more clean up

Author: alingTT

ttnn/cpp/ttnn/operations/transformer/sdpa_decode/device/kernels/dataflow/writer_decode_all.cpp

@@ -232,8 +232,7 @@ void kernel_main() {
          cur_head < cur_head_group * num_heads_per_core + num_heads_per_core;
          ++cur_head) {
         // Tree reduction: receive from children at each round
-        // IMPORTANT: We process round-by-round to synchronize with compute kernel
-        // Each round, we wait for one child (if any), read its data, and push to CBs
+        // Each round, we wait for one child (if any), read remote_sum, remote_max, remote_output, and push to CBs
         // The compute kernel processes each child's data before we move to the next round
         // Only receive from children that actually have data
         if (actual_num_children > 0) {
@@ -262,7 +261,6 @@ void kernel_main() {
                     constexpr uint32_t ml_read_size = PNHt * tile_bytes_intermed;
 
                     // Calculate offset based on round (child writes at round offset)
-                    // Note: Use get_write_ptr to match what the sender uses when writing to this buffer
                     uint32_t block_offset = round * (out_chunk_tiles + 2 * PNHt) * tile_bytes_intermed;
                     uint64_t intermed_l1_read_addr = get_noc_addr(get_read_ptr(cb_intermed_out)) + block_offset;