Skip to content

Commit 36b8af1

Browse files
author
Francesco Petrogalli
committed
[SveEmitter] Add SVE ACLE for svld1ro.
Reviewers: sdesmalen, efriedma Subscribers: tschuett, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D80740
1 parent 056a375 commit 36b8af1

File tree

2 files changed

+101
-0
lines changed

2 files changed

+101
-0
lines changed

‎clang/include/clang/Basic/arm_sve.td

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -470,6 +470,10 @@ def SVLDNT1_VNUM : MInst<"svldnt1_vnum[_{2}]", "dPcl", "csilUcUsUiUlhfd", [IsLoa
470470
// Load one quadword and replicate (scalar base)
471471
def SVLD1RQ : SInst<"svld1rq[_{2}]", "dPc", "csilUcUsUiUlhfd", MergeNone, "aarch64_sve_ld1rq">;
472472

473+
// Load one octoword and replicate (scalar base)
474+
let ArchGuard = "defined(__ARM_FEATURE_SVE_MATMUL_FP64)" in {
475+
def SVLD1RO : SInst<"svld1ro[_{2}]", "dPc", "csilUcUsUiUlhfd", MergeNone, "aarch64_sve_ld1ro">;
476+
}
473477
////////////////////////////////////////////////////////////////////////////////
474478
// Stores
475479

‎clang/test/CodeGen/aarch64-sve-intrinsics/acle_sve_ld1ro.c

Lines changed: 97 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,97 @@
1+
// RUN: %clang_cc1 -D__ARM_FEATURE_SVE_MATMUL_FP64 -D__ARM_FEATURE_SVE -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
2+
// RUN: %clang_cc1 -D__ARM_FEATURE_SVE_MATMUL_FP64 -D__ARM_FEATURE_SVE -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
3+
4+
#include <arm_sve.h>
5+
6+
#ifdef SVE_OVERLOADED_FORMS
7+
// A simple used,unused... macro, long enough to represent any SVE builtin.
8+
#define SVE_ACLE_FUNC(A1, A2_UNUSED, A3, A4_UNUSED) A1##A3
9+
#else
10+
#define SVE_ACLE_FUNC(A1, A2, A3, A4) A1##A2##A3##A4
11+
#endif
12+
13+
svint8_t test_svld1ro_s8(svbool_t pg, const int8_t *base) {
14+
// CHECK-LABEL: test_svld1ro_s8
15+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i8> @llvm.aarch64.sve.ld1ro.nxv16i8(<vscale x 16 x i1> %pg, i8* %base)
16+
// CHECK: ret <vscale x 16 x i8> %[[INTRINSIC]]
17+
return SVE_ACLE_FUNC(svld1ro, _s8, , )(pg, base);
18+
}
19+
20+
svint16_t test_svld1ro_s16(svbool_t pg, const int16_t *base) {
21+
// CHECK-LABEL: test_svld1ro_s16
22+
// CHECK: %[[PG:.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %pg)
23+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 8 x i16> @llvm.aarch64.sve.ld1ro.nxv8i16(<vscale x 8 x i1> %[[PG]], i16* %base)
24+
// CHECK: ret <vscale x 8 x i16> %[[INTRINSIC]]
25+
return SVE_ACLE_FUNC(svld1ro, _s16, , )(pg, base);
26+
}
27+
28+
svint32_t test_svld1ro_s32(svbool_t pg, const int32_t *base) {
29+
// CHECK-LABEL: test_svld1ro_s32
30+
// CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
31+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1ro.nxv4i32(<vscale x 4 x i1> %[[PG]], i32* %base)
32+
// CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
33+
return SVE_ACLE_FUNC(svld1ro, _s32, , )(pg, base);
34+
}
35+
36+
svint64_t test_svld1ro_s64(svbool_t pg, const int64_t *base) {
37+
// CHECK-LABEL: test_svld1ro_s64
38+
// CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
39+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.ld1ro.nxv2i64(<vscale x 2 x i1> %[[PG]], i64* %base)
40+
// CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
41+
return SVE_ACLE_FUNC(svld1ro, _s64, , )(pg, base);
42+
}
43+
44+
svuint8_t test_svld1ro_u8(svbool_t pg, const uint8_t *base) {
45+
// CHECK-LABEL: test_svld1ro_u8
46+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 16 x i8> @llvm.aarch64.sve.ld1ro.nxv16i8(<vscale x 16 x i1> %pg, i8* %base)
47+
// CHECK: ret <vscale x 16 x i8> %[[INTRINSIC]]
48+
return SVE_ACLE_FUNC(svld1ro, _u8, , )(pg, base);
49+
}
50+
51+
svuint16_t test_svld1ro_u16(svbool_t pg, const uint16_t *base) {
52+
// CHECK-LABEL: test_svld1ro_u16
53+
// CHECK: %[[PG:.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %pg)
54+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 8 x i16> @llvm.aarch64.sve.ld1ro.nxv8i16(<vscale x 8 x i1> %[[PG]], i16* %base)
55+
// CHECK: ret <vscale x 8 x i16> %[[INTRINSIC]]
56+
return SVE_ACLE_FUNC(svld1ro, _u16, , )(pg, base);
57+
}
58+
59+
svuint32_t test_svld1ro_u32(svbool_t pg, const uint32_t *base) {
60+
// CHECK-LABEL: test_svld1ro_u32
61+
// CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
62+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x i32> @llvm.aarch64.sve.ld1ro.nxv4i32(<vscale x 4 x i1> %[[PG]], i32* %base)
63+
// CHECK: ret <vscale x 4 x i32> %[[INTRINSIC]]
64+
return SVE_ACLE_FUNC(svld1ro, _u32, , )(pg, base);
65+
}
66+
67+
svuint64_t test_svld1ro_u64(svbool_t pg, const uint64_t *base) {
68+
// CHECK-LABEL: test_svld1ro_u64
69+
// CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
70+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x i64> @llvm.aarch64.sve.ld1ro.nxv2i64(<vscale x 2 x i1> %[[PG]], i64* %base)
71+
// CHECK: ret <vscale x 2 x i64> %[[INTRINSIC]]
72+
return SVE_ACLE_FUNC(svld1ro, _u64, , )(pg, base);
73+
}
74+
75+
svfloat16_t test_svld1ro_f16(svbool_t pg, const float16_t *base) {
76+
// CHECK-LABEL: test_svld1ro_f16
77+
// CHECK: %[[PG:.*]] = call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> %pg)
78+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 8 x half> @llvm.aarch64.sve.ld1ro.nxv8f16(<vscale x 8 x i1> %[[PG]], half* %base)
79+
// CHECK: ret <vscale x 8 x half> %[[INTRINSIC]]
80+
return SVE_ACLE_FUNC(svld1ro, _f16, , )(pg, base);
81+
}
82+
83+
svfloat32_t test_svld1ro_f32(svbool_t pg, const float32_t *base) {
84+
// CHECK-LABEL: test_svld1ro_f32
85+
// CHECK: %[[PG:.*]] = call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> %pg)
86+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 4 x float> @llvm.aarch64.sve.ld1ro.nxv4f32(<vscale x 4 x i1> %[[PG]], float* %base)
87+
// CHECK: ret <vscale x 4 x float> %[[INTRINSIC]]
88+
return SVE_ACLE_FUNC(svld1ro, _f32, , )(pg, base);
89+
}
90+
91+
svfloat64_t test_svld1ro_f64(svbool_t pg, const float64_t *base) {
92+
// CHECK-LABEL: test_svld1ro_f64
93+
// CHECK: %[[PG:.*]] = call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> %pg)
94+
// CHECK: %[[INTRINSIC:.*]] = call <vscale x 2 x double> @llvm.aarch64.sve.ld1ro.nxv2f64(<vscale x 2 x i1> %[[PG]], double* %base)
95+
// CHECK: ret <vscale x 2 x double> %[[INTRINSIC]]
96+
return SVE_ACLE_FUNC(svld1ro, _f64, , )(pg, base);
97+
}

0 commit comments

Comments
 (0)