fix bug in findLastSet

* findLastSet function: size_t could be  uint, ulong, ulonglong
* add default cuda 7.0 in flags.cmake

Change-Id: Ica9d09520e6a1468e5fc97e2c29c1ea7c73dafbb

cmake/flags.cmake

@@ -81,13 +81,9 @@ function(specify_cuda_arch cuda_version cuda_arch)
             list(APPEND __arch_flags " -gencode arch=compute_${cuda_arch},code=sm_${cuda_arch}")
           endif()
         endforeach()
-    elseif(${cuda_version} VERSION_GREATER "7.0")
-        foreach(capability 52 53)
-          if(${cuda_arch} STREQUAL ${capability})
+    elseif(${cuda_version} VERSION_GREATER "7.0" and ${cuda_arch} STREQUAL "53")
         list(APPEND __arch_flags " -gencode arch=compute_${cuda_arch},code=sm_${cuda_arch}")
     endif()
-        endforeach()
-    endif()
 endfunction()
 
 # Common gpu architectures: Kepler, Maxwell
@@ -95,6 +91,10 @@ foreach(capability 30 35 50)
     list(APPEND __arch_flags " -gencode arch=compute_${capability},code=sm_${capability}")
 endforeach()
 
+if (CUDA_VERSION VERSION_GREATER "7.0")
+      list(APPEND __arch_flags " -gencode arch=compute_52,code=sm_52")
+endif()
+
 # Modern gpu architectures: Pascal
 if (CUDA_VERSION VERSION_GREATER "8.0")
       list(APPEND __arch_flags " -gencode arch=compute_60,code=sm_60")

paddle/utils/Util.h

@@ -72,7 +72,11 @@ namespace paddle {
  * \f]
  */
 inline constexpr size_t findLastSet(size_t x) {
-  return x ? 8 * sizeof(unsigned long) - __builtin_clzl(x) : 0;  // NOLINT
+  return std::is_same<size_t , unsigned int>::value ?
+      (x ? 8 * sizeof(x) - __builtin_clz(x) : 0)
+    : (std::is_same<size_t , unsigned long>::value ? // NOLINT
+      (x ? 8 * sizeof(x) - __builtin_clzl(x) : 0)
+    : (x ? 8 * sizeof(x) - __builtin_clzll(x) : 0));
 }
 
 /**