From c66e4b030467ddadac71e5bd4803737cf94c0a07 Mon Sep 17 00:00:00 2001 From: Willem Jan Palenstijn Date: Tue, 22 Dec 2015 14:00:50 +0100 Subject: Reduce FP3D CUDA kernel runtime This reduces the chance of the Windows display driver watchdog triggering, and doesn't seem to hurt performance. --- cuda/3d/cone_fp.cu | 2 +- cuda/3d/par3d_fp.cu | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) (limited to 'cuda') diff --git a/cuda/3d/cone_fp.cu b/cuda/3d/cone_fp.cu index b36d2bc..13b184f 100644 --- a/cuda/3d/cone_fp.cu +++ b/cuda/3d/cone_fp.cu @@ -49,7 +49,7 @@ namespace astraCUDA3d { static const unsigned int g_anglesPerBlock = 4; // thickness of the slices we're splitting the volume up into -static const unsigned int g_blockSlices = 64; +static const unsigned int g_blockSlices = 32; static const unsigned int g_detBlockU = 32; static const unsigned int g_detBlockV = 32; diff --git a/cuda/3d/par3d_fp.cu b/cuda/3d/par3d_fp.cu index b14c494..3ce3d42 100644 --- a/cuda/3d/par3d_fp.cu +++ b/cuda/3d/par3d_fp.cu @@ -49,7 +49,7 @@ namespace astraCUDA3d { static const unsigned int g_anglesPerBlock = 4; // thickness of the slices we're splitting the volume up into -static const unsigned int g_blockSlices = 64; +static const unsigned int g_blockSlices = 32; static const unsigned int g_detBlockU = 32; static const unsigned int g_detBlockV = 32; -- cgit v1.2.3