Compile nbergst / akaze on TX2

[fehimebetul]

when I try to make nbergst/akaze fork on tx2 it gives

`cc: error: unrecognized command line option '-msse2' cc: error: unrecognized command line option '-msse3' cc: error: unrecognized command line option '-msse4' CMake Error at AKAZE_CUDA_generated_cuda_akaze.cu.o.cmake:207 (message): Error generating ~cuda-akaze/build/src/CMakeFiles/AKAZE_CUDA.dir/lib/./AKAZE_CUDA_generated_cuda_akaze.cu.o

src/CMakeFiles/AKAZE_CUDA.dir/build.make:70: recipe for target 'src/CMakeFiles/AKAZE_CUDA.dir/lib/AKAZE_CUDA_generated_cuda_akaze.cu.o' failed make[2]: [src/CMakeFiles/AKAZE_CUDA.dir/lib/AKAZE_CUDA_generated_cuda_akaze.cu.o] Error 1 CMakeFiles/Makefile2:168: recipe for target 'src/CMakeFiles/AKAZE_CUDA.dir/all' failed make[1]: [src/CMakeFiles/AKAZE_CUDA.dir/all] Error 2 Makefile:127: recipe for target 'all' failed make: *** [all] Error 2 ` when I removed the lines including "-msse2 -msse3 -msse4" on ./src/CMakelist.txt, in this time when making library there is no error but in runtime I mean when I try to command

./akaze_match ../../datasets/iguazu/img1.pgm ../../datasets/iguazu/img4.pgm ../../datasets/iguazu/H1to4p

it gives the following

allocating 83.657 Mbytes of gpu memory allocating 83.657 Mbytes of gpu memory safeCall() Runtime API error in file <~/vision-resource/cuda-akaze/src/lib/cudaImage.cu>, line 60 : unspecified launch failure TX2: Cuda 9.0 gcc 5.4.0 opencv 3.4.9

Thanks for your any help ...

[AliEmreKeskin]

Same for me with CUDA 10.2 and OpenCV 3.4.5.

[AliEmreKeskin]

Also this is what cuda-memcheck gives:

$ cuda-memcheck ./akaze_match ../../datasets/iguazu/img1.pgm  ../../datasets/iguazu/img3.pgm  ../../datasets/iguazu/H1to3p --descriptor 5
========= CUDA-MEMCHECK
allocating 83.657 Mbytes of gpu memory
safeCall() Runtime API error in file </.../cuda_akaze/src/lib/cuda_akaze.cu>, line 519 : all CUDA-capable devices are busy or unavailable.
========= Program hit cudaErrorDevicesUnavailable (error 46) due to "all CUDA-capable devices are busy or unavailable" on CUDA API call to cudaMallocPitch. 
=========     Saved host backtrace up to driver entry point at error
=========     Host Frame:/usr/lib/aarch64-linux-gnu/tegra/libcuda.so.1 [0x2fd95c]
=========     Host Frame:./akaze_match [0x5a8f8]
=========
========= ERROR SUMMARY: 1 error

As i can see from jtop my GPU is not busy:

NVIDIA Jetson TX2 - Jetpack 4.4 [L4T 32.4.3]
CPU1 [||||||||||||||                    Schedutil -  28%] 499MHz CPU4 [||||||||||||||                    Schedutil -  28%] 499MHz
CPU2 [  OFF                                                    ] CPU5 [||||||||||                        Schedutil -  20%] 498MHz
CPU3 [  OFF                                                    ] CPU6 [||||||||||||||                    Schedutil -  28%] 498MHz

Mem [||||||||||||||||||||||||||                                                                         2.0G/8.0GB] (lfb 1009x4MB)
Swp [                                                                                                    0.0GB/3.9GB] (cached 0MB)
EMC [|||||||                                                                                                            6%] 204MHz

GPU [                                                                                                                   0%] 114MHz
Dsk [#################################################################################################              21.7GB/27.4GB]
┌────────────────── [info] ──────────────────┬───────── [Sensor] ── [Temp] ─────────┬──────── [Power/mW] ─ [Cur]  [Avr] ──────────┐
│UpT: 0 days 2:37:23                         │          BCPU       40.50C           │         4V0 WIFI     0      19              │
│FAN [                            0%] Ta=  0%│          GPU        38.50C           │         SYS CPU      384    410             │
│Jetson Clocks: inactive                     │          MCPU       40.50C           │         SYS DDR      401    391             │
│NV Power[3]: MAXP CORE ARM                  │          PLL        40.50C           │         SYS GPU      153    153             │
│─────────────── [HW engines] ───────────────│          Tboard     37.00C           │         SYS SOC      384    411             │
│APE: 150MHz                                 │          Tdiode     37.75C           │         ALL          2192   2184            │
│NVENC: [OFF]          NVDEC: [OFF]          │          thermal    39.70C           │                                             │
│NVJPG: [OFF]                                │                                      │                                             │
└────────────────────────────────────────────┴──────────────────────────────────────┴─────────────────────────────────────────────┘
 1ALL  2GPU  3CPU  4MEM  5CTRL  6INFO  Quit                                                                        Raffaello Bonghi

[sthijntja]

Hi, are there any updates on this issue? I am struggling with the same issue.

[pablofdezalc]

Sorry, I am not actively working on this project at the moment. Unfortunately I cannot give you support about nbergst / akaze on TX2 since I did write the CUDA implementation myself. Have you tried contacting Niklas Bergström about this? His contact details are in his repo: https://github.com/nbergst/akaze

Regards, Pablo

[sthijntja]

Sorry, I am not actively working on this project at the moment. Unfortunately I cannot give you support about nbergst / akaze on TX2 since I did write the CUDA implementation myself. Have you tried contacting Niklas Bergström about this? His contact details are in his repo: https://github.com/nbergst/akaze

Regards, Pablo

Thank you for your reply! I have sent an email to Niklas. Hopefully he can help me further.

Regards, Stijn

[Snowpeanut]

Hi, I'm facing the same issue. Is there any update?

[Snowpeanut]

Limiting the max number of features seems to mitigate "the unspecified launch failure". But when I run with cuda-memcheck, I'm also getting "all CUDA-capable devices are busy or unavailable"