| 摘要 |
Method for detecting execution errors in a program, several identical replicas of which are executed in parallel or sequentially, this program comprising in particular instructions consisting of operations involving a simple operator acting on simple operands. Each operation instruction is replaced by several instructions making it possible to perform: - the coding of the operands consisting of input data for the program, by means of a specific coding injection; each coding injection is distinct from the coding injections associated with the other operation instructions within the other replicas, - transformation of the simple operator of this instruction, the transformations corresponding to identical operators within the various replicas being pairwise distinct, - the transcoding of the result obtained by applying the transformed operator to the coded operands, so as to obtain a result in the form of an operand coded by means of the specific coding of the operation instruction intended to use this result, the transcoding being performed by means of a transformation distinct from the transcoding transformations associated with the other operation instructions within the other replicas. The method lastly comprises an operation for decoding the final result of each replica by means of the inverse of a specific decoding injection distinct from the decoding injections of the other replicas. An execution error is flagged if, for certain replicas, the final result cannot be decoded by means of the corresponding inverse injection, and if, among the decoded final results of the other replicas, there does not exist one which is obtained by a predetermined minimum number of the other replicas.
|
