strict_float causes computational complexity issue with bounds_of_expr_in_scope

[abadams]

This repro:

#include "Halide.h"

using namespace Halide;
using namespace Halide::Internal;

int main(int argc, char **argv) {

    Param<float> p, p_min, p_max;
    Scope<Interval> scope;
    scope.push(p.name(), Interval{p_min, p_max});

    std::cout << "ops \t non-strict \t strict\n";

    for (int limit = 1; limit < 100; limit++) {
        Expr e = p;
        int ops = 0;
        for (int i = 0; i < limit; i++) {
            e = e * p + (i + 1);
            ops += 2;
        }

        auto t1 = std::chrono::high_resolution_clock::now();
        bounds_of_expr_in_scope(e, scope);
        auto t2 = std::chrono::high_resolution_clock::now();

        e = strictify_float(e);

        auto t3 = std::chrono::high_resolution_clock::now();
        bounds_of_expr_in_scope(e, scope);
        auto t4 = std::chrono::high_resolution_clock::now();

        // Print us per multiply op in the expr
        std::cout << ops << " \t "
                  << std::chrono::duration_cast<std::chrono::microseconds>(t2 - t1).count() << " \t "
                  << std::chrono::duration_cast<std::chrono::microseconds>(t4 - t3).count() << "\n";
    }

    return 0;
}

Produces this output:

ops 	 non-strict 	 strict
2 	 62 	 22
4 	 9 	 60
6 	 13 	 231
8 	 17 	 944
10 	 21 	 3709
12 	 13 	 7372
14 	 15 	 29597
16 	 17 	 118067
18 	 19 	 473886
20 	 21 	 1895721

Seems to be exponential.