Homebrew NSP to NSZ - C++

snuffbot1 · Wednesday at 9:45 PM

Here you go, a very small c++ version for converting nsp files to nsz and back again. Source code and exe are in the zip.

Example useage:
Flag Meaning
-C / -D Compress NSP→NSZ or decompress NSZ→NSP
-F Re-pad the PFS0 header to 0x20 alignment
-L zstd long-distance matching
-S / -B Solid stream (default) or block stream
-l N zstd level (default 18)
-s N Block size exponent 14..=32 (default 20)
-o DIR Output directory
-k PATH prod.keys (default ~/.switch/prod.keys)
-x Skip CRC32 check of known keys
-K, -P Accepted for nsz compatibility, no-op

Example compress/decompress nsp

nszrs -C -B -k prod.keys -s 20 -o out/ game.nsp
nszrs -D -F -k prod.keys -o out/ /out/game.nsz
nszrs -C -L -o out/ game.nsp

This is a command line tool, at some point I might make a gui, there's no xci conversion yet and I doubt I'll add it. I just made it for my own personal use but feel free to mod/share or whatever you want. It's called nszrs as I found some rust code version on the net and then converted it to c++ for my own use.

mathew77 · 2026-09-24T09:27:57+0100

@snuffbot1
Thank you for your lightweight C++ realization, I like the idea but unfortunately for right now it's very slow.

I tried to test it versus the nszgui by nicoboss, the results with 200 MB NSP file (level 22 blocks and long-distance) are:

His version took 9 seconds to compress with maximum CPU utilization across all cores:
Your version took 31 seconds, three and a half times long and the CPU load average was pretty low:
Code:
```
nszrs.exe -C -L -B -l 22 -k prod.keys test.nsp
```

Please optimize it if you could.

snuffbot1 · 2026-09-24T09:48:20+0100

mathew77 said:
@snuffbot1
Thank you for your lightweight C++ realization, I like the idea but unfortunately for right now it's very slow.
I tried to test it versus the nszgui by nicoboss, the results with 200 MB NSP file (level 22 blocks and long-distance) are:

1) His version took 9 seconds to compress with maximum CPU utilization across all cores:

View attachment 592261

2) Your version took 31 seconds, three and a half times long and the CPU load average was pretty low:

View attachment 592262

Code:

nszrs.exe -C -L -B -l 22 -k prod.keys test.nsp

Please optimize it if you could.

I'm adding xci support just now, I didn't look at nicboss version, but I assume he is starting a few threads so it's using more cpu usage rather that taking longer. I could add that I guess. I've got xcz decompress done and am about 90% though adding xci to xcz, so I'll have a look at multithreading/cpu core use once I'm done with that.

mathew77 · 2026-09-24T09:49:39+0100

@snuffbot1
..and add the ability to verify for hash mismatches (NCA & PFS0, SHA256) right after the compression, too, please!..

snuffbot1 · 2026-09-24T12:51:03+0100

mathew77 said:
@snuffbot1
..and add the ability to verify for hash mismatches (NCA & PFS0, SHA256) right after the compression, too, please!..

I'll think about it, I've just finished xci stuff just now so block and solid compress works. I'll add flags so you can remove update/logo partitions as it's pointless packing an xci with firmware update files included (defeats the point really).

Note to self:
Build windows fancy shmancy GUI with drag and drop support... dropdowns/switches and shit!

Post automatically merged: Yesterday at 3:05 PM

@mathew77

Try this command:
nszrs -C -S -l 20 -k prod.keys -o ./ game.nsp # solid

That maxed out my cpu and should be faster, I've also added xci > xcz < xci stuff and fixed a few bugs. Report issues and I'll fix them before making a gui. For verification you can use NXfile viewer for now, when I make a gui I'll probably add that as a separate feature from the exe I posted.

mathew77 · 2026-09-24T15:25:33+0100

snuffbot1 said:
solid

It's the deal breaker, I need the block compression, only.

nicoboss maxed out the CPU even with block compression method... It's also has an option to auto-verify the results afterwards.

The problem is, it's python-based, so heavy and cumbersome, I'd like C++ solution more.

snuffbot1 · 2026-09-24T15:34:34+0100

mathew77 said:
It's the deal breaker, I need the block compression, only. nicoboss maxed out the CPU even with block compression method... It's also has an option to auto-verify the results afterwards.

The problem is, it's python-based, so heavy and cumbersome, I'd like C++ solution more.

Solid produces much smaller files, why do you need block?

EDIT: Verification of nca's flag added. Will post when I finish checking stuff, most stuff is completed now, just some tweaks/bugs to fix...

Post automatically merged: Yesterday at 9:52 PM

@mathew77

Have a look here at ZSTD_c_nbWorkers section:
https://facebook.github.io/zstd/zstd_manual.html

To speed things up, how many workers are set in the python code you are currently using?

mathew77 · 2026-09-25T07:58:28+0100

snuffbot1 said:
verification of nca's flag added.

Thank you!..

snuffbot1 said:
why do you need block?

The main reason is that when installing large and high-compressed .NSZ files, like 15 GB with Level 22 in long-distance mode, they do not need the Application mode on a Switch, standard Applet mode works well. It's crucial for inexperience users = so my needs.

My ZSTD_c_nbWorkers is 8 though.

snuffbot1 · 2026-09-25T08:28:53+0100

mathew77 said:
Thank you!..

The main reason is that when installing large and high-compressed .NSZ files, like 15 GB with Level 22 in long-distance mode, they do not need the Application mode on a Switch, standard Applet mode works well. It's crucial for inexperience users = so my needs.

My ZSTD_c_nbWorkers is 8 though.

OK no problem, what I'll do is look at ZSTD_c_nbWorkers code in nsp/xci packing but I think it already automatically sets workers by itself, maybe I can add an override flag. I've been fixing some minor bugs and stuff but I am nearly completed now with the small exe (only started making this from scratch yesterday), I've still got a gui to make but I'll release the update command line exe later today. At the minute I am working something else. Verfication now works though I didn't add it to check automatically after compression/decompression. I just put it with it's own flag, like this : nszrs -V -k prod.keys game.xci
I suppose you could make a little batch script to run that automatically if you wanted. In the gui when I do that I want it as a seperate thing so that's why I've done it like that.

Post automatically merged: Today at 9:40 AM

@mathew77

I've added a flag for nsp solid compression so you can set your threads - please can you do some tests and report back.

nszrs -C -S -k prod.keys -o out/ game.nsp # auto threads
nszrs -C -S -t 1 -k prod.keys -o out/ game.nsp # single-threaded
nszrs -C -S -t 4 -k prod.keys -o out/ game.nsp # 4 workers

What to do is set the -t value to 8 or 20 or whatever and let me know what kind of speed you get without your CPU going mental or running out of memory, once Done I'll add code for block mode as currently this is single thread for now which is why it's slower than the python version you are using.

Other example commands:
Example commands
nszrs -C -B -k prod.keys -s 20 -o ./ game.xci # block
nszrs -C -S -k prod.keys -o ./ game.xci # solid
nszrs -C -k prod.keys -o ./ game.xci # default block
nszrs -C -B -k prod.keys -s 20 -o ./ game.nsp # block
nszrs -C -S -l 20 -K -P -k prod.keys -o ./ game.nsp # solid (better than block mode for compression)
nszrs -C -k prod.keys -o ./ game.nsp # default solid
nszrs -C -L -o ./ game.nsp

nszrs -C -k prod.keys -S -l 22 -x --keep -o ./solid game.xci
nszrs -C -B -k prod.keys -s 20 -o ./ game.xci # default (drops logo, empties update/normal)
nszrs -C -B -k prod.keys -s 18 --keep -o ./ game.xci # keeps everything
nszrs -C -S -l 20 -k prod.keys -o ./ game.xci

nszrs -D -F -k prod.keys -o ./ ./game.nsz
nszrs -D -k prod.keys -o ./ ./game.xcz
nszrs -D -k prod.keys -o ./ ./game.nsz # unpacks to identicle as source nsp

nszrs -C -S -k prod.keys -o out/ game.nsp # auto threads
nszrs -C -S -t 1 -k prod.keys -o out/ game.nsp # single-threaded
nszrs -C -S -t 4 -k prod.keys -o out/ game.nsp # 4 workers

nszrs -V -k prod.keys game.xci # verify file

mathew77 · 2026-09-25T10:02:17+0100

It's 1 min 3 sec for a 200 MB file, for now:

Code:

nszrs.exe -C -L -S -l 22 -k prod.keys test.nsp

No crashes at all, will wait for my beloved block compression mode!.. :grog:

snuffbot1 · 2026-09-25T10:18:24+0100

mathew77 said:
It's 1 min 3 sec for a 200 MB file, for now:

Code:

nszrs.exe -C -L -S -l 22 -k prod.keys test.nsp

View attachment 592418
No crashes at all, will wait for my beloved block compression mode!..

I've done it for block now, it works good. I just need to fix the percentage output, what kind of time were you getting before?

mathew77 · 2026-09-25T10:32:38+0100

@snuffbot1
Apprx. ~9 seconds with block compressing mode 22 level with long-distant mode per 200 MB, I guess.

snuffbot1 · 2026-09-25T10:40:59+0100

Here you go, this works with block compression now:

nszrs -C -B -t 20 -k prod.keys -o ./ game.xci

Test this and let me know how you get on, what speeds you expected and what you get.

mathew77 · 2026-09-25T10:46:42+0100

snuffbot1 said:
Test this and let me know how you get on, what speeds you expected and what you get.

Perfect!.. It's 9 seconds now!.. :wub:

There's still a little headroom to improvements, CPU load average is ~85-90% but for now it's no less then python version, and I love it, my gratitudes.

snuffbot1 · 2026-09-25T10:50:28+0100

mathew77 said:
Perfect!.. It's 9 seconds now!..

View attachment 592427

There's still a little headroom to improvements, CPU load average is ~85-90% but for now it's no less then python version, and I love it, my gratitudes.

Tell me what improvements you want, I've got my own few tweaks to do but I'll consider adding yours. Once that's done you'll get a nice gui to use it with so it will basically be drag and drop. Also increase threads to 40, it will be even faster.

duckbill007 · 2026-09-25T11:26:10+0100

Long distance mode is useless in case of block compression because your blocks are smaller than distance.
Also python code still uses native zstd library, so it is the same code. Also nicoboss' is the reference implementation, so I prefer to always use it.

Homebrew NSP to NSZ - C++

Well-Known Member

Attachments

Lovin' life.

Well-Known Member

Lovin' life.

Well-Known Member

Attachments

Lovin' life.

Well-Known Member

Lovin' life.

Well-Known Member

Attachments

Lovin' life.

Well-Known Member

Lovin' life.

Well-Known Member

Attachments

Lovin' life.

Well-Known Member

Well-Known Member

Similar threads

Popular threads in this forum