diff options
Diffstat (limited to 'lib/Target/X86/X86IntrinsicsInfo.h')
-rw-r--r-- | lib/Target/X86/X86IntrinsicsInfo.h | 781 |
1 files changed, 350 insertions, 431 deletions
diff --git a/lib/Target/X86/X86IntrinsicsInfo.h b/lib/Target/X86/X86IntrinsicsInfo.h index 151e1b9136c4..40141d894629 100644 --- a/lib/Target/X86/X86IntrinsicsInfo.h +++ b/lib/Target/X86/X86IntrinsicsInfo.h @@ -1,9 +1,8 @@ //===-- X86IntrinsicsInfo.h - X86 Intrinsics ------------*- C++ -*-===// // -// The LLVM Compiler Infrastructure -// -// This file is distributed under the University of Illinois Open Source -// License. See LICENSE.TXT for details. +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception // //===----------------------------------------------------------------------===// // @@ -20,21 +19,22 @@ namespace llvm { enum IntrinsicType : uint16_t { + CVTNEPS2BF16_MASK, GATHER, SCATTER, PREFETCH, RDSEED, RDRAND, RDPMC, RDTSC, XTEST, XGETBV, ADX, FPCLASSS, INTR_TYPE_1OP, INTR_TYPE_2OP, INTR_TYPE_3OP, INTR_TYPE_4OP, INTR_TYPE_3OP_IMM8, - CMP_MASK_CC,CMP_MASK_SCALAR_CC, VSHIFT, COMI, COMI_RM, - CVTPD2PS, CVTPD2PS_MASK, CVTPD2PS_RND_MASK, - INTR_TYPE_1OP_MASK, INTR_TYPE_1OP_MASK_RM, - INTR_TYPE_2OP_MASK, INTR_TYPE_2OP_MASK_RM, - INTR_TYPE_3OP_MASK, - IFMA_OP, VPERM_2OP, INTR_TYPE_SCALAR_MASK, - INTR_TYPE_SCALAR_MASK_RM, INTR_TYPE_3OP_SCALAR_MASK, + CMP_MASK_CC,CMP_MASK_SCALAR_CC, VSHIFT, COMI, COMI_RM, BLENDV, + CVTPD2PS_MASK, + INTR_TYPE_1OP_SAE, INTR_TYPE_2OP_SAE, + INTR_TYPE_1OP_MASK_SAE, INTR_TYPE_2OP_MASK_SAE, INTR_TYPE_3OP_MASK_SAE, + INTR_TYPE_1OP_MASK, INTR_TYPE_2OP_MASK, + IFMA_OP, VPERM_2OP, INTR_TYPE_SCALAR_MASK, INTR_TYPE_SCALAR_MASK_SAE, + INTR_TYPE_SCALAR_MASK_RND, + INTR_TYPE_3OP_SCALAR_MASK_SAE, COMPRESS_EXPAND_IN_REG, - TRUNCATE_TO_REG, CVTPS2PH_MASK, CVTPD2I_MASK, + TRUNCATE_TO_REG, CVTPS2PH_MASK, CVTPD2DQ_MASK, CVTQQ2PS_MASK, TRUNCATE_TO_MEM_VI8, TRUNCATE_TO_MEM_VI16, TRUNCATE_TO_MEM_VI32, - FIXUPIMM, FIXUPIMM_MASKZ, FIXUPIMMS, - FIXUPIMMS_MASKZ, GATHER_AVX2, + FIXUPIMM, FIXUPIMM_MASKZ, GATHER_AVX2, ROUNDP, ROUNDS }; @@ -64,47 +64,47 @@ struct IntrinsicData { * the alphabetical order. */ static const IntrinsicData IntrinsicsWithChain[] = { - X86_INTRINSIC_DATA(avx2_gather_d_d, GATHER_AVX2, X86::VPGATHERDDrm, 0), - X86_INTRINSIC_DATA(avx2_gather_d_d_256, GATHER_AVX2, X86::VPGATHERDDYrm, 0), - X86_INTRINSIC_DATA(avx2_gather_d_pd, GATHER_AVX2, X86::VGATHERDPDrm, 0), - X86_INTRINSIC_DATA(avx2_gather_d_pd_256, GATHER_AVX2, X86::VGATHERDPDYrm, 0), - X86_INTRINSIC_DATA(avx2_gather_d_ps, GATHER_AVX2, X86::VGATHERDPSrm, 0), - X86_INTRINSIC_DATA(avx2_gather_d_ps_256, GATHER_AVX2, X86::VGATHERDPSYrm, 0), - X86_INTRINSIC_DATA(avx2_gather_d_q, GATHER_AVX2, X86::VPGATHERDQrm, 0), - X86_INTRINSIC_DATA(avx2_gather_d_q_256, GATHER_AVX2, X86::VPGATHERDQYrm, 0), - X86_INTRINSIC_DATA(avx2_gather_q_d, GATHER_AVX2, X86::VPGATHERQDrm, 0), - X86_INTRINSIC_DATA(avx2_gather_q_d_256, GATHER_AVX2, X86::VPGATHERQDYrm, 0), - X86_INTRINSIC_DATA(avx2_gather_q_pd, GATHER_AVX2, X86::VGATHERQPDrm, 0), - X86_INTRINSIC_DATA(avx2_gather_q_pd_256, GATHER_AVX2, X86::VGATHERQPDYrm, 0), - X86_INTRINSIC_DATA(avx2_gather_q_ps, GATHER_AVX2, X86::VGATHERQPSrm, 0), - X86_INTRINSIC_DATA(avx2_gather_q_ps_256, GATHER_AVX2, X86::VGATHERQPSYrm, 0), - X86_INTRINSIC_DATA(avx2_gather_q_q, GATHER_AVX2, X86::VPGATHERQQrm, 0), - X86_INTRINSIC_DATA(avx2_gather_q_q_256, GATHER_AVX2, X86::VPGATHERQQYrm, 0), + X86_INTRINSIC_DATA(avx2_gather_d_d, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_d_d_256, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_d_pd, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_d_pd_256, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_d_ps, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_d_ps_256, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_d_q, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_d_q_256, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_q_d, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_q_d_256, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_q_pd, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_q_pd_256, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_q_ps, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_q_ps_256, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_q_q, GATHER_AVX2, 0, 0), + X86_INTRINSIC_DATA(avx2_gather_q_q_256, GATHER_AVX2, 0, 0), - X86_INTRINSIC_DATA(avx512_gather_dpd_512, GATHER, X86::VGATHERDPDZrm, 0), - X86_INTRINSIC_DATA(avx512_gather_dpi_512, GATHER, X86::VPGATHERDDZrm, 0), - X86_INTRINSIC_DATA(avx512_gather_dpq_512, GATHER, X86::VPGATHERDQZrm, 0), - X86_INTRINSIC_DATA(avx512_gather_dps_512, GATHER, X86::VGATHERDPSZrm, 0), - X86_INTRINSIC_DATA(avx512_gather_qpd_512, GATHER, X86::VGATHERQPDZrm, 0), - X86_INTRINSIC_DATA(avx512_gather_qpi_512, GATHER, X86::VPGATHERQDZrm, 0), - X86_INTRINSIC_DATA(avx512_gather_qpq_512, GATHER, X86::VPGATHERQQZrm, 0), - X86_INTRINSIC_DATA(avx512_gather_qps_512, GATHER, X86::VGATHERQPSZrm, 0), - X86_INTRINSIC_DATA(avx512_gather3div2_df, GATHER, X86::VGATHERQPDZ128rm, 0), - X86_INTRINSIC_DATA(avx512_gather3div2_di, GATHER, X86::VPGATHERQQZ128rm, 0), - X86_INTRINSIC_DATA(avx512_gather3div4_df, GATHER, X86::VGATHERQPDZ256rm, 0), - X86_INTRINSIC_DATA(avx512_gather3div4_di, GATHER, X86::VPGATHERQQZ256rm, 0), - X86_INTRINSIC_DATA(avx512_gather3div4_sf, GATHER, X86::VGATHERQPSZ128rm, 0), - X86_INTRINSIC_DATA(avx512_gather3div4_si, GATHER, X86::VPGATHERQDZ128rm, 0), - X86_INTRINSIC_DATA(avx512_gather3div8_sf, GATHER, X86::VGATHERQPSZ256rm, 0), - X86_INTRINSIC_DATA(avx512_gather3div8_si, GATHER, X86::VPGATHERQDZ256rm, 0), - X86_INTRINSIC_DATA(avx512_gather3siv2_df, GATHER, X86::VGATHERDPDZ128rm, 0), - X86_INTRINSIC_DATA(avx512_gather3siv2_di, GATHER, X86::VPGATHERDQZ128rm, 0), - X86_INTRINSIC_DATA(avx512_gather3siv4_df, GATHER, X86::VGATHERDPDZ256rm, 0), - X86_INTRINSIC_DATA(avx512_gather3siv4_di, GATHER, X86::VPGATHERDQZ256rm, 0), - X86_INTRINSIC_DATA(avx512_gather3siv4_sf, GATHER, X86::VGATHERDPSZ128rm, 0), - X86_INTRINSIC_DATA(avx512_gather3siv4_si, GATHER, X86::VPGATHERDDZ128rm, 0), - X86_INTRINSIC_DATA(avx512_gather3siv8_sf, GATHER, X86::VGATHERDPSZ256rm, 0), - X86_INTRINSIC_DATA(avx512_gather3siv8_si, GATHER, X86::VPGATHERDDZ256rm, 0), + X86_INTRINSIC_DATA(avx512_gather_dpd_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather_dpi_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather_dpq_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather_dps_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather_qpd_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather_qpi_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather_qpq_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather_qps_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3div2_df, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3div2_di, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3div4_df, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3div4_di, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3div4_sf, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3div4_si, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3div8_sf, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3div8_si, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3siv2_df, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3siv2_di, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3siv4_df, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3siv4_di, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3siv4_sf, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3siv4_si, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3siv8_sf, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_gather3siv8_si, GATHER, 0, 0), X86_INTRINSIC_DATA(avx512_gatherpf_dpd_512, PREFETCH, X86::VGATHERPF0DPDm, X86::VGATHERPF1DPDm), @@ -115,30 +115,30 @@ static const IntrinsicData IntrinsicsWithChain[] = { X86_INTRINSIC_DATA(avx512_gatherpf_qps_512, PREFETCH, X86::VGATHERPF0QPSm, X86::VGATHERPF1QPSm), - X86_INTRINSIC_DATA(avx512_mask_gather_dpd_512, GATHER, X86::VGATHERDPDZrm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather_dpi_512, GATHER, X86::VPGATHERDDZrm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather_dpq_512, GATHER, X86::VPGATHERDQZrm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather_dps_512, GATHER, X86::VGATHERDPSZrm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather_qpd_512, GATHER, X86::VGATHERQPDZrm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather_qpi_512, GATHER, X86::VPGATHERQDZrm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather_qpq_512, GATHER, X86::VPGATHERQQZrm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather_qps_512, GATHER, X86::VGATHERQPSZrm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3div2_df, GATHER, X86::VGATHERQPDZ128rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3div2_di, GATHER, X86::VPGATHERQQZ128rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3div4_df, GATHER, X86::VGATHERQPDZ256rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3div4_di, GATHER, X86::VPGATHERQQZ256rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3div4_sf, GATHER, X86::VGATHERQPSZ128rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3div4_si, GATHER, X86::VPGATHERQDZ128rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3div8_sf, GATHER, X86::VGATHERQPSZ256rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3div8_si, GATHER, X86::VPGATHERQDZ256rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3siv2_df, GATHER, X86::VGATHERDPDZ128rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3siv2_di, GATHER, X86::VPGATHERDQZ128rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3siv4_df, GATHER, X86::VGATHERDPDZ256rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3siv4_di, GATHER, X86::VPGATHERDQZ256rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3siv4_sf, GATHER, X86::VGATHERDPSZ128rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3siv4_si, GATHER, X86::VPGATHERDDZ128rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3siv8_sf, GATHER, X86::VGATHERDPSZ256rm, 0), - X86_INTRINSIC_DATA(avx512_mask_gather3siv8_si, GATHER, X86::VPGATHERDDZ256rm, 0), + X86_INTRINSIC_DATA(avx512_mask_gather_dpd_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather_dpi_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather_dpq_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather_dps_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather_qpd_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather_qpi_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather_qpq_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather_qps_512, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3div2_df, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3div2_di, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3div4_df, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3div4_di, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3div4_sf, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3div4_si, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3div8_sf, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3div8_si, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3siv2_df, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3siv2_di, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3siv4_df, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3siv4_di, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3siv4_sf, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3siv4_si, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3siv8_sf, GATHER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_gather3siv8_si, GATHER, 0, 0), X86_INTRINSIC_DATA(avx512_mask_pmov_db_mem_128, TRUNCATE_TO_MEM_VI8, X86ISD::VTRUNC, 0), @@ -249,47 +249,47 @@ static const IntrinsicData IntrinsicsWithChain[] = { X86_INTRINSIC_DATA(avx512_mask_pmovus_wb_mem_512, TRUNCATE_TO_MEM_VI8, X86ISD::VTRUNCUS, 0), - X86_INTRINSIC_DATA(avx512_mask_scatter_dpd_512, SCATTER, X86::VSCATTERDPDZmr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatter_dpi_512, SCATTER, X86::VPSCATTERDDZmr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatter_dpq_512, SCATTER, X86::VPSCATTERDQZmr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatter_dps_512, SCATTER, X86::VSCATTERDPSZmr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatter_qpd_512, SCATTER, X86::VSCATTERQPDZmr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatter_qpi_512, SCATTER, X86::VPSCATTERQDZmr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatter_qpq_512, SCATTER, X86::VPSCATTERQQZmr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatter_qps_512, SCATTER, X86::VSCATTERQPSZmr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatterdiv2_df, SCATTER, X86::VSCATTERQPDZ128mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatterdiv2_di, SCATTER, X86::VPSCATTERQQZ128mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatterdiv4_df, SCATTER, X86::VSCATTERQPDZ256mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatterdiv4_di, SCATTER, X86::VPSCATTERQQZ256mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatterdiv4_sf, SCATTER, X86::VSCATTERQPSZ128mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatterdiv4_si, SCATTER, X86::VPSCATTERQDZ128mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatterdiv8_sf, SCATTER, X86::VSCATTERQPSZ256mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scatterdiv8_si, SCATTER, X86::VPSCATTERQDZ256mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scattersiv2_df, SCATTER, X86::VSCATTERDPDZ128mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scattersiv2_di, SCATTER, X86::VPSCATTERDQZ128mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scattersiv4_df, SCATTER, X86::VSCATTERDPDZ256mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scattersiv4_di, SCATTER, X86::VPSCATTERDQZ256mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scattersiv4_sf, SCATTER, X86::VSCATTERDPSZ128mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scattersiv4_si, SCATTER, X86::VPSCATTERDDZ128mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scattersiv8_sf, SCATTER, X86::VSCATTERDPSZ256mr, 0), - X86_INTRINSIC_DATA(avx512_mask_scattersiv8_si, SCATTER, X86::VPSCATTERDDZ256mr, 0), + X86_INTRINSIC_DATA(avx512_mask_scatter_dpd_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatter_dpi_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatter_dpq_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatter_dps_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatter_qpd_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatter_qpi_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatter_qpq_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatter_qps_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatterdiv2_df, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatterdiv2_di, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatterdiv4_df, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatterdiv4_di, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatterdiv4_sf, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatterdiv4_si, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatterdiv8_sf, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scatterdiv8_si, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scattersiv2_df, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scattersiv2_di, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scattersiv4_df, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scattersiv4_di, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scattersiv4_sf, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scattersiv4_si, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scattersiv8_sf, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_mask_scattersiv8_si, SCATTER, 0, 0), - X86_INTRINSIC_DATA(avx512_scatter_dpd_512, SCATTER, X86::VSCATTERDPDZmr, 0), - X86_INTRINSIC_DATA(avx512_scatter_dpi_512, SCATTER, X86::VPSCATTERDDZmr, 0), - X86_INTRINSIC_DATA(avx512_scatter_dpq_512, SCATTER, X86::VPSCATTERDQZmr, 0), - X86_INTRINSIC_DATA(avx512_scatter_dps_512, SCATTER, X86::VSCATTERDPSZmr, 0), - X86_INTRINSIC_DATA(avx512_scatter_qpd_512, SCATTER, X86::VSCATTERQPDZmr, 0), - X86_INTRINSIC_DATA(avx512_scatter_qpi_512, SCATTER, X86::VPSCATTERQDZmr, 0), - X86_INTRINSIC_DATA(avx512_scatter_qpq_512, SCATTER, X86::VPSCATTERQQZmr, 0), - X86_INTRINSIC_DATA(avx512_scatter_qps_512, SCATTER, X86::VSCATTERQPSZmr, 0), - X86_INTRINSIC_DATA(avx512_scatterdiv2_df, SCATTER, X86::VSCATTERQPDZ128mr, 0), - X86_INTRINSIC_DATA(avx512_scatterdiv2_di, SCATTER, X86::VPSCATTERQQZ128mr, 0), - X86_INTRINSIC_DATA(avx512_scatterdiv4_df, SCATTER, X86::VSCATTERQPDZ256mr, 0), - X86_INTRINSIC_DATA(avx512_scatterdiv4_di, SCATTER, X86::VPSCATTERQQZ256mr, 0), - X86_INTRINSIC_DATA(avx512_scatterdiv4_sf, SCATTER, X86::VSCATTERQPSZ128mr, 0), - X86_INTRINSIC_DATA(avx512_scatterdiv4_si, SCATTER, X86::VPSCATTERQDZ128mr, 0), - X86_INTRINSIC_DATA(avx512_scatterdiv8_sf, SCATTER, X86::VSCATTERQPSZ256mr, 0), - X86_INTRINSIC_DATA(avx512_scatterdiv8_si, SCATTER, X86::VPSCATTERQDZ256mr, 0), + X86_INTRINSIC_DATA(avx512_scatter_dpd_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatter_dpi_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatter_dpq_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatter_dps_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatter_qpd_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatter_qpi_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatter_qpq_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatter_qps_512, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatterdiv2_df, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatterdiv2_di, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatterdiv4_df, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatterdiv4_di, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatterdiv4_sf, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatterdiv4_si, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatterdiv8_sf, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scatterdiv8_si, SCATTER, 0, 0), X86_INTRINSIC_DATA(avx512_scatterpf_dpd_512, PREFETCH, X86::VSCATTERPF0DPDm, X86::VSCATTERPF1DPDm), X86_INTRINSIC_DATA(avx512_scatterpf_dps_512, PREFETCH, X86::VSCATTERPF0DPSm, @@ -298,24 +298,24 @@ static const IntrinsicData IntrinsicsWithChain[] = { X86::VSCATTERPF1QPDm), X86_INTRINSIC_DATA(avx512_scatterpf_qps_512, PREFETCH, X86::VSCATTERPF0QPSm, X86::VSCATTERPF1QPSm), - X86_INTRINSIC_DATA(avx512_scattersiv2_df, SCATTER, X86::VSCATTERDPDZ128mr, 0), - X86_INTRINSIC_DATA(avx512_scattersiv2_di, SCATTER, X86::VPSCATTERDQZ128mr, 0), - X86_INTRINSIC_DATA(avx512_scattersiv4_df, SCATTER, X86::VSCATTERDPDZ256mr, 0), - X86_INTRINSIC_DATA(avx512_scattersiv4_di, SCATTER, X86::VPSCATTERDQZ256mr, 0), - X86_INTRINSIC_DATA(avx512_scattersiv4_sf, SCATTER, X86::VSCATTERDPSZ128mr, 0), - X86_INTRINSIC_DATA(avx512_scattersiv4_si, SCATTER, X86::VPSCATTERDDZ128mr, 0), - X86_INTRINSIC_DATA(avx512_scattersiv8_sf, SCATTER, X86::VSCATTERDPSZ256mr, 0), - X86_INTRINSIC_DATA(avx512_scattersiv8_si, SCATTER, X86::VPSCATTERDDZ256mr, 0), - X86_INTRINSIC_DATA(rdpmc, RDPMC, X86ISD::RDPMC_DAG, 0), + X86_INTRINSIC_DATA(avx512_scattersiv2_df, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scattersiv2_di, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scattersiv4_df, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scattersiv4_di, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scattersiv4_sf, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scattersiv4_si, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scattersiv8_sf, SCATTER, 0, 0), + X86_INTRINSIC_DATA(avx512_scattersiv8_si, SCATTER, 0, 0), + X86_INTRINSIC_DATA(rdpmc, RDPMC, X86::RDPMC, 0), X86_INTRINSIC_DATA(rdrand_16, RDRAND, X86ISD::RDRAND, 0), X86_INTRINSIC_DATA(rdrand_32, RDRAND, X86ISD::RDRAND, 0), X86_INTRINSIC_DATA(rdrand_64, RDRAND, X86ISD::RDRAND, 0), X86_INTRINSIC_DATA(rdseed_16, RDSEED, X86ISD::RDSEED, 0), X86_INTRINSIC_DATA(rdseed_32, RDSEED, X86ISD::RDSEED, 0), X86_INTRINSIC_DATA(rdseed_64, RDSEED, X86ISD::RDSEED, 0), - X86_INTRINSIC_DATA(rdtsc, RDTSC, X86ISD::RDTSC_DAG, 0), - X86_INTRINSIC_DATA(rdtscp, RDTSC, X86ISD::RDTSCP_DAG, 0), - X86_INTRINSIC_DATA(xgetbv, XGETBV, X86::XGETBV, 0), + X86_INTRINSIC_DATA(rdtsc, RDTSC, X86::RDTSC, 0), + X86_INTRINSIC_DATA(rdtscp, RDTSC, X86::RDTSCP, 0), + X86_INTRINSIC_DATA(xgetbv, XGETBV, X86::XGETBV, 0), X86_INTRINSIC_DATA(xtest, XTEST, X86ISD::XTEST, 0), }; @@ -340,9 +340,11 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(addcarry_64, ADX, X86ISD::ADC, X86ISD::ADD), X86_INTRINSIC_DATA(avx_addsub_pd_256, INTR_TYPE_2OP, X86ISD::ADDSUB, 0), X86_INTRINSIC_DATA(avx_addsub_ps_256, INTR_TYPE_2OP, X86ISD::ADDSUB, 0), + X86_INTRINSIC_DATA(avx_blendv_pd_256, BLENDV, X86ISD::BLENDV, 0), + X86_INTRINSIC_DATA(avx_blendv_ps_256, BLENDV, X86ISD::BLENDV, 0), X86_INTRINSIC_DATA(avx_cmp_pd_256, INTR_TYPE_3OP, X86ISD::CMPP, 0), X86_INTRINSIC_DATA(avx_cmp_ps_256, INTR_TYPE_3OP, X86ISD::CMPP, 0), - X86_INTRINSIC_DATA(avx_cvt_pd2_ps_256,CVTPD2PS, ISD::FP_ROUND, 0), + X86_INTRINSIC_DATA(avx_cvt_pd2_ps_256,INTR_TYPE_1OP, X86ISD::VFPROUND, 0), X86_INTRINSIC_DATA(avx_cvt_pd2dq_256, INTR_TYPE_1OP, X86ISD::CVTP2SI, 0), X86_INTRINSIC_DATA(avx_cvt_ps2dq_256, INTR_TYPE_1OP, X86ISD::CVTP2SI, 0), X86_INTRINSIC_DATA(avx_cvtt_pd2dq_256,INTR_TYPE_1OP, X86ISD::CVTTP2SI, 0), @@ -369,6 +371,9 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx2_packsswb, INTR_TYPE_2OP, X86ISD::PACKSS, 0), X86_INTRINSIC_DATA(avx2_packusdw, INTR_TYPE_2OP, X86ISD::PACKUS, 0), X86_INTRINSIC_DATA(avx2_packuswb, INTR_TYPE_2OP, X86ISD::PACKUS, 0), + X86_INTRINSIC_DATA(avx2_pavg_b, INTR_TYPE_2OP, X86ISD::AVG, 0), + X86_INTRINSIC_DATA(avx2_pavg_w, INTR_TYPE_2OP, X86ISD::AVG, 0), + X86_INTRINSIC_DATA(avx2_pblendvb, BLENDV, X86ISD::BLENDV, 0), X86_INTRINSIC_DATA(avx2_permd, VPERM_2OP, X86ISD::VPERMV, 0), X86_INTRINSIC_DATA(avx2_permps, VPERM_2OP, X86ISD::VPERMV, 0), X86_INTRINSIC_DATA(avx2_phadd_d, INTR_TYPE_2OP, X86ISD::HADD, 0), @@ -389,10 +394,10 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx2_pslli_d, VSHIFT, X86ISD::VSHLI, 0), X86_INTRINSIC_DATA(avx2_pslli_q, VSHIFT, X86ISD::VSHLI, 0), X86_INTRINSIC_DATA(avx2_pslli_w, VSHIFT, X86ISD::VSHLI, 0), - X86_INTRINSIC_DATA(avx2_psllv_d, INTR_TYPE_2OP, ISD::SHL, 0), - X86_INTRINSIC_DATA(avx2_psllv_d_256, INTR_TYPE_2OP, ISD::SHL, 0), - X86_INTRINSIC_DATA(avx2_psllv_q, INTR_TYPE_2OP, ISD::SHL, 0), - X86_INTRINSIC_DATA(avx2_psllv_q_256, INTR_TYPE_2OP, ISD::SHL, 0), + X86_INTRINSIC_DATA(avx2_psllv_d, INTR_TYPE_2OP, X86ISD::VSHLV, 0), + X86_INTRINSIC_DATA(avx2_psllv_d_256, INTR_TYPE_2OP, X86ISD::VSHLV, 0), + X86_INTRINSIC_DATA(avx2_psllv_q, INTR_TYPE_2OP, X86ISD::VSHLV, 0), + X86_INTRINSIC_DATA(avx2_psllv_q_256, INTR_TYPE_2OP, X86ISD::VSHLV, 0), X86_INTRINSIC_DATA(avx2_psra_d, INTR_TYPE_2OP, X86ISD::VSRA, 0), X86_INTRINSIC_DATA(avx2_psra_w, INTR_TYPE_2OP, X86ISD::VSRA, 0), X86_INTRINSIC_DATA(avx2_psrai_d, VSHIFT, X86ISD::VSRAI, 0), @@ -405,39 +410,45 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx2_psrli_d, VSHIFT, X86ISD::VSRLI, 0), X86_INTRINSIC_DATA(avx2_psrli_q, VSHIFT, X86ISD::VSRLI, 0), X86_INTRINSIC_DATA(avx2_psrli_w, VSHIFT, X86ISD::VSRLI, 0), - X86_INTRINSIC_DATA(avx2_psrlv_d, INTR_TYPE_2OP, ISD::SRL, 0), - X86_INTRINSIC_DATA(avx2_psrlv_d_256, INTR_TYPE_2OP, ISD::SRL, 0), - X86_INTRINSIC_DATA(avx2_psrlv_q, INTR_TYPE_2OP, ISD::SRL, 0), - X86_INTRINSIC_DATA(avx2_psrlv_q_256, INTR_TYPE_2OP, ISD::SRL, 0), + X86_INTRINSIC_DATA(avx2_psrlv_d, INTR_TYPE_2OP, X86ISD::VSRLV, 0), + X86_INTRINSIC_DATA(avx2_psrlv_d_256, INTR_TYPE_2OP, X86ISD::VSRLV, 0), + X86_INTRINSIC_DATA(avx2_psrlv_q, INTR_TYPE_2OP, X86ISD::VSRLV, 0), + X86_INTRINSIC_DATA(avx2_psrlv_q_256, INTR_TYPE_2OP, X86ISD::VSRLV, 0), X86_INTRINSIC_DATA(avx512_add_pd_512, INTR_TYPE_2OP, ISD::FADD, X86ISD::FADD_RND), X86_INTRINSIC_DATA(avx512_add_ps_512, INTR_TYPE_2OP, ISD::FADD, X86ISD::FADD_RND), X86_INTRINSIC_DATA(avx512_cmp_pd_128, CMP_MASK_CC, X86ISD::CMPM, 0), X86_INTRINSIC_DATA(avx512_cmp_pd_256, CMP_MASK_CC, X86ISD::CMPM, 0), - X86_INTRINSIC_DATA(avx512_cmp_pd_512, CMP_MASK_CC, X86ISD::CMPM, X86ISD::CMPM_RND), + X86_INTRINSIC_DATA(avx512_cmp_pd_512, CMP_MASK_CC, X86ISD::CMPM, X86ISD::CMPM_SAE), X86_INTRINSIC_DATA(avx512_cmp_ps_128, CMP_MASK_CC, X86ISD::CMPM, 0), X86_INTRINSIC_DATA(avx512_cmp_ps_256, CMP_MASK_CC, X86ISD::CMPM, 0), - X86_INTRINSIC_DATA(avx512_cmp_ps_512, CMP_MASK_CC, X86ISD::CMPM, X86ISD::CMPM_RND), - X86_INTRINSIC_DATA(avx512_cvtsi2sd64, INTR_TYPE_3OP, X86ISD::SCALAR_SINT_TO_FP_RND, 0), - X86_INTRINSIC_DATA(avx512_cvtsi2ss32, INTR_TYPE_3OP, X86ISD::SCALAR_SINT_TO_FP_RND, 0), - X86_INTRINSIC_DATA(avx512_cvtsi2ss64, INTR_TYPE_3OP, X86ISD::SCALAR_SINT_TO_FP_RND, 0), - X86_INTRINSIC_DATA(avx512_cvttsd2si, INTR_TYPE_1OP, X86ISD::CVTTS2SI, X86ISD::CVTTS2SI_RND), - X86_INTRINSIC_DATA(avx512_cvttsd2si64, INTR_TYPE_1OP, X86ISD::CVTTS2SI, X86ISD::CVTTS2SI_RND), - X86_INTRINSIC_DATA(avx512_cvttsd2usi, INTR_TYPE_1OP, X86ISD::CVTTS2UI, X86ISD::CVTTS2UI_RND), - X86_INTRINSIC_DATA(avx512_cvttsd2usi64, INTR_TYPE_1OP, X86ISD::CVTTS2UI, X86ISD::CVTTS2UI_RND), - X86_INTRINSIC_DATA(avx512_cvttss2si, INTR_TYPE_1OP, X86ISD::CVTTS2SI, X86ISD::CVTTS2SI_RND), - X86_INTRINSIC_DATA(avx512_cvttss2si64, INTR_TYPE_1OP, X86ISD::CVTTS2SI, X86ISD::CVTTS2SI_RND), - X86_INTRINSIC_DATA(avx512_cvttss2usi, INTR_TYPE_1OP, X86ISD::CVTTS2UI, X86ISD::CVTTS2UI_RND), - X86_INTRINSIC_DATA(avx512_cvttss2usi64, INTR_TYPE_1OP, X86ISD::CVTTS2UI, X86ISD::CVTTS2UI_RND), - X86_INTRINSIC_DATA(avx512_cvtusi2ss, INTR_TYPE_3OP, X86ISD::SCALAR_UINT_TO_FP_RND, 0), - X86_INTRINSIC_DATA(avx512_cvtusi642sd, INTR_TYPE_3OP, X86ISD::SCALAR_UINT_TO_FP_RND, 0), - X86_INTRINSIC_DATA(avx512_cvtusi642ss, INTR_TYPE_3OP, X86ISD::SCALAR_UINT_TO_FP_RND, 0), + X86_INTRINSIC_DATA(avx512_cmp_ps_512, CMP_MASK_CC, X86ISD::CMPM, X86ISD::CMPM_SAE), + X86_INTRINSIC_DATA(avx512_conflict_d_128, INTR_TYPE_1OP, X86ISD::CONFLICT, 0), + X86_INTRINSIC_DATA(avx512_conflict_d_256, INTR_TYPE_1OP, X86ISD::CONFLICT, 0), + X86_INTRINSIC_DATA(avx512_conflict_d_512, INTR_TYPE_1OP, X86ISD::CONFLICT, 0), + X86_INTRINSIC_DATA(avx512_conflict_q_128, INTR_TYPE_1OP, X86ISD::CONFLICT, 0), + X86_INTRINSIC_DATA(avx512_conflict_q_256, INTR_TYPE_1OP, X86ISD::CONFLICT, 0), + X86_INTRINSIC_DATA(avx512_conflict_q_512, INTR_TYPE_1OP, X86ISD::CONFLICT, 0), + X86_INTRINSIC_DATA(avx512_cvtsi2sd64, INTR_TYPE_2OP, X86ISD::SCALAR_SINT_TO_FP, X86ISD::SCALAR_SINT_TO_FP_RND), + X86_INTRINSIC_DATA(avx512_cvtsi2ss32, INTR_TYPE_2OP, X86ISD::SCALAR_SINT_TO_FP, X86ISD::SCALAR_SINT_TO_FP_RND), + X86_INTRINSIC_DATA(avx512_cvtsi2ss64, INTR_TYPE_2OP, X86ISD::SCALAR_SINT_TO_FP, X86ISD::SCALAR_SINT_TO_FP_RND), + X86_INTRINSIC_DATA(avx512_cvttsd2si, INTR_TYPE_1OP_SAE, X86ISD::CVTTS2SI, X86ISD::CVTTS2SI_SAE), + X86_INTRINSIC_DATA(avx512_cvttsd2si64, INTR_TYPE_1OP_SAE, X86ISD::CVTTS2SI, X86ISD::CVTTS2SI_SAE), + X86_INTRINSIC_DATA(avx512_cvttsd2usi, INTR_TYPE_1OP_SAE, X86ISD::CVTTS2UI, X86ISD::CVTTS2UI_SAE), + X86_INTRINSIC_DATA(avx512_cvttsd2usi64, INTR_TYPE_1OP_SAE, X86ISD::CVTTS2UI, X86ISD::CVTTS2UI_SAE), + X86_INTRINSIC_DATA(avx512_cvttss2si, INTR_TYPE_1OP_SAE, X86ISD::CVTTS2SI, X86ISD::CVTTS2SI_SAE), + X86_INTRINSIC_DATA(avx512_cvttss2si64, INTR_TYPE_1OP_SAE, X86ISD::CVTTS2SI, X86ISD::CVTTS2SI_SAE), + X86_INTRINSIC_DATA(avx512_cvttss2usi, INTR_TYPE_1OP_SAE, X86ISD::CVTTS2UI, X86ISD::CVTTS2UI_SAE), + X86_INTRINSIC_DATA(avx512_cvttss2usi64, INTR_TYPE_1OP_SAE, X86ISD::CVTTS2UI, X86ISD::CVTTS2UI_SAE), + X86_INTRINSIC_DATA(avx512_cvtusi2ss, INTR_TYPE_2OP, X86ISD::SCALAR_UINT_TO_FP, X86ISD::SCALAR_UINT_TO_FP_RND), + X86_INTRINSIC_DATA(avx512_cvtusi642sd, INTR_TYPE_2OP, X86ISD::SCALAR_UINT_TO_FP, X86ISD::SCALAR_UINT_TO_FP_RND), + X86_INTRINSIC_DATA(avx512_cvtusi642ss, INTR_TYPE_2OP, X86ISD::SCALAR_UINT_TO_FP, X86ISD::SCALAR_UINT_TO_FP_RND), X86_INTRINSIC_DATA(avx512_dbpsadbw_128, INTR_TYPE_3OP_IMM8, X86ISD::DBPSADBW, 0), X86_INTRINSIC_DATA(avx512_dbpsadbw_256, INTR_TYPE_3OP_IMM8, X86ISD::DBPSADBW, 0), X86_INTRINSIC_DATA(avx512_dbpsadbw_512, INTR_TYPE_3OP_IMM8, X86ISD::DBPSADBW, 0), X86_INTRINSIC_DATA(avx512_div_pd_512, INTR_TYPE_2OP, ISD::FDIV, X86ISD::FDIV_RND), X86_INTRINSIC_DATA(avx512_div_ps_512, INTR_TYPE_2OP, ISD::FDIV, X86ISD::FDIV_RND), - X86_INTRINSIC_DATA(avx512_exp2_pd, INTR_TYPE_1OP_MASK_RM, X86ISD::EXP2, 0), - X86_INTRINSIC_DATA(avx512_exp2_ps, INTR_TYPE_1OP_MASK_RM, X86ISD::EXP2, 0), + X86_INTRINSIC_DATA(avx512_exp2_pd, INTR_TYPE_1OP_MASK_SAE, X86ISD::EXP2, X86ISD::EXP2_SAE), + X86_INTRINSIC_DATA(avx512_exp2_ps, INTR_TYPE_1OP_MASK_SAE, X86ISD::EXP2, X86ISD::EXP2_SAE), X86_INTRINSIC_DATA(avx512_fpclass_pd_128, INTR_TYPE_2OP, X86ISD::VFPCLASS, 0), X86_INTRINSIC_DATA(avx512_fpclass_pd_256, INTR_TYPE_2OP, X86ISD::VFPCLASS, 0), X86_INTRINSIC_DATA(avx512_fpclass_pd_512, INTR_TYPE_2OP, X86ISD::VFPCLASS, 0), @@ -448,80 +459,32 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx512_kadd_d, INTR_TYPE_2OP, X86ISD::KADD, 0), X86_INTRINSIC_DATA(avx512_kadd_q, INTR_TYPE_2OP, X86ISD::KADD, 0), X86_INTRINSIC_DATA(avx512_kadd_w, INTR_TYPE_2OP, X86ISD::KADD, 0), - X86_INTRINSIC_DATA(avx512_mask_add_sd_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FADDS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_add_ss_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FADDS_RND, 0), + X86_INTRINSIC_DATA(avx512_mask_add_sd_round, INTR_TYPE_SCALAR_MASK, + X86ISD::FADDS, X86ISD::FADDS_RND), + X86_INTRINSIC_DATA(avx512_mask_add_ss_round, INTR_TYPE_SCALAR_MASK, + X86ISD::FADDS, X86ISD::FADDS_RND), X86_INTRINSIC_DATA(avx512_mask_cmp_sd, CMP_MASK_SCALAR_CC, - X86ISD::FSETCCM, X86ISD::FSETCCM_RND), + X86ISD::FSETCCM, X86ISD::FSETCCM_SAE), X86_INTRINSIC_DATA(avx512_mask_cmp_ss, CMP_MASK_SCALAR_CC, - X86ISD::FSETCCM, X86ISD::FSETCCM_RND), + X86ISD::FSETCCM, X86ISD::FSETCCM_SAE), - X86_INTRINSIC_DATA(avx512_mask_compress_b_128, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_b_256, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_b_512, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_d_128, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_d_256, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_d_512, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_pd_128, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_pd_256, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_pd_512, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_ps_128, COMPRESS_EXPAND_IN_REG, + X86_INTRINSIC_DATA(avx512_mask_compress, COMPRESS_EXPAND_IN_REG, X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_ps_256, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_ps_512, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_q_128, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_q_256, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_q_512, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_w_128, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_w_256, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_compress_w_512, COMPRESS_EXPAND_IN_REG, - X86ISD::COMPRESS, 0), - X86_INTRINSIC_DATA(avx512_mask_conflict_d_128, INTR_TYPE_1OP_MASK, - X86ISD::CONFLICT, 0), - X86_INTRINSIC_DATA(avx512_mask_conflict_d_256, INTR_TYPE_1OP_MASK, - X86ISD::CONFLICT, 0), - X86_INTRINSIC_DATA(avx512_mask_conflict_d_512, INTR_TYPE_1OP_MASK, - X86ISD::CONFLICT, 0), - X86_INTRINSIC_DATA(avx512_mask_conflict_q_128, INTR_TYPE_1OP_MASK, - X86ISD::CONFLICT, 0), - X86_INTRINSIC_DATA(avx512_mask_conflict_q_256, INTR_TYPE_1OP_MASK, - X86ISD::CONFLICT, 0), - X86_INTRINSIC_DATA(avx512_mask_conflict_q_512, INTR_TYPE_1OP_MASK, - X86ISD::CONFLICT, 0), - X86_INTRINSIC_DATA(avx512_mask_cvtdq2ps_512, INTR_TYPE_1OP_MASK, - ISD::SINT_TO_FP, X86ISD::SINT_TO_FP_RND), //er - X86_INTRINSIC_DATA(avx512_mask_cvtpd2dq_128, CVTPD2I_MASK, + X86_INTRINSIC_DATA(avx512_mask_cvtpd2dq_128, CVTPD2DQ_MASK, X86ISD::CVTP2SI, X86ISD::MCVTP2SI), X86_INTRINSIC_DATA(avx512_mask_cvtpd2dq_512, INTR_TYPE_1OP_MASK, X86ISD::CVTP2SI, X86ISD::CVTP2SI_RND), X86_INTRINSIC_DATA(avx512_mask_cvtpd2ps, CVTPD2PS_MASK, X86ISD::VFPROUND, X86ISD::VMFPROUND), - X86_INTRINSIC_DATA(avx512_mask_cvtpd2ps_512, CVTPD2PS_RND_MASK, - ISD::FP_ROUND, X86ISD::VFPROUND_RND), + X86_INTRINSIC_DATA(avx512_mask_cvtpd2ps_512, INTR_TYPE_1OP_MASK, + X86ISD::VFPROUND, X86ISD::VFPROUND_RND), X86_INTRINSIC_DATA(avx512_mask_cvtpd2qq_128, INTR_TYPE_1OP_MASK, X86ISD::CVTP2SI, 0), X86_INTRINSIC_DATA(avx512_mask_cvtpd2qq_256, INTR_TYPE_1OP_MASK, X86ISD::CVTP2SI, 0), X86_INTRINSIC_DATA(avx512_mask_cvtpd2qq_512, INTR_TYPE_1OP_MASK, X86ISD::CVTP2SI, X86ISD::CVTP2SI_RND), - X86_INTRINSIC_DATA(avx512_mask_cvtpd2udq_128, CVTPD2I_MASK, + X86_INTRINSIC_DATA(avx512_mask_cvtpd2udq_128, CVTPD2DQ_MASK, X86ISD::CVTP2UI, X86ISD::MCVTP2UI), X86_INTRINSIC_DATA(avx512_mask_cvtpd2udq_256, INTR_TYPE_1OP_MASK, X86ISD::CVTP2UI, 0), @@ -539,8 +502,8 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86ISD::CVTP2SI, 0), X86_INTRINSIC_DATA(avx512_mask_cvtps2dq_512, INTR_TYPE_1OP_MASK, X86ISD::CVTP2SI, X86ISD::CVTP2SI_RND), - X86_INTRINSIC_DATA(avx512_mask_cvtps2pd_512, INTR_TYPE_1OP_MASK, - ISD::FP_EXTEND, X86ISD::VFPEXT_RND), + X86_INTRINSIC_DATA(avx512_mask_cvtps2pd_512, INTR_TYPE_1OP_MASK_SAE, + ISD::FP_EXTEND, X86ISD::VFPEXT_SAE), X86_INTRINSIC_DATA(avx512_mask_cvtps2qq_128, INTR_TYPE_1OP_MASK, X86ISD::CVTP2SI, 0), X86_INTRINSIC_DATA(avx512_mask_cvtps2qq_256, INTR_TYPE_1OP_MASK, @@ -559,164 +522,116 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86ISD::CVTP2UI, 0), X86_INTRINSIC_DATA(avx512_mask_cvtps2uqq_512, INTR_TYPE_1OP_MASK, X86ISD::CVTP2UI, X86ISD::CVTP2UI_RND), - X86_INTRINSIC_DATA(avx512_mask_cvtqq2pd_512, INTR_TYPE_1OP_MASK, - ISD::SINT_TO_FP, X86ISD::SINT_TO_FP_RND), - X86_INTRINSIC_DATA(avx512_mask_cvtqq2ps_128, INTR_TYPE_1OP_MASK, - X86ISD::CVTSI2P, 0), - X86_INTRINSIC_DATA(avx512_mask_cvtqq2ps_256, INTR_TYPE_1OP_MASK, - ISD::SINT_TO_FP, 0), - X86_INTRINSIC_DATA(avx512_mask_cvtqq2ps_512, INTR_TYPE_1OP_MASK, - ISD::SINT_TO_FP, X86ISD::SINT_TO_FP_RND), - X86_INTRINSIC_DATA(avx512_mask_cvtsd2ss_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::VFPROUNDS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_cvtss2sd_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::VFPEXTS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_cvttpd2dq_128, CVTPD2I_MASK, + X86_INTRINSIC_DATA(avx512_mask_cvtqq2ps_128, CVTQQ2PS_MASK, + X86ISD::CVTSI2P, X86ISD::MCVTSI2P), + X86_INTRINSIC_DATA(avx512_mask_cvtsd2ss_round, INTR_TYPE_SCALAR_MASK_RND, + X86ISD::VFPROUNDS, X86ISD::VFPROUNDS_RND), + X86_INTRINSIC_DATA(avx512_mask_cvtss2sd_round, INTR_TYPE_SCALAR_MASK_SAE, + X86ISD::VFPEXTS, X86ISD::VFPEXTS_SAE), + X86_INTRINSIC_DATA(avx512_mask_cvttpd2dq_128, CVTPD2DQ_MASK, X86ISD::CVTTP2SI, X86ISD::MCVTTP2SI), - X86_INTRINSIC_DATA(avx512_mask_cvttpd2dq_512, INTR_TYPE_1OP_MASK, - X86ISD::CVTTP2SI, X86ISD::CVTTP2SI_RND), + X86_INTRINSIC_DATA(avx512_mask_cvttpd2dq_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::CVTTP2SI, X86ISD::CVTTP2SI_SAE), X86_INTRINSIC_DATA(avx512_mask_cvttpd2qq_128, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2SI, 0), X86_INTRINSIC_DATA(avx512_mask_cvttpd2qq_256, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2SI, 0), - X86_INTRINSIC_DATA(avx512_mask_cvttpd2qq_512, INTR_TYPE_1OP_MASK, - X86ISD::CVTTP2SI, X86ISD::CVTTP2SI_RND), - X86_INTRINSIC_DATA(avx512_mask_cvttpd2udq_128, CVTPD2I_MASK, + X86_INTRINSIC_DATA(avx512_mask_cvttpd2qq_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::CVTTP2SI, X86ISD::CVTTP2SI_SAE), + X86_INTRINSIC_DATA(avx512_mask_cvttpd2udq_128, CVTPD2DQ_MASK, X86ISD::CVTTP2UI, X86ISD::MCVTTP2UI), X86_INTRINSIC_DATA(avx512_mask_cvttpd2udq_256, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2UI, 0), - X86_INTRINSIC_DATA(avx512_mask_cvttpd2udq_512, INTR_TYPE_1OP_MASK, - X86ISD::CVTTP2UI, X86ISD::CVTTP2UI_RND), + X86_INTRINSIC_DATA(avx512_mask_cvttpd2udq_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::CVTTP2UI, X86ISD::CVTTP2UI_SAE), X86_INTRINSIC_DATA(avx512_mask_cvttpd2uqq_128, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2UI, 0), X86_INTRINSIC_DATA(avx512_mask_cvttpd2uqq_256, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2UI, 0), - X86_INTRINSIC_DATA(avx512_mask_cvttpd2uqq_512, INTR_TYPE_1OP_MASK, - X86ISD::CVTTP2UI, X86ISD::CVTTP2UI_RND), - X86_INTRINSIC_DATA(avx512_mask_cvttps2dq_512, INTR_TYPE_1OP_MASK, - X86ISD::CVTTP2SI, X86ISD::CVTTP2SI_RND), + X86_INTRINSIC_DATA(avx512_mask_cvttpd2uqq_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::CVTTP2UI, X86ISD::CVTTP2UI_SAE), + X86_INTRINSIC_DATA(avx512_mask_cvttps2dq_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::CVTTP2SI, X86ISD::CVTTP2SI_SAE), X86_INTRINSIC_DATA(avx512_mask_cvttps2qq_128, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2SI, 0), X86_INTRINSIC_DATA(avx512_mask_cvttps2qq_256, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2SI, 0), - X86_INTRINSIC_DATA(avx512_mask_cvttps2qq_512, INTR_TYPE_1OP_MASK, - X86ISD::CVTTP2SI, X86ISD::CVTTP2SI_RND), + X86_INTRINSIC_DATA(avx512_mask_cvttps2qq_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::CVTTP2SI, X86ISD::CVTTP2SI_SAE), X86_INTRINSIC_DATA(avx512_mask_cvttps2udq_128, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2UI, 0), X86_INTRINSIC_DATA(avx512_mask_cvttps2udq_256, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2UI, 0), - X86_INTRINSIC_DATA(avx512_mask_cvttps2udq_512, INTR_TYPE_1OP_MASK, - X86ISD::CVTTP2UI, X86ISD::CVTTP2UI_RND), + X86_INTRINSIC_DATA(avx512_mask_cvttps2udq_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::CVTTP2UI, X86ISD::CVTTP2UI_SAE), X86_INTRINSIC_DATA(avx512_mask_cvttps2uqq_128, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2UI, 0), X86_INTRINSIC_DATA(avx512_mask_cvttps2uqq_256, INTR_TYPE_1OP_MASK, X86ISD::CVTTP2UI, 0), - X86_INTRINSIC_DATA(avx512_mask_cvttps2uqq_512, INTR_TYPE_1OP_MASK, - X86ISD::CVTTP2UI, X86ISD::CVTTP2UI_RND), - X86_INTRINSIC_DATA(avx512_mask_cvtudq2ps_512, INTR_TYPE_1OP_MASK, - ISD::UINT_TO_FP, X86ISD::UINT_TO_FP_RND), - X86_INTRINSIC_DATA(avx512_mask_cvtuqq2pd_512, INTR_TYPE_1OP_MASK, - ISD::UINT_TO_FP, X86ISD::UINT_TO_FP_RND), - X86_INTRINSIC_DATA(avx512_mask_cvtuqq2ps_128, INTR_TYPE_1OP_MASK, - X86ISD::CVTUI2P, 0), - X86_INTRINSIC_DATA(avx512_mask_cvtuqq2ps_256, INTR_TYPE_1OP_MASK, - ISD::UINT_TO_FP, 0), - X86_INTRINSIC_DATA(avx512_mask_cvtuqq2ps_512, INTR_TYPE_1OP_MASK, - ISD::UINT_TO_FP, X86ISD::UINT_TO_FP_RND), - X86_INTRINSIC_DATA(avx512_mask_div_sd_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FDIVS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_div_ss_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FDIVS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_b_128, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_b_256, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_b_512, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_d_128, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_d_256, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_d_512, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_pd_128, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_pd_256, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_pd_512, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_ps_128, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_ps_256, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_ps_512, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_q_128, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_q_256, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_q_512, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_w_128, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_w_256, COMPRESS_EXPAND_IN_REG, - X86ISD::EXPAND, 0), - X86_INTRINSIC_DATA(avx512_mask_expand_w_512, COMPRESS_EXPAND_IN_REG, + X86_INTRINSIC_DATA(avx512_mask_cvttps2uqq_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::CVTTP2UI, X86ISD::CVTTP2UI_SAE), + X86_INTRINSIC_DATA(avx512_mask_cvtuqq2ps_128, CVTQQ2PS_MASK, + X86ISD::CVTUI2P, X86ISD::MCVTUI2P), + X86_INTRINSIC_DATA(avx512_mask_div_sd_round, INTR_TYPE_SCALAR_MASK, + X86ISD::FDIVS, X86ISD::FDIVS_RND), + X86_INTRINSIC_DATA(avx512_mask_div_ss_round, INTR_TYPE_SCALAR_MASK, + X86ISD::FDIVS, X86ISD::FDIVS_RND), + X86_INTRINSIC_DATA(avx512_mask_expand, COMPRESS_EXPAND_IN_REG, X86ISD::EXPAND, 0), X86_INTRINSIC_DATA(avx512_mask_fixupimm_pd_128, FIXUPIMM, X86ISD::VFIXUPIMM, 0), X86_INTRINSIC_DATA(avx512_mask_fixupimm_pd_256, FIXUPIMM, X86ISD::VFIXUPIMM, 0), - X86_INTRINSIC_DATA(avx512_mask_fixupimm_pd_512, FIXUPIMM, X86ISD::VFIXUPIMM, 0), + X86_INTRINSIC_DATA(avx512_mask_fixupimm_pd_512, FIXUPIMM, X86ISD::VFIXUPIMM, X86ISD::VFIXUPIMM_SAE), X86_INTRINSIC_DATA(avx512_mask_fixupimm_ps_128, FIXUPIMM, X86ISD::VFIXUPIMM, 0), X86_INTRINSIC_DATA(avx512_mask_fixupimm_ps_256, FIXUPIMM, X86ISD::VFIXUPIMM, 0), - X86_INTRINSIC_DATA(avx512_mask_fixupimm_ps_512, FIXUPIMM, X86ISD::VFIXUPIMM, 0), - X86_INTRINSIC_DATA(avx512_mask_fixupimm_sd, FIXUPIMMS, X86ISD::VFIXUPIMMS, 0), - X86_INTRINSIC_DATA(avx512_mask_fixupimm_ss, FIXUPIMMS, X86ISD::VFIXUPIMMS, 0), + X86_INTRINSIC_DATA(avx512_mask_fixupimm_ps_512, FIXUPIMM, X86ISD::VFIXUPIMM, X86ISD::VFIXUPIMM_SAE), + X86_INTRINSIC_DATA(avx512_mask_fixupimm_sd, FIXUPIMM, X86ISD::VFIXUPIMMS, X86ISD::VFIXUPIMMS_SAE), + X86_INTRINSIC_DATA(avx512_mask_fixupimm_ss, FIXUPIMM, X86ISD::VFIXUPIMMS, X86ISD::VFIXUPIMMS_SAE), X86_INTRINSIC_DATA(avx512_mask_fpclass_sd, FPCLASSS, X86ISD::VFPCLASSS, 0), X86_INTRINSIC_DATA(avx512_mask_fpclass_ss, FPCLASSS, X86ISD::VFPCLASSS, 0), - X86_INTRINSIC_DATA(avx512_mask_getexp_pd_128, INTR_TYPE_1OP_MASK_RM, - X86ISD::FGETEXP_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_getexp_pd_256, INTR_TYPE_1OP_MASK_RM, - X86ISD::FGETEXP_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_getexp_pd_512, INTR_TYPE_1OP_MASK_RM, - X86ISD::FGETEXP_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_getexp_ps_128, INTR_TYPE_1OP_MASK_RM, - X86ISD::FGETEXP_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_getexp_ps_256, INTR_TYPE_1OP_MASK_RM, - X86ISD::FGETEXP_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_getexp_ps_512, INTR_TYPE_1OP_MASK_RM, - X86ISD::FGETEXP_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_getexp_sd, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FGETEXPS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_getexp_ss, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FGETEXPS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_getmant_pd_128, INTR_TYPE_2OP_MASK, + X86_INTRINSIC_DATA(avx512_mask_getexp_pd_128, INTR_TYPE_1OP_MASK, + X86ISD::FGETEXP, 0), + X86_INTRINSIC_DATA(avx512_mask_getexp_pd_256, INTR_TYPE_1OP_MASK, + X86ISD::FGETEXP, 0), + X86_INTRINSIC_DATA(avx512_mask_getexp_pd_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::FGETEXP, X86ISD::FGETEXP_SAE), + X86_INTRINSIC_DATA(avx512_mask_getexp_ps_128, INTR_TYPE_1OP_MASK, + X86ISD::FGETEXP, 0), + X86_INTRINSIC_DATA(avx512_mask_getexp_ps_256, INTR_TYPE_1OP_MASK, + X86ISD::FGETEXP, 0), + X86_INTRINSIC_DATA(avx512_mask_getexp_ps_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::FGETEXP, X86ISD::FGETEXP_SAE), + X86_INTRINSIC_DATA(avx512_mask_getexp_sd, INTR_TYPE_SCALAR_MASK_SAE, + X86ISD::FGETEXPS, X86ISD::FGETEXPS_SAE), + X86_INTRINSIC_DATA(avx512_mask_getexp_ss, INTR_TYPE_SCALAR_MASK_SAE, + X86ISD::FGETEXPS, X86ISD::FGETEXPS_SAE), + X86_INTRINSIC_DATA(avx512_mask_getmant_pd_128, INTR_TYPE_2OP_MASK_SAE, X86ISD::VGETMANT, 0), - X86_INTRINSIC_DATA(avx512_mask_getmant_pd_256, INTR_TYPE_2OP_MASK, + X86_INTRINSIC_DATA(avx512_mask_getmant_pd_256, INTR_TYPE_2OP_MASK_SAE, X86ISD::VGETMANT, 0), - X86_INTRINSIC_DATA(avx512_mask_getmant_pd_512, INTR_TYPE_2OP_MASK, - X86ISD::VGETMANT, X86ISD::VGETMANT_RND), - X86_INTRINSIC_DATA(avx512_mask_getmant_ps_128, INTR_TYPE_2OP_MASK, + X86_INTRINSIC_DATA(avx512_mask_getmant_pd_512, INTR_TYPE_2OP_MASK_SAE, + X86ISD::VGETMANT, X86ISD::VGETMANT_SAE), + X86_INTRINSIC_DATA(avx512_mask_getmant_ps_128, INTR_TYPE_2OP_MASK_SAE, X86ISD::VGETMANT, 0), - X86_INTRINSIC_DATA(avx512_mask_getmant_ps_256, INTR_TYPE_2OP_MASK, + X86_INTRINSIC_DATA(avx512_mask_getmant_ps_256, INTR_TYPE_2OP_MASK_SAE, X86ISD::VGETMANT, 0), - X86_INTRINSIC_DATA(avx512_mask_getmant_ps_512, INTR_TYPE_2OP_MASK, - X86ISD::VGETMANT, X86ISD::VGETMANT_RND), - X86_INTRINSIC_DATA(avx512_mask_getmant_sd, INTR_TYPE_3OP_SCALAR_MASK, - X86ISD::VGETMANTS, X86ISD::VGETMANTS_RND), - X86_INTRINSIC_DATA(avx512_mask_getmant_ss, INTR_TYPE_3OP_SCALAR_MASK, - X86ISD::VGETMANTS, X86ISD::VGETMANTS_RND), - X86_INTRINSIC_DATA(avx512_mask_max_sd_round, INTR_TYPE_SCALAR_MASK, - X86ISD::FMAXS, X86ISD::FMAXS_RND), - X86_INTRINSIC_DATA(avx512_mask_max_ss_round, INTR_TYPE_SCALAR_MASK, - X86ISD::FMAXS, X86ISD::FMAXS_RND), - X86_INTRINSIC_DATA(avx512_mask_min_sd_round, INTR_TYPE_SCALAR_MASK, - X86ISD::FMINS, X86ISD::FMINS_RND), - X86_INTRINSIC_DATA(avx512_mask_min_ss_round, INTR_TYPE_SCALAR_MASK, - X86ISD::FMINS, X86ISD::FMINS_RND), - X86_INTRINSIC_DATA(avx512_mask_mul_sd_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FMULS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_mul_ss_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FMULS_RND, 0), + X86_INTRINSIC_DATA(avx512_mask_getmant_ps_512, INTR_TYPE_2OP_MASK_SAE, + X86ISD::VGETMANT, X86ISD::VGETMANT_SAE), + X86_INTRINSIC_DATA(avx512_mask_getmant_sd, INTR_TYPE_3OP_SCALAR_MASK_SAE, + X86ISD::VGETMANTS, X86ISD::VGETMANTS_SAE), + X86_INTRINSIC_DATA(avx512_mask_getmant_ss, INTR_TYPE_3OP_SCALAR_MASK_SAE, + X86ISD::VGETMANTS, X86ISD::VGETMANTS_SAE), + X86_INTRINSIC_DATA(avx512_mask_max_sd_round, INTR_TYPE_SCALAR_MASK_SAE, + X86ISD::FMAXS, X86ISD::FMAXS_SAE), + X86_INTRINSIC_DATA(avx512_mask_max_ss_round, INTR_TYPE_SCALAR_MASK_SAE, + X86ISD::FMAXS, X86ISD::FMAXS_SAE), + X86_INTRINSIC_DATA(avx512_mask_min_sd_round, INTR_TYPE_SCALAR_MASK_SAE, + X86ISD::FMINS, X86ISD::FMINS_SAE), + X86_INTRINSIC_DATA(avx512_mask_min_ss_round, INTR_TYPE_SCALAR_MASK_SAE, + X86ISD::FMINS, X86ISD::FMINS_SAE), + X86_INTRINSIC_DATA(avx512_mask_mul_sd_round, INTR_TYPE_SCALAR_MASK, + X86ISD::FMULS, X86ISD::FMULS_RND), + X86_INTRINSIC_DATA(avx512_mask_mul_ss_round, INTR_TYPE_SCALAR_MASK, + X86ISD::FMULS, X86ISD::FMULS_RND), X86_INTRINSIC_DATA(avx512_mask_pmov_db_128, TRUNCATE_TO_REG, X86ISD::VTRUNC, X86ISD::VMTRUNC), X86_INTRINSIC_DATA(avx512_mask_pmov_db_256, TRUNCATE_TO_REG, @@ -737,10 +652,6 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86ISD::VTRUNC, X86ISD::VMTRUNC), X86_INTRINSIC_DATA(avx512_mask_pmov_qd_128, TRUNCATE_TO_REG, X86ISD::VTRUNC, X86ISD::VMTRUNC), - X86_INTRINSIC_DATA(avx512_mask_pmov_qd_256, INTR_TYPE_1OP_MASK, - ISD::TRUNCATE, 0), - X86_INTRINSIC_DATA(avx512_mask_pmov_qd_512, INTR_TYPE_1OP_MASK, - ISD::TRUNCATE, 0), X86_INTRINSIC_DATA(avx512_mask_pmov_qw_128, TRUNCATE_TO_REG, X86ISD::VTRUNC, X86ISD::VMTRUNC), X86_INTRINSIC_DATA(avx512_mask_pmov_qw_256, TRUNCATE_TO_REG, @@ -749,10 +660,6 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { ISD::TRUNCATE, X86ISD::VMTRUNC), X86_INTRINSIC_DATA(avx512_mask_pmov_wb_128, TRUNCATE_TO_REG, X86ISD::VTRUNC, X86ISD::VMTRUNC), - X86_INTRINSIC_DATA(avx512_mask_pmov_wb_256, INTR_TYPE_1OP_MASK, - ISD::TRUNCATE, 0), - X86_INTRINSIC_DATA(avx512_mask_pmov_wb_512, INTR_TYPE_1OP_MASK, - ISD::TRUNCATE, 0), X86_INTRINSIC_DATA(avx512_mask_pmovs_db_128, TRUNCATE_TO_REG, X86ISD::VTRUNCS, X86ISD::VMTRUNCS), X86_INTRINSIC_DATA(avx512_mask_pmovs_db_256, TRUNCATE_TO_REG, @@ -825,62 +732,62 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86ISD::VTRUNCUS, 0), X86_INTRINSIC_DATA(avx512_mask_pmovus_wb_512, INTR_TYPE_1OP_MASK, X86ISD::VTRUNCUS, 0), - X86_INTRINSIC_DATA(avx512_mask_range_pd_128, INTR_TYPE_3OP_MASK, X86ISD::VRANGE, 0), - X86_INTRINSIC_DATA(avx512_mask_range_pd_256, INTR_TYPE_3OP_MASK, X86ISD::VRANGE, 0), - X86_INTRINSIC_DATA(avx512_mask_range_pd_512, INTR_TYPE_3OP_MASK, X86ISD::VRANGE, X86ISD::VRANGE_RND), - X86_INTRINSIC_DATA(avx512_mask_range_ps_128, INTR_TYPE_3OP_MASK, X86ISD::VRANGE, 0), - X86_INTRINSIC_DATA(avx512_mask_range_ps_256, INTR_TYPE_3OP_MASK, X86ISD::VRANGE, 0), - X86_INTRINSIC_DATA(avx512_mask_range_ps_512, INTR_TYPE_3OP_MASK, X86ISD::VRANGE, X86ISD::VRANGE_RND), - X86_INTRINSIC_DATA(avx512_mask_range_sd, INTR_TYPE_SCALAR_MASK, X86ISD::VRANGES, X86ISD::VRANGES_RND), - X86_INTRINSIC_DATA(avx512_mask_range_ss, INTR_TYPE_SCALAR_MASK, X86ISD::VRANGES, X86ISD::VRANGES_RND), - X86_INTRINSIC_DATA(avx512_mask_reduce_pd_128, INTR_TYPE_2OP_MASK, X86ISD::VREDUCE, 0), - X86_INTRINSIC_DATA(avx512_mask_reduce_pd_256, INTR_TYPE_2OP_MASK, X86ISD::VREDUCE, 0), - X86_INTRINSIC_DATA(avx512_mask_reduce_pd_512, INTR_TYPE_2OP_MASK, X86ISD::VREDUCE, X86ISD::VREDUCE_RND), - X86_INTRINSIC_DATA(avx512_mask_reduce_ps_128, INTR_TYPE_2OP_MASK, X86ISD::VREDUCE, 0), - X86_INTRINSIC_DATA(avx512_mask_reduce_ps_256, INTR_TYPE_2OP_MASK, X86ISD::VREDUCE, 0), - X86_INTRINSIC_DATA(avx512_mask_reduce_ps_512, INTR_TYPE_2OP_MASK, X86ISD::VREDUCE, X86ISD::VREDUCE_RND), - X86_INTRINSIC_DATA(avx512_mask_reduce_sd, INTR_TYPE_SCALAR_MASK, X86ISD::VREDUCES, X86ISD::VREDUCES_RND), - X86_INTRINSIC_DATA(avx512_mask_reduce_ss, INTR_TYPE_SCALAR_MASK, X86ISD::VREDUCES, X86ISD::VREDUCES_RND), - X86_INTRINSIC_DATA(avx512_mask_rndscale_pd_128, INTR_TYPE_2OP_MASK, X86ISD::VRNDSCALE, 0), - X86_INTRINSIC_DATA(avx512_mask_rndscale_pd_256, INTR_TYPE_2OP_MASK, X86ISD::VRNDSCALE, 0), - X86_INTRINSIC_DATA(avx512_mask_rndscale_pd_512, INTR_TYPE_2OP_MASK, X86ISD::VRNDSCALE, X86ISD::VRNDSCALE_RND), - X86_INTRINSIC_DATA(avx512_mask_rndscale_ps_128, INTR_TYPE_2OP_MASK, X86ISD::VRNDSCALE, 0), - X86_INTRINSIC_DATA(avx512_mask_rndscale_ps_256, INTR_TYPE_2OP_MASK, X86ISD::VRNDSCALE, 0), - X86_INTRINSIC_DATA(avx512_mask_rndscale_ps_512, INTR_TYPE_2OP_MASK, X86ISD::VRNDSCALE, X86ISD::VRNDSCALE_RND), + X86_INTRINSIC_DATA(avx512_mask_range_pd_128, INTR_TYPE_3OP_MASK_SAE, X86ISD::VRANGE, 0), + X86_INTRINSIC_DATA(avx512_mask_range_pd_256, INTR_TYPE_3OP_MASK_SAE, X86ISD::VRANGE, 0), + X86_INTRINSIC_DATA(avx512_mask_range_pd_512, INTR_TYPE_3OP_MASK_SAE, X86ISD::VRANGE, X86ISD::VRANGE_SAE), + X86_INTRINSIC_DATA(avx512_mask_range_ps_128, INTR_TYPE_3OP_MASK_SAE, X86ISD::VRANGE, 0), + X86_INTRINSIC_DATA(avx512_mask_range_ps_256, INTR_TYPE_3OP_MASK_SAE, X86ISD::VRANGE, 0), + X86_INTRINSIC_DATA(avx512_mask_range_ps_512, INTR_TYPE_3OP_MASK_SAE, X86ISD::VRANGE, X86ISD::VRANGE_SAE), + X86_INTRINSIC_DATA(avx512_mask_range_sd, INTR_TYPE_SCALAR_MASK, X86ISD::VRANGES, X86ISD::VRANGES_SAE), + X86_INTRINSIC_DATA(avx512_mask_range_ss, INTR_TYPE_SCALAR_MASK, X86ISD::VRANGES, X86ISD::VRANGES_SAE), + X86_INTRINSIC_DATA(avx512_mask_reduce_pd_128, INTR_TYPE_2OP_MASK_SAE, X86ISD::VREDUCE, 0), + X86_INTRINSIC_DATA(avx512_mask_reduce_pd_256, INTR_TYPE_2OP_MASK_SAE, X86ISD::VREDUCE, 0), + X86_INTRINSIC_DATA(avx512_mask_reduce_pd_512, INTR_TYPE_2OP_MASK_SAE, X86ISD::VREDUCE, X86ISD::VREDUCE_SAE), + X86_INTRINSIC_DATA(avx512_mask_reduce_ps_128, INTR_TYPE_2OP_MASK_SAE, X86ISD::VREDUCE, 0), + X86_INTRINSIC_DATA(avx512_mask_reduce_ps_256, INTR_TYPE_2OP_MASK_SAE, X86ISD::VREDUCE, 0), + X86_INTRINSIC_DATA(avx512_mask_reduce_ps_512, INTR_TYPE_2OP_MASK_SAE, X86ISD::VREDUCE, X86ISD::VREDUCE_SAE), + X86_INTRINSIC_DATA(avx512_mask_reduce_sd, INTR_TYPE_SCALAR_MASK, X86ISD::VREDUCES, X86ISD::VREDUCES_SAE), + X86_INTRINSIC_DATA(avx512_mask_reduce_ss, INTR_TYPE_SCALAR_MASK, X86ISD::VREDUCES, X86ISD::VREDUCES_SAE), + X86_INTRINSIC_DATA(avx512_mask_rndscale_pd_128, INTR_TYPE_2OP_MASK_SAE, X86ISD::VRNDSCALE, 0), + X86_INTRINSIC_DATA(avx512_mask_rndscale_pd_256, INTR_TYPE_2OP_MASK_SAE, X86ISD::VRNDSCALE, 0), + X86_INTRINSIC_DATA(avx512_mask_rndscale_pd_512, INTR_TYPE_2OP_MASK_SAE, X86ISD::VRNDSCALE, X86ISD::VRNDSCALE_SAE), + X86_INTRINSIC_DATA(avx512_mask_rndscale_ps_128, INTR_TYPE_2OP_MASK_SAE, X86ISD::VRNDSCALE, 0), + X86_INTRINSIC_DATA(avx512_mask_rndscale_ps_256, INTR_TYPE_2OP_MASK_SAE, X86ISD::VRNDSCALE, 0), + X86_INTRINSIC_DATA(avx512_mask_rndscale_ps_512, INTR_TYPE_2OP_MASK_SAE, X86ISD::VRNDSCALE, X86ISD::VRNDSCALE_SAE), X86_INTRINSIC_DATA(avx512_mask_rndscale_sd, INTR_TYPE_SCALAR_MASK, - X86ISD::VRNDSCALES, X86ISD::VRNDSCALES_RND), + X86ISD::VRNDSCALES, X86ISD::VRNDSCALES_SAE), X86_INTRINSIC_DATA(avx512_mask_rndscale_ss, INTR_TYPE_SCALAR_MASK, - X86ISD::VRNDSCALES, X86ISD::VRNDSCALES_RND), - X86_INTRINSIC_DATA(avx512_mask_scalef_pd_128, INTR_TYPE_2OP_MASK_RM, + X86ISD::VRNDSCALES, X86ISD::VRNDSCALES_SAE), + X86_INTRINSIC_DATA(avx512_mask_scalef_pd_128, INTR_TYPE_2OP_MASK, X86ISD::SCALEF, 0), - X86_INTRINSIC_DATA(avx512_mask_scalef_pd_256, INTR_TYPE_2OP_MASK_RM, + X86_INTRINSIC_DATA(avx512_mask_scalef_pd_256, INTR_TYPE_2OP_MASK, X86ISD::SCALEF, 0), - X86_INTRINSIC_DATA(avx512_mask_scalef_pd_512, INTR_TYPE_2OP_MASK_RM, + X86_INTRINSIC_DATA(avx512_mask_scalef_pd_512, INTR_TYPE_2OP_MASK, + X86ISD::SCALEF, X86ISD::SCALEF_RND), + X86_INTRINSIC_DATA(avx512_mask_scalef_ps_128, INTR_TYPE_2OP_MASK, X86ISD::SCALEF, 0), - X86_INTRINSIC_DATA(avx512_mask_scalef_ps_128, INTR_TYPE_2OP_MASK_RM, + X86_INTRINSIC_DATA(avx512_mask_scalef_ps_256, INTR_TYPE_2OP_MASK, X86ISD::SCALEF, 0), - X86_INTRINSIC_DATA(avx512_mask_scalef_ps_256, INTR_TYPE_2OP_MASK_RM, - X86ISD::SCALEF, 0), - X86_INTRINSIC_DATA(avx512_mask_scalef_ps_512, INTR_TYPE_2OP_MASK_RM, - X86ISD::SCALEF, 0), - X86_INTRINSIC_DATA(avx512_mask_scalef_sd, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::SCALEFS, 0), - X86_INTRINSIC_DATA(avx512_mask_scalef_ss, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::SCALEFS, 0), - X86_INTRINSIC_DATA(avx512_mask_sqrt_sd, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FSQRTS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_sqrt_ss, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FSQRTS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_sub_sd_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FSUBS_RND, 0), - X86_INTRINSIC_DATA(avx512_mask_sub_ss_round, INTR_TYPE_SCALAR_MASK_RM, - X86ISD::FSUBS_RND, 0), + X86_INTRINSIC_DATA(avx512_mask_scalef_ps_512, INTR_TYPE_2OP_MASK, + X86ISD::SCALEF, X86ISD::SCALEF_RND), + X86_INTRINSIC_DATA(avx512_mask_scalef_sd, INTR_TYPE_SCALAR_MASK, + X86ISD::SCALEFS, X86ISD::SCALEFS_RND), + X86_INTRINSIC_DATA(avx512_mask_scalef_ss, INTR_TYPE_SCALAR_MASK, + X86ISD::SCALEFS, X86ISD::SCALEFS_RND), + X86_INTRINSIC_DATA(avx512_mask_sqrt_sd, INTR_TYPE_SCALAR_MASK, + X86ISD::FSQRTS, X86ISD::FSQRTS_RND), + X86_INTRINSIC_DATA(avx512_mask_sqrt_ss, INTR_TYPE_SCALAR_MASK, + X86ISD::FSQRTS, X86ISD::FSQRTS_RND), + X86_INTRINSIC_DATA(avx512_mask_sub_sd_round, INTR_TYPE_SCALAR_MASK, + X86ISD::FSUBS, X86ISD::FSUBS_RND), + X86_INTRINSIC_DATA(avx512_mask_sub_ss_round, INTR_TYPE_SCALAR_MASK, + X86ISD::FSUBS, X86ISD::FSUBS_RND), X86_INTRINSIC_DATA(avx512_mask_vcvtph2ps_128, INTR_TYPE_1OP_MASK, X86ISD::CVTPH2PS, 0), X86_INTRINSIC_DATA(avx512_mask_vcvtph2ps_256, INTR_TYPE_1OP_MASK, X86ISD::CVTPH2PS, 0), - X86_INTRINSIC_DATA(avx512_mask_vcvtph2ps_512, INTR_TYPE_1OP_MASK, - X86ISD::CVTPH2PS, X86ISD::CVTPH2PS_RND), + X86_INTRINSIC_DATA(avx512_mask_vcvtph2ps_512, INTR_TYPE_1OP_MASK_SAE, + X86ISD::CVTPH2PS, X86ISD::CVTPH2PS_SAE), X86_INTRINSIC_DATA(avx512_mask_vcvtps2ph_128, CVTPS2PH_MASK, X86ISD::CVTPS2PH, X86ISD::MCVTPS2PH), X86_INTRINSIC_DATA(avx512_mask_vcvtps2ph_256, CVTPS2PH_MASK, @@ -893,28 +800,30 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx512_maskz_fixupimm_pd_256, FIXUPIMM_MASKZ, X86ISD::VFIXUPIMM, 0), X86_INTRINSIC_DATA(avx512_maskz_fixupimm_pd_512, FIXUPIMM_MASKZ, - X86ISD::VFIXUPIMM, 0), + X86ISD::VFIXUPIMM, X86ISD::VFIXUPIMM_SAE), X86_INTRINSIC_DATA(avx512_maskz_fixupimm_ps_128, FIXUPIMM_MASKZ, X86ISD::VFIXUPIMM, 0), X86_INTRINSIC_DATA(avx512_maskz_fixupimm_ps_256, FIXUPIMM_MASKZ, X86ISD::VFIXUPIMM, 0), X86_INTRINSIC_DATA(avx512_maskz_fixupimm_ps_512, FIXUPIMM_MASKZ, - X86ISD::VFIXUPIMM, 0), - X86_INTRINSIC_DATA(avx512_maskz_fixupimm_sd, FIXUPIMMS_MASKZ, - X86ISD::VFIXUPIMMS, 0), - X86_INTRINSIC_DATA(avx512_maskz_fixupimm_ss, FIXUPIMMS_MASKZ, - X86ISD::VFIXUPIMMS, 0), + X86ISD::VFIXUPIMM, X86ISD::VFIXUPIMM_SAE), + X86_INTRINSIC_DATA(avx512_maskz_fixupimm_sd, FIXUPIMM_MASKZ, + X86ISD::VFIXUPIMMS, X86ISD::VFIXUPIMMS_SAE), + X86_INTRINSIC_DATA(avx512_maskz_fixupimm_ss, FIXUPIMM_MASKZ, + X86ISD::VFIXUPIMMS, X86ISD::VFIXUPIMMS_SAE), - X86_INTRINSIC_DATA(avx512_max_pd_512, INTR_TYPE_2OP, X86ISD::FMAX, X86ISD::FMAX_RND), - X86_INTRINSIC_DATA(avx512_max_ps_512, INTR_TYPE_2OP, X86ISD::FMAX, X86ISD::FMAX_RND), - X86_INTRINSIC_DATA(avx512_min_pd_512, INTR_TYPE_2OP, X86ISD::FMIN, X86ISD::FMIN_RND), - X86_INTRINSIC_DATA(avx512_min_ps_512, INTR_TYPE_2OP, X86ISD::FMIN, X86ISD::FMIN_RND), + X86_INTRINSIC_DATA(avx512_max_pd_512, INTR_TYPE_2OP_SAE, X86ISD::FMAX, X86ISD::FMAX_SAE), + X86_INTRINSIC_DATA(avx512_max_ps_512, INTR_TYPE_2OP_SAE, X86ISD::FMAX, X86ISD::FMAX_SAE), + X86_INTRINSIC_DATA(avx512_min_pd_512, INTR_TYPE_2OP_SAE, X86ISD::FMIN, X86ISD::FMIN_SAE), + X86_INTRINSIC_DATA(avx512_min_ps_512, INTR_TYPE_2OP_SAE, X86ISD::FMIN, X86ISD::FMIN_SAE), X86_INTRINSIC_DATA(avx512_mul_pd_512, INTR_TYPE_2OP, ISD::FMUL, X86ISD::FMUL_RND), X86_INTRINSIC_DATA(avx512_mul_ps_512, INTR_TYPE_2OP, ISD::FMUL, X86ISD::FMUL_RND), X86_INTRINSIC_DATA(avx512_packssdw_512, INTR_TYPE_2OP, X86ISD::PACKSS, 0), X86_INTRINSIC_DATA(avx512_packsswb_512, INTR_TYPE_2OP, X86ISD::PACKSS, 0), X86_INTRINSIC_DATA(avx512_packusdw_512, INTR_TYPE_2OP, X86ISD::PACKUS, 0), X86_INTRINSIC_DATA(avx512_packuswb_512, INTR_TYPE_2OP, X86ISD::PACKUS, 0), + X86_INTRINSIC_DATA(avx512_pavg_b_512, INTR_TYPE_2OP, X86ISD::AVG, 0), + X86_INTRINSIC_DATA(avx512_pavg_w_512, INTR_TYPE_2OP, X86ISD::AVG, 0), X86_INTRINSIC_DATA(avx512_permvar_df_256, VPERM_2OP, X86ISD::VPERMV, 0), X86_INTRINSIC_DATA(avx512_permvar_df_512, VPERM_2OP, X86ISD::VPERMV, 0), X86_INTRINSIC_DATA(avx512_permvar_di_256, VPERM_2OP, X86ISD::VPERMV, 0), @@ -943,11 +852,11 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx512_pslli_d_512, VSHIFT, X86ISD::VSHLI, 0), X86_INTRINSIC_DATA(avx512_pslli_q_512, VSHIFT, X86ISD::VSHLI, 0), X86_INTRINSIC_DATA(avx512_pslli_w_512, VSHIFT, X86ISD::VSHLI, 0), - X86_INTRINSIC_DATA(avx512_psllv_d_512, INTR_TYPE_2OP, ISD::SHL, 0), - X86_INTRINSIC_DATA(avx512_psllv_q_512, INTR_TYPE_2OP, ISD::SHL, 0), - X86_INTRINSIC_DATA(avx512_psllv_w_128, INTR_TYPE_2OP, ISD::SHL, 0), - X86_INTRINSIC_DATA(avx512_psllv_w_256, INTR_TYPE_2OP, ISD::SHL, 0), - X86_INTRINSIC_DATA(avx512_psllv_w_512, INTR_TYPE_2OP, ISD::SHL, 0), + X86_INTRINSIC_DATA(avx512_psllv_d_512, INTR_TYPE_2OP, X86ISD::VSHLV, 0), + X86_INTRINSIC_DATA(avx512_psllv_q_512, INTR_TYPE_2OP, X86ISD::VSHLV, 0), + X86_INTRINSIC_DATA(avx512_psllv_w_128, INTR_TYPE_2OP, X86ISD::VSHLV, 0), + X86_INTRINSIC_DATA(avx512_psllv_w_256, INTR_TYPE_2OP, X86ISD::VSHLV, 0), + X86_INTRINSIC_DATA(avx512_psllv_w_512, INTR_TYPE_2OP, X86ISD::VSHLV, 0), X86_INTRINSIC_DATA(avx512_psra_d_512, INTR_TYPE_2OP, X86ISD::VSRA, 0), X86_INTRINSIC_DATA(avx512_psra_q_128, INTR_TYPE_2OP, X86ISD::VSRA, 0), X86_INTRINSIC_DATA(avx512_psra_q_256, INTR_TYPE_2OP, X86ISD::VSRA, 0), @@ -971,11 +880,11 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx512_psrli_d_512, VSHIFT, X86ISD::VSRLI, 0), X86_INTRINSIC_DATA(avx512_psrli_q_512, VSHIFT, X86ISD::VSRLI, 0), X86_INTRINSIC_DATA(avx512_psrli_w_512, VSHIFT, X86ISD::VSRLI, 0), - X86_INTRINSIC_DATA(avx512_psrlv_d_512, INTR_TYPE_2OP, ISD::SRL, 0), - X86_INTRINSIC_DATA(avx512_psrlv_q_512, INTR_TYPE_2OP, ISD::SRL, 0), - X86_INTRINSIC_DATA(avx512_psrlv_w_128, INTR_TYPE_2OP, ISD::SRL, 0), - X86_INTRINSIC_DATA(avx512_psrlv_w_256, INTR_TYPE_2OP, ISD::SRL, 0), - X86_INTRINSIC_DATA(avx512_psrlv_w_512, INTR_TYPE_2OP, ISD::SRL, 0), + X86_INTRINSIC_DATA(avx512_psrlv_d_512, INTR_TYPE_2OP, X86ISD::VSRLV, 0), + X86_INTRINSIC_DATA(avx512_psrlv_q_512, INTR_TYPE_2OP, X86ISD::VSRLV, 0), + X86_INTRINSIC_DATA(avx512_psrlv_w_128, INTR_TYPE_2OP, X86ISD::VSRLV, 0), + X86_INTRINSIC_DATA(avx512_psrlv_w_256, INTR_TYPE_2OP, X86ISD::VSRLV, 0), + X86_INTRINSIC_DATA(avx512_psrlv_w_512, INTR_TYPE_2OP, X86ISD::VSRLV, 0), X86_INTRINSIC_DATA(avx512_pternlog_d_128, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0), X86_INTRINSIC_DATA(avx512_pternlog_d_256, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0), X86_INTRINSIC_DATA(avx512_pternlog_d_512, INTR_TYPE_4OP, X86ISD::VPTERNLOG, 0), @@ -990,10 +899,10 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx512_rcp14_ps_512, INTR_TYPE_1OP_MASK, X86ISD::RCP14, 0), X86_INTRINSIC_DATA(avx512_rcp14_sd, INTR_TYPE_SCALAR_MASK, X86ISD::RCP14S, 0), X86_INTRINSIC_DATA(avx512_rcp14_ss, INTR_TYPE_SCALAR_MASK, X86ISD::RCP14S, 0), - X86_INTRINSIC_DATA(avx512_rcp28_pd, INTR_TYPE_1OP_MASK_RM, X86ISD::RCP28, 0), - X86_INTRINSIC_DATA(avx512_rcp28_ps, INTR_TYPE_1OP_MASK_RM, X86ISD::RCP28, 0), - X86_INTRINSIC_DATA(avx512_rcp28_sd, INTR_TYPE_SCALAR_MASK_RM, X86ISD::RCP28S, 0), - X86_INTRINSIC_DATA(avx512_rcp28_ss, INTR_TYPE_SCALAR_MASK_RM, X86ISD::RCP28S, 0), + X86_INTRINSIC_DATA(avx512_rcp28_pd, INTR_TYPE_1OP_MASK_SAE, X86ISD::RCP28, X86ISD::RCP28_SAE), + X86_INTRINSIC_DATA(avx512_rcp28_ps, INTR_TYPE_1OP_MASK_SAE, X86ISD::RCP28, X86ISD::RCP28_SAE), + X86_INTRINSIC_DATA(avx512_rcp28_sd, INTR_TYPE_SCALAR_MASK_SAE, X86ISD::RCP28S, X86ISD::RCP28S_SAE), + X86_INTRINSIC_DATA(avx512_rcp28_ss, INTR_TYPE_SCALAR_MASK_SAE, X86ISD::RCP28S, X86ISD::RCP28S_SAE), X86_INTRINSIC_DATA(avx512_rsqrt14_pd_128, INTR_TYPE_1OP_MASK, X86ISD::RSQRT14, 0), X86_INTRINSIC_DATA(avx512_rsqrt14_pd_256, INTR_TYPE_1OP_MASK, X86ISD::RSQRT14, 0), X86_INTRINSIC_DATA(avx512_rsqrt14_pd_512, INTR_TYPE_1OP_MASK, X86ISD::RSQRT14, 0), @@ -1002,14 +911,16 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx512_rsqrt14_ps_512, INTR_TYPE_1OP_MASK, X86ISD::RSQRT14, 0), X86_INTRINSIC_DATA(avx512_rsqrt14_sd, INTR_TYPE_SCALAR_MASK, X86ISD::RSQRT14S, 0), X86_INTRINSIC_DATA(avx512_rsqrt14_ss, INTR_TYPE_SCALAR_MASK, X86ISD::RSQRT14S, 0), - X86_INTRINSIC_DATA(avx512_rsqrt28_pd, INTR_TYPE_1OP_MASK_RM,X86ISD::RSQRT28, 0), - X86_INTRINSIC_DATA(avx512_rsqrt28_ps, INTR_TYPE_1OP_MASK_RM,X86ISD::RSQRT28, 0), - X86_INTRINSIC_DATA(avx512_rsqrt28_sd, INTR_TYPE_SCALAR_MASK_RM,X86ISD::RSQRT28S, 0), - X86_INTRINSIC_DATA(avx512_rsqrt28_ss, INTR_TYPE_SCALAR_MASK_RM,X86ISD::RSQRT28S, 0), + X86_INTRINSIC_DATA(avx512_rsqrt28_pd, INTR_TYPE_1OP_MASK_SAE,X86ISD::RSQRT28, X86ISD::RSQRT28_SAE), + X86_INTRINSIC_DATA(avx512_rsqrt28_ps, INTR_TYPE_1OP_MASK_SAE,X86ISD::RSQRT28, X86ISD::RSQRT28_SAE), + X86_INTRINSIC_DATA(avx512_rsqrt28_sd, INTR_TYPE_SCALAR_MASK_SAE,X86ISD::RSQRT28S, X86ISD::RSQRT28S_SAE), + X86_INTRINSIC_DATA(avx512_rsqrt28_ss, INTR_TYPE_SCALAR_MASK_SAE,X86ISD::RSQRT28S, X86ISD::RSQRT28S_SAE), + X86_INTRINSIC_DATA(avx512_sitofp_round, INTR_TYPE_1OP, ISD::SINT_TO_FP, X86ISD::SINT_TO_FP_RND), X86_INTRINSIC_DATA(avx512_sqrt_pd_512, INTR_TYPE_1OP, ISD::FSQRT, X86ISD::FSQRT_RND), X86_INTRINSIC_DATA(avx512_sqrt_ps_512, INTR_TYPE_1OP, ISD::FSQRT, X86ISD::FSQRT_RND), X86_INTRINSIC_DATA(avx512_sub_pd_512, INTR_TYPE_2OP, ISD::FSUB, X86ISD::FSUB_RND), X86_INTRINSIC_DATA(avx512_sub_ps_512, INTR_TYPE_2OP, ISD::FSUB, X86ISD::FSUB_RND), + X86_INTRINSIC_DATA(avx512_uitofp_round, INTR_TYPE_1OP, ISD::UINT_TO_FP, X86ISD::UINT_TO_FP_RND), X86_INTRINSIC_DATA(avx512_vcomi_sd, COMI_RM, X86ISD::COMI, X86ISD::UCOMI), X86_INTRINSIC_DATA(avx512_vcomi_ss, COMI_RM, X86ISD::COMI, X86ISD::UCOMI), X86_INTRINSIC_DATA(avx512_vcvtsd2si32, INTR_TYPE_1OP, X86ISD::CVTS2SI, X86ISD::CVTS2SI_RND), @@ -1071,6 +982,16 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(avx512_vpshufbitqmb_128, INTR_TYPE_2OP, X86ISD::VPSHUFBITQMB, 0), X86_INTRINSIC_DATA(avx512_vpshufbitqmb_256, INTR_TYPE_2OP, X86ISD::VPSHUFBITQMB, 0), X86_INTRINSIC_DATA(avx512_vpshufbitqmb_512, INTR_TYPE_2OP, X86ISD::VPSHUFBITQMB, 0), + // bfloat16 + X86_INTRINSIC_DATA(avx512bf16_cvtne2ps2bf16_128, INTR_TYPE_2OP, X86ISD::CVTNE2PS2BF16, 0), + X86_INTRINSIC_DATA(avx512bf16_cvtne2ps2bf16_256, INTR_TYPE_2OP, X86ISD::CVTNE2PS2BF16, 0), + X86_INTRINSIC_DATA(avx512bf16_cvtne2ps2bf16_512, INTR_TYPE_2OP, X86ISD::CVTNE2PS2BF16, 0), + X86_INTRINSIC_DATA(avx512bf16_cvtneps2bf16_256, INTR_TYPE_1OP, X86ISD::CVTNEPS2BF16, 0), + X86_INTRINSIC_DATA(avx512bf16_cvtneps2bf16_512, INTR_TYPE_1OP, X86ISD::CVTNEPS2BF16, 0), + X86_INTRINSIC_DATA(avx512bf16_dpbf16ps_128, INTR_TYPE_3OP, X86ISD::DPBF16PS, 0), + X86_INTRINSIC_DATA(avx512bf16_dpbf16ps_256, INTR_TYPE_3OP, X86ISD::DPBF16PS, 0), + X86_INTRINSIC_DATA(avx512bf16_dpbf16ps_512, INTR_TYPE_3OP, X86ISD::DPBF16PS, 0), + X86_INTRINSIC_DATA(avx512bf16_mask_cvtneps2bf16_128, CVTNEPS2BF16_MASK, X86ISD::CVTNEPS2BF16, X86ISD::MCVTNEPS2BF16), X86_INTRINSIC_DATA(bmi_bextr_32, INTR_TYPE_2OP, X86ISD::BEXTR, 0), X86_INTRINSIC_DATA(bmi_bextr_64, INTR_TYPE_2OP, X86ISD::BEXTR, 0), X86_INTRINSIC_DATA(bmi_bzhi_32, INTR_TYPE_2OP, X86ISD::BZHI, 0), @@ -1111,6 +1032,7 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(sse2_cvtps2dq, INTR_TYPE_1OP, X86ISD::CVTP2SI, 0), X86_INTRINSIC_DATA(sse2_cvtsd2si, INTR_TYPE_1OP, X86ISD::CVTS2SI, 0), X86_INTRINSIC_DATA(sse2_cvtsd2si64, INTR_TYPE_1OP, X86ISD::CVTS2SI, 0), + X86_INTRINSIC_DATA(sse2_cvtsd2ss, INTR_TYPE_2OP, X86ISD::VFPROUNDS, 0), X86_INTRINSIC_DATA(sse2_cvttpd2dq, INTR_TYPE_1OP, X86ISD::CVTTP2SI, 0), X86_INTRINSIC_DATA(sse2_cvttps2dq, INTR_TYPE_1OP, X86ISD::CVTTP2SI, 0), X86_INTRINSIC_DATA(sse2_cvttsd2si, INTR_TYPE_1OP, X86ISD::CVTTS2SI, 0), @@ -1123,6 +1045,8 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(sse2_packssdw_128, INTR_TYPE_2OP, X86ISD::PACKSS, 0), X86_INTRINSIC_DATA(sse2_packsswb_128, INTR_TYPE_2OP, X86ISD::PACKSS, 0), X86_INTRINSIC_DATA(sse2_packuswb_128, INTR_TYPE_2OP, X86ISD::PACKUS, 0), + X86_INTRINSIC_DATA(sse2_pavg_b, INTR_TYPE_2OP, X86ISD::AVG, 0), + X86_INTRINSIC_DATA(sse2_pavg_w, INTR_TYPE_2OP, X86ISD::AVG, 0), X86_INTRINSIC_DATA(sse2_pmadd_wd, INTR_TYPE_2OP, X86ISD::VPMADDWD, 0), X86_INTRINSIC_DATA(sse2_pmovmskb_128, INTR_TYPE_1OP, X86ISD::MOVMSK, 0), X86_INTRINSIC_DATA(sse2_pmulh_w, INTR_TYPE_2OP, ISD::MULHS, 0), @@ -1156,8 +1080,11 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(sse3_hadd_ps, INTR_TYPE_2OP, X86ISD::FHADD, 0), X86_INTRINSIC_DATA(sse3_hsub_pd, INTR_TYPE_2OP, X86ISD::FHSUB, 0), X86_INTRINSIC_DATA(sse3_hsub_ps, INTR_TYPE_2OP, X86ISD::FHSUB, 0), + X86_INTRINSIC_DATA(sse41_blendvpd, BLENDV, X86ISD::BLENDV, 0), + X86_INTRINSIC_DATA(sse41_blendvps, BLENDV, X86ISD::BLENDV, 0), X86_INTRINSIC_DATA(sse41_insertps, INTR_TYPE_3OP, X86ISD::INSERTPS, 0), X86_INTRINSIC_DATA(sse41_packusdw, INTR_TYPE_2OP, X86ISD::PACKUS, 0), + X86_INTRINSIC_DATA(sse41_pblendvb, BLENDV, X86ISD::BLENDV, 0), X86_INTRINSIC_DATA(sse41_phminposuw, INTR_TYPE_1OP, X86ISD::PHMINPOS, 0), X86_INTRINSIC_DATA(sse41_round_pd, ROUNDP, X86ISD::VRNDSCALE, 0), X86_INTRINSIC_DATA(sse41_round_ps, ROUNDP, X86ISD::VRNDSCALE, 0), @@ -1200,14 +1127,6 @@ static const IntrinsicData IntrinsicsWithoutChain[] = { X86_INTRINSIC_DATA(vgf2p8mulb_512, INTR_TYPE_2OP, X86ISD::GF2P8MULB, 0), - X86_INTRINSIC_DATA(xop_vpcomb, INTR_TYPE_3OP, X86ISD::VPCOM, 0), - X86_INTRINSIC_DATA(xop_vpcomd, INTR_TYPE_3OP, X86ISD::VPCOM, 0), - X86_INTRINSIC_DATA(xop_vpcomq, INTR_TYPE_3OP, X86ISD::VPCOM, 0), - X86_INTRINSIC_DATA(xop_vpcomub, INTR_TYPE_3OP, X86ISD::VPCOMU, 0), - X86_INTRINSIC_DATA(xop_vpcomud, INTR_TYPE_3OP, X86ISD::VPCOMU, 0), - X86_INTRINSIC_DATA(xop_vpcomuq, INTR_TYPE_3OP, X86ISD::VPCOMU, 0), - X86_INTRINSIC_DATA(xop_vpcomuw, INTR_TYPE_3OP, X86ISD::VPCOMU, 0), - X86_INTRINSIC_DATA(xop_vpcomw, INTR_TYPE_3OP, X86ISD::VPCOM, 0), X86_INTRINSIC_DATA(xop_vpermil2pd, INTR_TYPE_4OP, X86ISD::VPERMIL2, 0), X86_INTRINSIC_DATA(xop_vpermil2pd_256, INTR_TYPE_4OP, X86ISD::VPERMIL2, 0), X86_INTRINSIC_DATA(xop_vpermil2ps, INTR_TYPE_4OP, X86ISD::VPERMIL2, 0), |