wanderer/ BLAKE3
1
0
Fork 0
mirror of https://github.com/BLAKE3-team/BLAKE3 synced 2024-05-23 08:46:06 +02:00
Code Issues Releases Activity

SSE2 intrinsic: emulate _mm_blend_epi16 SSE4.1 intrinsic with SSE2 intrinsics 

Use a constant mask to blend according to (mask & b) | ((~mask) & a).

 * src/rust_sse2.rs: emulate _mm_blend_epi16 using SSE2 intrinsics
 * c/blake3_sse2.c: Likewise.
This commit is contained in:
Matthew Krupcale 2020-08-23 21:12:38 -04:00
parent d91f20dd29
commit 40a4a2b6b0
2 changed files with 39 additions and 24 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

31
c/blake3_sse2.c
View File

@ -80,6 +80,13 @@ INLINE void undiagonalize(__m128i *row0, __m128i *row2, __m128i *row3) {
*row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE(2, 1, 0, 3));
}
INLINE __m128i blend_epi16(__m128i a, __m128i b, const int imm8) {
__m128i mask = _mm_set1_epi16(imm8);
mask = _mm_mullo_epi16(mask, set4(0x40008000, 0x10002000, 0x04000800, 0x01000200));
mask = _mm_srai_epi16(mask, 15);
return _mm_or_si128(_mm_and_si128(mask, b), _mm_andnot_si128(mask, a));
}
INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
const uint8_t block[BLAKE3_BLOCK_LEN],
uint8_t block_len, uint64_t counter, uint8_t flags) {
@ -122,11 +129,11 @@ INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
diagonalize(&rows[0], &rows[2], &rows[3]);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -145,11 +152,11 @@ INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
diagonalize(&rows[0], &rows[2], &rows[3]);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -168,11 +175,11 @@ INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
diagonalize(&rows[0], &rows[2], &rows[3]);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -191,11 +198,11 @@ INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
diagonalize(&rows[0], &rows[2], &rows[3]);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -214,11 +221,11 @@ INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
diagonalize(&rows[0], &rows[2], &rows[3]);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -237,11 +244,11 @@ INLINE void compress_pre(__m128i rows[4], const uint32_t cv[8],
g1(&rows[0], &rows[1], &rows[2], &rows[3], t0);
t1 = _mm_shuffle_ps2(m2, m3, _MM_SHUFFLE(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(&rows[0], &rows[1], &rows[2], &rows[3], t1);
diagonalize(&rows[0], &rows[2], &rows[3]);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE(1, 3, 2, 0));
g1(&rows[0], &rows[1], &rows[2], &rows[3], t2);
t3 = _mm_unpackhi_epi32(m1, m3);

32
src/rust_sse2.rs
View File

@ -137,6 +137,14 @@ unsafe fn undiagonalize(row0: &mut __m128i, row2: &mut __m128i, row3: &mut __m12
*row2 = _mm_shuffle_epi32(*row2, _MM_SHUFFLE!(2, 1, 0, 3));
}
#[inline(always)]
unsafe fn blend_epi16(a: __m128i, b: __m128i, imm8: i32) -> __m128i {
let mut mask = _mm_set1_epi16(imm8 as i16);
mask = _mm_mullo_epi16(mask, set4(0x40008000, 0x10002000, 0x04000800, 0x01000200));
mask = _mm_srai_epi16(mask, 15);
_mm_or_si128(_mm_and_si128(mask, b), _mm_andnot_si128(mask, a))
}
#[inline(always)]
unsafe fn compress_pre(
cv: &CVWords,
@ -192,11 +200,11 @@ unsafe fn compress_pre(
g1(row0, row1, row2, row3, t0);
t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(row0, row1, row2, row3, t1);
diagonalize(row0, row2, row3);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
g1(row0, row1, row2, row3, t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -215,11 +223,11 @@ unsafe fn compress_pre(
g1(row0, row1, row2, row3, t0);
t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(row0, row1, row2, row3, t1);
diagonalize(row0, row2, row3);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
g1(row0, row1, row2, row3, t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -238,11 +246,11 @@ unsafe fn compress_pre(
g1(row0, row1, row2, row3, t0);
t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(row0, row1, row2, row3, t1);
diagonalize(row0, row2, row3);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
g1(row0, row1, row2, row3, t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -261,11 +269,11 @@ unsafe fn compress_pre(
g1(row0, row1, row2, row3, t0);
t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(row0, row1, row2, row3, t1);
diagonalize(row0, row2, row3);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
g1(row0, row1, row2, row3, t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -284,11 +292,11 @@ unsafe fn compress_pre(
g1(row0, row1, row2, row3, t0);
t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(row0, row1, row2, row3, t1);
diagonalize(row0, row2, row3);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
g1(row0, row1, row2, row3, t2);
t3 = _mm_unpackhi_epi32(m1, m3);
@ -307,11 +315,11 @@ unsafe fn compress_pre(
g1(row0, row1, row2, row3, t0);
t1 = shuffle2!(m2, m3, _MM_SHUFFLE!(3, 3, 2, 2));
tt = _mm_shuffle_epi32(m0, _MM_SHUFFLE!(0, 0, 3, 3));
t1 = _mm_blend_epi16(tt, t1, 0xCC);
t1 = blend_epi16(tt, t1, 0xCC);
g2(row0, row1, row2, row3, t1);
diagonalize(row0, row2, row3);
t2 = _mm_unpacklo_epi64(m3, m1);
tt = _mm_blend_epi16(t2, m2, 0xC0);
tt = blend_epi16(t2, m2, 0xC0);
t2 = _mm_shuffle_epi32(tt, _MM_SHUFFLE!(1, 3, 2, 0));
g1(row0, row1, row2, row3, t2);
t3 = _mm_unpackhi_epi32(m1, m3);