diff --git a/aie_kernels/relu.cc b/aie_kernels/relu.cc
index 62a53c7479..a2e87cffc4 100644
--- a/aie_kernels/relu.cc
+++ b/aie_kernels/relu.cc
@@ -19,25 +19,21 @@
 
 #include <aie_api/aie.hpp>
 
-
-void relu(bfloat16 * restrict a, bfloat16 * restrict c, const int TILE_SIZE) {
+void relu(bfloat16 *restrict a, bfloat16 *restrict c, const int TILE_SIZE) {
   const int v_factor = 32;
   v32bfloat16 zeroes = broadcast_zero_bfloat16();
 
   event0();
   for (size_t i = 0; i < TILE_SIZE; i += v_factor)
-  chess_prepare_for_pipelining
-  chess_loop_range(32, 32)
-  {
-    v32bfloat16 input = *(v32bfloat16 *)(a + i);
-    v32bfloat16 output = max(input, zeroes);
-    *(v32bfloat16 *)(c + i) = output;
-  }
+    chess_prepare_for_pipelining chess_loop_range(32, 32) {
+      v32bfloat16 input = *(v32bfloat16 *)(a + i);
+      v32bfloat16 output = max(input, zeroes);
+      *(v32bfloat16 *)(c + i) = output;
+    }
   event1();
   return;
 }
 
-
 extern "C" {
 
 void bf16_relu(bfloat16 *a_in, bfloat16 *c_out) { relu(a_in, c_out, 1024); }
diff --git a/programming_examples/basic/relu/aie2.py b/programming_examples/basic/relu/aie2.py
index f9553b57c3..8204706127 100644
--- a/programming_examples/basic/relu/aie2.py
+++ b/programming_examples/basic/relu/aie2.py
@@ -25,7 +25,6 @@ def my_relu():
     enable_tracing = True
     trace_size = 65536
 
-
     # Tile sizes
     n = 1024
     N_div_n = N // n
@@ -50,9 +49,7 @@ def device_body():
 
             # AIE Core Function declarations
 
-            bf16_relu = external_func(
-                "bf16_relu", inputs=[memRef_ty, memRef_ty]
-            )
+            bf16_relu = external_func("bf16_relu", inputs=[memRef_ty, memRef_ty])
 
             # Tile declarations
             ShimTile = tile(0, 0)
@@ -83,12 +80,10 @@ def device_body():
             outC = object_fifo("outC", MemTile, ShimTile, buffer_depth, memRef_C_MT_ty)
             object_fifo_link(outC_fifo_names[0:n_cores], outC)
 
-
             # Set up a circuit-switched flow from core to shim for tracing information
             if enable_tracing:
                 flow(cores[0], WireBundle.Trace, 0, ShimTile, WireBundle.DMA, 1)
 
-
             # Set up compute tiles
             for i in range(n_cores):
                 # Compute tile i
@@ -200,7 +195,6 @@ def sequence(A, C):
                     # Set start BD to our shim bd_Id (13)
                     ipu_write32(column=0, row=0, address=0x1D20C, value=trace_bd_id)
 
-
                 ipu_dma_memcpy_nd(
                     metadata="outC", bd_id=0, mem=C, sizes=[1, 1, 1, C_sz_in_i32s]
                 )
diff --git a/programming_examples/basic/relu/test.cpp b/programming_examples/basic/relu/test.cpp
index 53622644bc..14bb24babe 100644
--- a/programming_examples/basic/relu/test.cpp
+++ b/programming_examples/basic/relu/test.cpp
@@ -55,11 +55,10 @@ void write_out_trace(char *traceOutPtr, size_t trace_size, std::string path) {
   }
 }
 
-
 static inline std::bfloat16_t random_bfloat16_t(float scale, float bias) {
   // Random numbers should NOT be uniformly between 0 and 1, because that
   // would make the matrix product AB always close to 1.
-  return std::bfloat16_t((scale * (float)rand() / (float)(RAND_MAX))-bias);
+  return std::bfloat16_t((scale * (float)rand() / (float)(RAND_MAX)) - bias);
 }
 
 bool nearly_equal(std::bfloat16_t a, std::bfloat16_t b) {
@@ -98,9 +97,9 @@ int main(int argc, const char *argv[]) {
       "trace_sz,t", po::value<int>()->default_value(0),
       "the depth of the trace buffer")(
       "trace_file,f", po::value<std::string>()->default_value("trace.txt"),
-      "the output trace path")(
-      "verbosity,v", po::value<int>()->default_value(0),
-      "the verbosity of the output")(
+      "the output trace path")("verbosity,v",
+                               po::value<int>()->default_value(0),
+                               "the verbosity of the output")(
       "instr,i", po::value<std::string>()->required(),
       "path of file containing userspace instructions to be sent to the LX6");
   po::variables_map vm;
@@ -180,7 +179,6 @@ int main(int argc, const char *argv[]) {
   auto bo_out = xrt::bo(device, real_out_size, XRT_BO_FLAGS_HOST_ONLY,
                         kernel.group_id(3));
 
-
   if (verbosity >= 1)
     std::cout << "Writing data into buffer objects.\n";
 
@@ -254,8 +252,8 @@ int main(int argc, const char *argv[]) {
     npu_time_max = (npu_time > npu_time_max) ? npu_time : npu_time_max;
 
     if (trace_size > 0) {
-      write_out_trace(((char *)bufOut) + (OUT_SIZE*2), trace_size,
-                                     vm["trace_file"].as<std::string>());
+      write_out_trace(((char *)bufOut) + (OUT_SIZE * 2), trace_size,
+                      vm["trace_file"].as<std::string>());
     }
 
     if (VERIFY) {