Windows Build FAILED

[kintuX]

You have posted an empty file. Anyway I'm fixing it.

Oh, my bad.

Build succeeded.

[kintuX]

I'm getting
--- FATAL ERROR RUNNING CMAKE ---
w/ latest commit (a6b7830db)

Code: Select all

       "C:\luxcorerender\WindowsCompile\Build_CMake\LuxCore\LuxRays.sln" (default target) (1) ->
       "C:\luxcorerender\WindowsCompile\Build_CMake\LuxCore\src\slg\slg-film.vcxproj" (default target) (5) ->
       (ClCompile target) ->
         c:\luxcorerender\luxcore\deps\lut\lut.hpp(203): error C2872: 'enable_if_t': ambiguous symbol (compiling source
        file C:\luxcorerender\LuxCore\src\slg\film\imagepipeline\plugins\colorlut.cpp) [C:\luxcorerender\WindowsCompile
       \Build_CMake\LuxCore\src\slg\slg-film.vcxproj]
         c:\luxcorerender\luxcore\src\slg\film\imagepipeline\plugins\colorlut.cpp(41): error C2872: 'enable_if_t': ambi
       guous symbol [C:\luxcorerender\WindowsCompile\Build_CMake\LuxCore\src\slg\slg-film.vcxproj]

[kintuX]

Confirming the fix w/ commit 6e922d337.

Code: Select all

Build succeeded.
    0 Warning(s)
    0 Error(s)

[kintuX]

Got an error compiling after latest commit (11066e198)

Code: Select all

Build FAILED.

       "C:\luxcorerender\WindowsCompile\Build_CMake\LuxCore\LuxRays.sln" (default target) (1) ->
       "C:\luxcorerender\WindowsCompile\Build_CMake\LuxCore\src\luxcore\pyluxcore.vcxproj.metaproj" (default target) (1
       7) ->
       "C:\luxcorerender\WindowsCompile\Build_CMake\LuxCore\src\luxcore\pyluxcore.vcxproj" (default target) (25) ->
       (Link target) ->
         c:\luxcorerender\windowscompiledeps\include\boost\boost\lexical_cast\detail\converter_numeric.hpp(103): fatal
       error C1001: An internal error has occurred in the compiler. [C:\luxcorerender\WindowsCompile\Build_CMake\LuxCor
       e\src\luxcore\pyluxcore.vcxproj]

    0 Warning(s)
    1 Error(s)

Time Elapsed 00:04:00.63
--- FATAL ERROR RUNNING CMAKE ---

PS
" /CUDA" builds fine

[Dade]

Code: Select all

       error C1001: An internal error has occurred in the compiler. [C:\luxcorerender\WindowsCompile\Build_CMake\LuxCor
       e\src\luxcore\pyluxcore.vcxproj]

Emm, your compiler is broken.

[kintuX]

Code: Select all

       error C1001: An internal error has occurred in the compiler. [C:\luxcorerender\WindowsCompile\Build_CMake\LuxCor
       e\src\luxcore\pyluxcore.vcxproj]

Emm, your compiler is broken.

How?
How could this happen, I haven't changed or touched anything?
All compiling process done as usual, as described...

Code: Select all

cmake-build-x64.bat

fails

Code: Select all

cmake-build-x64.bat /cuda 

succeeds

sorry, for my ignorance - I'm simply confused

PS
and another thing I noticed is, that without "/ cuda" option enabled (19.05.2020 16:12 CET) & previous builds compiled & ran instantly on GPUs
Here are pyluxcore.pyd & luxcoreui.exe for windows to test and see... WTHIGO
https://mega.nz/file/JQhjhYqR#h2vqkpQKQ ... DuC_H_o1Jo
Although it is baffling, it's JUST AMAZING (as it used to be)

BlendLuxCore stats w/ Halt on 10s / resolution: Full HD (1080p)

CPU AMD 1950X (10s Halt = 32 samples, 6.1 M Samples/Sec)

GPUs GTX1070 & Q M5000 (10s Halt = 69 samples, 11.9 M Samples/Sec)

But when I switch to latest "/ cuda" enabled build... i can't find the patience to wait for compiling to end
oh, why


GPUs_CUDA... 5:35.78s later (10s Halt = 76 samples, 13.6 M Samples/Sec)

In such way I prefer "standard" build oppose to "cuda" since my workflow relies rather heavily on fast iterations (look dev) and not on faster final render. So in the end, re-compiling for every "drastic" change takes much more time - even a whole day when all iterations are summed up - combined. CPU comes in handy then, but this experience makes me feel sad, disappointed.

From the looks I guess OCL is way faster to compile (NVidia) so basically, what I wonder now is - Would it be possible to have a "switch" between this options?

[Dade]

I assume you have the OpenCL kernel already cached while the CUDA version is not. Anyway, it is one time compilation in both cases.

[kintuX]

Guess it could be, but...

I Started with a new scene. FBX import to 2.81 (since 2.83 crashed on import - is buggy & release postponed).
Anyways, it compiled almost instantly with "Cycles Settings" and was then moved to Blender 2.83beta where was very responsive (at time I didn't realized I had build "standard" OCL version). So where are caches stored, so I can clear/ delete all & make proper comparison?
And it would be great, if both versions could be built.

Also because now, after both were compiled (using different LuxCore commits) in Blender 2.83 - closed & restarted - CUDA is always slower.
With OCL the samples went up to 75, with 13.6 Samples/Sec & CUDA slowed down to 71samples, 12.7 M Samples/Sec. True, the scene is plain & simple, but nonetheless...

[acasta69]

So where are caches stored, so I can clear/ delete all & make proper comparison?

For LuxCore 2.3 it should be:

Code: Select all

C:\Users\<username>\AppData\Local\Temp\kernel_cache

For v2.4 there are two locations:

Code: Select all

C:\Users\<username>\AppData\Local\Temp\cuda_kernel_cache
C:\Users\<username>\AppData\Local\Temp\ocl_kernel_cache

[Dade]

@Acasta69, I'm starting to think that compilation problem on Azure is due to the use of a Linux OpenCL SDK on Windows. The error seems to suggest that and the compilation on Windows works fine here.
You should check the OpenCL headers, the definition of "cl_ulong" type should be like:

Code: Select all

typedef unsigned __int64        cl_ulong;

I think I'm using this OpenCL SDK: https://github.com/GPUOpen-LibrariesAnd ... K/releases