【Cherry-pick PR 36511】fix out_of_range bug of multinomial op's cuda k…

…ernel (#36511) (#36808)

Cherry-pick PR #36511

paddle/fluid/operators/multinomial_op.cu

@@ -33,31 +33,33 @@ namespace operators {
 
 template <typename T>
 __global__ void NormalizeProbability(T* norm_probs, const T* in_data,
-                                     T* sum_rows) {
+                                     T* sum_rows, int64_t num_distributions,
+                                     int64_t num_categories) {
   int id = threadIdx.x + blockIdx.x * blockDim.x +
            blockIdx.y * gridDim.x * blockDim.x;
-  PADDLE_ENFORCE(
-      in_data[id] >= 0.0,
-      "The input of multinomial distribution should be >= 0, but got %f.",
-      in_data[id]);
-  PADDLE_ENFORCE(sum_rows[blockIdx.y] > 0.0,
-                 "The sum of one multinomial distribution probability should "
-                 "be > 0, but got %f.",
-                 sum_rows[blockIdx.y]);
-  norm_probs[id] = in_data[id] / sum_rows[blockIdx.y];
+  if (id < num_distributions * num_categories) {
+    PADDLE_ENFORCE(
+        in_data[id] >= 0.0,
+        "The input of multinomial distribution should be >= 0, but got %f.",
+        in_data[id]);
+    int64_t row_id = id / num_categories;
+    PADDLE_ENFORCE(sum_rows[row_id] > 0.0,
+                   "The sum of one multinomial distribution probability should "
+                   "be > 0, but got %f.",
+                   sum_rows[row_id]);
+    norm_probs[id] = in_data[id] / sum_rows[row_id];
+  }
 }
 
 template <typename T>
 __global__ void GetCumulativeProbs(T* norm_probs_data,
                                    int64_t num_distributions,
                                    int64_t num_categories,
                                    T* cumulative_probs) {
-  for (int id = blockIdx.x; id < num_distributions; id += gridDim.x) {
-    thrust::inclusive_scan(thrust::device,
-                           norm_probs_data + id * num_categories,
-                           norm_probs_data + (id + 1) * num_categories,
-                           cumulative_probs + id * num_categories);
-  }
+  int id = blockIdx.x;
+  thrust::inclusive_scan(thrust::device, norm_probs_data + id * num_categories,
+                         norm_probs_data + (id + 1) * num_categories,
+                         cumulative_probs + id * num_categories);
 }
 
 template <typename T>
@@ -108,23 +110,19 @@ __global__ void sampleMultinomialWithReplacement(
   // use binary search to get the selected category sample id.
   // let cumulative_probs[id-1] < rng_data < cumulative_probs[id].
 
-  int idx = threadIdx.x + blockIdx.x * blockDim.x +
-            blockIdx.y * gridDim.x * blockDim.x;
-
   // for every distribution
-  for (int dist = blockIdx.y; dist < num_distributions; dist += gridDim.y) {
-    // for every sample
-    for (int sample = blockIdx.x * blockDim.x + threadIdx.x;
-         sample < num_samples; sample += blockDim.x * gridDim.x) {
-      T rng_number = rng_data[sample + dist * num_samples];
-
-      // Find the bucket that a uniform random number lies in
-      int selected_category = binarySearchFunctor<T>(
-          cumulative_probs + dist * num_categories,
-          norm_probs_data + dist * num_categories, num_categories, rng_number);
-
-      out_data[sample + dist * num_samples] = selected_category;
-    }
+  int dist = blockIdx.y;
+  // for every sample
+  int sample = blockIdx.x * blockDim.x + threadIdx.x;
+  if (sample < num_samples) {
+    T rng_number = rng_data[sample + dist * num_samples];
+
+    // Find the bucket that a uniform random number lies in
+    int selected_category = binarySearchFunctor<T>(
+        cumulative_probs + dist * num_categories,
+        norm_probs_data + dist * num_categories, num_categories, rng_number);
+
+    out_data[sample + dist * num_samples] = selected_category;
   }
 }
 
@@ -215,10 +213,11 @@ class MultinomialOpKernel<platform::CUDADeviceContext, T>
 
     // number of threads in a block is min(num_categories, 512)
     dim3 block_norm(num_categories < 512 ? num_categories : 512);
-    dim3 grid_norm((num_categories - 1) / block_norm.x + 1, num_distributions);
+    dim3 grid_norm((num_distributions * num_categories - 1) / block_norm.x + 1);
     NormalizeProbability<
         T><<<grid_norm, block_norm, 0, ctx.cuda_device_context().stream()>>>(
-        norm_probs_data, in_data, sum_rows_data);
+        norm_probs_data, in_data, sum_rows_data, num_distributions,
+        num_categories);
 
     // Get cumulative probability of each distribution. It's the same function
     // of

python/paddle/fluid/tests/unittests/test_multinomial_op.py

@@ -141,6 +141,14 @@ def test_dygraph3(self):
             "replacement is False. categories can't be sampled repeatedly")
         paddle.enable_static()
 
+    def test_dygraph4(self):
+        paddle.disable_static()
+        logits = -1 * paddle.ones([2800])
+        # Categorical.sample API will call multinomial op with replacement=True
+        cat = paddle.distribution.Categorical(logits.exp())
+        cat.sample([1])
+        paddle.enable_static()
+
     def test_static(self):
         paddle.enable_static()
         startup_program = fluid.Program()