Skip to content

Commit 1954686

Browse files
kmclaughlin-armkmclaughlin-arm
authored
[Clang][SME2] Add multi-vector unpack builtins (#75075)
Adds the following SME2 builtins: - svunpk (x2 & x4) See https://github.com/ARM-software/acle/pull/217/files Patch by David Sherwood <[email protected]>
1 parent 1e47a15 commit 1954686

File tree

3 files changed

+383
-0
lines changed

3 files changed

+383
-0
lines changed

‎clang/include/clang/Basic/arm_sve.td

Lines changed: 11 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2262,3 +2262,14 @@ let TargetGuard = "sme2" in {
22622262
def SVQCVTN_U16_U64_X4 : SInst<"svqcvtn_u16[_{d}_x4]", "b4.d", "Ul", MergeNone, "aarch64_sve_uqcvtn_x4", [IsStreaming], []>;
22632263
def SVQCVTN_U16_S64_X4 : SInst<"svqcvtn_u16[_{d}_x4]", "b4.d", "l", MergeNone, "aarch64_sve_sqcvtun_x4", [IsStreaming], []>;
22642264
}
2265+
2266+
//
2267+
// Multi-vector unpack
2268+
//
2269+
2270+
let TargetGuard = "sme2" in {
2271+
def SVSUNPK_X2 : SInst<"svunpk_{d}[_{1}_x2]", "2h", "sil", MergeNone, "aarch64_sve_sunpk_x2", [IsStreaming], []>;
2272+
def SVUUNPK_X2 : SInst<"svunpk_{d}[_{1}_x2]", "2h", "UsUiUl", MergeNone, "aarch64_sve_uunpk_x2", [IsStreaming], []>;
2273+
def SVSUNPK_X4 : SInst<"svunpk_{d}[_{3}_x4]", "42.h", "sil", MergeNone, "aarch64_sve_sunpk_x4", [IsStreaming], []>;
2274+
def SVUUNPK_X4 : SInst<"svunpk_{d}[_{3}_x4]", "42.h", "UsUiUl", MergeNone, "aarch64_sve_uunpk_x4", [IsStreaming], []>;
2275+
}

‎clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_unpkx2.c

Lines changed: 150 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,150 @@
1+
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
2+
3+
// REQUIRES: aarch64-registered-target
4+
5+
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
6+
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
7+
// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
8+
// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
9+
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -target-feature -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
10+
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
11+
#include <arm_sme_draft_spec_subject_to_change.h>
12+
13+
#ifdef SVE_OVERLOADED_FORMS
14+
// A simple used,unused... macro, long enough to represent any SVE builtin.
15+
#define SVE_ACLE_FUNC(A1,A2_UNUSED) A1
16+
#else
17+
#define SVE_ACLE_FUNC(A1,A2) A1##A2
18+
#endif
19+
20+
// CHECK-LABEL: @test_svunpk_s16_x2(
21+
// CHECK-NEXT: entry:
22+
// CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sunpk.x2.nxv8i16(<vscale x 16 x i8> [[ZN:%.*]])
23+
// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
24+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
25+
// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
26+
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
27+
// CHECK-NEXT: ret <vscale x 16 x i16> [[TMP4]]
28+
//
29+
// CPP-CHECK-LABEL: @_Z18test_svunpk_s16_x2u10__SVInt8_t(
30+
// CPP-CHECK-NEXT: entry:
31+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.sunpk.x2.nxv8i16(<vscale x 16 x i8> [[ZN:%.*]])
32+
// CPP-CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
33+
// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
34+
// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
35+
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
36+
// CPP-CHECK-NEXT: ret <vscale x 16 x i16> [[TMP4]]
37+
//
38+
svint16x2_t test_svunpk_s16_x2(svint8_t zn) __arm_streaming {
39+
return SVE_ACLE_FUNC(svunpk_s16,_s8_x2)(zn);
40+
}
41+
42+
// CHECK-LABEL: @test_svunpk_u16_x2(
43+
// CHECK-NEXT: entry:
44+
// CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uunpk.x2.nxv8i16(<vscale x 16 x i8> [[ZN:%.*]])
45+
// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
46+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
47+
// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
48+
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
49+
// CHECK-NEXT: ret <vscale x 16 x i16> [[TMP4]]
50+
//
51+
// CPP-CHECK-LABEL: @_Z18test_svunpk_u16_x2u11__SVUint8_t(
52+
// CPP-CHECK-NEXT: entry:
53+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.uunpk.x2.nxv8i16(<vscale x 16 x i8> [[ZN:%.*]])
54+
// CPP-CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 0
55+
// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> poison, <vscale x 8 x i16> [[TMP1]], i64 0)
56+
// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[TMP0]], 1
57+
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 16 x i16> @llvm.vector.insert.nxv16i16.nxv8i16(<vscale x 16 x i16> [[TMP2]], <vscale x 8 x i16> [[TMP3]], i64 8)
58+
// CPP-CHECK-NEXT: ret <vscale x 16 x i16> [[TMP4]]
59+
//
60+
svuint16x2_t test_svunpk_u16_x2(svuint8_t zn) __arm_streaming {
61+
return SVE_ACLE_FUNC(svunpk_u16,_u8_x2)(zn);
62+
}
63+
64+
// CHECK-LABEL: @test_svunpk_s32_x2(
65+
// CHECK-NEXT: entry:
66+
// CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sunpk.x2.nxv4i32(<vscale x 8 x i16> [[ZN:%.*]])
67+
// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
68+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
69+
// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
70+
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
71+
// CHECK-NEXT: ret <vscale x 8 x i32> [[TMP4]]
72+
//
73+
// CPP-CHECK-LABEL: @_Z18test_svunpk_s32_x2u11__SVInt16_t(
74+
// CPP-CHECK-NEXT: entry:
75+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.sunpk.x2.nxv4i32(<vscale x 8 x i16> [[ZN:%.*]])
76+
// CPP-CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
77+
// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
78+
// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
79+
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
80+
// CPP-CHECK-NEXT: ret <vscale x 8 x i32> [[TMP4]]
81+
//
82+
svint32x2_t test_svunpk_s32_x2(svint16_t zn) __arm_streaming {
83+
return SVE_ACLE_FUNC(svunpk_s32,_s16_x2)(zn);
84+
}
85+
86+
// CHECK-LABEL: @test_svunpk_u32_x2(
87+
// CHECK-NEXT: entry:
88+
// CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uunpk.x2.nxv4i32(<vscale x 8 x i16> [[ZN:%.*]])
89+
// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
90+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
91+
// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
92+
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
93+
// CHECK-NEXT: ret <vscale x 8 x i32> [[TMP4]]
94+
//
95+
// CPP-CHECK-LABEL: @_Z18test_svunpk_u32_x2u12__SVUint16_t(
96+
// CPP-CHECK-NEXT: entry:
97+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.uunpk.x2.nxv4i32(<vscale x 8 x i16> [[ZN:%.*]])
98+
// CPP-CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 0
99+
// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> poison, <vscale x 4 x i32> [[TMP1]], i64 0)
100+
// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[TMP0]], 1
101+
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x i32> @llvm.vector.insert.nxv8i32.nxv4i32(<vscale x 8 x i32> [[TMP2]], <vscale x 4 x i32> [[TMP3]], i64 4)
102+
// CPP-CHECK-NEXT: ret <vscale x 8 x i32> [[TMP4]]
103+
//
104+
svuint32x2_t test_svunpk_u32_x2(svuint16_t zn) __arm_streaming {
105+
return SVE_ACLE_FUNC(svunpk_u32,_u16_x2)(zn);
106+
}
107+
108+
// CHECK-LABEL: @test_svunpk_s64_x2(
109+
// CHECK-NEXT: entry:
110+
// CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sunpk.x2.nxv2i64(<vscale x 4 x i32> [[ZN:%.*]])
111+
// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
112+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
113+
// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
114+
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
115+
// CHECK-NEXT: ret <vscale x 4 x i64> [[TMP4]]
116+
//
117+
// CPP-CHECK-LABEL: @_Z18test_svunpk_s64_x2u11__SVInt32_t(
118+
// CPP-CHECK-NEXT: entry:
119+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.sunpk.x2.nxv2i64(<vscale x 4 x i32> [[ZN:%.*]])
120+
// CPP-CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
121+
// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
122+
// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
123+
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
124+
// CPP-CHECK-NEXT: ret <vscale x 4 x i64> [[TMP4]]
125+
//
126+
svint64x2_t test_svunpk_s64_x2(svint32_t zn) __arm_streaming {
127+
return SVE_ACLE_FUNC(svunpk_s64,_s32_x2)(zn);
128+
}
129+
130+
// CHECK-LABEL: @test_svunpk_u64_x2(
131+
// CHECK-NEXT: entry:
132+
// CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uunpk.x2.nxv2i64(<vscale x 4 x i32> [[ZN:%.*]])
133+
// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
134+
// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
135+
// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
136+
// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
137+
// CHECK-NEXT: ret <vscale x 4 x i64> [[TMP4]]
138+
//
139+
// CPP-CHECK-LABEL: @_Z18test_svunpk_u64_x2u12__SVUint32_t(
140+
// CPP-CHECK-NEXT: entry:
141+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.uunpk.x2.nxv2i64(<vscale x 4 x i32> [[ZN:%.*]])
142+
// CPP-CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 0
143+
// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP1]], i64 0)
144+
// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[TMP0]], 1
145+
// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], i64 2)
146+
// CPP-CHECK-NEXT: ret <vscale x 4 x i64> [[TMP4]]
147+
//
148+
svuint64x2_t test_svunpk_u64_x2(svuint32_t zn) __arm_streaming {
149+
return SVE_ACLE_FUNC(svunpk_u64,_u32_x2)(zn);
150+
}

0 commit comments

Comments
 (0)