AVX2 Code crashes when instructions are combined

Question

I was trying to shift zeros from Unsigned to signed range i.e,. [0,255] -> [127, -128]

To perform this below operation

__m256i shifted = _mm256_xor_si256(a, _mm256_set1_epi8(-128));
__m256i maskCount = _mm256_cmpgt_epi8(shifted, zerosShifted);

when I wrote the code like this, It Runs

    __m256i temp1 = _mm256_setzero_si256();
    __m256i temp2 = _mm256_set1_epi8(-128);
    __m256i zerosShifted = _mm256_xor_si256(temp1,temp2);

but when I wrote like this, It crashes saying Illegal Instruction

__m256i zerosShifted = _m256_xor_si256(_mm256_setzero_si256(),_mm256_set1_epi8(-128));

PS: I know there is no need to shift zeros to -128, I can set -128 directly to the register but just to know the root cause analysis.

The complete AVX2 code looks like this:

__m256i *pIn0, *pOut0;
            __m256i a;
            __m256i zerosShifted = _mm256_xor_si256(_mm256_setzero_si256(), _mm256_set1_epi8(-128));
            __m256i ones = _mm256_set1_epi8(1);
            __m256i zeros = _mm256_setzero_si256();
            for (int x = 1; x < height - 1 ; x++)
            {
                int y = 1;
                for (; y < width - 32; y+=32)
                {
                    a = _MM256_LOAD_SI256((__m256i*)&boundary_image_array[x*width + y]);

                    __m256i shifted = _mm256_xor_si256(a, _mm256_set1_epi8(-128));
                    __m256i maskCount = _mm256_cmpgt_epi8(shifted, zerosShifted);

                    maskCount = _mm256_and_si256(maskCount, ones);

                    __m256i imageAgg = _mm256_blendv_epi8(a, zeros, maskCount);

                    __m256i pixelCount = _mm256_sad_epu8(maskCount, zeros);
                    __m256i imgAggSum = _mm256_sad_epu8(imageAgg, zeros);

                    boundary_pixel_count_r += _mm256_extract_epi64(pixelCount, 0) + _mm256_extract_epi64(pixelCount, 1)
                        + _mm256_extract_epi64(pixelCount, 2) + _mm256_extract_epi64(pixelCount, 3);

                    boundary_grad_image_agg += _mm256_extract_epi64(imgAggSum, 0) + _mm256_extract_epi64(imgAggSum, 1)
                        + _mm256_extract_epi64(imgAggSum, 2) + _mm256_extract_epi64(imgAggSum, 3);
                }
                for (; y < width - 1 ; y++)
                {
                    //! Determine the boundary stats:
                    if (boundary_image_array[x*width + y] > 0)
                    {
                        boundary_pixel_count_r++;
                        boundary_grad_image_agg += boundary_image_array[x*width + y];
                    }
                }
            }