Increased block sizes on Fermi

c4008bb2 · Peter Eastman · 2bce659f · c4008bb2 · c4008bb2
Commit c4008bb2 authored Aug 11, 2010 by Peter Eastman
2 changed files
--- a/plugins/amoeba/platforms/cuda/src/kernels/kCalculateAmoebaCudaMutualInducedAndGkFields.cu
+++ b/plugins/amoeba/platforms/cuda/src/kernels/kCalculateAmoebaCudaMutualInducedAndGkFields.cu
@@ -578,7 +578,7 @@ static void cudaComputeAmoebaMutualInducedAndGkFieldMatrixMultiply( amoebaGpuCon
    if( threadsPerBlock == 0 ){
        unsigned int maxThreads;
        if (gpu->sm_version >= SM_20)
-            maxThreads = 256;
+            maxThreads = 384;
        else if (gpu->sm_version >= SM_12)
            maxThreads = 128;
        else

--- a/plugins/amoeba/platforms/cuda/src/kernels/kCalculateAmoebaCudaMutualInducedAndGkFields.h
+++ b/plugins/amoeba/platforms/cuda/src/kernels/kCalculateAmoebaCudaMutualInducedAndGkFields.h
@@ -28,7 +28,7 @@

 __global__
 #if (__CUDA_ARCH__ >= 200)
-__launch_bounds__(256, 1)
+__launch_bounds__(384, 1)
 #elif (__CUDA_ARCH__ >= 130)
 __launch_bounds__(128, 1)
 #else