gpuopen-professionalcompute-libraries / amdovx-core Goto Github PK

Hello,
While I am experimenting with the amdovx-core, I tried to run the code for skintonedetect.gdf which is given in the ReadMe.md in runvx folder.
After the code has been executed , I am able to see the input window with the following image rather than the given input image (http://cdn5.applesencia.com/wp-content/blogs.dir/17/files/2013/04/raja-koduri-640x480.jpg) :

Hi,

I am trying a channel_combine with the following:

data out = image:1920,1080,RGB2
node org.khronos.openvx.channel_combine out_r out_g out_b out

or

data out = image:1920,1080,RGBX
node org.khronos.openvx.channel_combine out_r out_g out_b out_a out

Both produce errors. 1st one complains about missing 4th argument and the second about not being able to create RGBX. Is that normal?

Thanks

when i run runvx.exe in command line, such as : runvx.exe file canny.gdf

results is
OK: suing AMD OpenVX 0.9.1
OK enable graph scheduling in separate threads
ERROR: This build does't support CAMERA option

can you share where to modify source file or setting to solve this issue?

thanks.

Hello

I found that vxNonLinearFilterNode is not implemented

How can I setup a GDF file to work with the camera as an input?
Anyone has a good idea?
Cheers

I'm going through the Graph Manager code. As far as my understanding, merging of functions are not yet supported. Am I right?
If so, are you planning to release any updated version?

-- The C compiler identification is GNU 6.3.1
-- The CXX compiler identification is GNU 6.3.1
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Configuring done
-- Generating done
-- Build files have been written to: /home/arpu/Work/githubsources/amdovx-core/build
arpu@localhost  ~/Work/githubsources/amdovx-core/build   master ?  export LC_ALL=C  ✓  6930  17:24:07
arpu@localhost  ~/Work/githubsources/amdovx-core/build   master ?  make  ✓  6931  17:24:09
Scanning dependencies of target openvx
[ 1%] Building CXX object openvx/CMakeFiles/openvx.dir/ago/ago_drama.cpp.o
[ 3%] Building CXX object openvx/CMakeFiles/openvx.dir/ago/ago_drama_alloc.cpp.o
[ 5%] Building CXX object openvx/CMakeFiles/openvx.dir/ago/ago_drama_analyze.cpp.o
[ 7%] Building CXX object openvx/CMakeFiles/openvx.dir/ago/ago_drama_divide.cpp.o
[ 9%] Building CXX object openvx/CMakeFiles/openvx.dir/ago/ago_drama_merge.cpp.o
[ 11%] Building CXX object openvx/CMakeFiles/openvx.dir/ago/ago_drama_remove.cpp.o
[ 13%] Building CXX object openvx/CMakeFiles/openvx.dir/ago/ago_haf_cpu.cpp.o
[ 15%] Building CXX object openvx/CMakeFiles/openvx.dir/ago/ago_haf_cpu_arithmetic.cpp.o
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp: In function 'int HafCpu_Add_S16_U8U8(vx_uint32, vx_uint32, vx_int16*, vx_uint32, vx_uint8*, vx_uint32, vx_uint8*, vx_uint32)':
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:797:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels1L = _mm_cvtepu8_epi16(pixels1L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:834:43: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels1L = _mm_cvtepu8_epi16(pixels1L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp: In function 'int HafCpu_Sub_S16_U8U8(vx_uint32, vx_uint32, vx_int16*, vx_uint32, vx_uint8*, vx_uint32, vx_uint8*, vx_uint32)':
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:986:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels1L = _mm_cvtepu8_epi16(pixels1L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1022:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels1L = _mm_cvtepu8_epi16(pixels1L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp: In function 'int HafCpu_Add_S16_S16U8_Wrap(vx_uint32, vx_uint32, vx_int16*, vx_uint32, vx_int16*, vx_uint32, vx_uint8*, vx_uint32)':
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1172:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels2L = _mm_cvtepu8_epi16(pixels2L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1207:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels2L = _mm_cvtepu8_epi16(pixels2L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp: In function 'int HafCpu_Add_S16_S16U8_Sat(vx_uint32, vx_uint32, vx_int16*, vx_uint32, vx_int16*, vx_uint32, vx_uint8*, vx_uint32)':
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1356:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels2L = _mm_cvtepu8_epi16(pixels2L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1392:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels2L = _mm_cvtepu8_epi16(pixels2L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp: In function 'int HafCpu_Sub_S16_S16U8_Wrap(vx_uint32, vx_uint32, vx_int16*, vx_uint32, vx_int16*, vx_uint32, vx_uint8*, vx_uint32)':
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1542:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels2L = _mm_cvtepu8_epi16(pixels2L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1578:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels2L = _mm_cvtepu8_epi16(pixels2L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp: In function 'int HafCpu_Sub_S16_S16U8_Sat(vx_uint32, vx_uint32, vx_int16*, vx_uint32, vx_int16*, vx_uint32, vx_uint8*, vx_uint32)':
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1728:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels2L = _mm_cvtepu8_epi16(pixels2L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1764:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels2L = _mm_cvtepu8_epi16(pixels2L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp: In function 'int HafCpu_Sub_S16_U8S16_Wrap(vx_uint32, vx_uint32, vx_int16*, vx_uint32, vx_uint8*, vx_uint32, vx_int16*, vx_uint32)':
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1910:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels1L = _mm_cvtepu8_epi16(pixels1L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:1945:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels1L = _mm_cvtepu8_epi16(pixels1L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp: In function 'int HafCpu_Sub_S16_U8S16_Sat(vx_uint32, vx_uint32, vx_int16*, vx_uint32, vx_uint8*, vx_uint32, vx_int16*, vx_uint32)':
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:2094:42: error: '_mm_cvtepu8_epi16' was not declared in this scope
pixels1L = _mm_cvtepu8_epi16(pixels1L);
^
/home/arpu/Work/githubsources/amdovx-core/openvx/ago/ago_haf_cpu_arithmetic.cpp:2130:42: error: '_mm_cvtepu8_epi16' was not declared in this scope

The headers have been updated to OpenVX 1.1 but the function vxuNonLinearFilter is unimplemented.

Hi - I am trying to build with OpenCL and am getting the following errors:
amdovx-core-master/openvx/ago/ago_util_opencl.cpp: In function ‘int agoGpuOclSetKernelArgs(cl_kernel, vx_uint32&, AgoData*, bool, vx_uint32, vx_uint32)’:
amdovx-core-master/openvx/ago/ago_util_opencl.cpp:697:9: error: ‘union cl_uint2’ has no member named ‘s0’
value.s0 = data->u.thr.threshold_lower;
^
amdovx-core-master/openvx/ago/ago_util_opencl.cpp:700:10: error: ‘union cl_uint2’ has no member named ‘s1’
value.s1 = data->u.thr.threshold_upper;
^
amdovx-core-master/openvx/ago/ago_util_opencl.cpp: In function ‘int agoGpuOclDataInputSync(_vx_graph*, cl_kernel, vx_uint32&, AgoData*, vx_uint32, vx_uint32, bool, bool, bool)’:
amdovx-core-master/openvx/ago/ago_util_opencl.cpp:931:9: error: ‘union cl_uint2’ has no member named ‘s0’
value.s0 = data->u.thr.threshold_lower;
^
amdovx-core-master/openvx/ago/ago_util_opencl.cpp:934:10: error: ‘union cl_uint2’ has no member named ‘s1’
value.s1 = data->u.thr.threshold_upper;
^

What am I doing wrong?
Thanks!

I wanna run canny.gdf example and executing runvx.exe file canny.gdf I got:

c:\amdovx-core\build\runvx\Debug>runvx.exe -v file canny.gdf

***** VIDEOINPUT LIBRARY - 0.1995 - TFW07 *****

OK: using AMD OpenVX 0.9.1
OK: enabled graph scheduling in separate threads

include canny.gdf
data input = image:640,480,RGB2
data output = image:640,480,U008
read input raja-koduri-640x480.jpg
view input inputWindow
ERROR: display of image type ( ☺) is not support. Exiting.

How to resolve it?
Thank you.

Hi,

It seems that all VX functions defined in vx_nodes.cpp have the params array ordered as [input1,input2, ... inputK, output]. For example, vxAbsDiffNode() has its params[] being [in1, in2, out]. However, in ago_kernel_api.cpp, functions such as agoKernel_AbsDiff_U8_U8U8() actually assume a reverted order, for instance [oImg, iImg0, iImg1], which seems incorrect to me. Please let me know whether I missed something or there is an issue with this logic.

Thank you.

Best,
Thanh

Is it possible to get it comipled amdovx-core and amdove-modules for arm64 (aarch64) architecture?

I have a severe memory leak at openvx tutorial exercise1 (and solution1 too).
https://github.com/rgiduthuri/openvx_tutorial

linked with OpenVX.dll from this repo (no opencl)

Visual Studio 2013 community. x64. both debug and release.

Call stack of allocation

OpenVX.dll!agoOptimizeDramaAlloc(_vx_graph * agraph) Line 554 C++
OpenVX.dll!agoOptimizeDrama(_vx_graph * agraph) Line 565 C++
OpenVX.dll!agoOptimizeGraph(_vx_graph * agraph) Line 208 C++
OpenVX.dll!vxVerifyGraph(_vx_graph * graph) Line 1657 C++
OpenVX.dll!vxuHarrisCorners(_vx_context * context, _vx_image * input, _vx_scalar * strength_thresh, _vx_scalar * min_distance, _vx_scalar * sensitivity, int gradient_size, int block_size, _vx_array * corners, _vx_scalar * num_corners) Line 873 C++
exercise1.exe!main(int argc, char * * argv) Line 286 C++

No break in openvx\api\vx_api.cpp line 674-675 (function vxQueryImage, switch, VX_IMAGE_ATTRIBUTE_FORMAT), thus VX_IMAGE_ATTRIBUTE_PLANES always works instead.

        case VX_IMAGE_ATTRIBUTE_FORMAT:
            if (size == sizeof(vx_uint32)) {
                *(vx_df_image *)ptr = image->u.img.format;
                status = VX_SUCCESS;
            }
        case VX_IMAGE_ATTRIBUTE_PLANES:

should be

        case VX_IMAGE_ATTRIBUTE_FORMAT:
            if (size == sizeof(vx_uint32)) {
                *(vx_df_image *)ptr = image->u.img.format;
                status = VX_SUCCESS;
            }
            break;  // ***
        case VX_IMAGE_ATTRIBUTE_PLANES:

Im trying to write a dilation node using the opencv extended nodes lib.
For that I need to specify a kernel matrix.

#compute dilation
data kernelMatrix = matrix:UINT8,3,3:INIT,0;1;0;1;1;1;0;1;0
data anchorX = scalar:INT32,1
data anchorY = scalar:INT32,1
data iter = scalar:INT32,3
data border = scalar:INT32,4
node org.opencv.dilate not dilation kernelMatrix anchorX anchorY iter border

Gives me the error:
E:\Projects\SignsFramework\x64\Release>runvx.exe vx_ext_cv.gdf
runvx.exe 0.9.2
OK: using AMD OpenVX 0.9.2
OK: loaded 55 kernels from vx_ext_cv.dll
init done
opengl support available
ERROR: matrix creation failed => -12 (VX_ERROR_INVALID_REFERENCE)

Probably a syntax error: but how can it be fixed?
Cheers,
Bruno

Hi - When I use the vxThresholdNode the stride doesn't seem to be set for the output image. I am unable to use this output image as input to another node because of this. The input image to vxThresholdNode is of VX_DF_IMAGE_U8 type. Should I be doing something else to ensure that I get an output image with the stride set correctly?

Thanks!
Mayura

Error:
amdovx-core/runvx/vxArray.cpp: In member function ‘virtual int CVxParamArray::Initialize(vx_context, vx_graph, const char*)’:
/home/sheldon/repositories/sheldonrobinson/amdovx-core/runvx/vxArray.cpp:67:42: error: ‘strcmp’ was not declared in this scope
if (strcmp(itemType, it->first.c_str()) == 0){

I believe "#include " should be "#include " in runvx/vxUtils.h
--- a/runvx/vxUtils.h
+++ b/runvx/vxUtils.h
@@ -42,7 +42,7 @@ THE SOFTWARE.
#include <inttypes.h>

#include
-#include
+#include
#include
#include
#include

recreated from amdovx-modules

the problem looks like the local_work_size is not set for clEnqueueNDRangeKernel
and the kernel request the size
the docs says
The work-group size to be used for kernel can also be specified in the program source using the attribute((reqd_work_group_size(X, Y, Z)))qualifier. In this case the size of work group specified by local_work_size must match the value specified by the reqd_work_group_size attribute qualifier.

https://www.khronos.org/registry/OpenCL/sdk/1.0/docs/man/xhtml/clEnqueueNDRangeKernel.html

@rgiduthuri so the local_work_size should be the same as the reqd_work_group_size from the kernel right?

Do you have a roadmap to support OpenVX Neural Network Extension here?