From: yoni206 Date: Thu, 10 Sep 2020 01:32:55 +0000 (-0700) Subject: bv2int: improvement in lazy failures (#5020) X-Git-Tag: cvc5-1.0.0~2879 X-Git-Url: https://git.libre-soc.org/?a=commitdiff_plain;h=98f808f01735c6cebfbf945fcb072da3ba6cff6c;p=cvc5.git bv2int: improvement in lazy failures (#5020) Previously, in case we encountered a term we cannot translate to integers (e.g. a bit-vector array), we would fail. This PR improves that behavior by keeping the original term, and potentially wrapping it BV_TO_NAT and INT_TO_BV operators. A test which includes a bit-vector array is included, which was not supported before. In addition, the treatment of uninterpreted functions is refactored and documented better. --- diff --git a/src/preprocessing/passes/bv_to_int.cpp b/src/preprocessing/passes/bv_to_int.cpp index c0e22a0ee..e91e375c9 100644 --- a/src/preprocessing/passes/bv_to_int.cpp +++ b/src/preprocessing/passes/bv_to_int.cpp @@ -688,11 +688,9 @@ Node BVToInt::bvToInt(Node n) /* * We replace all BV-sorts of the domain with INT * If the range is a BV sort, we replace it with INT - * We cache both the term itself (e.g., f(a)) and the function - * symbol f. */ - //Construct the function itself + // construct the new function symbol. Node bvUF = current.getOperator(); Node intUF; TypeNode tn = current.getOperator().getType(); @@ -703,12 +701,14 @@ Node BVToInt::bvToInt(Node n) } else { + // The function symbol has not been converted yet vector bvDomain = tn.getArgTypes(); vector intDomain; /** * if the original range is a bit-vector sort, * the new range should be an integer sort. * Otherwise, we keep the original range. + * Similarly for the domains. */ TypeNode intRange = bvRange.isBitVector() ? d_nm->integerType() : bvRange; @@ -725,8 +725,12 @@ Node BVToInt::bvToInt(Node n) "bv2int function"); // Insert the function symbol itself to the cache d_bvToIntCache[bvUF] = intUF; + + // introduce a `define-fun` in the smt-engine to keep + // the correspondence between the original + // function symbol and the new one. + defineBVUFAsIntUF(bvUF); } - if (childrenTypesChanged(current) && options::ufHo()) { /** * higher order logic allows comparing between functions * The current translation does not support this, @@ -734,48 +738,53 @@ Node BVToInt::bvToInt(Node n) * of the bounds that were relevant for the original * bit-vectors. */ - throw TypeCheckingException( - current.toExpr(), - string("Cannot translate to Int: ") + current.toString()); + if (childrenTypesChanged(current) && options::ufHo()) + { + throw TypeCheckingException( + current.toExpr(), + string("Cannot translate to Int: ") + current.toString()); } - else { - translated_children.insert(translated_children.begin(), intUF); - // Insert the term to the cache - d_bvToIntCache[current] = - d_nm->mkNode(kind::APPLY_UF, translated_children); - /** - * Add range constraints if necessary. - * If the original range was a BV sort, the current application of - * the function Must be within the range determined by the - * bitwidth. - */ - if (bvRange.isBitVector()) - { - d_rangeAssertions.insert( - mkRangeConstraint(d_bvToIntCache[current], - current.getType().getBitVectorSize())); - } + + // Now that the translated function symbol was + // created, we translate the applicatio and add to the cache. + // Additionally, we add + // range constraints induced by the original BV width of the + // the functions range (codomain).. + translated_children.insert(translated_children.begin(), intUF); + // Insert the translated application term to the cache + d_bvToIntCache[current] = + d_nm->mkNode(kind::APPLY_UF, translated_children); + // Add range constraints if necessary. + // If the original range was a BV sort, the current application of + // the function Must be within the range determined by the + // bitwidth. + if (bvRange.isBitVector()) + { + Node m = d_bvToIntCache[current]; + d_rangeAssertions.insert( + mkRangeConstraint(d_bvToIntCache[current], + current.getType().getBitVectorSize())); } - break; + break; } default: { - if (childrenTypesChanged(current)) { - /** - * This is "failing on demand": - * We throw an exception if we encounter a case - * that we do not know how to translate, - * only if we actually need to construct a new - * node for such a case. - */ - throw TypeCheckingException( - current.toExpr(), - string("Cannot translate to Int: ") + current.toString()); + // In the default case, we have reached an operator that we do not + // translate directly to integers. The children whose types have + // changed from bv to int should be adjusted back to bv and then + // this term is reconstructed. + TypeNode resultingType; + if (current.getType().isBitVector()) + { + resultingType = d_nm->integerType(); } - else { - d_bvToIntCache[current] = - d_nm->mkNode(oldKind, translated_children); + else + { + resultingType = current.getType(); } + Node reconstruction = + reconstructNode(current, resultingType, translated_children); + d_bvToIntCache[current] = reconstruction; break; } } @@ -787,12 +796,60 @@ Node BVToInt::bvToInt(Node n) return d_bvToIntCache[n]; } -bool BVToInt::childrenTypesChanged(Node n) { +void BVToInt::defineBVUFAsIntUF(Node bvUF) +{ + // This function should only be called after translating + // the function symbol to a new function symbol + // with the right domain and range. + Assert(d_bvToIntCache.find(bvUF) != d_bvToIntCache.end()); + + // get domain and range of the original function + TypeNode tn = bvUF.getType(); + vector bvDomain = tn.getArgTypes(); + TypeNode bvRange = tn.getRangeType(); + + // get the translated function symbol + Node intUF = d_bvToIntCache[bvUF]; + + // create a symbolic application to be used in define-fun + + // symbolic arguments of original function + vector args; + // children of the new symbolic application + vector achildren; + achildren.push_back(intUF); + int i = 0; + for (TypeNode d : bvDomain) + { + // Each bit-vector argument is casted to a natural number + // Other arguments are left intact. + Node fresh_bound_var = d_nm->mkBoundVar(d); + args.push_back(fresh_bound_var.toExpr()); + Node castedArg = args[i]; + if (d.isBitVector()) + { + castedArg = castToType(castedArg, d_nm->integerType()); + } + achildren.push_back(castedArg); + i++; + } + Node intApplication = d_nm->mkNode(kind::APPLY_UF, achildren); + // If the range is BV, the application needs to be casted back. + intApplication = castToType(intApplication, bvRange); + // add the function definition to the smt engine. + smt::currentSmtEngine()->defineFunction( + bvUF.toExpr(), args, intApplication.toExpr(), true); +} + +bool BVToInt::childrenTypesChanged(Node n) +{ bool result = false; - for (Node child : n) { + for (const Node& child : n) + { TypeNode originalType = child.getType(); TypeNode newType = d_bvToIntCache[child].get().getType(); - if (! newType.isSubtypeOf(originalType)) { + if (!newType.isSubtypeOf(originalType)) + { result = true; break; } @@ -800,6 +857,54 @@ bool BVToInt::childrenTypesChanged(Node n) { return result; } +Node BVToInt::castToType(Node n, TypeNode tn) +{ + // If there is no reason to cast, return the + // original node. + if (n.getType().isSubtypeOf(tn)) + { + return n; + } + // We only case int to bv or vice verse. + Assert((n.getType().isBitVector() && tn.isInteger()) + || (n.getType().isInteger() && tn.isBitVector())); + if (n.getType().isInteger()) + { + Assert(tn.isBitVector()); + unsigned bvsize = tn.getBitVectorSize(); + Node intToBVOp = d_nm->mkConst(IntToBitVector(bvsize)); + return d_nm->mkNode(intToBVOp, n); + } + Assert(n.getType().isBitVector()); + Assert(tn.isInteger()); + return d_nm->mkNode(kind::BITVECTOR_TO_NAT, n); +} + +Node BVToInt::reconstructNode(Node originalNode, + TypeNode resultType, + const vector& translated_children) +{ + // first, we adjust the children of the node as needed. + // re-construct the term with the adjusted children. + kind::Kind_t oldKind = originalNode.getKind(); + NodeBuilder<> builder(oldKind); + if (originalNode.getMetaKind() == kind::metakind::PARAMETERIZED) + { + builder << originalNode.getOperator(); + } + for (uint i = 0; i < originalNode.getNumChildren(); i++) + { + Node originalChild = originalNode[i]; + Node translatedChild = translated_children[i]; + Node adjustedChild = castToType(translatedChild, originalChild.getType()); + builder << adjustedChild; + } + Node reconstruction = builder.constructNode(); + // cast to tn in case the reconstruction is a bit-vector. + reconstruction = castToType(reconstruction, resultType); + return reconstruction; +} + BVToInt::BVToInt(PreprocessingPassContext* preprocContext) : PreprocessingPass(preprocContext, "bv-to-int"), d_binarizeCache(preprocContext->getUserContext()), diff --git a/src/preprocessing/passes/bv_to_int.h b/src/preprocessing/passes/bv_to_int.h index 2777a36a6..bec8a9a21 100644 --- a/src/preprocessing/passes/bv_to_int.h +++ b/src/preprocessing/passes/bv_to_int.h @@ -253,6 +253,47 @@ class BVToInt : public PreprocessingPass */ void addFinalizeRangeAssertions(AssertionPipeline* assertionsToPreprocess); + /** + * Reconstructs a node whose main operator cannot be + * translated to integers. + * Reconstruction is done by casting to integers/bit-vectors + * as needed. + * For example, if node is (select A x) where A + * is a bit-vector array, we do not change A to be + * an integer array, even though x was translated + * to integers. + * In this case we cast x to (bv2nat x) during + * the reconstruction. + * + * @param originalNode the node that we are reconstructing + * @param resultType the desired type for the reconstruction + * @param translated_children the children of originalNode + * after their translation to integers. + * @return A node with originalNode's operator that has type resultType. + */ + Node reconstructNode(Node originalNode, + TypeNode resultType, + const vector& translated_children); + + /** + * A useful utility function. + * if n is an integer and tn is bit-vector, + * applies the IntToBitVector operator on n. + * if n is a vit-vector and tn is integer, + * applies BitVector_TO_NAT operator. + * Otherwise, keeps n intact. + */ + Node castToType(Node n, TypeNode tn); + + /** + * When a UF f is translated to a UF g, + * we add a define-fun command to the smt-engine + * to relate between f and g. + * This is useful, for example, when asking + * for a model-value of a term that includes the + * original UF f. + */ + void defineBVUFAsIntUF(Node bvUF); /** * Caches for the different functions */ diff --git a/test/regress/CMakeLists.txt b/test/regress/CMakeLists.txt index ea012a279..ed07c7474 100644 --- a/test/regress/CMakeLists.txt +++ b/test/regress/CMakeLists.txt @@ -2083,6 +2083,7 @@ set(regress_2_tests regress2/bv_to_int_inc1.smt2 regress2/bv_to_int_mask_array_1.smt2 regress2/bv_to_int_mask_array_2.smt2 + regress2/bv_to_int_mask_array_3.smt2 regress2/bv_to_int_shifts.smt2 regress2/error0.smt2 regress2/error1.smtv1.smt2 diff --git a/test/regress/regress2/bv_to_int_mask_array_3.smt2 b/test/regress/regress2/bv_to_int_mask_array_3.smt2 new file mode 100644 index 000000000..7eef1ad4e --- /dev/null +++ b/test/regress/regress2/bv_to_int_mask_array_3.smt2 @@ -0,0 +1,8 @@ +; COMMAND-LINE: --solve-bv-as-int=sum --bvand-integer-granularity=1 --no-check-models --no-check-unsat-cores +; EXPECT: sat +(set-logic ALL) +(declare-fun A () (Array (_ BitVec 4) (_ BitVec 4))) +(declare-fun x () (_ BitVec 4)) +(declare-fun y () (_ BitVec 4)) +(assert (distinct (select A x) (select A y))) +(check-sat)