diff options
Diffstat (limited to 'intern/cycles/kernel/kernels/cuda/kernel_config.h')
| -rw-r--r-- | intern/cycles/kernel/kernels/cuda/kernel_config.h | 9 |
1 files changed, 7 insertions, 2 deletions
diff --git a/intern/cycles/kernel/kernels/cuda/kernel_config.h b/intern/cycles/kernel/kernels/cuda/kernel_config.h index 9fa39dc9ebb..7ae205b7e14 100644 --- a/intern/cycles/kernel/kernels/cuda/kernel_config.h +++ b/intern/cycles/kernel/kernels/cuda/kernel_config.h @@ -81,8 +81,13 @@ # error "Unknown or unsupported CUDA architecture, can't determine launch bounds" #endif -/* compute number of threads per block and minimum blocks per multiprocessor - * given the maximum number of registers per thread */ +/* For split kernel using all registers seems fastest for now, but this + * is unlikely to be optimal once we resolve other bottlenecks. */ + +#define CUDA_KERNEL_SPLIT_MAX_REGISTERS CUDA_THREAD_MAX_REGISTERS + +/* Compute number of threads per block and minimum blocks per multiprocessor + * given the maximum number of registers per thread. */ #define CUDA_LAUNCH_BOUNDS(threads_block_width, thread_num_registers) \ __launch_bounds__( \ |