[Mlir-commits] [mlir] [MLIR][Presburger] Implement function to evaluate the number of terms in a generating function. (PR #78078)

[Arjun P]

================
@@ -245,3 +245,226 @@ QuasiPolynomial mlir::presburger::detail::getCoefficientInRationalFunction(
   }
   return coefficients[power].simplify();
 }
+
+static std::vector<Fraction> convolution(std::vector<Fraction> a,
+                                         std::vector<Fraction> b) {
+  // The length of the convolution is the maximum of the lengths
+  // of the two sequences. We pad the shorter one with zeroes.
+  unsigned convlen = std::max(a.size(), b.size());
+  for (unsigned k = a.size(); k < convlen; ++k)
+    a.push_back(0);
+  for (unsigned k = b.size(); k < convlen; ++k)
+    b.push_back(0);
+
+  std::vector<Fraction> convolution;
+  convolution.reserve(convlen);
+  convolution.clear();
+  for (unsigned k = 0; k < convlen; ++k) {
+    Fraction sum(0, 1);
+    for (unsigned l = 0; l <= k; ++l)
+      sum = sum + a[l] * b[k - l];
+    convolution.push_back(sum);
+  }
+  return convolution;
+}
+
+/// Substitute x_i = (s+1)^μ_i in one term of a generating function,
+/// returning a quasipolynomial which represents the exponent of the
+/// numerator of the result.
+QuasiPolynomial substituteMuInTerm(unsigned numParams, ParamPoint v,
+                                   std::vector<Point> ds, Point mu) {
+  unsigned numDims = mu.size();
+  // First, the exponent in the numerator becomes
+  // - (μ • u_1) * (floor(first col of v))
+  // - (μ • u_2) * (floor(second col of v)) - ...
+  // - (μ • u_d) * (floor(d'th col of v))
+  // So we store the negation of the  dot produts.
+
+  // We have d terms, each of whose coefficient is the negative dot product,
+  SmallVector<Fraction> coefficients;
+  coefficients.reserve(numDims);
+  for (const Point &d : ds)
+    coefficients.push_back(-dotProduct(mu, d));
+
+  // Then, the affine fn is a single floor expression, given by the
+  // corresponding column of v.
+  ParamPoint vTranspose = v.transpose();
+  std::vector<std::vector<SmallVector<Fraction>>> affine;
+  affine.reserve(numDims);
+  for (unsigned j = 0; j < numDims; ++j)
+    affine.push_back({SmallVector<Fraction>(vTranspose.getRow(j))});
+
+  QuasiPolynomial num(numParams, coefficients, affine);
+  num = num.simplify();
+
+  return num;
+}
+
+/// We have a generating function of the form
+/// f_p(x) = \sum_i sign_i * (x^n_i(p)) / (\prod_j (1 - x^d_{ij})
+///
+/// where sign_i is ±1,
+/// n_i \in Q^p -> Q^d is the sum of the vectors d_{ij}, weighted by the
+/// floors of d affine functions.
+/// d_{ij} \in Q^d are vectors.
+///
+/// We need to find the number of terms of the form x^t in the expansion of
+/// this function.
+/// However, direct substitution causes the denominator to become zero.
+///
+/// We therefore use the following procedure instead:
+/// 1. Substitute x_i = (s+1)^μ_i for some vector μ. This makes the generating
+/// a function of a scalar s.
+/// 2. Write each term in this function as P(s)/Q(s), where P and Q are
+/// polynomials. P has coefficients as quasipolynomials in d parameters, while
+/// Q has coefficients as scalars.
+/// 3. Find the constant term in the expansion of each term P(s)/Q(s). This is
+/// equivalent to substituting s = 0.
+///
+/// Step (1) We need to find a μ_i such that we can substitute x_i =
+/// (s+1)^μ_i. After this substitution, the exponent of (s+1) in the
+/// denominator is (μ_i • d_{ij}) in each term. Clearly, this cannot become
+/// zero. Thus we find a vector μ that is not orthogonal to any of the
----------------
Superty wrote:

Hence we find

https://github.com/llvm/llvm-project/pull/78078