blob: 191f8c6097a2cc586064a97ddb01c37f5af1aaf8 [file] [log] [blame] [edit]
/*===----- amxfp16transposeintrin.h - AMX-FP16 and AMX-TRANSPOSE ------------===
*
* Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
* See https://llvm.org/LICENSE.txt for license information.
* SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
*
*===------------------------------------------------------------------------===
*/
#ifndef __IMMINTRIN_H
#error \
"Never use <amxfp16transposeintrin.h> directly; use <immintrin.h> instead."
#endif /* __IMMINTRIN_H */
#ifndef __AMX_FP16TRANSPOSEINTRIN_H
#define __AMX_FP16TRANSPOSEINTRIN_H
#ifdef __x86_64__
/* Define the default attributes for the functions in this file. */
#define __DEFAULT_FN_ATTRS \
__attribute__((__always_inline__, __nodebug__, \
__target__("amx-fp16,amx-transpose")))
/// Compute transpose and dot-product of FP16 (16-bit) floating-point pairs in
/// tiles \a a and \a b, accumulating the intermediate single-precision
/// (32-bit) floating-point elements with elements in \a dst, and store the
/// 32-bit result back to tile \a dst.
///
/// \headerfile <immintrin.h>
///
/// \code
/// void _tile_tdpfp16ps (__tile dst, __tile a, __tile b)
/// \endcode
///
/// \code{.operation}
/// FOR m := 0 TO dst.rows - 1
/// tmp := dst.row[m]
/// FOR k := 0 TO (a.colsb / 4) - 1
/// FOR n := 0 TO (dst.colsb / 4) - 1
/// tmp.fp32[n] += FP32(a.row[m].fp16[2*k+0]) *
/// FP32(b.row[k].fp16[2*n+0])
/// tmp.fp32[n] += FP32(a.row[m].fp16[2*k+1]) *
/// FP32(b.row[k].fp16[2*n+1])
/// ENDFOR
/// ENDFOR
/// write_row_and_zero(dst, m, tmp, dst.colsb)
/// ENDFOR
/// zero_upper_rows(dst, dst.rows)
/// zero_tileconfig_start()
/// \endcode
///
/// This intrinsic corresponds to the \c TTDPFP16PS instruction.
///
/// \param dst
/// The destination tile. Max size is 1024 Bytes.
/// \param a
/// The 1st source tile. Max size is 1024 Bytes.
/// \param b
/// The 2nd source tile. Max size is 1024 Bytes.
#define _tile_tdpfp16ps(dst, a, b) __builtin_ia32_ttdpfp16ps((dst), (a), (b))
/// This is internal intrinsic. C/C++ user should avoid calling it directly.
static __inline__ _tile1024i __DEFAULT_FN_ATTRS
_tile_tdpfp16ps_internal(unsigned short m, unsigned short n, unsigned short k,
_tile1024i dst, _tile1024i src1, _tile1024i src2) {
return __builtin_ia32_ttdpfp16ps_internal(m, n, k, dst, src1, src2);
}
/// Compute transpose and dot-product of FP16 (16-bit) floating-point pairs in
/// tiles src0 and src1, accumulating the intermediate single-precision
/// (32-bit) floating-point elements with elements in "dst", and store the
/// 32-bit result back to tile "dst".
///
/// \headerfile <immintrin.h>
///
/// This intrinsic corresponds to the <c> TTDPFP16PS </c> instruction.
///
/// \param dst
/// The destination tile. Max size is 1024 Bytes.
/// \param src0
/// The 1st source tile. Max size is 1024 Bytes.
/// \param src1
/// The 2nd source tile. Max size is 1024 Bytes.
__DEFAULT_FN_ATTRS
static __inline__ void __tile_tdpfp16ps(__tile1024i *dst, __tile1024i src0,
__tile1024i src1) {
dst->tile = _tile_tdpfp16ps_internal(src0.row, src1.col, src0.col, dst->tile,
src0.tile, src1.tile);
}
#undef __DEFAULT_FN_ATTRS
#endif /* __x86_64__ */
#endif /* __AMX_FP16TRANSPOSEINTRIN_H */