| 1 | /*===-------- amxmovrsintrin.h - AMX MOVRS intrinsics -*- C++ -*---------=== |
| 2 | * |
| 3 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | * See https://llvm.org/LICENSE.txt for license information. |
| 5 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | * |
| 7 | * ===-------------------------------------------------------------------=== */ |
| 8 | |
| 9 | #ifndef __IMMINTRIN_H |
| 10 | #error "Never use <amxmovrsintrin.h> directly; include <immintrin.h> instead." |
| 11 | #endif /* __IMMINTRIN_H */ |
| 12 | |
| 13 | #ifndef __AMXMOVRSINTRIN_H |
| 14 | #define __AMXMOVRSINTRIN_H |
| 15 | #ifdef __x86_64__ |
| 16 | |
| 17 | #define __DEFAULT_FN_ATTRS_MOVRS \ |
| 18 | __attribute__((__always_inline__, __nodebug__, __target__("amx-movrs"))) |
| 19 | |
| 20 | #define _tile_loaddrs(dst, base, stride) \ |
| 21 | __builtin_ia32_tileloaddrs64((dst), ((const void *)(base)), \ |
| 22 | (__SIZE_TYPE__)(stride)) |
| 23 | #define _tile_stream_loaddrs(dst, base, stride) \ |
| 24 | __builtin_ia32_tileloaddrst164((dst), ((const void *)(base)), \ |
| 25 | (__SIZE_TYPE__)(stride)) |
| 26 | static __inline__ _tile1024i __DEFAULT_FN_ATTRS_MOVRS |
| 27 | _tile_loaddrs_internal(unsigned short m, unsigned short n, const void *base, |
| 28 | __SIZE_TYPE__ stride) { |
| 29 | return __builtin_ia32_tileloaddrs64_internal(m, n, base, |
| 30 | (__SIZE_TYPE__)(stride)); |
| 31 | } |
| 32 | static __inline__ _tile1024i __DEFAULT_FN_ATTRS_MOVRS |
| 33 | _tile_loaddrst1_internal(unsigned short m, unsigned short n, const void *base, |
| 34 | __SIZE_TYPE__ stride) { |
| 35 | return __builtin_ia32_tileloaddrst164_internal(m, n, base, |
| 36 | (__SIZE_TYPE__)(stride)); |
| 37 | } |
| 38 | static __inline__ void __DEFAULT_FN_ATTRS_MOVRS |
| 39 | __tile_loaddrs(__tile1024i *dst, const void *base, __SIZE_TYPE__ stride) { |
| 40 | dst->tile = _tile_loaddrs_internal(dst->row, dst->col, base, stride); |
| 41 | } |
| 42 | static __inline__ void __DEFAULT_FN_ATTRS_MOVRS __tile_stream_loaddrs( |
| 43 | __tile1024i *dst, const void *base, __SIZE_TYPE__ stride) { |
| 44 | dst->tile = _tile_loaddrst1_internal(dst->row, dst->col, base, stride); |
| 45 | } |
| 46 | #undef __DEFAULT_FN_ATTRS_MOVRS |
| 47 | #endif /* __x86_64__ */ |
| 48 | #endif /* __AMXMOVRSINTRIN_H */ |