Probably another processes didn't allow to show full performance in your run.
It's better to close all another prgrams including internet browser before run. Also check in Task Manager that CPU load is about 0% before run.
-slp test also didn't use large pages. Try to run " 7-Zip File Manager with administrator's rights as written in post above.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Igor, I tested both arm64 and arm32 on:
SD800 - Nokia Lumia 2520 - Windows RT 8.1 (Unlocked Test Mode) arm32 - works fine (it requires both installer and program bins to be signed with cert to satisfy Test mode's running wim32 programs)
SD835 - HP Envy x2 arm64 - both work OK, but arm32 version should:
1) be installed to \Program Files (Arm) NOT to \Program Files (x86) (when x86 is already present)
2) (the same for x86, when arm32 is already present)
3) use only Uninstall in MAIN and WOW6432Node nodes (since WowAA32Node Uninstall node doesn't work on W10 arm64).
Please, do NOT drop arm32 version in future, since Windows RT 8-x arm32 can't run arm64 win32 bins.
Because this key DOESN'T display anything in Unistall list in CP on W10 arm64 by MS's fault: [HKEY_LOCAL_MACHINE\SOFTWARE\WowAA32Node\Microsoft\Windows\CurrentVersion\Uninstall]...
...I made the attached REG.
Of course 7-Zip\Uninstall.exe isn't aware of any of that.
Thanks.
Is it Xcode compiler?
7-Zip versions 16.02 and 20.02 have different optimizations for lzma compression.
But 16.02 results also are interesting.
Please run also full benchmark in Apple:
7zb-mm=*-mmt=*
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
OPTFLAGS=-O2-mcpu=vortex-mtune=vortex
# don't use -DNDEBUG in order to use MacOSX wxWidgetsALLFLAGS=-archarm64 ${OPTFLAGS} \
-DENV_MACOSX-DNDEBUG \
-D_FILE_OFFSET_BITS=64-D_LARGEFILE_SOURCE \
-D_REENTRANT-DENV_UNIX \
$(LOCAL_FLAGS)
# ForoldMacOSX :
# CXX=/Developer/usr/bin/llvm-g++
# CC=/Developer/usr/bin/llvm-gcc
# FornewMacOSX :
XX=/usr/bin/clang++CC=/usr/bin/clangLINK_SHARED=-bundleLOCAL_LIBS=-frameworkCoreFoundation-liconvLOCAL_LIBS_DLL=$(LOCAL_LIBS)OBJ_CRC32=$(OBJ_CRC32_C)OBJ_AES=
Disregard the test results attached to this post... there were OS processes eating more than a full CPU due to a bug, now fixed and doing a second run...
Thanks for these results.
7-Zip benchmark compresses only synthetic data that doesn't represent real world data very well. Also 7-Zip algorithms strolgly depends from memory and cache and have low IPC and it likes big freqency also (for decoding mostly). That is why A12Z in 7-Zip is not so good as A12Z in some another workloads where there is high IPC.
Actually Apple A9 core was already pretty good for 7-Zip workloads.
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Probably another processes didn't allow to show full performance in your run.
It's better to close all another prgrams including internet browser before run. Also check in Task Manager that CPU load is about 0% before run.
-slptest also didn't use large pages. Try to run " 7-Zip File Manager with administrator's rights as written in post above.Igor, I tested both arm64 and arm32 on:
SD800 - Nokia Lumia 2520 - Windows RT 8.1 (Unlocked Test Mode) arm32 - works fine (it requires both installer and program bins to be signed with cert to satisfy Test mode's running wim32 programs)
SD835 - HP Envy x2 arm64 - both work OK, but arm32 version should:
1) be installed to \Program Files (Arm) NOT to \Program Files (x86) (when x86 is already present)
2) (the same for x86, when arm32 is already present)
3) use only Uninstall in MAIN and WOW6432Node nodes (since WowAA32Node Uninstall node doesn't work on W10 arm64).
Please, do NOT drop arm32 version in future, since Windows RT 8-x arm32 can't run arm64 win32 bins.
Last edit: Final Aeon 2020-08-13
Because this key DOESN'T display anything in Unistall list in CP on W10 arm64 by MS's fault:
[HKEY_LOCAL_MACHINE\SOFTWARE\WowAA32Node\Microsoft\Windows\CurrentVersion\Uninstall]...
...I made the attached REG.
Of course 7-Zip\Uninstall.exe isn't aware of any of that.
Here is the test of p7zip(16.02) on A12Z that you may interest in. It looks like it's the fastest result of these ARM tests.
Thanks.
Is it Xcode compiler?
7-Zip versions 16.02 and 20.02 have different optimizations for lzma compression.
But 16.02 results also are interesting.
Please run also full benchmark in Apple:
With makefile.machine:
Disregard the test results attached to this post... there were OS processes eating more than a full CPU due to a bug, now fixed and doing a second run...
Last edit: never_released 2020-08-16
Second run (with the same binary):
Last edit: never_released 2020-08-16
Thanks for these results.
7-Zip benchmark compresses only synthetic data that doesn't represent real world data very well. Also 7-Zip algorithms strolgly depends from memory and cache and have low IPC and it likes big freqency also (for decoding mostly). That is why A12Z in 7-Zip is not so good as A12Z in some another workloads where there is high IPC.
Actually Apple A9 core was already pretty good for 7-Zip workloads.