griefith/test2
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Cycles: reduce CUDA stack memory access for Maxwell and up, increasing max registers.

Browse Source 
For non-branched path tracing with a GTX 960 and CUDA 7.5, this gives a small reduction
in stack usage but mainly: 8% faster render on BMW, 5% on pabellon, 13% on classroom.
This commit is contained in:
Brecht Van Lommel
2016-06-19 18:06:22 +02:00
parent 9b618c1a5a
commit e26eb9c93b
1 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
intern/cycles/kernel/kernels/cuda/kernel.cu
View File

@@ -77,8 +77,8 @@
# define CUDA_KERNEL_MAX_REGISTERS 63
# define CUDA_KERNEL_BRANCHED_MAX_REGISTERS 63
/* 5.0, 5.2 and 5.3 */
#elif __CUDA_ARCH__ == 500 || __CUDA_ARCH__ == 520 || __CUDA_ARCH__ == 530
/* 5.0, 5.2, 5.3, 6.0, 6.1 */
#elif __CUDA_ARCH__ >= 500
# define CUDA_MULTIPRESSOR_MAX_REGISTERS 65536
# define CUDA_MULTIPROCESSOR_MAX_BLOCKS 32
# define CUDA_BLOCK_MAX_THREADS 1024
@@ -86,7 +86,7 @@
/* tunable parameters */
# define CUDA_THREADS_BLOCK_WIDTH 16
# define CUDA_KERNEL_MAX_REGISTERS 40
# define CUDA_KERNEL_MAX_REGISTERS 48
# define CUDA_KERNEL_BRANCHED_MAX_REGISTERS 63
/* unknown architecture */