Changeset 712 for soft/giet_vm/applications/transpose

Timestamp:

Oct 7, 2015, 11:56:33 AM (9 years ago)

Author:

alain

Message:

Introduce the giet_fbf_size() and giet_fbf_alloc() system calls.

File:

• soft/giet_vm/applications/transpose/transpose.c (modified) (15 diffs)

soft/giet_vm/applications/transpose/transpose.c

@@ -7 +7 @@
 // It can run on a multi-processors, multi-clusters architecture, with one thread
 // per processor, and uses the POSIX threads API. 
+// It does not use the CMA to display the result image.
 //
 // The main() function can be launched on any processor P[x,y,l].
@@ -28 +29 @@
 
 #include "stdio.h"
+#include "stdlib.h"
 #include "user_barrier.h"
 #include "malloc.h"
@@ -36 +38 @@
 #define PROCS_MAX             4                           // max number of procs per cluster
 #define CLUSTER_MAX           (X_MAX * Y_MAX)             // max number of clusters
-#define IMAGE_SIZE            256                         // image size : nlines = npixels
-#define INPUT_FILE_PATH       "/misc/lena_256.raw"        // pathname on virtual disk
-#define OUTPUT_FILE_PATH      "/home/lena_transposed.raw" // pathname on virtual disk
+#define IMAGE_SIZE            256                         // default image size 
+#define INPUT_FILE_PATH       "/misc/lena_256.raw"        // default input file pathname 
+#define OUTPUT_FILE_PATH      "/home/lena_transposed.raw" // default output file pathname 
 
 // macro to use a shared TTY
@@ -65 +67 @@
 
 // checksum variables 
-unsigned check_line_before[IMAGE_SIZE];
-unsigned check_line_after[IMAGE_SIZE];
+unsigned check_line_before[1024];
+unsigned check_line_after[1024];
 
 // lock protecting shared TTY
@@ -73 +75 @@
 // synchronisation barrier (all threads)
 giet_sqt_barrier_t barrier;
+
+// input & output files pathname and size
+char          input_file_name[256];
+char          output_file_name[256];
+unsigned int  image_size;
 
 ////////////////////////////////////////////
@@ -97 +104 @@
     unsigned int nclusters     = x_size * y_size;               // number of clusters
     unsigned int nthreads      = x_size * y_size * nprocs;      // number of threads
-    unsigned int npixels       = IMAGE_SIZE * IMAGE_SIZE;       // pixels per image
+    unsigned int npixels       = image_size * image_size;       // pixels per image
     int          fd_in         = 0;                             // initial file descriptor
     int          fd_out        = 0;                             // output file descriptor
@@ -122 +129 @@
 
         // open input file
-        fd_in = giet_fat_open( INPUT_FILE_PATH , O_RDONLY );  // read_only
+        fd_in = giet_fat_open( input_file_name , O_RDONLY );  // read_only
         if ( fd_in < 0 ) 
         { 
             printf("\n[TRANSPOSE ERROR] Proc [%d,%d,%d] cannot open file %s\n",
-                   x_id , y_id , p_id , INPUT_FILE_PATH );
+                   x_id , y_id , p_id , input_file_name );
             giet_pthread_exit(" open() failure");
         }
@@ -132 +139 @@
         {
             printf("\n[TRANSPOSE] Proc [0,0,0] open file %s / fd = %d\n",
-                   INPUT_FILE_PATH , fd_in );
+                   input_file_name , fd_in );
         }
 
         // open output file
-        fd_out = giet_fat_open( OUTPUT_FILE_PATH , O_CREATE );   // create if required
+        fd_out = giet_fat_open( output_file_name , O_CREATE );   // create if required
         if ( fd_out < 0 ) 
         { 
             printf("\n[TRANSPOSE ERROR] Proc [%d,%d,%d] cannot open file %s\n",
-                            x_id , y_id , p_id , OUTPUT_FILE_PATH );
+                            x_id , y_id , p_id , output_file_name );
             giet_pthread_exit(" open() failure");
         }
@@ -146 +153 @@
         {
             printf("\n[TRANSPOSE] Proc [0,0,0] open file %s / fd = %d\n",
-                   OUTPUT_FILE_PATH , fd_out );
+                   output_file_name , fd_out );
         }
 
@@ -184 +191 @@
 
     // parallel transpose from buf_in to buf_out
-    // each thread makes the transposition for nlt lines (nlt = IMAGE_SIZE/nthreads)
+    // each thread makes the transposition for nlt lines (nlt = image_size/nthreads)
     // from line [thread_id*nlt] to line [(thread_id + 1)*nlt - 1]
     // (p,l) are the absolute pixel coordinates in the source image
@@ -190 +197 @@
     TRSP_START[x_id][y_id][p_id] = giet_proctime();
 
-    unsigned int nlt   = IMAGE_SIZE / nthreads;    // number of lines per thread
-    unsigned int nlc   = IMAGE_SIZE / nclusters;   // number of lines per cluster
+    unsigned int nlt   = image_size / nthreads;    // number of lines per thread
+    unsigned int nlc   = image_size / nclusters;   // number of lines per cluster
 
     unsigned int src_cluster;
@@ -208 +215 @@
      
         // in each iteration we transfer one byte
-        for ( p = 0 ; p < IMAGE_SIZE ; p++ )
+        for ( p = 0 ; p < image_size ; p++ )
         {
             // read one byte from local buf_in
             src_cluster = l / nlc;
-            src_index   = (l % nlc)*IMAGE_SIZE + p;
+            src_index   = (l % nlc)*image_size + p;
             byte        = buf_in[src_cluster][src_index];
 
@@ -220 +227 @@
             // write one byte to remote buf_out
             dst_cluster = p / nlc; 
-            dst_index   = (p % nlc)*IMAGE_SIZE + l;
+            dst_index   = (p % nlc)*image_size + l;
             buf_out[dst_cluster][dst_index] = byte;
         }
@@ -436 +443 @@
                          "[TRANSPOSE ERROR] y_size must be 1,2,4,8,16");
 
-    giet_pthread_assert( (nprocs * x_size * y_size <= IMAGE_SIZE ),
-                         "[TRANSPOSE ERROR] number of threads larger than number of lines");
-
+    // compute number of threads
     unsigned int nthreads = x_size * y_size * nprocs;
 
@@ -445 +450 @@
     lock_init( &tty_lock);
 
-    printf("\n[TRANSPOSE] start at cycle %d on %d cores\n", giet_proctime(), nthreads );
+    // get FBF ownership and FBF size
+    unsigned int   width;
+    unsigned int   height;
+    giet_fbf_alloc();
+    giet_fbf_size( &width , &height );
+
+    // enter interactive part if required
+    printf("\n[TRANSPOSE] start at cycle %d on %d cores / FBF = %d * %d pixels\n",
+           giet_proctime(), nthreads , width , height );
+
+    // input_file_name, output_file_name, and size  acquisition
+    printf("\n[TRANSPOSE] enter path for input file / default is : %s\n> ", INPUT_FILE_PATH );  
+    giet_tty_gets( input_file_name , 256 );
+    printf("\n");
+
+    if ( strcmp( input_file_name , "" ) == 0 ) strcpy( input_file_name , INPUT_FILE_PATH );
+
+    printf("\n[TRANSPOSE] enter path for output file / default is : %s\n> ", OUTPUT_FILE_PATH );  
+    giet_tty_gets( output_file_name , 256 );
+    printf("\n");
+
+    if ( strcmp( output_file_name , "" ) == 0 ) strcpy( output_file_name , OUTPUT_FILE_PATH );
+
+    printf("\n[TRANSPOSE] enter image size / default is : %d\n> ", IMAGE_SIZE );  
+    giet_tty_getw( &image_size );
+    printf("\n");
+    
+    if ( image_size == 0 ) image_size = IMAGE_SIZE;
+
+    printf("\n[TRANSPOSE] input = %s / output = %s / size = %d\n",
+           input_file_name, output_file_name, image_size );
+
+    giet_pthread_assert( (nprocs * x_size * y_size <= image_size ),
+                         "[TRANSPOSE ERROR] number of threads larger than number of lines");
 
     // distributed heap initialisation