging on a stable figure — -W auto line counting is still in progress. Wait for the prefix to flip to a space; the estimate stabilizes once the count is final.