This is a warning from nvcc that it output by default. I don't understand why updating Theano help with that. This happen when we compile gpu code in the old back-end. As we are transitioning to the new back-end, I don't think we will do a work around that in the old back-end.
The reason why you don't get anymore is probably because Theano compilation cache contain all what you need, so we don't compile anymore. But if you change your modele to request compilation unit that isn'T in the cache, it should apprear again. The medium term solution is to use the new back-end. Don't hesitate to use it: https://github.com/Theano/Theano/wiki/Converting-to-the-new-gpu-back-end%28gpuarray%29 Fred On Mon, Jan 9, 2017 at 10:01 AM, Roach Sinai <[email protected]> wrote: > Sorry for these reply. > > pip install --upgrade --no-deps git+git://github.com/Theano/Theano.git > > > Bleeding-edge solve the problem! > 在 2017年1月9日星期一 UTC+8下午8:57:22，Roach Sinai写道： > >> Hi, Pascal. Why I run the same python script, also *Creating library. *Every >> time I run the same script! >> >> DEBUG: nvcc STDOUT nvcc warning : The 'compute_20', 'sm_20', and 'sm_21' >> architectures are deprecated, and may be removed in a future release (Use >> -Wno-deprecated-gpu-targets to suppress warning). >> mod.cu >> 正在创建库 C:/Users/DD/AppData/Local/Theano/compiledir_Windows-10-10.0. >> 14393-SP0-Intel64_Family_6_Model_58_Stepping_9_GenuineInt >> el-3.5.2-64/tmp0cr8i5sp/m91973e5c136ea49268a916ff971b7377.lib 和对象 >> C:/Users/DD/AppData/Local/Theano/compiledir_Windows-10-10.0. >> 14393-SP0-Intel64_Family_6_Model_58_Stepping_9_GenuineInt >> el-3.5.2-64/tmp0cr8i5sp/m91973e5c136ea49268a916ff971b7377.exp >> >> Can't find the reason. Could you give me some advices? >> >> 在 2015年1月15日星期四 UTC+8上午8:04:50，Pascal Lamblin写道： >>> >>> Hi, >>> >>> On Wed, Jan 14, 2015, yifan xie wrote: >>> > first time post, so please bare with me. >>> > >>> > regarding the mentioned message, with the bleeding edge version, I >>> also get >>> > the same message (running same configuration with the original report >>> -- >>> > Win 8.1 and CUDA 6.0 >>> > >>> > The message itself doesn't do any harm, although I am not sure whether >>> it >>> > slow theano down? there seems to be occasion that I get this repeating >>> > message for a long time (then perhaps it is just data being loaded to >>> the >>> > GPU)? >>> >>> The message itself is harmless, it simply indicates that a module >>> containing >>> CUDA code is being compiled. >>> These modules contain automatically-generated code that perform the >>> operations defined by your model, on the GPU. >>> >>> These compiled modules are cached for a couple of weeks, so that if you >>> execute >>> the same code on the same machine, it does not get recompiled every >>> time. >>> This is why the first time you execute some Theano code, more code will >>> be >>> generated and compiled, but the following times it should be much >>> faster. >>> >>> > Another issue to do with this, it create LOTS of tmp folders in the >>> > "AppData/Local/Theano/<platform info>/" directory, see the following >>> > screenshot, I get 30+ M of tmp data (mostly created by process related >>> to >>> > this message) in two run of the logistic regression example. >>> > >>> > <https://lh5.googleusercontent.com/-N7czc9nR0eI/VLb2Hr9k8_I/ >>> AAAAAAAAGDY/kbdVtxqMeLI/s1600/Capture.PNG> >>> >>> Yes, this is the cache for compiled modules. >>> Why do you think it is an issue? >>> >>> -- >>> Pascal >>> >> -- > > --- > You received this message because you are subscribed to the Google Groups > "theano-users" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- --- You received this message because you are subscribed to the Google Groups "theano-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
