fix lars (#36431)

paddle/fluid/operators/optimizers/lars_momentum_op.cu

@@ -165,8 +165,10 @@ __global__ void L2NormKernel(
   int tid = threadIdx.x + blockDim.x * blockIdx.x;
   int grid_stride = LARS_BLOCK_SIZE * gridDim.x;
   const MT rescale_pow = rescale_grad * rescale_grad;
+  if (threadIdx.x == 0) {
     s_buffer[0] = static_cast<MT>(0);
     s_buffer[1] = static_cast<MT>(0);
+  }
   MT p_tmp = static_cast<MT>(0);
   MT g_tmp = static_cast<MT>(0);
 
@@ -175,8 +177,12 @@ __global__ void L2NormKernel(
       p_tmp = static_cast<MT>(p_data[tid]);
       g_tmp = static_cast<MT>(g_data[tid]);
     }
-    s_buffer[0] += math::blockReduceSum<MT>(p_tmp * p_tmp, FINAL_MASK);
-    s_buffer[1] += math::blockReduceSum<MT>(g_tmp * g_tmp, FINAL_MASK);
+    MT tmp0 = math::blockReduceSum<MT>(p_tmp * p_tmp, FINAL_MASK);
+    MT tmp1 = math::blockReduceSum<MT>(g_tmp * g_tmp, FINAL_MASK);
+    if (threadIdx.x == 0) {
+      s_buffer[0] += tmp0;
+      s_buffer[1] += tmp1;
+    }
   } else {
     /* Avoid occupy too much temp buffer. Slice the whole data into 2 parts,
     the front of data whose quantity is excatly multiple of grid-thread
@@ -185,8 +191,12 @@ __global__ void L2NormKernel(
       p_tmp = static_cast<MT>(p_data[tid]);
       g_tmp = static_cast<MT>(g_data[tid]);
       tid += grid_stride;
-      s_buffer[0] += math::blockReduceSum<MT>(p_tmp * p_tmp, FINAL_MASK);
-      s_buffer[1] += math::blockReduceSum<MT>(g_tmp * g_tmp, FINAL_MASK);
+      MT tmp0 = math::blockReduceSum<MT>(p_tmp * p_tmp, FINAL_MASK);
+      MT tmp1 = math::blockReduceSum<MT>(g_tmp * g_tmp, FINAL_MASK);
+      if (threadIdx.x == 0) {
+        s_buffer[0] += tmp0;
+        s_buffer[1] += tmp1;
+      }
       __syncthreads();
     }
     MT p_val = 0;
@@ -195,8 +205,12 @@ __global__ void L2NormKernel(
       p_val = static_cast<MT>(p_data[tid]);
       g_val = static_cast<MT>(g_data[tid]);
     }
-    s_buffer[0] += math::blockReduceSum<MT>(p_val * p_val, FINAL_MASK);
-    s_buffer[1] += math::blockReduceSum<MT>(g_val * g_val, FINAL_MASK);
+    MT tmp0 = math::blockReduceSum<MT>(p_val * p_val, FINAL_MASK);
+    MT tmp1 = math::blockReduceSum<MT>(g_val * g_val, FINAL_MASK);
+    if (threadIdx.x == 0) {
+      s_buffer[0] += tmp0;
+      s_buffer[1] += tmp1;
+    }
   }
   __syncthreads();
 
@@ -208,8 +222,15 @@ __global__ void L2NormKernel(
   cg->sync();  // Grid sync for writring partial result to gloabl memory
   MT p_part_sum = threadIdx.x < gridDim.x ? p_buffer[threadIdx.x] : 0;
   MT g_part_sum = threadIdx.x < gridDim.x ? g_buffer[threadIdx.x] : 0;
-  *p_n = Sqrt(math::blockReduceSum<MT>(p_part_sum, FINAL_MASK));
-  *g_n = Sqrt(rescale_pow * math::blockReduceSum<MT>(g_part_sum, FINAL_MASK));
+  MT tmp0 = math::blockReduceSum<MT>(p_part_sum, FINAL_MASK);
+  MT tmp1 = math::blockReduceSum<MT>(g_part_sum, FINAL_MASK);
+  if (threadIdx.x == 0) {
+    s_buffer[0] = tmp0;
+    s_buffer[1] = tmp1;
+  }
+  __syncthreads();
+  *p_n = Sqrt(s_buffer[0]);
+  *g_n = Sqrt(rescale_pow * s_buffer[1]);
 #endif
 }