Skip to content

Commit

Permalink
Add VPCLMULQDQ Intrinsics (#943)
Browse files Browse the repository at this point in the history
  • Loading branch information
DevJPM authored Dec 1, 2020
1 parent 0a91a62 commit b3b03be
Show file tree
Hide file tree
Showing 2 changed files with 269 additions and 0 deletions.
266 changes: 266 additions & 0 deletions crates/core_arch/src/x86/avx512vpclmulqdq.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,266 @@
//! Vectorized Carry-less Multiplication (VCLMUL)
//!
//! The reference is [Intel 64 and IA-32 Architectures Software Developer's
//! Manual Volume 2: Instruction Set Reference, A-Z][intel64_ref] (p. 4-241).
//!
//! [intel64_ref]: http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-instruction-set-reference-manual-325383.pdf
use crate::core_arch::x86::__m256i;
use crate::core_arch::x86::__m512i;

#[cfg(test)]
use crate::stdarch_test::assert_instr;

#[allow(improper_ctypes)]
extern "C" {
#[link_name = "llvm.x86.pclmulqdq.256"]
fn pclmulqdq_256(a: __m256i, round_key: __m256i, imm8: u8) -> __m256i;
#[link_name = "llvm.x86.pclmulqdq.512"]
fn pclmulqdq_512(a: __m512i, round_key: __m512i, imm8: u8) -> __m512i;
}

// for some odd reason on x86_64 we generate the correct long name instructions
// but on i686 we generate the short name + imm8
// so we need to special-case on that...

/// Performs a carry-less multiplication of two 64-bit polynomials over the
/// finite field GF(2^k) - in each of the 4 128-bit lanes.
///
/// The immediate byte is used for determining which halves of each lane `a` and `b`
/// should be used. Immediate bits other than 0 and 4 are ignored.
/// All lanes share immediate byte.
///
/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_clmulepi64_epi128)
#[inline]
#[target_feature(enable = "avx512vpclmulqdq,avx512f")]
// technically according to Intel's documentation we don't need avx512f here, however LLVM gets confused otherwise
#[cfg_attr(test, assert_instr(vpclmul, imm8 = 0))]
#[rustc_args_required_const(2)]
pub unsafe fn _mm512_clmulepi64_epi128(a: __m512i, b: __m512i, imm8: i32) -> __m512i {
macro_rules! call {
($imm8:expr) => {
pclmulqdq_512(a, b, $imm8)
};
}
constify_imm8!(imm8, call)
}

/// Performs a carry-less multiplication of two 64-bit polynomials over the
/// finite field GF(2^k) - in each of the 2 128-bit lanes.
///
/// The immediate byte is used for determining which halves of each lane `a` and `b`
/// should be used. Immediate bits other than 0 and 4 are ignored.
/// All lanes share immediate byte.
///
/// [Intel's documentation](https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm256_clmulepi64_epi128)
#[inline]
#[target_feature(enable = "avx512vpclmulqdq,avx512vl")]
#[cfg_attr(test, assert_instr(vpclmul, imm8 = 0))]
#[rustc_args_required_const(2)]
pub unsafe fn _mm256_clmulepi64_epi128(a: __m256i, b: __m256i, imm8: i32) -> __m256i {
macro_rules! call {
($imm8:expr) => {
pclmulqdq_256(a, b, $imm8)
};
}
constify_imm8!(imm8, call)
}

#[cfg(test)]
mod tests {
// The constants in the tests below are just bit patterns. They should not
// be interpreted as integers; signedness does not make sense for them, but
// __mXXXi happens to be defined in terms of signed integers.
#![allow(overflowing_literals)]

use stdarch_test::simd_test;

use crate::core_arch::x86::*;

macro_rules! verify_kat_pclmul {
($broadcast:ident, $clmul:ident, $assert:ident) => {
// Constants taken from https://software.intel.com/sites/default/files/managed/72/cc/clmul-wp-rev-2.02-2014-04-20.pdf
let a = _mm_set_epi64x(0x7b5b546573745665, 0x63746f725d53475d);
let a = $broadcast(a);
let b = _mm_set_epi64x(0x4869285368617929, 0x5b477565726f6e5d);
let b = $broadcast(b);
let r00 = _mm_set_epi64x(0x1d4d84c85c3440c0, 0x929633d5d36f0451);
let r00 = $broadcast(r00);
let r01 = _mm_set_epi64x(0x1bd17c8d556ab5a1, 0x7fa540ac2a281315);
let r01 = $broadcast(r01);
let r10 = _mm_set_epi64x(0x1a2bf6db3a30862f, 0xbabf262df4b7d5c9);
let r10 = $broadcast(r10);
let r11 = _mm_set_epi64x(0x1d1e1f2c592e7c45, 0xd66ee03e410fd4ed);
let r11 = $broadcast(r11);

$assert($clmul(a, b, 0x00), r00);
$assert($clmul(a, b, 0x10), r01);
$assert($clmul(a, b, 0x01), r10);
$assert($clmul(a, b, 0x11), r11);

let a0 = _mm_set_epi64x(0x0000000000000000, 0x8000000000000000);
let a0 = $broadcast(a0);
let r = _mm_set_epi64x(0x4000000000000000, 0x0000000000000000);
let r = $broadcast(r);
$assert($clmul(a0, a0, 0x00), r);
}
}

macro_rules! unroll {
($target:ident[4] = $op:ident($source:ident,4);) => {
$target[3] = $op($source, 3);
$target[2] = $op($source, 2);
unroll! {$target[2] = $op($source,2);}
};
($target:ident[2] = $op:ident($source:ident,2);) => {
$target[1] = $op($source, 1);
$target[0] = $op($source, 0);
};
(assert_eq_m128i($op:ident($vec_res:ident,4),$lin_res:ident[4]);) => {
assert_eq_m128i($op($vec_res, 3), $lin_res[3]);
assert_eq_m128i($op($vec_res, 2), $lin_res[2]);
unroll! {assert_eq_m128i($op($vec_res,2),$lin_res[2]);}
};
(assert_eq_m128i($op:ident($vec_res:ident,2),$lin_res:ident[2]);) => {
assert_eq_m128i($op($vec_res, 1), $lin_res[1]);
assert_eq_m128i($op($vec_res, 0), $lin_res[0]);
};
}

// this function tests one of the possible 4 instances
// with different inputs across lanes
#[target_feature(enable = "avx512vpclmulqdq,avx512f")]
unsafe fn verify_512_helper(
linear: unsafe fn(__m128i, __m128i) -> __m128i,
vectorized: unsafe fn(__m512i, __m512i) -> __m512i,
) {
let a = _mm512_set_epi64(
0xDCB4DB3657BF0B7D,
0x18DB0601068EDD9F,
0xB76B908233200DC5,
0xE478235FA8E22D5E,
0xAB05CFFA2621154C,
0x1171B47A186174C9,
0x8C6B6C0E7595CEC9,
0xBE3E7D4934E961BD,
);
let b = _mm512_set_epi64(
0x672F6F105A94CEA7,
0x8298B8FFCA5F829C,
0xA3927047B3FB61D8,
0x978093862CDE7187,
0xB1927AB22F31D0EC,
0xA9A5DA619BE4D7AF,
0xCA2590F56884FDC6,
0x19BE9F660038BDB5,
);

let mut a_decomp = [_mm_setzero_si128(); 4];
unroll! {a_decomp[4] = _mm512_extracti32x4_epi32(a,4);}
let mut b_decomp = [_mm_setzero_si128(); 4];
unroll! {b_decomp[4] = _mm512_extracti32x4_epi32(b,4);}

let r = vectorized(a, b);
let mut e_decomp = [_mm_setzero_si128(); 4];
for i in 0..4 {
e_decomp[i] = linear(a_decomp[i], b_decomp[i]);
}
unroll! {assert_eq_m128i(_mm512_extracti32x4_epi32(r,4),e_decomp[4]);}
}

// this function tests one of the possible 4 instances
// with different inputs across lanes for the VL version
#[target_feature(enable = "avx512vpclmulqdq,avx512vl")]
unsafe fn verify_256_helper(
linear: unsafe fn(__m128i, __m128i) -> __m128i,
vectorized: unsafe fn(__m256i, __m256i) -> __m256i,
) {
let a = _mm512_set_epi64(
0xDCB4DB3657BF0B7D,
0x18DB0601068EDD9F,
0xB76B908233200DC5,
0xE478235FA8E22D5E,
0xAB05CFFA2621154C,
0x1171B47A186174C9,
0x8C6B6C0E7595CEC9,
0xBE3E7D4934E961BD,
);
let b = _mm512_set_epi64(
0x672F6F105A94CEA7,
0x8298B8FFCA5F829C,
0xA3927047B3FB61D8,
0x978093862CDE7187,
0xB1927AB22F31D0EC,
0xA9A5DA619BE4D7AF,
0xCA2590F56884FDC6,
0x19BE9F660038BDB5,
);

let mut a_decomp = [_mm_setzero_si128(); 2];
unroll! {a_decomp[2] = _mm512_extracti32x4_epi32(a,2);}
let mut b_decomp = [_mm_setzero_si128(); 2];
unroll! {b_decomp[2] = _mm512_extracti32x4_epi32(b,2);}

let r = vectorized(
_mm512_extracti64x4_epi64(a, 0),
_mm512_extracti64x4_epi64(b, 0),
);
let mut e_decomp = [_mm_setzero_si128(); 2];
for i in 0..2 {
e_decomp[i] = linear(a_decomp[i], b_decomp[i]);
}
unroll! {assert_eq_m128i(_mm256_extracti128_si256(r,2),e_decomp[2]);}
}

#[simd_test(enable = "avx512vpclmulqdq,avx512f")]
unsafe fn test_mm512_clmulepi64_epi128() {
verify_kat_pclmul!(
_mm512_broadcast_i32x4,
_mm512_clmulepi64_epi128,
assert_eq_m512i
);

verify_512_helper(
|a, b| _mm_clmulepi64_si128(a, b, 0x00),
|a, b| _mm512_clmulepi64_epi128(a, b, 0x00),
);
verify_512_helper(
|a, b| _mm_clmulepi64_si128(a, b, 0x01),
|a, b| _mm512_clmulepi64_epi128(a, b, 0x01),
);
verify_512_helper(
|a, b| _mm_clmulepi64_si128(a, b, 0x10),
|a, b| _mm512_clmulepi64_epi128(a, b, 0x10),
);
verify_512_helper(
|a, b| _mm_clmulepi64_si128(a, b, 0x11),
|a, b| _mm512_clmulepi64_epi128(a, b, 0x11),
);
}

#[simd_test(enable = "avx512vpclmulqdq,avx512vl")]
unsafe fn test_mm256_clmulepi64_epi128() {
verify_kat_pclmul!(
_mm256_broadcastsi128_si256,
_mm256_clmulepi64_epi128,
assert_eq_m256i
);

verify_256_helper(
|a, b| _mm_clmulepi64_si128(a, b, 0x00),
|a, b| _mm256_clmulepi64_epi128(a, b, 0x00),
);
verify_256_helper(
|a, b| _mm_clmulepi64_si128(a, b, 0x01),
|a, b| _mm256_clmulepi64_epi128(a, b, 0x01),
);
verify_256_helper(
|a, b| _mm_clmulepi64_si128(a, b, 0x10),
|a, b| _mm256_clmulepi64_epi128(a, b, 0x10),
);
verify_256_helper(
|a, b| _mm_clmulepi64_si128(a, b, 0x11),
|a, b| _mm256_clmulepi64_epi128(a, b, 0x11),
);
}
}
3 changes: 3 additions & 0 deletions crates/core_arch/src/x86/mod.rs
Original file line number Diff line number Diff line change
Expand Up @@ -675,6 +675,9 @@ pub use self::avx512ifma::*;
mod avx512vaes;
pub use self::avx512vaes::*;

mod avx512vpclmulqdq;
pub use self::avx512vpclmulqdq::*;

mod bt;
pub use self::bt::*;

Expand Down

0 comments on commit b3b03be

Please sign in to comment.