libmwemu 0.24.1

use crate::*;

// PMAXSB/PMAXSD - Maximum of Packed Unsigned Integers
//
// Performs SIMD compare of packed unsigned integers and returns maximum values.
//
// PMAXSB: Maximum of 8 packed signed byte integers (SSE4.1)
// PMAXSD: Maximum of 4 packed signed dword integers (SSE4.1)
//
// Opcodes:
// 66 0F 38 3A /r      PMAXSB xmm1, xmm2/m128   - Max of packed signed bytes
// 66 0F 38 3B /r      PMAXSD xmm1, xmm2/m128   - Max of packed signed dwords

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

// ============================================================================
// PMAXSB Tests - 8x Signed Byte Maximum
// ============================================================================

#[test]
fn test_pmaxsb_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, 0x38, 0x3c, 0xc1, // PMAXSB 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_pmaxsb_max_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, 0x38, 0x3c, 0xc1,
        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, 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, 0x80, 0x00, 0x80]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsb_mixed_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, 0x38, 0x3c, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let data1 = [0x00, 0x01, 0x00, 0x02, 0x00, 0x03, 0x00, 0x04,
                 0x00, 0x05, 0x00, 0x06, 0x00, 0x07, 0x00, 0x08];
    let data2 = [0x00, 0x08, 0x00, 0x07, 0x00, 0x06, 0x00, 0x05,
                 0x00, 0x04, 0x00, 0x03, 0x00, 0x02, 0x00, 0x01];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &data1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &data2);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsb_large_small() {
    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, 0x38, 0x3c, 0xc1,
        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, &[0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00, 0x01, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsb_xmm4_xmm5() {
    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, 0x20, // MOVDQA XMM4, [RAX]
        0x66, 0x0f, 0x6f, 0x2b, // MOVDQA XMM5, [RBX]
        0x66, 0x0f, 0x38, 0x3c, 0xe5, // PMAXSB XMM4, XMM5
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x11, 0x11, 0x22, 0x22, 0x33, 0x33, 0x44, 0x44, 0x55, 0x55, 0x66, 0x66, 0x77, 0x77, 0x88, 0x88]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0xEE, 0xEE, 0xDD, 0xDD, 0xCC, 0xCC, 0xBB, 0xBB, 0xAA, 0xAA, 0x99, 0x99, 0x88, 0x88, 0x77, 0x77]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsb_from_memory() {
    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, 0x38, 0x3c, 0x03, // PMAXSB XMM0, [RBX]
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x00, 0x10, 0x00, 0x20, 0x00, 0x30, 0x00, 0x40, 0x00, 0x50, 0x00, 0x60, 0x00, 0x70, 0x00, 0x80]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0xFF, 0x0F, 0xFF, 0x1F, 0xFF, 0x2F, 0xFF, 0x3F, 0xFF, 0x4F, 0xFF, 0x5F, 0xFF, 0x6F, 0xFF, 0x7F]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsb_identical() {
    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, 0x38, 0x3c, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00, 0x42, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsb_alternating() {
    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, 0x38, 0x3c, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let data1 = [0x01, 0x00, 0xFF, 0xFF, 0x01, 0x00, 0xFF, 0xFF,
                 0x01, 0x00, 0xFF, 0xFF, 0x01, 0x00, 0xFF, 0xFF];
    let data2 = [0xFF, 0xFF, 0x01, 0x00, 0xFF, 0xFF, 0x01, 0x00,
                 0xFF, 0xFF, 0x01, 0x00, 0xFF, 0xFF, 0x01, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &data1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &data2);
    emu.run(None).unwrap();
}

// ============================================================================
// PMAXSD Tests - 4x Signed Dword Maximum
// ============================================================================

#[test]
fn test_pmaxsd_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,
        0x66, 0x0f, 0x6f, 0x0b,
        0x66, 0x0f, 0x38, 0x3d, 0xc1, // PMAXSD XMM0, XMM1
        0xf4,
    ]);

    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_pmaxsd_max_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, 0x38, 0x3d, 0xc1,
        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, 0x80, 0x00, 0x00, 0x00, 0x80, 0x00, 0x00, 0x00, 0x80, 0x00, 0x00, 0x00, 0x80]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsd_mixed_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, 0x38, 0x3d, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let data1 = [0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x02,
                 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x04];
    let data2 = [0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x03,
                 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x01];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &data1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &data2);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsd_large_small() {
    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, 0x38, 0x3d, 0xc1,
        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, &[0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsd_xmm1_xmm2() {
    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, 0x08, // MOVDQA XMM1, [RAX]
        0x66, 0x0f, 0x6f, 0x13, // MOVDQA XMM2, [RBX]
        0x66, 0x0f, 0x38, 0x3d, 0xca, // PMAXSD XMM1, XMM2
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x11, 0x11, 0x11, 0x11, 0x22, 0x22, 0x22, 0x22, 0x33, 0x33, 0x33, 0x33, 0x44, 0x44, 0x44, 0x44]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0xEE, 0xEE, 0xEE, 0xEE, 0xDD, 0xDD, 0xDD, 0xDD, 0xCC, 0xCC, 0xCC, 0xCC, 0xBB, 0xBB, 0xBB, 0xBB]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsd_from_memory() {
    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, 0x38, 0x3d, 0x03, // PMAXSD XMM0, [RBX]
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x04, 0x00]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0xFF, 0xFF, 0x00, 0x00, 0xFF, 0xFF, 0x01, 0x00, 0xFF, 0xFF, 0x02, 0x00, 0xFF, 0xFF, 0x03, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsd_identical() {
    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, 0x38, 0x3d, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x42, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0x42, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00, 0x42, 0x00, 0x00, 0x00]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsd_alternating() {
    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, 0x38, 0x3d, 0xc1,
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    let data1 = [0x01, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF,
                 0x01, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0xFF, 0xFF];
    let data2 = [0xFF, 0xFF, 0xFF, 0xFF, 0x01, 0x00, 0x00, 0x00,
                 0xFF, 0xFF, 0xFF, 0xFF, 0x01, 0x00, 0x00, 0x00];
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &data1);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &data2);
    emu.run(None).unwrap();
}

// ============================================================================
// Extended Register Tests (XMM8-XMM15)
// ============================================================================

#[test]
fn test_pmaxsb_xmm10_xmm11() {
    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, 0x44, 0x0f, 0x6f, 0x10, // MOVDQA XMM10, [RAX]
        0x66, 0x44, 0x0f, 0x6f, 0x1b, // MOVDQA XMM11, [RBX]
        0x66, 0x45, 0x0f, 0x38, 0x3c, 0xd3, // PMAXSB XMM10, XMM11
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x01, 0x00, 0x02, 0x00, 0x03, 0x00, 0x04, 0x00, 0x05, 0x00, 0x06, 0x00, 0x07, 0x00, 0x08, 0x00]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0xFF, 0xFF, 0xFE, 0xFF, 0xFD, 0xFF, 0xFC, 0xFF, 0xFB, 0xFF, 0xFA, 0xFF, 0xF9, 0xFF, 0xF8, 0xFF]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsd_xmm12_xmm13() {
    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, 0x44, 0x0f, 0x6f, 0x20, // MOVDQA XMM12, [RAX]
        0x66, 0x44, 0x0f, 0x6f, 0x2b, // MOVDQA XMM13, [RBX]
        0x66, 0x45, 0x0f, 0x38, 0x3d, 0xe5, // PMAXSD XMM12, XMM13
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0xFF, 0xFF, 0xFF, 0xFF, 0xFE, 0xFF, 0xFF, 0xFF, 0xFD, 0xFF, 0xFF, 0xFF, 0xFC, 0xFF, 0xFF, 0xFF]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsb_xmm8_from_memory() {
    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, 0x44, 0x0f, 0x6f, 0x00, // MOVDQA XMM8, [RAX]
        0x66, 0x44, 0x0f, 0x38, 0x3c, 0x03, // PMAXSB XMM8, [RBX]
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x00, 0x10, 0x00, 0x20, 0x00, 0x30, 0x00, 0x40, 0x00, 0x50, 0x00, 0x60, 0x00, 0x70, 0x00, 0x80]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0xFF, 0x0F, 0xFF, 0x1F, 0xFF, 0x2F, 0xFF, 0x3F, 0xFF, 0x4F, 0xFF, 0x5F, 0xFF, 0x6F, 0xFF, 0x7F]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsd_xmm14_from_memory() {
    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, 0x44, 0x0f, 0x6f, 0x30, // MOVDQA XMM14, [RAX]
        0x66, 0x44, 0x0f, 0x38, 0x3d, 0x33, // PMAXSD XMM14, [RBX]
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x04, 0x00]);
    emu.maps.write_bytes_slice(ALIGNED_ADDR2, &[0xFF, 0xFF, 0x00, 0x00, 0xFF, 0xFF, 0x01, 0x00, 0xFF, 0xFF, 0x02, 0x00, 0xFF, 0xFF, 0x03, 0x00]);
    emu.run(None).unwrap();
}

// ============================================================================
// Combined/Sequence Tests
// ============================================================================

#[test]
fn test_pmaxsb_pmaxsd_sequence() {
    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, // MOVDQA XMM0, [RAX]
        0x66, 0x0f, 0x6f, 0x08, // MOVDQA XMM1, [RAX]
        0x66, 0x0f, 0x38, 0x3c, 0xc1, // PMAXSB XMM0, XMM1
        0x66, 0x0f, 0x38, 0x3d, 0xc1, // PMAXSD XMM0, XMM1
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x01]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsb_same_register() {
    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, // MOVDQA XMM0, [RAX]
        0x66, 0x0f, 0x38, 0x3c, 0xc0, // PMAXSB XMM0, XMM0
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsd_same_register() {
    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, // MOVDQA XMM0, [RAX]
        0x66, 0x0f, 0x38, 0x3d, 0xc0, // PMAXSD XMM0, XMM0
        0xf4,
    ]);

    emu.load_code_bytes(&full_code);
    emu.maps.write_bytes_slice(ALIGNED_ADDR, &[0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42, 0x42]);
    emu.run(None).unwrap();
}

#[test]
fn test_pmaxsb_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, 0x38, 0x3c, 0xc1,
        0xf4,
    ]);

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

#[test]
fn test_pmaxsd_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, 0x38, 0x3d, 0xc1,
        0xf4,
    ]);

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