add typecheck for basic MPI types

tudasc · Jun 16, 2021 · b8295b9 · b8295b9
1 parent 15b4efc
commit b8295b9
Showing 1 changed file with 78 additions and 4 deletions.
diff --git a/lib/mpi_interceptor/InterceptorFunctions.h b/lib/mpi_interceptor/InterceptorFunctions.h
@@ -17,7 +17,9 @@
 #include <sys/resource.h>
 #include <sys/time.h>
 
-int ta_check_buffer(const char* mpi_name, const void* called_from, const void* buf, int mpi_count, int const_adr);
+int ta_mpi_type_to_type_id(MPI_Datatype mpi_type);
+int ta_check_buffer(const char* mpi_name, const void* called_from, const void* buf, MPI_Datatype mpi_type,
+                    int mpi_count, int const_adr);
 void ta_print_loc(const void* call_adr);
 
 typedef struct CallCounter {
@@ -39,12 +41,12 @@ static MPICounter mcounter = {0, 0, 0};
 
 void ta_check_send(const char* name, const void* called_from, const void* sendbuf, int count, MPI_Datatype dtype) {
   ++counter.send;
-  ta_check_buffer(name, called_from, sendbuf, count, 1);
+  ta_check_buffer(name, called_from, sendbuf, dtype, count, 1);
 }
 
 void ta_check_recv(const char* name, const void* called_from, void* recvbuf, int count, MPI_Datatype dtype) {
   ++counter.recv;
-  ta_check_buffer(name, called_from, recvbuf, count, 0);
+  ta_check_buffer(name, called_from, recvbuf, dtype, count, 0);
 }
 
 void ta_check_send_and_recv(const char* name, const void* called_from, const void* sendbuf, int sendcount,
@@ -80,7 +82,68 @@ const char* ta_get_error_message(typeart_status status) {
   }
 }
 
-int ta_check_buffer(const char* mpi_name, const void* called_from, const void* buf, int mpi_count, int const_adr) {
+int ta_mpi_map_int_type(size_t int_sizeof) {
+  if (int_sizeof == 1) {
+    return TA_INT8;
+  } else if (int_sizeof == 2) {
+    return TA_INT16;
+  } else if (int_sizeof == 4) {
+    return TA_INT32;
+  } else if (int_sizeof == 8) {
+    return TA_INT64;
+  } else {
+    fprintf(stderr, "[Error] Unsupperted integer width %lu!", int_sizeof);
+    return TA_UNKNOWN_TYPE;
+  }
+}
+
+// Given an MPI type, returns the corresponding TypeArt type.
+// Note: this function cannot distinguish between TA_FP128 und TA_PPC_TP128,
+// therefore TA_FP128 is always returned in case of an 16 byte floating point
+// MPI type. This should be considered by the caller for performing typechecks.
+int ta_mpi_type_to_type_id(MPI_Datatype mpi_type) {
+  if (mpi_type == MPI_CHAR) {
+    fprintf(stderr, "[Error] MPI_CHAR is currently unsupported!");
+  } else if (mpi_type == MPI_UNSIGNED_CHAR) {
+    fprintf(stderr, "[Error] MPI_UNSIGNED_CHAR is currently unsupported!");
+  } else if (mpi_type == MPI_SIGNED_CHAR) {
+    fprintf(stderr, "[Error] MPI_SIGNED_CHAR is currently unsupported!");
+  } else if (mpi_type == MPI_SHORT) {
+    return ta_mpi_map_int_type(sizeof(short));
+  } else if (mpi_type == MPI_UNSIGNED_SHORT) {
+    fprintf(stderr, "[Error] Unsigned integers are currently not supported!");
+  } else if (mpi_type == MPI_INT) {
+    return ta_mpi_map_int_type(sizeof(int));
+  } else if (mpi_type == MPI_UNSIGNED) {
+    fprintf(stderr, "[Error] Unsigned integers are currently not supported!");
+  } else if (mpi_type == MPI_LONG) {
+    return ta_mpi_map_int_type(sizeof(long int));
+  } else if (mpi_type == MPI_UNSIGNED_LONG) {
+    fprintf(stderr, "[Error] Unsigned integers are currently not supported!");
+  } else if (mpi_type == MPI_LONG_LONG_INT) {
+    return ta_mpi_map_int_type(sizeof(long long int));
+  } else if (mpi_type == MPI_FLOAT) {
+    return TA_FLOAT;
+  } else if (mpi_type == MPI_DOUBLE) {
+    return TA_DOUBLE;
+  } else if (mpi_type == MPI_LONG_DOUBLE) {
+    if (sizeof(long double) == sizeof(double)) {
+      return TA_DOUBLE;
+    } else if (sizeof(long double) == 10) {
+      return TA_X86_FP80;
+    } else if (sizeof(long double) == 16) {
+      return TA_FP128;
+    } else {
+      fprintf(stderr, "[Error] long double has unexpected size %zu!", sizeof(long double));
+    }
+  } else {
+    fprintf(stderr, "[Error] Unsupported MPI datatype found!");
+  }
+  return TA_UNKNOWN_TYPE;
+}
+
+int ta_check_buffer(const char* mpi_name, const void* called_from, const void* buf, MPI_Datatype mpi_type,
+                    int mpi_count, int const_adr) {
   if (mpi_count <= 0) {
     ++mcounter.null_count;
     return 1;
@@ -103,6 +166,17 @@ int ta_check_buffer(const char* mpi_name, const void* called_from, const void* b
     ta_print_loc(called_from);
     return 0;
   }
+  const char* type_name     = typeart_get_type_name(typeId);
+  const int mpi_type_id     = ta_mpi_type_to_type_id(mpi_type);
+  const char* mpi_type_name = typeart_get_type_name(mpi_type_id);
+  printf("R[%d][Info][%d] %s: buffer %p has type %s, MPI type is %s\n", rank, const_adr, mpi_name, buf, type_name,
+         mpi_type_name);
+  if (typeId != mpi_type_id && !(typeId == TA_PPC_FP128 && mpi_type_id == TA_FP128)) {
+    printf("R[%d][Error][%d] %s: buffer %p at loc %p has type %s while the MPI type is %s\n", rank, const_adr, mpi_name,
+           buf, called_from, type_name, mpi_type_name);
+    ta_print_loc(called_from);
+    return -1;
+  }
   // if (mpi_count > count) {
   // TODO: Count check not really sensible without taking the MPI type into account
   //  printf("R[%d][Error][%d] Call '%s' buffer %p too small\n", rank, const_adr, mpi_name, buf);