[Beignet] Sample program crash with beignet 0.3 on IVB
Zhigang Gong
zhigang.gong at linux.intel.com
Wed Jan 8 20:38:06 PST 2014
This is a known bug as clang generated some llvm.memcpy or llvm.memset
intrinsics which are not supported in beignet. We are fixing it now.
On Wed, Jan 08, 2014 at 09:08:53PM +0100, Robert Jobbagy wrote:
> I updated opencv to 2.4.8 release (it has lot of bug fixes for ocl module )
> and I use beignet opencl-1.2 head version and I gave this seg fault :
>
> ASSERTION FAILED: Unsupported intrinsics
> at file beignet_opencl_1_2/backend/src/llvm/llvm_gen_backend.cpp,
> function void gbe::GenWriter::regAllocateCallInst(llvm::CallInst&), line
> 1982
> Stack dump:
> 0. Running pass 'Function Pass Manager' on module '/tmp/fileHPFMF4.ll'.
> 1. Running pass 'Gen Back-End' on function '@edgesHysteresisLocal'
>
> Can I help you with more information ?
>
>
>
> 2013/12/5 Robert Jobbagy <jobbagy.robert at gmail.com>
>
> > if I use opencv 2.4.7.1 branch head , I gave differenct sigsev :
> >
> > #1 0x00007ffff733fb1e in (anonymous namespace)::opencl_fn9<62, int,
> > _cl_command_queue*, _cl_mem*, void const*, unsigned long, unsigned long,
> > unsigned long, unsigned int, _cl_event* const*, _cl_event**>::switch_fn
> > (p1=0x73fb00, p2=0x7d25a0, p3=0x7442e0, p4=1, p5=0, p6=307200, p7=0,
> > p8=0x0, p9=0x0) at
> > /home/evil85/workspace/cpp/opencv/modules/ocl/src/cl_runtime/cl_runtime_opencl_impl.hpp:266
> > No locals.
> > #2 0x00007ffff72ab34d in set_to_withoutmask_run (dst=..., scalar=...,
> > kernelName=...)
> > at
> > /home/evil85/workspace/cpp/opencv/modules/ocl/src/matrix_operations.cpp:449
> > p = {<std::_Vector_base<unsigned char, std::allocator<unsigned
> > char> >> = {
> > _M_impl = {<std::allocator<unsigned char>> =
> > {<__gnu_cxx::new_allocator<unsigned char>> = {<No data fields>}, <No data
> > fields>},
> > _M_start = 0x7442e0 "", _M_finish = 0x7442e1 "",
> > _M_end_of_storage = 0x7442e1 ""}}, <No data fields>}
> > localThreads = {16, 16, 1}
> > step_in_pixel = 640
> > buildOptions = {static npos = <optimized out>,
> > _M_dataplus = {<std::allocator<char>> =
> > {<__gnu_cxx::new_allocator<char>> = {<No data fields>}, <No data fields>},
> > _M_p = 0x7442b8 "-D GENTYPE=uchar "}}
> > args = {<std::_Vector_base<std::pair<unsigned long, void const*>,
> > std::allocator<std::pair<unsigned long, void const*> > >> = {
> > _M_impl = {<std::allocator<std::pair<unsigned long, void
> > const*> >> = {<__gnu_cxx::new_allocator<std::pair<unsigned long, void
> > const*> >> = {<No data fields>}, <No data fields>}, _M_start = 0x0,
> > _M_finish = 0x0, _M_end_of_storage = 0x0}}, <No data fields>}
> > globalThreads = {176, 480, 1}
> > offset_in_pixel = 0
> > typeMap = {0x7ffff7345909 "uchar", 0x7ffff734590f "char",
> > 0x7ffff7345914 "ushort", 0x7ffff734591b "short", 0x7ffff7345921 "int",
> > 0x7ffff7345925 "float", 0x7ffff734592b "double"}
> > channelMap = " 244"
> > mat = {flags = 1124024320, dims = 2, rows = 1, cols = 1, data =
> > 0x7d27a0 "", refcount = 0x7d27a4, datastart = 0x7d27a0 "",
> > dataend = 0x7d27a1 "", datalimit = 0x7d27a1 "", allocator = 0x0,
> > size = {p = 0x7fffffffcec8}, step = {p = 0x7fffffffcf10, buf = {1,
> > 1}}}
> >
> >
> > 2013/12/4 Robert Jobbagy <jobbagy.robert at gmail.com>
> >
> >> Earlier I made an issue in opencv track here :
> >> http://code.opencv.org/issues/3412
> >>
> >> Thanks your patch.
> >>
> >> I will try compile with llvm 3.3
> >>
> >>
> >> 2013/12/4 Robert Jobbagy <jobbagy.robert at gmail.com>
> >>
> >>> I use Ubuntu 13.10 default 3.2 llvm
> >>>
> >>>
> >>> 2013/12/4 Zhigang Gong <zhigang.gong at linux.intel.com>
> >>>
> >>>> I just tried the same way as you. But the sympton is not the same, I
> >>>> met a
> >>>> region check bug which I just sent a patch to fix.
> >>>>
> >>>> After that fix, the squares example could run to complete but the result
> >>>> is incorrect, the acuracy check is also failed. I will continue to
> >>>> investigate
> >>>> this issue.
> >>>>
> >>>> The llvm/clang I am using is 3.3.1. What's yours?
> >>>>
> >>>> On Wed, Dec 04, 2013 at 09:32:06AM +0100, Robert Jobbagy wrote:
> >>>> > I attached my output log.
> >>>> >
> >>>> >
> >>>> > 2013/12/4 zhigang gong <zhigang.gong at gmail.com>
> >>>> >
> >>>> > >
> >>>> > >
> >>>> > > On Wed, Dec 4, 2013 at 3:23 AM, Robert Jobbagy <
> >>>> jobbagy.robert at gmail.com>wrote:
> >>>> > >
> >>>> > >> Thanks your help, it works.
> >>>> > >>
> >>>> > >> I tried this opencl example :
> >>>> > >>
> >>>> > >> http://wiki.tiker.net/OpenCLHowTo#Testing
> >>>> > >>
> >>>> > >> and it shows these supported extensions by IVB:
> >>>> > >>
> >>>> > >> NAME: Intel HD Graphics Family
> >>>> > >> VENDOR: Intel
> >>>> > >> PROFILE: FULL_PROFILE
> >>>> > >> VERSION: OpenCL 1.2 beignet 0.3
> >>>> > >> EXTENSIONS: cl_khr_global_int32_base_atomics
> >>>> > >> cl_khr_global_int32_extended_atomics
> >>>> cl_khr_local_int32_base_atomics
> >>>> > >> cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store
> >>>> > >> cl_khr_icd cl_khr_gl_sharing
> >>>> > >> DRIVER_VERSION: 0.3
> >>>> > >>
> >>>> > >>
> >>>> > >> but when I use opencv with opencl I gave CL_BUILD_PROGRAM_FAILURE
> >>>> error.
> >>>> > >> I think the problem is opencv want to use cl_khr_fp64 extension and
> >>>> > >> beignet doesnt support yet or not ?
> >>>> > >>
> >>>> > > Could you provide more details about the failure? The complete
> >>>> error
> >>>> > > information, and the specific procedure to reproduce this failure.
> >>>> > >
> >>>> > > Thanks,
> >>>> > > Zhigang
> >>>> > >
> >>>> > >>
> >>>> > >> What can I do ?
> >>>> > >>
> >>>> > >> Thanks again.
> >>>> > >>
> >>>> > >>
> >>>> > >> 2013/12/2 Zhigang Gong <zhigang.gong at linux.intel.com>
> >>>> > >>
> >>>> > >>> On Sat, Nov 30, 2013 at 10:57:00PM +0100, Robert Jobbagy wrote:
> >>>> > >>> > Thanks your help,
> >>>> > >>> >
> >>>> > >>> > I used mesa 9.2.1 and it's contains this patch.
> >>>> > >>> You need to use the eaxct version specified by that commit log,
> >>>> not just
> >>>> > >>> includes that patch.
> >>>> > >>> So you may need to use git checkout 92e62...
> >>>> > >>> >
> >>>> > >>> > I set mesa source and gone previous errors but I gave these
> >>>> compile
> >>>> > >>> errors.
> >>>> > >>> > I used opencl 1.2 branch:
> >>>> > >>> >
> >>>> > >>> >
> >>>> > >>>
> >>>> /home/evil85/Development/beignet_opencl_1_2/src/intel/intel_dri_resource_sharing.c:
> >>>> > >>> > In function ‘intel_get_gl_obj_from_texture’:
> >>>> > >>> >
> >>>> > >>>
> >>>> /home/evil85/Development/beignet_opencl_1_2/src/intel/intel_dri_resource_sharing.c:103:38:
> >>>> > >>> > error: dereferencing pointer to incomplete type
> >>>> > >>> > obj = __mesa_lookup_texture(&intel->ctx, texture);
> >>>> > >>> > ^
> >>>> > >>> >
> >>>> > >>>
> >>>> /home/evil85/Development/beignet_opencl_1_2/src/intel/intel_dri_resource_sharing.c:109:42:
> >>>> > >>> > error: dereferencing pointer to incomplete type
> >>>> > >>> > __mesa_test_texobj_completeness(&intel->ctx, obj);
> >>>> > >>>
> >>>> > >>>
> >>>> > >>>
> >>>> > >
> >>>> >
> >>>> >
> >>>> > --
> >>>> > Best Regards,
> >>>> >
> >>>> > Robert
> >>>>
> >>>> > VIDIOC_QUERYMENU: Invalid argument
> >>>> > VIDIOC_QUERYMENU: Invalid argument
> >>>> > VIDIOC_QUERYMENU: Invalid argument
> >>>> > VIDIOC_QUERYMENU: Invalid argument
> >>>> > VIDIOC_QUERYMENU: Invalid argument
> >>>> > VIDIOC_QUERYMENU: Invalid argument
> >>>> > VIDIOC_QUERYMENU: Invalid argument
> >>>> > platinfo size: 1
> >>>> > Vendor:Intel
> >>>> > Name: Experiment Intel Gen OCL Driver
> >>>> > PlatformProfile: FULL_PROFILE
> >>>> > Version: OpenCL 1.2 beignet 0.3
> >>>> > Major version: 1 Minor ver: 2
> >>>> > How many devices: 1
> >>>> > Device Name:Intel HD Graphics Family
> >>>> > DriverVer: 0.3
> >>>> > Intel: 1 Integraded? 0
> >>>> > Device profile: FULL_PROFILE
> >>>> > Device Type: 4
> >>>> > Device Vendor: Intel
> >>>> > Device Version: OpenCL 1.2 beignet 0.3
> >>>> > warming up ...
> >>>> > Devinfo compilationExtra: -D INTEL_DEVICE build options: -D
> >>>> GENTYPE4=uchar4
> >>>> > all build options: -D INTEL_DEVICE -D GENTYPE4=uchar4
> >>>> > source: convertC3C4 programSTR: #if defined (DOUBLE_SUPPORT)
> >>>> > #pragma OPENCL EXTENSION cl_khr_fp64:enable
> >>>> > #endif
> >>>> > __kernel void convertC3C4(__global const GENTYPE4 * restrict src,
> >>>> __global GENTYPE4 *dst, int cols, int rows,
> >>>> > int dstStep_in_piexl,int pixel_end)
> >>>> > {
> >>>> > int id = get_global_id(0);
> >>>> > int3 pixelid = (int3)(mul24(id,3),mad24(id,3,1),mad24(id,3,2));
> >>>> > pixelid = clamp(pixelid,0,pixel_end);
> >>>> > GENTYPE4 pixel0, pixel1, pixel2, outpix0,outpix1,outpix2,outpix3;
> >>>> > pixel0 = src[pixelid.x];
> >>>> > pixel1 = src[pixelid.y];
> >>>> > pixel2 = src[pixelid.z];
> >>>> > outpix0 = (GENTYPE4)(pixel0.x,pixel0.y,pixel0.z,0);
> >>>> > outpix1 = (GENTYPE4)(pixel0.w,pixel1.x,pixel1.y,0);
> >>>> > outpix2 = (GENTYPE4)(pixel1.z,pixel1.w,pixel2.x,0);
> >>>> > outpix3 = (GENTYPE4)(pixel2.y,pixel2.z,pixel2.w,0);
> >>>> > int4 outy = (id<<2)/cols;
> >>>> > int4 outx = (id<<2)%cols;
> >>>> > outx += (int4)(0, 1, 2, 3);
> >>>> > outy = select(outy, outy+1, outx>=cols);
> >>>> > outx = select(outx, outx-cols, outx>=cols);
> >>>> > outy = select(outy, outy + 1, outx >= cols);
> >>>> > outx = select(outx, outx-cols, outx >= cols);
> >>>> > outy = select(outy, outy + 1, outx >= cols);
> >>>> > outx = select(outx, outx-cols, outx >= cols);
> >>>> > int4 addr = mad24(outy,(int4)dstStep_in_piexl,outx);
> >>>> > if(outx.w<cols && outy.w<rows)
> >>>> > {
> >>>> > dst[addr.x] = outpix0;
> >>>> > dst[addr.y] = outpix1;
> >>>> > dst[addr.z] = outpix2;
> >>>> > dst[addr.w] = outpix3;
> >>>> > }
> >>>> > else if(outx.z<cols && outy.z<rows)
> >>>> > {
> >>>> > dst[addr.x] = outpix0;
> >>>> > dst[addr.y] = outpix1;
> >>>> > dst[addr.z] = outpix2;
> >>>> > }
> >>>> > else if(outx.y<cols && outy.y<rows)
> >>>> > {
> >>>> > dst[addr.x] = outpix0;
> >>>> > dst[addr.y] = outpix1;
> >>>> > }
> >>>> > else if(outx.x<cols && outy.x<rows)
> >>>> > {
> >>>> > dst[addr.x] = outpix0;
> >>>> > }
> >>>> > }
> >>>> > __kernel void convertC4C3(__global const GENTYPE4 * restrict src,
> >>>> __global GENTYPE4 *dst, int cols, int rows,
> >>>> > int srcStep_in_pixel,int pixel_end)
> >>>> > {
> >>>> > int id = get_global_id(0)<<2;
> >>>> > int y = id / cols;
> >>>> > int x = id % cols;
> >>>> > int4 x4 = (int4)(x,x+1,x+2,x+3);
> >>>> > int4 y4 = select((int4)y,(int4)(y+1),x4>=(int4)cols);
> >>>> > x4 = select(x4,x4-(int4)cols,x4>=(int4)cols);
> >>>> > y4 = select(y4, y4 + 1,x4>=(int4)cols);
> >>>> > x4 = select(x4, x4 - (int4)cols,x4>=(int4)cols);
> >>>> > y4 = select(y4, y4 + 1,x4>=(int4)cols);
> >>>> > x4 = select(x4, x4-(int4)cols,x4>=(int4)cols);
> >>>> > y4=clamp(y4,(int4)0,(int4)(rows-1));
> >>>> > int4 addr = mad24(y4, (int4)srcStep_in_pixel, x4);
> >>>> > GENTYPE4 pixel0,pixel1,pixel2,pixel3, outpixel1, outpixel2;
> >>>> > pixel0 = src[addr.x];
> >>>> > pixel1 = src[addr.y];
> >>>> > pixel2 = src[addr.z];
> >>>> > pixel3 = src[addr.w];
> >>>> > pixel0.w = pixel1.x;
> >>>> > outpixel1.x = pixel1.y;
> >>>> > outpixel1.y = pixel1.z;
> >>>> > outpixel1.z = pixel2.x;
> >>>> > outpixel1.w = pixel2.y;
> >>>> > outpixel2.x = pixel2.z;
> >>>> > outpixel2.y = pixel3.x;
> >>>> > outpixel2.z = pixel3.y;
> >>>> > outpixel2.w = pixel3.z;
> >>>> > int4 outaddr = mul24(id>>2 , 3);
> >>>> > outaddr.y++;
> >>>> > outaddr.z+=2;
> >>>> > if(outaddr.z <= pixel_end)
> >>>> > {
> >>>> > dst[outaddr.x] = pixel0;
> >>>> > dst[outaddr.y] = outpixel1;
> >>>> > dst[outaddr.z] = outpixel2;
> >>>> > }
> >>>> > else if(outaddr.y <= pixel_end)
> >>>> > {
> >>>> > dst[outaddr.x] = pixel0;
> >>>> > dst[outaddr.y] = outpixel1;
> >>>> > }
> >>>> > else if(outaddr.x <= pixel_end)
> >>>> > {
> >>>> > dst[outaddr.x] = pixel0;
> >>>> > }
> >>>> > }
> >>>> > hash: 63881381b364c3374626fe67e3b07e05
> >>>> > options from params: -D INTEL_DEVICE -D GENTYPE4=uchar4
> >>>> > foptions: entry datasize: 15296
> >>>> > #2 device: 0x7f68ddebd760 createBinaryStatus: 0 options: -D
> >>>> INTEL_DEVICE -D GENTYPE4=uchar4
> >>>> > binary size: 15296 binary[0]:
> >>>> > Device info: Intel HD Graphics Family Profile: FULL_PROFILE
> >>>> > Dev Version: OpenCL 1.2 beignet 0.3 Extensions:
> >>>> cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics
> >>>> cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics
> >>>> cl_khr_byte_addressable_store cl_khr_icd
> >>>> > DeviceType: 4
> >>>> > Platform Info: Experiment Intel Gen OCL Driver profile: FULL_PROFILE
> >>>> > Extensions: cl_khr_global_int32_base_atomics
> >>>> cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics
> >>>> cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store
> >>>> cl_khr_icd Vendor: Intel Version: OpenCL 1.2 beignet 0.3
> >>>> > #2 status: -11
> >>>> >
> >>>> > BUILD LOG: convertC3C4: -D INTEL_DEVICE -D GENTYPE4=uchar4
> >>>> >
> >>>> > status: -11
> >>>> > OpenCV Error: Gpu API call (CL_BUILD_PROGRAM_FAILURE) in
> >>>> getOrBuildProgram, file
> >>>> /home/evil85/workspace/cpp/opencv-2.4.7/modules/ocl/src/cl_programcache.cpp,
> >>>> line 460
> >>>> > terminate called after throwing an instance of 'cv::Exception'
> >>>> > what():
> >>>> /home/evil85/workspace/cpp/opencv-2.4.7/modules/ocl/src/cl_programcache.cpp:460:
> >>>> error: (-217) CL_BUILD_PROGRAM_FAILURE in function getOrBuildProgram
> >>>> >
> >>>> > Aborted (core dumped)
> >>>>
> >>>> > _______________________________________________
> >>>> > Beignet mailing list
> >>>> > Beignet at lists.freedesktop.org
> >>>> > http://lists.freedesktop.org/mailman/listinfo/beignet
> >>>>
> >>>>
> >>>
> >>>
> >>> --
> >>> Best Regards,
> >>>
> >>> Robert
> >>>
> >>
> >>
> >>
> >> --
> >> Best Regards,
> >>
> >> Robert
> >>
> >
> >
> >
> > --
> > Best Regards,
> >
> > Robert
> >
>
>
>
> --
> Best Regards,
>
> Robert
> _______________________________________________
> Beignet mailing list
> Beignet at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/beignet
More information about the Beignet
mailing list