[PATCH 11/16] lib: enable lzma decompression support for SPL build

[Tom Rini]

On Fri, Jan 10, 2020 at 04:21:35PM +0800, Weijie Gao wrote:
> On Thu, 2020-01-09 at 12:45 -0500, Tom Rini wrote:
> > On Thu, Jan 09, 2020 at 04:10:12PM +0800, Weijie Gao wrote:
> > > On Wed, 2020-01-08 at 08:22 +0100, Heinrich Schuchardt wrote:
> > > > 
> > > > On 1/8/20 4:01 AM, Weijie Gao wrote:
> > > > > This patch enables LZMA decompression support for SPL build
> > > > >
> > > > > Signed-off-by: Weijie Gao <weijie.gao at mediatek.com>
> > > > > ---
> > > > >   lib/Kconfig  | 5 +++++
> > > > >   lib/Makefile | 1 +
> > > > >   2 files changed, 6 insertions(+)
> > > > >
> > > > > diff --git a/lib/Kconfig b/lib/Kconfig
> > > > > index 965cf7bc03..2f38e37802 100644
> > > > > --- a/lib/Kconfig
> > > > > +++ b/lib/Kconfig
> > > > > @@ -418,6 +418,11 @@ config SPL_LZ4
> > > > >   	  fast compression and decompression speed. It belongs to the LZ77
> > > > >   	  family of byte-oriented compression schemes.
> > > > >
> > > > > +config SPL_LZMA
> > > > > +	bool "Enable LZMA decompression support for SPL build"
> > > > > +	help
> > > > > +	  This enables support for LZMA compression altorithm for SPL boot.
> > > > 
> > > > %s/altorithm/algorithm/g
> > > > 
> > > > Why do we need yet another algorithm for compression?
> > > > 
> > > > Doesn't Zstandard provide a much faster decompression?
> > > > Cf. https://sites.google.com/site/powturbo/home/web-compression
> > > > 
> > > > Best regards
> > > > 
> > > > Heinrich
> > > > 
> > > > > +
> > > > >   config SPL_LZO
> > > > >   	bool "Enable LZO decompression support in SPL"
> > > > >   	help
> > > > > diff --git a/lib/Makefile b/lib/Makefile
> > > > > index 1fb650cd90..8e51eec01c 100644
> > > > > --- a/lib/Makefile
> > > > > +++ b/lib/Makefile
> > > > > @@ -64,6 +64,7 @@ obj-$(CONFIG_$(SPL_)ZLIB) += zlib/
> > > > >   obj-$(CONFIG_$(SPL_)ZSTD) += zstd/
> > > > >   obj-$(CONFIG_$(SPL_)GZIP) += gunzip.o
> > > > >   obj-$(CONFIG_$(SPL_)LZO) += lzo/
> > > > > +obj-$(CONFIG_$(SPL_)LZMA) += lzma/
> > > > >   obj-$(CONFIG_$(SPL_)LZ4) += lz4_wrapper.o
> > > > >
> > > > >   obj-$(CONFIG_LIBAVB) += libavb/
> > > > >
> > > 
> > > Hi Heinrich,
> > > 
> > > I need an algorithm with the maximum compression ratio to reduce the
> > > size of the u-boot payload as much as possible.
> > > 
> > > I believe currently lzma and xz are the best choices.
> > > Refer to:
> > > https://catchchallenger.first-world.info/wiki/Quick_Benchmark:_Gzip_vs_Bzip2_vs_LZMA_vs_XZ_vs_LZ4_vs_LZO
> > > 
> > > Although they will take longer time for compression and decompression,
> > > considering that the u-boot payload is usually small (<= 1MiB), there
> > > will not be significant effect on decompression time.
> > 
> > Where does zstd fall in the list however?  Can you please try that and
> > report back?  Thanks!
> > 
> 
> My purpose is to reduce the size of u-boot binary, and my plan is to
> compress the payload and the SPL decompress it at runtime.
> I need smallest size here, not fastest decompression speed.
> 
> So there are two points must be taken into consideration:
> the size of the decompressor (i.e. u-boot-spl.bin) and the size of the
> compressed payload (u-boot.bin.gz/bz2/...).
> 
> Below is my test results, including all compression algorithms supported
> by uImage, plus zstd:
> 
> Platform: mips - mt7628
> 
> Uncompressed size of u-boot.bin: 302736 (295KiB)
> 
> Commands for compression:
> gzip -c -f -k -9
> bzip2 -c -f -k -9
> lzma -c -z -k -9
> lzop -c -f -9
> lz4 -f -9
> zstd --ultra -f
> 
> Results:
> algo. | spl.bin size  | compressed size | % of uncomp | total size
> ------------------------------------------------------------------------
> gzip  | 26232 (25.6K) | 141207 (137KiB) | 46.64%      | 167439 (163KiB)
> ------------------------------------------------------------------------
> bzip2 | 30616 (29.8K) | 135403 (132KiB) | 44.72%      | 166019 (162KiB)
> ------------------------------------------------------------------------
> lzma  | 18920 (18.4K) | 112660 (110KiB) | 37.21%      | 131580 (128KiB)
> ------------------------------------------------------------------------
> lzo   | 13288 (12.9K) | 152409 (148KiB) | 50.34%      | 165697 (161KiB)
> ------------------------------------------------------------------------
> lz4   | 12408 (12.1K) | 167311 (163KiB) | 55.27%      | 179719 (175KiB)
> ------------------------------------------------------------------------
> zstd  | 44824 (43.7K) | 149765 (146KiB) | 49.47%      | 194589 (190KiB)
> 
> As you can see lzma has the best compression ratio. The is the reason I
> chose lzma here.

Thanks, that's exactly the kind of data I was looking for.

Reviewed-by: Tom Rini <trini at konsulko.com>

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 833 bytes
Desc: not available
URL: <https://lists.denx.de/pipermail/u-boot/attachments/20200110/e5894b33/attachment.sig>