updates to reliability debug code

This commit was SVN r9390.

ompi/mca/bml/base/bml_base_btl.c

@@ -64,6 +64,8 @@ int mca_bml_base_btl_array_reserve(mca_bml_base_btl_array_t* array, size_t size)
 
 #if OMPI_ENABLE_DEBUG_RELIABILITY
 
+extern double mca_bml_base_error_rate;
+
 struct mca_bml_base_context_t {
     size_t index;
     mca_btl_base_completion_fn_t cbfunc;
@@ -78,7 +80,6 @@ static void mca_bml_base_completion(
                                     int status)
 {
     mca_bml_base_context_t* ctx = (mca_bml_base_context_t*) des->des_cbdata;
-    uint32_t csum;
     /* restore original state */
     ((unsigned char*)des->des_src[0].seg_addr.pval)[ctx->index] ^= ~0;
     des->des_cbdata = ctx->cbdata;
@@ -95,8 +96,8 @@ int mca_bml_base_send(
 { 
     static int count;
     des->des_context = bml_btl;
-    if(count <= 0) {
+    if(count <= 0 && mca_bml_base_error_rate > 0) {
-        count = (int) ((1000.0 * rand())/(RAND_MAX+1.0));
+        count = (int) ((mca_bml_base_error_rate * rand())/(RAND_MAX+1.0));
         if(count % 2) {
             /* local completion - network "drops" packet */
             opal_output(0, "dropping data\n");
@@ -106,14 +107,9 @@ int mca_bml_base_send(
             /* corrupt data */
             mca_bml_base_context_t* ctx = (mca_bml_base_context_t*) 
                 malloc(sizeof(mca_bml_base_context_t));
-            opal_output(0, "corrupting data\n");
             if(NULL != ctx) {
+                opal_output(0, "corrupting data\n");
                 ctx->index = (size_t) ((des->des_src[0].seg_len * rand() * 1.0) / (RAND_MAX + 1.0));
-                if(des->des_src[0].seg_len > 40 ) { 
-                    unsigned char temp;
-                    ctx->index =  40;
-                    temp =  ((unsigned char*)des->des_src[0].seg_addr.pval)[ctx->index];
-                }                
                 ctx->cbfunc = des->des_cbfunc;
                 ctx->cbdata = des->des_cbdata;
                 ((unsigned char*)des->des_src[0].seg_addr.pval)[ctx->index] ^= ~0;

ompi/mca/bml/base/bml_base_open.c

@@ -19,6 +19,7 @@
 
 #include "ompi_config.h"
 #include <stdio.h>
+#include <unistd.h>
 #include "ompi/mca/bml/bml.h" 
 #include "ompi/mca/bml/base/base.h"
 #include "ompi/mca/btl/base/base.h"
@@ -27,6 +28,9 @@
 
 opal_list_t mca_bml_base_components_available;
 
+#if OMPI_ENABLE_DEBUG_RELIABILITY
+double mca_bml_base_error_rate;
+#endif
 
 int mca_bml_base_open( void ) { 
     
@@ -36,6 +40,28 @@ int mca_bml_base_open( void ) {
                                 true)) {  
         return OMPI_ERROR; 
     }
+
+#if OMPI_ENABLE_DEBUG_RELIABILITY
+    do {
+        int param, value;
+
+        mca_base_param_register_int("bml", NULL, "error_rate", "error_rate", 0);
+        param = mca_base_param_find("bml", NULL, "error_rate");
+        mca_base_param_lookup_int(param, &value);
+        mca_bml_base_error_rate = value;
+
+        mca_base_param_register_int("bml", NULL, "srand", "srand", 1);
+        param = mca_base_param_find("bml", NULL, "srand");
+        mca_base_param_lookup_int(param, &value);
+
+        /* seed random number generator */
+        if(value) {
+            struct timeval tv;
+            gettimeofday(&tv, NULL);
+            srand(getpid() * tv.tv_usec);
+        }
+    } while (0);
+#endif
     return mca_btl_base_open(); 
 }
 

ompi/mca/bml/bml.h

@@ -35,7 +35,7 @@
 #include "ompi/class/ompi_free_list.h"
 #include "ompi/mca/pml/pml.h"
 
-#define OMPI_ENABLE_DEBUG_RELIABILITY 0
+#define OMPI_ENABLE_DEBUG_RELIABILITY 1
 
 /*
  * BML types