Final changes to

lib/THC/THCTensorRandom.cuh

@@ -289,7 +289,7 @@ sampleMultinomialWithoutReplacement(curandStateMtgp32* state,
     // The warp determines the distribution
     long curDist = curDistBase + threadIdx.y;
 
-    // All threads participate in this
+    // All threads must participate in this
     T r = ScalarConvert<float, T>::to(curand_uniform(&state[blockIdx.x]));
 
     if (threadIdx.x == 0 && curDist < distributions) {

lib/THC/generic/THCTensorRandom.cu

@@ -210,11 +210,11 @@ THC_API void THCTensor_(multinomial)(struct THCState *state,
       
       sampleMultinomialWithReplacement
         <<<grid, block, 0, THCState_getCurrentStream(state)>>>(
-							       THCTensor_(data)(state, uniform_idx),
-							       n_sample,
-							       THCudaLongTensor_data(state, self),
-							       numDist, numCategories,
-							       THCTensor_(data)(state, prefixSum));
+           THCTensor_(data)(state, uniform_idx),
+	   n_sample,
+	   THCudaLongTensor_data(state, self),
+	   numDist, numCategories,
+	   THCTensor_(data)(state, prefixSum));
       
       THCTensor_(free)(state, uniform_idx);
     } else {
@@ -245,13 +245,13 @@ THC_API void THCTensor_(multinomial)(struct THCState *state,
         // recalculate our distribution
         sampleMultinomialWithoutReplacement
           <<<grid, block, 0, THCState_getCurrentStream(state)>>>(
-								 gen->gen_states,
-								 n_sample,	     
-								 sample,	     
-								 THCudaLongTensor_data(state, self),
-								 numDist, numCategories,
-								 THCTensor_(data)(state, origDist),
-								 THCTensor_(data)(state, prefixSum));
+	     gen->gen_states,
+	     n_sample,	     
+	     sample,	     
+	     THCudaLongTensor_data(state, self),
+	     numDist, numCategories,
+	     THCTensor_(data)(state, origDist),
+	     THCTensor_(data)(state, prefixSum));
       }
     }