Skip to content

Commit e88ba6d

Browse files
momchil-velikovmomchil-velikov
authored
[AArch64] Add intrinsics for multi-vector to ZA array vector accumulators (#88266)
According to the specification in ARM-software/acle#309 this adds the intrinsics void_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za"); void_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za"); void_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za"); void_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za"); as well as the corresponding `bf16` variants.
1 parent cbf1535 commit e88ba6d

File tree

6 files changed

+389
-9
lines changed

6 files changed

+389
-9
lines changed

‎clang/include/clang/Basic/arm_sme.td

+10
Original file line numberDiff line numberDiff line change
@@ -298,6 +298,16 @@ multiclass ZAAddSub<string n_suffix> {
298298
def NAME # _ZA64_VG1X2_F64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x2", "vm2", "d", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x2", [IsStreaming, IsInOutZA], []>;
299299
def NAME # _ZA64_VG1X4_F64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x4", "vm4", "d", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x4", [IsStreaming, IsInOutZA], []>;
300300
}
301+
302+
let TargetGuard = "sme-f16f16|sme-f8f16" in {
303+
def NAME # _ZA16_VG1X2_F16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x2", "vm2", "h", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x2", [IsStreaming, IsInOutZA], []>;
304+
def NAME # _ZA16_VG1X4_F16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x4", "vm4", "h", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x4", [IsStreaming, IsInOutZA], []>;
305+
}
306+
307+
let TargetGuard = "sme2,b16b16" in {
308+
def NAME # _ZA16_VG1X2_BF16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x2", "vm2", "b", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x2", [IsStreaming, IsInOutZA], []>;
309+
def NAME # _ZA16_VG1X4_BF16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x4", "vm4", "b", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x4", [IsStreaming, IsInOutZA], []>;
310+
}
301311
}
302312

303313
defm SVADD : ZAAddSub<"add">;

‎clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c

+193
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,193 @@
1+
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 4
2+
// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s
3+
// RUN: %clang_cc1 -x c++ -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-f8f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CXX
4+
// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-f8f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s
5+
// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -x c++ -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CXX
6+
7+
// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -o /dev/null
8+
9+
// REQUIRES: aarch64-registered-target
10+
11+
#include <arm_sme.h>
12+
13+
#ifdef SVE_OVERLOADED_FORMS
14+
#define SVE_ACLE_FUNC(A1,A2_UNUSED,A3) A1##A3
15+
#else
16+
#define SVE_ACLE_FUNC(A1,A2,A3) A1##A2##A3
17+
#endif
18+
19+
// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x2_f16(
20+
// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
21+
// CHECK-NEXT: entry:
22+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
23+
// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
24+
// CHECK-NEXT: tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
25+
// CHECK-NEXT: ret void
26+
//
27+
// CHECK-CXX-LABEL: define dso_local void @_Z25test_svadd_za16_vg1x2_f16j13svfloat16x2_t(
28+
// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
29+
// CHECK-CXX-NEXT: entry:
30+
// CHECK-CXX-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
31+
// CHECK-CXX-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
32+
// CHECK-CXX-NEXT: tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
33+
// CHECK-CXX-NEXT: ret void
34+
//
35+
void test_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za") {
36+
SVE_ACLE_FUNC(svadd_za16,_f16,_vg1x2)(slice, zn);
37+
}
38+
39+
// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x4_f16(
40+
// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
41+
// CHECK-NEXT: entry:
42+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
43+
// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
44+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
45+
// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
46+
// CHECK-NEXT: tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
47+
// CHECK-NEXT: ret void
48+
//
49+
// CHECK-CXX-LABEL: define dso_local void @_Z25test_svadd_za16_vg1x4_f16j13svfloat16x4_t(
50+
// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
51+
// CHECK-CXX-NEXT: entry:
52+
// CHECK-CXX-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
53+
// CHECK-CXX-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
54+
// CHECK-CXX-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
55+
// CHECK-CXX-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
56+
// CHECK-CXX-NEXT: tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
57+
// CHECK-CXX-NEXT: ret void
58+
//
59+
void test_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za") {
60+
SVE_ACLE_FUNC(svadd_za16,_f16,_vg1x4)(slice, zn);
61+
}
62+
63+
// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x2_f16(
64+
// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
65+
// CHECK-NEXT: entry:
66+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
67+
// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
68+
// CHECK-NEXT: tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
69+
// CHECK-NEXT: ret void
70+
//
71+
// CHECK-CXX-LABEL: define dso_local void @_Z25test_svsub_za16_vg1x2_f16j13svfloat16x2_t(
72+
// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
73+
// CHECK-CXX-NEXT: entry:
74+
// CHECK-CXX-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
75+
// CHECK-CXX-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
76+
// CHECK-CXX-NEXT: tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
77+
// CHECK-CXX-NEXT: ret void
78+
//
79+
void test_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za") {
80+
SVE_ACLE_FUNC(svsub_za16,_f16,_vg1x2)(slice, zn);
81+
}
82+
83+
// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x4_f16(
84+
// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
85+
// CHECK-NEXT: entry:
86+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
87+
// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
88+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
89+
// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
90+
// CHECK-NEXT: tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
91+
// CHECK-NEXT: ret void
92+
//
93+
// CHECK-CXX-LABEL: define dso_local void @_Z25test_svsub_za16_vg1x4_f16j13svfloat16x4_t(
94+
// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
95+
// CHECK-CXX-NEXT: entry:
96+
// CHECK-CXX-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
97+
// CHECK-CXX-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
98+
// CHECK-CXX-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
99+
// CHECK-CXX-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
100+
// CHECK-CXX-NEXT: tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
101+
// CHECK-CXX-NEXT: ret void
102+
//
103+
void test_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za") {
104+
SVE_ACLE_FUNC(svsub_za16,_f16,_vg1x4)(slice, zn);
105+
}
106+
107+
// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x2_bf16(
108+
// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
109+
// CHECK-NEXT: entry:
110+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
111+
// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
112+
// CHECK-NEXT: tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
113+
// CHECK-NEXT: ret void
114+
//
115+
// CHECK-CXX-LABEL: define dso_local void @_Z26test_svadd_za16_vg1x2_bf16j14svbfloat16x2_t(
116+
// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
117+
// CHECK-CXX-NEXT: entry:
118+
// CHECK-CXX-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
119+
// CHECK-CXX-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
120+
// CHECK-CXX-NEXT: tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
121+
// CHECK-CXX-NEXT: ret void
122+
//
123+
void test_svadd_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streaming __arm_inout("za") {
124+
SVE_ACLE_FUNC(svadd_za16,_bf16,_vg1x2)(slice, zn);
125+
}
126+
127+
// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x4_bf16(
128+
// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
129+
// CHECK-NEXT: entry:
130+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
131+
// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
132+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
133+
// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
134+
// CHECK-NEXT: tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
135+
// CHECK-NEXT: ret void
136+
//
137+
// CHECK-CXX-LABEL: define dso_local void @_Z26test_svadd_za16_vg1x4_bf16j14svbfloat16x4_t(
138+
// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
139+
// CHECK-CXX-NEXT: entry:
140+
// CHECK-CXX-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
141+
// CHECK-CXX-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
142+
// CHECK-CXX-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
143+
// CHECK-CXX-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
144+
// CHECK-CXX-NEXT: tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
145+
// CHECK-CXX-NEXT: ret void
146+
//
147+
void test_svadd_za16_vg1x4_bf16(uint32_t slice, svbfloat16x4_t zn) __arm_streaming __arm_inout("za") {
148+
SVE_ACLE_FUNC(svadd_za16,_bf16,_vg1x4)(slice, zn);
149+
}
150+
151+
// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x2_bf16(
152+
// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
153+
// CHECK-NEXT: entry:
154+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
155+
// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
156+
// CHECK-NEXT: tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
157+
// CHECK-NEXT: ret void
158+
//
159+
// CHECK-CXX-LABEL: define dso_local void @_Z26test_svsub_za16_vg1x2_bf16j14svbfloat16x2_t(
160+
// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
161+
// CHECK-CXX-NEXT: entry:
162+
// CHECK-CXX-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
163+
// CHECK-CXX-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
164+
// CHECK-CXX-NEXT: tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
165+
// CHECK-CXX-NEXT: ret void
166+
//
167+
void test_svsub_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streaming __arm_inout("za") {
168+
SVE_ACLE_FUNC(svsub_za16,_bf16,_vg1x2)(slice, zn);
169+
}
170+
171+
// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x4_bf16(
172+
// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
173+
// CHECK-NEXT: entry:
174+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
175+
// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
176+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
177+
// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
178+
// CHECK-NEXT: tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
179+
// CHECK-NEXT: ret void
180+
//
181+
// CHECK-CXX-LABEL: define dso_local void @_Z26test_svsub_za16_vg1x4_bf16j14svbfloat16x4_t(
182+
// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
183+
// CHECK-CXX-NEXT: entry:
184+
// CHECK-CXX-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
185+
// CHECK-CXX-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
186+
// CHECK-CXX-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
187+
// CHECK-CXX-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
188+
// CHECK-CXX-NEXT: tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
189+
// CHECK-CXX-NEXT: ret void
190+
//
191+
void test_svsub_za16_vg1x4_bf16(uint32_t slice, svbfloat16x4_t zn) __arm_streaming __arm_inout("za") {
192+
SVE_ACLE_FUNC(svsub_za16,_bf16,_vg1x4)(slice, zn);
193+
}

‎clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c

+29
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,29 @@
1+
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -fsyntax-only -verify -emit-llvm %s
2+
3+
// REQUIRES: aarch64-registered-target
4+
5+
#include <arm_sme.h>
6+
7+
void test_features(uint32_t slice, svfloat16x2_t zn2, svfloat16x4_t zn4,
8+
svbfloat16x2_t bzn2, svbfloat16x4_t bzn4) __arm_streaming __arm_inout("za") {
9+
// expected-error@+1 {{'svadd_za16_f16_vg1x2' needs target feature sme-f16f16|sme-f8f16}}
10+
svadd_za16_f16_vg1x2(slice, zn2);
11+
// expected-error@+1 {{'svadd_za16_f16_vg1x4' needs target feature sme-f16f16|sme-f8f16}}
12+
svadd_za16_f16_vg1x4(slice, zn4);
13+
// expected-error@+1 {{'svsub_za16_f16_vg1x2' needs target feature sme-f16f16|sme-f8f16}}
14+
svsub_za16_f16_vg1x2(slice, zn2);
15+
// expected-error@+1 {{'svsub_za16_f16_vg1x4' needs target feature sme-f16f16|sme-f8f16}}
16+
svsub_za16_f16_vg1x4(slice, zn4);
17+
18+
// expected-error@+1 {{'svadd_za16_bf16_vg1x2' needs target feature sme2,b16b16}}
19+
svadd_za16_bf16_vg1x2(slice, bzn2);
20+
// expected-error@+1 {{'svadd_za16_bf16_vg1x4' needs target feature sme2,b16b16}}
21+
svadd_za16_bf16_vg1x4(slice, bzn4);
22+
// expected-error@+1 {{'svsub_za16_bf16_vg1x2' needs target feature sme2,b16b16}}
23+
svsub_za16_bf16_vg1x2(slice, bzn2);
24+
// expected-error@+1 {{'svsub_za16_bf16_vg1x4' needs target feature sme2,b16b16}}
25+
svsub_za16_bf16_vg1x4(slice, bzn4);
26+
}
27+
28+
29+

‎llvm/include/llvm/IR/IntrinsicsAArch64.td

+1-1
Original file line numberDiff line numberDiff line change
@@ -3481,7 +3481,7 @@ let TargetPrefix = "aarch64" in {
34813481
// Multi-vector add/sub and accumulate into ZA
34823482
//
34833483
foreach intr = ["add", "sub"] in {
3484-
foreach za = ["za32", "za64"] in {
3484+
foreach za = ["za16","za32", "za64"] in {
34853485
def int_aarch64_sme_ # intr # _ # za # _vg1x2 : SME2_ZA_Write_VG2_Intrinsic;
34863486
def int_aarch64_sme_ # intr # _ # za # _vg1x4 : SME2_ZA_Write_VG4_Intrinsic;
34873487
}

0 commit comments

Comments
 (0)