Merge commit 'e4d15223dcba0fb55e58ae9822f1b97a2f9d97d7'

Author: soumith

torch/lib/THC/THCCachingHostAllocator.h

@@ -22,7 +22,7 @@
 THC_API THAllocator THCCachingHostAllocator;
 
 // Records an event in the specified stream. The allocation 'ptr' will not be
-// re-used until the event has occured.
+// re-used until the event has occurred.
 THC_API cudaError_t THCCachingHostAllocator_recordEvent(void *ptr, THCStream *stream);
 
 // Releases cached pinned memory allocations via cudaHostFree

torch/lib/THC/THCTensorConv.cu

@@ -663,7 +663,7 @@ THC_API void THCudaTensor_conv2DRevgerm(THCState *state, THCudaTensor *output, f
  float *output_data = THCudaTensor_data(state, output);
 
  // kernel is called multiple times
- // (the arbitrary split below is just here to make sure we dont go over 256 threads)
+ // (the arbitrary split below is just here to make sure we don't go over 256 threads)
  for (int sl=0; sl<nbatch; sl+=6) {
  // auto compute nb of blocks and threads
  dim3 blocks(nKernelPlane, nInputPlane);

torch/lib/THC/THCTensorMode.cuh

@@ -245,7 +245,7 @@ __global__ void computeMode(
  // Finally, we need to find the "an" index of the mode in the input Tensor. The API does
  // not constrain which index we pick, so it can be any of the indices that contain the mode.
  // We will do a reduction to find the index. We go back to using the (index, flag) buffer
- // arrangment. First, we mark indices that are equal to the mode, i.e B[i] = true if
+ // arrangement. First, we mark indices that are equal to the mode, i.e B[i] = true if
  // input[i] == mode, and initialize C[i] to be the index
  //
  // Again we reduce 2 elements in the thread's registers prior to the block-wide reduction