Closed yoosan closed 8 years ago
Hmmm... not much to go on there. Can you do the following please:
uname -a
cat /etc/lsb-release
clinfo
train.lua
, then rerun, using luajit
in place of th
:path = {}
function path.join(a,b)
return a .. '/' .. b
end
function path.exists(a)
print('exists', a)
return true
end
rungdb.sh
, with the following contents:#!/bin/bash
gdb $1 -ex "catch throw" -ex "run $2 $3 $4 $5 $6 $7 $8 $9"
Run as before, but put ./rungdb.sh
in front of luajit train.lua
. When it crashes, type:
bt
... and provide the output
Thanks!!
I run this code on my iMac(OS X Yosemite 10.10). So i haven't the file /etc/lsb-release
. Since I don't have knowledge of OpenCL and more experience of torch. Is there any package I haven't installed? I mean, does the library OpenCL exists in my Mac? I will read code carefully.
Hi, after googling a bit, for 'mac os x opencl', looks like:
clinfo
commandHowever, I think the easiest might be to install EasyCL, and try that. Can you do the following please? :
git clone --recursive https://github.com/hughperkins/EasyCL.git
cd EasyCL
mkdir build
cd build
ccmake ..
# press 'c' for configure
# set BUILD_TESTS to 'ON',
# change BUILD_TYPE to 'Debug'
# press 'c' for configure, then 'g' for generate
# => ccmake should exit back to commandline, then do:
make -j 4
cp ../test/*.cl .
./gpuinfo
./easycl_unittests
and provide the output of the last two commands, and also of any build errors.
OK, some results as follows:
clinfo
not found.[1] 3769 segmentation fault luajit -l cltorch -e 'cltorch.test()'
This is another report. th> cltorch.getDeviceProperties(1) { deviceType : "GPU" localMemSizeKB : 16 globalMemSizeMB : 384 deviceVersion : "OpenCL 1.0 " platformVendor : "Apple" deviceName : "Radeon HD 4850" maxComputeUnits : 10 globalMemCachelineSizeKB : 0 openClCVersion : "OpenCL C 1.0 " maxClockFrequency : 503 maxMemAllocSizeMB : 128 maxWorkGroupSize : 1024 }
EasyCL also works, when i run ./gpuinfo
then i get
num platforms: 1
platform index: 0: platform id: 0x7fff0000 platform vendor: Apple platform name: Apple platform num devices: 2
device index: 0 device id: 0xffffffff device type: 2 global memory size: 16384MB local memory size: 32KB global cache size: 0KB global cacheline size: 8388608 max memory alloc size: 4096MB max compute units: 4 max workgroup size: 1024 max workitem dimensions: 3 max workitem sizes: 1024 1 1 device name: Intel(R) Core(TM) i5 CPU 750 @ 2.67GHz opencl c version: OpenCL C 1.2 opencl device version: OpenCL 1.2 frequency MHz: 2660
device index: 1 device id: 0x1021a00 device type: 4 global memory size: 384MB local memory size: 16KB global cache size: 0KB global cacheline size: 0 max memory alloc size: 128MB max compute units: 10 max workgroup size: 1024 max workitem dimensions: 3 max workitem sizes: 1024 1024 1024 device name: Radeon HD 4850 opencl c version: OpenCL C 1.0 opencl device version: OpenCL 1.0 frequency MHz: 503
Ah. Unfortunately, cltorch needs a GPU that supports OpenCL 1.1.
Awesome! I will find some way to change my OpenCL version. sigh... I will post here If I get a good result. Thank you, hughperkins!
Here I will post a issue. I run the code char-rnn and then I got a error:
[1] 8608 segmentation fault th train.lua -opencl 1
I have installed packagecltorch
andclnn
. It works when I testtorch.rand(2,3):cl()
. I don't know where makes error.