diff options
Diffstat (limited to 'src/vppinfra')
-rw-r--r-- | src/vppinfra/CMakeLists.txt | 9 | ||||
-rw-r--r-- | src/vppinfra/vector/compress.h | 78 | ||||
-rw-r--r-- | src/vppinfra/vector/mask_compare.h (renamed from src/vppinfra/vector_funcs.h) | 72 | ||||
-rw-r--r-- | src/vppinfra/vector/test/compress.c (renamed from src/vppinfra/test_vector_funcs_compress.c) | 4 | ||||
-rw-r--r-- | src/vppinfra/vector/test/mask_compare.c (renamed from src/vppinfra/test_vector_funcs_mask_compare.c) | 4 | ||||
-rw-r--r-- | src/vppinfra/vector/test/test.c (renamed from src/vppinfra/test_vector_funcs.c) | 2 | ||||
-rw-r--r-- | src/vppinfra/vector/test/test.h (renamed from src/vppinfra/test_vector_funcs.h) | 0 |
7 files changed, 90 insertions, 79 deletions
diff --git a/src/vppinfra/CMakeLists.txt b/src/vppinfra/CMakeLists.txt index 64420b96676..8f9d0746b42 100644 --- a/src/vppinfra/CMakeLists.txt +++ b/src/vppinfra/CMakeLists.txt @@ -188,7 +188,8 @@ set(VPPINFRA_HEADERS vector_altivec.h vector_avx2.h vector_avx512.h - vector_funcs.h + vector/mask_compare.h + vector/compress.h vector.h vector_neon.h vector_sse42.h @@ -264,13 +265,13 @@ if(VPP_BUILD_VPPINFRA_TESTS) endforeach() set(test_files - test_vector_funcs_compress.c - test_vector_funcs_mask_compare.c + vector/test/compress.c + vector/test/mask_compare.c ) add_vpp_executable(test_vector_funcs SOURCES - test_vector_funcs.c + vector/test/test.c ${test_files} LINK_LIBRARIES vppinfra ) diff --git a/src/vppinfra/vector/compress.h b/src/vppinfra/vector/compress.h new file mode 100644 index 00000000000..1d5d84e77ea --- /dev/null +++ b/src/vppinfra/vector/compress.h @@ -0,0 +1,78 @@ +/* SPDX-License-Identifier: Apache-2.0 + * Copyright(c) 2021 Cisco Systems, Inc. + */ + +#ifndef included_vector_compress_h +#define included_vector_compress_h +#include <vppinfra/clib.h> +#include <vppinfra/memcpy.h> + +static_always_inline u32 * +clib_compress_u32_x64 (u32 *dst, u32 *src, u64 mask) +{ +#if defined(CLIB_HAVE_VEC512_COMPRESS) + u32x16u *sv = (u32x16u *) src; + for (int i = 0; i < 4; i++) + { + int cnt = _popcnt32 ((u16) mask); + u32x16_compress_store (sv[i], mask, dst); + dst += cnt; + mask >>= 16; + } + +#elif defined(CLIB_HAVE_VEC256_COMPRESS) + u32x8u *sv = (u32x8u *) src; + for (int i = 0; i < 8; i++) + { + int cnt = _popcnt32 ((u8) mask); + u32x8_compress_store (sv[i], mask, dst); + dst += cnt; + mask >>= 8; + } +#else + while (mask) + { + u16 bit = count_trailing_zeros (mask); + mask = clear_lowest_set_bit (mask); + dst++[0] = src[bit]; + } +#endif + return dst; +} + +/** \brief Compress array of 32-bit elemments into destination array based on + * mask + + @param dst destination array of u32 elements + @param src source array of u32 elements + @param mask array of u64 values representing compress mask + @param n_elts number of elements in the source array + @return number of elements stored in destionation array +*/ + +static_always_inline u32 +clib_compress_u32 (u32 *dst, u32 *src, u64 *mask, u32 n_elts) +{ + u32 *dst0 = dst; + while (n_elts >= 64) + { + if (mask[0] == ~0ULL) + { + clib_memcpy_u32 (dst, src, 64); + dst += 64; + } + else + dst = clib_compress_u32_x64 (dst, src, mask[0]); + + mask++; + src += 64; + n_elts -= 64; + } + + if (PREDICT_TRUE (n_elts == 0)) + return dst - dst0; + + return clib_compress_u32_x64 (dst, src, mask[0] & pow2_mask (n_elts)) - dst0; +} + +#endif diff --git a/src/vppinfra/vector_funcs.h b/src/vppinfra/vector/mask_compare.h index fed1b5cc3d8..cac48a31f47 100644 --- a/src/vppinfra/vector_funcs.h +++ b/src/vppinfra/vector/mask_compare.h @@ -2,8 +2,8 @@ * Copyright(c) 2021 Cisco Systems, Inc. */ -#ifndef included_vector_funcs_h -#define included_vector_funcs_h +#ifndef included_vector_mask_compare_h +#define included_vector_mask_compare_h #include <vppinfra/clib.h> #include <vppinfra/memcpy.h> @@ -163,72 +163,4 @@ clib_mask_compare_u32 (u32 v, u32 *a, u64 *bitmap, u32 n_elts) bitmap[0] = clib_mask_compare_u32_x64 (v, a, n_elts) & pow2_mask (n_elts); } -static_always_inline u32 * -clib_compress_u32_x64 (u32 *dst, u32 *src, u64 mask) -{ -#if defined(CLIB_HAVE_VEC512_COMPRESS) - u32x16u *sv = (u32x16u *) src; - for (int i = 0; i < 4; i++) - { - int cnt = _popcnt32 ((u16) mask); - u32x16_compress_store (sv[i], mask, dst); - dst += cnt; - mask >>= 16; - } - -#elif defined(CLIB_HAVE_VEC256_COMPRESS) - u32x8u *sv = (u32x8u *) src; - for (int i = 0; i < 8; i++) - { - int cnt = _popcnt32 ((u8) mask); - u32x8_compress_store (sv[i], mask, dst); - dst += cnt; - mask >>= 8; - } -#else - while (mask) - { - u16 bit = count_trailing_zeros (mask); - mask = clear_lowest_set_bit (mask); - dst++[0] = src[bit]; - } -#endif - return dst; -} - -/** \brief Compare array of 32-bit elemments into destination array based on - * mask - - @param dst destination array of u32 elements - @param src source array of u32 elements - @param mask array of u64 values representing compress mask - @param n_elts number of elements in the source array - @return number of elements stored in destionation array -*/ - -static_always_inline u32 -clib_compress_u32 (u32 *dst, u32 *src, u64 *mask, u32 n_elts) -{ - u32 *dst0 = dst; - while (n_elts >= 64) - { - if (mask[0] == ~0ULL) - { - clib_memcpy_u32 (dst, src, 64); - dst += 64; - } - else - dst = clib_compress_u32_x64 (dst, src, mask[0]); - - mask++; - src += 64; - n_elts -= 64; - } - - if (PREDICT_TRUE (n_elts == 0)) - return dst - dst0; - - return clib_compress_u32_x64 (dst, src, mask[0] & pow2_mask (n_elts)) - dst0; -} - #endif diff --git a/src/vppinfra/test_vector_funcs_compress.c b/src/vppinfra/vector/test/compress.c index c0815601e45..7e3eba9892d 100644 --- a/src/vppinfra/test_vector_funcs_compress.c +++ b/src/vppinfra/vector/test/compress.c @@ -3,8 +3,8 @@ */ #include <vppinfra/format.h> -#include <vppinfra/vector_funcs.h> -#include <vppinfra/test_vector_funcs.h> +#include <vppinfra/vector/test/test.h> +#include <vppinfra/vector/compress.h> __clib_test_fn u32 clib_compress_u32_wrapper (u32 *dst, u32 *src, u64 *mask, u32 n_elts) diff --git a/src/vppinfra/test_vector_funcs_mask_compare.c b/src/vppinfra/vector/test/mask_compare.c index 009f87712e4..64df0ee084a 100644 --- a/src/vppinfra/test_vector_funcs_mask_compare.c +++ b/src/vppinfra/vector/test/mask_compare.c @@ -3,8 +3,8 @@ */ #include <vppinfra/format.h> -#include <vppinfra/vector_funcs.h> -#include <vppinfra/test_vector_funcs.h> +#include <vppinfra/vector/test/test.h> +#include <vppinfra/vector/mask_compare.h> __clib_test_fn void clib_mask_compare_u16_wrapper (u16 v, u16 *a, u64 *mask, u32 n_elts) diff --git a/src/vppinfra/test_vector_funcs.c b/src/vppinfra/vector/test/test.c index 85f159a17c6..0e90bacce49 100644 --- a/src/vppinfra/test_vector_funcs.c +++ b/src/vppinfra/vector/test/test.c @@ -3,7 +3,7 @@ */ #include <vppinfra/format.h> -#include <vppinfra/test_vector_funcs.h> +#include <vppinfra/vector/test/test.h> test_registration_t *test_registrations[CLIB_MARCH_TYPE_N_VARIANTS] = {}; diff --git a/src/vppinfra/test_vector_funcs.h b/src/vppinfra/vector/test/test.h index bc499fb24e8..bc499fb24e8 100644 --- a/src/vppinfra/test_vector_funcs.h +++ b/src/vppinfra/vector/test/test.h |