Skip to content

Commit 9f93a99

Browse files
CarolineConcattoCarolineConcatto
committed
[Clang][SVE2.1] Add builtins for 2-way svdot (vectors, indexed)
As described in: ARM-software/acle#257 Patch by: David Sherwood <[email protected]> Reviewed By: dtemirbulatov Differential Revision: https://reviews.llvm.org/D151439
1 parent 03d1c99 commit 9f93a99

File tree

3 files changed

+123
-0
lines changed

3 files changed

+123
-0
lines changed

‎clang/include/clang/Basic/arm_sve.td

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1952,6 +1952,13 @@ def SVSTNT1B_VNUM_X4 : MInst<"svstnt1_vnum[_{2}_x4]", "v}pl4", "cUc", [IsStructS
19521952
def SVSTNT1H_VNUM_X4 : MInst<"svstnt1_vnum[_{2}_x4]", "v}pl4", "sUshb", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
19531953
def SVSTNT1W_VNUM_X4 : MInst<"svstnt1_vnum[_{2}_x4]", "v}pl4", "iUif", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
19541954
def SVSTNT1D_VNUM_X4 : MInst<"svstnt1_vnum[_{2}_x4]", "v}pl4", "lUld", [IsStructStore], MemEltTyDefault, "aarch64_sve_stnt1_pn_x4">;
1955+
1956+
def SVDOT_X2_S : SInst<"svdot[_{d}_{2}_{3}]", "ddhh", "i", MergeNone, "aarch64_sve_sdot_x2", [], []>;
1957+
def SVDOT_X2_U : SInst<"svdot[_{d}_{2}_{3}]", "ddhh", "Ui", MergeNone, "aarch64_sve_udot_x2", [], []>;
1958+
def SVDOT_X2_F : SInst<"svdot[_{d}_{2}_{3}]", "ddhh", "f", MergeNone, "aarch64_sve_fdot_x2", [], []>;
1959+
def SVDOT_LANE_X2_S : SInst<"svdot_lane[_{d}_{2}_{3}]", "ddhhi", "i", MergeNone, "aarch64_sve_sdot_lane_x2", [], [ImmCheck<3, ImmCheck0_3>]>;
1960+
def SVDOT_LANE_X2_U : SInst<"svdot_lane[_{d}_{2}_{3}]", "ddhhi", "Ui", MergeNone, "aarch64_sve_udot_lane_x2", [], [ImmCheck<3, ImmCheck0_3>]>;
1961+
def SVDOT_LANE_X2_F : SInst<"svdot_lane[_{d}_{2}_{3}]", "ddhhi", "f", MergeNone, "aarch64_sve_fdot_lane_x2", [], [ImmCheck<3, ImmCheck0_3>]>;
19551962
}
19561963

19571964
let TargetGuard = "sve2p1" in {

‎clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_dot.c

Lines changed: 107 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,107 @@
1+
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
2+
// REQUIRES: aarch64-registered-target
3+
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
4+
// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
5+
// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
6+
// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2p1 -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 -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
8+
#include <arm_sve.h>
9+
10+
#ifdef SVE_OVERLOADED_FORMS
11+
// A simple used,unused... macro, long enough to represent any SVE builtin.
12+
#define SVE_ACLE_FUNC(A1,A2_UNUSED,A3) A1##A3
13+
#else
14+
#define SVE_ACLE_FUNC(A1,A2,A3) A1##A2##A3
15+
#endif
16+
17+
// CHECK-LABEL: @test_svdot_s32_x2(
18+
// CHECK-NEXT: entry:
19+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.sdot.x2.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
20+
// CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
21+
//
22+
// CPP-CHECK-LABEL: @_Z17test_svdot_s32_x2u11__SVInt32_tu11__SVInt16_tu11__SVInt16_t(
23+
// CPP-CHECK-NEXT: entry:
24+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.sdot.x2.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
25+
// CPP-CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
26+
//
27+
svint32_t test_svdot_s32_x2(svint32_t op1, svint16_t op2, svint16_t op3)
28+
{
29+
return SVE_ACLE_FUNC(svdot,_s32_s16_s16,)(op1, op2, op3);
30+
}
31+
32+
// CHECK-LABEL: @test_svdot_u32_x2(
33+
// CHECK-NEXT: entry:
34+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.udot.x2.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
35+
// CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
36+
//
37+
// CPP-CHECK-LABEL: @_Z17test_svdot_u32_x2u12__SVUint32_tu12__SVUint16_tu12__SVUint16_t(
38+
// CPP-CHECK-NEXT: entry:
39+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.udot.x2.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]])
40+
// CPP-CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
41+
//
42+
svuint32_t test_svdot_u32_x2(svuint32_t op1, svuint16_t op2, svuint16_t op3)
43+
{
44+
return SVE_ACLE_FUNC(svdot,_u32_u16_u16,)(op1, op2, op3);
45+
}
46+
47+
// CHECK-LABEL: @test_svdot_f32_x2(
48+
// CHECK-NEXT: entry:
49+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.fdot.x2.nxv4f32(<vscale x 4 x float> [[OP1:%.*]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
50+
// CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
51+
//
52+
// CPP-CHECK-LABEL: @_Z17test_svdot_f32_x2u13__SVFloat32_tu13__SVFloat16_tu13__SVFloat16_t(
53+
// CPP-CHECK-NEXT: entry:
54+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.fdot.x2.nxv4f32(<vscale x 4 x float> [[OP1:%.*]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]])
55+
// CPP-CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
56+
//
57+
svfloat32_t test_svdot_f32_x2(svfloat32_t op1, svfloat16_t op2, svfloat16_t op3)
58+
{
59+
return SVE_ACLE_FUNC(svdot,_f32_f16_f16,)(op1, op2, op3);
60+
}
61+
62+
63+
64+
// CHECK-LABEL: @test_svdot_lane_s32_x2(
65+
// CHECK-NEXT: entry:
66+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.sdot.lane.x2.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]], i32 3)
67+
// CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
68+
//
69+
// CPP-CHECK-LABEL: @_Z22test_svdot_lane_s32_x2u11__SVInt32_tu11__SVInt16_tu11__SVInt16_t(
70+
// CPP-CHECK-NEXT: entry:
71+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.sdot.lane.x2.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]], i32 3)
72+
// CPP-CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
73+
//
74+
svint32_t test_svdot_lane_s32_x2(svint32_t op1, svint16_t op2, svint16_t op3)
75+
{
76+
return SVE_ACLE_FUNC(svdot_lane,_s32_s16_s16,)(op1, op2, op3, 3);
77+
}
78+
79+
// CHECK-LABEL: @test_svdot_lane_u32_x2(
80+
// CHECK-NEXT: entry:
81+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.udot.lane.x2.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]], i32 3)
82+
// CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
83+
//
84+
// CPP-CHECK-LABEL: @_Z22test_svdot_lane_u32_x2u12__SVUint32_tu12__SVUint16_tu12__SVUint16_t(
85+
// CPP-CHECK-NEXT: entry:
86+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.udot.lane.x2.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]], <vscale x 8 x i16> [[OP3:%.*]], i32 3)
87+
// CPP-CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]
88+
//
89+
svuint32_t test_svdot_lane_u32_x2(svuint32_t op1, svuint16_t op2, svuint16_t op3)
90+
{
91+
return SVE_ACLE_FUNC(svdot_lane,_u32_u16_u16,)(op1, op2, op3, 3);
92+
}
93+
94+
// CHECK-LABEL: @test_svdot_lane_f32_x2(
95+
// CHECK-NEXT: entry:
96+
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.fdot.lane.x2.nxv4f32(<vscale x 4 x float> [[OP1:%.*]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]], i32 3)
97+
// CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
98+
//
99+
// CPP-CHECK-LABEL: @_Z22test_svdot_lane_f32_x2u13__SVFloat32_tu13__SVFloat16_tu13__SVFloat16_t(
100+
// CPP-CHECK-NEXT: entry:
101+
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.fdot.lane.x2.nxv4f32(<vscale x 4 x float> [[OP1:%.*]], <vscale x 8 x half> [[OP2:%.*]], <vscale x 8 x half> [[OP3:%.*]], i32 3)
102+
// CPP-CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
103+
//
104+
svfloat32_t test_svdot_lane_f32_x2(svfloat32_t op1, svfloat16_t op2, svfloat16_t op3)
105+
{
106+
return SVE_ACLE_FUNC(svdot_lane,_f32_f16_f16,)(op1, op2, op3, 3);
107+
}

‎clang/test/Sema/aarch64-sve2p1-intrinsics/acle_sve2p1_imm.cpp

Lines changed: 9 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -107,3 +107,12 @@ void test_cntp(svcount_t c) {
107107
svcntp_c14(c, 3); // expected-error {{argument should be a multiple of 2}}
108108
}
109109

110+
void test_svdot_lane_2way(svint32_t s32, svuint32_t u32, svint16_t s16, svuint16_t u16,
111+
svfloat32_t f32, svfloat16_t f16) {
112+
svdot_lane_s32_s16_s16(s32, s16, s16, 1); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
113+
svdot_lane_u32_u16_u16(u32, u16, u16, 1); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
114+
svdot_lane_f32_f16_f16(f32, f16, f16, 1); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
115+
svdot_lane_s32_s16_s16(s32, s16, s16, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
116+
svdot_lane_u32_u16_u16(u32, u16, u16, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
117+
svdot_lane_f32_f16_f16(f32, f16, f16, 4); // expected-error {{argument value 4 is outside the valid range [0, 3]}}
118+
}

0 commit comments

Comments
 (0)