From 492f54fb5e776420c43d1a5e86e140b756f07719 Mon Sep 17 00:00:00 2001
From: Kiran Varaganti <Kiran.Varaganti@amd.com>
Date: Sat, 22 May 2021 20:46:57 +0530
Subject: [PATCH] Fix a bug in bench_gemm.c

When op(A) or op(B) = transpose - the leading dimensions of these matrices altered.
Commented out the statements "if(transa) lda = ..." similarly for matrix B and corrected this
mistake in both column and row storages.
Provide a provision to call BLIS interfaces when row-major inputs are used.

Change-Id: Id2041af219a64567471c14190f283274d1df2f7f
---
 bench/bench_gemm.c | 31 +++++++++++++++++++------------
 1 file changed, 19 insertions(+), 12 deletions(-)

diff --git a/bench/bench_gemm.c b/bench/bench_gemm.c
index 3cea4c98e..5390a8d66 100755
--- a/bench/bench_gemm.c
+++ b/bench/bench_gemm.c
@@ -172,16 +172,16 @@ int main( int argc, char** argv )
             // Since this bench app is run on logs generated by AOCL trace logs
             // - we have relaxed the checks on the input parameters.
 
-            // if A is transpose - A(lda x m), lda = max(1,k)
-            // if A is non-transpose - A (lda x k), lda = max(1,m)
-            // if B is transpose - B (ldb x k), ldb = max(1,n)
-            // if B is non-transpose - B (ldb x n), ldb = max(1,k)
-            //    C is ldc x n - ldc = max(1, m)
-            if(transa) lda = k;
+            // if A is transpose - A(lda x m), lda >= max(1,k)
+            // if A is non-transpose - A (lda x k), lda >= max(1,m)
+            // if B is transpose - B (ldb x k), ldb >= max(1,n)
+            // if B is non-transpose - B (ldb x n), ldb >= max(1,k)
+            //    C is ldc x n - ldc >= max(1, m)
+            //if(transa) lda = k; // We will end up overwriting lda
             bli_set_dims_with_trans( transa, m, k, &m_trans, &n_trans);
             bli_obj_create( dt, m_trans, n_trans, 1, lda, &a);
 
-            if(transb) ldb = n;
+            //if(transb) ldb = n; // we will end up overwriting ldb, ldb >= n
             bli_set_dims_with_trans( transb, k, n, &m_trans, &n_trans);
             bli_obj_create( dt, m_trans, n_trans, 1, ldb, &b);
 
@@ -197,11 +197,17 @@ int main( int argc, char** argv )
             // Since this bench app is run on logs generated by AOCL trace logs
             // - we have relaxed the checks on the input parameters.
 
-            if(transa) lda = m;
+	    // if A is transpose - A(k x lda), lda >= max(1,m)
+            // if A is non-transpose - A (m x lda), lda >= max(1,k)
+            // if B is transpose - B (n x ldb), ldb >= max(1,k)
+            // if B is non-transpose - B (k x ldb ), ldb >= max(1,n)
+            //    C is m x ldc - ldc >= max(1, n)
+
+            //if(transa) lda = m; // this will overwrite lda
             bli_set_dims_with_trans(transa, m, k, &m_trans, &n_trans);
             bli_obj_create( dt, m_trans, n_trans, lda, 1, &a);
 
-            if(transb) ldb = k;
+            //if(transb) ldb = k; // this will overwrite ldb
             bli_set_dims_with_trans(transb, k, n, &m_trans, &n_trans);
             bli_obj_create( dt, m_trans, n_trans, ldb, 1, &b);
 
@@ -213,15 +219,16 @@ int main( int argc, char** argv )
             printf("Invalid storage scheme\n");
             continue;
         }
-
-#ifndef CBLAS
+#ifndef BLIS // Incase if we are using blis interface we don't have to check for col-storage.
+     #ifndef CBLAS
         if(bli_obj_col_stride(&c) == 1)
         {
             printf("BLAS APIs doesn't support row-storage: Enable CBLAS\n");
             continue;
         }
+     #endif
 #endif
-
+	
 #ifdef AOCL_MATRIX_INITIALISATION
         bli_randm( &a );
         bli_randm( &b );