[clang] [X86][AVX10.2] Add comments for the avx10_2convertintrin.h file (PR #120766)

Mikołaj Piróg via cfe-commits cfe-commits at lists.llvm.org
Wed Feb 5 05:16:54 PST 2025


https://github.com/mikolaj-pirog updated https://github.com/llvm/llvm-project/pull/120766

>From 18dfaec1d43fe38dc6b3fe754153426e5c974f0f Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Miko=C5=82aj=20Pir=C3=B3g?= <mikolajpirog at gmail.com>
Date: Thu, 23 Jan 2025 23:27:17 +0100
Subject: [PATCH 1/7] rename

---
 clang/lib/Headers/avx10_2convertintrin.h | 2943 +++++++++++++++++++++-
 1 file changed, 2879 insertions(+), 64 deletions(-)

diff --git a/clang/lib/Headers/avx10_2convertintrin.h b/clang/lib/Headers/avx10_2convertintrin.h
index c67a5b890f1957..82d434c18df131 100644
--- a/clang/lib/Headers/avx10_2convertintrin.h
+++ b/clang/lib/Headers/avx10_2convertintrin.h
@@ -24,24 +24,157 @@
   __attribute__((__always_inline__, __nodebug__, __target__("avx10.2-256"),    \
                  __min_vector_width__(256)))
 
+// clang-format off
+
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed
+///    single-precision (32-bit) floating-point elements to a 128-bit vector
+///    containing FP16 elements.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF i < 4
+/// 		dst.fp16[i] := convert_fp32_to_fp16(__B.fp32[i])
+/// 	ELSE
+/// 		dst.fp16[i] := convert_fp32_to_fp16(__A.fp32[i - 4])
+/// 	FI
+///
+/// ENDFOR
+///
+/// dst[MAX:127] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PS2PHX instruction.
+///
+/// \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 fp16]. Lower 4 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvtx2ps_ph(__m128 __A,
                                                                __m128 __B) {
   return (__m128h)__builtin_ia32_vcvt2ps2phx128_mask(
       (__v4sf)__A, (__v4sf)__B, (__v8hf)_mm_setzero_ph(), (__mmask8)(-1));
 }
 
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed
+/// single-precision (32-bit) floating-point elements to a 128-bit vector
+/// containing FP16 elements. Merging mask \a __U is used to determine if given
+/// element should be taken from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		IF i < 4
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__B.fp32[i])
+/// 		ELSE
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__A.fp32[i - 4])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp16[i] := __W.fp16[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:127] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PS2PHX instruction.
+///
+/// \param __W
+///    A 128-bit vector of [8 x fp16].
+/// \param __U
+///    A 8-bit merging mask.
+/// \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 fp16]. Lower elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128h __DEFAULT_FN_ATTRS128
 _mm_mask_cvtx2ps_ph(__m128h __W, __mmask8 __U, __m128 __A, __m128 __B) {
   return (__m128h)__builtin_ia32_vcvt2ps2phx128_mask(
       (__v4sf)__A, (__v4sf)__B, (__v8hf)__W, (__mmask8)__U);
 }
 
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed
+/// single-precision (32-bit) floating-point elements to a 128-bit vector
+/// containing FP16 elements. Zeroing mask \a __U is used to determine if given
+/// element should be zeroed instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		IF i < 4
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__B.fp32[i])
+/// 		ELSE
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__A.fp32[i - 4])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp16[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:127] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PS2PHX instruction.
+///
+/// \param __U
+///    A 8-bit zeroing mask.
+/// \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 fp16]. Lower elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    then zero is taken instead.
 static __inline__ __m128h __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtx2ps_ph(__mmask8 __U, __m128 __A, __m128 __B) {
   return (__m128h)__builtin_ia32_vcvt2ps2phx128_mask(
       (__v4sf)__A, (__v4sf)__B, (__v8hf)_mm_setzero_ph(), (__mmask8)__U);
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed
+/// single-precision (32-bit) floating-point elements to a 256-bit vector
+/// containing FP16 elements.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15 
+/// 	IF i < 8
+/// 		dst.fp16[i] := convert_fp32_to_fp16(__B.fp32[i])
+/// 	ELSE
+/// 		dst.fp16[i] := convert_fp32_to_fp16(__A.fp32[i - 8])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PS2PHX instruction.
+///
+/// \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 fp16]. Lower elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m256h __DEFAULT_FN_ATTRS256 _mm256_cvtx2ps_ph(__m256 __A,
                                                                   __m256 __B) {
   return (__m256h)__builtin_ia32_vcvt2ps2phx256_mask(
@@ -49,6 +182,44 @@ static __inline__ __m256h __DEFAULT_FN_ATTRS256 _mm256_cvtx2ps_ph(__m256 __A,
       _MM_FROUND_CUR_DIRECTION);
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed
+/// single-precision (32-bit) floating-point elements to a 256-bit vector
+/// containing FP16 elements. Merging mask \a __U is used to determine if given
+/// element should be taken from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__B.fp32[i])
+/// 		ELSE
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__A.fp32[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp16[i] := __W.fp16[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PS2PHX instruction.
+///
+/// \param __W
+///    A 256-bit vector of [16 x fp16].
+/// \param __U
+///    A 16-bit merging mask.
+/// \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 fp16]. Lower elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m256h __DEFAULT_FN_ATTRS256
 _mm256_mask_cvtx2ps_ph(__m256h __W, __mmask16 __U, __m256 __A, __m256 __B) {
   return (__m256h)__builtin_ia32_vcvt2ps2phx256_mask(
@@ -56,6 +227,42 @@ _mm256_mask_cvtx2ps_ph(__m256h __W, __mmask16 __U, __m256 __A, __m256 __B) {
       _MM_FROUND_CUR_DIRECTION);
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed
+/// single-precision (32-bit) floating-point elements to a 256-bit vector
+/// containing FP16 elements. Zeroing mask \a __U is used to determine if given
+/// element should be zeroed instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__B.fp32[i])
+/// 		ELSE
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__A.fp32[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp16[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PS2PHX instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \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 fp16]. Lower elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    then zero is taken instead.
 static __inline__ __m256h __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtx2ps_ph(__mmask16 __U, __m256 __A, __m256 __B) {
   return (__m256h)__builtin_ia32_vcvt2ps2phx256_mask(
@@ -63,32 +270,239 @@ _mm256_maskz_cvtx2ps_ph(__mmask16 __U, __m256 __A, __m256 __B) {
       _MM_FROUND_CUR_DIRECTION);
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed
+/// single-precision (32-bit) floating-point elements to a 256-bit vector
+/// containing FP16 elements. Rounding mode \a __R needs to be provided.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15 
+/// 	IF i < 8
+/// 		dst.fp16[i] := convert_fp32_to_fp16(__B.fp32[i])
+/// 	ELSE
+/// 		dst.fp16[i] := convert_fp32_to_fp16(__A.fp32[i - 8])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PS2PHX instruction.
+///
+/// \param __A
+///    A 256-bit vector of [8 x float].
+/// \param __B
+///    A 256-bit vector of [8 x float].
+/// \param __R
+///    Rounding mode. Valid inputs are: _MM_FROUND_CUR_DIRECTION or
+///    result of bitwise or of _MM_FROUND_NO_EXC with at most one of the following:
+///    _MM_FROUND_TO_NEAREST_INT, _MM_FROUND_TO_NEG_INF, _MM_FROUND_TO_POS_INF,
+///    _MM_FROUND_TO_ZERO.
+/// \returns
+///    A 256-bit vector of [16 x fp16]. Lower elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
 #define _mm256_cvtx_round2ps_ph(A, B, R)                                       \
   ((__m256h)__builtin_ia32_vcvt2ps2phx256_mask(                                \
       (__v8sf)(A), (__v8sf)(B), (__v16hf)_mm256_undefined_ph(),                \
       (__mmask16)(-1), (const int)(R)))
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed
+/// single-precision (32-bit) floating-point elements to a 256-bit vector
+/// containing FP16 elements. Merging mask \a __U is used to determine if given
+/// element should be taken from \a __W instead. Rounding mode \a __R needs to
+/// be provided.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__B.fp32[i])
+/// 		ELSE
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__A.fp32[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp16[i] := __W.fp16[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PS2PHX instruction.
+///
+/// \param __W
+///    A 256-bit vector of [16 x fp16].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [8 x float].
+/// \param __B
+///    A 256-bit vector of [8 x float].
+/// \param __R
+///    Rounding mode. Valid inputs are: _MM_FROUND_CUR_DIRECTION or
+///    result of bitwise or of _MM_FROUND_NO_EXC with at most one of the following:
+///    _MM_FROUND_TO_NEAREST_INT, _MM_FROUND_TO_NEG_INF, _MM_FROUND_TO_POS_INF,
+///    _MM_FROUND_TO_ZERO.
+/// \returns
+///    A 256-bit vector of [16 x fp16]. Lower elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 #define _mm256_mask_cvtx_round2ps_ph(W, U, A, B, R)                            \
   ((__m256h)__builtin_ia32_vcvt2ps2phx256_mask(                                \
       (__v8sf)(A), (__v8sf)(B), (__v16hf)(W), (__mmask16)(U), (const int)(R)))
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed
+/// single-precision (32-bit) floating-point elements to a 256-bit vector
+/// containing FP16 elements. Zeroing mask \a __U is used to determine if given
+/// element should be zeroed instead. Rounding mode \a __R needs to be provided.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__B.fp32[i])
+/// 		ELSE
+/// 			dst.fp16[i] := convert_fp32_to_fp16(__A.fp32[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp16[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PS2PHX instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \param __A
+///    A 256-bit vector of [8 x float].
+/// \param __B
+///    A 256-bit vector of [8 x float].
+/// \param __R
+///    Rounding mode. Valid inputs are: _MM_FROUND_CUR_DIRECTION or
+///    result of bitwise or of _MM_FROUND_NO_EXC with at most one of the following:
+///    _MM_FROUND_TO_NEAREST_INT, _MM_FROUND_TO_NEG_INF, _MM_FROUND_TO_POS_INF,
+///    _MM_FROUND_TO_ZERO.
+/// \returns
+///    A 256-bit vector of [16 x fp16]. Lower elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    then zero is taken instead.
 #define _mm256_maskz_cvtx_round2ps_ph(U, A, B, R)                              \
   ((__m256h)__builtin_ia32_vcvt2ps2phx256_mask(                                \
       (__v8sf)(A), (__v8sf)(B), (__v16hf)(_mm256_setzero_ph()),                \
       (__mmask16)(U), (const int)(R)))
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtbiasph_bf8(__m128i __A,
-                                                                  __m128h __B) {
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements and 8-bit integers stored in the lower half of
+/// packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    sum of elements from \a __A and \a __B; higher order elements are zeroed.
+static __inline__ __m128i __DEFAULT_FN_ATTRS128
+_mm_cvtbiasph_bf8(__m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8_128_mask(
       (__v16qi)__A, (__v8hf)__B, (__v16qi)_mm_undefined_si128(), (__mmask8)-1);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements and 8-bit integers stored in the lower half of
+/// packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	ELSE
+/// 		dst.fp8[i] := _W[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the sum of
+///    elements from \a __A and \a __B; higher order elements are zeroed. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvtbiasph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8_128_mask(
       (__v16qi)__A, (__v8hf)__B, (__v16qi)(__m128i)__W, (__mmask8)__U);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+///    Zeroing mask \a __U is used to determine if given element should be zeroed
+///    instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 ELSE
+///	 	dst.fp8[i] := 0
+///	 FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
+///
+/// \param __U
+///    A 8-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the sum of
+///    elements from \a __A and \a __B; higher order elements are zeroed. If
+///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtbiasph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8_128_mask(
@@ -96,6 +510,29 @@ _mm_maskz_cvtbiasph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
       (__mmask8)__U);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x bf8]. Elements correspond to the
+///    sum of elements from \a __A and \a __B.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_cvtbiasph_bf8(__m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8_256_mask(
@@ -103,12 +540,78 @@ _mm256_cvtbiasph_bf8(__m256i __A, __m256h __B) {
       (__mmask16)-1);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements and 8-bit integers stored in the lower half of
+/// packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	ELSE
+/// 		dst.fp8[i] := _W[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Elements correspond to the sum of
+///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_bf8(
     __m128i __W, __mmask16 __U, __m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8_256_mask(
       (__v32qi)__A, (__v16hf)__B, (__v16qi)(__m128i)__W, (__mmask16)__U);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+///    Merging mask \a __U is used to determine if given element should be taken
+///    from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 ELSE
+///	 	dst.fp8[i] := 0
+///	 FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x bf8]. Elements correspond to the sum of
+///    elements from \a __A and \a __B. If corresponding mask bit is not set,
+///    then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtbiasph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8_256_mask(
@@ -116,18 +619,108 @@ _mm256_maskz_cvtbiasph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
       (__mmask16)__U);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+///    Results are saturated.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the
+///    sum of elements from \a __A and \a __B; higher order elements are zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_cvtbiassph_bf8(__m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8s_128_mask(
       (__v16qi)__A, (__v8hf)__B, (__v16qi)_mm_undefined_si128(), (__mmask8)-1);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements and 8-bit integers stored in the lower half of
+/// packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+/// Results are saturated. Merging mask \a __U is used to determine if given
+/// element should be taken from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	ELSE
+/// 		dst.fp8[i] := _W[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the sum of
+///    elements from \a __A and \a __B; higher order elements are zeroed. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvtbiassph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8s_128_mask(
       (__v16qi)__A, (__v8hf)__B, (__v16qi)(__m128i)__W, (__mmask8)__U);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+///    Results are saturated. Zeroing mask \a __U is used to determine if given
+///    element should be zeroed instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 ELSE
+///	 	dst.fp8[i] := 0
+///	 FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8S instruction.
+///
+/// \param __U
+///    A 8-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the sum of
+///    elements from \a __A and \a __B; higher order elements are zeroed. If
+///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtbiassph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8s_128_mask(
@@ -135,6 +728,30 @@ _mm_maskz_cvtbiassph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
       (__mmask8)__U);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+///    Results are saturated.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8S instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x bf8]. Elements correspond to the
+///    sum of elements from \a __A and \a __B.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_cvtbiassph_bf8(__m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8s_256_mask(
@@ -142,12 +759,78 @@ _mm256_cvtbiassph_bf8(__m256i __A, __m256h __B) {
       (__mmask16)-1);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements and 8-bit integers stored in the lower half of
+/// packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+/// Results are saturated. Merging mask \a __U is used to determine if given
+/// element should be taken from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	ELSE
+/// 		dst.fp8[i] := _W[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Elements correspond to the sum of
+///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_bf8(
     __m128i __W, __mmask16 __U, __m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8s_256_mask(
       (__v32qi)__A, (__v16hf)__B, (__v16qi)(__m128i)__W, (__mmask16)__U);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E5M2.
+///    Results are saturated. Merging mask \a __U is used to determine if given
+///    element should be taken from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 ELSE
+///	 	dst.fp8[i] := 0
+///	 FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2BF8S instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x bf8]. Elements correspond to the sum of
+///    elements from \a __A and \a __B. If corresponding mask bit is not set,
+///    then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtbiassph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2bf8s_256_mask(
@@ -155,18 +838,107 @@ _mm256_maskz_cvtbiassph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
       (__mmask16)__U);
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtbiasph_hf8(__m128i __A,
-                                                                  __m128h __B) {
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8 instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the
+///    sum of elements from \a __A and \a __B; higher order elements are zeroed.
+static __inline__ __m128i __DEFAULT_FN_ATTRS128
+_mm_cvtbiasph_hf8(__m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8_128_mask(
       (__v16qi)__A, (__v8hf)__B, (__v16qi)_mm_undefined_si128(), (__mmask8)-1);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements and 8-bit integers stored in the lower half of
+/// packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	ELSE
+/// 		dst.fp8[i] := _W[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the sum of
+///    elements from \a __A and \a __B; higher order elements are zeroed. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvtbiasph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8_128_mask(
       (__v16qi)__A, (__v8hf)__B, (__v16qi)(__m128i)__W, (__mmask8)__U);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+///    Zeroing mask \a __U is used to determine if given element should be zeroed
+///    instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 ELSE
+///	 	dst.fp8[i] := 0
+///	 FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8 instruction.
+///
+/// \param __U
+///    A 8-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the sum of
+///    elements from \a __A and \a __B; higher order elements are zeroed. If
+///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtbiasph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8_128_mask(
@@ -174,6 +946,29 @@ _mm_maskz_cvtbiasph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
       (__mmask8)__U);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	dst.fp8[i] := add_fp16_int8_convert_to_hf8(__A.fp16[i], __B.int8[2 * i])
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8 instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x half].
+/// \param __B
+///    A 256-bit vector of [16 x i16].
+/// \returns
+///    A 128-bit vector of [16 x hf8]. Elements correspond to the
+///    sum of elements from \a __A and \a __B.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_cvtbiasph_hf8(__m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8_256_mask(
@@ -181,12 +976,78 @@ _mm256_cvtbiasph_hf8(__m256i __A, __m256h __B) {
       (__mmask16)-1);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements and 8-bit integers stored in the lower half of
+/// packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	ELSE
+/// 		dst.fp8[i] := _W[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Elements correspond to the sum of
+///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_hf8(
     __m128i __W, __mmask16 __U, __m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8_256_mask(
       (__v32qi)__A, (__v16hf)__B, (__v16qi)(__m128i)__W, (__mmask16)__U);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E4M3 
+///    Merging mask \a __U is used to determine if given element should be taken
+///    from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+///	 	dst.fp8[i] := add_fp16_int8_convert_to_hf8(__A.fp16[i], __B.int8[2 * i])
+///	 ELSE
+///	 	dst.fp8[i] := 0
+///	 FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8 instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \param __A
+///    A 256-bit vector of [16 x half].
+/// \param __B
+///    A 256-bit vector of [16 x i16].
+/// \returns
+///    A 128-bit vector of [16 x hf8]. Elements correspond to the sum of
+///    elements from \a __A and \a __B. If corresponding mask bit is not set,
+///    then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtbiasph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8_256_mask(
@@ -194,18 +1055,108 @@ _mm256_maskz_cvtbiasph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
       (__mmask16)__U);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+///    Results are saturated.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8S`instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the
+///    sum of elements from \a __A and \a __B; higher order elements are zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_cvtbiassph_hf8(__m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8s_128_mask(
       (__v16qi)__A, (__v8hf)__B, (__v16qi)_mm_undefined_si128(), (__mmask8)-1);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements and 8-bit integers stored in the lower half of
+/// packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+/// Results are saturated. Merging mask \a __U is used to determine if given
+/// element should be taken from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	ELSE
+/// 		dst.fp8[i] := _W[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the sum of
+///    elements from \a __A and \a __B; higher order elements are zeroed. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvtbiassph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8s_128_mask(
       (__v16qi)__A, (__v8hf)__B, (__v16qi)(__m128i)__W, (__mmask8)__U);
 }
 
+/// Add two 128-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+///    Results are saturated. Zeroing mask \a __U is used to determine if given
+///    element should be zeroed instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 ELSE
+///	 	dst.fp8[i] := 0
+///	 FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8S instruction.
+///
+/// \param __U
+///    A 8-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x int16].
+/// \returns
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the sum of
+///    elements from \a __A and \a __B; higher order elements are zeroed. If
+///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtbiassph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8s_128_mask(
@@ -213,6 +1164,30 @@ _mm_maskz_cvtbiassph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
       (__mmask8)__U);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+///    Results are saturated.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8S instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x hf8]. Elements correspond to the
+///    sum of elements from \a __A and \a __B.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_cvtbiassph_hf8(__m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8s_256_mask(
@@ -220,12 +1195,78 @@ _mm256_cvtbiassph_hf8(__m256i __A, __m256h __B) {
       (__mmask16)-1);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements and 8-bit integers stored in the lower half of
+/// packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+/// Results are saturated. Merging mask \a __U is used to determine if given
+/// element should be taken from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	ELSE
+/// 		dst.fp8[i] := _W[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Elements correspond to the sum of
+///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_hf8(
     __m128i __W, __mmask16 __U, __m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8s_256_mask(
       (__v32qi)__A, (__v16hf)__B, (__v16qi)(__m128i)__W, (__mmask16)__U);
 }
 
+/// Add two 256-bit vectors, \a __A and \a __B, containing packed FP16
+///    floating-point elements and 8-bit integers stored in the lower half of
+///    packed 16-bit integers, respectively. Results are converted to FP8 E4M3.
+///    Results are saturated. Merging mask \a __U is used to determine if given
+///    element should be taken from \a __W instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 ELSE
+///	 	dst.fp8[i] := 0
+///	 FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTBIASPH2HF8S instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x int16].
+/// \returns
+///    A 128-bit vector of [16 x hf8]. Elements correspond to the sum of
+///    elements from \a __A and \a __B. If corresponding mask bit is not set,
+///    then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtbiassph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
   return (__m128i)__builtin_ia32_vcvtbiasph2hf8s_256_mask(
@@ -233,17 +1274,119 @@ _mm256_maskz_cvtbiassph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
       (__mmask16)__U);
 }
 
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E5M2 FP8 elements.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15 
+/// 	IF i < 8
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8 instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_bf8(__m128h __A,
-                                                               __m128h __B) {
-  return (__m128i)__builtin_ia32_vcvt2ph2bf8_128((__v8hf)(__A), (__v8hf)(__B));
-}
-
+                                                                  __m128h __B) {
+  return (__m128i)__builtin_ia32_vcvt2ph2bf8_128((__v8hf)(__A),
+                                                   (__v8hf)(__B));
+}
+
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E5M2 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvt2ph_bf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
   return (__m128i)__builtin_ia32_selectb_128(
       (__mmask16)__U, (__v16qi)_mm_cvt2ph_bf8(__A, __B), (__v16qi)__W);
 }
 
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E5M2 FP8 elements.
+/// Zeroing mask \a __U is used to determine if given element should be zeroed
+/// instead.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8 instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    zero is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvt2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
   return (__m128i)__builtin_ia32_selectb_128(
@@ -251,18 +1394,119 @@ _mm_maskz_cvt2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
       (__v16qi)(__m128i)_mm_setzero_si128());
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_cvt2ph_bf8(__m256h __A,
-                                                                  __m256h __B) {
-  return (__m256i)__builtin_ia32_vcvt2ph2bf8_256((__v16hf)(__A),
-                                                 (__v16hf)(__B));
-}
-
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E5M2 FP8 elements.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF i < 16 
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8 instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_cvt2ph_bf8(__m256i __W, __mmask32 __U, __m256h __A, __m256h __B) {
+_mm256_cvt2ph_bf8(__m256h __A, __m256h __B) {
+  return (__m256i)__builtin_ia32_vcvt2ph2bf8_256((__v16hf)(__A),
+                                                   (__v16hf)(__B));
+}
+
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E5M2 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF __U[i]
+/// 		IF i < 16 
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8[i] := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8 instruction.
+///
+/// \param __W
+///    A 256-bit vector of [32 x fp8].
+/// \param __U
+///    A 32-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_bf8(
+    __m256i __W, __mmask32 __U, __m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_selectb_256(
       (__mmask16)__U, (__v32qi)_mm256_cvt2ph_bf8(__A, __B), (__v32qi)__W);
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E5M2 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be zeroed
+/// instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF __U[i]
+/// 		dst.fp8[i] := 0
+/// 	ELSE
+/// 		IF i < 16 
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		FI
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8 instruction.
+///
+/// \param __U
+///    A 32-bit zeroing mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    zero is taken instead.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvt2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_selectb_256(
@@ -270,17 +1514,120 @@ _mm256_maskz_cvt2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
       (__v32qi)(__m256i)_mm256_setzero_si256());
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvts2ph_bf8(__m128h __A,
-                                                                __m128h __B) {
-  return (__m128i)__builtin_ia32_vcvt2ph2bf8s_128((__v8hf)(__A), (__v8hf)(__B));
-}
-
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E5M2 FP8 elements.
+/// Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15 
+/// 	IF i < 8
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8S instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
+static __inline__ __m128i __DEFAULT_FN_ATTRS128
+_mm_cvts2ph_bf8(__m128h __A, __m128h __B) {
+  return (__m128i)__builtin_ia32_vcvt2ph2bf8s_128((__v8hf)(__A),
+                                                    (__v8hf)(__B));
+}
+
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E5M2 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead. Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation}
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvts2ph_bf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
   return (__m128i)__builtin_ia32_selectb_128(
       (__mmask16)__U, (__v16qi)_mm_cvts2ph_bf8(__A, __B), (__v16qi)__W);
 }
 
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E5M2 FP8 elements.
+/// Zeroing mask \a __U is used to determine if given element should be zeroed
+/// instead. Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 16 
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8S instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    zero is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvts2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
   return (__m128i)__builtin_ia32_selectb_128(
@@ -288,18 +1635,120 @@ _mm_maskz_cvts2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
       (__v16qi)(__m128i)_mm_setzero_si128());
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E5M2 FP8 elements.
+/// Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF i < 16 
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8S instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_cvts2ph_bf8(__m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_vcvt2ph2bf8s_256((__v16hf)(__A),
-                                                  (__v16hf)(__B));
-}
-
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_cvts2ph_bf8(__m256i __W, __mmask32 __U, __m256h __A, __m256h __B) {
+                                                    (__v16hf)(__B));
+}
+
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E5M2 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead. Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF __U[i]
+/// 		IF i < 16 
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8[i] := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8S instruction.
+///
+/// \param __W
+///    A 256-bit vector of [32 x fp8].
+/// \param __U
+///    A 32-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_bf8(
+    __m256i __W, __mmask32 __U, __m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_selectb_256(
       (__mmask16)__U, (__v32qi)_mm256_cvts2ph_bf8(__A, __B), (__v32qi)__W);
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E5M2 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be zeroed
+/// instead. Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF __U[i]
+/// 		dst.fp8[i] := 0
+/// 	ELSE
+/// 		IF i < 16 
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		FI
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2BF8S instruction.
+///
+/// \param __U
+///    A 32-bit zeroing mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    zero is taken instead.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvts2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_selectb_256(
@@ -307,17 +1756,119 @@ _mm256_maskz_cvts2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
       (__v32qi)(__m256i)_mm256_setzero_si256());
 }
 
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E4M3 FP8 elements.
+///
+/// \code{.operation]
+/// FOR i := 0 to 16 
+/// 	IF i < 8
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8 instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_hf8(__m128h __A,
-                                                               __m128h __B) {
-  return (__m128i)__builtin_ia32_vcvt2ph2hf8_128((__v8hf)(__A), (__v8hf)(__B));
-}
-
+                                                                  __m128h __B) {
+  return (__m128i)__builtin_ia32_vcvt2ph2hf8_128((__v8hf)(__A),
+                                                   (__v8hf)(__B));
+}
+
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E4M3 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 16 
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvt2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
   return (__m128i)__builtin_ia32_selectb_128(
       (__mmask16)__U, (__v16qi)_mm_cvt2ph_hf8(__A, __B), (__v16qi)__W);
 }
 
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E4M3 FP8 elements.
+/// Zeroing mask \a __U is used to determine if given element should be zeroed
+/// instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 16 
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8 instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    zero is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvt2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
   return (__m128i)__builtin_ia32_selectb_128(
@@ -325,18 +1876,119 @@ _mm_maskz_cvt2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
       (__v16qi)(__m128i)_mm_setzero_si128());
 }
 
-static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_cvt2ph_hf8(__m256h __A,
-                                                                  __m256h __B) {
-  return (__m256i)__builtin_ia32_vcvt2ph2hf8_256((__v16hf)(__A),
-                                                 (__v16hf)(__B));
-}
-
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E4M3 FP8 elements.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF i < 16 
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8 instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_cvt2ph_hf8(__m256i __W, __mmask32 __U, __m256h __A, __m256h __B) {
+_mm256_cvt2ph_hf8(__m256h __A, __m256h __B) {
+  return (__m256i)__builtin_ia32_vcvt2ph2hf8_256((__v16hf)(__A),
+                                                   (__v16hf)(__B));
+}
+
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E4M3 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF __U[i]
+/// 		IF i < 16 
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8[i] := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8 instruction.
+///
+/// \param __W
+///    A 256-bit vector of [32 x fp8].
+/// \param __U
+///    A 32-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_hf8(
+    __m256i __W, __mmask32 __U, __m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_selectb_256(
       (__mmask16)__U, (__v32qi)_mm256_cvt2ph_hf8(__A, __B), (__v32qi)__W);
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E4M3 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be zeroed
+/// instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF __U[i]
+/// 		dst.fp8[i] := 0
+/// 	ELSE
+/// 		IF i < 16 
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		FI
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8 instruction.
+///
+/// \param __U
+///    A 32-bit zeroing mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    zero is taken instead.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvt2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_selectb_256(
@@ -344,17 +1996,120 @@ _mm256_maskz_cvt2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
       (__v32qi)(__m256i)_mm256_setzero_si256());
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvts2ph_hf8(__m128h __A,
-                                                                __m128h __B) {
-  return (__m128i)__builtin_ia32_vcvt2ph2hf8s_128((__v8hf)(__A), (__v8hf)(__B));
-}
-
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E4M3 FP8 elements.
+/// Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 16 
+/// 	IF i < 8
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8S instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
+static __inline__ __m128i __DEFAULT_FN_ATTRS128
+_mm_cvts2ph_hf8(__m128h __A, __m128h __B) {
+  return (__m128i)__builtin_ia32_vcvt2ph2hf8s_128((__v8hf)(__A),
+                                                    (__v8hf)(__B));
+}
+
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E4M3 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead. Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 16 
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvts2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
   return (__m128i)__builtin_ia32_selectb_128(
       (__mmask16)__U, (__v16qi)_mm_cvts2ph_hf8(__A, __B), (__v16qi)__W);
 }
 
+/// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 128-bit vector containing E4M3 FP8 elements.
+/// Zeroing mask \a __U is used to determine if given element should be zeroed
+/// instead. Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 16 
+/// 	IF __U[i]
+/// 		IF i < 8
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8S instruction.
+///
+/// \param __U
+///    A 16-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \param __B
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    zero is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvts2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
   return (__m128i)__builtin_ia32_selectb_128(
@@ -362,18 +2117,120 @@ _mm_maskz_cvts2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
       (__v16qi)(__m128i)_mm_setzero_si128());
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E4M3 FP8 elements.
+/// Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF i < 16 
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8S instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_cvts2ph_hf8(__m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_vcvt2ph2hf8s_256((__v16hf)(__A),
-                                                  (__v16hf)(__B));
-}
-
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_mask_cvts2ph_hf8(__m256i __W, __mmask32 __U, __m256h __A, __m256h __B) {
+                                                    (__v16hf)(__B));
+}
+
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E4M3 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead. Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF __U[i]
+/// 		IF i < 16 
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		FI
+/// 	ELSE
+/// 		dst.fp8[i] := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8S instruction.
+///
+/// \param __W
+///    A 256-bit vector of [32 x fp8].
+/// \param __U
+///    A 32-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
+static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_hf8(
+    __m256i __W, __mmask32 __U, __m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_selectb_256(
       (__mmask16)__U, (__v32qi)_mm256_cvts2ph_hf8(__A, __B), (__v32qi)__W);
 }
 
+/// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
+/// floating-point elements to a 256-bit vector containing E4M3 FP8 elements.
+/// Merging mask \a __U is used to determine if given element should be zeroed
+/// instead. Resulting elements are saturated in case of overflow.
+///
+/// \code{.operation]
+/// FOR i := 0 to 32 
+/// 	IF __U[i]
+/// 		dst.fp8[i] := 0
+/// 	ELSE
+/// 		IF i < 16 
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		ELSE
+/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		FI
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVT2PH2HF8S instruction.
+///
+/// \param __U
+///    A 32-bit zeroing mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \param __B
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    (converted) elements from \a __B; higher order elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    zero is taken instead.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvts2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
   return (__m256i)__builtin_ia32_selectb_256(
@@ -381,208 +2238,1166 @@ _mm256_maskz_cvts2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
       (__v32qi)(__m256i)_mm256_setzero_si256());
 }
 
+/// Convert 128-bit vector \a __A, containing packed FP8 E4M3 floating-point
+/// elements to a 128-bit vector containing FP16 elements. The conversion is exact.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTHF82PH instruction.
+///
+/// \param __A
+///    A 128-bit vector of [16 x fp8].
+/// \returns
+///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvthf8(__m128i __A) {
   return (__m128h)__builtin_ia32_vcvthf8_2ph128_mask(
       (__v16qi)__A, (__v8hf)(__m128h)_mm_undefined_ph(), (__mmask8)-1);
 }
 
-static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_mask_cvthf8(__m128h __W,
-                                                                __mmask8 __U,
-                                                                __m128i __A) {
+/// Convert 128-bit vector \a __A, containing packed FP8 E4M3 floating-point
+/// elements to a 128-bit vector containing FP16 elements. The conversion is
+/// exact. Merging mask \a __U is used to determine if given element should be
+/// taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp16[i] := convert_fp8_to_fp16(__A.fp8[i])
+/// 	ELSE
+/// 		dst.fp16[i] := __W.fp16[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTHF82PH instruction.
+///
+/// \param __W
+///    A 128-bit vector of [8 x fp16].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [16 x fp8].
+/// \returns
+///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
+static __inline__ __m128h __DEFAULT_FN_ATTRS128
+_mm_mask_cvthf8(__m128h __W, __mmask8 __U, __m128i __A) {
   return (__m128h)__builtin_ia32_vcvthf8_2ph128_mask(
       (__v16qi)__A, (__v8hf)(__m128h)__W, (__mmask8)__U);
 }
 
-static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_maskz_cvthf8(__mmask8 __U,
-                                                                 __m128i __A) {
+/// Convert 128-bit vector \a __A, containing packed FP8 E4M3 floating-point
+/// elements to a 128-bit vector containing FP16 elements. The conversion is
+/// exact. Zeroing mask \a __U is used to determine if given element should be
+/// zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	ELSE
+/// 		dst.fp16[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTHF82PH instruction.
+///
+/// \param __U
+///    A 8-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [16 x fp8].
+/// \returns
+///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    zero is taken instead.
+static __inline__ __m128h __DEFAULT_FN_ATTRS128
+_mm_maskz_cvthf8(__mmask8 __U, __m128i __A) {
   return (__m128h)__builtin_ia32_vcvthf8_2ph128_mask(
       (__v16qi)__A, (__v8hf)(__m128h)_mm_setzero_ph(), (__mmask8)__U);
 }
 
-static __inline__ __m256h __DEFAULT_FN_ATTRS256 _mm256_cvthf8(__m128i __A) {
+/// Convert 256-bit vector \a __A, containing packed FP8 E4M3 floating-point
+/// elements to a 256-bit vector containing FP16 elements. The conversion is exact.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTHF82PH instruction.
+///
+/// \param __A
+///    A 256-bit vector of [32 x fp8].
+/// \returns
+///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A.
+static __inline__ __m256h __DEFAULT_FN_ATTRS256
+_mm256_cvthf8(__m128i __A) {
   return (__m256h)__builtin_ia32_vcvthf8_2ph256_mask(
       (__v16qi)__A, (__v16hf)(__m256h)_mm256_undefined_ph(), (__mmask16)-1);
 }
 
+/// Convert 256-bit vector \a __A, containing packed FP8 E4M3 floating-point
+/// elements to a 256-bit vector containing FP16 elements. The conversion is
+/// exact. Merging mask \a __U is used to determine if given element should be
+/// taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		dst.fp16[i] := convert_fp8_to_fp16(__A.fp8[i])
+/// 	ELSE
+/// 		dst.fp16[i] := __W.fp16[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTHF82PH instruction.
+///
+/// \param __W
+///    A 256-bit vector of [16 x fp16].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [32 x fp8].
+/// \returns
+///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m256h __DEFAULT_FN_ATTRS256
 _mm256_mask_cvthf8(__m256h __W, __mmask16 __U, __m128i __A) {
   return (__m256h)__builtin_ia32_vcvthf8_2ph256_mask(
       (__v16qi)__A, (__v16hf)(__m256h)__W, (__mmask16)__U);
 }
 
+/// Convert 256-bit vector \a __A, containing packed FP8 E4M3 floating-point
+/// elements to a 256-bit vector containing FP16 elements. The conversion is
+/// exact. Zeroing mask \a __U is used to determine if given element should be
+/// zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	ELSE
+/// 		dst.fp16[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:256] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTHF82PH instruction.
+///
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [32 x fp8].
+/// \returns
+///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    zero is taken instead.
 static __inline__ __m256h __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvthf8(__mmask16 __U, __m128i __A) {
   return (__m256h)__builtin_ia32_vcvthf8_2ph256_mask(
       (__v16qi)__A, (__v16hf)(__m256h)_mm256_setzero_ph(), (__mmask16)__U);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
+/// resulting vector are zeroed.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8 instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the (converted)
+///    elements from \a __A; upper elements are zeroed. 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_bf8(__m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)_mm_undefined_si128(), (__mmask8)-1);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Merging mask \a __U is used to determine if
+/// given element should be taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    (converted) elements from \a __A; upper elements are zeroed. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvtph_bf8(__m128i __W, __mmask8 __U, __m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)__W, (__mmask8)__U);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Zeroing mask \a __U is used to determine if
+/// given element should be zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := 0
+/// 	ELSE
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8 instruction.
+///
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    (converted) elements from \a __A; upper elements are zeroed. If
+///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtph_bf8(__mmask8 __U, __m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)_mm_setzero_si128(), (__mmask8)__U);
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_cvtph_bf8(__m256h __A) {
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8 instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the (converted)
+///    elements from \a __A.
+static __inline__ __m128i __DEFAULT_FN_ATTRS256
+_mm256_cvtph_bf8(__m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)_mm_undefined_si128(), (__mmask16)-1);
 }
 
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Merging mask \a __U is
+/// used to determine if given element should be taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_mask_cvtph_bf8(__m128i __W, __mmask16 __U, __m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)__W, (__mmask16)__U);
 }
 
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Zeroing mask \a __U is
+/// used to determine if given element should be zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8 instruction.
+///
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    then element is zeroed instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtph_bf8(__mmask16 __U, __m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)_mm_setzero_si128(), (__mmask16)__U);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Results are saturated.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8S instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the (converted)
+///    elements from \a __A; upper elements are zeroed. 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_bf8(__m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8s_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)_mm_undefined_si128(), (__mmask8)-1);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Results are saturated. Merging mask \a __U is
+/// used to determine if given element should be taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    (converted) elements from \a __A; upper elements are zeroed. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvtsph_bf8(__m128i __W, __mmask8 __U, __m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8s_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)__W, (__mmask8)__U);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Results are saturated. Zeroing mask \a __U is
+/// used to determine if given element should be zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8S instruction.
+///
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    (converted) elements from \a __A; upper elements are zeroed. If
+///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtsph_bf8(__mmask8 __U, __m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8s_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)_mm_setzero_si128(), (__mmask8)__U);
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_cvtsph_bf8(__m256h __A) {
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Results are saturated.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8S instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the (converted)
+///    elements from \a __A.
+static __inline__ __m128i __DEFAULT_FN_ATTRS256
+_mm256_cvtsph_bf8(__m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8s_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)_mm_undefined_si128(), (__mmask16)-1);
 }
 
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Results are saturated.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_mask_cvtsph_bf8(__m128i __W, __mmask16 __U, __m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8s_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)__W, (__mmask16)__U);
 }
 
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Results are saturated.
+/// Zeroing mask \a __U is used to determine if given element should be zeroed
+/// instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2BF8S instruction.
+///
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    then element is zeroed instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtsph_bf8(__mmask16 __U, __m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8s_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)_mm_setzero_si128(), (__mmask16)__U);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
+/// resulting vector are zeroed.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8 instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the (converted)
+///    elements from \a __A; upper elements are zeroed. 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_hf8(__m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)_mm_undefined_si128(), (__mmask8)-1);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Merging mask \a __U is used to determine if
+/// given element should be taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    (converted) elements from \a __A; upper elements are zeroed. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvtph_hf8(__m128i __W, __mmask8 __U, __m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)__W, (__mmask8)__U);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Zeroing mask \a __U is used to determine if
+/// given element should be zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8 instruction.
+///
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    (converted) elements from \a __A; upper elements are zeroed. If
+///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtph_hf8(__mmask8 __U, __m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)_mm_setzero_si128(), (__mmask8)__U);
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_cvtph_hf8(__m256h __A) {
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8 instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the (converted)
+///    elements from \a __A.
+static __inline__ __m128i __DEFAULT_FN_ATTRS256
+_mm256_cvtph_hf8(__m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)_mm_undefined_si128(), (__mmask16)-1);
 }
 
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Merging mask \a __U is
+/// used to determine if given element should be taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8 instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_mask_cvtph_hf8(__m128i __W, __mmask16 __U, __m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)__W, (__mmask16)__U);
 }
 
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Zeroing mask \a __U is
+/// used to determine if given element should be zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8 instruction.
+///
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    then element is zeroed instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtph_hf8(__mmask16 __U, __m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)_mm_setzero_si128(), (__mmask16)__U);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Results are saturated.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8S instruction.
+///
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the (converted)
+///    elements from \a __A; upper elements are zeroed. 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_hf8(__m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8s_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)_mm_undefined_si128(), (__mmask8)-1);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Results are saturated. Merging mask \a __U is
+/// used to determine if given element should be taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    (converted) elements from \a __A; upper elements are zeroed. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_mask_cvtsph_hf8(__m128i __W, __mmask8 __U, __m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8s_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)__W, (__mmask8)__U);
 }
 
+/// Convert 128-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Upper elements of
+/// resulting vector are zeroed. Results are saturated. Zeroing mask \a __U is
+/// used to determine if given element should be zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8S instruction.
+///
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    (converted) elements from \a __A; upper elements are zeroed. If
+///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtsph_hf8(__mmask8 __U, __m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8s_128_mask(
       (__v8hf)__A, (__v16qi)(__m128i)_mm_setzero_si128(), (__mmask8)__U);
 }
 
-static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_cvtsph_hf8(__m256h __A) {
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Results are saturated.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8S instruction.
+///
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the (converted)
+///    elements from \a __A.
+static __inline__ __m128i __DEFAULT_FN_ATTRS256
+_mm256_cvtsph_hf8(__m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8s_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)_mm_undefined_si128(), (__mmask16)-1);
 }
 
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Results are saturated.
+/// Merging mask \a __U is used to determine if given element should be taken
+/// from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8 := __W.fp8[i]
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:128] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8S instruction.
+///
+/// \param __W
+///    A 128-bit vector of [16 x fp8].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [8 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If
+///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_mask_cvtsph_hf8(__m128i __W, __mmask16 __U, __m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8s_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)__W, (__mmask16)__U);
 }
 
+/// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
+/// to a 128-bit vector containing E4M3 FP8 elements. Results are saturated.
+/// Zeroing mask \a __U is used to determine if given element should be zeroed
+/// instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	ELSE
+/// 		dst.fp8[i] := 0
+/// 	FI
+/// ENDFOR
+///
+/// dst[MAX:64] := 0
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic corresponds to the \c VCVTPH2HF8S instruction.
+///
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [16 x fp16].
+/// \returns
+///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set,
+///    then element is zeroed instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtsph_hf8(__mmask16 __U, __m256h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8s_256_mask(
       (__v16hf)__A, (__v16qi)(__m128i)_mm_setzero_si128(), (__mmask16)__U);
 }
 
+/// Convert 128-bit vector \a __A, containing packed FP8 E5M2 floating-point
+/// elements to a 128-bit vector containing FP16 elements. The conversion is exact.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// ENDFOR
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic does not correspond to a single instruction.
+///
+/// \param __A
+///    A 128-bit vector of [16 x fp8].
+/// \returns
+///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvtbf8_ph(__m128i __A) {
   return _mm_castsi128_ph(_mm_slli_epi16(_mm_cvtepi8_epi16(__A), 8));
 }
 
+/// Convert 128-bit vector \a __A, containing packed FP8 E5M2 floating-point
+/// elements to a 128-bit vector containing FP16 elements. The conversion is
+/// exact. Merging mask \a __U is used to determine if given element should be
+/// taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	ELSE
+/// 		dst.fp16[i] := __W.fp16[i]
+/// 	FI
+/// ENDFOR
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic does not correspond to a single instruction.
+///
+/// \param __W
+///    A 128-bit vector of [8 x fp16].
+/// \param __U
+///    A 8-bit merging mask.
+/// \param __A
+///    A 128-bit vector of [16 x fp8].
+/// \returns
+///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m128h __DEFAULT_FN_ATTRS128
-_mm_mask_cvtbf8_ph(__m128h __S, __mmask8 __U, __m128i __A) {
+_mm_mask_cvtbf8_ph(__m128h __W, __mmask8 __U, __m128i __A) {
   return _mm_castsi128_ph(
-      _mm_mask_slli_epi16((__m128i)__S, __U, _mm_cvtepi8_epi16(__A), 8));
-}
-
+      _mm_mask_slli_epi16((__m128i)__W, __U, _mm_cvtepi8_epi16(__A), 8));
+}
+
+/// Convert 128-bit vector \a __A, containing packed FP8 E5M2 floating-point
+/// elements to a 128-bit vector containing FP16 elements. The conversion is
+/// exact. Zeroing mask \a __U is used to determine if given element should be
+/// zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 7
+/// 	IF __U[i]
+/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	ELSE
+/// 		dst.fp16[i] := 0
+/// 	FI
+/// ENDFOR
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic does not correspond to a single instruction.
+///
+/// \param __U
+///    A 8-bit zeroing mask.
+/// \param __A
+///    A 128-bit vector of [16 x fp8].
+/// \returns
+///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    zero is taken instead.
 static __inline__ __m128h __DEFAULT_FN_ATTRS128
 _mm_maskz_cvtbf8_ph(__mmask8 __U, __m128i __A) {
   return _mm_castsi128_ph(_mm_slli_epi16(_mm_maskz_cvtepi8_epi16(__U, __A), 8));
 }
 
+/// Convert 256-bit vector \a __A, containing packed FP8 E4M3 floating-point
+/// elements to a 256-bit vector containing FP16 elements. The conversion is exact.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15
+/// 	dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// ENDFOR
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic does not correspond to a single instruction.
+///
+/// \param __A
+///    A 256-bit vector of [32 x fp8].
+/// \returns
+///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A.
 static __inline__ __m256h __DEFAULT_FN_ATTRS256 _mm256_cvtbf8_ph(__m128i __A) {
   return _mm256_castsi256_ph(_mm256_slli_epi16(_mm256_cvtepi8_epi16(__A), 8));
 }
 
+/// Convert 256-bit vector \a __A, containing packed FP8 E5M2 floating-point
+/// elements to a 256-bit vector containing FP16 elements. The conversion is
+/// exact. Merging mask \a __U is used to determine if given element should be
+/// taken from \a __W instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	ELSE
+/// 		dst.fp16[i] := __W.fp16[i]
+/// 	FI
+/// ENDFOR
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic does not correspond to a single instruction.
+///
+/// \param __W
+///    A 256-bit vector of [16 x fp16].
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [32 x fp8].
+/// \returns
+///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    element from \a __W is taken instead.
 static __inline__ __m256h __DEFAULT_FN_ATTRS256
-_mm256_mask_cvtbf8_ph(__m256h __S, __mmask16 __U, __m128i __A) {
+_mm256_mask_cvtbf8_ph(__m256h __W, __mmask16 __U, __m128i __A) {
   return _mm256_castsi256_ph(
-      _mm256_mask_slli_epi16((__m256i)__S, __U, _mm256_cvtepi8_epi16(__A), 8));
-}
-
+      _mm256_mask_slli_epi16((__m256i)__W, __U, _mm256_cvtepi8_epi16(__A), 8));
+}
+
+/// Convert 256-bit vector \a __A, containing packed FP8 E5M2 floating-point
+/// elements to a 256-bit vector containing FP16 elements. The conversion is
+/// exact. Zeroing mask \a __U is used to determine if given element should be
+/// zeroed instead.
+///
+/// \code{.operation]
+/// FOR i := 0 to 15 
+/// 	IF __U[i]
+/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	ELSE
+/// 		dst.fp16[i] := 0
+/// 	FI
+/// ENDFOR
+/// \endcode
+///
+/// \headerfile <immintrin.h>
+///
+/// This intrinsic does not correspond to a single instruction.
+///
+/// \param __U
+///    A 16-bit merging mask.
+/// \param __A
+///    A 256-bit vector of [32 x fp8].
+/// \returns
+///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
+///    (converted) elements from \a __A. If corresponding mask bit is not set, then
+///    zero is taken instead.
 static __inline__ __m256h __DEFAULT_FN_ATTRS256
 _mm256_maskz_cvtbf8_ph(__mmask16 __U, __m128i __A) {
   return _mm256_castsi256_ph(
       _mm256_slli_epi16(_mm256_maskz_cvtepi8_epi16(__U, __A), 8));
 }
 
+// clang-format on
+
 #undef __DEFAULT_FN_ATTRS128
 #undef __DEFAULT_FN_ATTRS256
 

>From 66f4f5e6c96a00f0d1a0cc1a362d7479f4cc44d3 Mon Sep 17 00:00:00 2001
From: "Pirog, Mikolaj Maciej" <mikolaj.maciej.pirog at intel.com>
Date: Wed, 29 Jan 2025 13:56:47 +0100
Subject: [PATCH 2/7] Replace all fp8 -> bf8/hf8

---
 clang/lib/Headers/avx10_2convertintrin.h | 498 +++++++++++------------
 1 file changed, 249 insertions(+), 249 deletions(-)

diff --git a/clang/lib/Headers/avx10_2convertintrin.h b/clang/lib/Headers/avx10_2convertintrin.h
index 82d434c18df131..52cd08d6d2d59d 100644
--- a/clang/lib/Headers/avx10_2convertintrin.h
+++ b/clang/lib/Headers/avx10_2convertintrin.h
@@ -408,7 +408,7 @@ _mm256_maskz_cvtx2ps_ph(__mmask16 __U, __m256 __A, __m256 __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -423,7 +423,7 @@ _mm256_maskz_cvtx2ps_ph(__mmask16 __U, __m256 __A, __m256 __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x int16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the
 ///    sum of elements from \a __A and \a __B; higher order elements are zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
 _mm_cvtbiasph_bf8(__m128i __A, __m128h __B) {
@@ -440,9 +440,9 @@ _mm_cvtbiasph_bf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 /// 	ELSE
-/// 		dst.fp8[i] := _W[i]
+/// 		dst.bf8[i] := _W[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -454,7 +454,7 @@ _mm_cvtbiasph_bf8(__m128i __A, __m128h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
@@ -462,7 +462,7 @@ _mm_cvtbiasph_bf8(__m128i __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x int16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the sum of
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -480,9 +480,9 @@ _mm_mask_cvtbiasph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 ///	 ELSE
-///	 	dst.fp8[i] := 0
+///	 	dst.bf8[i] := 0
 ///	 FI
 /// ENDFOR
 ///
@@ -516,7 +516,7 @@ _mm_maskz_cvtbiasph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -549,9 +549,9 @@ _mm256_cvtbiasph_bf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 /// 	ELSE
-/// 		dst.fp8[i] := _W[i]
+/// 		dst.bf8[i] := _W[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -563,7 +563,7 @@ _mm256_cvtbiasph_bf8(__m256i __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
@@ -571,7 +571,7 @@ _mm256_cvtbiasph_bf8(__m256i __A, __m256h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x int16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Elements correspond to the sum of
+///    A 128-bit vector of [16 x bf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_bf8(
@@ -589,9 +589,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_bf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 ///	 ELSE
-///	 	dst.fp8[i] := 0
+///	 	dst.bf8[i] := 0
 ///	 FI
 /// ENDFOR
 ///
@@ -626,7 +626,7 @@ _mm256_maskz_cvtbiasph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -658,9 +658,9 @@ _mm_cvtbiassph_bf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 /// 	ELSE
-/// 		dst.fp8[i] := _W[i]
+/// 		dst.bf8[i] := _W[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -672,7 +672,7 @@ _mm_cvtbiassph_bf8(__m128i __A, __m128h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2BF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
@@ -680,7 +680,7 @@ _mm_cvtbiassph_bf8(__m128i __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x int16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the sum of
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -698,9 +698,9 @@ _mm_mask_cvtbiassph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 ///	 ELSE
-///	 	dst.fp8[i] := 0
+///	 	dst.bf8[i] := 0
 ///	 FI
 /// ENDFOR
 ///
@@ -735,7 +735,7 @@ _mm_maskz_cvtbiassph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -768,9 +768,9 @@ _mm256_cvtbiassph_bf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 /// 	ELSE
-/// 		dst.fp8[i] := _W[i]
+/// 		dst.bf8[i] := _W[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -782,7 +782,7 @@ _mm256_cvtbiassph_bf8(__m256i __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2BF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
@@ -790,7 +790,7 @@ _mm256_cvtbiassph_bf8(__m256i __A, __m256h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x int16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Elements correspond to the sum of
+///    A 128-bit vector of [16 x bf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_bf8(
@@ -808,9 +808,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_bf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
 ///	 ELSE
-///	 	dst.fp8[i] := 0
+///	 	dst.bf8[i] := 0
 ///	 FI
 /// ENDFOR
 ///
@@ -844,7 +844,7 @@ _mm256_maskz_cvtbiassph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -876,9 +876,9 @@ _mm_cvtbiasph_hf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 /// 	ELSE
-/// 		dst.fp8[i] := _W[i]
+/// 		dst.hf8[i] := _W[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -890,7 +890,7 @@ _mm_cvtbiasph_hf8(__m128i __A, __m128h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2HF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
@@ -898,7 +898,7 @@ _mm_cvtbiasph_hf8(__m128i __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x int16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the sum of
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -916,9 +916,9 @@ _mm_mask_cvtbiasph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 ///	 ELSE
-///	 	dst.fp8[i] := 0
+///	 	dst.hf8[i] := 0
 ///	 FI
 /// ENDFOR
 ///
@@ -952,7 +952,7 @@ _mm_maskz_cvtbiasph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.fp8[i] := add_fp16_int8_convert_to_hf8(__A.fp16[i], __B.int8[2 * i])
+/// 	dst.hf8[i] := add_fp16_int8_convert_to_hf8(__A.fp16[i], __B.int8[2 * i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -985,9 +985,9 @@ _mm256_cvtbiasph_hf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 /// 	ELSE
-/// 		dst.fp8[i] := _W[i]
+/// 		dst.hf8[i] := _W[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -999,7 +999,7 @@ _mm256_cvtbiasph_hf8(__m256i __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2HF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
@@ -1007,7 +1007,7 @@ _mm256_cvtbiasph_hf8(__m256i __A, __m256h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x int16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Elements correspond to the sum of
+///    A 128-bit vector of [16 x hf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_hf8(
@@ -1025,9 +1025,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_hf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.fp8[i] := add_fp16_int8_convert_to_hf8(__A.fp16[i], __B.int8[2 * i])
+///	 	dst.hf8[i] := add_fp16_int8_convert_to_hf8(__A.fp16[i], __B.int8[2 * i])
 ///	 ELSE
-///	 	dst.fp8[i] := 0
+///	 	dst.hf8[i] := 0
 ///	 FI
 /// ENDFOR
 ///
@@ -1062,7 +1062,7 @@ _mm256_maskz_cvtbiasph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -1094,9 +1094,9 @@ _mm_cvtbiassph_hf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 /// 	ELSE
-/// 		dst.fp8[i] := _W[i]
+/// 		dst.hf8[i] := _W[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -1108,7 +1108,7 @@ _mm_cvtbiassph_hf8(__m128i __A, __m128h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2HF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
@@ -1116,7 +1116,7 @@ _mm_cvtbiassph_hf8(__m128i __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x int16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the sum of
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -1134,9 +1134,9 @@ _mm_mask_cvtbiassph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 ///	 ELSE
-///	 	dst.fp8[i] := 0
+///	 	dst.hf8[i] := 0
 ///	 FI
 /// ENDFOR
 ///
@@ -1171,7 +1171,7 @@ _mm_maskz_cvtbiassph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -1204,9 +1204,9 @@ _mm256_cvtbiassph_hf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 /// 	ELSE
-/// 		dst.fp8[i] := _W[i]
+/// 		dst.hf8[i] := _W[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -1218,7 +1218,7 @@ _mm256_cvtbiassph_hf8(__m256i __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2HF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
@@ -1226,7 +1226,7 @@ _mm256_cvtbiassph_hf8(__m256i __A, __m256h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x int16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Elements correspond to the sum of
+///    A 128-bit vector of [16 x hf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_hf8(
@@ -1244,9 +1244,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_hf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.fp8[i] := add_convert_fp16_to_fp8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
 ///	 ELSE
-///	 	dst.fp8[i] := 0
+///	 	dst.hf8[i] := 0
 ///	 FI
 /// ENDFOR
 ///
@@ -1280,9 +1280,9 @@ _mm256_maskz_cvtbiassph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF i < 8
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 8])
 /// 	FI
 /// ENDFOR
 ///
@@ -1298,7 +1298,7 @@ _mm256_maskz_cvtbiassph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_bf8(__m128h __A,
@@ -1316,12 +1316,12 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_bf8(__m128h __A,
 /// FOR i := 0 to 15
 /// 	IF __U[i]
 /// 		IF i < 8
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 8])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.bf8 := __W.bf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -1333,7 +1333,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_bf8(__m128h __A,
 /// This intrinsic corresponds to the \c VCVT2PH2BF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
@@ -1341,7 +1341,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_bf8(__m128h __A,
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
@@ -1360,12 +1360,12 @@ _mm_mask_cvt2ph_bf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		IF i < 8
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 8])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.bf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -1383,7 +1383,7 @@ _mm_mask_cvt2ph_bf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    zero is taken instead.
@@ -1400,9 +1400,9 @@ _mm_maskz_cvt2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// \code{.operation]
 /// FOR i := 0 to 32 
 /// 	IF i < 16 
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 16])
 /// 	FI
 /// ENDFOR
 ///
@@ -1418,7 +1418,7 @@ _mm_maskz_cvt2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x bf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
@@ -1436,12 +1436,12 @@ _mm256_cvt2ph_bf8(__m256h __A, __m256h __B) {
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		IF i < 16 
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 16])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8[i] := __W.fp8[i]
+/// 		dst.bf8[i] := __W.bf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -1453,7 +1453,7 @@ _mm256_cvt2ph_bf8(__m256h __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVT2PH2BF8 instruction.
 ///
 /// \param __W
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x bf8].
 /// \param __U
 ///    A 32-bit merging mask.
 /// \param __A
@@ -1461,7 +1461,7 @@ _mm256_cvt2ph_bf8(__m256h __A, __m256h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x bf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
@@ -1479,12 +1479,12 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_bf8(
 /// \code{.operation]
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
-/// 		dst.fp8[i] := 0
+/// 		dst.bf8[i] := 0
 /// 	ELSE
 /// 		IF i < 16 
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 16])
 /// 		FI
 /// 	FI
 /// ENDFOR
@@ -1503,7 +1503,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_bf8(
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x bf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set,
 ///    zero is taken instead.
@@ -1521,9 +1521,9 @@ _mm256_maskz_cvt2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF i < 8
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 8])
 /// 	FI
 /// ENDFOR
 ///
@@ -1539,7 +1539,7 @@ _mm256_maskz_cvt2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -1557,12 +1557,12 @@ _mm_cvts2ph_bf8(__m128h __A, __m128h __B) {
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		IF i < 8
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 8])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.bf8 := __W.bf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -1574,7 +1574,7 @@ _mm_cvts2ph_bf8(__m128h __A, __m128h __B) {
 /// This intrinsic corresponds to the \c VCVT2PH2BF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
@@ -1582,7 +1582,7 @@ _mm_cvts2ph_bf8(__m128h __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
@@ -1601,12 +1601,12 @@ _mm_mask_cvts2ph_bf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 8])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.bf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -1624,7 +1624,7 @@ _mm_mask_cvts2ph_bf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    zero is taken instead.
@@ -1642,9 +1642,9 @@ _mm_maskz_cvts2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// \code{.operation]
 /// FOR i := 0 to 32 
 /// 	IF i < 16 
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 16])
 /// 	FI
 /// ENDFOR
 ///
@@ -1660,7 +1660,7 @@ _mm_maskz_cvts2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x bf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
@@ -1678,12 +1678,12 @@ _mm256_cvts2ph_bf8(__m256h __A, __m256h __B) {
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		IF i < 16 
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 16])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8[i] := __W.fp8[i]
+/// 		dst.bf8[i] := __W.bf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -1695,7 +1695,7 @@ _mm256_cvts2ph_bf8(__m256h __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVT2PH2BF8S instruction.
 ///
 /// \param __W
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x bf8].
 /// \param __U
 ///    A 32-bit merging mask.
 /// \param __A
@@ -1703,7 +1703,7 @@ _mm256_cvts2ph_bf8(__m256h __A, __m256h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x bf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
@@ -1721,12 +1721,12 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_bf8(
 /// \code{.operation]
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
-/// 		dst.fp8[i] := 0
+/// 		dst.bf8[i] := 0
 /// 	ELSE
 /// 		IF i < 16 
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 			dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i - 16])
 /// 		FI
 /// 	FI
 /// ENDFOR
@@ -1745,7 +1745,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_bf8(
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x bf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set,
 ///    zero is taken instead.
@@ -1762,9 +1762,9 @@ _mm256_maskz_cvts2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// \code{.operation]
 /// FOR i := 0 to 16 
 /// 	IF i < 8
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 8])
 /// 	FI
 /// ENDFOR
 ///
@@ -1780,7 +1780,7 @@ _mm256_maskz_cvts2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_hf8(__m128h __A,
@@ -1798,12 +1798,12 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_hf8(__m128h __A,
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 8])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.hf8 := __W.hf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -1815,7 +1815,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_hf8(__m128h __A,
 /// This intrinsic corresponds to the \c VCVT2PH2HF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
@@ -1823,7 +1823,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_hf8(__m128h __A,
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
@@ -1842,12 +1842,12 @@ _mm_mask_cvt2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 8])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.hf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -1865,7 +1865,7 @@ _mm_mask_cvt2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    zero is taken instead.
@@ -1882,9 +1882,9 @@ _mm_maskz_cvt2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// \code{.operation]
 /// FOR i := 0 to 32 
 /// 	IF i < 16 
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 16])
 /// 	FI
 /// ENDFOR
 ///
@@ -1900,7 +1900,7 @@ _mm_maskz_cvt2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x hf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
@@ -1918,12 +1918,12 @@ _mm256_cvt2ph_hf8(__m256h __A, __m256h __B) {
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		IF i < 16 
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 16])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8[i] := __W.fp8[i]
+/// 		dst.hf8[i] := __W.hf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -1935,7 +1935,7 @@ _mm256_cvt2ph_hf8(__m256h __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVT2PH2HF8 instruction.
 ///
 /// \param __W
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x hf8].
 /// \param __U
 ///    A 32-bit merging mask.
 /// \param __A
@@ -1943,7 +1943,7 @@ _mm256_cvt2ph_hf8(__m256h __A, __m256h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x hf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
@@ -1961,12 +1961,12 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_hf8(
 /// \code{.operation]
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
-/// 		dst.fp8[i] := 0
+/// 		dst.hf8[i] := 0
 /// 	ELSE
 /// 		IF i < 16 
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 16])
 /// 		FI
 /// 	FI
 /// ENDFOR
@@ -1985,7 +1985,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_hf8(
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x hf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set,
 ///    zero is taken instead.
@@ -2003,9 +2003,9 @@ _mm256_maskz_cvt2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// \code{.operation]
 /// FOR i := 0 to 16 
 /// 	IF i < 8
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 8])
 /// 	FI
 /// ENDFOR
 ///
@@ -2021,7 +2021,7 @@ _mm256_maskz_cvt2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -2039,12 +2039,12 @@ _mm_cvts2ph_hf8(__m128h __A, __m128h __B) {
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 8])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.hf8 := __W.hf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -2056,7 +2056,7 @@ _mm_cvts2ph_hf8(__m128h __A, __m128h __B) {
 /// This intrinsic corresponds to the \c VCVT2PH2HF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
@@ -2064,7 +2064,7 @@ _mm_cvts2ph_hf8(__m128h __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
@@ -2083,12 +2083,12 @@ _mm_mask_cvts2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 8])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 8])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.hf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -2106,7 +2106,7 @@ _mm_mask_cvts2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// \param __B
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower 8 elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower 8 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    zero is taken instead.
@@ -2124,9 +2124,9 @@ _mm_maskz_cvts2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// \code{.operation]
 /// FOR i := 0 to 32 
 /// 	IF i < 16 
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 16])
 /// 	FI
 /// ENDFOR
 ///
@@ -2142,7 +2142,7 @@ _mm_maskz_cvts2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x hf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A.
 static __inline__ __m256i __DEFAULT_FN_ATTRS256
@@ -2160,12 +2160,12 @@ _mm256_cvts2ph_hf8(__m256h __A, __m256h __B) {
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		IF i < 16 
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 16])
 /// 		FI
 /// 	ELSE
-/// 		dst.fp8[i] := __W.fp8[i]
+/// 		dst.hf8[i] := __W.hf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -2177,7 +2177,7 @@ _mm256_cvts2ph_hf8(__m256h __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVT2PH2HF8S instruction.
 ///
 /// \param __W
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x hf8].
 /// \param __U
 ///    A 32-bit merging mask.
 /// \param __A
@@ -2185,7 +2185,7 @@ _mm256_cvts2ph_hf8(__m256h __A, __m256h __B) {
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x hf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
 ///    element from \a __W is taken instead.
@@ -2203,12 +2203,12 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_hf8(
 /// \code{.operation]
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
-/// 		dst.fp8[i] := 0
+/// 		dst.hf8[i] := 0
 /// 	ELSE
 /// 		IF i < 16 
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__B.fp16[i])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 		ELSE
-/// 			dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i - 16])
+/// 			dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i - 16])
 /// 		FI
 /// 	FI
 /// ENDFOR
@@ -2227,7 +2227,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_hf8(
 /// \param __B
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 256-bit vector of [32 x fp8]. Lower 16 elements correspond to the
+///    A 256-bit vector of [32 x hf8]. Lower 16 elements correspond to the
 ///    (converted) elements from \a __B; higher order elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set,
 ///    zero is taken instead.
@@ -2243,7 +2243,7 @@ _mm256_maskz_cvts2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 7
-/// 	dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	dst.fp16[i] := convert_hf8_to_fp16(__B.hf8[i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -2254,7 +2254,7 @@ _mm256_maskz_cvts2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVTHF82PH instruction.
 ///
 /// \param __A
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \returns
 ///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A.
@@ -2271,7 +2271,7 @@ static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvthf8(__m128i __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_fp8_to_fp16(__A.fp8[i])
+/// 		dst.fp16[i] := convert_hf8_to_fp16(__A.hf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := __W.fp16[i]
 /// 	FI
@@ -2289,7 +2289,7 @@ static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvthf8(__m128i __A) {
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \returns
 ///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
@@ -2308,7 +2308,7 @@ _mm_mask_cvthf8(__m128h __W, __mmask8 __U, __m128i __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 		dst.fp16[i] := convert_hf8_to_fp16(__B.hf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := 0
 /// 	FI
@@ -2324,7 +2324,7 @@ _mm_mask_cvthf8(__m128h __W, __mmask8 __U, __m128i __A) {
 /// \param __U
 ///    A 8-bit zeroing mask.
 /// \param __A
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \returns
 ///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
@@ -2340,7 +2340,7 @@ _mm_maskz_cvthf8(__mmask8 __U, __m128i __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 15
-/// 	dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	dst.fp16[i] := convert_hf8_to_fp16(__B.hf8[i])
 /// ENDFOR
 ///
 /// dst[MAX:256] := 0
@@ -2351,7 +2351,7 @@ _mm_maskz_cvthf8(__mmask8 __U, __m128i __A) {
 /// This intrinsic corresponds to the \c VCVTHF82PH instruction.
 ///
 /// \param __A
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x hf8].
 /// \returns
 ///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A.
@@ -2369,7 +2369,7 @@ _mm256_cvthf8(__m128i __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_fp8_to_fp16(__A.fp8[i])
+/// 		dst.fp16[i] := convert_hf8_to_fp16(__A.hf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := __W.fp16[i]
 /// 	FI
@@ -2387,7 +2387,7 @@ _mm256_cvthf8(__m128i __A) {
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x hf8].
 /// \returns
 ///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
@@ -2406,7 +2406,7 @@ _mm256_mask_cvthf8(__m256h __W, __mmask16 __U, __m128i __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 		dst.fp16[i] := convert_hf8_to_fp16(__B.hf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := 0
 /// 	FI
@@ -2422,7 +2422,7 @@ _mm256_mask_cvthf8(__m256h __W, __mmask16 __U, __m128i __A) {
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x hf8].
 /// \returns
 ///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
@@ -2439,7 +2439,7 @@ _mm256_maskz_cvthf8(__mmask16 __U, __m128i __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 7
-/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -2452,7 +2452,7 @@ _mm256_maskz_cvthf8(__mmask16 __U, __m128i __A) {
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the (converted)
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the (converted)
 ///    elements from \a __A; upper elements are zeroed. 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_bf8(__m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8_128_mask(
@@ -2467,9 +2467,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_bf8(__m128h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.bf8 := __W.bf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -2481,14 +2481,14 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_bf8(__m128h __A) {
 /// This intrinsic corresponds to the \c VCVTPH2BF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
 ///
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the
 ///    (converted) elements from \a __A; upper elements are zeroed. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -2505,9 +2505,9 @@ _mm_mask_cvtph_bf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := 0
+/// 		dst.bf8[i] := 0
 /// 	ELSE
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// 	FI
 /// ENDFOR
 ///
@@ -2524,7 +2524,7 @@ _mm_mask_cvtph_bf8(__m128i __W, __mmask8 __U, __m128h __A) {
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
 ///
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the
 ///    (converted) elements from \a __A; upper elements are zeroed. If
 ///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -2538,7 +2538,7 @@ _mm_maskz_cvtph_bf8(__mmask8 __U, __m128h __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 15
-/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -2551,7 +2551,7 @@ _mm_maskz_cvtph_bf8(__mmask8 __U, __m128h __A) {
 /// \param __A
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the (converted)
+///    A 128-bit vector of [16 x bf8]. Resulting elements correspond to the (converted)
 ///    elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_cvtph_bf8(__m256h __A) {
@@ -2566,9 +2566,9 @@ _mm256_cvtph_bf8(__m256h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.bf8 := __W.bf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -2580,13 +2580,13 @@ _mm256_cvtph_bf8(__m256h __A) {
 /// This intrinsic corresponds to the \c VCVTPH2BF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
 ///    A 256-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
@@ -2602,9 +2602,9 @@ _mm256_mask_cvtph_bf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.bf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -2620,7 +2620,7 @@ _mm256_mask_cvtph_bf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// \param __A
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set,
 ///    then element is zeroed instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
@@ -2635,7 +2635,7 @@ _mm256_maskz_cvtph_bf8(__mmask16 __U, __m256h __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 7
-/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -2648,7 +2648,7 @@ _mm256_maskz_cvtph_bf8(__mmask16 __U, __m256h __A) {
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the (converted)
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the (converted)
 ///    elements from \a __A; upper elements are zeroed. 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_bf8(__m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2bf8s_128_mask(
@@ -2663,9 +2663,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_bf8(__m128h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.bf8 := __W.bf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -2677,13 +2677,13 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_bf8(__m128h __A) {
 /// This intrinsic corresponds to the \c VCVTPH2BF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the
 ///    (converted) elements from \a __A; upper elements are zeroed. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -2700,9 +2700,9 @@ _mm_mask_cvtsph_bf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.bf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -2718,7 +2718,7 @@ _mm_mask_cvtsph_bf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the
 ///    (converted) elements from \a __A; upper elements are zeroed. If
 ///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -2732,7 +2732,7 @@ _mm_maskz_cvtsph_bf8(__mmask8 __U, __m128h __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 15
-/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -2745,7 +2745,7 @@ _mm_maskz_cvtsph_bf8(__mmask8 __U, __m128h __A) {
 /// \param __A
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the (converted)
+///    A 128-bit vector of [16 x bf8]. Resulting elements correspond to the (converted)
 ///    elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_cvtsph_bf8(__m256h __A) {
@@ -2761,9 +2761,9 @@ _mm256_cvtsph_bf8(__m256h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.bf8 := __W.bf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -2775,13 +2775,13 @@ _mm256_cvtsph_bf8(__m256h __A) {
 /// This intrinsic corresponds to the \c VCVTPH2BF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
 ///    A 256-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
@@ -2798,9 +2798,9 @@ _mm256_mask_cvtsph_bf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.bf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -2816,7 +2816,7 @@ _mm256_mask_cvtsph_bf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// \param __A
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    A 128-bit vector of [16 x bf8]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set,
 ///    then element is zeroed instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
@@ -2831,7 +2831,7 @@ _mm256_maskz_cvtsph_bf8(__mmask16 __U, __m256h __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 7
-/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -2844,7 +2844,7 @@ _mm256_maskz_cvtsph_bf8(__mmask16 __U, __m256h __A) {
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the (converted)
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the (converted)
 ///    elements from \a __A; upper elements are zeroed. 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_hf8(__m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8_128_mask(
@@ -2859,9 +2859,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_hf8(__m128h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.hf8 := __W.hf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -2873,14 +2873,14 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_hf8(__m128h __A) {
 /// This intrinsic corresponds to the \c VCVTPH2HF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
 ///
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the
 ///    (converted) elements from \a __A; upper elements are zeroed. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -2897,9 +2897,9 @@ _mm_mask_cvtph_hf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.hf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -2916,7 +2916,7 @@ _mm_mask_cvtph_hf8(__m128i __W, __mmask8 __U, __m128h __A) {
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
 ///
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the
 ///    (converted) elements from \a __A; upper elements are zeroed. If
 ///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -2930,7 +2930,7 @@ _mm_maskz_cvtph_hf8(__mmask8 __U, __m128h __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 15
-/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -2943,7 +2943,7 @@ _mm_maskz_cvtph_hf8(__mmask8 __U, __m128h __A) {
 /// \param __A
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the (converted)
+///    A 128-bit vector of [16 x hf8]. Resulting elements correspond to the (converted)
 ///    elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_cvtph_hf8(__m256h __A) {
@@ -2958,9 +2958,9 @@ _mm256_cvtph_hf8(__m256h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.hf8 := __W.hf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -2972,13 +2972,13 @@ _mm256_cvtph_hf8(__m256h __A) {
 /// This intrinsic corresponds to the \c VCVTPH2HF8 instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
 ///    A 256-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
@@ -2994,9 +2994,9 @@ _mm256_mask_cvtph_hf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.hf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -3012,7 +3012,7 @@ _mm256_mask_cvtph_hf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// \param __A
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set,
 ///    then element is zeroed instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
@@ -3027,7 +3027,7 @@ _mm256_maskz_cvtph_hf8(__mmask16 __U, __m256h __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 7
-/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -3040,7 +3040,7 @@ _mm256_maskz_cvtph_hf8(__mmask16 __U, __m256h __A) {
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the (converted)
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the (converted)
 ///    elements from \a __A; upper elements are zeroed. 
 static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_hf8(__m128h __A) {
   return (__m128i)__builtin_ia32_vcvtph2hf8s_128_mask(
@@ -3055,9 +3055,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_hf8(__m128h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.hf8 := __W.hf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -3069,13 +3069,13 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_hf8(__m128h __A) {
 /// This intrinsic corresponds to the \c VCVTPH2HF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the
 ///    (converted) elements from \a __A; upper elements are zeroed. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -3092,9 +3092,9 @@ _mm_mask_cvtsph_hf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.hf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -3110,7 +3110,7 @@ _mm_mask_cvtsph_hf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// \param __A
 ///    A 128-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Lower elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the
 ///    (converted) elements from \a __A; upper elements are zeroed. If
 ///    corresponding mask bit is not set, then element is zeroed.
 static __inline__ __m128i __DEFAULT_FN_ATTRS128
@@ -3124,7 +3124,7 @@ _mm_maskz_cvtsph_hf8(__mmask8 __U, __m128h __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 15
-/// 	dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 	dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -3137,7 +3137,7 @@ _mm_maskz_cvtsph_hf8(__mmask8 __U, __m128h __A) {
 /// \param __A
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the (converted)
+///    A 128-bit vector of [16 x hf8]. Resulting elements correspond to the (converted)
 ///    elements from \a __A.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
 _mm256_cvtsph_hf8(__m256h __A) {
@@ -3153,9 +3153,9 @@ _mm256_cvtsph_hf8(__m256h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8 := __W.fp8[i]
+/// 		dst.hf8 := __W.hf8[i]
 /// 	FI
 /// ENDFOR
 ///
@@ -3167,13 +3167,13 @@ _mm256_cvtsph_hf8(__m256h __A) {
 /// This intrinsic corresponds to the \c VCVTPH2HF8S instruction.
 ///
 /// \param __W
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x hf8].
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
 ///    A 256-bit vector of [8 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If
 ///    corresponding mask bit is not set, then element from \a __W is taken instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
@@ -3190,9 +3190,9 @@ _mm256_mask_cvtsph_hf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
-/// 		dst.fp8[i] := convert_fp16_to_fp8(__A.fp16[i])
+/// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// 	ELSE
-/// 		dst.fp8[i] := 0
+/// 		dst.hf8[i] := 0
 /// 	FI
 /// ENDFOR
 ///
@@ -3208,7 +3208,7 @@ _mm256_mask_cvtsph_hf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// \param __A
 ///    A 256-bit vector of [16 x fp16].
 /// \returns
-///    A 128-bit vector of [16 x fp8]. Resulting elements correspond to the
+///    A 128-bit vector of [16 x hf8]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set,
 ///    then element is zeroed instead.
 static __inline__ __m128i __DEFAULT_FN_ATTRS256
@@ -3222,7 +3222,7 @@ _mm256_maskz_cvtsph_hf8(__mmask16 __U, __m256h __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 7
-/// 	dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
 /// ENDFOR
 /// \endcode
 ///
@@ -3231,7 +3231,7 @@ _mm256_maskz_cvtsph_hf8(__mmask16 __U, __m256h __A) {
 /// This intrinsic does not correspond to a single instruction.
 ///
 /// \param __A
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \returns
 ///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A.
@@ -3247,7 +3247,7 @@ static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvtbf8_ph(__m128i __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := __W.fp16[i]
 /// 	FI
@@ -3263,7 +3263,7 @@ static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvtbf8_ph(__m128i __A) {
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \returns
 ///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
@@ -3282,7 +3282,7 @@ _mm_mask_cvtbf8_ph(__m128h __W, __mmask8 __U, __m128i __A) {
 /// \code{.operation]
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := 0
 /// 	FI
@@ -3296,7 +3296,7 @@ _mm_mask_cvtbf8_ph(__m128h __W, __mmask8 __U, __m128i __A) {
 /// \param __U
 ///    A 8-bit zeroing mask.
 /// \param __A
-///    A 128-bit vector of [16 x fp8].
+///    A 128-bit vector of [16 x bf8].
 /// \returns
 ///    A 128-bit vector of [8 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
@@ -3311,7 +3311,7 @@ _mm_maskz_cvtbf8_ph(__mmask8 __U, __m128i __A) {
 ///
 /// \code{.operation]
 /// FOR i := 0 to 15
-/// 	dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 	dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
 /// ENDFOR
 /// \endcode
 ///
@@ -3320,7 +3320,7 @@ _mm_maskz_cvtbf8_ph(__mmask8 __U, __m128i __A) {
 /// This intrinsic does not correspond to a single instruction.
 ///
 /// \param __A
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x bf8].
 /// \returns
 ///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A.
@@ -3336,7 +3336,7 @@ static __inline__ __m256h __DEFAULT_FN_ATTRS256 _mm256_cvtbf8_ph(__m128i __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := __W.fp16[i]
 /// 	FI
@@ -3352,7 +3352,7 @@ static __inline__ __m256h __DEFAULT_FN_ATTRS256 _mm256_cvtbf8_ph(__m128i __A) {
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x bf8].
 /// \returns
 ///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then
@@ -3371,7 +3371,7 @@ _mm256_mask_cvtbf8_ph(__m256h __W, __mmask16 __U, __m128i __A) {
 /// \code{.operation]
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_fp8_to_fp16(__B.fp8[i])
+/// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := 0
 /// 	FI
@@ -3385,7 +3385,7 @@ _mm256_mask_cvtbf8_ph(__m256h __W, __mmask16 __U, __m128i __A) {
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
-///    A 256-bit vector of [32 x fp8].
+///    A 256-bit vector of [32 x bf8].
 /// \returns
 ///    A 256-bit vector of [16 x fp16]. Resulting elements correspond to the
 ///    (converted) elements from \a __A. If corresponding mask bit is not set, then

>From 0560ea527fcc22a218e2e9dc3bde5e3e428c4f63 Mon Sep 17 00:00:00 2001
From: "Pirog, Mikolaj Maciej" <mikolaj.maciej.pirog at intel.com>
Date: Wed, 29 Jan 2025 14:36:36 +0100
Subject: [PATCH 3/7] 2i -> 2 * i

---
 clang/lib/Headers/avx10_2convertintrin.h | 44 ++++++++++++------------
 1 file changed, 22 insertions(+), 22 deletions(-)

diff --git a/clang/lib/Headers/avx10_2convertintrin.h b/clang/lib/Headers/avx10_2convertintrin.h
index 52cd08d6d2d59d..63cf2f4f01683b 100644
--- a/clang/lib/Headers/avx10_2convertintrin.h
+++ b/clang/lib/Headers/avx10_2convertintrin.h
@@ -408,7 +408,7 @@ _mm256_maskz_cvtx2ps_ph(__mmask16 __U, __m256 __A, __m256 __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -440,7 +440,7 @@ _mm_cvtbiasph_bf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// 	ELSE
 /// 		dst.bf8[i] := _W[i]
 /// 	FI
@@ -480,7 +480,7 @@ _mm_mask_cvtbiasph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 ///	 ELSE
 ///	 	dst.bf8[i] := 0
 ///	 FI
@@ -516,7 +516,7 @@ _mm_maskz_cvtbiasph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -549,7 +549,7 @@ _mm256_cvtbiasph_bf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// 	ELSE
 /// 		dst.bf8[i] := _W[i]
 /// 	FI
@@ -589,7 +589,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_bf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 ///	 ELSE
 ///	 	dst.bf8[i] := 0
 ///	 FI
@@ -626,7 +626,7 @@ _mm256_maskz_cvtbiasph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -658,7 +658,7 @@ _mm_cvtbiassph_bf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// 	ELSE
 /// 		dst.bf8[i] := _W[i]
 /// 	FI
@@ -698,7 +698,7 @@ _mm_mask_cvtbiassph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 ///	 ELSE
 ///	 	dst.bf8[i] := 0
 ///	 FI
@@ -735,7 +735,7 @@ _mm_maskz_cvtbiassph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -768,7 +768,7 @@ _mm256_cvtbiassph_bf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// 	ELSE
 /// 		dst.bf8[i] := _W[i]
 /// 	FI
@@ -808,7 +808,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_bf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
 ///	 ELSE
 ///	 	dst.bf8[i] := 0
 ///	 FI
@@ -844,7 +844,7 @@ _mm256_maskz_cvtbiassph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -876,7 +876,7 @@ _mm_cvtbiasph_hf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// 	ELSE
 /// 		dst.hf8[i] := _W[i]
 /// 	FI
@@ -916,7 +916,7 @@ _mm_mask_cvtbiasph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 ///	 ELSE
 ///	 	dst.hf8[i] := 0
 ///	 FI
@@ -985,7 +985,7 @@ _mm256_cvtbiasph_hf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// 	ELSE
 /// 		dst.hf8[i] := _W[i]
 /// 	FI
@@ -1062,7 +1062,7 @@ _mm256_maskz_cvtbiasph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -1094,7 +1094,7 @@ _mm_cvtbiassph_hf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// 	ELSE
 /// 		dst.hf8[i] := _W[i]
 /// 	FI
@@ -1134,7 +1134,7 @@ _mm_mask_cvtbiassph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 ///	 ELSE
 ///	 	dst.hf8[i] := 0
 ///	 FI
@@ -1171,7 +1171,7 @@ _mm_maskz_cvtbiassph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -1204,7 +1204,7 @@ _mm256_cvtbiassph_hf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 /// 	ELSE
 /// 		dst.hf8[i] := _W[i]
 /// 	FI
@@ -1244,7 +1244,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_hf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2i])
+///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
 ///	 ELSE
 ///	 	dst.hf8[i] := 0
 ///	 FI

>From 8ebd5cc5ab0ff37a2843fff1801db20c70275c98 Mon Sep 17 00:00:00 2001
From: "Pirog, Mikolaj Maciej" <mikolaj.maciej.pirog at intel.com>
Date: Mon, 3 Feb 2025 18:12:29 +0100
Subject: [PATCH 4/7] Fix wrong instr order

---
 clang/lib/Headers/avx10_2convertintrin.h | 136 +++++++++++------------
 1 file changed, 68 insertions(+), 68 deletions(-)

diff --git a/clang/lib/Headers/avx10_2convertintrin.h b/clang/lib/Headers/avx10_2convertintrin.h
index 63cf2f4f01683b..d2a16b416bb2af 100644
--- a/clang/lib/Headers/avx10_2convertintrin.h
+++ b/clang/lib/Headers/avx10_2convertintrin.h
@@ -408,7 +408,7 @@ _mm256_maskz_cvtx2ps_ph(__mmask16 __U, __m256 __A, __m256 __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -419,9 +419,9 @@ _mm256_maskz_cvtx2ps_ph(__mmask16 __U, __m256 __A, __m256 __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
 ///
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the
 ///    sum of elements from \a __A and \a __B; higher order elements are zeroed.
@@ -440,7 +440,7 @@ _mm_cvtbiasph_bf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// 	ELSE
 /// 		dst.bf8[i] := _W[i]
 /// 	FI
@@ -458,9 +458,9 @@ _mm_cvtbiasph_bf8(__m128i __A, __m128h __B) {
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
@@ -480,7 +480,7 @@ _mm_mask_cvtbiasph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 ///	 ELSE
 ///	 	dst.bf8[i] := 0
 ///	 FI
@@ -496,9 +496,9 @@ _mm_mask_cvtbiasph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \param __U
 ///    A 8-bit zeroing mask.
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
@@ -516,7 +516,7 @@ _mm_maskz_cvtbiasph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -527,9 +527,9 @@ _mm_maskz_cvtbiasph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
 ///
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Elements correspond to the
 ///    sum of elements from \a __A and \a __B.
@@ -549,7 +549,7 @@ _mm256_cvtbiasph_bf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// 	ELSE
 /// 		dst.bf8[i] := _W[i]
 /// 	FI
@@ -567,9 +567,9 @@ _mm256_cvtbiasph_bf8(__m256i __A, __m256h __B) {
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
@@ -589,7 +589,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_bf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 ///	 ELSE
 ///	 	dst.bf8[i] := 0
 ///	 FI
@@ -605,9 +605,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_bf8(
 /// \param __U
 ///    A 16-bit zeroing mask.
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set,
@@ -626,7 +626,7 @@ _mm256_maskz_cvtbiasph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -637,9 +637,9 @@ _mm256_maskz_cvtbiasph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2BF8 instruction.
 ///
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the
 ///    sum of elements from \a __A and \a __B; higher order elements are zeroed.
@@ -658,7 +658,7 @@ _mm_cvtbiassph_bf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// 	ELSE
 /// 		dst.bf8[i] := _W[i]
 /// 	FI
@@ -676,9 +676,9 @@ _mm_cvtbiassph_bf8(__m128i __A, __m128h __B) {
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
@@ -698,7 +698,7 @@ _mm_mask_cvtbiassph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 ///	 ELSE
 ///	 	dst.bf8[i] := 0
 ///	 FI
@@ -714,9 +714,9 @@ _mm_mask_cvtbiassph_bf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \param __U
 ///    A 8-bit zeroing mask.
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
@@ -735,7 +735,7 @@ _mm_maskz_cvtbiassph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -746,9 +746,9 @@ _mm_maskz_cvtbiassph_bf8(__mmask8 __U, __m128i __A, __m128h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2BF8S instruction.
 ///
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Elements correspond to the
 ///    sum of elements from \a __A and \a __B.
@@ -768,7 +768,7 @@ _mm256_cvtbiassph_bf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 		dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// 	ELSE
 /// 		dst.bf8[i] := _W[i]
 /// 	FI
@@ -786,9 +786,9 @@ _mm256_cvtbiassph_bf8(__m256i __A, __m256h __B) {
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
@@ -808,7 +808,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_bf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.fp16[i], __B.int8[2 * i])
+///	 	dst.bf8[i] := add_convert_fp16_to_bf8_bias(__A.int8[2 * i], __B.fp16[i])
 ///	 ELSE
 ///	 	dst.bf8[i] := 0
 ///	 FI
@@ -824,9 +824,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_bf8(
 /// \param __U
 ///    A 16-bit zeroing mask.
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x bf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set,
@@ -844,7 +844,7 @@ _mm256_maskz_cvtbiassph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -855,9 +855,9 @@ _mm256_maskz_cvtbiassph_bf8(__mmask16 __U, __m256i __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2HF8 instruction.
 ///
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the
 ///    sum of elements from \a __A and \a __B; higher order elements are zeroed.
@@ -876,7 +876,7 @@ _mm_cvtbiasph_hf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// 	ELSE
 /// 		dst.hf8[i] := _W[i]
 /// 	FI
@@ -894,9 +894,9 @@ _mm_cvtbiasph_hf8(__m128i __A, __m128h __B) {
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
@@ -916,7 +916,7 @@ _mm_mask_cvtbiasph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 ///	 ELSE
 ///	 	dst.hf8[i] := 0
 ///	 FI
@@ -932,9 +932,9 @@ _mm_mask_cvtbiasph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \param __U
 ///    A 8-bit zeroing mask.
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
@@ -952,7 +952,7 @@ _mm_maskz_cvtbiasph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.hf8[i] := add_fp16_int8_convert_to_hf8(__A.fp16[i], __B.int8[2 * i])
+/// 	dst.hf8[i] := add_fp16_int8_convert_to_hf8(__A.int8[2 * i], __B.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -985,7 +985,7 @@ _mm256_cvtbiasph_hf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// 	ELSE
 /// 		dst.hf8[i] := _W[i]
 /// 	FI
@@ -1003,9 +1003,9 @@ _mm256_cvtbiasph_hf8(__m256i __A, __m256h __B) {
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
@@ -1025,7 +1025,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiasph_hf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.hf8[i] := add_fp16_int8_convert_to_hf8(__A.fp16[i], __B.int8[2 * i])
+///	 	dst.hf8[i] := add_fp16_int8_convert_to_hf8(__A.int8[2 * i], __B.fp16[i])
 ///	 ELSE
 ///	 	dst.hf8[i] := 0
 ///	 FI
@@ -1062,7 +1062,7 @@ _mm256_maskz_cvtbiasph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:64] := 0
@@ -1073,9 +1073,9 @@ _mm256_maskz_cvtbiasph_hf8(__mmask16 __U, __m256i __A, __m256h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2HF8S`instruction.
 ///
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the
 ///    sum of elements from \a __A and \a __B; higher order elements are zeroed.
@@ -1094,7 +1094,7 @@ _mm_cvtbiassph_hf8(__m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// 	ELSE
 /// 		dst.hf8[i] := _W[i]
 /// 	FI
@@ -1112,9 +1112,9 @@ _mm_cvtbiassph_hf8(__m128i __A, __m128h __B) {
 /// \param __U
 ///    A 8-bit merging mask.
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
@@ -1134,7 +1134,7 @@ _mm_mask_cvtbiassph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 ///	 ELSE
 ///	 	dst.hf8[i] := 0
 ///	 FI
@@ -1150,9 +1150,9 @@ _mm_mask_cvtbiassph_hf8(__m128i __W, __mmask8 __U, __m128i __A, __m128h __B) {
 /// \param __U
 ///    A 8-bit zeroing mask.
 /// \param __A
-///    A 128-bit vector of [8 x fp16].
-/// \param __B
 ///    A 128-bit vector of [8 x int16].
+/// \param __B
+///    A 128-bit vector of [8  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Lower elements correspond to the sum of
 ///    elements from \a __A and \a __B; higher order elements are zeroed. If
@@ -1171,7 +1171,7 @@ _mm_maskz_cvtbiassph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// ENDFOR
 ///
 /// dst[MAX:128] := 0
@@ -1182,9 +1182,9 @@ _mm_maskz_cvtbiassph_hf8(__mmask8 __U, __m128i __A, __m128h __B) {
 /// This intrinsic corresponds to the \c VCVTBIASPH2HF8S instruction.
 ///
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Elements correspond to the
 ///    sum of elements from \a __A and \a __B.
@@ -1204,7 +1204,7 @@ _mm256_cvtbiassph_hf8(__m256i __A, __m256h __B) {
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+/// 		dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 /// 	ELSE
 /// 		dst.hf8[i] := _W[i]
 /// 	FI
@@ -1222,9 +1222,9 @@ _mm256_cvtbiassph_hf8(__m256i __A, __m256h __B) {
 /// \param __U
 ///    A 16-bit merging mask.
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set, then
@@ -1244,7 +1244,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_hf8(
 /// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
-///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.fp16[i], __B.int8[2 * i])
+///	 	dst.hf8[i] := add_convert_fp16_to_hf8_bias(__A.int8[2 * i], __B.fp16[i])
 ///	 ELSE
 ///	 	dst.hf8[i] := 0
 ///	 FI
@@ -1260,9 +1260,9 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS256 _mm256_mask_cvtbiassph_hf8(
 /// \param __U
 ///    A 16-bit zeroing mask.
 /// \param __A
-///    A 256-bit vector of [16 x fp16].
-/// \param __B
 ///    A 256-bit vector of [16 x int16].
+/// \param __B
+///    A 256-bit vector of [16  x fp16].
 /// \returns
 ///    A 128-bit vector of [16 x hf8]. Elements correspond to the sum of
 ///    elements from \a __A and \a __B. If corresponding mask bit is not set,

>From 758dd4c07c6adea9f6e87cf5a2f458ed20bd2eec Mon Sep 17 00:00:00 2001
From: "Pirog, Mikolaj Maciej" <mikolaj.maciej.pirog at intel.com>
Date: Tue, 4 Feb 2025 05:15:43 -0800
Subject: [PATCH 5/7] Fix wrong brackets

---
 clang/lib/Headers/avx10_2convertintrin.h | 112 +++++++++++------------
 1 file changed, 56 insertions(+), 56 deletions(-)

diff --git a/clang/lib/Headers/avx10_2convertintrin.h b/clang/lib/Headers/avx10_2convertintrin.h
index d2a16b416bb2af..587aa4b2737b3f 100644
--- a/clang/lib/Headers/avx10_2convertintrin.h
+++ b/clang/lib/Headers/avx10_2convertintrin.h
@@ -1312,7 +1312,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_bf8(__m128h __A,
 /// Merging mask \a __U is used to determine if given element should be taken
 /// from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
 /// 		IF i < 8
@@ -1397,7 +1397,7 @@ _mm_maskz_cvt2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
 /// floating-point elements to a 256-bit vector containing E5M2 FP8 elements.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF i < 16 
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
@@ -1432,7 +1432,7 @@ _mm256_cvt2ph_bf8(__m256h __A, __m256h __B) {
 /// Merging mask \a __U is used to determine if given element should be taken
 /// from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		IF i < 16 
@@ -1476,7 +1476,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_bf8(
 /// Merging mask \a __U is used to determine if given element should be zeroed
 /// instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		dst.bf8[i] := 0
@@ -1597,7 +1597,7 @@ _mm_mask_cvts2ph_bf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// Zeroing mask \a __U is used to determine if given element should be zeroed
 /// instead. Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
@@ -1639,7 +1639,7 @@ _mm_maskz_cvts2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// floating-point elements to a 256-bit vector containing E5M2 FP8 elements.
 /// Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF i < 16 
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
@@ -1674,7 +1674,7 @@ _mm256_cvts2ph_bf8(__m256h __A, __m256h __B) {
 /// Merging mask \a __U is used to determine if given element should be taken
 /// from \a __W instead. Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		IF i < 16 
@@ -1718,7 +1718,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_bf8(
 /// Merging mask \a __U is used to determine if given element should be zeroed
 /// instead. Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		dst.bf8[i] := 0
@@ -1759,7 +1759,7 @@ _mm256_maskz_cvts2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// Convert two 128-bit vectors, \a __A and \a __B, containing packed FP16
 /// floating-point elements to a 128-bit vector containing E4M3 FP8 elements.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 16 
 /// 	IF i < 8
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -1794,7 +1794,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_hf8(__m128h __A,
 /// Merging mask \a __U is used to determine if given element should be taken
 /// from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
@@ -1838,7 +1838,7 @@ _mm_mask_cvt2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// Zeroing mask \a __U is used to determine if given element should be zeroed
 /// instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
@@ -1879,7 +1879,7 @@ _mm_maskz_cvt2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// Convert two 256-bit vectors, \a __A and \a __B, containing packed FP16
 /// floating-point elements to a 256-bit vector containing E4M3 FP8 elements.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF i < 16 
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -1914,7 +1914,7 @@ _mm256_cvt2ph_hf8(__m256h __A, __m256h __B) {
 /// Merging mask \a __U is used to determine if given element should be taken
 /// from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		IF i < 16 
@@ -1958,7 +1958,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_hf8(
 /// Merging mask \a __U is used to determine if given element should be zeroed
 /// instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		dst.hf8[i] := 0
@@ -2000,7 +2000,7 @@ _mm256_maskz_cvt2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// floating-point elements to a 128-bit vector containing E4M3 FP8 elements.
 /// Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 16 
 /// 	IF i < 8
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -2035,7 +2035,7 @@ _mm_cvts2ph_hf8(__m128h __A, __m128h __B) {
 /// Merging mask \a __U is used to determine if given element should be taken
 /// from \a __W instead. Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
@@ -2079,7 +2079,7 @@ _mm_mask_cvts2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// Zeroing mask \a __U is used to determine if given element should be zeroed
 /// instead. Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 16 
 /// 	IF __U[i]
 /// 		IF i < 8
@@ -2121,7 +2121,7 @@ _mm_maskz_cvts2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// floating-point elements to a 256-bit vector containing E4M3 FP8 elements.
 /// Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF i < 16 
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -2156,7 +2156,7 @@ _mm256_cvts2ph_hf8(__m256h __A, __m256h __B) {
 /// Merging mask \a __U is used to determine if given element should be taken
 /// from \a __W instead. Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		IF i < 16 
@@ -2200,7 +2200,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_hf8(
 /// Merging mask \a __U is used to determine if given element should be zeroed
 /// instead. Resulting elements are saturated in case of overflow.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 32 
 /// 	IF __U[i]
 /// 		dst.hf8[i] := 0
@@ -2241,7 +2241,7 @@ _mm256_maskz_cvts2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// Convert 128-bit vector \a __A, containing packed FP8 E4M3 floating-point
 /// elements to a 128-bit vector containing FP16 elements. The conversion is exact.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	dst.fp16[i] := convert_hf8_to_fp16(__B.hf8[i])
 /// ENDFOR
@@ -2268,7 +2268,7 @@ static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvthf8(__m128i __A) {
 /// exact. Merging mask \a __U is used to determine if given element should be
 /// taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.fp16[i] := convert_hf8_to_fp16(__A.hf8[i])
@@ -2305,7 +2305,7 @@ _mm_mask_cvthf8(__m128h __W, __mmask8 __U, __m128i __A) {
 /// exact. Zeroing mask \a __U is used to determine if given element should be
 /// zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.fp16[i] := convert_hf8_to_fp16(__B.hf8[i])
@@ -2338,7 +2338,7 @@ _mm_maskz_cvthf8(__mmask8 __U, __m128i __A) {
 /// Convert 256-bit vector \a __A, containing packed FP8 E4M3 floating-point
 /// elements to a 256-bit vector containing FP16 elements. The conversion is exact.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	dst.fp16[i] := convert_hf8_to_fp16(__B.hf8[i])
 /// ENDFOR
@@ -2366,7 +2366,7 @@ _mm256_cvthf8(__m128i __A) {
 /// exact. Merging mask \a __U is used to determine if given element should be
 /// taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		dst.fp16[i] := convert_hf8_to_fp16(__A.hf8[i])
@@ -2403,7 +2403,7 @@ _mm256_mask_cvthf8(__m256h __W, __mmask16 __U, __m128i __A) {
 /// exact. Zeroing mask \a __U is used to determine if given element should be
 /// zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		dst.fp16[i] := convert_hf8_to_fp16(__B.hf8[i])
@@ -2437,7 +2437,7 @@ _mm256_maskz_cvthf8(__mmask16 __U, __m128i __A) {
 /// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
 /// resulting vector are zeroed.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// ENDFOR
@@ -2464,7 +2464,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_bf8(__m128h __A) {
 /// resulting vector are zeroed. Merging mask \a __U is used to determine if
 /// given element should be taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
@@ -2502,7 +2502,7 @@ _mm_mask_cvtph_bf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// resulting vector are zeroed. Zeroing mask \a __U is used to determine if
 /// given element should be zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.bf8[i] := 0
@@ -2536,7 +2536,7 @@ _mm_maskz_cvtph_bf8(__mmask8 __U, __m128h __A) {
 /// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
 /// to a 128-bit vector containing E5M2 FP8 elements.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// ENDFOR
@@ -2563,7 +2563,7 @@ _mm256_cvtph_bf8(__m256h __A) {
 /// to a 128-bit vector containing E5M2 FP8 elements. Merging mask \a __U is
 /// used to determine if given element should be taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
@@ -2599,7 +2599,7 @@ _mm256_mask_cvtph_bf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// to a 128-bit vector containing E5M2 FP8 elements. Zeroing mask \a __U is
 /// used to determine if given element should be zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
@@ -2633,7 +2633,7 @@ _mm256_maskz_cvtph_bf8(__mmask16 __U, __m256h __A) {
 /// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
 /// resulting vector are zeroed. Results are saturated.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// ENDFOR
@@ -2660,7 +2660,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_bf8(__m128h __A) {
 /// resulting vector are zeroed. Results are saturated. Merging mask \a __U is
 /// used to determine if given element should be taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
@@ -2697,7 +2697,7 @@ _mm_mask_cvtsph_bf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// resulting vector are zeroed. Results are saturated. Zeroing mask \a __U is
 /// used to determine if given element should be zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
@@ -2730,7 +2730,7 @@ _mm_maskz_cvtsph_bf8(__mmask8 __U, __m128h __A) {
 /// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
 /// to a 128-bit vector containing E5M2 FP8 elements. Results are saturated.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
 /// ENDFOR
@@ -2758,7 +2758,7 @@ _mm256_cvtsph_bf8(__m256h __A) {
 /// Merging mask \a __U is used to determine if given element should be taken
 /// from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
@@ -2795,7 +2795,7 @@ _mm256_mask_cvtsph_bf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// Zeroing mask \a __U is used to determine if given element should be zeroed
 /// instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__A.fp16[i])
@@ -2829,7 +2829,7 @@ _mm256_maskz_cvtsph_bf8(__mmask16 __U, __m256h __A) {
 /// to a 128-bit vector containing E5M2 FP8 elements. Upper elements of
 /// resulting vector are zeroed.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// ENDFOR
@@ -2856,7 +2856,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtph_hf8(__m128h __A) {
 /// resulting vector are zeroed. Merging mask \a __U is used to determine if
 /// given element should be taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
@@ -2894,7 +2894,7 @@ _mm_mask_cvtph_hf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// resulting vector are zeroed. Zeroing mask \a __U is used to determine if
 /// given element should be zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
@@ -2928,7 +2928,7 @@ _mm_maskz_cvtph_hf8(__mmask8 __U, __m128h __A) {
 /// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
 /// to a 128-bit vector containing E4M3 FP8 elements.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// ENDFOR
@@ -2955,7 +2955,7 @@ _mm256_cvtph_hf8(__m256h __A) {
 /// to a 128-bit vector containing E4M3 FP8 elements. Merging mask \a __U is
 /// used to determine if given element should be taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
@@ -2991,7 +2991,7 @@ _mm256_mask_cvtph_hf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// to a 128-bit vector containing E4M3 FP8 elements. Zeroing mask \a __U is
 /// used to determine if given element should be zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
@@ -3025,7 +3025,7 @@ _mm256_maskz_cvtph_hf8(__mmask16 __U, __m256h __A) {
 /// to a 128-bit vector containing E4M3 FP8 elements. Upper elements of
 /// resulting vector are zeroed. Results are saturated.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// ENDFOR
@@ -3052,7 +3052,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvtsph_hf8(__m128h __A) {
 /// resulting vector are zeroed. Results are saturated. Merging mask \a __U is
 /// used to determine if given element should be taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
@@ -3089,7 +3089,7 @@ _mm_mask_cvtsph_hf8(__m128i __W, __mmask8 __U, __m128h __A) {
 /// resulting vector are zeroed. Results are saturated. Zeroing mask \a __U is
 /// used to determine if given element should be zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
@@ -3122,7 +3122,7 @@ _mm_maskz_cvtsph_hf8(__mmask8 __U, __m128h __A) {
 /// Convert 256-bit vector \a __A containing packed FP16 floating-point elements
 /// to a 128-bit vector containing E4M3 FP8 elements. Results are saturated.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
 /// ENDFOR
@@ -3150,7 +3150,7 @@ _mm256_cvtsph_hf8(__m256h __A) {
 /// Merging mask \a __U is used to determine if given element should be taken
 /// from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	IF __U[i]
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
@@ -3187,7 +3187,7 @@ _mm256_mask_cvtsph_hf8(__m128i __W, __mmask16 __U, __m256h __A) {
 /// Zeroing mask \a __U is used to determine if given element should be zeroed
 /// instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__A.fp16[i])
@@ -3220,7 +3220,7 @@ _mm256_maskz_cvtsph_hf8(__mmask16 __U, __m256h __A) {
 /// Convert 128-bit vector \a __A, containing packed FP8 E5M2 floating-point
 /// elements to a 128-bit vector containing FP16 elements. The conversion is exact.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
 /// ENDFOR
@@ -3244,7 +3244,7 @@ static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvtbf8_ph(__m128i __A) {
 /// exact. Merging mask \a __U is used to determine if given element should be
 /// taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
@@ -3279,7 +3279,7 @@ _mm_mask_cvtbf8_ph(__m128h __W, __mmask8 __U, __m128i __A) {
 /// exact. Zeroing mask \a __U is used to determine if given element should be
 /// zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
 /// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
@@ -3309,7 +3309,7 @@ _mm_maskz_cvtbf8_ph(__mmask8 __U, __m128i __A) {
 /// Convert 256-bit vector \a __A, containing packed FP8 E4M3 floating-point
 /// elements to a 256-bit vector containing FP16 elements. The conversion is exact.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15
 /// 	dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
 /// ENDFOR
@@ -3333,7 +3333,7 @@ static __inline__ __m256h __DEFAULT_FN_ATTRS256 _mm256_cvtbf8_ph(__m128i __A) {
 /// exact. Merging mask \a __U is used to determine if given element should be
 /// taken from \a __W instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
@@ -3368,7 +3368,7 @@ _mm256_mask_cvtbf8_ph(__m256h __W, __mmask16 __U, __m128i __A) {
 /// exact. Zeroing mask \a __U is used to determine if given element should be
 /// zeroed instead.
 ///
-/// \code{.operation]
+/// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])

>From 74c4a2d52605f1dfe1ea50692aa281015f5c2408 Mon Sep 17 00:00:00 2001
From: "Pirog, Mikolaj Maciej" <mikolaj.maciej.pirog at intel.com>
Date: Wed, 5 Feb 2025 04:49:14 -0800
Subject: [PATCH 6/7] Fix wrong loop bounds

---
 clang/lib/Headers/avx10_2convertintrin.h | 38 ++++++++++++------------
 1 file changed, 19 insertions(+), 19 deletions(-)

diff --git a/clang/lib/Headers/avx10_2convertintrin.h b/clang/lib/Headers/avx10_2convertintrin.h
index 587aa4b2737b3f..a9261908241e8a 100644
--- a/clang/lib/Headers/avx10_2convertintrin.h
+++ b/clang/lib/Headers/avx10_2convertintrin.h
@@ -1398,7 +1398,7 @@ _mm_maskz_cvt2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// floating-point elements to a 256-bit vector containing E5M2 FP8 elements.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF i < 16 
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 	ELSE
@@ -1433,7 +1433,7 @@ _mm256_cvt2ph_bf8(__m256h __A, __m256h __B) {
 /// from \a __W instead.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF __U[i]
 /// 		IF i < 16 
 /// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
@@ -1477,7 +1477,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_bf8(
 /// instead.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF __U[i]
 /// 		dst.bf8[i] := 0
 /// 	ELSE
@@ -1598,7 +1598,7 @@ _mm_mask_cvts2ph_bf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// instead. Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 16 
+/// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		IF i < 8
 /// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
@@ -1640,7 +1640,7 @@ _mm_maskz_cvts2ph_bf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF i < 16 
 /// 		dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
 /// 	ELSE
@@ -1675,7 +1675,7 @@ _mm256_cvts2ph_bf8(__m256h __A, __m256h __B) {
 /// from \a __W instead. Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF __U[i]
 /// 		IF i < 16 
 /// 			dst.bf8[i] := convert_fp16_to_bf8(__B.fp16[i])
@@ -1719,7 +1719,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_bf8(
 /// instead. Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF __U[i]
 /// 		dst.bf8[i] := 0
 /// 	ELSE
@@ -1760,7 +1760,7 @@ _mm256_maskz_cvts2ph_bf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// floating-point elements to a 128-bit vector containing E4M3 FP8 elements.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 16 
+/// FOR i := 0 to 15 
 /// 	IF i < 8
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 	ELSE
@@ -1795,7 +1795,7 @@ static __inline__ __m128i __DEFAULT_FN_ATTRS128 _mm_cvt2ph_hf8(__m128h __A,
 /// from \a __W instead.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 16 
+/// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		IF i < 8
 /// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -1839,7 +1839,7 @@ _mm_mask_cvt2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// instead.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 16 
+/// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		IF i < 8
 /// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -1880,7 +1880,7 @@ _mm_maskz_cvt2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// floating-point elements to a 256-bit vector containing E4M3 FP8 elements.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF i < 16 
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 	ELSE
@@ -1915,7 +1915,7 @@ _mm256_cvt2ph_hf8(__m256h __A, __m256h __B) {
 /// from \a __W instead.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF __U[i]
 /// 		IF i < 16 
 /// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -1959,7 +1959,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvt2ph_hf8(
 /// instead.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF __U[i]
 /// 		dst.hf8[i] := 0
 /// 	ELSE
@@ -2001,7 +2001,7 @@ _mm256_maskz_cvt2ph_hf8(__mmask32 __U, __m256h __A, __m256h __B) {
 /// Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 16 
+/// FOR i := 0 to 15 
 /// 	IF i < 8
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 	ELSE
@@ -2036,7 +2036,7 @@ _mm_cvts2ph_hf8(__m128h __A, __m128h __B) {
 /// from \a __W instead. Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 16 
+/// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		IF i < 8
 /// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -2080,7 +2080,7 @@ _mm_mask_cvts2ph_hf8(__m128i __W, __mmask16 __U, __m128h __A, __m128h __B) {
 /// instead. Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 16 
+/// FOR i := 0 to 15 
 /// 	IF __U[i]
 /// 		IF i < 8
 /// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -2122,7 +2122,7 @@ _mm_maskz_cvts2ph_hf8(__mmask16 __U, __m128h __A, __m128h __B) {
 /// Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF i < 16 
 /// 		dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
 /// 	ELSE
@@ -2157,7 +2157,7 @@ _mm256_cvts2ph_hf8(__m256h __A, __m256h __B) {
 /// from \a __W instead. Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF __U[i]
 /// 		IF i < 16 
 /// 			dst.hf8[i] := convert_fp16_to_hf8(__B.fp16[i])
@@ -2201,7 +2201,7 @@ static __inline__ __m256i __DEFAULT_FN_ATTRS256 _mm256_mask_cvts2ph_hf8(
 /// instead. Resulting elements are saturated in case of overflow.
 ///
 /// \code{.operation}
-/// FOR i := 0 to 32 
+/// FOR i := 0 to 31 
 /// 	IF __U[i]
 /// 		dst.hf8[i] := 0
 /// 	ELSE

>From f42a5885ccc7a53acb0e466454ddbc4482c1e40d Mon Sep 17 00:00:00 2001
From: "Pirog, Mikolaj Maciej" <mikolaj.maciej.pirog at intel.com>
Date: Wed, 5 Feb 2025 05:15:34 -0800
Subject: [PATCH 7/7] Rename wrong var

---
 clang/lib/Headers/avx10_2convertintrin.h | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)

diff --git a/clang/lib/Headers/avx10_2convertintrin.h b/clang/lib/Headers/avx10_2convertintrin.h
index a9261908241e8a..cc2cd1b3e2b629 100644
--- a/clang/lib/Headers/avx10_2convertintrin.h
+++ b/clang/lib/Headers/avx10_2convertintrin.h
@@ -3222,7 +3222,7 @@ _mm256_maskz_cvtsph_hf8(__mmask16 __U, __m256h __A) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 7
-/// 	dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
+/// 	dst.fp16[i] := convert_bf8_to_fp16(__A.bf8[i])
 /// ENDFOR
 /// \endcode
 ///
@@ -3247,7 +3247,7 @@ static __inline__ __m128h __DEFAULT_FN_ATTRS128 _mm_cvtbf8_ph(__m128i __A) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
+/// 		dst.fp16[i] := convert_bf8_to_fp16(__A.bf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := __W.fp16[i]
 /// 	FI
@@ -3282,7 +3282,7 @@ _mm_mask_cvtbf8_ph(__m128h __W, __mmask8 __U, __m128i __A) {
 /// \code{.operation}
 /// FOR i := 0 to 7
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
+/// 		dst.fp16[i] := convert_bf8_to_fp16(__A.bf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := 0
 /// 	FI
@@ -3311,7 +3311,7 @@ _mm_maskz_cvtbf8_ph(__mmask8 __U, __m128i __A) {
 ///
 /// \code{.operation}
 /// FOR i := 0 to 15
-/// 	dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
+/// 	dst.fp16[i] := convert_bf8_to_fp16(__A.bf8[i])
 /// ENDFOR
 /// \endcode
 ///
@@ -3336,7 +3336,7 @@ static __inline__ __m256h __DEFAULT_FN_ATTRS256 _mm256_cvtbf8_ph(__m128i __A) {
 /// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
+/// 		dst.fp16[i] := convert_bf8_to_fp16(__A.bf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := __W.fp16[i]
 /// 	FI
@@ -3371,7 +3371,7 @@ _mm256_mask_cvtbf8_ph(__m256h __W, __mmask16 __U, __m128i __A) {
 /// \code{.operation}
 /// FOR i := 0 to 15 
 /// 	IF __U[i]
-/// 		dst.fp16[i] := convert_bf8_to_fp16(__B.bf8[i])
+/// 		dst.fp16[i] := convert_bf8_to_fp16(__A.bf8[i])
 /// 	ELSE
 /// 		dst.fp16[i] := 0
 /// 	FI



More information about the cfe-commits mailing list