Add arena helper functions and improve block insertion #210

AW-AlanWu · 2025-05-31T08:46:30Z

This PR introduces several helper functions into the arena allocator and refactors the block insertion strategy in arena_alloc to improve performance. As a result, total allocation times have been measurably reduced.

Below is the benchmark script and corresponding results demonstrating the speedup:

Hardware info

CPU: AMD Ryzen 5 7535HS with Radeon Graphics
Cores/Threads: 12 threads
RAM: 30Gi
OS: Ubuntu 24.04.2 LTS
Kernel: 6.11.0-26-generic
Compiler: gcc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0

Benchmark Script on Stage-0

Benchmark Script

#!/bin/bash
export LC_ALL=C

n=15
warmup=5

total_user=0
total_sys=0
total_elapsed=0
total_rss=0

tmp_file="$(mktemp)"

echo "Warming up $warmup times..."
for i in $(seq 1 $warmup); do
    ./out/shecc ./src/main.c >/dev/null 2>&1
done

echo "Running $n benchmarks..."
for i in $(seq 1 $n); do
    /usr/bin/time -f "%U %S %e %M" -o "$tmp_file" ./out/shecc ./src/main.c >/dev/null 2>&1

    read user sys elapsed rss < "$tmp_file"

    echo "Run $i: user=${user}s sys=${sys}s elapsed=${elapsed}s maxrss=${rss}KB"

    total_user=$(awk "BEGIN {printf \"%.6f\", $total_user + $user}")
    total_sys=$(awk "BEGIN {printf \"%.6f\", $total_sys + $sys}")
    total_elapsed=$(awk "BEGIN {printf \"%.6f\", $total_elapsed + $elapsed}")
    total_rss=$(awk "BEGIN {print $total_rss + $rss}")
done

rm -f "$tmp_file"

avg_user=$(awk "BEGIN {printf \"%.6f\", $total_user / $n}")
avg_sys=$(awk "BEGIN {printf \"%.6f\", $total_sys / $n}")
avg_elapsed=$(awk "BEGIN {printf \"%.6f\", $total_elapsed / $n}")
avg_rss=$(awk "BEGIN {printf \"%.2f\", $total_rss / $n}")

echo "----------------------------------"
echo "Average user time:    ${avg_user}s"
echo "Average system time:  ${avg_sys}s"
echo "Average elapsed time: ${avg_elapsed}s"
echo "Average max RSS:      ${avg_rss} KB"

Before (Stage-0)

Result

Warming up 5 times...
Running 15 benchmarks...
Run 1: user=0.16s sys=0.15s elapsed=0.32s maxrss=291848KB
Run 2: user=0.16s sys=0.15s elapsed=0.32s maxrss=291528KB
Run 3: user=0.14s sys=0.16s elapsed=0.30s maxrss=291656KB
Run 4: user=0.14s sys=0.16s elapsed=0.31s maxrss=291912KB
Run 5: user=0.16s sys=0.16s elapsed=0.32s maxrss=291912KB
Run 6: user=0.15s sys=0.17s elapsed=0.33s maxrss=291652KB
Run 7: user=0.15s sys=0.16s elapsed=0.32s maxrss=291848KB
Run 8: user=0.16s sys=0.15s elapsed=0.32s maxrss=291720KB
Run 9: user=0.15s sys=0.15s elapsed=0.31s maxrss=291524KB
Run 10: user=0.15s sys=0.16s elapsed=0.31s maxrss=291848KB
Run 11: user=0.15s sys=0.16s elapsed=0.32s maxrss=291720KB
Run 12: user=0.15s sys=0.15s elapsed=0.31s maxrss=291912KB
Run 13: user=0.16s sys=0.14s elapsed=0.31s maxrss=291720KB
Run 14: user=0.14s sys=0.16s elapsed=0.31s maxrss=291784KB
Run 15: user=0.16s sys=0.14s elapsed=0.31s maxrss=291656KB
----------------------------------
Average user time:    0.152000s
Average system time:  0.154667s
Average elapsed time: 0.314667s
Average max RSS:      291749.33 KB

After (Stage-0)

Result

Warming up 5 times...
Running 15 benchmarks...
Run 1: user=0.08s sys=0.15s elapsed=0.23s maxrss=294344KB
Run 2: user=0.09s sys=0.14s elapsed=0.24s maxrss=294344KB
Run 3: user=0.06s sys=0.16s elapsed=0.22s maxrss=294408KB
Run 4: user=0.06s sys=0.15s elapsed=0.22s maxrss=294536KB
Run 5: user=0.06s sys=0.16s elapsed=0.22s maxrss=294344KB
Run 6: user=0.06s sys=0.15s elapsed=0.22s maxrss=294344KB
Run 7: user=0.05s sys=0.16s elapsed=0.22s maxrss=294472KB
Run 8: user=0.07s sys=0.15s elapsed=0.22s maxrss=294408KB
Run 9: user=0.06s sys=0.16s elapsed=0.23s maxrss=294408KB
Run 10: user=0.06s sys=0.15s elapsed=0.22s maxrss=294600KB
Run 11: user=0.06s sys=0.15s elapsed=0.22s maxrss=294472KB
Run 12: user=0.07s sys=0.17s elapsed=0.25s maxrss=294088KB
Run 13: user=0.07s sys=0.15s elapsed=0.23s maxrss=294024KB
Run 14: user=0.08s sys=0.14s elapsed=0.23s maxrss=294088KB
Run 15: user=0.06s sys=0.15s elapsed=0.21s maxrss=294536KB
----------------------------------
Average user time:    0.066000s
Average system time:  0.152667s
Average elapsed time: 0.225333s
Average max RSS:      294361.07 KB

Benchmark Script on Stage-1

Benchmark Script

#!/bin/bash
export LC_ALL=C

n=15
warmup=5

total_user=0
total_sys=0
total_elapsed=0
total_rss=0

tmp_file="$(mktemp)"

echo "Warming up $warmup times..."
for i in $(seq 1 $warmup); do
    qemu-arm ./out/shecc-stage1.elf ./src/main.c >/dev/null 2>&1
done

echo "Running $n benchmarks..."
for i in $(seq 1 $n); do
    /usr/bin/time -f "%U %S %e %M" -o "$tmp_file" qemu-arm ./out/shecc-stage1.elf ./src/main.c >/dev/null 2>&1

    read user sys elapsed rss < "$tmp_file"

    echo "Run $i: user=${user}s sys=${sys}s elapsed=${elapsed}s maxrss=${rss}KB"

    total_user=$(awk "BEGIN {printf \"%.6f\", $total_user + $user}")
    total_sys=$(awk "BEGIN {printf \"%.6f\", $total_sys + $sys}")
    total_elapsed=$(awk "BEGIN {printf \"%.6f\", $total_elapsed + $elapsed}")
    total_rss=$(awk "BEGIN {print $total_rss + $rss}")
done

rm -f "$tmp_file"

avg_user=$(awk "BEGIN {printf \"%.6f\", $total_user / $n}")
avg_sys=$(awk "BEGIN {printf \"%.6f\", $total_sys / $n}")
avg_elapsed=$(awk "BEGIN {printf \"%.6f\", $total_elapsed / $n}")
avg_rss=$(awk "BEGIN {printf \"%.2f\", $total_rss / $n}")

echo "----------------------------------"
echo "Average user time:    ${avg_user}s"
echo "Average system time:  ${avg_sys}s"
echo "Average elapsed time: ${avg_elapsed}s"
echo "Average max RSS:      ${avg_rss} KB"

Before (Stage-1)

Result

Warming up 5 times...
Running 15 benchmarks...
Run 1: user=1.36s sys=1.12s elapsed=2.49s maxrss=638864KB
Run 2: user=1.35s sys=1.14s elapsed=2.50s maxrss=639104KB
Run 3: user=1.38s sys=1.10s elapsed=2.49s maxrss=638936KB
Run 4: user=1.35s sys=1.14s elapsed=2.49s maxrss=638864KB
Run 5: user=1.36s sys=1.10s elapsed=2.47s maxrss=638664KB
Run 6: user=1.36s sys=1.09s elapsed=2.46s maxrss=638884KB
Run 7: user=1.34s sys=1.10s elapsed=2.46s maxrss=638920KB
Run 8: user=1.36s sys=1.10s elapsed=2.47s maxrss=638796KB
Run 9: user=1.39s sys=1.10s elapsed=2.50s maxrss=638600KB
Run 10: user=1.34s sys=1.10s elapsed=2.45s maxrss=638800KB
Run 11: user=1.40s sys=1.08s elapsed=2.49s maxrss=638684KB
Run 12: user=1.36s sys=1.09s elapsed=2.47s maxrss=639060KB
Run 13: user=1.41s sys=1.07s elapsed=2.49s maxrss=638712KB
Run 14: user=1.36s sys=1.12s elapsed=2.49s maxrss=639016KB
Run 15: user=1.35s sys=1.10s elapsed=2.48s maxrss=638500KB
----------------------------------
Average user time:    1.364667s
Average system time:  1.103333s
Average elapsed time: 2.480000s
Average max RSS:      638826.93 KB

After (Stage-1)

Result

Warming up 5 times...
Running 15 benchmarks...
Run 1: user=0.73s sys=1.03s elapsed=1.77s maxrss=435132KB
Run 2: user=0.72s sys=1.06s elapsed=1.80s maxrss=434956KB
Run 3: user=0.70s sys=1.05s elapsed=1.76s maxrss=435108KB
Run 4: user=0.69s sys=1.05s elapsed=1.76s maxrss=435104KB
Run 5: user=0.67s sys=1.06s elapsed=1.74s maxrss=435448KB
Run 6: user=0.69s sys=1.04s elapsed=1.74s maxrss=435168KB
Run 7: user=0.68s sys=1.05s elapsed=1.75s maxrss=435396KB
Run 8: user=0.68s sys=1.06s elapsed=1.75s maxrss=435556KB
Run 9: user=0.69s sys=1.09s elapsed=1.79s maxrss=434892KB
Run 10: user=0.69s sys=1.05s elapsed=1.74s maxrss=435560KB
Run 11: user=0.69s sys=1.06s elapsed=1.76s maxrss=434988KB
Run 12: user=0.72s sys=1.02s elapsed=1.75s maxrss=434968KB
Run 13: user=0.68s sys=1.05s elapsed=1.75s maxrss=435432KB
Run 14: user=0.69s sys=1.05s elapsed=1.76s maxrss=434868KB
Run 15: user=0.71s sys=1.04s elapsed=1.76s maxrss=435092KB
----------------------------------
Average user time:    0.695333s
Average system time:  1.050667s
Average elapsed time: 1.758667s
Average max RSS:      435177.87 KB

Summary by Bito

This pull request enhances arena memory management by introducing new helper functions and refactoring the block insertion strategy in `arena_alloc`, leading to improved performance and reduced allocation times. Additionally, error handling has been improved with clearer failure messages, enhancing code robustness.

src/globals.c

DrXiao

For the new functions arena_calloc(), arena_realloc(), arena_strdup() and arena_memdup(), they have been added but are not used by any callers, and I am unsure whether it is appropriate to keep these unused functions in the codebase.

src/globals.c

AW-AlanWu · 2025-06-01T14:14:41Z

Based on my current observations, most of the memory allocation in shecc follow a unified allocate-and-release model. This makes almost all scenarios suitable for replacing the current use of malloc, calloc, and free with an arena allocator. I plan to start working on this replacement soon.

In the short term, I'm also working on improving the current strbuf_t implementation by introducing a general-purpose dynamic array in shecc. This new array structure can not only replace the existing strbuf_t, but also serve as a replacement for many of the fixed-size arrays currently used in shecc. In scenarios with frequent push operations, using arena_realloc can lead to performance improvements.

Additionally, I’m exploring optimizations like improving cache locality by replacing the linked list used in the internal structure of the hashmap.

jserv

Make git commit messages always informative based on the motivations and proposed code changes.

bito-code-review · 2025-06-01T19:09:57Z

Bito Review Skipped - No Changes Detected

Bito didn't review this pull request because we did not detect any changes in the pull request to review.

jserv

Avoid using backticks in commit messages. Backticks can be easily confused with single quotes on some terminals, reducing readability. Plain text or single quotes provide sufficient clarity and emphasis.

ChAoSUnItY

Overall changes are fine. Can you also show how this patch would affect stage 1 binary compilation performance?

AW-AlanWu · 2025-06-01T21:19:06Z

Overall changes are fine. Can you also show how this patch would affect stage 1 binary compilation performance?

The Stage-1 benchmark has already been added to the initial conversation

Since I don't have ARM hardware on hand, I can only use QEMU-ARM for simulation to run benchmarks.
Are the results obtained this way acceptable?

jserv · 2025-06-02T00:18:30Z

Since I don't have ARM hardware on hand, I can only use QEMU-ARM for simulation to run benchmarks. Are the results obtained this way acceptable?

No, wait for the empirical confirmation from @ChAoSUnItY .

Refactor 'arena_alloc' to insert newly allocated blocks at the head of the block list, improving allocation efficiency. This reduces pointer chasing during allocation and enables recently freed memory to be reused more quickly. Introduce 'arena_calloc', 'arena_memdup', 'arena_strdup', and 'arena_realloc' to extend the arena allocator's API: - 'arena_calloc': Allocate zero-initialized memory blocks. - 'arena_memdup': Duplicate arbitrary data into arena-managed memory. - 'arena_strdup': Copy null-terminated strings into the arena. - 'arena_realloc': Resize existing arena-managed allocations. These improvements lay the foundation for replacing standard 'malloc', 'calloc', 'realloc', and 'free' with the arena allocator, providing better performance, consistent memory handling, and simpler batch deallocation throughout the project.

ChAoSUnItY

I've confirmed the following benchmark result generated from stage 1 compilation on Raspberry Pi 4B:

Before

Result

Warming up 5 times...
Running 15 benchmarks...
Run 1: user=5.28s sys=1.92s elapsed=7.21s maxrss=628864KB
Run 2: user=5.12s sys=2.02s elapsed=7.15s maxrss=628864KB
Run 3: user=5.16s sys=1.96s elapsed=7.13s maxrss=628864KB
Run 4: user=5.26s sys=1.92s elapsed=7.20s maxrss=628864KB
Run 5: user=5.17s sys=2.00s elapsed=7.18s maxrss=628864KB
Run 6: user=5.19s sys=1.97s elapsed=7.16s maxrss=628864KB
Run 7: user=5.25s sys=1.95s elapsed=7.21s maxrss=628736KB
Run 8: user=5.20s sys=1.96s elapsed=7.18s maxrss=628864KB
Run 9: user=5.13s sys=1.96s elapsed=7.11s maxrss=628864KB
Run 10: user=5.24s sys=1.89s elapsed=7.14s maxrss=628864KB
Run 11: user=5.26s sys=1.93s elapsed=7.21s maxrss=628864KB
Run 12: user=5.33s sys=1.90s elapsed=7.24s maxrss=628864KB
Run 13: user=5.21s sys=1.99s elapsed=7.21s maxrss=628864KB
Run 14: user=5.25s sys=1.96s elapsed=7.22s maxrss=628864KB
Run 15: user=5.06s sys=2.11s elapsed=7.18s maxrss=628864KB
----------------------------------
Average user time:    5.207333s
Average system time:  1.962667s
Average elapsed time: 7.182000s
Average max RSS:      628855.47 KB

After

Result

Warming up 5 times...
Running 15 benchmarks...
Run 1: user=0.74s sys=1.71s elapsed=2.47s maxrss=424832KB
Run 2: user=0.73s sys=1.77s elapsed=2.51s maxrss=424960KB
Run 3: user=0.74s sys=1.71s elapsed=2.47s maxrss=424960KB
Run 4: user=0.80s sys=1.66s elapsed=2.47s maxrss=424832KB
Run 5: user=0.81s sys=1.64s elapsed=2.46s maxrss=424960KB
Run 6: user=0.72s sys=1.73s elapsed=2.47s maxrss=424960KB
Run 7: user=0.70s sys=1.74s elapsed=2.46s maxrss=424960KB
Run 8: user=0.77s sys=1.69s elapsed=2.47s maxrss=424832KB
Run 9: user=0.77s sys=1.68s elapsed=2.47s maxrss=424960KB
Run 10: user=0.75s sys=1.73s elapsed=2.49s maxrss=424960KB
Run 11: user=0.66s sys=1.80s elapsed=2.46s maxrss=424960KB
Run 12: user=0.66s sys=1.79s elapsed=2.46s maxrss=424960KB
Run 13: user=0.71s sys=1.75s elapsed=2.47s maxrss=424832KB
Run 14: user=0.75s sys=1.70s elapsed=2.47s maxrss=424960KB
Run 15: user=0.70s sys=1.77s elapsed=2.48s maxrss=424960KB
----------------------------------
Average user time:    0.734000s
Average system time:  1.724667s
Average elapsed time: 2.472000s
Average max RSS:      424925.87 KB

Benchmark machine

  `.::///+:/-.        --///+//-:``    chaos@raspberrypi 
 `+oooooooooooo:   `+oooooooooooo:    ----------------- 
  /oooo++//ooooo:  ooooo+//+ooooo.    OS: Raspbian GNU/Linux 12 (bookworm) aarch64 
  `+ooooooo:-:oo-  +o+::/ooooooo:     Host: Raspberry Pi 4 Model B Rev 1.5 
   `:oooooooo+``    `.oooooooo+-      Kernel: 6.6.51+rpt-rpi-v8 
     `:++ooo/.        :+ooo+/.`       Uptime: 32 mins 
        ...`  `.----.` ``..           Packages: 1604 (dpkg) 
     .::::-``:::::::::.`-:::-`        Shell: bash 5.2.15 
    -:::-`   .:::::::-`  `-:::-       Resolution: 1920x1080 
   `::.  `.--.`  `` `.---.``.::`      DE: wlroots 
       .::::::::`  -::::::::` `       Theme: PiXflat [GTK3] 
 .::` .:::::::::- `::::::::::``::.    Icons: PiXflat [GTK3] 
-:::` ::::::::::.  ::::::::::.`:::-   Terminal: lxterminal 
::::  -::::::::.   `-::::::::  ::::   Terminal Font: Monospace 10 
-::-   .-:::-.``....``.-::-.   -::-   CPU: (4) @ 1.800GHz 
 .. ``       .::::::::.     `..`..    Memory: 692MiB / 3791MiB 
   -:::-`   -::::::::::`  .:::::`
   :::::::` -::::::::::` :::::::.                             
   .:::::::  -::::::::. ::::::::                              
    `-:::::`   ..--.`   ::::::.
      `...`  `...--..`  `...`
            .::::::::::
             `.-::::-`

jserv · 2025-06-02T07:57:03Z

Thank @AW-AlanWu for contributing!

jserv reviewed Jun 1, 2025

View reviewed changes

src/globals.c Outdated Show resolved Hide resolved

jserv requested review from ChAoSUnItY, DrXiao and fennecJ June 1, 2025 06:10

AW-AlanWu force-pushed the chore/arena branch 2 times, most recently from 9575ef2 to 53bec08 Compare June 1, 2025 07:57

DrXiao reviewed Jun 1, 2025

View reviewed changes

src/globals.c Outdated Show resolved Hide resolved

src/globals.c Outdated Show resolved Hide resolved

AW-AlanWu force-pushed the chore/arena branch from 53bec08 to 8485f58 Compare June 1, 2025 13:51

jserv requested changes Jun 1, 2025

View reviewed changes

AW-AlanWu force-pushed the chore/arena branch from 8485f58 to fe1de6d Compare June 1, 2025 19:06

jserv requested changes Jun 1, 2025

View reviewed changes

AW-AlanWu force-pushed the chore/arena branch from fe1de6d to e9426d7 Compare June 1, 2025 20:20

ChAoSUnItY reviewed Jun 1, 2025

View reviewed changes

AW-AlanWu force-pushed the chore/arena branch from e9426d7 to 4be720b Compare June 2, 2025 07:25

ChAoSUnItY approved these changes Jun 2, 2025

View reviewed changes

jserv merged commit c508a38 into sysprog21:master Jun 2, 2025
6 checks passed

AW-AlanWu deleted the chore/arena branch June 2, 2025 07:59

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add arena helper functions and improve block insertion #210

Add arena helper functions and improve block insertion #210

Uh oh!

AW-AlanWu commented May 31, 2025 •

edited by bito-code-review bot

Loading

Uh oh!

Uh oh!

DrXiao left a comment

Uh oh!

Uh oh!

Uh oh!

AW-AlanWu commented Jun 1, 2025 •

edited

Loading

Uh oh!

jserv left a comment

Uh oh!

bito-code-review bot commented Jun 1, 2025

Uh oh!

jserv left a comment

Uh oh!

ChAoSUnItY left a comment

Uh oh!

AW-AlanWu commented Jun 1, 2025

Uh oh!

jserv commented Jun 2, 2025

Uh oh!

ChAoSUnItY left a comment •

edited

Loading

Uh oh!

Uh oh!

jserv commented Jun 2, 2025

Uh oh!

Uh oh!

Add arena helper functions and improve block insertion #210

Add arena helper functions and improve block insertion #210

Uh oh!

Conversation

AW-AlanWu commented May 31, 2025 • edited by bito-code-review bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Hardware info

Benchmark Script on Stage-0

Before (Stage-0)

After (Stage-0)

Benchmark Script on Stage-1

Before (Stage-1)

After (Stage-1)

Summary by Bito

Uh oh!

Uh oh!

DrXiao left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

AW-AlanWu commented Jun 1, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jserv left a comment

Choose a reason for hiding this comment

Uh oh!

bito-code-review bot commented Jun 1, 2025

Uh oh!

jserv left a comment

Choose a reason for hiding this comment

Uh oh!

ChAoSUnItY left a comment

Choose a reason for hiding this comment

Uh oh!

AW-AlanWu commented Jun 1, 2025

Uh oh!

jserv commented Jun 2, 2025

Uh oh!

ChAoSUnItY left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Before

After

Uh oh!

Uh oh!

jserv commented Jun 2, 2025

Uh oh!

Uh oh!

AW-AlanWu commented May 31, 2025 •

edited by bito-code-review bot

Loading

AW-AlanWu commented Jun 1, 2025 •

edited

Loading

ChAoSUnItY left a comment •

edited

Loading