diff options
Diffstat (limited to 'src/vppinfra/vector/mask_compare.h')
| -rw-r--r-- | src/vppinfra/vector/mask_compare.h | 166 |
1 files changed, 166 insertions, 0 deletions
diff --git a/src/vppinfra/vector/mask_compare.h b/src/vppinfra/vector/mask_compare.h new file mode 100644 index 00000000000..cac48a31f47 --- /dev/null +++ b/src/vppinfra/vector/mask_compare.h @@ -0,0 +1,166 @@ +/* SPDX-License-Identifier: Apache-2.0 + * Copyright(c) 2021 Cisco Systems, Inc. + */ + +#ifndef included_vector_mask_compare_h +#define included_vector_mask_compare_h +#include <vppinfra/clib.h> +#include <vppinfra/memcpy.h> + +static_always_inline u64 +clib_mask_compare_u16_x64 (u16 v, u16 *a, u32 n_elts) +{ + u64 mask = 0; +#if defined(CLIB_HAVE_VEC512) + u16x32 v32 = u16x32_splat (v); + u16x32u *av = (u16x32u *) a; + mask = ((u64) u16x32_is_equal_mask (av[0], v32) | + (u64) u16x32_is_equal_mask (av[1], v32) << 32); +#elif defined(CLIB_HAVE_VEC256) + u16x16 v16 = u16x16_splat (v); + u16x16u *av = (u16x16u *) a; + i8x32 x; + + x = i8x32_pack (v16 == av[0], v16 == av[1]); + mask = i8x32_msb_mask ((i8x32) u64x4_permute (x, 0, 2, 1, 3)); + x = i8x32_pack (v16 == av[2], v16 == av[3]); + mask |= (u64) i8x32_msb_mask ((i8x32) u64x4_permute (x, 0, 2, 1, 3)) << 32; +#elif defined(CLIB_HAVE_VEC128) && defined(__ARM_NEON) + u16x8 v8 = u16x8_splat (v); + u16x8 m = { 1, 2, 4, 8, 16, 32, 64, 128 }; + u16x8u *av = (u16x8u *) a; + + /* compare each u16 elemment with v8, result gives 0xffff in each element + of the resulting vector if comparison result is true. + Bitwise AND with m will give us one bit set for true result and offset + of that bit represend element index. Finally vaddvq_u16() gives us sum + of all elements of the vector which will give us u8 bitmap. */ + + for (int i = 0; i < 8; i++) + mask |= (u64) vaddvq_u16 ((av[i] == v8) & m) << (i * 8); + +#elif defined(CLIB_HAVE_VEC128) && defined(CLIB_HAVE_VEC128_MSB_MASK) + u16x8 v8 = u16x8_splat (v); + u16x8u *av = (u16x8u *) a; + mask = ((u64) i8x16_msb_mask (i8x16_pack (v8 == av[0], v8 == av[1])) | + (u64) i8x16_msb_mask (i8x16_pack (v8 == av[2], v8 == av[3])) << 16 | + (u64) i8x16_msb_mask (i8x16_pack (v8 == av[4], v8 == av[5])) << 32 | + (u64) i8x16_msb_mask (i8x16_pack (v8 == av[6], v8 == av[7])) << 48); +#else + for (int i = 0; i < n_elts; i++) + if (a[i] == v) + mask |= 1ULL << i; +#endif + return mask; +} + +/** \brief Compare 16-bit elemments with provied value and return bitmap + + @param v value to compare elements with + @param a array of u16 elements + @param mask array of u64 where reuslting mask will be stored + @param n_elts number of elements in the array + @return none +*/ + +static_always_inline void +clib_mask_compare_u16 (u16 v, u16 *a, u64 *mask, u32 n_elts) +{ + while (n_elts >= 64) + { + mask++[0] = clib_mask_compare_u16_x64 (v, a, 64); + n_elts -= 64; + a += 64; + } + + if (PREDICT_TRUE (n_elts == 0)) + return; + + mask[0] = clib_mask_compare_u16_x64 (v, a, n_elts) & pow2_mask (n_elts); +} + +static_always_inline u64 +clib_mask_compare_u32_x64 (u32 v, u32 *a, u32 n_elts) +{ + u64 mask = 0; +#if defined(CLIB_HAVE_VEC512) + u32x16 v16 = u32x16_splat (v); + u32x16u *av = (u32x16u *) a; + mask = ((u64) u32x16_is_equal_mask (av[0], v16) | + (u64) u32x16_is_equal_mask (av[1], v16) << 16 | + (u64) u32x16_is_equal_mask (av[2], v16) << 32 | + (u64) u32x16_is_equal_mask (av[3], v16) << 48); +#elif defined(CLIB_HAVE_VEC256) + u32x8 v8 = u32x8_splat (v); + u32x8u *av = (u32x8u *) a; + u32x8 m = { 0, 4, 1, 5, 2, 6, 3, 7 }; + i8x32 c; + + c = i8x32_pack (i16x16_pack ((i32x8) (v8 == av[0]), (i32x8) (v8 == av[1])), + i16x16_pack ((i32x8) (v8 == av[2]), (i32x8) (v8 == av[3]))); + mask = i8x32_msb_mask ((i8x32) u32x8_permute ((u32x8) c, m)); + + c = i8x32_pack (i16x16_pack ((i32x8) (v8 == av[4]), (i32x8) (v8 == av[5])), + i16x16_pack ((i32x8) (v8 == av[6]), (i32x8) (v8 == av[7]))); + mask |= (u64) i8x32_msb_mask ((i8x32) u32x8_permute ((u32x8) c, m)) << 32; + +#elif defined(CLIB_HAVE_VEC128) && defined(__ARM_NEON) + u32x4 v4 = u32x4_splat (v); + u32x4 m = { 1, 2, 4, 8 }; + u32x4u *av = (u32x4u *) a; + + /* compare each u32 elemment with v4, result gives -1 in each element + of the resulting vector if comparison result is true. + Bitwise AND with m will give us one bit set for true result and offset + of that bit represend element index. Finally vaddvq_u32() gives us sum + of all elements of the vector which will give us u8 bitmap. */ + + for (int i = 0; i < 16; i++) + mask |= (u64) vaddvq_u32 ((av[i] == v4) & m) << (i * 4); + +#elif defined(CLIB_HAVE_VEC128) && defined(CLIB_HAVE_VEC128_MSB_MASK) + u32x4 v4 = u32x4_splat (v); + u32x4u *av = (u32x4u *) a; + + for (int i = 0; i < 4; i++) + { + i16x8 p1 = i16x8_pack (v4 == av[0], v4 == av[1]); + i16x8 p2 = i16x8_pack (v4 == av[2], v4 == av[3]); + mask |= (u64) i8x16_msb_mask (i8x16_pack (p1, p2)) << (i * 16); + av += 4; + } + +#else + for (int i = 0; i < n_elts; i++) + if (a[i] == v) + mask |= 1ULL << i; +#endif + return mask; +} + +/** \brief Compare 32-bit elemments with provied value and return bitmap + + @param v value to compare elements with + @param a array of u32 elements + @param mask array of u64 where reuslting mask will be stored + @param n_elts number of elements in the array + @return none +*/ + +static_always_inline void +clib_mask_compare_u32 (u32 v, u32 *a, u64 *bitmap, u32 n_elts) +{ + while (n_elts >= 64) + { + bitmap++[0] = clib_mask_compare_u32_x64 (v, a, 64); + n_elts -= 64; + a += 64; + } + + if (PREDICT_TRUE (n_elts == 0)) + return; + + bitmap[0] = clib_mask_compare_u32_x64 (v, a, n_elts) & pow2_mask (n_elts); +} + +#endif |