Merge pull request #2642 from MARD1NO/remove_redundant_sync

use shfl_xor_sync to reduce redundant shfl broadcast

Author: qingqing01

custom_ops/gpu_ops/per_token_quant_fp8.cu

@@ -50,13 +50,11 @@ __global__ void quant_per_token_per_block(const T *input,
                 max_value_thread = max(abs(load_vec_float[vid]), max_value_thread);
             }
             // get max value per warp
-            max_value_thread = max(__shfl_down_sync(0xffffffff, max_value_thread, 16), max_value_thread);
-            max_value_thread = max(__shfl_down_sync(0xffffffff, max_value_thread, 8), max_value_thread);
-            max_value_thread = max(__shfl_down_sync(0xffffffff, max_value_thread, 4), max_value_thread);
-            max_value_thread = max(__shfl_down_sync(0xffffffff, max_value_thread, 2), max_value_thread);
-            max_value_thread = max(__shfl_down_sync(0xffffffff, max_value_thread, 1), max_value_thread);
-            // broadcast max_value
-            max_value_thread = __shfl_sync(0xFFFFFFFF, max_value_thread, 0);
+            max_value_thread = max(__shfl_xor_sync(0xffffffff, max_value_thread, 16), max_value_thread);
+            max_value_thread = max(__shfl_xor_sync(0xffffffff, max_value_thread, 8), max_value_thread);
+            max_value_thread = max(__shfl_xor_sync(0xffffffff, max_value_thread, 4), max_value_thread);
+            max_value_thread = max(__shfl_xor_sync(0xffffffff, max_value_thread, 2), max_value_thread);
+            max_value_thread = max(__shfl_xor_sync(0xffffffff, max_value_thread, 1), max_value_thread);
             max_value_thread = max(max_value_thread, epsilon);
             float scale_to_store = max_value_thread / MAX_VALUE;
             // quant