On Mon, Aug 2, 2021 at 2:44 PM liuhongt <hongtao....@intel.com> wrote:
>
> From: "Guo, Xuepeng" <xuepeng....@intel.com>
>
> gcc/ChangeLog:
>
> * common/config/i386/cpuinfo.h (get_available_features):
> Detect FEATURE_AVX512FP16.
> * common/config/i386/i386-common.c
> (OPTION_MASK_ISA_AVX512FP16_SET,
> OPTION_MASK_ISA_AVX512FP16_UNSET,
> OPTION_MASK_ISA2_AVX512FP16_SET,
> OPTION_MASK_ISA2_AVX512FP16_UNSET): New.
> (OPTION_MASK_ISA2_AVX512BW_UNSET,
> OPTION_MASK_ISA2_AVX512BF16_UNSET): Add AVX512FP16.
> (ix86_handle_option): Handle -mavx512fp16.
> * common/config/i386/i386-cpuinfo.h (enum processor_features):
> Add FEATURE_AVX512FP16.
> * common/config/i386/i386-isas.h: Add entry for AVX512FP16.
> * config.gcc: Add avx512fp16intrin.h.
> * config/i386/avx512fp16intrin.h: New intrinsic header.
> * config/i386/cpuid.h: Add bit_AVX512FP16.
> * config/i386/i386-builtin-types.def: (FLOAT16): New primitive type.
> * config/i386/i386-builtins.c: Support _Float16 type for i386
> backend.
> (ix86_init_float16_builtins): New function.
> (ix86_float16_type_node): New.
> * config/i386/i386-c.c (ix86_target_macros_internal): Define
> __AVX512FP16__.
> * config/i386/i386-expand.c (ix86_expand_branch): Support
> HFmode.
> (ix86_prepare_fp_compare_args): Adjust TARGET_SSE_MATH &&
> SSE_FLOAT_MODE_P to SSE_FLOAT_MODE_SSEMATH_OR_HF_P.
> (ix86_expand_fp_movcc): Ditto.
> * config/i386/i386-isa.def: Add PTA define for AVX512FP16.
> * config/i386/i386-options.c (isa2_opts): Add -mavx512fp16.
> (ix86_valid_target_attribute_inner_p): Add avx512fp16 attribute.
> * config/i386/i386.c (ix86_get_ssemov): Use
> vmovdqu16/vmovw/vmovsh for HFmode/HImode scalar or vector.
> (ix86_get_excess_precision): Use
> FLT_EVAL_METHOD_PROMOTE_TO_FLOAT16 when TARGET_AVX512FP16
> existed.
> (sse_store_index): Use SFmode cost for HFmode cost.
> (inline_memory_move_cost): Add HFmode, and perfer SSE cost over
> GPR cost for HFmode.
> (ix86_hard_regno_mode_ok): Allow HImode in sse register.
> (ix86_mangle_type): Add manlging for _Float16 type.
> (inline_secondary_memory_needed): No memory is needed for
> 16bit movement between gpr and sse reg under
> TARGET_AVX512FP16.
> (ix86_multiplication_cost): Adjust TARGET_SSE_MATH &&
> SSE_FLOAT_MODE_P to SSE_FLOAT_MODE_SSEMATH_OR_HF_P.
> (ix86_division_cost): Ditto.
> (ix86_rtx_costs): Ditto.
> (ix86_add_stmt_cost): Ditto.
> (ix86_optab_supported_p): Ditto.
> * config/i386/i386.h (VALID_AVX512F_SCALAR_MODE): Add HFmode.
> (SSE_FLOAT_MODE_SSEMATH_OR_HF_P): Add HFmode.
> (PTA_SAPPHIRERAPIDS): Add PTA_AVX512FP16.
> * config/i386/i386.md (mode): Add HFmode.
> (MODE_SIZE): Add HFmode.
> (isa): Add avx512fp16.
> (enabled): Handle avx512fp16.
> (ssemodesuffix): Add sh suffix for HFmode.
> (comm): Add mult, div.
> (plusminusmultdiv): New code iterator.
> (insn): Add mult, div.
> (*movhf_internal): Adjust for avx512fp16 instruction.
> (*movhi_internal): Ditto.
> (*cmpi<unord>hf): New define_insn for HFmode.
> (*ieee_s<ieee_maxmin>hf3): Likewise.
> (extendhf<mode>2): Likewise.
> (trunc<mode>hf2): Likewise.
> (float<floatunssuffix><mode>hf2): Likewise.
> (*<insn>hf): Likewise.
> (cbranchhf4): New expander.
> (movhfcc): Likewise.
> (<insn>hf3): Likewise.
> (mulhf3): Likewise.
> (divhf3): Likewise.
> * config/i386/i386.opt: Add mavx512fp16.
> * config/i386/immintrin.h: Include avx512fp16intrin.h.
> * doc/invoke.texi: Add mavx512fp16.
> * doc/extend.texi: Add avx512fp16 Usage Notes.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.target/i386/avx-1.c: Add -mavx512fp16 in dg-options.
> * gcc.target/i386/avx-2.c: Ditto.
> * gcc.target/i386/avx512-check.h: Check cpuid for AVX512FP16.
> * gcc.target/i386/funcspec-56.inc: Add new target attribute check.
> * gcc.target/i386/sse-13.c: Add -mavx512fp16.
> * gcc.target/i386/sse-14.c: Ditto.
> * gcc.target/i386/sse-22.c: Ditto.
> * gcc.target/i386/sse-23.c: Ditto.
> * lib/target-supports.exp: (check_effective_target_avx512fp16): New.
> * g++.target/i386/float16-1.C: New test.
> * g++.target/i386/float16-2.C: Ditto.
> * g++.target/i386/float16-3.C: Ditto.
> * gcc.target/i386/avx512fp16-12a.c: Ditto.
> * gcc.target/i386/avx512fp16-12b.c: Ditto.
> * gcc.target/i386/float16-3a.c: Ditto.
> * gcc.target/i386/float16-3b.c: Ditto.
> * gcc.target/i386/float16-4a.c: Ditto.
> * gcc.target/i386/float16-4b.c: Ditto.
> * gcc.target/i386/pr54855-12.c: Ditto.
> * g++.dg/other/i386-2.C: Ditto.
> * g++.dg/other/i386-3.C: Ditto.
>
> Co-Authored-By: H.J. Lu <hongjiu...@intel.com>
> Co-Authored-By: Liu Hongtao <hongtao....@intel.com>
> Co-Authored-By: Wang Hongyu <hongyu.w...@intel.com>
> Co-Authored-By: Xu Dianhong <dianhong...@intel.com>
Ping.
Hi uros, this is the updated patch according to your comments.
> ---
> gcc/common/config/i386/cpuinfo.h | 2 +
> gcc/common/config/i386/i386-common.c | 26 ++-
> gcc/common/config/i386/i386-cpuinfo.h | 1 +
> gcc/common/config/i386/i386-isas.h | 1 +
> gcc/config.gcc | 2 +-
> gcc/config/i386/avx512fp16intrin.h | 53 ++++++
> gcc/config/i386/cpuid.h | 1 +
> gcc/config/i386/i386-builtin-types.def | 1 +
> gcc/config/i386/i386-builtins.c | 23 +++
> gcc/config/i386/i386-c.c | 2 +
> gcc/config/i386/i386-expand.c | 5 +-
> gcc/config/i386/i386-isa.def | 1 +
> gcc/config/i386/i386-options.c | 4 +-
> gcc/config/i386/i386.c | 133 ++++++++++----
> gcc/config/i386/i386.h | 11 +-
> gcc/config/i386/i386.md | 172 ++++++++++++++++--
> gcc/config/i386/i386.opt | 4 +
> gcc/config/i386/immintrin.h | 4 +
> gcc/doc/extend.texi | 8 +
> gcc/doc/invoke.texi | 10 +-
> gcc/testsuite/g++.dg/other/i386-2.C | 2 +-
> gcc/testsuite/g++.dg/other/i386-3.C | 2 +-
> gcc/testsuite/g++.target/i386/float16-1.C | 8 +
> gcc/testsuite/g++.target/i386/float16-2.C | 14 ++
> gcc/testsuite/g++.target/i386/float16-3.C | 10 +
> gcc/testsuite/gcc.target/i386/avx-1.c | 2 +-
> gcc/testsuite/gcc.target/i386/avx-2.c | 2 +-
> gcc/testsuite/gcc.target/i386/avx512-check.h | 3 +
> .../gcc.target/i386/avx512fp16-12a.c | 21 +++
> .../gcc.target/i386/avx512fp16-12b.c | 27 +++
> gcc/testsuite/gcc.target/i386/float16-3a.c | 10 +
> gcc/testsuite/gcc.target/i386/float16-3b.c | 10 +
> gcc/testsuite/gcc.target/i386/float16-4a.c | 10 +
> gcc/testsuite/gcc.target/i386/float16-4b.c | 10 +
> gcc/testsuite/gcc.target/i386/funcspec-56.inc | 2 +
> gcc/testsuite/gcc.target/i386/pr54855-12.c | 14 ++
> gcc/testsuite/gcc.target/i386/sse-13.c | 2 +-
> gcc/testsuite/gcc.target/i386/sse-14.c | 2 +-
> gcc/testsuite/gcc.target/i386/sse-22.c | 4 +-
> gcc/testsuite/gcc.target/i386/sse-23.c | 2 +-
> gcc/testsuite/lib/target-supports.exp | 13 +-
> 41 files changed, 558 insertions(+), 76 deletions(-)
> create mode 100644 gcc/config/i386/avx512fp16intrin.h
> create mode 100644 gcc/testsuite/g++.target/i386/float16-1.C
> create mode 100644 gcc/testsuite/g++.target/i386/float16-2.C
> create mode 100644 gcc/testsuite/g++.target/i386/float16-3.C
> create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-12a.c
> create mode 100644 gcc/testsuite/gcc.target/i386/avx512fp16-12b.c
> create mode 100644 gcc/testsuite/gcc.target/i386/float16-3a.c
> create mode 100644 gcc/testsuite/gcc.target/i386/float16-3b.c
> create mode 100644 gcc/testsuite/gcc.target/i386/float16-4a.c
> create mode 100644 gcc/testsuite/gcc.target/i386/float16-4b.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr54855-12.c
>
> diff --git a/gcc/common/config/i386/cpuinfo.h
> b/gcc/common/config/i386/cpuinfo.h
> index 458f41de776..1835ac64e67 100644
> --- a/gcc/common/config/i386/cpuinfo.h
> +++ b/gcc/common/config/i386/cpuinfo.h
> @@ -731,6 +731,8 @@ get_available_features (struct __processor_model
> *cpu_model,
> set_feature (FEATURE_AVX5124FMAPS);
> if (edx & bit_AVX512VP2INTERSECT)
> set_feature (FEATURE_AVX512VP2INTERSECT);
> + if (edx & bit_AVX512FP16)
> + set_feature (FEATURE_AVX512FP16);
> }
>
> __cpuid_count (7, 1, eax, ebx, ecx, edx);
> diff --git a/gcc/common/config/i386/i386-common.c
> b/gcc/common/config/i386/i386-common.c
> index 76ab1a14e54..00c65ba15ab 100644
> --- a/gcc/common/config/i386/i386-common.c
> +++ b/gcc/common/config/i386/i386-common.c
> @@ -82,6 +82,8 @@ along with GCC; see the file COPYING3. If not see
> #define OPTION_MASK_ISA2_AVX5124VNNIW_SET OPTION_MASK_ISA2_AVX5124VNNIW
> #define OPTION_MASK_ISA_AVX512VBMI2_SET \
> (OPTION_MASK_ISA_AVX512VBMI2 | OPTION_MASK_ISA_AVX512F_SET)
> +#define OPTION_MASK_ISA_AVX512FP16_SET OPTION_MASK_ISA_AVX512BW_SET
> +#define OPTION_MASK_ISA2_AVX512FP16_SET OPTION_MASK_ISA2_AVX512FP16
> #define OPTION_MASK_ISA_AVX512VNNI_SET \
> (OPTION_MASK_ISA_AVX512VNNI | OPTION_MASK_ISA_AVX512F_SET)
> #define OPTION_MASK_ISA2_AVXVNNI_SET OPTION_MASK_ISA2_AVXVNNI
> @@ -231,6 +233,8 @@ along with GCC; see the file COPYING3. If not see
> #define OPTION_MASK_ISA2_AVX5124FMAPS_UNSET OPTION_MASK_ISA2_AVX5124FMAPS
> #define OPTION_MASK_ISA2_AVX5124VNNIW_UNSET OPTION_MASK_ISA2_AVX5124VNNIW
> #define OPTION_MASK_ISA_AVX512VBMI2_UNSET OPTION_MASK_ISA_AVX512VBMI2
> +#define OPTION_MASK_ISA_AVX512FP16_UNSET OPTION_MASK_ISA_AVX512BW_UNSET
> +#define OPTION_MASK_ISA2_AVX512FP16_UNSET OPTION_MASK_ISA2_AVX512FP16
> #define OPTION_MASK_ISA_AVX512VNNI_UNSET OPTION_MASK_ISA_AVX512VNNI
> #define OPTION_MASK_ISA2_AVXVNNI_UNSET OPTION_MASK_ISA2_AVXVNNI
> #define OPTION_MASK_ISA_AVX512VPOPCNTDQ_UNSET OPTION_MASK_ISA_AVX512VPOPCNTDQ
> @@ -313,7 +317,8 @@ along with GCC; see the file COPYING3. If not see
> (OPTION_MASK_ISA2_AVX512BF16_UNSET \
> | OPTION_MASK_ISA2_AVX5124FMAPS_UNSET \
> | OPTION_MASK_ISA2_AVX5124VNNIW_UNSET \
> - | OPTION_MASK_ISA2_AVX512VP2INTERSECT_UNSET)
> + | OPTION_MASK_ISA2_AVX512VP2INTERSECT_UNSET \
> + | OPTION_MASK_ISA2_AVX512FP16_UNSET)
> #define OPTION_MASK_ISA2_GENERAL_REGS_ONLY_UNSET \
> (OPTION_MASK_ISA2_AVX512F_UNSET)
> #define OPTION_MASK_ISA2_AVX_UNSET OPTION_MASK_ISA2_AVX2_UNSET
> @@ -326,7 +331,9 @@ along with GCC; see the file COPYING3. If not see
> (OPTION_MASK_ISA2_SSE3_UNSET | OPTION_MASK_ISA2_KL_UNSET)
> #define OPTION_MASK_ISA2_SSE_UNSET OPTION_MASK_ISA2_SSE2_UNSET
>
> -#define OPTION_MASK_ISA2_AVX512BW_UNSET OPTION_MASK_ISA2_AVX512BF16_UNSET
> +#define OPTION_MASK_ISA2_AVX512BW_UNSET \
> + (OPTION_MASK_ISA2_AVX512BF16_UNSET \
> + | OPTION_MASK_ISA2_AVX512FP16_UNSET)
>
> /* Set 1 << value as value of -malign-FLAG option. */
>
> @@ -853,6 +860,21 @@ ix86_handle_option (struct gcc_options *opts,
> }
> return true;
>
> + case OPT_mavx512fp16:
> + if (value)
> + {
> + opts->x_ix86_isa_flags2 |= OPTION_MASK_ISA2_AVX512FP16_SET;
> + opts->x_ix86_isa_flags2_explicit |= OPTION_MASK_ISA2_AVX512FP16_SET;
> + opts->x_ix86_isa_flags |= OPTION_MASK_ISA_AVX512FP16_SET;
> + opts->x_ix86_isa_flags_explicit |= OPTION_MASK_ISA_AVX512FP16_SET;
> + }
> + else
> + {
> + opts->x_ix86_isa_flags2 &= ~OPTION_MASK_ISA2_AVX512FP16_UNSET;
> + opts->x_ix86_isa_flags2_explicit |=
> OPTION_MASK_ISA2_AVX512FP16_UNSET;
> + }
> + return true;
> +
> case OPT_mavx512vnni:
> if (value)
> {
> diff --git a/gcc/common/config/i386/i386-cpuinfo.h
> b/gcc/common/config/i386/i386-cpuinfo.h
> index e68dd656046..4e0659fc7b2 100644
> --- a/gcc/common/config/i386/i386-cpuinfo.h
> +++ b/gcc/common/config/i386/i386-cpuinfo.h
> @@ -228,6 +228,7 @@ enum processor_features
> FEATURE_AESKLE,
> FEATURE_WIDEKL,
> FEATURE_AVXVNNI,
> + FEATURE_AVX512FP16,
> CPU_FEATURE_MAX
> };
>
> diff --git a/gcc/common/config/i386/i386-isas.h
> b/gcc/common/config/i386/i386-isas.h
> index 898c18f3dda..a6783660278 100644
> --- a/gcc/common/config/i386/i386-isas.h
> +++ b/gcc/common/config/i386/i386-isas.h
> @@ -169,4 +169,5 @@ ISA_NAMES_TABLE_START
> ISA_NAMES_TABLE_ENTRY("aeskle", FEATURE_AESKLE, P_NONE, NULL)
> ISA_NAMES_TABLE_ENTRY("widekl", FEATURE_WIDEKL, P_NONE, "-mwidekl")
> ISA_NAMES_TABLE_ENTRY("avxvnni", FEATURE_AVXVNNI, P_NONE, "-mavxvnni")
> + ISA_NAMES_TABLE_ENTRY("avx512fp16", FEATURE_AVX512FP16, P_NONE,
> "-mavx512fp16")
> ISA_NAMES_TABLE_END
> diff --git a/gcc/config.gcc b/gcc/config.gcc
> index 3df9b52cf25..a354351408c 100644
> --- a/gcc/config.gcc
> +++ b/gcc/config.gcc
> @@ -416,7 +416,7 @@ i[34567]86-*-* | x86_64-*-*)
> tsxldtrkintrin.h amxtileintrin.h amxint8intrin.h
> amxbf16intrin.h x86gprintrin.h uintrintrin.h
> hresetintrin.h keylockerintrin.h avxvnniintrin.h
> - mwaitintrin.h"
> + mwaitintrin.h avx512fp16intrin.h"
> ;;
> ia64-*-*)
> extra_headers=ia64intrin.h
> diff --git a/gcc/config/i386/avx512fp16intrin.h
> b/gcc/config/i386/avx512fp16intrin.h
> new file mode 100644
> index 00000000000..38d63161ba6
> --- /dev/null
> +++ b/gcc/config/i386/avx512fp16intrin.h
> @@ -0,0 +1,53 @@
> +/* Copyright (C) 2019 Free Software Foundation, Inc.
> +
> + This file is part of GCC.
> +
> + GCC is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3, or (at your option)
> + any later version.
> +
> + GCC is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + Under Section 7 of GPL version 3, you are granted additional
> + permissions described in the GCC Runtime Library Exception, version
> + 3.1, as published by the Free Software Foundation.
> +
> + You should have received a copy of the GNU General Public License and
> + a copy of the GCC Runtime Library Exception along with this program;
> + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
> + <http://www.gnu.org/licenses/>. */
> +
> +#ifndef _IMMINTRIN_H_INCLUDED
> +#error "Never use <avx512fp16intrin.h> directly; include <immintrin.h>
> instead."
> +#endif
> +
> +#ifndef __AVX512FP16INTRIN_H_INCLUDED
> +#define __AVX512FP16INTRIN_H_INCLUDED
> +
> +#ifndef __AVX512FP16__
> +#pragma GCC push_options
> +#pragma GCC target("avx512fp16")
> +#define __DISABLE_AVX512FP16__
> +#endif /* __AVX512FP16__ */
> +
> +/* Internal data types for implementing the intrinsics. */
> +typedef _Float16 __v8hf __attribute__ ((__vector_size__ (16)));
> +typedef _Float16 __v16hf __attribute__ ((__vector_size__ (32)));
> +typedef _Float16 __v32hf __attribute__ ((__vector_size__ (64)));
> +
> +/* The Intel API is flexible enough that we must allow aliasing with other
> + vector types, and their scalar components. */
> +typedef _Float16 __m128h __attribute__ ((__vector_size__ (16),
> __may_alias__));
> +typedef _Float16 __m256h __attribute__ ((__vector_size__ (32),
> __may_alias__));
> +typedef _Float16 __m512h __attribute__ ((__vector_size__ (64),
> __may_alias__));
> +
> +#ifdef __DISABLE_AVX512FP16__
> +#undef __DISABLE_AVX512FP16__
> +#pragma GCC pop_options
> +#endif /* __DISABLE_AVX512FP16__ */
> +
> +#endif /* __AVX512FP16INTRIN_H_INCLUDED */
> diff --git a/gcc/config/i386/cpuid.h b/gcc/config/i386/cpuid.h
> index aebc17c6827..82b8050028b 100644
> --- a/gcc/config/i386/cpuid.h
> +++ b/gcc/config/i386/cpuid.h
> @@ -126,6 +126,7 @@
> #define bit_AVX5124VNNIW (1 << 2)
> #define bit_AVX5124FMAPS (1 << 3)
> #define bit_AVX512VP2INTERSECT (1 << 8)
> +#define bit_AVX512FP16 (1 << 23)
> #define bit_IBT (1 << 20)
> #define bit_UINTR (1 << 5)
> #define bit_PCONFIG (1 << 18)
> diff --git a/gcc/config/i386/i386-builtin-types.def
> b/gcc/config/i386/i386-builtin-types.def
> index 3ca313c19ec..1768b88d748 100644
> --- a/gcc/config/i386/i386-builtin-types.def
> +++ b/gcc/config/i386/i386-builtin-types.def
> @@ -68,6 +68,7 @@ DEF_PRIMITIVE_TYPE (UINT8, unsigned_char_type_node)
> DEF_PRIMITIVE_TYPE (UINT16, short_unsigned_type_node)
> DEF_PRIMITIVE_TYPE (INT64, long_long_integer_type_node)
> DEF_PRIMITIVE_TYPE (UINT64, long_long_unsigned_type_node)
> +DEF_PRIMITIVE_TYPE (FLOAT16, ix86_float16_type_node)
> DEF_PRIMITIVE_TYPE (FLOAT, float_type_node)
> DEF_PRIMITIVE_TYPE (DOUBLE, double_type_node)
> DEF_PRIMITIVE_TYPE (FLOAT80, float80_type_node)
> diff --git a/gcc/config/i386/i386-builtins.c b/gcc/config/i386/i386-builtins.c
> index 204e2903126..668f09f12a0 100644
> --- a/gcc/config/i386/i386-builtins.c
> +++ b/gcc/config/i386/i386-builtins.c
> @@ -125,6 +125,7 @@ BDESC_VERIFYS (IX86_BUILTIN_MAX,
> /* Table for the ix86 builtin non-function types. */
> static GTY(()) tree ix86_builtin_type_tab[(int) IX86_BT_LAST_CPTR + 1];
>
> +tree ix86_float16_type_node = NULL_TREE;
> /* Retrieve an element from the above table, building some of
> the types lazily. */
>
> @@ -1343,6 +1344,26 @@ ix86_init_builtins_va_builtins_abi (void)
> BUILT_IN_VA_COPY, BUILT_IN_NORMAL, NULL, fnattr_sysv);
> }
>
> +static void
> +ix86_init_float16_builtins (void)
> +{
> + /* Provide the _Float16 type and float16_type_node if needed so that
> + it can be used in AVX512FP16 intrinsics and builtins. */
> + if (!float16_type_node)
> + {
> + ix86_float16_type_node = make_node (REAL_TYPE);
> + TYPE_PRECISION (ix86_float16_type_node) = 16;
> + SET_TYPE_MODE (ix86_float16_type_node, HFmode);
> + layout_type (ix86_float16_type_node);
> + }
> + else
> + ix86_float16_type_node = float16_type_node;
> +
> + if (!maybe_get_identifier ("_Float16") && TARGET_SSE2)
> + lang_hooks.types.register_builtin_type (ix86_float16_type_node,
> + "_Float16");
> +}
> +
> static void
> ix86_init_builtin_types (void)
> {
> @@ -1371,6 +1392,8 @@ ix86_init_builtin_types (void)
> it. */
> lang_hooks.types.register_builtin_type (float128_type_node, "__float128");
>
> + ix86_init_float16_builtins ();
> +
> const_string_type_node
> = build_pointer_type (build_qualified_type
> (char_type_node, TYPE_QUAL_CONST));
> diff --git a/gcc/config/i386/i386-c.c b/gcc/config/i386/i386-c.c
> index 5ed0de006fb..cc64f855ecc 100644
> --- a/gcc/config/i386/i386-c.c
> +++ b/gcc/config/i386/i386-c.c
> @@ -598,6 +598,8 @@ ix86_target_macros_internal (HOST_WIDE_INT isa_flag,
> def_or_undef (parse_in, "__PTWRITE__");
> if (isa_flag2 & OPTION_MASK_ISA2_AVX512BF16)
> def_or_undef (parse_in, "__AVX512BF16__");
> + if (isa_flag2 & OPTION_MASK_ISA2_AVX512FP16)
> + def_or_undef (parse_in, "__AVX512FP16__");
> if (TARGET_MMX_WITH_SSE)
> def_or_undef (parse_in, "__MMX_WITH_SSE__");
> if (isa_flag2 & OPTION_MASK_ISA2_ENQCMD)
> diff --git a/gcc/config/i386/i386-expand.c b/gcc/config/i386/i386-expand.c
> index 69ea79e6123..b7d050a1e42 100644
> --- a/gcc/config/i386/i386-expand.c
> +++ b/gcc/config/i386/i386-expand.c
> @@ -2314,6 +2314,7 @@ ix86_expand_branch (enum rtx_code code, rtx op0, rtx
> op1, rtx label)
>
> switch (mode)
> {
> + case E_HFmode:
> case E_SFmode:
> case E_DFmode:
> case E_XFmode:
> @@ -2627,7 +2628,7 @@ ix86_prepare_fp_compare_args (enum rtx_code code, rtx
> *pop0, rtx *pop1)
> bool unordered_compare = ix86_unordered_fp_compare (code);
> rtx op0 = *pop0, op1 = *pop1;
> machine_mode op_mode = GET_MODE (op0);
> - bool is_sse = TARGET_SSE_MATH && SSE_FLOAT_MODE_P (op_mode);
> + bool is_sse = SSE_FLOAT_MODE_SSEMATH_OR_HF_P (op_mode);
>
> /* All of the unordered compare instructions only work on registers.
> The same is true of the fcomi compare instructions. The XFmode
> @@ -4112,7 +4113,7 @@ ix86_expand_fp_movcc (rtx operands[])
> rtx op0 = XEXP (operands[1], 0);
> rtx op1 = XEXP (operands[1], 1);
>
> - if (TARGET_SSE_MATH && SSE_FLOAT_MODE_P (mode))
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> {
> machine_mode cmode;
>
> diff --git a/gcc/config/i386/i386-isa.def b/gcc/config/i386/i386-isa.def
> index a0d46cbc892..83d9302ea3d 100644
> --- a/gcc/config/i386/i386-isa.def
> +++ b/gcc/config/i386/i386-isa.def
> @@ -108,3 +108,4 @@ DEF_PTA(HRESET)
> DEF_PTA(KL)
> DEF_PTA(WIDEKL)
> DEF_PTA(AVXVNNI)
> +DEF_PTA(AVX512FP16)
> diff --git a/gcc/config/i386/i386-options.c b/gcc/config/i386/i386-options.c
> index 3416a4f1752..df191763e4b 100644
> --- a/gcc/config/i386/i386-options.c
> +++ b/gcc/config/i386/i386-options.c
> @@ -223,7 +223,8 @@ static struct ix86_target_opts isa2_opts[] =
> { "-mhreset", OPTION_MASK_ISA2_HRESET },
> { "-mkl", OPTION_MASK_ISA2_KL },
> { "-mwidekl", OPTION_MASK_ISA2_WIDEKL },
> - { "-mavxvnni", OPTION_MASK_ISA2_AVXVNNI }
> + { "-mavxvnni", OPTION_MASK_ISA2_AVXVNNI },
> + { "-mavx512fp16", OPTION_MASK_ISA2_AVX512FP16 }
> };
> static struct ix86_target_opts isa_opts[] =
> {
> @@ -1045,6 +1046,7 @@ ix86_valid_target_attribute_inner_p (tree fndecl, tree
> args, char *p_strings[],
> IX86_ATTR_ISA ("amx-bf16", OPT_mamx_bf16),
> IX86_ATTR_ISA ("hreset", OPT_mhreset),
> IX86_ATTR_ISA ("avxvnni", OPT_mavxvnni),
> + IX86_ATTR_ISA ("avx512fp16", OPT_mavx512fp16),
>
> /* enum options */
> IX86_ATTR_ENUM ("fpmath=", OPT_mfpmath_),
> diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
> index dc673c89bc8..71bbcf968c5 100644
> --- a/gcc/config/i386/i386.c
> +++ b/gcc/config/i386/i386.c
> @@ -5497,6 +5497,14 @@ ix86_output_ssemov (rtx_insn *insn, rtx *operands)
> case MODE_SI:
> return "%vmovd\t{%1, %0|%0, %1}";
>
> + case MODE_HI:
> + if (GENERAL_REG_P (operands[0]))
> + return "vmovw\t{%1, %k0|%k0, %1}";
> + else if (GENERAL_REG_P (operands[1]))
> + return "vmovw\t{%k1, %0|%0, %k1}";
> + else
> + return "vmovw\t{%1, %0|%0, %1}";
> +
> case MODE_DF:
> if (TARGET_AVX && REG_P (operands[0]) && REG_P (operands[1]))
> return "vmovsd\t{%d1, %0|%0, %d1}";
> @@ -5509,6 +5517,12 @@ ix86_output_ssemov (rtx_insn *insn, rtx *operands)
> else
> return "%vmovss\t{%1, %0|%0, %1}";
>
> + case MODE_HF:
> + if (REG_P (operands[0]) && REG_P (operands[1]))
> + return "vmovsh\t{%d1, %0|%0, %d1}";
> + else
> + return "vmovsh\t{%1, %0|%0, %1}";
> +
> case MODE_V1DF:
> gcc_assert (!TARGET_AVX);
> return "movlpd\t{%1, %0|%0, %1}";
> @@ -13955,7 +13969,7 @@ output_387_binary_op (rtx_insn *insn, rtx *operands)
>
> if (is_sse)
> {
> - p = (GET_MODE (operands[0]) == SFmode) ? "ss" : "sd";
> + p = (GET_MODE (operands[0]) == SFmode ? "ss" : "sd");
> strcat (buf, p);
>
> if (TARGET_AVX)
> @@ -19132,10 +19146,19 @@ inline_secondary_memory_needed (machine_mode mode,
> reg_class_t class1,
> if (!TARGET_SSE2)
> return true;
>
> + if (!(INTEGER_CLASS_P (class1) || INTEGER_CLASS_P (class2)))
> + return true;
> +
> + int msize = GET_MODE_SIZE (mode);
> +
> /* Between SSE and general, we have moves no larger than word size. */
> - if (!(INTEGER_CLASS_P (class1) || INTEGER_CLASS_P (class2))
> - || GET_MODE_SIZE (mode) < GET_MODE_SIZE (SImode)
> - || GET_MODE_SIZE (mode) > UNITS_PER_WORD)
> + if (msize > UNITS_PER_WORD)
> + return true;
> +
> + /* In addition to SImode moves, AVX512FP16 also enables HImode moves.
> */
> + int minsize = GET_MODE_SIZE (TARGET_AVX512FP16 ? HImode : SImode);
> +
> + if (msize < minsize)
> return true;
>
> /* If the target says that inter-unit moves are more expensive
> @@ -19229,21 +19252,26 @@ ix86_can_change_mode_class (machine_mode from,
> machine_mode to,
> static inline int
> sse_store_index (machine_mode mode)
> {
> - switch (GET_MODE_SIZE (mode))
> - {
> - case 4:
> - return 0;
> - case 8:
> - return 1;
> - case 16:
> - return 2;
> - case 32:
> - return 3;
> - case 64:
> - return 4;
> - default:
> - return -1;
> - }
> + /* NB: Use SFmode cost for HFmode instead of adding HFmode load/store
> + costs to processor_costs, which requires changes to all entries in
> + processor cost table. */
> + if (mode == E_HFmode)
> + mode = E_SFmode;
> + switch (GET_MODE_SIZE (mode))
> + {
> + case 4:
> + return 0;
> + case 8:
> + return 1;
> + case 16:
> + return 2;
> + case 32:
> + return 3;
> + case 64:
> + return 4;
> + default:
> + return -1;
> + }
> }
>
> /* Return the cost of moving data of mode M between a
> @@ -19270,6 +19298,7 @@ inline_memory_move_cost (machine_mode mode, enum
> reg_class regclass, int in)
> int index;
> switch (mode)
> {
> + case E_HFmode:
> case E_SFmode:
> index = 0;
> break;
> @@ -19370,11 +19399,31 @@ inline_memory_move_cost (machine_mode mode, enum
> reg_class regclass, int in)
> }
> break;
> case 2:
> - if (in == 2)
> - return MAX (ix86_cost->hard_register.int_load[1],
> - ix86_cost->hard_register.int_store[1]);
> - return in ? ix86_cost->hard_register.int_load[1]
> - : ix86_cost->hard_register.int_store[1];
> + {
> + int cost;
> + if (in == 2)
> + cost = MAX (ix86_cost->hard_register.int_load[1],
> + ix86_cost->hard_register.int_store[1]);
> + else
> + cost = in ? ix86_cost->hard_register.int_load[1]
> + : ix86_cost->hard_register.int_store[1];
> + if (mode == E_HFmode)
> + {
> + /* Prefer SSE over GPR for HFmode. */
> + int sse_cost;
> + int index = sse_store_index (mode);
> + if (in == 2)
> + sse_cost = MAX (ix86_cost->hard_register.sse_load[index],
> + ix86_cost->hard_register.sse_store[index]);
> + else
> + sse_cost = (in
> + ? ix86_cost->hard_register.sse_load [index]
> + : ix86_cost->hard_register.sse_store [index]);
> + if (sse_cost >= cost)
> + cost = sse_cost + 1;
> + }
> + return cost;
> + }
> default:
> if (in == 2)
> cost = MAX (ix86_cost->hard_register.int_load[2],
> @@ -19548,6 +19597,8 @@ ix86_hard_regno_mode_ok (unsigned int regno,
> machine_mode mode)
> - XI mode
> - any of 512-bit wide vector mode
> - any scalar mode. */
> + /* For AVX512FP16, vmovw supports movement of HImode
> + between gpr and sse registser. */
> if (TARGET_AVX512F
> && (mode == XImode
> || VALID_AVX512F_REG_MODE (mode)
> @@ -19831,7 +19882,7 @@ ix86_multiplication_cost (const struct
> processor_costs *cost,
> if (VECTOR_MODE_P (mode))
> inner_mode = GET_MODE_INNER (mode);
>
> - if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> return inner_mode == DFmode ? cost->mulsd : cost->mulss;
> else if (X87_FLOAT_MODE_P (mode))
> return cost->fmul;
> @@ -19883,7 +19934,7 @@ ix86_division_cost (const struct processor_costs
> *cost,
> if (VECTOR_MODE_P (mode))
> inner_mode = GET_MODE_INNER (mode);
>
> - if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> return inner_mode == DFmode ? cost->divsd : cost->divss;
> else if (X87_FLOAT_MODE_P (mode))
> return cost->fdiv;
> @@ -20303,7 +20354,7 @@ ix86_rtx_costs (rtx x, machine_mode mode, int
> outer_code_i, int opno,
> return true;
> }
>
> - if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> {
> *total = cost->addss;
> return false;
> @@ -20336,7 +20387,7 @@ ix86_rtx_costs (rtx x, machine_mode mode, int
> outer_code_i, int opno,
> /* FALLTHRU */
>
> case NEG:
> - if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> {
> *total = cost->sse_op;
> return false;
> @@ -20418,14 +20469,14 @@ ix86_rtx_costs (rtx x, machine_mode mode, int
> outer_code_i, int opno,
> return false;
>
> case FLOAT_EXTEND:
> - if (!(SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH))
> + if (!SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> *total = 0;
> else
> *total = ix86_vec_cost (mode, cost->addss);
> return false;
>
> case FLOAT_TRUNCATE:
> - if (!(SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH))
> + if (!SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> *total = cost->fadd;
> else
> *total = ix86_vec_cost (mode, cost->addss);
> @@ -20435,7 +20486,7 @@ ix86_rtx_costs (rtx x, machine_mode mode, int
> outer_code_i, int opno,
> /* SSE requires memory load for the constant operand. It may make
> sense to account for this. Of course the constant operand may or
> may not be reused. */
> - if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> *total = cost->sse_op;
> else if (X87_FLOAT_MODE_P (mode))
> *total = cost->fabs;
> @@ -20444,7 +20495,7 @@ ix86_rtx_costs (rtx x, machine_mode mode, int
> outer_code_i, int opno,
> return false;
>
> case SQRT:
> - if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> *total = mode == SFmode ? cost->sqrtss : cost->sqrtsd;
> else if (X87_FLOAT_MODE_P (mode))
> *total = cost->fsqrt;
> @@ -21928,6 +21979,10 @@ ix86_mangle_type (const_tree type)
>
> switch (TYPE_MODE (type))
> {
> + case E_HFmode:
> + /* _Float16 is "DF16_".
> + Align with clang's decision in https://reviews.llvm.org/D33719. */
> + return "DF16_";
> case E_TFmode:
> /* __float128 is "g". */
> return "g";
> @@ -22551,7 +22606,7 @@ ix86_add_stmt_cost (class vec_info *vinfo, void
> *data, int count,
> case MINUS_EXPR:
> if (kind == scalar_stmt)
> {
> - if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> stmt_cost = ix86_cost->addss;
> else if (X87_FLOAT_MODE_P (mode))
> stmt_cost = ix86_cost->fadd;
> @@ -22569,7 +22624,7 @@ ix86_add_stmt_cost (class vec_info *vinfo, void
> *data, int count,
> stmt_cost = ix86_multiplication_cost (ix86_cost, mode);
> break;
> case NEGATE_EXPR:
> - if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> stmt_cost = ix86_cost->sse_op;
> else if (X87_FLOAT_MODE_P (mode))
> stmt_cost = ix86_cost->fchs;
> @@ -22625,7 +22680,7 @@ ix86_add_stmt_cost (class vec_info *vinfo, void
> *data, int count,
> case BIT_XOR_EXPR:
> case BIT_AND_EXPR:
> case BIT_NOT_EXPR:
> - if (SSE_FLOAT_MODE_P (mode) && TARGET_SSE_MATH)
> + if (SSE_FLOAT_MODE_SSEMATH_OR_HF_P (mode))
> stmt_cost = ix86_cost->sse_op;
> else if (VECTOR_MODE_P (mode))
> stmt_cost = ix86_vec_cost (mode, ix86_cost->sse_op);
> @@ -23327,14 +23382,18 @@ ix86_get_excess_precision (enum
> excess_precision_type type)
> /* The fastest type to promote to will always be the native type,
> whether that occurs with implicit excess precision or
> otherwise. */
> - return FLT_EVAL_METHOD_PROMOTE_TO_FLOAT;
> + return TARGET_AVX512FP16
> + ? FLT_EVAL_METHOD_PROMOTE_TO_FLOAT16
> + : FLT_EVAL_METHOD_PROMOTE_TO_FLOAT;
> case EXCESS_PRECISION_TYPE_STANDARD:
> case EXCESS_PRECISION_TYPE_IMPLICIT:
> /* Otherwise, the excess precision we want when we are
> in a standards compliant mode, and the implicit precision we
> provide would be identical were it not for the unpredictable
> cases. */
> - if (!TARGET_80387)
> + if (TARGET_AVX512FP16 && TARGET_SSE_MATH)
> + return FLT_EVAL_METHOD_PROMOTE_TO_FLOAT16;
> + else if (!TARGET_80387)
> return FLT_EVAL_METHOD_PROMOTE_TO_FLOAT;
> else if (!TARGET_MIX_SSE_I387)
> {
> diff --git a/gcc/config/i386/i386.h b/gcc/config/i386/i386.h
> index b1e66ee192e..8fcd5693624 100644
> --- a/gcc/config/i386/i386.h
> +++ b/gcc/config/i386/i386.h
> @@ -1000,7 +1000,8 @@ extern const char *host_detect_local_cpu (int argc,
> const char **argv);
>
> #define VALID_AVX512F_SCALAR_MODE(MODE)
> \
> ((MODE) == DImode || (MODE) == DFmode || (MODE) == SImode \
> - || (MODE) == SFmode)
> + || (MODE) == SFmode \
> + || (TARGET_AVX512FP16 && ((MODE) == HImode || (MODE) == HFmode)))
>
> #define VALID_AVX512F_REG_MODE(MODE) \
> ((MODE) == V8DImode || (MODE) == V8DFmode || (MODE) == V64QImode \
> @@ -1039,7 +1040,7 @@ extern const char *host_detect_local_cpu (int argc,
> const char **argv);
>
> #define VALID_FP_MODE_P(MODE) \
> ((MODE) == SFmode || (MODE) == DFmode || (MODE) == XFmode \
> - || (MODE) == SCmode || (MODE) == DCmode || (MODE) == XCmode)
> \
> + || (MODE) == SCmode || (MODE) == DCmode || (MODE) == XCmode)
>
> #define VALID_INT_MODE_P(MODE) \
> ((MODE) == QImode || (MODE) == HImode
> \
> @@ -1072,6 +1073,10 @@ extern const char *host_detect_local_cpu (int argc,
> const char **argv);
> #define SSE_FLOAT_MODE_P(MODE) \
> ((TARGET_SSE && (MODE) == SFmode) || (TARGET_SSE2 && (MODE) == DFmode))
>
> +#define SSE_FLOAT_MODE_SSEMATH_OR_HF_P(MODE) \
> + ((SSE_FLOAT_MODE_P (MODE) && TARGET_SSE_MATH)
> \
> + || (TARGET_AVX512FP16 && (MODE) == HFmode))
> +
> #define FMA4_VEC_FLOAT_MODE_P(MODE) \
> (TARGET_FMA4 && ((MODE) == V4SFmode || (MODE) == V2DFmode \
> || (MODE) == V8SFmode || (MODE) == V4DFmode))
> @@ -2265,7 +2270,7 @@ constexpr wide_int_bitmask PTA_TIGERLAKE =
> PTA_ICELAKE_CLIENT | PTA_MOVDIRI
> constexpr wide_int_bitmask PTA_SAPPHIRERAPIDS = PTA_COOPERLAKE | PTA_MOVDIRI
> | PTA_MOVDIR64B | PTA_AVX512VP2INTERSECT | PTA_ENQCMD | PTA_CLDEMOTE
> | PTA_PTWRITE | PTA_WAITPKG | PTA_SERIALIZE | PTA_TSXLDTRK | PTA_AMX_TILE
> - | PTA_AMX_INT8 | PTA_AMX_BF16 | PTA_UINTR | PTA_AVXVNNI;
> + | PTA_AMX_INT8 | PTA_AMX_BF16 | PTA_UINTR | PTA_AVXVNNI | PTA_AVX512FP16;
> constexpr wide_int_bitmask PTA_KNL = PTA_BROADWELL | PTA_AVX512PF
> | PTA_AVX512ER | PTA_AVX512F | PTA_AVX512CD | PTA_PREFETCHWT1;
> constexpr wide_int_bitmask PTA_BONNELL = PTA_CORE2 | PTA_MOVBE;
> diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
> index d475347172d..777d11261ac 100644
> --- a/gcc/config/i386/i386.md
> +++ b/gcc/config/i386/i386.md
> @@ -496,7 +496,7 @@ (define_attr "type"
>
> ;; Main data type used by the insn
> (define_attr "mode"
> - "unknown,none,QI,HI,SI,DI,TI,OI,XI,SF,DF,XF,TF,V16SF,V8SF,V4DF,V4SF,
> + "unknown,none,QI,HI,SI,DI,TI,OI,XI,HF,SF,DF,XF,TF,V16SF,V8SF,V4DF,V4SF,
> V2DF,V2SF,V1DF,V8DF"
> (const_string "unknown"))
>
> @@ -832,8 +832,7 @@ (define_attr "isa"
> "base,x64,nox64,x64_sse2,x64_sse4,x64_sse4_noavx,
> sse_noavx,sse2,sse2_noavx,sse3,sse3_noavx,sse4,sse4_noavx,
> avx,noavx,avx2,noavx2,bmi,bmi2,fma4,fma,avx512f,noavx512f,
> avx512bw,noavx512bw,avx512dq,noavx512dq,
> - avx512vl,noavx512vl,
> - avxvnni,avx512vnnivl"
> + avx512vl,noavx512vl,avxvnni,avx512vnnivl,avx512fp16"
> (const_string "base"))
>
> ;; Define instruction set of MMX instructions
> @@ -885,6 +884,8 @@ (define_attr "enabled" ""
> (eq_attr "isa" "avxvnni") (symbol_ref "TARGET_AVXVNNI")
> (eq_attr "isa" "avx512vnnivl")
> (symbol_ref "TARGET_AVX512VNNI && TARGET_AVX512VL")
> + (eq_attr "isa" "avx512fp16")
> + (symbol_ref "TARGET_AVX512FP16")
>
> (eq_attr "mmx_isa" "native")
> (symbol_ref "!TARGET_MMX_WITH_SSE")
> @@ -906,6 +907,7 @@ (define_asm_attributes
> (set_attr "type" "multi")])
>
> (define_code_iterator plusminus [plus minus])
> +(define_code_iterator plusminusmultdiv [plus minus mult div])
>
> (define_code_iterator sat_plusminus [ss_plus us_plus ss_minus us_minus])
>
> @@ -921,7 +923,8 @@ (define_code_attr multdiv_mnemonic
>
> ;; Mark commutative operators as such in constraints.
> (define_code_attr comm [(plus "%") (ss_plus "%") (us_plus "%")
> - (minus "") (ss_minus "") (us_minus "")])
> + (minus "") (ss_minus "") (us_minus "")
> + (mult "%") (div "")])
>
> ;; Mapping of max and min
> (define_code_iterator maxmin [smax smin umax umin])
> @@ -1021,7 +1024,8 @@ (define_code_attr insn
> (minus "sub") (ss_minus "sssub") (us_minus "ussub")
> (sign_extend "extend") (zero_extend "zero_extend")
> (ashift "ashl") (lshiftrt "lshr") (ashiftrt "ashr")
> - (rotate "rotl") (rotatert "rotr")])
> + (rotate "rotl") (rotatert "rotr")
> + (mult "mul") (div "div")])
>
> ;; All integer modes.
> (define_mode_iterator SWI1248x [QI HI SI DI])
> @@ -1089,8 +1093,9 @@ (define_mode_iterator SWI48DWI [SI DI (TI
> "TARGET_64BIT")])
> ;; compile time constant, it is faster to use <MODE_SIZE> than
> ;; GET_MODE_SIZE (<MODE>mode). For XFmode which depends on
> ;; command line options just use GET_MODE_SIZE macro.
> -(define_mode_attr MODE_SIZE [(QI "1") (HI "2") (SI "4") (DI "8") (TI "16")
> - (SF "4") (DF "8") (XF "GET_MODE_SIZE (XFmode)")
> +(define_mode_attr MODE_SIZE [(QI "1") (HI "2") (SI "4") (DI "8")
> + (TI "16") (HF "2") (SF "4") (DF "8")
> + (XF "GET_MODE_SIZE (XFmode)")
> (V16QI "16") (V32QI "32") (V64QI "64")
> (V8HI "16") (V16HI "32") (V32HI "64")
> (V4SI "16") (V8SI "32") (V16SI "64")
> @@ -1222,8 +1227,8 @@ (define_mode_iterator MODEF [SF DF])
> ;; All x87 floating point modes
> (define_mode_iterator X87MODEF [SF DF XF])
>
> -;; All x87 floating point modes plus HF
> -(define_mode_iterator X87MODEFH [SF DF XF HF])
> +;; All x87 floating point modes plus HFmode
> +(define_mode_iterator X87MODEFH [HF SF DF XF])
>
> ;; All SSE floating point modes
> (define_mode_iterator SSEMODEF [SF DF TF])
> @@ -1231,7 +1236,7 @@ (define_mode_attr ssevecmodef [(SF "V4SF") (DF "V2DF")
> (TF "TF")])
>
> ;; SSE instruction suffix for various modes
> (define_mode_attr ssemodesuffix
> - [(SF "ss") (DF "sd")
> + [(HF "sh") (SF "ss") (DF "sd")
> (V16SF "ps") (V8DF "pd")
> (V8SF "ps") (V4DF "pd")
> (V4SF "ps") (V2DF "pd")
> @@ -1496,6 +1501,23 @@ (define_expand "cstorexf4"
> DONE;
> })
>
> +(define_expand "cbranchhf4"
> + [(set (reg:CC FLAGS_REG)
> + (compare:CC (match_operand:HF 1 "cmp_fp_expander_operand")
> + (match_operand:HF 2 "cmp_fp_expander_operand")))
> + (set (pc) (if_then_else
> + (match_operator 0 "ix86_fp_comparison_operator"
> + [(reg:CC FLAGS_REG)
> + (const_int 0)])
> + (label_ref (match_operand 3))
> + (pc)))]
> + "TARGET_AVX512FP16"
> +{
> + ix86_expand_branch (GET_CODE (operands[0]),
> + operands[1], operands[2], operands[3]);
> + DONE;
> +})
> +
> (define_expand "cbranch<mode>4"
> [(set (reg:CC FLAGS_REG)
> (compare:CC (match_operand:MODEF 1 "cmp_fp_expander_operand")
> @@ -1705,6 +1727,17 @@ (define_insn "*cmpi<unord><MODEF:mode>"
> (eq_attr "alternative" "0")
> (symbol_ref "true")
> (symbol_ref "false"))))])
> +
> +(define_insn "*cmpi<unord>hf"
> + [(set (reg:CCFP FLAGS_REG)
> + (compare:CCFP
> + (match_operand:HF 0 "register_operand" "v")
> + (match_operand:HF 1 "nonimmediate_operand" "vm")))]
> + "TARGET_AVX512FP16"
> + "v<unord>comish\t{%1, %0|%0, %1}"
> + [(set_attr "type" "ssecomi")
> + (set_attr "prefix" "evex")
> + (set_attr "mode" "HF")])
>
> ;; Push/pop instructions.
>
> @@ -2436,8 +2469,8 @@ (define_insn "*movsi_internal"
> (symbol_ref "true")))])
>
> (define_insn "*movhi_internal"
> - [(set (match_operand:HI 0 "nonimmediate_operand" "=r,r ,r ,m ,*k,*k
> ,*r,*m,*k")
> - (match_operand:HI 1 "general_operand" "r
> ,rn,rm,rn,*r,*km,*k,*k,CBC"))]
> + [(set (match_operand:HI 0 "nonimmediate_operand" "=r,r ,r ,m ,*k,*k
> ,*r,*m,*k,?r,?v,*v,*v,*m")
> + (match_operand:HI 1 "general_operand" "r
> ,rn,rm,rn,*r,*km,*k,*k,CBC,v, r, v, m, v"))]
> "!(MEM_P (operands[0]) && MEM_P (operands[1]))
> && ix86_hardreg_mov_ok (operands[0], operands[1])"
>
> @@ -2463,6 +2496,9 @@ (define_insn "*movhi_internal"
> gcc_unreachable ();
> }
>
> + case TYPE_SSEMOV:
> + return ix86_output_ssemov (insn, operands);
> +
> case TYPE_MSKLOG:
> if (operands[1] == const0_rtx)
> return "kxorw\t%0, %0, %0";
> @@ -2477,8 +2513,15 @@ (define_insn "*movhi_internal"
> return "mov{w}\t{%1, %0|%0, %1}";
> }
> }
> - [(set (attr "type")
> - (cond [(eq_attr "alternative" "4,5,6,7")
> + [(set (attr "isa")
> + (cond [(eq_attr "alternative" "9,10,11,12,13")
> + (const_string "avx512fp16")
> + ]
> + (const_string "*")))
> + (set (attr "type")
> + (cond [(eq_attr "alternative" "9,10,11,12,13")
> + (const_string "ssemov")
> + (eq_attr "alternative" "4,5,6,7")
> (const_string "mskmov")
> (eq_attr "alternative" "8")
> (const_string "msklog")
> @@ -2503,6 +2546,8 @@ (define_insn "*movhi_internal"
> (set (attr "mode")
> (cond [(eq_attr "type" "imovx")
> (const_string "SI")
> + (eq_attr "alternative" "11")
> + (const_string "HF")
> (and (eq_attr "alternative" "1,2")
> (match_operand:HI 1 "aligned_operand"))
> (const_string "SI")
> @@ -3727,7 +3772,10 @@ (define_insn "*movhf_internal"
> (eq_attr "alternative" "2")
> (const_string "sselog1")
> (eq_attr "alternative" "4,5,6,7")
> - (const_string "sselog")
> + (if_then_else
> + (match_test ("TARGET_AVX512FP16"))
> + (const_string "ssemov")
> + (const_string "sselog"))
> ]
> (const_string "ssemov")))
> (set (attr "memory")
> @@ -3750,9 +3798,15 @@ (define_insn "*movhf_internal"
> (eq_attr "alternative" "2")
> (const_string "V4SF")
> (eq_attr "alternative" "4,5,6,7")
> - (const_string "TI")
> + (if_then_else
> + (match_test "TARGET_AVX512FP16")
> + (const_string "HI")
> + (const_string "TI"))
> (eq_attr "alternative" "3")
> - (const_string "SF")
> + (if_then_else
> + (match_test "TARGET_AVX512FP16")
> + (const_string "HF")
> + (const_string "SF"))
> ]
> (const_string "*")))])
>
> @@ -4493,6 +4547,17 @@ (define_split
> emit_move_insn (operands[0], CONST0_RTX (V2DFmode));
> })
>
> +(define_insn "extendhf<mode>2"
> + [(set (match_operand:MODEF 0 "nonimm_ssenomem_operand" "=v")
> + (float_extend:MODEF
> + (match_operand:HF 1 "nonimmediate_operand" "vm")))]
> + "TARGET_AVX512FP16"
> + "vcvtsh2<ssemodesuffix>\t{%1, %0, %0|%0, %0, %1}"
> + [(set_attr "type" "ssecvt")
> + (set_attr "prefix" "evex")
> + (set_attr "mode" "<MODE>")])
> +
> +
> (define_expand "extend<mode>xf2"
> [(set (match_operand:XF 0 "nonimmediate_operand")
> (float_extend:XF (match_operand:MODEF 1 "general_operand")))]
> @@ -4670,6 +4735,18 @@ (define_insn "truncxf<mode>2"
> (symbol_ref "flag_unsafe_math_optimizations")
> ]
> (symbol_ref "true")))])
> +
> +;; Conversion from {SF,DF}mode to HFmode.
> +
> +(define_insn "trunc<mode>hf2"
> + [(set (match_operand:HF 0 "register_operand" "=v")
> + (float_truncate:HF
> + (match_operand:MODEF 1 "nonimmediate_operand" "vm")))]
> + "TARGET_AVX512FP16"
> + "vcvt<ssemodesuffix>2sh\t{%1, %d0|%d0, %1}"
> + [(set_attr "type" "ssecvt")
> + (set_attr "prefix" "evex")
> + (set_attr "mode" "HF")])
>
> ;; Signed conversion to DImode.
>
> @@ -5046,6 +5123,16 @@ (define_insn "*float<SWI48:mode><MODEF:mode>2"
> (symbol_ref "TARGET_INTER_UNIT_CONVERSIONS")]
> (symbol_ref "true")))])
>
> +(define_insn "float<floatunssuffix><mode>hf2"
> + [(set (match_operand:HF 0 "register_operand" "=v")
> + (any_float:HF
> + (match_operand:SWI48 1 "nonimmediate_operand" "rm")))]
> + "TARGET_AVX512FP16"
> + "vcvt<floatsuffix>si2sh<rex64suffix>\t{%1, %d0|%d0, %1}"
> + [(set_attr "type" "sseicvt")
> + (set_attr "prefix" "evex")
> + (set_attr "mode" "HF")])
> +
> (define_insn "*floatdi<MODEF:mode>2_i387"
> [(set (match_operand:MODEF 0 "register_operand" "=f")
> (float:MODEF (match_operand:DI 1 "nonimmediate_operand" "m")))]
> @@ -7626,6 +7713,13 @@ (define_expand "<insn>xf3"
> (match_operand:XF 2 "register_operand")))]
> "TARGET_80387")
>
> +(define_expand "<insn>hf3"
> + [(set (match_operand:HF 0 "register_operand")
> + (plusminus:HF
> + (match_operand:HF 1 "register_operand")
> + (match_operand:HF 2 "nonimmediate_operand")))]
> + "TARGET_AVX512FP16")
> +
> (define_expand "<insn><mode>3"
> [(set (match_operand:MODEF 0 "register_operand")
> (plusminus:MODEF
> @@ -8203,6 +8297,12 @@ (define_expand "mulxf3"
> (match_operand:XF 2 "register_operand")))]
> "TARGET_80387")
>
> +(define_expand "mulhf3"
> + [(set (match_operand:HF 0 "register_operand")
> + (mult:HF (match_operand:HF 1 "register_operand")
> + (match_operand:HF 2 "nonimmediate_operand")))]
> + "TARGET_AVX512FP16")
> +
> (define_expand "mul<mode>3"
> [(set (match_operand:MODEF 0 "register_operand")
> (mult:MODEF (match_operand:MODEF 1 "register_operand")
> @@ -8220,6 +8320,12 @@ (define_expand "divxf3"
> (match_operand:XF 2 "register_operand")))]
> "TARGET_80387")
>
> +(define_expand "divhf3"
> + [(set (match_operand:HF 0 "register_operand")
> + (div:HF (match_operand:HF 1 "register_operand")
> + (match_operand:HF 2 "nonimmediate_operand")))]
> + "TARGET_AVX512FP16")
> +
> (define_expand "div<mode>3"
> [(set (match_operand:MODEF 0 "register_operand")
> (div:MODEF (match_operand:MODEF 1 "register_operand")
> @@ -16312,6 +16418,17 @@ (define_insn "*fop_<mode>_comm"
> (symbol_ref "true")
> (symbol_ref "false"))))])
>
> +(define_insn "*<insn>hf"
> + [(set (match_operand:HF 0 "register_operand" "=v")
> + (plusminusmultdiv:HF
> + (match_operand:HF 1 "nonimmediate_operand" "<comm>v")
> + (match_operand:HF 2 "nonimmediate_operand" "vm")))]
> + "TARGET_AVX512FP16
> + && !(MEM_P (operands[1]) && MEM_P (operands[2]))"
> + "v<insn>sh\t{%2, %1, %0|%0, %1, %2}"
> + [(set_attr "prefix" "evex")
> + (set_attr "mode" "HF")])
> +
> (define_insn "*rcpsf2_sse"
> [(set (match_operand:SF 0 "register_operand" "=x,x,x")
> (unspec:SF [(match_operand:SF 1 "nonimmediate_operand" "0,x,m")]
> @@ -19178,6 +19295,15 @@ (define_peephole2
> gcc_unreachable ();
> })
>
> +(define_expand "movhfcc"
> + [(set (match_operand:HF 0 "register_operand")
> + (if_then_else:HF
> + (match_operand 1 "comparison_operator")
> + (match_operand:HF 2 "register_operand")
> + (match_operand:HF 3 "register_operand")))]
> + "TARGET_AVX512FP16"
> + "if (ix86_expand_fp_movcc (operands)) DONE; else FAIL;")
> +
> (define_expand "mov<mode>cc"
> [(set (match_operand:X87MODEF 0 "register_operand")
> (if_then_else:X87MODEF
> @@ -19346,6 +19472,18 @@ (define_insn "<code><mode>3"
> ;; Their operands are not commutative, and thus they may be used in the
> ;; presence of -0.0 and NaN.
>
> +(define_insn "*ieee_s<ieee_maxmin>hf3"
> + [(set (match_operand:HF 0 "register_operand" "=v")
> + (unspec:HF
> + [(match_operand:HF 1 "register_operand" "v")
> + (match_operand:HF 2 "nonimmediate_operand" "vm")]
> + IEEE_MAXMIN))]
> + "TARGET_AVX512FP16"
> + "v<ieee_maxmin>sh\t{%2, %1, %0|%0, %1, %2}"
> + [(set_attr "prefix" "evex")
> + (set_attr "type" "sseadd")
> + (set_attr "mode" "HF")])
> +
> (define_insn "*ieee_s<ieee_maxmin><mode>3"
> [(set (match_operand:MODEF 0 "register_operand" "=x,v")
> (unspec:MODEF
> diff --git a/gcc/config/i386/i386.opt b/gcc/config/i386/i386.opt
> index 7b8547bb1c3..ad366974b5b 100644
> --- a/gcc/config/i386/i386.opt
> +++ b/gcc/config/i386/i386.opt
> @@ -1166,3 +1166,7 @@ Emit GNU_PROPERTY_X86_ISA_1_NEEDED GNU property.
> mmwait
> Target Mask(ISA2_MWAIT) Var(ix86_isa_flags2) Save
> Support MWAIT and MONITOR built-in functions and code generation.
> +
> +mavx512fp16
> +Target Mask(ISA2_AVX512FP16) Var(ix86_isa_flags2) Save
> +Support MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, AVX, AVX2, AVX512F and
> AVX512FP16 built-in functions and code generation.
> diff --git a/gcc/config/i386/immintrin.h b/gcc/config/i386/immintrin.h
> index f129de4bbe5..2421a78637b 100644
> --- a/gcc/config/i386/immintrin.h
> +++ b/gcc/config/i386/immintrin.h
> @@ -94,6 +94,10 @@
>
> #include <avx512vp2intersectvlintrin.h>
>
> +#ifdef __SSE2__
> +#include <avx512fp16intrin.h>
> +#endif
> +
> #include <shaintrin.h>
>
> #include <fmaintrin.h>
> diff --git a/gcc/doc/extend.texi b/gcc/doc/extend.texi
> index 3a1978efc97..09040bfca33 100644
> --- a/gcc/doc/extend.texi
> +++ b/gcc/doc/extend.texi
> @@ -1164,6 +1164,14 @@ to inconsistent behavior between software emulation
> and AVX512-FP16
> instructions. Using @option{-fexcess-precision=16} and will force round
> back after each operation.
>
> +Using @option{-mavx512fp16} will generate AVX512-FP16 instructions instead of
> +software emulation. The default behavior of @code{FLT_EVAL_METHOD} is to
> round
> +after each operation. The same is true with
> @option{-fexcess-precision=standard}
> +and @option{-mfpmath=sse}. If there is no @option{-mfpmath=sse},
> +@option{-fexcess-precision=standard} alone does the same thing as before,
> +It is useful for code that does not have @code{_Float16} and runs on the x87
> +FPU.
> +
> @node Decimal Float
> @section Decimal Floating Types
> @cindex decimal floating types
> diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
> index 32697e6117c..bb9f7ca956e 100644
> --- a/gcc/doc/invoke.texi
> +++ b/gcc/doc/invoke.texi
> @@ -1393,6 +1393,7 @@ See RS/6000 and PowerPC Options.
> -mavx5124fmaps -mavx512vnni -mavx5124vnniw -mprfchw -mrdpid @gol
> -mrdseed -msgx -mavx512vp2intersect -mserialize -mtsxldtrk@gol
> -mamx-tile -mamx-int8 -mamx-bf16 -muintr -mhreset -mavxvnni@gol
> +-mavx512fp16 @gol
> -mcldemote -mms-bitfields -mno-align-stringops -minline-all-stringops @gol
> -minline-stringops-dynamically -mstringop-strategy=@var{alg} @gol
> -mkl -mwidekl @gol
> @@ -31154,6 +31155,9 @@ preferred alignment to
> @option{-mpreferred-stack-boundary=2}.
> @itemx -mavx512bf16
> @opindex mavx512bf16
> @need 200
> +@itemx -mavx512fp16
> +@opindex mavx512fp16
> +@need 200
> @itemx -mgfni
> @opindex mgfni
> @need 200
> @@ -31232,9 +31236,9 @@ WBNOINVD, FMA4, PREFETCHW, RDPID, PREFETCHWT1,
> RDSEED, SGX, XOP, LWP,
> XSAVEOPT, XSAVEC, XSAVES, RTM, HLE, TBM, MWAITX, CLZERO, PKU, AVX512VBMI2,
> GFNI, VAES, WAITPKG, VPCLMULQDQ, AVX512BITALG, MOVDIRI, MOVDIR64B,
> AVX512BF16,
> ENQCMD, AVX512VPOPCNTDQ, AVX5124FMAPS, AVX512VNNI, AVX5124VNNIW, SERIALIZE,
> -UINTR, HRESET, AMXTILE, AMXINT8, AMXBF16, KL, WIDEKL, AVXVNNI or CLDEMOTE
> -extended instruction sets. Each has a corresponding @option{-mno-} option to
> -disable use of these instructions.
> +UINTR, HRESET, AMXTILE, AMXINT8, AMXBF16, KL, WIDEKL, AVXVNNI, AVX512FP16
> +or CLDEMOTE extended instruction sets. Each has a corresponding
> +@option{-mno-} option to disable use of these instructions.
>
> These extensions are also available as built-in functions: see
> @ref{x86 Built-in Functions}, for details of the functions enabled and
> diff --git a/gcc/testsuite/g++.dg/other/i386-2.C
> b/gcc/testsuite/g++.dg/other/i386-2.C
> index 62b2132957a..fba3d1ac684 100644
> --- a/gcc/testsuite/g++.dg/other/i386-2.C
> +++ b/gcc/testsuite/g++.dg/other/i386-2.C
> @@ -1,5 +1,5 @@
> /* { dg-do compile { target i?86-*-* x86_64-*-* } } */
> -/* { dg-options "-O -pedantic-errors -march=k8 -msse4a -m3dnow -mavx -mavx2
> -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp
> -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr
> -mxsaveopt -mavx512f -mavx512er -mavx512cd -mavx512pf -msha -mprefetchwt1
> -mxsavec -mxsaves -mclflushopt -mavx512dq -mavx512bw -mavx512vl -mavx512ifma
> -mavx512vbmi -mavx512vbmi2 -mavx5124fmaps -mavx5124vnniw -mavx512vpopcntdq
> -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mavx512bitalg -mpconfig
> -mwbnoinvd -mavx512bf16 -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk
> -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni" } */
> +/* { dg-options "-O -pedantic-errors -march=k8 -msse4a -m3dnow -mavx -mavx2
> -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm -mlwp
> -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr
> -mxsaveopt -mavx512f -mavx512er -mavx512cd -mavx512pf -msha -mprefetchwt1
> -mxsavec -mxsaves -mclflushopt -mavx512dq -mavx512bw -mavx512vl -mavx512ifma
> -mavx512vbmi -mavx512vbmi2 -mavx5124fmaps -mavx5124vnniw -mavx512vpopcntdq
> -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mavx512bitalg -mpconfig
> -mwbnoinvd -mavx512bf16 -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk
> -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavx512fp16" } */
>
> /* Test that {,x,e,p,t,s,w,a,b,i}mmintrin.h, mm3dnow.h, fma4intrin.h,
> xopintrin.h, abmintrin.h, bmiintrin.h, tbmintrin.h, lwpintrin.h,
> diff --git a/gcc/testsuite/g++.dg/other/i386-3.C
> b/gcc/testsuite/g++.dg/other/i386-3.C
> index 843aa2bdb2f..5cc0fa83457 100644
> --- a/gcc/testsuite/g++.dg/other/i386-3.C
> +++ b/gcc/testsuite/g++.dg/other/i386-3.C
> @@ -1,5 +1,5 @@
> /* { dg-do compile { target i?86-*-* x86_64-*-* } } */
> -/* { dg-options "-O -fkeep-inline-functions -march=k8 -msse4a -m3dnow -mavx
> -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm
> -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr
> -mxsaveopt -mavx512f -mavx512er -mavx512cd -mavx512pf -msha -mprefetchwt1
> -mxsavec -mxsaves -mclflushopt -mavx512dq -mavx512bw -mavx512vl -mavx512ifma
> -mavx512vbmi -mavx512vbmi2 -mavx5124fmaps -mavx5124vnniw -mavx512vpopcntdq
> -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mavx512bitalg -mpconfig
> -mwbnoinvd -mavx512bf16 -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk
> -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni" } */
> +/* { dg-options "-O -fkeep-inline-functions -march=k8 -msse4a -m3dnow -mavx
> -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi -mbmi2 -mtbm
> -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw -madx -mfxsr
> -mxsaveopt -mavx512f -mavx512er -mavx512cd -mavx512pf -msha -mprefetchwt1
> -mxsavec -mxsaves -mclflushopt -mavx512dq -mavx512bw -mavx512vl -mavx512ifma
> -mavx512vbmi -mavx512vbmi2 -mavx5124fmaps -mavx5124vnniw -mavx512vpopcntdq
> -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni -mavx512bitalg -mpconfig
> -mwbnoinvd -mavx512bf16 -menqcmd -mavx512vp2intersect -mserialize -mtsxldtrk
> -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl -mavxvnni -mavx512fp16" } */
>
> /* Test that {,x,e,p,t,s,w,a,b,i}mmintrin.h, mm3dnow.h, fma4intrin.h,
> xopintrin.h, abmintrin.h, bmiintrin.h, tbmintrin.h, lwpintrin.h,
> diff --git a/gcc/testsuite/g++.target/i386/float16-1.C
> b/gcc/testsuite/g++.target/i386/float16-1.C
> new file mode 100644
> index 00000000000..95d1ac27c4f
> --- /dev/null
> +++ b/gcc/testsuite/g++.target/i386/float16-1.C
> @@ -0,0 +1,8 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mno-sse2" } */
> +
> +_Float16/* { dg-error "does not name a type" } */
> +foo (_Float16 x)
> +{
> + return x;
> +}
> diff --git a/gcc/testsuite/g++.target/i386/float16-2.C
> b/gcc/testsuite/g++.target/i386/float16-2.C
> new file mode 100644
> index 00000000000..99eb797eff1
> --- /dev/null
> +++ b/gcc/testsuite/g++.target/i386/float16-2.C
> @@ -0,0 +1,14 @@
> +/* { dg-do assemble { target avx512fp16 } } */
> +/* { dg-options "-O2 -mavx512fp16" } */
> +
> +union flt
> +{
> + _Float16 flt;
> + short s;
> +};
> +
> +_Float16
> +foo (union flt x)
> +{
> + return x.flt;
> +}
> diff --git a/gcc/testsuite/g++.target/i386/float16-3.C
> b/gcc/testsuite/g++.target/i386/float16-3.C
> new file mode 100644
> index 00000000000..940878503f1
> --- /dev/null
> +++ b/gcc/testsuite/g++.target/i386/float16-3.C
> @@ -0,0 +1,10 @@
> +/* { dg-do assemble { target avx512fp16 } } */
> +/* { dg-options "-O0 -mavx512fp16" } */
> +
> +template <typename> void a(char *) {}
> +char b, d;
> +void c()
> +{
> + a<unsigned char>(&d);
> + a<_Float16>(&b);
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/avx-1.c
> b/gcc/testsuite/gcc.target/i386/avx-1.c
> index 6178e38ce02..f3676077743 100644
> --- a/gcc/testsuite/gcc.target/i386/avx-1.c
> +++ b/gcc/testsuite/gcc.target/i386/avx-1.c
> @@ -1,5 +1,5 @@
> /* { dg-do compile } */
> -/* { dg-options "-O2 -Werror-implicit-function-declaration -march=k8 -m3dnow
> -mavx -mavx2 -maes -mpclmul -mgfni -mavx512bw" } */
> +/* { dg-options "-O2 -Werror-implicit-function-declaration -march=k8 -m3dnow
> -mavx -mavx2 -maes -mpclmul -mgfni -mavx512bw -mavx512fp16" } */
> /* { dg-add-options bind_pic_locally } */
>
> #include <mm_malloc.h>
> diff --git a/gcc/testsuite/gcc.target/i386/avx-2.c
> b/gcc/testsuite/gcc.target/i386/avx-2.c
> index 986fbd819e4..1751c52565c 100644
> --- a/gcc/testsuite/gcc.target/i386/avx-2.c
> +++ b/gcc/testsuite/gcc.target/i386/avx-2.c
> @@ -1,5 +1,5 @@
> /* { dg-do compile } */
> -/* { dg-options "-O0 -Werror-implicit-function-declaration -march=k8 -m3dnow
> -mavx -mavx2 -msse4a -maes -mpclmul -mavx512bw" } */
> +/* { dg-options "-O0 -Werror-implicit-function-declaration -march=k8 -m3dnow
> -mavx -mavx2 -msse4a -maes -mpclmul -mavx512bw -mavx512fp16" } */
> /* { dg-add-options bind_pic_locally } */
>
> #include <mm_malloc.h>
> diff --git a/gcc/testsuite/gcc.target/i386/avx512-check.h
> b/gcc/testsuite/gcc.target/i386/avx512-check.h
> index 0a377dba1d5..0ad9064f637 100644
> --- a/gcc/testsuite/gcc.target/i386/avx512-check.h
> +++ b/gcc/testsuite/gcc.target/i386/avx512-check.h
> @@ -87,6 +87,9 @@ main ()
> #ifdef AVX512VNNI
> && (ecx & bit_AVX512VNNI)
> #endif
> +#ifdef AVX512FP16
> + && (edx & bit_AVX512FP16)
> +#endif
> #ifdef VAES
> && (ecx & bit_VAES)
> #endif
> diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-12a.c
> b/gcc/testsuite/gcc.target/i386/avx512fp16-12a.c
> new file mode 100644
> index 00000000000..88887556d68
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-12a.c
> @@ -0,0 +1,21 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mavx512fp16" } */
> +
> +_Float16
> +__attribute__ ((noinline, noclone))
> +do_max (_Float16 __A, _Float16 __B)
> +{
> + return __A > __B ? __A : __B;
> +}
> +
> +_Float16
> +__attribute__ ((noinline, noclone))
> +do_min (_Float16 __A, _Float16 __B)
> +{
> + return __A < __B ? __A : __B;
> +}
> +
> +/* { dg-final { scan-assembler-times "vmaxsh\[ \\t\]" 1 } } */
> +/* { dg-final { scan-assembler-times "vminsh\[ \\t\]" 1 } } */
> +/* { dg-final { scan-assembler-not "vmovsh\[ \\t\]" { target { ! ia32 } } }
> } */
> +/* { dg-final { scan-assembler-not "vcomish\[ \\t\]" } } */
> diff --git a/gcc/testsuite/gcc.target/i386/avx512fp16-12b.c
> b/gcc/testsuite/gcc.target/i386/avx512fp16-12b.c
> new file mode 100644
> index 00000000000..c9e23bf95c2
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/avx512fp16-12b.c
> @@ -0,0 +1,27 @@
> +/* { dg-do run { target avx512fp16 } } */
> +/* { dg-options "-O2 -mavx512fp16" } */
> +
> +#include <string.h>
> +
> +static void do_test (void);
> +
> +#define DO_TEST do_test
> +#define AVX512FP16
> +#include "avx512-check.h"
> +#include "avx512fp16-12a.c"
> +
> +static void
> +do_test (void)
> +{
> + _Float16 x = 0.1f;
> + _Float16 y = -3.2f;
> + _Float16 z;
> +
> + z = do_max (x, y);
> + if (z != x)
> + abort ();
> +
> + z = do_min (x, y);
> + if (z != y)
> + abort ();
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/float16-3a.c
> b/gcc/testsuite/gcc.target/i386/float16-3a.c
> new file mode 100644
> index 00000000000..3846c8e9b6e
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/float16-3a.c
> @@ -0,0 +1,10 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mavx512fp16" } */
> +
> +_Float16
> +foo (int x)
> +{
> + return x;
> +}
> +
> +/* { dg-final { scan-assembler-times "vcvtsi2shl\[ \t\]+\[^\n\r]*%xmm0" 1 }
> } */
> diff --git a/gcc/testsuite/gcc.target/i386/float16-3b.c
> b/gcc/testsuite/gcc.target/i386/float16-3b.c
> new file mode 100644
> index 00000000000..247dd6e7e33
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/float16-3b.c
> @@ -0,0 +1,10 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mavx512fp16" } */
> +
> +_Float16
> +foo (unsigned int x)
> +{
> + return x;
> +}
> +
> +/* { dg-final { scan-assembler-times "vcvtusi2shl\[ \t\]+\[^\n\r]*%xmm0" 1 }
> } */
> diff --git a/gcc/testsuite/gcc.target/i386/float16-4a.c
> b/gcc/testsuite/gcc.target/i386/float16-4a.c
> new file mode 100644
> index 00000000000..631082581f3
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/float16-4a.c
> @@ -0,0 +1,10 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-options "-O2 -mavx512fp16" } */
> +
> +_Float16
> +foo (long long x)
> +{
> + return x;
> +}
> +
> +/* { dg-final { scan-assembler-times "vcvtsi2shq\[ \t\]+\[^\n\r]*%xmm0" 1 }
> } */
> diff --git a/gcc/testsuite/gcc.target/i386/float16-4b.c
> b/gcc/testsuite/gcc.target/i386/float16-4b.c
> new file mode 100644
> index 00000000000..828d8530769
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/float16-4b.c
> @@ -0,0 +1,10 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-options "-O2 -mavx512fp16" } */
> +
> +_Float16
> +foo (unsigned long long x)
> +{
> + return x;
> +}
> +
> +/* { dg-final { scan-assembler-times "vcvtusi2shq\[ \t\]+\[^\n\r]*%xmm0" 1 }
> } */
> diff --git a/gcc/testsuite/gcc.target/i386/funcspec-56.inc
> b/gcc/testsuite/gcc.target/i386/funcspec-56.inc
> index 79265c7c94f..8499fdf2db9 100644
> --- a/gcc/testsuite/gcc.target/i386/funcspec-56.inc
> +++ b/gcc/testsuite/gcc.target/i386/funcspec-56.inc
> @@ -79,6 +79,7 @@ extern void test_hreset (void)
> __attribute__((__target__("hreset")));
> extern void test_keylocker (void)
> __attribute__((__target__("kl")));
> extern void test_widekl (void)
> __attribute__((__target__("widekl")));
> extern void test_avxvnni (void)
> __attribute__((__target__("avxvnni")));
> +extern void test_avx512fp16 (void)
> __attribute__((__target__("avx512fp16")));
>
> extern void test_no_sgx (void)
> __attribute__((__target__("no-sgx")));
> extern void test_no_avx5124fmaps(void)
> __attribute__((__target__("no-avx5124fmaps")));
> @@ -159,6 +160,7 @@ extern void test_no_hreset (void)
> __attribute__((__target__("no-hreset")));
> extern void test_no_keylocker (void)
> __attribute__((__target__("no-kl")));
> extern void test_no_widekl (void)
> __attribute__((__target__("no-widekl")));
> extern void test_no_avxvnni (void)
> __attribute__((__target__("no-avxvnni")));
> +extern void test_no_avx512fp16 (void)
> __attribute__((__target__("no-avx512fp16")));
>
> extern void test_arch_nocona (void)
> __attribute__((__target__("arch=nocona")));
> extern void test_arch_core2 (void)
> __attribute__((__target__("arch=core2")));
> diff --git a/gcc/testsuite/gcc.target/i386/pr54855-12.c
> b/gcc/testsuite/gcc.target/i386/pr54855-12.c
> new file mode 100644
> index 00000000000..2f8af392c83
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr54855-12.c
> @@ -0,0 +1,14 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mavx512fp16" } */
> +/* { dg-final { scan-assembler-times "vmaxsh\[ \\t\]" 1 } } */
> +/* { dg-final { scan-assembler-not "vcomish\[ \\t\]" } } */
> +/* { dg-final { scan-assembler-not "vmovsh\[ \\t\]" { target { ! ia32 } } }
> } */
> +
> +#include <immintrin.h>
> +
> +_Float16
> +foo (_Float16 x, _Float16 y)
> +{
> + x = x > y ? x : y;
> + return x;
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/sse-13.c
> b/gcc/testsuite/gcc.target/i386/sse-13.c
> index 7029771334b..f5f5c113612 100644
> --- a/gcc/testsuite/gcc.target/i386/sse-13.c
> +++ b/gcc/testsuite/gcc.target/i386/sse-13.c
> @@ -1,5 +1,5 @@
> /* { dg-do compile } */
> -/* { dg-options "-O2 -Werror-implicit-function-declaration -march=k8 -msse4a
> -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi
> -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw
> -madx -mfxsr -mxsaveopt -mavx512f -mavx512er -mavx512cd -mavx512pf -msha
> -mprefetchwt1 -mxsavec -mxsaves -mclflushopt -mavx512vl -mavx512dq -mavx512bw
> -mavx512vbmi -mavx512vbmi2 -mavx512ifma -mavx5124fmaps -mavx5124vnniw
> -mavx512vpopcntdq -mavx512vp2intersect -mclwb -mmwaitx -mclzero -mpku -msgx
> -mrdpid -mgfni -mavx512bitalg -mpconfig -mwbnoinvd -mavx512bf16 -menqcmd
> -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl
> -mavxvnni" } */
> +/* { dg-options "-O2 -Werror-implicit-function-declaration -march=k8 -msse4a
> -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi
> -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw
> -madx -mfxsr -mxsaveopt -mavx512f -mavx512er -mavx512cd -mavx512pf -msha
> -mprefetchwt1 -mxsavec -mxsaves -mclflushopt -mavx512vl -mavx512dq -mavx512bw
> -mavx512vbmi -mavx512vbmi2 -mavx512ifma -mavx5124fmaps -mavx5124vnniw
> -mavx512vpopcntdq -mavx512vp2intersect -mclwb -mmwaitx -mclzero -mpku -msgx
> -mrdpid -mgfni -mavx512bitalg -mpconfig -mwbnoinvd -mavx512bf16 -menqcmd
> -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl
> -mavxvnni -mavx512fp16" } */
> /* { dg-add-options bind_pic_locally } */
>
> #include <mm_malloc.h>
> diff --git a/gcc/testsuite/gcc.target/i386/sse-14.c
> b/gcc/testsuite/gcc.target/i386/sse-14.c
> index 4ce0ffffaf3..747d504cedb 100644
> --- a/gcc/testsuite/gcc.target/i386/sse-14.c
> +++ b/gcc/testsuite/gcc.target/i386/sse-14.c
> @@ -1,5 +1,5 @@
> /* { dg-do compile } */
> -/* { dg-options "-O0 -Werror-implicit-function-declaration -march=k8 -msse4a
> -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi
> -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw
> -madx -mfxsr -mxsaveopt -mavx512f -mavx512er -mavx512cd -mavx512pf -msha
> -mprefetchwt1 -mxsavec -mxsaves -mclflushopt -mavx512dq -mavx512bw -mavx512vl
> -mavx512ifma -mavx512vbmi -mavx512vbmi2 -mavx5124fmaps -mavx5124vnniw
> -mavx512vpopcntdq -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni
> -mpconfig -mwbnoinvd -mavx512vl -mavx512bf16 -menqcmd -mavx512vp2intersect
> -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl
> -mavxvnni" } */
> +/* { dg-options "-O0 -Werror-implicit-function-declaration -march=k8 -msse4a
> -m3dnow -mavx -mavx2 -mfma4 -mxop -maes -mpclmul -mpopcnt -mabm -mlzcnt -mbmi
> -mbmi2 -mtbm -mlwp -mfsgsbase -mrdrnd -mf16c -mfma -mrtm -mrdseed -mprfchw
> -madx -mfxsr -mxsaveopt -mavx512f -mavx512er -mavx512cd -mavx512pf -msha
> -mprefetchwt1 -mxsavec -mxsaves -mclflushopt -mavx512dq -mavx512bw -mavx512vl
> -mavx512ifma -mavx512vbmi -mavx512vbmi2 -mavx5124fmaps -mavx5124vnniw
> -mavx512vpopcntdq -mclwb -mmwaitx -mclzero -mpku -msgx -mrdpid -mgfni
> -mpconfig -mwbnoinvd -mavx512vl -mavx512bf16 -menqcmd -mavx512vp2intersect
> -mserialize -mtsxldtrk -mamx-tile -mamx-int8 -mamx-bf16 -mkl -mwidekl
> -mavxvnni -mavx512fp16" } */
> /* { dg-add-options bind_pic_locally } */
>
> #include <mm_malloc.h>
> diff --git a/gcc/testsuite/gcc.target/i386/sse-22.c
> b/gcc/testsuite/gcc.target/i386/sse-22.c
> index 6e8b6f3fa1b..33411969901 100644
> --- a/gcc/testsuite/gcc.target/i386/sse-22.c
> +++ b/gcc/testsuite/gcc.target/i386/sse-22.c
> @@ -103,7 +103,7 @@
>
>
> #ifndef DIFFERENT_PRAGMAS
> -#pragma GCC target
> ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,rtm,rdseed,prfchw,adx,fxsr,xsaveopt,avx512f,avx512er,avx512cd,avx512pf,sha,prefetchwt1,avx512vl,avx512bw,avx512dq,avx512vbmi,avx512vbmi2,avx512ifma,avx5124fmaps,avx5124vnniw,avx512vpopcntdq,gfni,avx512bitalg,avx512bf16,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni")
> +#pragma GCC target
> ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,rtm,rdseed,prfchw,adx,fxsr,xsaveopt,avx512f,avx512er,avx512cd,avx512pf,sha,prefetchwt1,avx512vl,avx512bw,avx512dq,avx512vbmi,avx512vbmi2,avx512ifma,avx5124fmaps,avx5124vnniw,avx512vpopcntdq,gfni,avx512bitalg,avx512bf16,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avx512fp16")
> #endif
>
> /* Following intrinsics require immediate arguments. They
> @@ -220,7 +220,7 @@ test_4 (_mm_cmpestrz, int, __m128i, int, __m128i, int, 1)
>
> /* immintrin.h (AVX/AVX2/RDRND/FSGSBASE/F16C/RTM/AVX512F/SHA) */
> #ifdef DIFFERENT_PRAGMAS
> -#pragma GCC target
> ("avx,avx2,rdrnd,fsgsbase,f16c,rtm,avx512f,avx512er,avx512cd,avx512pf,sha,avx512vl,avx512bw,avx512dq,avx512ifma,avx512vbmi,avx512vbmi2,avx5124fmaps,avx5124vnniw,avx512vpopcntdq,gfni,avx512bitalg,avx512bf16,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni")
> +#pragma GCC target
> ("avx,avx2,rdrnd,fsgsbase,f16c,rtm,avx512f,avx512er,avx512cd,avx512pf,sha,avx512vl,avx512bw,avx512dq,avx512ifma,avx512vbmi,avx512vbmi2,avx5124fmaps,avx5124vnniw,avx512vpopcntdq,gfni,avx512bitalg,avx512bf16,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avx512fp16")
> #endif
> #include <immintrin.h>
> test_1 (_cvtss_sh, unsigned short, float, 1)
> diff --git a/gcc/testsuite/gcc.target/i386/sse-23.c
> b/gcc/testsuite/gcc.target/i386/sse-23.c
> index 7faa053ace8..86590ca5ffb 100644
> --- a/gcc/testsuite/gcc.target/i386/sse-23.c
> +++ b/gcc/testsuite/gcc.target/i386/sse-23.c
> @@ -708,6 +708,6 @@
> #define __builtin_ia32_vpclmulqdq_v2di(A, B, C)
> __builtin_ia32_vpclmulqdq_v2di(A, B, 1)
> #define __builtin_ia32_vpclmulqdq_v8di(A, B, C)
> __builtin_ia32_vpclmulqdq_v8di(A, B, 1)
>
> -#pragma GCC target
> ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,fma,rtm,rdseed,prfchw,adx,fxsr,xsaveopt,avx512f,avx512er,avx512cd,avx512pf,sha,prefetchwt1,xsavec,xsaves,clflushopt,avx512bw,avx512dq,avx512vl,avx512vbmi,avx512ifma,avx5124fmaps,avx5124vnniw,avx512vpopcntdq,clwb,mwaitx,clzero,pku,sgx,rdpid,gfni,avx512vbmi2,vpclmulqdq,avx512bitalg,pconfig,wbnoinvd,avx512bf16,enqcmd,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni")
> +#pragma GCC target
> ("sse4a,3dnow,avx,avx2,fma4,xop,aes,pclmul,popcnt,abm,lzcnt,bmi,bmi2,tbm,lwp,fsgsbase,rdrnd,f16c,fma,rtm,rdseed,prfchw,adx,fxsr,xsaveopt,avx512f,avx512er,avx512cd,avx512pf,sha,prefetchwt1,xsavec,xsaves,clflushopt,avx512bw,avx512dq,avx512vl,avx512vbmi,avx512ifma,avx5124fmaps,avx5124vnniw,avx512vpopcntdq,clwb,mwaitx,clzero,pku,sgx,rdpid,gfni,avx512vbmi2,vpclmulqdq,avx512bitalg,pconfig,wbnoinvd,avx512bf16,enqcmd,avx512vp2intersect,serialize,tsxldtrk,amx-tile,amx-int8,amx-bf16,kl,widekl,avxvnni,avx512fp16")
>
> #include <x86intrin.h>
> diff --git a/gcc/testsuite/lib/target-supports.exp
> b/gcc/testsuite/lib/target-supports.exp
> index 42ac9d0ac1a..10765365d7b 100644
> --- a/gcc/testsuite/lib/target-supports.exp
> +++ b/gcc/testsuite/lib/target-supports.exp
> @@ -3020,7 +3020,7 @@ proc check_effective_target_has_q_floating_suffix { } {
>
> proc check_effective_target_float16 {} {
> return [check_no_compiler_messages_nocache float16 object {
> - _Float16 x;
> + _Float16 foo (_Float16 x) { return x; }
> } [add_options_for_float16 ""]]
> }
>
> @@ -8714,6 +8714,17 @@ proc check_prefer_avx128 { } {
> }
>
>
> +# Return 1 if avx512fp16 instructions can be compiled.
> +
> +proc check_effective_target_avx512fp16 { } {
> + return [check_no_compiler_messages avx512fp16 object {
> + void foo (void)
> + {
> + asm volatile ("vmovw %edi, %xmm0");
> + }
> + } "-O2 -mavx512fp16" ]
> +}
> +
> # Return 1 if avx512f instructions can be compiled.
>
> proc check_effective_target_avx512f { } {
> --
> 2.27.0
>
--
BR,
Hongtao
