Cr8-xyz/blender
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Cycles: OpenCL kernel split

Browse Source 
This commit contains all the work related on the AMD megakernel split work
which was mainly done by Varun Sundar, George Kyriazis and Lenny Wang, plus
some help from Sergey Sharybin, Martijn Berger, Thomas Dinges and likely
someone else which we're forgetting to mention.

Currently only AMD cards are enabled for the new split kernel, but it is
possible to force split opencl kernel to be used by setting the following
environment variable: CYCLES_OPENCL_SPLIT_KERNEL_TEST=1.

Not all the features are supported yet, and that being said no motion blur,
camera blur, SSS and volumetrics for now. Also transparent shadows are
disabled on AMD device because of some compiler bug.

This kernel is also only implements regular path tracing and supporting
branched one will take a bit. Branched path tracing is exposed to the
interface still, which is a bit misleading and will be hidden there soon.

More feature will be enabled once they're ported to the split kernel and
tested.

Neither regular CPU nor CUDA has any difference, they're generating the
same exact code, which means no regressions/improvements there.

Based on the research paper:

  https://research.nvidia.com/sites/default/files/publications/laine2013hpg_paper.pdf

Here's the documentation:

  https://docs.google.com/document/d/1LuXW-CV-sVJkQaEGZlMJ86jZ8FmoPfecaMdR-oiWbUY/edit

Design discussion of the patch:

  https://developer.blender.org/T44197

Differential Revision: https://developer.blender.org/D1200
This commit is contained in:
George Kyriazis
2015-05-09 19:34:30 +05:00
committed by Sergey Sharybin
parent f680c1b54a
commit 7f4479da42
57 changed files with 5826 additions and 870 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
intern/cycles/kernel/kernel_passes.h
View File

@@ -19,23 +19,49 @@ CCL_NAMESPACE_BEGIN
ccl_device_inline void kernel_write_pass_float(ccl_global float *buffer, int sample, float value)
{
ccl_global float *buf = buffer;
#if defined(__SPLIT_KERNEL__) && defined(__WORK_STEALING__)
atomic_add_float(buf, value);
#else
*buf = (sample == 0)? value: *buf + value;
#endif // __SPLIT_KERNEL__ && __WORK_STEALING__
}
ccl_device_inline void kernel_write_pass_float3(ccl_global float *buffer, int sample, float3 value)
{
#if defined(__SPLIT_KERNEL__) && defined(__WORK_STEALING__)
ccl_global float *buf_x = buffer + 0;
ccl_global float *buf_y = buffer + 1;
ccl_global float *buf_z = buffer + 2;
atomic_add_float(buf_x, value.x);
atomic_add_float(buf_y, value.y);
atomic_add_float(buf_z, value.z);
#else
ccl_global float3 *buf = (ccl_global float3*)buffer;
*buf = (sample == 0)? value: *buf + value;
#endif // __SPLIT_KERNEL__ && __WORK_STEALING__
}
ccl_device_inline void kernel_write_pass_float4(ccl_global float *buffer, int sample, float4 value)
{
#if defined(__SPLIT_KERNEL__) && defined(__WORK_STEALING__)
ccl_global float *buf_x = buffer + 0;
ccl_global float *buf_y = buffer + 1;
ccl_global float *buf_z = buffer + 2;
ccl_global float *buf_w = buffer + 3;
atomic_add_float(buf_x, value.x);
atomic_add_float(buf_y, value.y);
atomic_add_float(buf_z, value.z);
atomic_add_float(buf_w, value.w);
#else
ccl_global float4 *buf = (ccl_global float4*)buffer;
*buf = (sample == 0)? value: *buf + value;
#endif // __SPLIT_KERNEL__ && __WORK_STEALING__
}
ccl_device_inline void kernel_write_data_passes(KernelGlobals *kg, ccl_global float *buffer, PathRadiance *L,
ShaderData *sd, int sample, PathState *state, float3 throughput)
ShaderData *sd, int sample, ccl_addr_space PathState *state, float3 throughput)
{
#ifdef __PASSES__
int path_flag = state->flag;
@@ -49,18 +75,18 @@ ccl_device_inline void kernel_write_data_passes(KernelGlobals *kg, ccl_global fl
return;
if(!(path_flag & PATH_RAY_SINGLE_PASS_DONE)) {
if(!(sd->flag & SD_TRANSPARENT) ||
if(!(ccl_fetch(sd, flag) & SD_TRANSPARENT) ||
kernel_data.film.pass_alpha_threshold == 0.0f ||
average(shader_bsdf_alpha(kg, sd)) >= kernel_data.film.pass_alpha_threshold)
{
if(sample == 0) {
if(flag & PASS_DEPTH) {
float depth = camera_distance(kg, sd->P);
float depth = camera_distance(kg, ccl_fetch(sd, P));
kernel_write_pass_float(buffer + kernel_data.film.pass_depth, sample, depth);
}
if(flag & PASS_OBJECT_ID) {
float id = object_pass_id(kg, sd->object);
float id = object_pass_id(kg, ccl_fetch(sd, object));
kernel_write_pass_float(buffer + kernel_data.film.pass_object_id, sample, id);
}
if(flag & PASS_MATERIAL_ID) {
@@ -70,7 +96,7 @@ ccl_device_inline void kernel_write_data_passes(KernelGlobals *kg, ccl_global fl
}
if(flag & PASS_NORMAL) {
float3 normal = sd->N;
float3 normal = ccl_fetch(sd, N);
kernel_write_pass_float3(buffer + kernel_data.film.pass_normal, sample, normal);
}
if(flag & PASS_UV) {
@@ -101,7 +127,7 @@ ccl_device_inline void kernel_write_data_passes(KernelGlobals *kg, ccl_global fl
float mist_start = kernel_data.film.mist_start;
float mist_inv_depth = kernel_data.film.mist_inv_depth;
float depth = camera_distance(kg, sd->P);
float depth = camera_distance(kg, ccl_fetch(sd, P));
float mist = saturate((depth - mist_start)*mist_inv_depth);
/* falloff */