chromium/third_party/llvm-build/Release+Asserts/lib/clang/20/include/avx512vlbf16intrin.h

/*===--------- avx512vlbf16intrin.h - AVX512_BF16 intrinsics ---------------===
 *
 * 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
 *
 *===-----------------------------------------------------------------------===
 */
#ifndef __IMMINTRIN_H
#error "Never use <avx512vlbf16intrin.h> directly; include <immintrin.h> instead."
#endif

#ifdef __SSE2__

#ifndef __AVX512VLBF16INTRIN_H
#define __AVX512VLBF16INTRIN_H

#define __DEFAULT_FN_ATTRS128 …
#define __DEFAULT_FN_ATTRS256 …

/// Convert Two Packed Single Data to One Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNE2PS2BF16 </c> instructions.
///
/// \param __A
///    A 128-bit vector of [4 x float].
/// \param __B
///    A 128-bit vector of [4 x float].
/// \returns A 128-bit vector of [8 x bfloat] whose lower 64 bits come from
///    conversion of __B, and higher 64 bits come from conversion of __A.
static __inline__ __m128bh __DEFAULT_FN_ATTRS128
_mm_cvtne2ps_pbh(__m128 __A, __m128 __B) { … }

/// Convert Two Packed Single Data to One Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNE2PS2BF16 </c> instructions.
///
/// \param __A
///    A 128-bit vector of [4 x float].
/// \param __B
///    A 128-bit vector of [4 x float].
/// \param __W
///    A 128-bit vector of [8 x bfloat].
/// \param __U
///    A 8-bit mask value specifying what is chosen for each element.
///    A 1 means conversion of __A or __B. A 0 means element from __W.
/// \returns A 128-bit vector of [8 x bfloat] whose lower 64 bits come from
///    conversion of __B, and higher 64 bits come from conversion of __A.
static __inline__ __m128bh __DEFAULT_FN_ATTRS128
_mm_mask_cvtne2ps_pbh(__m128bh __W, __mmask8 __U, __m128 __A, __m128 __B) { … }

/// Convert Two Packed Single Data to One Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNE2PS2BF16 </c> instructions.
///
/// \param __A
///    A 128-bit vector of [4 x float].
/// \param __B
///    A 128-bit vector of [4 x float].
/// \param __U
///    A 8-bit mask value specifying what is chosen for each element.
///    A 1 means conversion of __A or __B. A 0 means element is zero.
/// \returns A 128-bit vector of [8 x bfloat] whose lower 64 bits come from
///    conversion of __B, and higher 64 bits come from conversion of __A.
static __inline__ __m128bh __DEFAULT_FN_ATTRS128
_mm_maskz_cvtne2ps_pbh(__mmask8 __U, __m128 __A, __m128 __B) { … }

/// Convert Two Packed Single Data to One Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNE2PS2BF16 </c> instructions.
///
/// \param __A
///    A 256-bit vector of [8 x float].
/// \param __B
///    A 256-bit vector of [8 x float].
/// \returns A 256-bit vector of [16 x bfloat] whose lower 128 bits come from
///    conversion of __B, and higher 128 bits come from conversion of __A.
static __inline__ __m256bh __DEFAULT_FN_ATTRS256
_mm256_cvtne2ps_pbh(__m256 __A, __m256 __B) { … }

/// Convert Two Packed Single Data to One Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNE2PS2BF16 </c> instructions.
///
/// \param __A
///    A 256-bit vector of [8 x float].
/// \param __B
///    A 256-bit vector of [8 x float].
/// \param __W
///    A 256-bit vector of [16 x bfloat].
/// \param __U
///    A 16-bit mask value specifying what is chosen for each element.
///    A 1 means conversion of __A or __B. A 0 means element from __W.
/// \returns A 256-bit vector of [16 x bfloat] whose lower 128 bits come from
///    conversion of __B, and higher 128 bits come from conversion of __A.
static __inline__ __m256bh __DEFAULT_FN_ATTRS256
_mm256_mask_cvtne2ps_pbh(__m256bh __W, __mmask16 __U, __m256 __A, __m256 __B) { … }

/// Convert Two Packed Single Data to One Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNE2PS2BF16 </c> instructions.
///
/// \param __A
///    A 256-bit vector of [8 x float].
/// \param __B
///    A 256-bit vector of [8 x float].
/// \param __U
///    A 16-bit mask value specifying what is chosen for each element.
///    A 1 means conversion of __A or __B. A 0 means element is zero.
/// \returns A 256-bit vector of [16 x bfloat] whose lower 128 bits come from
///    conversion of __B, and higher 128 bits come from conversion of __A.
static __inline__ __m256bh __DEFAULT_FN_ATTRS256
_mm256_maskz_cvtne2ps_pbh(__mmask16 __U, __m256 __A, __m256 __B) { … }

/// Convert Packed Single Data to Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNEPS2BF16 </c> instructions.
///
/// \param __A
///    A 128-bit vector of [4 x float].
/// \returns A 128-bit vector of [8 x bfloat] whose lower 64 bits come from
///    conversion of __A, and higher 64 bits are 0.
#define _mm_cvtneps_pbh(A) …

/// Convert Packed Single Data to Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNEPS2BF16 </c> instructions.
///
/// \param __A
///    A 128-bit vector of [4 x float].
/// \param __W
///    A 128-bit vector of [8 x bfloat].
/// \param __U
///    A 4-bit mask value specifying what is chosen for each element.
///    A 1 means conversion of __A. A 0 means element from __W.
/// \returns A 128-bit vector of [8 x bfloat] whose lower 64 bits come from
///    conversion of __A, and higher 64 bits are 0.
static __inline__ __m128bh __DEFAULT_FN_ATTRS128
_mm_mask_cvtneps_pbh(__m128bh __W, __mmask8 __U, __m128 __A) { … }

/// Convert Packed Single Data to Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNEPS2BF16 </c> instructions.
///
/// \param __A
///    A 128-bit vector of [4 x float].
/// \param __U
///    A 4-bit mask value specifying what is chosen for each element.
///    A 1 means conversion of __A. A 0 means element is zero.
/// \returns A 128-bit vector of [8 x bfloat] whose lower 64 bits come from
///    conversion of __A, and higher 64 bits are 0.
static __inline__ __m128bh __DEFAULT_FN_ATTRS128
_mm_maskz_cvtneps_pbh(__mmask8 __U, __m128 __A) { … }

/// Convert Packed Single Data to Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNEPS2BF16 </c> instructions.
///
/// \param __A
///    A 256-bit vector of [8 x float].
/// \returns A 128-bit vector of [8 x bfloat] comes from conversion of __A.
#define _mm256_cvtneps_pbh(A) …

/// Convert Packed Single Data to Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNEPS2BF16 </c> instructions.
///
/// \param __A
///    A 256-bit vector of [8 x float].
/// \param __W
///    A 256-bit vector of [8 x bfloat].
/// \param __U
///    A 8-bit mask value specifying what is chosen for each element.
///    A 1 means conversion of __A. A 0 means element from __W.
/// \returns A 128-bit vector of [8 x bfloat] comes from conversion of __A.
static __inline__ __m128bh __DEFAULT_FN_ATTRS256
_mm256_mask_cvtneps_pbh(__m128bh __W, __mmask8 __U, __m256 __A) { … }

/// Convert Packed Single Data to Packed BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNEPS2BF16 </c> instructions.
///
/// \param __A
///    A 256-bit vector of [8 x float].
/// \param __U
///    A 8-bit mask value specifying what is chosen for each element.
///    A 1 means conversion of __A. A 0 means element is zero.
/// \returns A 128-bit vector of [8 x bfloat] comes from conversion of __A.
static __inline__ __m128bh __DEFAULT_FN_ATTRS256
_mm256_maskz_cvtneps_pbh(__mmask8 __U, __m256 __A) { … }

/// Dot Product of BF16 Pairs Accumulated into Packed Single Precision.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VDPBF16PS </c> instructions.
///
/// \param __A
///    A 128-bit vector of [8 x bfloat].
/// \param __B
///    A 128-bit vector of [8 x bfloat].
/// \param __D
///    A 128-bit vector of [4 x float].
/// \returns A 128-bit vector of [4 x float] comes from  Dot Product of
///  __A, __B and __D
static __inline__ __m128 __DEFAULT_FN_ATTRS128
_mm_dpbf16_ps(__m128 __D, __m128bh __A, __m128bh __B) { … }

/// Dot Product of BF16 Pairs Accumulated into Packed Single Precision.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VDPBF16PS </c> instructions.
///
/// \param __A
///    A 128-bit vector of [8 x bfloat].
/// \param __B
///    A 128-bit vector of [8 x bfloat].
/// \param __D
///    A 128-bit vector of [4 x float].
/// \param __U
///    A 8-bit mask value specifying what is chosen for each element.
///    A 1 means __A and __B's dot product accumulated with __D. A 0 means __D.
/// \returns A 128-bit vector of [4 x float] comes from  Dot Product of
///  __A, __B and __D
static __inline__ __m128 __DEFAULT_FN_ATTRS128
_mm_mask_dpbf16_ps(__m128 __D, __mmask8 __U, __m128bh __A, __m128bh __B) { … }

/// Dot Product of BF16 Pairs Accumulated into Packed Single Precision.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VDPBF16PS </c> instructions.
///
/// \param __A
///    A 128-bit vector of [8 x bfloat].
/// \param __B
///    A 128-bit vector of [8 x bfloat].
/// \param __D
///    A 128-bit vector of [4 x float].
/// \param __U
///    A 8-bit mask value specifying what is chosen for each element.
///    A 1 means __A and __B's dot product accumulated with __D. A 0 means 0.
/// \returns A 128-bit vector of [4 x float] comes from  Dot Product of
///  __A, __B and __D
static __inline__ __m128 __DEFAULT_FN_ATTRS128
_mm_maskz_dpbf16_ps(__mmask8 __U, __m128 __D, __m128bh __A, __m128bh __B) { … }

/// Dot Product of BF16 Pairs Accumulated into Packed Single Precision.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VDPBF16PS </c> instructions.
///
/// \param __A
///    A 256-bit vector of [16 x bfloat].
/// \param __B
///    A 256-bit vector of [16 x bfloat].
/// \param __D
///    A 256-bit vector of [8 x float].
/// \returns A 256-bit vector of [8 x float] comes from  Dot Product of
///  __A, __B and __D
static __inline__ __m256 __DEFAULT_FN_ATTRS256
_mm256_dpbf16_ps(__m256 __D, __m256bh __A, __m256bh __B) { … }

/// Dot Product of BF16 Pairs Accumulated into Packed Single Precision.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VDPBF16PS </c> instructions.
///
/// \param __A
///    A 256-bit vector of [16 x bfloat].
/// \param __B
///    A 256-bit vector of [16 x bfloat].
/// \param __D
///    A 256-bit vector of [8 x float].
/// \param __U
///    A 16-bit mask value specifying what is chosen for each element.
///    A 1 means __A and __B's dot product accumulated with __D. A 0 means __D.
/// \returns A 256-bit vector of [8 x float] comes from  Dot Product of
///  __A, __B and __D
static __inline__ __m256 __DEFAULT_FN_ATTRS256
_mm256_mask_dpbf16_ps(__m256 __D, __mmask8 __U, __m256bh __A, __m256bh __B) { … }

/// Dot Product of BF16 Pairs Accumulated into Packed Single Precision.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VDPBF16PS </c> instructions.
///
/// \param __A
///    A 256-bit vector of [16 x bfloat].
/// \param __B
///    A 256-bit vector of [16 x bfloat].
/// \param __D
///    A 256-bit vector of [8 x float].
/// \param __U
///    A 8-bit mask value specifying what is chosen for each element.
///    A 1 means __A and __B's dot product accumulated with __D. A 0 means 0.
/// \returns A 256-bit vector of [8 x float] comes from  Dot Product of
///  __A, __B and __D
static __inline__ __m256 __DEFAULT_FN_ATTRS256
_mm256_maskz_dpbf16_ps(__mmask8 __U, __m256 __D, __m256bh __A, __m256bh __B) { … }

/// Convert One Single float Data to One BF16 Data.
///
/// \headerfile <x86intrin.h>
///
/// This intrinsic corresponds to the <c> VCVTNEPS2BF16 </c> instructions.
///
/// \param __A
///    A float data.
/// \returns A bf16 data whose sign field and exponent field keep unchanged,
///    and fraction field is truncated to 7 bits.
static __inline__ __bf16 __DEFAULT_FN_ATTRS128 _mm_cvtness_sbh(float __A) { … }

/// Convert Packed BF16 Data to Packed float Data.
///
/// \headerfile <x86intrin.h>
///
/// \param __A
///    A 128-bit vector of [4 x bfloat].
/// \returns A 128-bit vector of [4 x float] come from conversion of __A
static __inline__ __m128 __DEFAULT_FN_ATTRS128 _mm_cvtpbh_ps(__m128bh __A) { … }

/// Convert Packed BF16 Data to Packed float Data.
///
/// \headerfile <x86intrin.h>
///
/// \param __A
///    A 128-bit vector of [8 x bfloat].
/// \returns A 256-bit vector of [8 x float] come from conversion of __A
static __inline__ __m256 __DEFAULT_FN_ATTRS256 _mm256_cvtpbh_ps(__m128bh __A) { … }

/// Convert Packed BF16 Data to Packed float Data using zeroing mask.
///
/// \headerfile <x86intrin.h>
///
/// \param __U
///    A 4-bit mask. Elements are zeroed out when the corresponding mask
///    bit is not set.
/// \param __A
///    A 128-bit vector of [4 x bfloat].
/// \returns A 128-bit vector of [4 x float] come from conversion of __A
static __inline__ __m128 __DEFAULT_FN_ATTRS128
_mm_maskz_cvtpbh_ps(__mmask8 __U, __m128bh __A) { … }

/// Convert Packed BF16 Data to Packed float Data using zeroing mask.
///
/// \headerfile <x86intrin.h>
///
/// \param __U
///    A 8-bit mask. Elements are zeroed out when the corresponding mask
///    bit is not set.
/// \param __A
///    A 128-bit vector of [8 x bfloat].
/// \returns A 256-bit vector of [8 x float] come from conversion of __A
static __inline__ __m256 __DEFAULT_FN_ATTRS256
_mm256_maskz_cvtpbh_ps(__mmask8 __U, __m128bh __A) { … }

/// Convert Packed BF16 Data to Packed float Data using merging mask.
///
/// \headerfile <x86intrin.h>
///
/// \param __S
///    A 128-bit vector of [4 x float]. Elements are copied from __S when
///     the corresponding mask bit is not set.
/// \param __U
///    A 4-bit mask. Elements are zeroed out when the corresponding mask
///    bit is not set.
/// \param __A
///    A 128-bit vector of [4 x bfloat].
/// \returns A 128-bit vector of [4 x float] come from conversion of __A
static __inline__ __m128 __DEFAULT_FN_ATTRS128
_mm_mask_cvtpbh_ps(__m128 __S, __mmask8 __U, __m128bh __A) { … }

/// Convert Packed BF16 Data to Packed float Data using merging mask.
///
/// \headerfile <x86intrin.h>
///
/// \param __S
///    A 256-bit vector of [8 x float]. Elements are copied from __S when
///     the corresponding mask bit is not set.
/// \param __U
///    A 8-bit mask. Elements are zeroed out when the corresponding mask
///    bit is not set.
/// \param __A
///    A 128-bit vector of [8 x bfloat].
/// \returns A 256-bit vector of [8 x float] come from conversion of __A
static __inline__ __m256 __DEFAULT_FN_ATTRS256
_mm256_mask_cvtpbh_ps(__m256 __S, __mmask8 __U, __m128bh __A) { … }

#undef __DEFAULT_FN_ATTRS128
#undef __DEFAULT_FN_ATTRS256

#endif
#endif