From 8795174f3a0c005c48ce7d140574bf3c98aada30 Mon Sep 17 00:00:00 2001
From: anna_schlittenhardt <anna.schlittenhardt@stud.uni-due.de>
Date: Sat, 19 Dec 2020 14:06:13 +0100
Subject: [PATCH] added code for 61x67 matrix

---
 src/trans.c | 63 +++++++++++++++++++++++++++++++++++++++++++----------
 1 file changed, 51 insertions(+), 12 deletions(-)

diff --git a/src/trans.c b/src/trans.c
index ecbe6e2..5edb029 100644
--- a/src/trans.c
+++ b/src/trans.c
@@ -27,15 +27,31 @@ int is_transpose(int M, int N, int A[N][M], int B[M][N]);
 char transpose_submit_desc[] = "Transpose submission";
 void transpose_submit(int M, int N, int A[N][M], int B[M][N])
 {   
-    //343 misses (goal: <300; 0 points if >600)
+    
+    int blocksize=0;
+    switch (N)
+    {
+    case 32:
+        blocksize = 8;
+        break;
+    case 64:
+        blocksize = 4;
+        break;
+    case 67:
+        blocksize = 18;
+        break;
+    default:
+        break;
+    }
+
     if(N == 32){
         int n, m, nn, mm, diag_pos, diag_value;
         
-        for (n=0; n<N;n+=8){
-            for (m=0;m<M; m+=8){
+        for (n=0; n<N;n+=blocksize){
+            for (m=0;m<M; m+=blocksize){
 
-                for(nn=n; nn<n+8;nn++){
-                    for(mm=m; mm<m+8;mm++){
+                for(nn=n; nn<n+blocksize;nn++){
+                    for(mm=m; mm<m+blocksize;mm++){
                        
                         if(nn!=mm){
                             B[mm][nn] = A[nn][mm]; 
@@ -55,11 +71,34 @@ void transpose_submit(int M, int N, int A[N][M], int B[M][N])
     if(N == 64){
         int n, m, nn, mm, diag_pos, diag_value;
         
-        for (n=0; n<N;n+=4){
-            for (m=0;m<M; m+=4){
+        for (n=0; n<N;n+=blocksize){
+            for (m=0;m<M; m+=blocksize){
 
-                for(nn=n; nn<n+4;nn++){
-                    for(mm=m; mm<m+4;mm++){
+                for(nn=n; nn<n+blocksize;nn++){
+                    for(mm=m; mm<m+blocksize;mm++){
+                       
+                        if(nn!=mm){
+                            B[mm][nn] = A[nn][mm]; 
+                        }else{
+                            diag_pos = nn;
+						    diag_value = A[nn][mm];
+                        }
+                    }
+                    if(n==m){
+                        B[diag_pos][diag_pos]= diag_value;
+                    }
+                }
+            }
+        }
+    }
+        if(N == 67){
+        int n, m, nn, mm, diag_pos, diag_value;
+        
+        for (n=0; n<N;n+=blocksize){
+            for (m=0;m<M; m+=blocksize){
+
+                for(nn=n; (nn<n+blocksize)&&(nn<N);nn++){
+                    for(mm=m; (mm<m+blocksize)&&(mm<M);mm++){
                        
                         if(nn!=mm){
                             B[mm][nn] = A[nn][mm]; 
@@ -85,7 +124,7 @@ void transpose_submit(int M, int N, int A[N][M], int B[M][N])
 /* 
  * trans - A simple baseline transpose function, not optimized for the cache.
  */
-char trans_desc[] = "Simple row-wise scan transpose";
+/*char trans_desc[] = "Simple row-wise scan transpose";
 void trans(int M, int N, int A[N][M], int B[M][N])
 {
     int i, j, tmp;
@@ -97,7 +136,7 @@ void trans(int M, int N, int A[N][M], int B[M][N])
         }
     }    
 
-}
+}*/
 
 /*
  * registerFunctions - This function registers your transpose
@@ -112,7 +151,7 @@ void registerFunctions()
     registerTransFunction(transpose_submit, transpose_submit_desc); 
 
     /* Register any additional transpose functions */
-    registerTransFunction(trans, trans_desc); 
+    //registerTransFunction(trans, trans_desc); 
 
 }