PATHGPU with OpenCL and CUDA support

[u3dreal]

Meanwhile on osx HighSierra...

Code: Select all

[LuxRays][28.533] EmbreeAccel build time: 0ms
[LuxRays][28.533] [Device GeForce GT 750M CUDAIntersect] BVH mesh vertices buffer size: 768bytes
[LuxRays][28.533] [Device GeForce GT 750M CUDAIntersect] BVH nodes buffer size: 1536bytes
[LuxRays][28.533] [BVHKernel] Compiler options: -D LUXRAYS_CUDA_DEVICE  -D LUXRAYS_OPENCL_KERNEL -D PARAM_RAY_EPSILON_MIN=1e-05f -D PARAM_RAY_EPSILON_MAX=0.1f
[LuxRays][28.533] [BVHKernel] Compiling kernels
[LuxRays][28.674] [BVHKernel] CUDA program compilation error: 
epsilon_funcs.cl(32): error: identifier "PARAM_RAY_EPSILON_MIN" is undefined

epsilon_funcs.cl(32): error: identifier "PARAM_RAY_EPSILON_MAX" is undefined

ray_funcs.cl(77): error: identifier "VSTORE3F" is undefined

ray_funcs.cl(90): error: identifier "VSTORE3F" is undefined

ray_funcs.cl(102): error: identifier "VSTORE3F" is undefined

bvh.cl(162): error: identifier "VLOAD3F" is undefined

6 errors detected in the compilation of "BVHKernel".

RenderSession starting error: 
BVHKernel CUDA program compilation error
[LuxRays][28.684] WARNING: there is a memory leak in LuxRays HardwareD

[juangea]

The out of core is awesome!

This brings a question, if OOC rendering needs buckets, will Ligth Trace be compatible with buckets?

I really like buckets, specially the multi pass version, not sure if it’s faster, but feels faster, but right now is a bit unstable, it can give uncomplete buckets when it finishes.

Anyways, OOC is awesome!

[lacilaci]

The out of core is awesome!

This brings a question, if OOC rendering needs buckets, will Ligth Trace be compatible with buckets?

I really like buckets, specially the multi pass version, not sure if it’s faster, but feels faster, but right now is a bit unstable, it can give uncomplete buckets when it finishes.

Anyways, OOC is awesome!

I think it's gonna be something like octane, where you don't really have buckets.. but at the same time you kinda do . I guess lighttracing can be sampled separately as it is running on cpu and is just added on top of rendering... ?

[Dade]

I think it's gonna be something like octane, where you don't really have buckets.. but at the same time you kinda do .

Yes, I'm playing with the idea of a special Random/Sobol sampler mode to improve "locality" (i.e. a TILEPATHOCL-like rendering without using TILEPATHOCL)

I guess lighttracing can be sampled separately as it is running on cpu and is just added on top of rendering... ?

Yes, it is the main reason to have a special Random/Sobol sampler out-of-core mode instead of just use TILEPATHOCL (where you can not enable light tracing).

[mischterlampe]

These news are very good news.

I tried to test it by myself but I get an error.

I downloaded the cuda win from today.
I choosed for the render device "OpenCL" is that correct?

When I hit render I get that error

"Hardware device selection string has the wrong length, must be 4 instead of 2"

I'm using a RTX 2070 and a RTX 2060 super.

The console keeps printing this.

Code: Select all

==================================================
[Engine/Viewport] New session
[Exporter] Creating session
[SDL][136.703] Camera type: perspective
[SDL][136.703] Camera position: Point[14.7279, -6.50511, 8.01803]
[SDL][136.703] Camera target: Point[13.9091, -6.14344, 7.57226]
[SDL][136.703] Camera clipping plane disabled
[SDL][136.703] Light definition: 1690932647784
[SDL][136.703] Light definition: __WORLD_BACKGROUND_LIGHT__
[LuxCore][136.719] Configuration:
[LuxCore][136.719]   path.pathdepth.total = 7
[LuxCore][136.719]   path.pathdepth.diffuse = 5
[LuxCore][136.719]   path.pathdepth.glossy = 5
[LuxCore][136.719]   path.pathdepth.specular = 6
[LuxCore][136.719]   path.hybridbackforward.enable = 0
[LuxCore][136.719]   path.hybridbackforward.partition = 0
[LuxCore][136.735]   path.hybridbackforward.glossinessthreshold = 0.048999998718500137
[LuxCore][136.735]   rtpath.resolutionreduction.preview = 4
[LuxCore][136.735]   rtpath.resolutionreduction.preview.step = 2
[LuxCore][136.735]   rtpath.resolutionreduction = 2
[LuxCore][136.735]   opencl.cpu.use = 0
[LuxCore][136.735]   opencl.gpu.use = 1
[LuxCore][136.735]   opencl.devices.select = "11"
[LuxCore][136.735]   opencl.native.threads.count = 0
[LuxCore][136.735]   renderengine.type = "RTPATHOCL"
[LuxCore][136.735]   sampler.type = "TILEPATHSAMPLER"
[LuxCore][136.735]   film.width = 1526
[LuxCore][136.735]   film.height = 1044
[LuxCore][136.735]   film.filter.type = "BLACKMANHARRIS"
[LuxCore][136.735]   film.filter.width = 1.5
[LuxCore][136.735]   lightstrategy.type = "LOG_POWER"
[LuxCore][136.735]   scene.epsilon.min = 9.9999997473787516e-06
[LuxCore][136.735]   scene.epsilon.max = 0.10000000149011612
[LuxCore][136.735]   film.opencl.enable = 1
[LuxCore][136.735]   film.opencl.device = 0
[LuxCore][136.735]   path.forceblackbackground.enable = 0
[LuxCore][136.735]   renderengine.seed = 1
[LuxCore][136.735]   film.outputs.0.type = "RGB_IMAGEPIPELINE"
[LuxCore][136.735]   film.outputs.0.index = 0
[LuxCore][136.735]   film.outputs.0.filename = "RGB_IMAGEPIPELINE_0.png"
[LuxCore][136.735]   film.outputs.1.type = "ALBEDO"
[LuxCore][136.735]   film.outputs.1.filename = "ALBEDO.exr"
[LuxCore][136.735]   film.outputs.2.type = "AVG_SHADING_NORMAL"
[LuxCore][136.735]   film.outputs.2.filename = "AVG_SHADING_NORMAL.exr"
[LuxCore][136.735]   film.imagepipelines.000.0.type = "NOP"
[LuxCore][136.735]   film.imagepipelines.000.1.type = "TONEMAP_LINEAR"
[LuxCore][136.735]   film.imagepipelines.000.1.scale = 1
[LuxCore][136.735]   film.imagepipelines.000.radiancescales.0.enabled = 1
[LuxCore][136.735]   film.imagepipelines.000.radiancescales.0.globalscale = 1
[LuxCore][136.735]   film.imagepipelines.000.radiancescales.0.rgbscale = 1 1 1
[LuxCore][136.750]   batch.haltspp = 0
[LuxCore][136.750]   batch.halttime = 0
[LuxCore][136.750] File Name Resolver Configuration:
Export took 0.0 s
[LuxCore][136.750] Film resolution: 1526x1044
[SDL][136.750] Film output definition: RGB_IMAGEPIPELINE [image.png]
[SDL][136.750] Image pipeline: film.imagepipelines.000
[SDL][136.750] Image pipeline step 0: NOP
[SDL][136.750] Image pipeline step 1: TONEMAP_LINEAR
[SDL][136.750] Film output definition: RGB_IMAGEPIPELINE [RGB_IMAGEPIPELINE_0.png]
[SDL][136.750] Film output definition: ALBEDO [ALBEDO.exr]
[SDL][136.750] Film output definition: AVG_SHADING_NORMAL [AVG_SHADING_NORMAL.exr]
[LuxRays][136.750] OpenCL Platform 0: NVIDIA Corporation
[LuxRays][136.750] CUDA driver version: 11.0
[LuxRays][136.750] CUDA device count: 2
[LuxRays][136.750] Device 0 name: Native
[LuxRays][136.750] Device 0 type: NATIVE_THREAD
[LuxRays][136.750] Device 0 compute units: 1
[LuxRays][136.750] Device 0 preferred float vector width: 4
[LuxRays][136.750] Device 0 max allocable memory: 17592186044415MBytes
[LuxRays][136.750] Device 0 max allocable memory block size: 17592186044415MBytes
[LuxRays][136.750] Device 0 has out of core memory support: 0
[LuxRays][136.750] Device 1 name: GeForce RTX 2070
[LuxRays][136.750] Device 1 type: OPENCL_GPU
[LuxRays][136.766] Device 1 compute units: 36
[LuxRays][136.766] Device 1 preferred float vector width: 1
[LuxRays][136.766] Device 1 max allocable memory: 8192MBytes
[LuxRays][136.766] Device 1 max allocable memory block size: 2048MBytes
[LuxRays][136.766] Device 1 has out of core memory support: 0
[LuxRays][136.766] Device 2 name: GeForce RTX 2060 SUPER
[LuxRays][136.766] Device 2 type: OPENCL_GPU
[LuxRays][136.766] Device 2 compute units: 34
[LuxRays][136.766] Device 2 preferred float vector width: 1
[LuxRays][136.766] Device 2 max allocable memory: 8192MBytes
[LuxRays][136.766] Device 2 max allocable memory block size: 2048MBytes
[LuxRays][136.766] Device 2 has out of core memory support: 0
[LuxRays][136.766] Device 3 name: GeForce RTX 2070
[LuxRays][136.766] Device 3 type: CUDA_GPU
[LuxRays][136.766] Device 3 compute units: 64
[LuxRays][136.766] Device 3 preferred float vector width: 1
[LuxRays][136.766] Device 3 max allocable memory: 8192MBytes
[LuxRays][136.766] Device 3 max allocable memory block size: 17592186044415MBytes
[LuxRays][136.766] Device 3 has out of core memory support: 1
[LuxRays][136.766] Device 4 name: GeForce RTX 2060 SUPER
[LuxRays][136.766] Device 4 type: CUDA_GPU
[LuxRays][136.766] Device 4 compute units: 64
[LuxRays][136.766] Device 4 preferred float vector width: 1
[LuxRays][136.766] Device 4 max allocable memory: 8192MBytes
[LuxRays][136.766] Device 4 max allocable memory block size: 17592186044415MBytes
[LuxRays][136.766] Device 4 has out of core memory support: 1
ERROR: Hardware device selection string has the wrong length, must be 4 instead of 2
Traceback (most recent call last):
  File "C:\Users\philip\AppData\Roaming\Blender Foundation\Blender\2.83\scripts\addons\BlendLuxCore\engine\viewport.py", line 51, in view_update
    engine.session = engine.exporter.create_session(depsgraph, context, engine=engine)
  File "C:\Users\philip\AppData\Roaming\Blender Foundation\Blender\2.83\scripts\addons\BlendLuxCore\export\__init__.py", line 251, in create_session
    return pyluxcore.RenderSession(renderconfig)
RuntimeError: Hardware device selection string has the wrong length, must be 4 instead of 2
[SDL][136.782] Camera type: perspective
[SDL][136.782] Camera position: Point[0, 0, 0]
[SDL][136.782] Camera target: Point[0, 0, 0]
[SDL][136.782] Camera clipping plane disabled
[LuxCore][136.782] Configuration:
[LuxCore][136.782]   renderengine.type = "RTPATHOCL"
[LuxCore][136.782]   sampler.type = "TILEPATHSAMPLER"
[LuxCore][136.782]   scene.epsilon.min = 9.9999997473787516e-06
[LuxCore][136.782]   scene.epsilon.max = 0.10000000149011612
[LuxCore][136.782] File Name Resolver Configuration:
==================================================

any idea?

[acasta69]

"Hardware device selection string has the wrong length, must be 4 instead of 2"

It might be that Cuda support has to be implemented in the BlendLuxCore addon also:
viewtopic.php?f=5&t=1950&p=22196#p22196

[Odilkhan Yakubov]

Wow! Very promising. Is it possible to test out with GTX 1060 6GB? As mentioned about bucket rendering, I'd like bucket rendering too. It is faster than Progressive refine as Cycles named a long time ago.

[B.Y.O.B.]

Wow! Very promising. Is it possible to test out with GTX 1060 6GB? As mentioned about bucket rendering, I'd like bucket rendering too. It is faster than Progressive refine as Cycles named a long time ago.

Just because progressive refine is super slow in Cycles doesn't mean that that's an inherent property of progressive refine compared to buckets.

[juangea]

My feeling is that it's also a bit faster in lux, and the progressive tile seemed even more convinient, but it has some flaws right now, however I don't have a proper correct comparison to be completely sure about tile rendering being actually faster for real.

Anyways I like it more in general, but we don't use it because it has some problems, like leaving unfinished tiles or non properly finished tiles or parts, and of course, not supporting Light Tracing

[Odilkhan Yakubov]

Wow! Very promising. Is it possible to test out with GTX 1060 6GB? As mentioned about bucket rendering, I'd like bucket rendering too. It is faster than Progressive refine as Cycles named a long time ago.

Just because progressive refine is super slow in Cycles doesn't mean that that's an inherent property of progressive refine compared to buckets.

Yes, of course. Just remind it