libmwemu 0.24.0

use crate::*;

// PSADBW - Sum of Absolute Differences (SSE/SSE2)
//
// Computes the absolute value of the difference of 8 unsigned byte integers,
// then sums the 8 differences to produce an unsigned word integer result.
//
// For 128-bit operands:
// - Low 8 bytes: abs differences summed to bits [15:0], bits [63:16] = 0
// - High 8 bytes: abs differences summed to bits [79:64], bits [127:80] = 0
//
// Formula: For each group of 8 bytes:
//   result = SUM(ABS(src1[i] - src2[i])) for i = 0 to 7
//
// Opcodes (SSE2 - 128-bit XMM):
// 66 0F F6 /r      PSADBW xmm1, xmm2/m128   - Sum absolute differences of bytes

const ALIGNED_ADDR: u64 = 0x3000;
const ALIGNED_ADDR2: u64 = 0x3100;

// ============================================================================
// PSADBW Tests - Sum of Absolute Differences
// ============================================================================

#[test]
fn test_psadbw_all_zeros() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00, // MOVDQA XMM0, [RAX]
        0x66, 0x0f, 0x6f, 0x0b, // MOVDQA XMM1, [RBX]
        0x66, 0x0f, 0xf6, 0xc1, // PSADBW XMM0, XMM1
        0xf4, // HLT
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_identical_values() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50, 0x50]);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_simple_difference() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,
                0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01];
    let src2 = [0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02,
                0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02, 0x02];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_max_difference() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
                0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF];
    let src2 = [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_mixed_differences() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x00, 0xFF, 0x80, 0x40, 0x00, 0x01, 0xFE, 0x7F,
                0xAA, 0x55, 0xFF, 0x00, 0x80, 0x90, 0xA0, 0x10];
    let src2 = [0xFF, 0x00, 0x00, 0xC0, 0xFF, 0xFE, 0x01, 0x80,
                0x55, 0xAA, 0x00, 0xFF, 0x7F, 0x6F, 0x5F, 0xEF];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_incremental_differences() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
                0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F];
    let src2 = [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_alternating_pattern() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0xFF, 0x00, 0xFF, 0x00, 0xFF, 0x00, 0xFF, 0x00,
                0xFF, 0x00, 0xFF, 0x00, 0xFF, 0x00, 0xFF, 0x00];
    let src2 = [0x00, 0xFF, 0x00, 0xFF, 0x00, 0xFF, 0x00, 0xFF,
                0x00, 0xFF, 0x00, 0xFF, 0x00, 0xFF, 0x00, 0xFF];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_one_byte_different() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    let src2 = [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_low_group_only() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    let src2 = [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_high_group_only() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08];
    let src2 = [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_both_groups_different() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,
                0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10, 0x10];
    let src2 = [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_memory_operand() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0xf6, 0x03,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_xmm_self() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0xf6, 0xc0,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80]);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_different_registers() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x10,
        0x66, 0x0f, 0x6f, 0x1b,
        0x66, 0x0f, 0xf6, 0xd3,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_sequential_operations() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0x66, 0x0f, 0xf6, 0xc1,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0, 0xA0]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60, 0x60]);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_halfway_values() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_small_differences() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
                0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F];
    let src2 = [0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18,
                0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F, 0x20];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_large_differences() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_reverse_order_same_result() {
    let mut emu = emu64();
    // abs(a - b) = abs(b - a)
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
                0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F];
    let src2 = [0xFF, 0xFE, 0xFD, 0xFC, 0xFB, 0xFA, 0xF9, 0xF8,
                0xF7, 0xF6, 0xF5, 0xF4, 0xF3, 0xF2, 0xF1, 0xF0];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_varying_differences() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x00, 0x10, 0x20, 0x30, 0x40, 0x50, 0x60, 0x70,
                0x80, 0x90, 0xA0, 0xB0, 0xC0, 0xD0, 0xE0, 0xF0];
    let src2 = [0x01, 0x11, 0x22, 0x33, 0x44, 0x55, 0x66, 0x77,
                0x88, 0x99, 0xAA, 0xBB, 0xCC, 0xDD, 0xEE, 0xFF];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_powers_of_two() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80,
                0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80];
    let src2 = [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_asymmetric_groups() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
                0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF];
    let src2 = [0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01,
                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_boundary_values() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0xFE, 0xFF, 0x01, 0x00, 0x80, 0x7F, 0xC0, 0xBF,
                0x01, 0x02, 0xFE, 0xFD, 0xAA, 0x55, 0x00, 0xFF];
    let src2 = [0x01, 0x00, 0xFE, 0xFF, 0x7F, 0x80, 0x3F, 0x40,
                0xFE, 0xFD, 0x01, 0x02, 0x55, 0xAA, 0xFF, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_zero_result_different_bytes() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x10, 0x20, 0x30, 0x40, 0x50, 0x60, 0x70, 0x80,
                0x90, 0xA0, 0xB0, 0xC0, 0xD0, 0xE0, 0xF0, 0x00];
    let src2 = [0x10, 0x20, 0x30, 0x40, 0x50, 0x60, 0x70, 0x80,
                0x90, 0xA0, 0xB0, 0xC0, 0xD0, 0xE0, 0xF0, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_random_pattern_1() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0x12, 0x34, 0x56, 0x78, 0x9A, 0xBC, 0xDE, 0xF0,
                0x11, 0x22, 0x33, 0x44, 0x55, 0x66, 0x77, 0x88];
    let src2 = [0x21, 0x43, 0x65, 0x87, 0xA9, 0xCB, 0xED, 0x0F,
                0x10, 0x20, 0x30, 0x40, 0x50, 0x60, 0x70, 0x80];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}

#[test]
fn test_psadbw_random_pattern_2() {
    let mut emu = emu64();
    let code = [0x48, 0xb8];
    let mut full_code = code.to_vec();
    full_code.extend_from_slice(&ALIGNED_ADDR.to_le_bytes());
    full_code.extend_from_slice(&[0x48, 0xbb]);
    full_code.extend_from_slice(&ALIGNED_ADDR2.to_le_bytes());
    full_code.extend_from_slice(&[
        0x66, 0x0f, 0x6f, 0x00,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0xf6, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let src1 = [0xAA, 0x55, 0xAA, 0x55, 0xAA, 0x55, 0xAA, 0x55,
                0x33, 0xCC, 0x33, 0xCC, 0x33, 0xCC, 0x33, 0xCC];
    let src2 = [0x55, 0xAA, 0x55, 0xAA, 0x55, 0xAA, 0x55, 0xAA,
                0xCC, 0x33, 0xCC, 0x33, 0xCC, 0x33, 0xCC, 0x33];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &src1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &src2);
    emu.run(None).unwrap();
}