We find this problem since the version 1.6 always when Aling photos within the depth maps generation process
The same problem with versions 1.6.2, 1.6.3 and 1.6.5
Obviusly that represents a significant lower performance, even duplicating the time processing, with respect to the last version (1.5.5.9097), that was working fine in our Workstation, being that really harmful when processing big projects
Graphic adapter: NVIDIA Quadro K4000
Processors: 2x Intel(R) Xeon(R) CPU E5-2640 v2 @ 2.00GHz, 2001 Mhz, 16 logical processors
Next attached a small section of the window process information:
Initializing...
Found 1 GPUs in 0.369 sec (CUDA: 0.059 sec, OpenCL: 0.31 sec)
Using device: Quadro K4000, 4 compute units, 3072 MB global memory, OpenCL 1.1
driver version: 347.52, platform version: OpenCL 1.1 CUDA 7.0.23
max work group size 1024
max work item sizes [1024, 1024, 64]
max mem alloc size 768 MB
warp size 32
Loading photos...
loaded photos in 2.989 seconds
Generating depth maps...
[CPU] estimating 960x2688x320 disparity using 960x896x8u tiles
Building kernels for Quadro K4000...
Device 1
Program build log:
<kernel>:7528:21: error: invalid operands to binary expression ('__attribute__((address_space(16776963))) float' and '__attribute__((address_space(16776963))) float')
float diff12 = (d1 && d2 ? fabs(d1 - d2) : FLT_MAX);
~~ ^ ~~
<kernel>:7529:21: error: invalid operands to binary expression ('__attribute__((address_space(16776963))) float' and '__attribute__((address_space(16776963))) float')
float diff03 = (d0 && d3 ? fabs(d0 - d3) : FLT_MAX);
~~ ^ ~~
Failed to compile binary dumped to C:/Users/../AppData/Local/Agisoft/Metashape Pro/cache/kernel_cache/program_stereo4318Quadro K4000_FAILED
GPU rectifying failed: ciErrNum: CL_BUILD_PROGRAM_FAILURE (-11) at line 895
using CPU implementation...