Skip to content

Commit

Permalink
big tech doesnt want you to know this
Browse files Browse the repository at this point in the history
  • Loading branch information
@alion02
alion02 committed Dec 6, 2024
1 parent 0a6b27c commit a89346d
Showing 1 changed file with 180 additions and 88 deletions.
268 changes: 180 additions & 88 deletions src/day5.rs
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,97 +8,189 @@ use super::*;
static mut MATRIX: [u8x32; 45] = [u8x32::from_array([0; 32]); 45];

#[target_feature(enable = "avx2,bmi1,bmi2,cmpxchg16b,lzcnt,movbe,popcnt")]
#[allow(unreachable_code)]
unsafe fn inner1(s: &[u8]) -> u32 {
let matrix = &mut MATRIX;
{
MATRIX[0] = Simd::splat(0);
MATRIX[1] = Simd::splat(0);
MATRIX[2] = Simd::splat(0);
MATRIX[3] = Simd::splat(0);
MATRIX[4] = Simd::splat(0);
MATRIX[5] = Simd::splat(0);
MATRIX[6] = Simd::splat(0);
MATRIX[7] = Simd::splat(0);
MATRIX[8] = Simd::splat(0);
MATRIX[9] = Simd::splat(0);
MATRIX[10] = Simd::splat(0);
MATRIX[11] = Simd::splat(0);
MATRIX[12] = Simd::splat(0);
MATRIX[13] = Simd::splat(0);
MATRIX[14] = Simd::splat(0);
MATRIX[15] = Simd::splat(0);
MATRIX[16] = Simd::splat(0);
MATRIX[17] = Simd::splat(0);
MATRIX[18] = Simd::splat(0);
MATRIX[19] = Simd::splat(0);
MATRIX[20] = Simd::splat(0);
MATRIX[21] = Simd::splat(0);
MATRIX[22] = Simd::splat(0);
MATRIX[23] = Simd::splat(0);
MATRIX[24] = Simd::splat(0);
MATRIX[25] = Simd::splat(0);
MATRIX[26] = Simd::splat(0);
MATRIX[27] = Simd::splat(0);
MATRIX[28] = Simd::splat(0);
MATRIX[29] = Simd::splat(0);
MATRIX[30] = Simd::splat(0);
MATRIX[31] = Simd::splat(0);
MATRIX[32] = Simd::splat(0);
MATRIX[33] = Simd::splat(0);
MATRIX[34] = Simd::splat(0);
MATRIX[35] = Simd::splat(0);
MATRIX[36] = Simd::splat(0);
MATRIX[37] = Simd::splat(0);
MATRIX[38] = Simd::splat(0);
MATRIX[39] = Simd::splat(0);
MATRIX[40] = Simd::splat(0);
MATRIX[41] = Simd::splat(0);
MATRIX[42] = Simd::splat(0);
MATRIX[43] = Simd::splat(0);
MATRIX[44] = Simd::splat(0);
}
// {
// MATRIX[0] = Simd::splat(0);
// MATRIX[1] = Simd::splat(0);
// MATRIX[2] = Simd::splat(0);
// MATRIX[3] = Simd::splat(0);
// MATRIX[4] = Simd::splat(0);
// MATRIX[5] = Simd::splat(0);
// MATRIX[6] = Simd::splat(0);
// MATRIX[7] = Simd::splat(0);
// MATRIX[8] = Simd::splat(0);
// MATRIX[9] = Simd::splat(0);
// MATRIX[10] = Simd::splat(0);
// MATRIX[11] = Simd::splat(0);
// MATRIX[12] = Simd::splat(0);
// MATRIX[13] = Simd::splat(0);
// MATRIX[14] = Simd::splat(0);
// MATRIX[15] = Simd::splat(0);
// MATRIX[16] = Simd::splat(0);
// MATRIX[17] = Simd::splat(0);
// MATRIX[18] = Simd::splat(0);
// MATRIX[19] = Simd::splat(0);
// MATRIX[20] = Simd::splat(0);
// MATRIX[21] = Simd::splat(0);
// MATRIX[22] = Simd::splat(0);
// MATRIX[23] = Simd::splat(0);
// MATRIX[24] = Simd::splat(0);
// MATRIX[25] = Simd::splat(0);
// MATRIX[26] = Simd::splat(0);
// MATRIX[27] = Simd::splat(0);
// MATRIX[28] = Simd::splat(0);
// MATRIX[29] = Simd::splat(0);
// MATRIX[30] = Simd::splat(0);
// MATRIX[31] = Simd::splat(0);
// MATRIX[32] = Simd::splat(0);
// MATRIX[33] = Simd::splat(0);
// MATRIX[34] = Simd::splat(0);
// MATRIX[35] = Simd::splat(0);
// MATRIX[36] = Simd::splat(0);
// MATRIX[37] = Simd::splat(0);
// MATRIX[38] = Simd::splat(0);
// MATRIX[39] = Simd::splat(0);
// MATRIX[40] = Simd::splat(0);
// MATRIX[41] = Simd::splat(0);
// MATRIX[42] = Simd::splat(0);
// MATRIX[43] = Simd::splat(0);
// MATRIX[44] = Simd::splat(0);
// }

let matrix: &mut [u8x16; 90] = transmute(matrix);
let r = s.as_ptr_range();
let mut ptr = r.start;
macro_rules! fast_bt {
($bt:literal, $i0:expr, $i1:expr) => {
asm!(
concat!($bt, " dword ptr[{base} + {i0:r}], {i1:e}"),
base = in(reg) matrix,
i0 = in(reg) $i0,
i1 = in(reg) $i1,
options(nostack),
);
};
}
loop {
let chunk = (ptr as *const u8x16).read_unaligned();
if _mm_testc_si128(
chunk.into(),
u8x16::from_array([16, 0, 0, 16, 0, 0, 16, 0, 0, 16, 0, 0, 0, 0, 0, 0]).into(),
) == 0
{
break;
}
let normalized = chunk
- Simd::from_array([
b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0, 0, 0, 0, 0,
]);
let shuffled = _mm_shuffle_epi8(
normalized.into(),
i8x16::from_array([0, 1, -1, -1, 3, 4, -1, -1, 6, 7, -1, -1, 9, 10, -1, -1]).into(),
);
let indices: u32x4 = _mm_maddubs_epi16(
u8x16::from_array([160, 16, 0, 0, 10, 1, 0, 0, 160, 16, 0, 0, 10, 1, 0, 0]).into(),
shuffled,
)
.into();
fast_bt!("bts", indices[0], indices[1]);
fast_bt!("bts", indices[2], indices[3]);
ptr = ptr.add(12);
}
matrix.iter().copied().sum::<u8x16>().reduce_sum() as u32
asm!(
"vmovdqa ymmword ptr[{table}], {zero}",
"vmovdqa ymmword ptr[{table} + 32], {zero}",
"vmovdqa ymmword ptr[{table} + 64], {zero}",
"vmovdqa ymmword ptr[{table} + 96], {zero}",
"vmovdqa ymmword ptr[{table} + 128], {zero}",
"vmovdqa ymmword ptr[{table} + 160], {zero}",
"vmovdqa ymmword ptr[{table} + 192], {zero}",
"vmovdqa ymmword ptr[{table} + 224], {zero}",
"vmovdqa ymmword ptr[{table} + 256], {zero}",
"vmovdqa ymmword ptr[{table} + 288], {zero}",
"vmovdqa ymmword ptr[{table} + 320], {zero}",
"vmovdqa ymmword ptr[{table} + 352], {zero}",
"vmovdqa ymmword ptr[{table} + 384], {zero}",
"vmovdqa ymmword ptr[{table} + 416], {zero}",
"vmovdqa ymmword ptr[{table} + 448], {zero}",
"vmovdqa ymmword ptr[{table} + 480], {zero}",
"vmovdqa ymmword ptr[{table} + 512], {zero}",
"vmovdqa ymmword ptr[{table} + 544], {zero}",
"vmovdqa ymmword ptr[{table} + 576], {zero}",
"vmovdqa ymmword ptr[{table} + 608], {zero}",
"vmovdqa ymmword ptr[{table} + 640], {zero}",
"vmovdqa ymmword ptr[{table} + 672], {zero}",
"vmovdqa ymmword ptr[{table} + 704], {zero}",
"vmovdqa ymmword ptr[{table} + 736], {zero}",
"vmovdqa ymmword ptr[{table} + 768], {zero}",
"vmovdqa ymmword ptr[{table} + 800], {zero}",
"vmovdqa ymmword ptr[{table} + 832], {zero}",
"vmovdqa ymmword ptr[{table} + 864], {zero}",
"vmovdqa ymmword ptr[{table} + 896], {zero}",
"vmovdqa ymmword ptr[{table} + 928], {zero}",
"vmovdqa ymmword ptr[{table} + 960], {zero}",
"vmovdqa ymmword ptr[{table} + 992], {zero}",
"vmovdqa ymmword ptr[{table} + 1024], {zero}",
"vmovdqa ymmword ptr[{table} + 1056], {zero}",
"vmovdqa ymmword ptr[{table} + 1088], {zero}",
"vmovdqa ymmword ptr[{table} + 1120], {zero}",
"vmovdqa ymmword ptr[{table} + 1152], {zero}",
"vmovdqa ymmword ptr[{table} + 1184], {zero}",
"vmovdqa ymmword ptr[{table} + 1216], {zero}",
"vmovdqa ymmword ptr[{table} + 1248], {zero}",
"vmovdqa ymmword ptr[{table} + 1280], {zero}",
"vmovdqa ymmword ptr[{table} + 1312], {zero}",
"vmovdqa ymmword ptr[{table} + 1344], {zero}",
"vmovdqa ymmword ptr[{table} + 1376], {zero}",
"vmovdqa ymmword ptr[{table} + 1408], {zero}",
"vmovdqu {chunk:x}, xmmword ptr[{ptr}]",
"vpsubb {chunk:x}, {chunk:x}, {normalize1:x}",
"vpshufb {chunk:x}, {chunk:x}, {shuffle1}",
"vpmaddubsw {chunk:x}, {mults1}, {chunk:x}",
"vmovd {t1:e}, {chunk:x}",
"vpextrd {t2:e}, {chunk:x}, 1",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vpextrd {t1:e}, {chunk:x}, 2",
"vpextrd {t2:e}, {chunk:x}, 3",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vmovdqu {chunk}, ymmword ptr[{ptr} + 11]",
"20:",
"vpsubb {chunk}, {chunk}, {normalize2:y}",
"vpshufb {chunk}, {chunk}, {shuffle2}",
"vpmaddubsw {chunk}, {mults2}, {chunk}",
"vmovd {t1:e}, {chunk:x}",
"vpextrw {t2:e}, {chunk:x}, 2",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vpextrw {t1:e}, {chunk:x}, 3",
"vpextrw {t2:e}, {chunk:x}, 4",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vpextrw {t1:e}, {chunk:x}, 5",
"vextracti128 {upperchunk:x}, {chunk}, 1",
"vmovd {t2:e}, {upperchunk:x}",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vpextrw {t1:e}, {upperchunk:x}, 2",
"vpextrw {t2:e}, {upperchunk:x}, 3",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vpextrw {t1:e}, {upperchunk:x}, 4",
"vpextrw {t2:e}, {upperchunk:x}, 5",
"bts dword ptr[{table} + {t1}], {t2:e}",
"add {ptr}, 30",
"vmovdqu {chunk}, ymmword ptr[{ptr} + 11]",
"vptest {chunk}, {newline_mask}",
"jb 20b",
"vpcmpgtb {chunk}, {newline_mask}, {chunk}",
"vpmovmskb {t1}, {chunk}",
"tzcnt {t1:e}, {t1:e}",
"vmovdqu {chunk}, ymmword ptr[{ptr} + {t1} - 20]",
"vpsubb {chunk}, {chunk}, {normalize2:y}",
"vpshufb {chunk}, {chunk}, {shuffle2}",
"vpmaddubsw {chunk}, {mults2}, {chunk}",
"vmovd {t1:e}, {chunk:x}",
"vpextrw {t2:e}, {chunk:x}, 2",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vpextrw {t1:e}, {chunk:x}, 3",
"vpextrw {t2:e}, {chunk:x}, 4",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vpextrw {t1:e}, {chunk:x}, 5",
"vextracti128 {upperchunk:x}, {chunk}, 1",
"vmovd {t2:e}, {upperchunk:x}",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vpextrw {t1:e}, {upperchunk:x}, 2",
"vpextrw {t2:e}, {upperchunk:x}, 3",
"bts dword ptr[{table} + {t1}], {t2:e}",
"vpextrw {t1:e}, {upperchunk:x}, 4",
"vpextrw {t2:e}, {upperchunk:x}, 5",
"bts dword ptr[{table} + {t1}], {t2:e}",
table = in(reg) matrix,
ptr = inout(reg) s.as_ptr() => _,
zero = in(ymm_reg) u8x32::splat(0),
normalize1 = in(xmm_reg) u8x16::from_array([b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0, 0, 0, 0, 0]),
shuffle1 = in(xmm_reg) i8x16::from_array([0, 1, -1, -1, 3, 4, -1, -1, 6, 7, -1, -1, 9, 10, -1, -1]),
mults1 = in(xmm_reg) u8x16::from_array([160, 16, 0, 0, 10, 1, 0, 0, 160, 16, 0, 0, 10, 1, 0, 0]),
normalize2 = in(ymm_reg) u8x32::from_array([
0, b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0,
b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0, b'1', b'0', 0, 0,
]),
shuffle2 = in(ymm_reg) i8x32::from_array([
1, 2, -1, -1, 4, 5, 7, 8, 10, 11, -1, -1, -1, -1, -1, -1,
0, 1, -1, -1, 3, 4, 6, 7, 9, 10, -1, -1, -1, -1, -1, -1,
]),
mults2 = in(ymm_reg) u8x32::from_array([
160, 16, 0, 0, 10, 1, 160, 16, 10, 1, 0, 0, 0, 0, 0, 0,
160, 16, 0, 0, 10, 1, 160, 16, 10, 1, 0, 0, 0, 0, 0, 0,
]),
newline_mask = in(ymm_reg) u8x32::from_array([
0, 0, 0, 0, 0, 0, 0, 16, 0, 0, 0, 0, 0, 16, 0, 0, 0, 0, 0, 16, 0, 0, 0, 0, 0, 16, 0, 0, 0, 0, 0, 16,
]),
t1 = out(reg) _,
t2 = out(reg) _,
chunk = out(ymm_reg) _,
upperchunk = out(ymm_reg) _,
);

0
}

#[target_feature(enable = "avx2,bmi1,bmi2,cmpxchg16b,lzcnt,movbe,popcnt")]
Expand Down

0 comments on commit a89346d

Please sign in to comment.