Kernels fail to compile when using PT with volume, and when using BPT without subsurface.
The body of kernel_lamp_emission should be skipped during regular BPT iteration.
In kernel_holdout_... houldout and kernel_write_passes should be skipped during indirect BPT iteration.
In same kernel shader_bsdf_blur should be skipped during regular BPT iteration.

Apr 14 2017, 1:29 AM

Apr 12 2017

Hristo Gueorguiev (nirved) updated the diff for D2598: Cycles: Split kernel - sort shaders.

Renamed KSES_* defines to more meaningful ones
Moved SHADER_EVAL_SORT_SIZE to kernel_types.h
Moved local variables to __kernel function
Fixed style issues

Apr 12 2017, 10:59 AM

Hristo Gueorguiev (nirved) added inline comments to D2598: Cycles: Split kernel - sort shaders.

Apr 12 2017, 10:45 AM

Apr 4 2017

Hristo Gueorguiev (nirved) created D2598: Cycles: Split kernel - sort shaders.

Apr 4 2017, 12:30 PM

Apr 1 2017

Hristo Gueorguiev (nirved) added inline comments to D2590: Cycles Denoising: Part 1: Preparation changes.

Apr 1 2017, 12:48 PM

Mar 29 2017

Hristo Gueorguiev (nirved) accepted D2586: Cycles: Make all #include statements relative to cycles source directory.

Mar 29 2017, 10:06 AM

Mar 24 2017

Hristo Gueorguiev (nirved) added a comment to T50985: Rendering volume scatter with GPU OpenCL comes to an halt after a few seconds.

Cause is unsafe compiler options. Changing it to a safe one does performance penalty 1%.

P457 Fix for T50985

1	diff --git a/intern/cycles/device/opencl/opencl_base.cpp b/intern/cycles/device/opencl/opencl_base.cpp
2	index 0328dfe..c3a499d 100644
3	--- a/intern/cycles/device/opencl/opencl_base.cpp
4	+++ b/intern/cycles/device/opencl/opencl_base.cpp
5	@@ -612,7 +612,7 @@ void OpenCLDeviceBase::shader(DeviceTask& task)
6
7	string OpenCLDeviceBase::kernel_build_options(const string *debug_src)
8	{
9	- string build_options = "-cl-fast-relaxed-math ";
10	+ string build_options = "-cl-no-signed-zeros -cl-mad-enable ";
11
12	if(platform_name == "NVIDIA CUDA") {
13	build_options += "-D__KERNEL_OPENCL_NVIDIA__ "

Mar 24 2017, 9:31 PM · Cycles, BF Blender

Hristo Gueorguiev (nirved) created P457 Fix for T50985.

Mar 24 2017, 9:28 PM

Mar 23 2017

Hristo Gueorguiev (nirved) abandoned D2314: Cycles: OpenCL split kernel SSS.

Already in master rB57e26627c4.

Mar 23 2017, 12:13 PM

Hristo Gueorguiev (nirved) abandoned D2310: Cycles: OpenCL split kernel volume.

In D2310#58322, @Mai Lavelle (maiself) wrote:

Subsurface scattering works ok when using a single program - all kernels in one file.

Not sure why this would be, do you have an explanation or a fix?

This was due to a combination of division by zero and unsafe OpenCL optimization options.

Mar 23 2017, 12:12 PM

Mar 13 2017

Hristo Gueorguiev (nirved) committed rBf169ff8b8861: Fix T50925: Add AO approximation to split kernel (authored by Hristo Gueorguiev (nirved)).

Fix T50925: Add AO approximation to split kernel

Mar 13 2017, 11:16 AM

Hristo Gueorguiev (nirved) closed T50925: AO bounce simplification diffences between GPU and CPU as "Resolved" by committing rBf169ff8b8861: Fix T50925: Add AO approximation to split kernel.

Mar 13 2017, 11:16 AM · BF Blender

Mar 10 2017

Hristo Gueorguiev (nirved) added 1 commit(s) for D2310: Cycles: OpenCL split kernel volume: rB57e26627c485: Cycles: SSS and Volume rendering in split kernel.

Mar 10 2017, 2:46 PM

Hristo Gueorguiev (nirved) added an edge to rB57e26627c485: Cycles: SSS and Volume rendering in split kernel: D2310: Cycles: OpenCL split kernel volume.

Mar 10 2017, 2:46 PM

Mar 9 2017

Hristo Gueorguiev (nirved) added a comment to T50888: Master Broken, Cuda kernel compat errors compile, OpenCL GPU just sits there on one tile never rendering anything..

@James W E Bird (3dLuver) Please also share the git hash (first 8 letters) of the master which you use.

Mar 9 2017, 6:26 PM · BF Blender

Mar 8 2017

Hristo Gueorguiev (nirved) committed rB0c70c2eb5123: Cycles: add single program debug option for split kernel (authored by Hristo Gueorguiev (nirved)).

Cycles: add single program debug option for split kernel

Mar 8 2017, 6:12 PM

Hristo Gueorguiev (nirved) committed rBa4f8766265bf: Cycles: split kernel_shadow_blocked to AO & DL parts (authored by Hristo Gueorguiev (nirved)).

Cycles: split kernel_shadow_blocked to AO & DL parts

Mar 8 2017, 6:12 PM

Hristo Gueorguiev (nirved) committed rB2e42f8e73f9a: Cycles: Speedup transparent shadows in split kernel (authored by Hristo Gueorguiev (nirved)).

Cycles: Speedup transparent shadows in split kernel

Mar 8 2017, 6:12 PM

Hristo Gueorguiev (nirved) committed rBc33ab5fe6b7c: Cycles: SSS and Volume rendering in split kernel (authored by Hristo Gueorguiev (nirved)).

Cycles: SSS and Volume rendering in split kernel

Mar 8 2017, 6:12 PM

Feb 3 2017

Hristo Gueorguiev (nirved) added a comment to D2310: Cycles: OpenCL split kernel volume.

@Mai Lavelle (maiself)
Volume scattering doesn't work due to NO_TRANSPARENT_SHADOWS.
Subsurface scattering works ok when using a single program - all kernels in one file.

Feb 3 2017, 3:24 PM

Dec 10 2016

Hristo Gueorguiev (nirved) updated the diff for D2314: Cycles: OpenCL split kernel SSS.

Update against latest master.

Dec 10 2016, 10:31 PM

Hristo Gueorguiev (nirved) updated the diff for D2310: Cycles: OpenCL split kernel volume.

Update against latest master.

Dec 10 2016, 10:26 PM

Nov 14 2016

Hristo Gueorguiev (nirved) added a comment to D2348: Cycles: Refactor split kernel and implement for CPU.

Changing the behavior of atomic_inc is misleading, both CUDA & OpenCL return the old value. Further, using atomic_inc_uint32 and atomic_add_uint32 produces less readable code, since we have to subtract what was just added. Maybe there should be indication in atomic_ops.h, i.e. atomic_add_and_get_uint32.

Nov 14 2016, 3:02 PM