Hi David, First of all, I noticed I put a wrong title to my mail : the aditionnal memory transfert is from device to host. The buffer d_c_buf is read back, in the file matrix-multiply.py in the examples packaged with pyopencl. As you will see, it's size is 2 times smaller than the size of d_a_buf and d_b_buf. The 2 memcpyHtoDasync of 512 MB are the transfert of the matrix A and B (filling d_a_buf and d_b_buf) and there should be one memcpyDtoHasync of 256 MB (reading back d_c_buf to get the matrix C). It seems the 3 buffer d_a_buf, d_b_buf and d_c_buf are read back another time. Regards, Nicolas David Garcia a écrit :

Hi Andreas, I didn't know what settings you had in your temp_cl_profiler.conf file, so made a run with default settings. $ python matrix-multiply.py $ cat opencl_profile.log # OPENCL_PROFILE_LOG_VERSION 1.0 # OPENCL_DEVICE 0 Tesla C1060 # TIMESTAMPFACTOR fcec360a5d451e4 method,gputime,cputime,occupancy method=[ memcpyHtoDasync ] gputime=[ 3487.264 ] cputime=[ 4092.000 ] method=[ memcpyHtoDasync ] gputime=[ 2722.880 ] cputime=[ 3283.000 ] method=[ matrixMul ] gputime=[ 91482.430 ] cputime=[ 13.000 ] occupancy=[ 0.750 ] method=[ matrixMul ] gputime=[ 91467.359 ] cputime=[ 12.000 ] occupancy=[ 0.750 ] method=[ matrixMul ] gputime=[ 91516.477 ] cputime=[ 13.000 ] occupancy=[ 0.750 ] method=[ memcpyDtoHasync ] gputime=[ 1119.680 ] cputime=[ 1893.000 ] method=[ memcpyDtoHasync ] gputime=[ 2664.768 ] cputime=[ 3464.000 ] method=[ memcpyDtoHasync ] gputime=[ 1117.952 ] cputime=[ 1823.000 ] method=[ memcpyDtoHasync ] gputime=[ 2690.976 ] cputime=[ 3514.000 ] The 3 last lines are the extra memory transfert. gputime and cputime are not accurate because of other jobs running on the host computer at the moment. For instance gputime for the first, 2nd, 7th and last row should be equal (512 Mb transfert). My nvidia driver is 190.29 and the device is a tesla c1060. Feel free to ask additionnal information. Regards, Nicolas Andreas Klöckner a écrit :

Andreas Klöckner a écrit : Hi, The generated config file look like this : gpustarttimestamp local_load local_store branch divergent_branch ndrangesize workgroupsize dynsmemperblock stapmemperworkgroup regperworkitem memtransfersize streamid I don't think this will help much :( Nicolas

Hi Nick, I agree this would be pretty exciting ! Imagination Technologies (Imgtec) has publicly announced they *will* support OpenCL in the future though they didn’t disclose any availability date. Since they’re behind the majority of the current embedded 3D IP (including Nokia, Apple, Motorola Droid etc…), this should represent a fairly large number of target devices. TI, Freescale, Marvell, NXP, Samsung are all Imgtec licensees. Both Apple and Intel have a large stake (respectively 9.5% and 16%) in Imagination. Then it’s a matter of whether the OS will make it available to the user. Nokia has an history of supporting python on its handheld (I got an E61 then an N900 that both support python). Don’t know about Apple though I would be more doubtful. Don’t know about other manufacturers either. Below is a quote from IMGTEC (source: http://www.khronos.org/news/press/releases/imagination-technologies-announce s-powervr-sgx545-graphics-ip-core-with-ope/) Las Vegas, USA, 8th January 2010: Imagination Technologies, a leading multimedia chip technologies company, announces POWERVR SGX545, the first and only DirectX10.1 capable embedded graphics IP core available for immediate licensing. SGX545 will also deliver OpenGL ES 2.x and OpenGL 3.2 to deliver class leading 3D graphics performance, and will also support OpenCL 1.0 full profile capability which will enable mobile and embedded applications to take maximum advantage of the capabilities offered by these GPU APIs for both 3D graphics and general purpose applications. ARM provides its own 3D core known as Mali (with currently 27 licencees including Samsung) but they didn’t tell their plans about OpenCL yet. Regarding Google, sources (http://www.geeky-gadgets.com/google-nexus-one-phone-hardware-specifications -revealed-16-12-2009/) reveal the nexus one is powered by a snapdragon processor from Qualcomm. Since Qualcomm acquired AMD mobile 3D graphics assets back in January 2009, they must be brewing their own, yet-to-be-announced solution. Best regards, Hervé From: pyopencl-bounces(a)tiker.net [mailto:pyopencl-bounces@tiker.net] On Behalf Of Nick Gaens Sent: mardi 9 février 2010 18:36 To: pyopencl(a)tiker.net Subject: [PyOpenCL] PyOpenCL for handhelds Hello all, I'm wondering if there are handhelds / mobile devices out there that are known to support OpenCL. I've read the following on Khronos' page <http://www.khronos.org/opencl/> : OpenCL (Open Computing Language) is the first open, royalty-free standard for general-purpose parallel programming of heterogeneous systems. OpenCL provides a uniform programming environment for software developers to write efficient, portable code for high-performance compute servers, desktop computer systems and handheld devices using a diverse mix of multi-core CPUs, GPUs, Cell-type architectures and other parallel processors such as DSPs. So, such devices exist, based upon this small introductionary piece of text. Although I cannot determine which range of devices is actually supporting OpenCL. Just above the previous quote, though, there is this piece of text (same page): OpenCL is being created by the Khronos Group with the participation of many industry-leading companies and institutions including 3DLABS, Activision Blizzard, AMD, Apple, ARM, Broadcom, Codeplay, Electronic Arts, Ericsson, Freescale, Fujitsu, GE, Graphic Remedy, HI, IBM, Intel, Imagination Technologies, Los Alamos National Laboratory, Motorola, Movidia, Nokia, NVIDIA, Petapath, QNX, Qualcomm, RapidMind, Samsung, Seaweed, S3, ST Microelectronics, Takumi, Texas Instruments, Toshiba and Vivante. I've marked the companies that I know of being developers of mobile device hardware in one way or another, but Google e.g. doesn't know any interesting combination of "OpenCL" and "mobile device" or "handheld" :-(. Very strange, since there are that much developers involved here. The reason why I'm asking this on this mailing list is that I'm willing to target some mobile device or platform and am going to do some (in-depth) research (MSc) about OpenCL on that mobile platform and since Python is available on almost every known platform out there, PyOpenCL gives me a splendid way to actually build such OpenCL-powered applications. Can someone elaborate on this matter? Maybe one can point me in some relevant direction? Thx in advance, Nick