Skip to content

Commit 75991cc

Browse files
dinosauredinosaure
authored
Merge pull request #46 from robur-coop/bench
Bench
2 parents 49b067e + edae3f7 commit 75991cc

File tree

9 files changed

+1628
-103
lines changed

9 files changed

+1628
-103
lines changed

‎bench/BENCH.md‎

Lines changed: 165 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,165 @@
1+
## httpcats
2+
3+
wrk.4.2.0-3 (ArchLinux, https://aur.archlinux.org/packages/wrk)
4+
miou https://github.com/robur-coop/miou.git#94915ce5006fc9388a267465bfd2b437a4e15f59
5+
httpcats https://github.com/robur-coop/httpcats.git#7f2385724496d4081912b1e5538299605b557f48
6+
with TCP_NODELAY
7+
8+
taskset -c 16-28 wrk -c <clients> -t <threads> -d120s
9+
10+
DOMAINS=<domains-1> taskset -c 0-7 ./smiou.exe
11+
12+
| domains | attempt | clients | threads | Latency (Stdev) | Req/s (Stdev) |
13+
| 1 | 1 | 12 | 12 | 260.84us (64.53%) | 45795.22 (63.23%) |
14+
| 1 | 2 | 12 | 12 | 260.56us (66.24%) | 45847.17 (66.58%) |
15+
| 1 | 3 | 12 | 12 | 256.28us (66.56%) | 46612.88 (70.53%) |
16+
| 2 | 1 | 12 | 12 | 132.81us (99.35%) | 90907.80 (58.99%) |
17+
| 2 | 2 | 12 | 12 | 131.72us (99.27%) | 91523.50 (59.99%) |
18+
| 2 | 3 | 12 | 12 | 132.38us (99.48%) | 91458.78 (58.92%) |
19+
| 4 | 1 | 12 | 12 | 83.68us (98.32%) | 177059.83 (66.87%) |
20+
| 4 | 2 | 12 | 12 | 89.53us (97.94%) | 170969.87 (91.66%) |
21+
| 4 | 3 | 12 | 12 | 104.59us (98.55%) | 171324.89 (91.14%) |
22+
| 8 | 1 | 12 | 12 | 65.15us (98.20%) | 247210.23 (83.32%) |
23+
| 8 | 2 | 12 | 12 | 65.73us (98.15%) | 246538.37 (51.54%) |
24+
| 8 | 3 | 12 | 12 | 69.74us (97.32%) | 273641.00 (67.48%) |
25+
| 8 | 1 | 32 | 12 | 120.31us (96.43%) | 299448.34 (81.56%) |
26+
| 8 | 2 | 32 | 12 | 113.72us (96.86%) | 292947.73 (83.19%) |
27+
| 8 | 3 | 32 | 12 | 136.14us (95.53%) | 303005.38 (67.37%) |
28+
| 8 | 1 | 64 | 12 | 275.87us (94.62%) | 286181.09 (64.00%) |
29+
| 8 | 2 | 64 | 12 | 276.06us (94.53%) | 290020.32 (59.68%) |
30+
| 8 | 3 | 64 | 12 | 275.89us (94.62%) | 286757.01 (63.00%) |
31+
| 8 | 1 | 128 | 12 | 516.42us (94.06%) | 264090.45 (70.12%) |
32+
| 8 | 2 | 128 | 12 | 526.42us (93.43%) | 266677.63 (64.27%) |
33+
| 8 | 3 | 128 | 12 | 521.38us (93.92%) | 262602.01 (57.61%) |
34+
| 8 | 1 | 256 | 12 | 1.15ms (88.04%) | 231427.96 (65.24%) |
35+
| 8 | 2 | 256 | 12 | 1.15ms (92.42%) | 229988.62 (73.04%) |
36+
| 8 | 3 | 256 | 12 | 1.13ms (91.92%) | 235226.27 (70.42%) |
37+
| 8 | 1 | 32 | 4 | 172.61us (94.89%) | 304639.66 (60.28%) |
38+
| 8 | 2 | 32 | 4 | 152.10us (96.72%) | 272204.15 (56.72%) |
39+
| 8 | 3 | 32 | 4 | 153.39us (96.09%) | 297484.78 (59.45%) |
40+
| 8 | 1 | 64 | 4 | 289.15us (94.51%) | 287882.94 (66.44%) |
41+
| 8 | 2 | 64 | 4 | 299.36us (94.73%) | 287477.42 (56.71%) |
42+
| 8 | 3 | 64 | 4 | 293.06us (94.32%) | 287417.56 (58.00%) |
43+
44+
## vif
45+
46+
wrk.4.2.0-3 (ArchLinux, https://aur.archlinux.org/packages/wrk)
47+
miou https://github.com/robur-coop/miou.git#94915ce5006fc9388a267465bfd2b437a4e15f59
48+
httpcats https://github.com/robur-coop/httpcats.git#7f2385724496d4081912b1e5538299605b557f48
49+
vif https://github.com/robur-coop/vif.git#85d5ad0343db768ce509fc06c91919ee0cb33f1c
50+
with TCP_NODELAY
51+
52+
taskset -c 16-28 wrk -c <clients> -t <threads> -d120s
53+
54+
DOMAINS=<domains-1> taskset -c 0-7 ./svif.exe
55+
56+
| domains | attempt | clients | threads | Latency (Stdev) | Req/s (Stdev) |
57+
| 1 | 1 | 12 | 12 | 292.07us (63.81%) | 40917.43 (63.71%) |
58+
| 1 | 2 | 12 | 12 | 289.84us (64.35%) | 41230.61 (69.31%) |
59+
| 1 | 3 | 12 | 12 | 299.81us (65.10%) | 39857.65 (68.21%) |
60+
| 2 | 1 | 12 | 12 | 146.20us (75.40%) | 81411.42 (67.55%) |
61+
| 2 | 2 | 12 | 12 | 147.84us (73.81%) | 80503.94 (75.70%) |
62+
| 2 | 3 | 12 | 12 | 144.36us (53.25%) | 82455.91 (66.75%) |
63+
| 4 | 1 | 12 | 12 | 78.42us (99.40%) | 160341.42 (91.40%) |
64+
| 4 | 2 | 12 | 12 | 79.79us (99.38%) | 158982.14 (91.67%) |
65+
| 4 | 3 | 12 | 12 | 81.23us (99.25%) | 158549.19 (91.25%) |
66+
| 8 | 1 | 12 | 12 | 85.01us (98.77%) | 156577.27 (91.67%) |
67+
| 8 | 2 | 12 | 12 | 77.52us (98.42%) | 191322.44 (87.11%) |
68+
| 8 | 3 | 12 | 12 | 75.83us (98.47%) | 191159.60 (91.61%) |
69+
| 8 | 1 | 32 | 12 | 137.61us (98.53%) | 187495.81 (74.88%) |
70+
| 8 | 2 | 32 | 12 | 132.52us (98.65%) | 187875.86 (58.58%) |
71+
| 8 | 3 | 32 | 12 | 135.82us (98.90%) | 186670.32 (83.33%) |
72+
| 8 | 1 | 64 | 12 | 335.85us (97.57%) | 175018.96 (70.84%) |
73+
| 8 | 2 | 64 | 12 | 363.29us (97.65%) | 173275.45 (63.74%) |
74+
| 8 | 3 | 64 | 12 | 356.96us (98.02%) | 173977.34 (62.85%) |
75+
| 8 | 1 | 128 | 12 | 761.92us (85.79%) | 159131.89 (68.64%) |
76+
| 8 | 2 | 128 | 12 | 770.80us (92.74%) | 157596.40 (69.30%) |
77+
| 8 | 3 | 128 | 12 | 770.81us (88.89%) | 157751.74 (63.53%) |
78+
| 8 | 1 | 256 | 12 | 1.88ms (78.61%) | 134303.06 (63.92%) |
79+
| 8 | 2 | 256 | 12 | 1.88ms (75.20%) | 133830.88 (76.46%) |
80+
| 8 | 3 | 256 | 12 | 1.90ms (68.15%) | 133068.67 (75.15%) |
81+
| 8 | 1 | 32 | 4 | 178.77us (97.92%) | 191074.67 (65.33%) |
82+
| 8 | 2 | 32 | 4 | 177.04us (98.61%) | 191065.05 (56.00%) |
83+
| 8 | 3 | 32 | 4 | 184.92us (98.17%) | 189689.10 (70.78%) |
84+
| 8 | 1 | 64 | 4 | 374.37us (97.95%) | 177832.95 (62.95%) |
85+
| 8 | 2 | 64 | 4 | 371.00us (93.07%) | 178383.60 (66.37%) |
86+
| 8 | 3 | 64 | 4 | 373.27us (97.32%) | 177777.87 (61.33%) |
87+
88+
## httpun+eio
89+
90+
wrk.4.2.0-3 (ArchLinux, https://aur.archlinux.org/packages/wrk)
91+
eio.1.3
92+
httpun.0.2.0
93+
with TCP_NODELAY
94+
95+
taskset -c 16-28 wrk -c <clients> -t <threads> -d120s
96+
97+
DOMAINS=<domains-1> taskset -c 0-7 ./seio.exe
98+
Detected <domains-1> cores
99+
Detected 1024 max open files
100+
Detected 4096 somaxconn
101+
102+
| domains | attempt | clients | threads | Latency (Stdev) | Req/s (Stdev) |
103+
| 1 | 1 | 12 | 12 | 2.15ms (83.14%) | 23566.44 (89.99%) |
104+
| 1 | 2 | 12 | 12 | 2.21ms (83.28%) | 23240.62 (90.08%) |
105+
| 1 | 3 | 12 | 12 | 2.25ms (83.27%) | 22992.66 (90.17%) |
106+
| 2 | 1 | 12 | 12 | 2.27ms (86.13%) | 44246.58 (89.76%) |
107+
| 2 | 2 | 12 | 12 | 2.24ms (85.89%) | 43691.81 (89.96%) |
108+
| 2 | 3 | 12 | 12 | 2.26ms (86.24%) | 43713.11 (89.33%) |
109+
| 4 | 1 | 12 | 12 | 2.68ms (87.31%) | 74439.51 (89.31%) |
110+
| 4 | 2 | 12 | 12 | 2.59ms (87.51%) | 71189.82 (88.83%) |
111+
| 4 | 3 | 12 | 12 | 2.49ms (85.79%) | 53957.00 (89.82%) |
112+
| 8 | 1 | 12 | 12 | 2.70ms (85.25%) | 97812.17 (88.40%) |
113+
| 8 | 2 | 12 | 12 | 2.58ms (85.60%) | 90427.52 (88.92%) |
114+
| 8 | 3 | 12 | 12 | 2.59ms (85.99%) | 90127.41 (88.31%) |
115+
| 8 | 1 | 32 | 12 | 2.68ms (86.05%) | 98589.41 (89.57%) |
116+
| 8 | 2 | 32 | 12 | 2.71ms (86.69%) | 97215.50 (89.18%) |
117+
| 8 | 3 | 32 | 12 | 2.66ms (85.97%) | 95854.64 (90.03%) |
118+
| 8 | 1 | 64 | 12 | 3.47ms (86.04%) | 105166.43 (90.12%) |
119+
| 8 | 2 | 64 | 12 | 3.42ms (86.15%) | 96920.86 (90.56%) |
120+
| 8 | 3 | 64 | 12 | 3.30ms (86.15%) | 90518.37 (90.33%) |
121+
| 8 | 1 | 128 | 12 | 3.87ms (86.02%) | 108686.93 (90.24%) |
122+
| 8 | 2 | 128 | 12 | 3.94ms (85.81%) | 103329.04 (89.89%) |
123+
| 8 | 3 | 128 | 12 | 3.93ms (86.32%) | 96327.52 (90.02%) |
124+
| 8 | 1 | 256 | 12 | 4.65ms (85.72%) | 105082.82 (89.57%) |
125+
| 8 | 2 | 256 | 12 | 4.75ms (85.82%) | 103277.00 (89.83%) |
126+
| 8 | 3 | 256 | 12 | 4.90ms (85.76%) | 99308.92 (89.94%) |
127+
| 8 | 1 | 32 | 4 | 2.93ms (85.50%) | 97593.32 (89.71%) |
128+
| 8 | 2 | 32 | 4 | 3.03ms (85.05%) | 96591.98 (89.90%) |
129+
| 8 | 3 | 32 | 4 | 3.09ms (85.66%) | 94456.97 (89.54%) |
130+
| 8 | 1 | 64 | 4 | 3.25ms (86.35%) | 96787.99 (90.10%) |
131+
| 8 | 2 | 64 | 4 | 3.29ms (85.82%) | 90975.61 (90.16%) |
132+
| 8 | 3 | 64 | 4 | 3.26ms (85.48%) | 94306.00 (90.08%) |
133+
134+
## nginx
135+
136+
nginx/1.28.2
137+
taskset -c 16-28 wrk -c <clients> -t <threads> -d120s http://localhost:8080/plaintext
138+
sudo taskset -c 0-7 nginx -p . -c nginx.conf
139+
140+
| domains | attempt | clients | threads | Latency (Stdev) | Req/s (Stdev) |
141+
| 1 | 1 | 12 | 12 | 61.90us (96.34%) | 190329.81 (72.99%) |
142+
| 1 | 2 | 12 | 12 | 61.77us (88.40%) | 190647.78 (80.34%) |
143+
| 1 | 3 | 12 | 12 | 60.99us (94.98%) | 192098.10 (65.36%) |
144+
| 2 | 1 | 12 | 12 | 34.16us (65.39%) | 339631.42 (84.21%) |
145+
| 2 | 2 | 12 | 12 | 34.97us (72.29%) | 332398.77 (72.80%) |
146+
| 2 | 3 | 12 | 12 | 34.01us (70.37%) | 341239.19 (81.95%) |
147+
| 4 | 1 | 12 | 12 | 24.92us (87.03%) | 459989.61 (60.64%) |
148+
| 4 | 1 | 12 | 12 | 24.51us (89.10%) | 471448.33 (57.33%) |
149+
| 4 | 1 | 12 | 12 | 24.44us (92.83%) | 468375.92 (66.81%) |
150+
| 8 | 1 | 12 | 12 | 19.79us (98.03%) | 574249.64 (60.07%) |
151+
| 8 | 2 | 12 | 12 | 19.81us (96.86%) | 572347.61 (62.68%) |
152+
| 8 | 3 | 12 | 12 | 19.50us (99.00%) | 583214.18 (62.95%) |
153+
| 8 | 1 | 32 | 12 | 29.57us (90.13%) | 759297.39 (69.18%) |
154+
| 8 | 2 | 32 | 12 | 27.12us (92.37%) | 816192.41 (67.34%) |
155+
| 8 | 3 | 32 | 12 | 28.53us (90.08%) | 783033.37 (69.58%) |
156+
| 8 | 1 | 64 | 12 | 60.96us (90.05%) | 987026.17 (64.24%) |
157+
| 8 | 2 | 64 | 12 | 59.09us (75.04%) | 989338.02 (66.37%) |
158+
| 8 | 3 | 64 | 12 | 64.13us (86.68%) | 946537.36 (68.31%) |
159+
| 8 | 1 | 128 | 12 | 111.31us (96.34%) | 1069342.17 (69.67%) |
160+
| 8 | 2 | 128 | 12 | 116.47us (89.25%) | 1038051.76 (63.20%) |
161+
| 8 | 3 | 128 | 12 | 118.27us (90.50%) | 1061696.25 (64.73%) |
162+
| 8 | 1 | 256 | 12 | 226.64us (75.06%) | 1094142.47 (71.55%) |
163+
| 8 | 2 | 256 | 12 | 233.47us (86.52%) | 1070324.12 (78.43%) |
164+
| 8 | 3 | 256 | 12 | 229.38us (84.58%) | 1087358.25 (78.39%) |
165+

‎bench/PROTOCOL.md‎

Lines changed: 188 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,188 @@
1+
# Benchmark Protocol
2+
3+
This document describes the methodology used to benchmark HTTP server
4+
implementations. The goal is to compare throughput (requests/second) and
5+
latency across several OCaml HTTP server stacks and nginx as a baseline.
6+
7+
## Implementations under test
8+
9+
| Name | Stack | Runtime | Binary |
10+
|------|-------|---------|--------|
11+
| **httpcats** | [httpcats](https://github.com/robur-coop/httpcats) (h1/h2) | [miou](https://github.com/robur-coop/miou) | `smiou.exe` |
12+
| **vif** | [vif](https://github.com/robur-coop/vif) (built on httpcats) | miou | `svif.exe` |
13+
| **httpun+eio** | [httpun](https://github.com/anmonteiro/httpun) 0.2.0 | [eio](https://github.com/ocaml-multicore/eio) 1.3 | `seio.exe` |
14+
| **nginx** | nginx/1.28.2 | native | `nginx` |
15+
16+
All OCaml servers respond to `GET /plaintext` with `Hello, World!`
17+
(`text/plain`) and `GET /json` with `{"message":"Hello, World!"}`
18+
(`application/json`). nginx only serves `/plaintext`.
19+
20+
All servers set `TCP_NODELAY` on accepted connections.
21+
22+
## Hardware
23+
24+
- **CPU**: AMD Ryzen 9 7950X — 16 cores / 32 threads (SMT enabled)
25+
- Base clock: 4.5 GHz, boost up to 5.88 GHz
26+
- L1d/L1i: 512 KiB each (16 instances), L2: 16 MiB (16 instances), L3: 64 MiB
27+
(2 × 32 MiB CCDs)
28+
- Single NUMA node (all 32 logical CPUs on node 0)
29+
- **RAM**: 64 GB DDR5
30+
- **OS**: Arch Linux, kernel `6.18.13-zen1-1-zen` (ZEN preemptive, SMP)
31+
- **Compiler**: GCC 15.2.1
32+
33+
### CPU topology note
34+
35+
The Ryzen 9 7950X has 2 CCDs (Core Complex Dies), each with 8 cores. Cores 0-7
36+
correspond to physical cores on CCD0 (logical CPUs 0-15 with SMT), and cores
37+
8-15 to CCD1 (logical CPUs 16-31). The benchmark pins the server to cores 0-7
38+
and wrk to cores 16-27, which places them on **separate CCDs**. This is
39+
intentional: it avoids L3 cache contention between the server and the load
40+
generator, at the cost of cross-CCD (Infinity Fabric) latency for any
41+
inter-process communication — which is acceptable since they only communicate
42+
over TCP loopback.
43+
44+
## Network configuration
45+
46+
All benchmarks run over **localhost** (`127.0.0.1:8080`). There is no physical
47+
NIC involved; traffic goes through the kernel's loopback interface.
48+
49+
### Loopback interface
50+
51+
```
52+
lo: mtu 65536, qdisc noqueue, state UNKNOWN
53+
```
54+
55+
The default loopback MTU of 65536 bytes allows large TCP segments without
56+
fragmentation.
57+
58+
### Relevant sysctl settings
59+
60+
| Parameter | Value | Note |
61+
|-----------|-------|------|
62+
| `net.core.somaxconn` | 4096 | Listen backlog upper bound |
63+
| `net.ipv4.tcp_max_syn_backlog` | 4096 | SYN queue size |
64+
| `net.ipv4.tcp_fin_timeout` | 60 | Default FIN-WAIT-2 timeout |
65+
| `net.ipv4.tcp_tw_reuse` | 2 | Kernel default (reuse for loopback) |
66+
| `net.ipv4.ip_local_port_range` | 32768–60999 | ~28k ephemeral ports |
67+
| `net.core.rmem_max` | 4194304 | Max receive buffer (4 MiB) |
68+
| `net.core.wmem_max` | 4194304 | Max send buffer (4 MiB) |
69+
| `net.core.netdev_max_backlog` | 1000 | Default NIC backlog (not relevant for loopback) |
70+
71+
### File descriptor limit
72+
73+
```
74+
ulimit -n = 524288
75+
```
76+
77+
This is high enough to support all tested concurrency levels (up to 256
78+
clients) without hitting file descriptor exhaustion.
79+
80+
### `TCP_NODELAY`
81+
82+
All servers explicitly set `TCP_NODELAY` on accepted connections to disable
83+
Nagle's algorithm, ensuring that small HTTP responses are sent immediately
84+
without waiting for additional data to coalesce.
85+
86+
## Load generator
87+
88+
- **wrk** version 4.2.0-3 (ArchLinux AUR package)
89+
- Each test runs for **120 seconds** (`-d120s`)
90+
- Each configuration is repeated **3 times** (3 attempts) to assess
91+
reproducibility
92+
93+
```
94+
taskset -c 16-27 wrk -c <clients> -t <threads> -d120s http://localhost:8080/plaintext
95+
```
96+
97+
## CPU pinning
98+
99+
To eliminate scheduling noise, the server and the load generator are pinned to
100+
**disjoint CPU sets** using `taskset`:
101+
102+
| Component | CPU cores |
103+
|-----------|-----------|
104+
| Server | `0-7` (up to 8 cores) |
105+
| wrk | `16-27` (12 cores) |
106+
107+
The cores in between (8-15) are left unused to avoid cache/interconnect
108+
contention between the server and the load generator.
109+
110+
### Server launch commands
111+
112+
```sh
113+
# httpcats (miou)
114+
DOMAINS=<N-1> taskset -c 0-7 ./smiou.exe
115+
116+
# vif (miou)
117+
DOMAINS=<N-1> taskset -c 0-7 ./svif.exe
118+
119+
# httpun+eio
120+
DOMAINS=<N-1> taskset -c 0-7 ./seio.exe
121+
122+
# nginx
123+
sudo taskset -c 0-7 nginx -p . -c nginx.conf
124+
```
125+
126+
The `DOMAINS` environment variable controls the number of OCaml domains
127+
(OS-level threads) used by the runtime. For miou-based servers, `DOMAINS=N-1`
128+
means the server runs on N domains total (1 main domain + N-1 additional
129+
domains). For eio, the value is passed to `Eio.Net.run_server
130+
~additional_domains`.
131+
132+
For nginx, the number of worker processes is set in `nginx.conf`
133+
(`worker_processes 8`) with `worker_cpu_affinity auto`.
134+
135+
## Test matrix
136+
137+
### Variable: number of domains (scalability)
138+
139+
With a fixed load of **12 clients / 12 threads**, the number of domains is
140+
varied across `{1, 2, 4, 8}` to measure how each implementation scales with
141+
parallelism.
142+
143+
### Variable: number of clients (concurrency)
144+
145+
With a fixed number of **8 domains**, the number of concurrent clients is
146+
varied across `{12, 32, 64, 128, 256}` (with 12 wrk threads) and `{32, 64}`
147+
(with 4 wrk threads) to measure behavior under increasing connection pressure.
148+
149+
## Metrics collected
150+
151+
From each wrk run, two metrics are recorded:
152+
- **Average latency** and its **±σ percentage** (percentage of requests within
153+
±1 standard deviation of the mean)
154+
- **Average requests/second** and its **±σ percentage**
155+
156+
The ±σ percentage is reported directly by wrk as the `+/- Stdev` column. A
157+
higher percentage indicates a more stable/predictable distribution.
158+
159+
## nginx configuration
160+
161+
nginx is configured for maximum raw throughput with the following key settings
162+
(see `nginx.conf`):
163+
- `worker_processes 8` with `worker_cpu_affinity auto`
164+
- `worker_connections 32768`
165+
- `access_log off` and `server_tokens off`
166+
- `keepalive_requests 300000` (default is 100)
167+
- `listen 8080 reuseport deferred fastopen=4096`
168+
- Response: `return 200 "Hello, World!"`
169+
170+
This is intentionally an aggressive configuration to establish an upper bound
171+
for throughput on this hardware.
172+
173+
## Reproducibility
174+
175+
To reproduce these benchmarks:
176+
1. Build the OCaml servers:
177+
```sh
178+
dune build smiou.exe seio.exe svif.exe
179+
```
180+
2. Start one server at a time, pinned to cores 0-7
181+
3. Run wrk from a separate terminal, pinned to cores 16-28:
182+
```sh
183+
taskset -c 16-27 wrk -c 12 -t 12 -d120s http://localhost:8080/plaintext
184+
```
185+
4. Repeat each configuration 3 times, recording the wrk summary output
186+
5. Stop the server between configuration changes
187+
188+
Results are recorded in `BENCH.md` and visualized in `index.html`.

‎bench/dune‎

Lines changed: 14 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,14 @@
1+
(executable
2+
(name smiou)
3+
(enabled_if (= %{profile} "bench"))
4+
(libraries yojson httpcats))
5+
6+
(executable
7+
(name seio)
8+
(enabled_if (= %{profile} "bench"))
9+
(libraries httpun httpun-eio eio eio_main eio_linux yojson unix))
10+
11+
(executable
12+
(name svif)
13+
(enabled_if (= %{profile} "bench"))
14+
(libraries vif))

0 commit comments

Comments
 (0)