+1

examples_ptx/radixSort/Makefile_gpu

@@ -2,7 +2,7 @@ PROG=radixSort
 ISPC_SRC=radixSort.ispc
 
 CU_SRC=radixSort.cu
-NVCC_FLAGS=-Xptxas=-O1
+# NVCC_FLAGS=-Xptxas=-O1
 CXX_SRC=radixSort.cpp  radixSort.cpp
 PTXCC_REGMAX=64
 

examples_ptx/radixSort/radixSort.cu

@@ -109,6 +109,7 @@ void sortPass(
       const int key = mask & ((unsigned int)keys[i] >> bit);
       int scatter;
       /* not a vector friendly loop */
+#pragma unroll 1  /* needed, otherwise compiler unroll and optimizes the result :S */
       for (int iv = 0; iv < programCount; iv++)
         if (programIndex == iv)
           scatter = digitOffsets[key]++;