[X86][AMX] Assemble new AMX interface to an experiment header file.

[LuoYuanke]

Summary

Since the new AMX interface is not mature right now, we assemble the new
interface to an experiment file.

[pengfei]

clang/lib/Headers/amxintrin_experiment.h

13

It's better to add such macros:

#ifndef __AMXINTRIN_EXPERIMENT_H
#define __AMXINTRIN_EXPERIMENT_H
#ifdef __x86_64__

clang/test/CodeGen/X86/amx_api.c

4–5

Better to add #include <immintrin.h> at the same time. Or to simplify, add it in amxintrin_experiment.h.

[LuoYuanke]

Address Pengfei's comments.

clang/lib/Headers/amxintrin_experiment.h

13

Sure. Thanks.

[pengfei]

Hi @craig.topper, this patch is intent upon LLVM 12.0 release stablization, since the new AMX interface is on development and has some know issues. Do you think it is necessary and enough for the release? If it does, we can commit it to both trunk and 12.x branch.

clang/lib/Headers/CMakeLists.txt

 set(files
   adxintrin.h
   altivec.h
   ammintrin.h
   amxintrin.h
+  amxintrin_experiment.h
   arm_acle.h
   arm_cmse.h
   armintr.h
   arm64intr.h
   avx2intrin.h
   avx512bf16intrin.h
   avx512bwintrin.h
   avx512bitalgintrin.h

clang/lib/Headers/amxintrin.h

 ///    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.
 #define _tile_dpbf16ps(dst, src0, src1)                                        \
   __builtin_ia32_tdpbf16ps((dst), (src0), (src1))
 
-#define __DEFAULT_FN_ATTRS_INT8                                                \
-  __attribute__((__always_inline__, __nodebug__, __target__("amx-int8")))
-
-typedef int _tile1024i __attribute__((__vector_size__(1024), __aligned__(64)));
-static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8
-_tile_loadd_internal(unsigned short m, unsigned short n, const void *base,
-                     __SIZE_TYPE__ stride) {
-  return __builtin_ia32_tileloadd64_internal(m, n, base,
-                                             (__SIZE_TYPE__)(stride));
-}
-
-static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8
-_tile_dpbssd_internal(unsigned short m, unsigned short n, unsigned short k,
-                      _tile1024i dst, _tile1024i src1, _tile1024i src2) {
-  return __builtin_ia32_tdpbssd_internal(m, n, k, dst, src1, src2);
-}
-
-static __inline__ void __DEFAULT_FN_ATTRS_INT8
-_tile_stored_internal(unsigned short m, unsigned short n, void *base,
-                      __SIZE_TYPE__ stride, _tile1024i tile) {
-  return __builtin_ia32_tilestored64_internal(m, n, base,
-                                              (__SIZE_TYPE__)(stride), tile);
-}
-
-typedef struct __tile1024i_str {
-  const unsigned short row;
-  const unsigned short col;
-  _tile1024i tile;
-} __tile1024i;
-
-__DEFAULT_FN_ATTRS_TILE
-static void __tile_loadd(__tile1024i *dst, const void *base,
-                         __SIZE_TYPE__ stride) {
-  dst->tile = _tile_loadd_internal(dst->row, dst->col, base, stride);
-}
-
-__DEFAULT_FN_ATTRS_INT8
-static void __tile_dpbssd(__tile1024i *dst, __tile1024i src1,
-                          __tile1024i src2) {
-  dst->tile = _tile_dpbssd_internal(src1.row, src2.col, src1.col, dst->tile,
-                                    src1.tile, src2.tile);
-}
-
-__DEFAULT_FN_ATTRS_TILE
-static void __tile_stored(void *base, __SIZE_TYPE__ stride, __tile1024i src) {
-  _tile_stored_internal(src.row, src.col, base, stride, src.tile);
-}
-
-__DEFAULT_FN_ATTRS_TILE
-static void __tile_zero(__tile1024i *dst) {
-  dst->tile = __builtin_ia32_tilezero_internal(dst->row, dst->col);
-}
-
 #endif /* __x86_64__ */
 #endif /* __AMXINTRIN_H */

clang/lib/Headers/amxintrin_experiment.h

+/*===---------- amxintrin_experiment.h - AMX intrinsics -*- C/C++ -*---------===
+ *
+ * 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
+ *
+ *===------------------------------------------------------------------------===
+ */
+
+/*
+ * This file is experiment interface for AMX new programming model.
+ */
+
+#ifndef __AMXINTRIN_EXPERIMENT_H
+#define __AMXINTRIN_EXPERIMENT_H
+#ifdef __x86_64__
+
+#define __DEFAULT_FN_ATTRS_TILE                                                \
+  __attribute__((__always_inline__, __nodebug__, __target__("amx-tile")))
+
+#define __DEFAULT_FN_ATTRS_INT8                                                \
+  __attribute__((__always_inline__, __nodebug__, __target__("amx-int8")))
+
+typedef int _tile1024i __attribute__((__vector_size__(1024), __aligned__(64)));
+static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8
+_tile_loadd_internal(unsigned short m, unsigned short n, const void *base,
+                     __SIZE_TYPE__ stride) {
+  return __builtin_ia32_tileloadd64_internal(m, n, base,
+                                             (__SIZE_TYPE__)(stride));
+}
+
+static __inline__ _tile1024i __DEFAULT_FN_ATTRS_INT8
+_tile_dpbssd_internal(unsigned short m, unsigned short n, unsigned short k,
+                      _tile1024i dst, _tile1024i src1, _tile1024i src2) {
+  return __builtin_ia32_tdpbssd_internal(m, n, k, dst, src1, src2);
+}
+
+static __inline__ void __DEFAULT_FN_ATTRS_INT8
+_tile_stored_internal(unsigned short m, unsigned short n, void *base,
+                      __SIZE_TYPE__ stride, _tile1024i tile) {
+  return __builtin_ia32_tilestored64_internal(m, n, base,
+                                              (__SIZE_TYPE__)(stride), tile);
+}
+
+typedef struct __tile1024i_str {
+  const unsigned short row;
+  const unsigned short col;
+  _tile1024i tile;
+} __tile1024i;
+
+__DEFAULT_FN_ATTRS_TILE
+static void __tile_loadd(__tile1024i *dst, const void *base,
+                         __SIZE_TYPE__ stride) {
+  dst->tile = _tile_loadd_internal(dst->row, dst->col, base, stride);
+}
+
+__DEFAULT_FN_ATTRS_INT8
+static void __tile_dpbssd(__tile1024i *dst, __tile1024i src1,
+                          __tile1024i src2) {
+  dst->tile = _tile_dpbssd_internal(src1.row, src2.col, src1.col, dst->tile,
+                                    src1.tile, src2.tile);
+}
+
+__DEFAULT_FN_ATTRS_TILE
+static void __tile_stored(void *base, __SIZE_TYPE__ stride, __tile1024i src) {
+  _tile_stored_internal(src.row, src.col, base, stride, src.tile);
+}
+
+__DEFAULT_FN_ATTRS_TILE
+static void __tile_zero(__tile1024i *dst) {
+  dst->tile = __builtin_ia32_tilezero_internal(dst->row, dst->col);
+}
+
+#endif /* __x86_64__ */
+#endif /* __AMXINTRIN_EXPERIMENT_H */

clang/test/CodeGen/X86/amx_api.c

 // RUN: %clang_cc1 %s -ffreestanding -triple=x86_64-unknown-unknown  -target-feature +avx512f  -target-feature +amx-int8  \
 // RUN: -target-feature +amx-bf16 -emit-llvm -o - -Werror -pedantic | FileCheck %s --check-prefixes=CHECK
 
 #include <immintrin.h>
+#include <amxintrin_experiment.h>
 
 char buf[1024];
 #define STRIDE 32
 
 char buf2[1024];
 
 // This is an example code and integration test.
 void test_api(int cond, short row, short col) {