From 76c190e7cb7e419919194695d9d8040b6a6e6d1d Mon Sep 17 00:00:00 2001
From: L-roro <luis.romerorodriguez@epfl.ch>
Date: Wed, 30 Jul 2025 11:34:44 -0600
Subject: [PATCH 1/2] Unify pragmas for aie2p/mm.cc

---
 aie_kernels/aie2p/mm.cc | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

diff --git a/aie_kernels/aie2p/mm.cc b/aie_kernels/aie2p/mm.cc
index 0295088e639..488b9cb108f 100644
--- a/aie_kernels/aie2p/mm.cc
+++ b/aie_kernels/aie2p/mm.cc
@@ -16,6 +16,8 @@
 #define REL_WRITE 0
 #define REL_READ 1
 
+
+#include "../aie_kernel_utils.h"
 #include <aie_api/aie.hpp>
 
 #include "zero.cc"
@@ -74,14 +76,15 @@ static inline void matmul_vectorized_2x2_mmul(const T_in *__restrict pA,
 
   event0();
 
-  for (unsigned z = 0; z < rowA; z += 2)
-    chess_prepare_for_pipelining chess_loop_range(4, ) {
+  AIE_PREPARE_FOR_PIPELINING
+  AIE_LOOP_MIN_ITERATION_COUNT(4)
+  for (unsigned z = 0; z < rowA; z += 2) {
       T_out *__restrict pC1 = pC + (z * colB) * MMUL::size_C;
       T_out *__restrict pC2 = pC + ((z + 1) * colB) * MMUL::size_C;
 
       for (unsigned j = 0; j < colB; j += 2)
 #ifdef OPT_PERF_ENABLED
-        chess_flatten_loop
+      AIE_LOOP_FLATTEN
 #endif
         {
           const T_in *__restrict pA1 = pA + (z * colA) * MMUL::size_A;
@@ -119,7 +122,7 @@ static inline void matmul_vectorized_2x2_mmul(const T_in *__restrict pA,
 
           for (unsigned i = 0; i < colA; ++i)
 #ifdef OPT_PERF_ENABLED
-            chess_flatten_loop
+      AIE_LOOP_FLATTEN
 #endif
             {
               A0 = aie::load_v<MMUL::size_A>(pA1);

From e30496f1c89f0a6c129bbb055ad2d422ac29a06e Mon Sep 17 00:00:00 2001
From: L-roro <115032777+L-roro@users.noreply.github.com>
Date: Wed, 30 Jul 2025 18:43:23 +0100
Subject: [PATCH 2/2] Remove accidental line indent

---
 aie_kernels/aie2p/mm.cc | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/aie_kernels/aie2p/mm.cc b/aie_kernels/aie2p/mm.cc
index 488b9cb108f..14ab45e934e 100644
--- a/aie_kernels/aie2p/mm.cc
+++ b/aie_kernels/aie2p/mm.cc
@@ -16,7 +16,6 @@
 #define REL_WRITE 0
 #define REL_READ 1
 
-
 #include "../aie_kernel_utils.h"
 #include <aie_api/aie.hpp>
 
@@ -440,4 +439,4 @@ extern "C" {
 combos(matmul_vectorized_c_func) combos(matmul_scalar_c_func)
     combos(zero_vectorized_c_func) combos(zero_scalar_c_func)
 
-} // extern "C"
\ No newline at end of file
+} // extern "C"