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

Cycles: CUDA faster rendering of small tiles, using multiple samples like OpenCL.

Browse Source 
The work size is still very conservative, and this doesn't help for progressive
refine. For that we will need to render multiple tiles at the same time. But this
should already help for denoising renders that require too much memory with big
tiles, and just generally soften the performance dropoff with small tiles.

Differential Revision: https://developer.blender.org/D2856
This commit is contained in:
Brecht Van Lommel
2017-09-27 01:38:19 +02:00
parent 77f300e2a9
commit 6da6f8d33f
3 changed files with 53 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
intern/cycles/kernel/kernel_passes.h
View File

@@ -16,19 +16,23 @@
CCL_NAMESPACE_BEGIN
#if defined(__SPLIT_KERNEL__) || defined(__KERNEL_CUDA__)
#define __ATOMIC_PASS_WRITE__
#endif
ccl_device_inline void kernel_write_pass_float(ccl_global float *buffer, float value)
{
ccl_global float *buf = buffer;
#if defined(__SPLIT_KERNEL__)
#ifdef __ATOMIC_PASS_WRITE__
atomic_add_and_fetch_float(buf, value);
#else
*buf += value;
#endif /* __SPLIT_KERNEL__ */
#endif
}
ccl_device_inline void kernel_write_pass_float3(ccl_global float *buffer, float3 value)
{
#if defined(__SPLIT_KERNEL__)
#ifdef __ATOMIC_PASS_WRITE__
ccl_global float *buf_x = buffer + 0;
ccl_global float *buf_y = buffer + 1;
ccl_global float *buf_z = buffer + 2;
@@ -39,12 +43,12 @@ ccl_device_inline void kernel_write_pass_float3(ccl_global float *buffer, float3
#else
ccl_global float3 *buf = (ccl_global float3*)buffer;
*buf += value;
#endif /* __SPLIT_KERNEL__ */
#endif
}
ccl_device_inline void kernel_write_pass_float4(ccl_global float *buffer, float4 value)
{
#if defined(__SPLIT_KERNEL__)
#ifdef __ATOMIC_PASS_WRITE__
ccl_global float *buf_x = buffer + 0;
ccl_global float *buf_y = buffer + 1;
ccl_global float *buf_z = buffer + 2;
@@ -57,7 +61,7 @@ ccl_device_inline void kernel_write_pass_float4(ccl_global float *buffer, float4
#else
ccl_global float4 *buf = (ccl_global float4*)buffer;
*buf += value;
#endif /* __SPLIT_KERNEL__ */
#endif
}
#ifdef __DENOISING_FEATURES__
@@ -70,7 +74,7 @@ ccl_device_inline void kernel_write_pass_float_variance(ccl_global float *buffer
kernel_write_pass_float(buffer+1, value*value);
}
# if defined(__SPLIT_KERNEL__)
# ifdef __ATOMIC_PASS_WRITE__
# define kernel_write_pass_float3_unaligned kernel_write_pass_float3
# else
ccl_device_inline void kernel_write_pass_float3_unaligned(ccl_global float *buffer, float3 value)