10b57cec5SDimitry Andric /*===------ avx512vlvp2intersectintrin.h - VL VP2INTERSECT intrinsics ------=== 20b57cec5SDimitry Andric * 30b57cec5SDimitry Andric * 40b57cec5SDimitry Andric * Permission is hereby granted, free of charge, to any person obtaining a copy 50b57cec5SDimitry Andric * of this software and associated documentation files (the "Software"), to deal 60b57cec5SDimitry Andric * in the Software without restriction, including without limitation the rights 70b57cec5SDimitry Andric * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 80b57cec5SDimitry Andric * copies of the Software, and to permit persons to whom the Software is 90b57cec5SDimitry Andric * furnished to do so, subject to the following conditions: 100b57cec5SDimitry Andric * 110b57cec5SDimitry Andric * The above copyright notice and this permission notice shall be included in 120b57cec5SDimitry Andric * all copies or substantial portions of the Software. 130b57cec5SDimitry Andric * 140b57cec5SDimitry Andric * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 150b57cec5SDimitry Andric * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 160b57cec5SDimitry Andric * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 170b57cec5SDimitry Andric * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 180b57cec5SDimitry Andric * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 190b57cec5SDimitry Andric * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 200b57cec5SDimitry Andric * THE SOFTWARE. 210b57cec5SDimitry Andric * 220b57cec5SDimitry Andric *===-----------------------------------------------------------------------=== 230b57cec5SDimitry Andric */ 240b57cec5SDimitry Andric #ifndef __IMMINTRIN_H 250b57cec5SDimitry Andric #error "Never use <avx512vlvp2intersectintrin.h> directly; include <immintrin.h> instead." 260b57cec5SDimitry Andric #endif 270b57cec5SDimitry Andric 280b57cec5SDimitry Andric #ifndef _AVX512VLVP2INTERSECT_H 290b57cec5SDimitry Andric #define _AVX512VLVP2INTERSECT_H 300b57cec5SDimitry Andric 310b57cec5SDimitry Andric #define __DEFAULT_FN_ATTRS128 \ 32*5f757f3fSDimitry Andric __attribute__((__always_inline__, __nodebug__, \ 33*5f757f3fSDimitry Andric __target__("avx512vl,avx512vp2intersect,no-evex512"), \ 340b57cec5SDimitry Andric __min_vector_width__(128))) 350b57cec5SDimitry Andric 360b57cec5SDimitry Andric #define __DEFAULT_FN_ATTRS256 \ 37*5f757f3fSDimitry Andric __attribute__((__always_inline__, __nodebug__, \ 38*5f757f3fSDimitry Andric __target__("avx512vl,avx512vp2intersect,no-evex512"), \ 390b57cec5SDimitry Andric __min_vector_width__(256))) 400b57cec5SDimitry Andric /// Store, in an even/odd pair of mask registers, the indicators of the 410b57cec5SDimitry Andric /// locations of value matches between dwords in operands __a and __b. 420b57cec5SDimitry Andric /// 430b57cec5SDimitry Andric /// \headerfile <x86intrin.h> 440b57cec5SDimitry Andric /// 450b57cec5SDimitry Andric /// This intrinsic corresponds to the <c> VP2INTERSECTD </c> instruction. 460b57cec5SDimitry Andric /// 470b57cec5SDimitry Andric /// \param __a 480b57cec5SDimitry Andric /// A 256-bit vector of [8 x i32]. 490b57cec5SDimitry Andric /// \param __b 500b57cec5SDimitry Andric /// A 256-bit vector of [8 x i32] 510b57cec5SDimitry Andric /// \param __m0 520b57cec5SDimitry Andric /// A pointer point to 8-bit mask 530b57cec5SDimitry Andric /// \param __m1 540b57cec5SDimitry Andric /// A pointer point to 8-bit mask 550b57cec5SDimitry Andric static __inline__ void __DEFAULT_FN_ATTRS256 560b57cec5SDimitry Andric _mm256_2intersect_epi32(__m256i __a, __m256i __b, __mmask8 *__m0, __mmask8 *__m1) { 570b57cec5SDimitry Andric __builtin_ia32_vp2intersect_d_256((__v8si)__a, (__v8si)__b, __m0, __m1); 580b57cec5SDimitry Andric } 590b57cec5SDimitry Andric 600b57cec5SDimitry Andric /// Store, in an even/odd pair of mask registers, the indicators of the 610b57cec5SDimitry Andric /// locations of value matches between quadwords in operands __a and __b. 620b57cec5SDimitry Andric /// 630b57cec5SDimitry Andric /// \headerfile <x86intrin.h> 640b57cec5SDimitry Andric /// 650b57cec5SDimitry Andric /// This intrinsic corresponds to the <c> VP2INTERSECTQ </c> instruction. 660b57cec5SDimitry Andric /// 670b57cec5SDimitry Andric /// \param __a 680b57cec5SDimitry Andric /// A 256-bit vector of [4 x i64]. 690b57cec5SDimitry Andric /// \param __b 700b57cec5SDimitry Andric /// A 256-bit vector of [4 x i64] 710b57cec5SDimitry Andric /// \param __m0 720b57cec5SDimitry Andric /// A pointer point to 8-bit mask 730b57cec5SDimitry Andric /// \param __m1 740b57cec5SDimitry Andric /// A pointer point to 8-bit mask 750b57cec5SDimitry Andric static __inline__ void __DEFAULT_FN_ATTRS256 760b57cec5SDimitry Andric _mm256_2intersect_epi64(__m256i __a, __m256i __b, __mmask8 *__m0, __mmask8 *__m1) { 770b57cec5SDimitry Andric __builtin_ia32_vp2intersect_q_256((__v4di)__a, (__v4di)__b, __m0, __m1); 780b57cec5SDimitry Andric } 790b57cec5SDimitry Andric 800b57cec5SDimitry Andric /// Store, in an even/odd pair of mask registers, the indicators of the 810b57cec5SDimitry Andric /// locations of value matches between dwords in operands __a and __b. 820b57cec5SDimitry Andric /// 830b57cec5SDimitry Andric /// \headerfile <x86intrin.h> 840b57cec5SDimitry Andric /// 850b57cec5SDimitry Andric /// This intrinsic corresponds to the <c> VP2INTERSECTD </c> instruction. 860b57cec5SDimitry Andric /// 870b57cec5SDimitry Andric /// \param __a 880b57cec5SDimitry Andric /// A 128-bit vector of [4 x i32]. 890b57cec5SDimitry Andric /// \param __b 900b57cec5SDimitry Andric /// A 128-bit vector of [4 x i32] 910b57cec5SDimitry Andric /// \param __m0 920b57cec5SDimitry Andric /// A pointer point to 8-bit mask 930b57cec5SDimitry Andric /// \param __m1 940b57cec5SDimitry Andric /// A pointer point to 8-bit mask 950b57cec5SDimitry Andric static __inline__ void __DEFAULT_FN_ATTRS128 960b57cec5SDimitry Andric _mm_2intersect_epi32(__m128i __a, __m128i __b, __mmask8 *__m0, __mmask8 *__m1) { 970b57cec5SDimitry Andric __builtin_ia32_vp2intersect_d_128((__v4si)__a, (__v4si)__b, __m0, __m1); 980b57cec5SDimitry Andric } 990b57cec5SDimitry Andric 1000b57cec5SDimitry Andric /// Store, in an even/odd pair of mask registers, the indicators of the 1010b57cec5SDimitry Andric /// locations of value matches between quadwords in operands __a and __b. 1020b57cec5SDimitry Andric /// 1030b57cec5SDimitry Andric /// \headerfile <x86intrin.h> 1040b57cec5SDimitry Andric /// 1050b57cec5SDimitry Andric /// This intrinsic corresponds to the <c> VP2INTERSECTQ </c> instruction. 1060b57cec5SDimitry Andric /// 1070b57cec5SDimitry Andric /// \param __a 1080b57cec5SDimitry Andric /// A 128-bit vector of [2 x i64]. 1090b57cec5SDimitry Andric /// \param __b 1100b57cec5SDimitry Andric /// A 128-bit vector of [2 x i64] 1110b57cec5SDimitry Andric /// \param __m0 1120b57cec5SDimitry Andric /// A pointer point to 8-bit mask 1130b57cec5SDimitry Andric /// \param __m1 1140b57cec5SDimitry Andric /// A pointer point to 8-bit mask 1150b57cec5SDimitry Andric static __inline__ void __DEFAULT_FN_ATTRS128 1160b57cec5SDimitry Andric _mm_2intersect_epi64(__m128i __a, __m128i __b, __mmask8 *__m0, __mmask8 *__m1) { 1170b57cec5SDimitry Andric __builtin_ia32_vp2intersect_q_128((__v2di)__a, (__v2di)__b, __m0, __m1); 1180b57cec5SDimitry Andric } 1190b57cec5SDimitry Andric 1200b57cec5SDimitry Andric #undef __DEFAULT_FN_ATTRS128 1210b57cec5SDimitry Andric #undef __DEFAULT_FN_ATTRS256 1220b57cec5SDimitry Andric 1230b57cec5SDimitry Andric #endif 124