Segmentation fault in Adams2019

[jxl1080]

Adams2019 issues segmentation fault with Halide generator class below:

#include "Halide.h"
#include <stdio.h>
using namespace Halide;
class mMatmul_matmul_out1_fcn_halide_generator : public Halide::Generator <mMatmul_matmul_out1_fcn_halide_generator> {

public:
    Input<Buffer<uint8_t>> A1_in{"A1_in", 2};
    Input<Buffer<uint8_t>> B1_in{"B1_in", 2};
    Output<Buffer<uint16_t>> matmul_out1_fcn{"matmul_out1_fcn", 2};

    void generate() {
        RDom r(0, 81);
        matmul_out1_func(d1, d2) = sum(cast<uint16_t>(A1_in(d1, r))*cast<uint16_t>(B1_in(r, d2)));
        matmul_out1_fcn(d1, d2) = matmul_out1_func(d1, d2);
    }

    void schedule() {
    // Schedule is determined by autoscheduler. Need to set estimate on buffer
        if(using_autoscheduler()) {
            A1_in.dim(1).set_estimate(0, 81);
            A1_in.dim(0).set_estimate(0, 81);
            B1_in.dim(1).set_estimate(0, 81);
            B1_in.dim(0).set_estimate(0, 81);
            matmul_out1_fcn.set_estimate(d1, 0, 81).set_estimate(d2, 0, 81);
        }  else {
            // Default schedule
        }
    }

private:
    Var d1{"d1"};
    Var d2{"d2"};
    Func matmul_out1_func{"matmul_out1_func"};

};
HALIDE_REGISTER_GENERATOR(mMatmul_matmul_out1_fcn_halide_generator, mMatmul_matmul_out1_fcn_halide_gen)

I used binary 'Halide-18.0.0-x86-64-linux-b741d9c5b2705c45fd7c8462eb0b4bfa35a3e6d8.tar.gz' downloaded from: https://buildbot.halide-lang.org/

My command for compiling the Halide Genertor Class is:
$ g++ mMatmul_matmul_out1_fcn_halide.cpp -std=c++17 ....../Halide-18.0.0-x86-64-linux/share/Halide/tools/GenGen.cpp -L ....../Halide-18.0.0-x86-64-linux/lib -lHalide -I ....../Halide-18.0.0-x86-64-linux/include -o mMatmul_matmul_out1_fcn_halide

My command for running generator with Adams2019 is:
$ export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:....../Halide-18.0.0-x86-64-linux/lib
$ ./mMatmul_matmul_out1_fcn_halide -f myPipeline -g mMatmul_matmul_out1_fcn_halide_gen -e h,o target=hexagon-32-noos-hvx-no_runtime autoscheduler.parallelism=2 autoscheduler=Adams2019 -p ....../Halide-18.0.0-x86-64-linux/lib/libautoschedule_adams2019.so -o ./

I have the following observations:

1. If I make the RDom length and the length used in set_estimate less than or equal to 80, the segmentation fault will disappear.
2. If I change the output buffer type from uint16_t to uint8_t and remove the two 'cast' in generate function, the segmentation fault will disappear.