Pattern matching on byte slices produces sub-optimal ASM

I tried this code:

#[unsafe(no_mangle)]
fn test1(data: &[u8]) -> Option<Value> {
    match data {
        [.., 100, 0, 2, 111, 112, 97, 10] => Some(Value::Test0),
        [.., 45, 254, 12, 74, 13, 65, 10] => Some(Value::Test1),
        [.., 75, 11, 64, 12, 128, 65, 11] => Some(Value::Test2),
        [.., 101, 55, 32, 77, 12, 66, 88] => Some(Value::Test3),
        [.., 75, 234, 56, 23, 86, 45, 86] => Some(Value::Test4),
        [.., 51, 35, 75, 14, 65, 45, 155] => Some(Value::Test5),
        [.., 66, 23, 54, 123, 76, 23, 66] => Some(Value::Test6),
        [.., 60, 55, 87, 235, 67, 14, 88] => Some(Value::Test7),
        [.., 69, 22, 4 , 1  , 0, 55, 129] => Some(Value::Test8),
        [.., 99, 23, 77, 13, 66, 86, 193] => Some(Value::Test9),
        _ => None
    }
}

#[unsafe(no_mangle)]
fn test2(data: &[u8]) -> Option<Value> {
    match data {
        data if data.ends_with(&[100, 0, 2, 111, 112, 97, 10]) => Some(Value::Test0),
        data if data.ends_with(&[45, 254, 12, 74, 13, 65, 10]) => Some(Value::Test1),
        data if data.ends_with(&[75, 11, 64, 12, 128, 65, 11]) => Some(Value::Test2),
        data if data.ends_with(&[101, 55, 32, 77, 12, 66, 88]) => Some(Value::Test3),
        data if data.ends_with(&[75, 234, 56, 23, 86, 45, 86]) => Some(Value::Test4),
        data if data.ends_with(&[51, 35, 75, 14, 65, 45, 155]) => Some(Value::Test5),
        data if data.ends_with(&[66, 23, 54, 123, 76, 23, 66]) => Some(Value::Test6),
        data if data.ends_with(&[60, 55, 87, 235, 67, 14, 88]) => Some(Value::Test7),
        data if data.ends_with(&[69, 22, 4 , 1  , 0, 55, 129]) => Some(Value::Test8),
        data if data.ends_with(&[99, 23, 77, 13, 66, 86, 193]) => Some(Value::Test9),
        _ => None
    }
}

enum Value {
    Test0,
    Test1,
    Test2,
    Test3,
    Test4,
    Test5,
    Test6,
    Test7,
    Test8,
    Test9
}

I expected to see this happen: test1 and test2 are equivalent, and should generate same compiled code.

Instead, this happened: the generated code of test1 is inferior to test2

Meta

rustc --version --verbose:

rustc 1.89.0 (29483883e 2025-08-04)
binary: rustc
commit-hash: 29483883eed69d5fb4db01964cdf2af4d86e9cb2
commit-date: 2025-08-04
host: x86_64-unknown-linux-gnu
release: 1.89.0
LLVM version: 20.1.7

Godbolt link: https://rust.godbolt.org/z/76aT93b1z
Related RPLCS discussion: https://discord.com/channels/273534239310479360/1388176726994980925

Start with the supplied test1 and test2 reproducer and compare their generated assembly using the linked Godbolt example. Investigate why the byte-slice pattern match produces inferior code, then verify that equivalent patterns generate comparable optimized assembly while preserving the reported matching behavior.