diff options
Diffstat (limited to 'src/vppinfra/vector_sse42.h')
| -rw-r--r-- | src/vppinfra/vector_sse42.h | 197 |
1 files changed, 0 insertions, 197 deletions
diff --git a/src/vppinfra/vector_sse42.h b/src/vppinfra/vector_sse42.h index 1bdb34b866e..7e75ad28710 100644 --- a/src/vppinfra/vector_sse42.h +++ b/src/vppinfra/vector_sse42.h @@ -146,43 +146,6 @@ u64x2_interleave_lo (u64x2 a, u64x2 b) return (u64x2) _mm_unpacklo_epi64 ((__m128i) a, (__m128i) b); } -/* 64 bit interleaves. */ -always_inline u8x8 -u8x8_interleave_hi (u8x8 a, u8x8 b) -{ - return (u8x8) _m_punpckhbw ((__m64) a, (__m64) b); -} - -always_inline u8x8 -u8x8_interleave_lo (u8x8 a, u8x8 b) -{ - return (u8x8) _m_punpcklbw ((__m64) a, (__m64) b); -} - -always_inline u16x4 -u16x4_interleave_hi (u16x4 a, u16x4 b) -{ - return (u16x4) _m_punpckhwd ((__m64) a, (__m64) b); -} - -always_inline u16x4 -u16x4_interleave_lo (u16x4 a, u16x4 b) -{ - return (u16x4) _m_punpcklwd ((__m64) a, (__m64) b); -} - -always_inline u32x2 -u32x2_interleave_hi (u32x2 a, u32x2 b) -{ - return (u32x2) _m_punpckhdq ((__m64) a, (__m64) b); -} - -always_inline u32x2 -u32x2_interleave_lo (u32x2 a, u32x2 b) -{ - return (u32x2) _m_punpckldq ((__m64) a, (__m64) b); -} - /* 128 bit packs. */ #define _(f, t, fn) \ always_inline t t##_pack (f lo, f hi) \ @@ -255,54 +218,6 @@ _(i, 16, 8, left, sll) _(i, 32, 4, left, sll) _(i, 64, 2, left, sll) _(i, 16, 8, right, sra) _(i, 32, 4, right, sra) #undef _ -/* 64 bit shifts. */ - always_inline u16x4 -u16x4_shift_left (u16x4 x, u16x4 i) -{ - return (u16x4) _m_psllw ((__m64) x, (__m64) i); -}; - -always_inline u32x2 -u32x2_shift_left (u32x2 x, u32x2 i) -{ - return (u32x2) _m_pslld ((__m64) x, (__m64) i); -}; - -always_inline u16x4 -u16x4_shift_right (u16x4 x, u16x4 i) -{ - return (u16x4) _m_psrlw ((__m64) x, (__m64) i); -}; - -always_inline u32x2 -u32x2_shift_right (u32x2 x, u32x2 i) -{ - return (u32x2) _m_psrld ((__m64) x, (__m64) i); -}; - -always_inline i16x4 -i16x4_shift_left (i16x4 x, i16x4 i) -{ - return (i16x4) _m_psllw ((__m64) x, (__m64) i); -}; - -always_inline i32x2 -i32x2_shift_left (i32x2 x, i32x2 i) -{ - return (i32x2) _m_pslld ((__m64) x, (__m64) i); -}; - -always_inline i16x4 -i16x4_shift_right (i16x4 x, i16x4 i) -{ - return (i16x4) _m_psraw ((__m64) x, (__m64) i); -}; - -always_inline i32x2 -i32x2_shift_right (i32x2 x, i32x2 i) -{ - return (i32x2) _m_psrad ((__m64) x, (__m64) i); -}; #define u8x16_word_shift_left(a,n) (u8x16) _mm_slli_si128((__m128i) a, n) #define u8x16_word_shift_right(a,n) (u8x16) _mm_srli_si128((__m128i) a, n) @@ -366,118 +281,6 @@ _(u64, 2, right, left); #undef _ -#ifndef __clang__ -#define _(t,n,lr1,lr2) \ - always_inline t##x##n \ - t##x##n##_word_rotate2_##lr1 (t##x##n w0, t##x##n w1, int i) \ - { \ - int m = sizeof (t##x##n) / sizeof (t); \ - ASSERT (i >= 0 && i < m); \ - return (t##x##n##_word_shift_##lr1 (w0, i) \ - | t##x##n##_word_shift_##lr2 (w1, m - i)); \ - } \ - \ - always_inline t##x##n \ - t##x##n##_word_rotate_##lr1 (t##x##n w0, int i) \ - { return t##x##n##_word_rotate2_##lr1 (w0, w0, i); } - -_(u8, 16, left, right); -_(u8, 16, right, left); -_(u16, 8, left, right); -_(u16, 8, right, left); -_(u32, 4, left, right); -_(u32, 4, right, left); -_(u64, 2, left, right); -_(u64, 2, right, left); - -#undef _ -#endif - -#define u32x4_select(A,MASK) \ -({ \ - u32x4 _x, _y; \ - _x = (A); \ - asm volatile ("pshufd %[mask], %[x], %[y]" \ - : /* outputs */ [y] "=x" (_y) \ - : /* inputs */ [x] "x" (_x), [mask] "i" (MASK)); \ - _y; \ -}) - -#define u32x4_splat_word(x,i) \ - u32x4_select ((x), (((i) << (2*0)) \ - | ((i) << (2*1)) \ - | ((i) << (2*2)) \ - | ((i) << (2*3)))) - -/* Extract low order 32 bit word. */ -always_inline u32 -u32x4_get0 (u32x4 x) -{ - u32 result; - asm volatile ("movd %[x], %[result]": /* outputs */ [result] "=r" (result) - : /* inputs */ [x] "x" (x)); - return result; -} - -always_inline u32x4 -u32x4_set0 (u32 x) -{ - u32x4 result; - asm volatile ("movd %[x], %[result]": /* outputs */ [result] "=x" (result) - : /* inputs */ [x] "r" (x)); - return result; -} - -always_inline i32x4 -i32x4_set0 (i32 x) -{ - return (i32x4) u32x4_set0 ((u32) x); -} - -always_inline i32 -i32x4_get0 (i32x4 x) -{ - return (i32) u32x4_get0 ((u32x4) x); -} - -/* Converts all ones/zeros compare mask to bitmap. */ -always_inline u32 -u8x16_compare_byte_mask (u8x16 x) -{ - return _mm_movemask_epi8 ((__m128i) x); -} - -extern u8 u32x4_compare_word_mask_table[256]; - -always_inline u32 -u32x4_compare_word_mask (u32x4 x) -{ - u32 m = u8x16_compare_byte_mask ((u8x16) x); - return (u32x4_compare_word_mask_table[(m >> 0) & 0xff] - | (u32x4_compare_word_mask_table[(m >> 8) & 0xff] << 2)); -} - -always_inline u32 -u8x16_zero_byte_mask (u8x16 x) -{ - u8x16 zero = { 0 }; - return u8x16_compare_byte_mask (x == zero); -} - -always_inline u32 -u16x8_zero_byte_mask (u16x8 x) -{ - u16x8 zero = { 0 }; - return u8x16_compare_byte_mask ((u8x16) (x == zero)); -} - -always_inline u32 -u32x4_zero_byte_mask (u32x4 x) -{ - u32x4 zero = { 0 }; - return u8x16_compare_byte_mask ((u8x16) (x == zero)); -} - always_inline u32 u8x16_max_scalar (u8x16 x) { |