Skip to content

Commit 6e751b0

Browse files
CarolineConcattoCarolineConcatto
authored
[Clang][LLVM][AArch64]Add support for svrint{32|64}{z|x} intrinsics (#169661)
This patch add supports in Clang for these assembly intrinsics: FRINT32X FRINT32Z FRINT64X FRINT64Z By adding support for these intrinsics: // Variant is available for _f64 svfloat32_t svrint32x[_f32]_z(svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint32x[_f32]_x(svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint32x[_f32]_m(svfloat32_t inactive, svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint32z[_f32]_z(svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint32z[_f32]_x(svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint32z[_f32]_m(svfloat32_t inactive, svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint64x[_f32]_z(svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint64x[_f32]_x(svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint64x[_f32]_m(svfloat32_t inactive, svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint64z[_f32]_z(svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint64z[_f32]_x(svbool_t pg, svfloat32_t zn); // Variant is available for _f64 svfloat32_t svrint64z[_f32]_m(svfloat32_t inactive, svbool_t pg, svfloat32_t zn); ``` according to the ACLE[1] [1]ARM-software/acle#412
1 parent 8fc3f6d commit 6e751b0

File tree

8 files changed

+932
-10
lines changed

8 files changed

+932
-10
lines changed

‎clang/include/clang/Basic/arm_sve.td‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -778,6 +778,13 @@ defm SVRINTX : SInstZPZ<"svrintx", "hfd", "aarch64_sve_frintx">;
778778
defm SVRINTZ : SInstZPZ<"svrintz", "hfd", "aarch64_sve_frintz">;
779779
defm SVSQRT : SInstZPZ<"svsqrt", "hfd", "aarch64_sve_fsqrt">;
780780

781+
let SVETargetGuard = "sve2p2|sme2p2", SMETargetGuard = "sve2p2|sme2p2" in {
782+
defm SVRINT32X : SInstZPZ<"svrint32x", "fd", "aarch64_sve_frint32x">;
783+
defm SVRINT32Z : SInstZPZ<"svrint32z", "fd", "aarch64_sve_frint32z">;
784+
defm SVRINT64X : SInstZPZ<"svrint64x", "fd", "aarch64_sve_frint64x">;
785+
defm SVRINT64Z : SInstZPZ<"svrint64z", "fd", "aarch64_sve_frint64z">;
786+
}
787+
781788
let SMETargetGuard = "sme2,ssve-fexpa" in {
782789
def SVEXPA : SInst<"svexpa[_{d}]", "du", "hfd", MergeNone, "aarch64_sve_fexpa_x", [VerifyRuntimeMode]>;
783790
}

‎clang/test/CodeGen/AArch64/sve2p2-intrinsics/acle_sve_rintx.c‎

Lines changed: 445 additions & 0 deletions
Large diffs are not rendered by default.

‎clang/test/Sema/AArch64/arm_sve_feature_dependent_sve_AND_LP_sve2p2_OR_sme2p2_RP___sme_AND_LP_sve2p2_OR_sme2p2_RP.c‎

Lines changed: 144 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -36,6 +36,54 @@ void test(void) {
3636
svlastp_b16(svbool_t_val, svbool_t_val);
3737
svlastp_b32(svbool_t_val, svbool_t_val);
3838
svlastp_b64(svbool_t_val, svbool_t_val);
39+
svrint32x_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
40+
svrint32x_f32_x(svbool_t_val, svfloat32_t_val);
41+
svrint32x_f32_z(svbool_t_val, svfloat32_t_val);
42+
svrint32x_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
43+
svrint32x_f64_x(svbool_t_val, svfloat64_t_val);
44+
svrint32x_f64_z(svbool_t_val, svfloat64_t_val);
45+
svrint32x_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
46+
svrint32x_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
47+
svrint32x_x(svbool_t_val, svfloat32_t_val);
48+
svrint32x_x(svbool_t_val, svfloat64_t_val);
49+
svrint32x_z(svbool_t_val, svfloat32_t_val);
50+
svrint32x_z(svbool_t_val, svfloat64_t_val);
51+
svrint32z_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
52+
svrint32z_f32_x(svbool_t_val, svfloat32_t_val);
53+
svrint32z_f32_z(svbool_t_val, svfloat32_t_val);
54+
svrint32z_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
55+
svrint32z_f64_x(svbool_t_val, svfloat64_t_val);
56+
svrint32z_f64_z(svbool_t_val, svfloat64_t_val);
57+
svrint32z_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
58+
svrint32z_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
59+
svrint32z_x(svbool_t_val, svfloat32_t_val);
60+
svrint32z_x(svbool_t_val, svfloat64_t_val);
61+
svrint32z_z(svbool_t_val, svfloat32_t_val);
62+
svrint32z_z(svbool_t_val, svfloat64_t_val);
63+
svrint64x_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
64+
svrint64x_f32_x(svbool_t_val, svfloat32_t_val);
65+
svrint64x_f32_z(svbool_t_val, svfloat32_t_val);
66+
svrint64x_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
67+
svrint64x_f64_x(svbool_t_val, svfloat64_t_val);
68+
svrint64x_f64_z(svbool_t_val, svfloat64_t_val);
69+
svrint64x_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
70+
svrint64x_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
71+
svrint64x_x(svbool_t_val, svfloat32_t_val);
72+
svrint64x_x(svbool_t_val, svfloat64_t_val);
73+
svrint64x_z(svbool_t_val, svfloat32_t_val);
74+
svrint64x_z(svbool_t_val, svfloat64_t_val);
75+
svrint64z_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
76+
svrint64z_f32_x(svbool_t_val, svfloat32_t_val);
77+
svrint64z_f32_z(svbool_t_val, svfloat32_t_val);
78+
svrint64z_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
79+
svrint64z_f64_x(svbool_t_val, svfloat64_t_val);
80+
svrint64z_f64_z(svbool_t_val, svfloat64_t_val);
81+
svrint64z_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
82+
svrint64z_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
83+
svrint64z_x(svbool_t_val, svfloat32_t_val);
84+
svrint64z_x(svbool_t_val, svfloat64_t_val);
85+
svrint64z_z(svbool_t_val, svfloat32_t_val);
86+
svrint64z_z(svbool_t_val, svfloat64_t_val);
3987
}
4088

4189
void test_streaming(void) __arm_streaming{
@@ -65,6 +113,54 @@ void test_streaming(void) __arm_streaming{
65113
svlastp_b16(svbool_t_val, svbool_t_val);
66114
svlastp_b32(svbool_t_val, svbool_t_val);
67115
svlastp_b64(svbool_t_val, svbool_t_val);
116+
svrint32x_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
117+
svrint32x_f32_x(svbool_t_val, svfloat32_t_val);
118+
svrint32x_f32_z(svbool_t_val, svfloat32_t_val);
119+
svrint32x_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
120+
svrint32x_f64_x(svbool_t_val, svfloat64_t_val);
121+
svrint32x_f64_z(svbool_t_val, svfloat64_t_val);
122+
svrint32x_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
123+
svrint32x_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
124+
svrint32x_x(svbool_t_val, svfloat32_t_val);
125+
svrint32x_x(svbool_t_val, svfloat64_t_val);
126+
svrint32x_z(svbool_t_val, svfloat32_t_val);
127+
svrint32x_z(svbool_t_val, svfloat64_t_val);
128+
svrint32z_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
129+
svrint32z_f32_x(svbool_t_val, svfloat32_t_val);
130+
svrint32z_f32_z(svbool_t_val, svfloat32_t_val);
131+
svrint32z_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
132+
svrint32z_f64_x(svbool_t_val, svfloat64_t_val);
133+
svrint32z_f64_z(svbool_t_val, svfloat64_t_val);
134+
svrint32z_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
135+
svrint32z_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
136+
svrint32z_x(svbool_t_val, svfloat32_t_val);
137+
svrint32z_x(svbool_t_val, svfloat64_t_val);
138+
svrint32z_z(svbool_t_val, svfloat32_t_val);
139+
svrint32z_z(svbool_t_val, svfloat64_t_val);
140+
svrint64x_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
141+
svrint64x_f32_x(svbool_t_val, svfloat32_t_val);
142+
svrint64x_f32_z(svbool_t_val, svfloat32_t_val);
143+
svrint64x_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
144+
svrint64x_f64_x(svbool_t_val, svfloat64_t_val);
145+
svrint64x_f64_z(svbool_t_val, svfloat64_t_val);
146+
svrint64x_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
147+
svrint64x_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
148+
svrint64x_x(svbool_t_val, svfloat32_t_val);
149+
svrint64x_x(svbool_t_val, svfloat64_t_val);
150+
svrint64x_z(svbool_t_val, svfloat32_t_val);
151+
svrint64x_z(svbool_t_val, svfloat64_t_val);
152+
svrint64z_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
153+
svrint64z_f32_x(svbool_t_val, svfloat32_t_val);
154+
svrint64z_f32_z(svbool_t_val, svfloat32_t_val);
155+
svrint64z_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
156+
svrint64z_f64_x(svbool_t_val, svfloat64_t_val);
157+
svrint64z_f64_z(svbool_t_val, svfloat64_t_val);
158+
svrint64z_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
159+
svrint64z_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
160+
svrint64z_x(svbool_t_val, svfloat32_t_val);
161+
svrint64z_x(svbool_t_val, svfloat64_t_val);
162+
svrint64z_z(svbool_t_val, svfloat32_t_val);
163+
svrint64z_z(svbool_t_val, svfloat64_t_val);
68164
}
69165

70166
void test_streaming_compatible(void) __arm_streaming_compatible{
@@ -94,4 +190,52 @@ void test_streaming_compatible(void) __arm_streaming_compatible{
94190
svlastp_b16(svbool_t_val, svbool_t_val);
95191
svlastp_b32(svbool_t_val, svbool_t_val);
96192
svlastp_b64(svbool_t_val, svbool_t_val);
193+
svrint32x_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
194+
svrint32x_f32_x(svbool_t_val, svfloat32_t_val);
195+
svrint32x_f32_z(svbool_t_val, svfloat32_t_val);
196+
svrint32x_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
197+
svrint32x_f64_x(svbool_t_val, svfloat64_t_val);
198+
svrint32x_f64_z(svbool_t_val, svfloat64_t_val);
199+
svrint32x_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
200+
svrint32x_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
201+
svrint32x_x(svbool_t_val, svfloat32_t_val);
202+
svrint32x_x(svbool_t_val, svfloat64_t_val);
203+
svrint32x_z(svbool_t_val, svfloat32_t_val);
204+
svrint32x_z(svbool_t_val, svfloat64_t_val);
205+
svrint32z_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
206+
svrint32z_f32_x(svbool_t_val, svfloat32_t_val);
207+
svrint32z_f32_z(svbool_t_val, svfloat32_t_val);
208+
svrint32z_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
209+
svrint32z_f64_x(svbool_t_val, svfloat64_t_val);
210+
svrint32z_f64_z(svbool_t_val, svfloat64_t_val);
211+
svrint32z_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
212+
svrint32z_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
213+
svrint32z_x(svbool_t_val, svfloat32_t_val);
214+
svrint32z_x(svbool_t_val, svfloat64_t_val);
215+
svrint32z_z(svbool_t_val, svfloat32_t_val);
216+
svrint32z_z(svbool_t_val, svfloat64_t_val);
217+
svrint64x_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
218+
svrint64x_f32_x(svbool_t_val, svfloat32_t_val);
219+
svrint64x_f32_z(svbool_t_val, svfloat32_t_val);
220+
svrint64x_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
221+
svrint64x_f64_x(svbool_t_val, svfloat64_t_val);
222+
svrint64x_f64_z(svbool_t_val, svfloat64_t_val);
223+
svrint64x_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
224+
svrint64x_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
225+
svrint64x_x(svbool_t_val, svfloat32_t_val);
226+
svrint64x_x(svbool_t_val, svfloat64_t_val);
227+
svrint64x_z(svbool_t_val, svfloat32_t_val);
228+
svrint64x_z(svbool_t_val, svfloat64_t_val);
229+
svrint64z_f32_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
230+
svrint64z_f32_x(svbool_t_val, svfloat32_t_val);
231+
svrint64z_f32_z(svbool_t_val, svfloat32_t_val);
232+
svrint64z_f64_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
233+
svrint64z_f64_x(svbool_t_val, svfloat64_t_val);
234+
svrint64z_f64_z(svbool_t_val, svfloat64_t_val);
235+
svrint64z_m(svfloat32_t_val, svbool_t_val, svfloat32_t_val);
236+
svrint64z_m(svfloat64_t_val, svbool_t_val, svfloat64_t_val);
237+
svrint64z_x(svbool_t_val, svfloat32_t_val);
238+
svrint64z_x(svbool_t_val, svfloat64_t_val);
239+
svrint64z_z(svbool_t_val, svfloat32_t_val);
240+
svrint64z_z(svbool_t_val, svfloat64_t_val);
97241
}

‎llvm/include/llvm/IR/IntrinsicsAArch64.td‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2155,6 +2155,10 @@ def int_aarch64_sve_frintn : AdvSIMD_Merged1VectorArg_Intrinsic;
21552155
def int_aarch64_sve_frintp : AdvSIMD_Merged1VectorArg_Intrinsic;
21562156
def int_aarch64_sve_frintx : AdvSIMD_Merged1VectorArg_Intrinsic;
21572157
def int_aarch64_sve_frintz : AdvSIMD_Merged1VectorArg_Intrinsic;
2158+
def int_aarch64_sve_frint32x : AdvSIMD_Merged1VectorArg_Intrinsic;
2159+
def int_aarch64_sve_frint32z : AdvSIMD_Merged1VectorArg_Intrinsic;
2160+
def int_aarch64_sve_frint64x : AdvSIMD_Merged1VectorArg_Intrinsic;
2161+
def int_aarch64_sve_frint64z : AdvSIMD_Merged1VectorArg_Intrinsic;
21582162
def int_aarch64_sve_frsqrte_x : AdvSIMD_1VectorArg_Intrinsic;
21592163
def int_aarch64_sve_frsqrts_x : AdvSIMD_2VectorArg_Intrinsic;
21602164
def int_aarch64_sve_fscale : AdvSIMD_SVE_SCALE_Intrinsic;

‎llvm/lib/Target/AArch64/AArch64ISelLowering.cpp‎

Lines changed: 20 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -271,9 +271,13 @@ static bool isMergePassthruOpcode(unsigned Opc) {
271271
case AArch64ISD::FFLOOR_MERGE_PASSTHRU:
272272
case AArch64ISD::FNEARBYINT_MERGE_PASSTHRU:
273273
case AArch64ISD::FRINT_MERGE_PASSTHRU:
274+
case AArch64ISD::FRINT32_MERGE_PASSTHRU:
275+
case AArch64ISD::FRINT64_MERGE_PASSTHRU:
274276
case AArch64ISD::FROUND_MERGE_PASSTHRU:
275277
case AArch64ISD::FROUNDEVEN_MERGE_PASSTHRU:
276278
case AArch64ISD::FTRUNC_MERGE_PASSTHRU:
279+
case AArch64ISD::FTRUNC32_MERGE_PASSTHRU:
280+
case AArch64ISD::FTRUNC64_MERGE_PASSTHRU:
277281
case AArch64ISD::FP_ROUND_MERGE_PASSTHRU:
278282
case AArch64ISD::FP_EXTEND_MERGE_PASSTHRU:
279283
case AArch64ISD::SINT_TO_FP_MERGE_PASSTHRU:
@@ -6610,6 +6614,14 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
66106614
case Intrinsic::aarch64_sve_frintx:
66116615
return DAG.getNode(AArch64ISD::FRINT_MERGE_PASSTHRU, DL, Op.getValueType(),
66126616
Op.getOperand(2), Op.getOperand(3), Op.getOperand(1));
6617+
case Intrinsic::aarch64_sve_frint32x:
6618+
return DAG.getNode(AArch64ISD::FRINT32_MERGE_PASSTHRU, DL,
6619+
Op.getValueType(), Op.getOperand(2), Op.getOperand(3),
6620+
Op.getOperand(1));
6621+
case Intrinsic::aarch64_sve_frint64x:
6622+
return DAG.getNode(AArch64ISD::FRINT64_MERGE_PASSTHRU, DL,
6623+
Op.getValueType(), Op.getOperand(2), Op.getOperand(3),
6624+
Op.getOperand(1));
66136625
case Intrinsic::aarch64_sve_frinta:
66146626
return DAG.getNode(AArch64ISD::FROUND_MERGE_PASSTHRU, DL, Op.getValueType(),
66156627
Op.getOperand(2), Op.getOperand(3), Op.getOperand(1));
@@ -6620,6 +6632,14 @@ SDValue AArch64TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op,
66206632
case Intrinsic::aarch64_sve_frintz:
66216633
return DAG.getNode(AArch64ISD::FTRUNC_MERGE_PASSTHRU, DL, Op.getValueType(),
66226634
Op.getOperand(2), Op.getOperand(3), Op.getOperand(1));
6635+
case Intrinsic::aarch64_sve_frint32z:
6636+
return DAG.getNode(AArch64ISD::FTRUNC32_MERGE_PASSTHRU, DL,
6637+
Op.getValueType(), Op.getOperand(2), Op.getOperand(3),
6638+
Op.getOperand(1));
6639+
case Intrinsic::aarch64_sve_frint64z:
6640+
return DAG.getNode(AArch64ISD::FTRUNC64_MERGE_PASSTHRU, DL,
6641+
Op.getValueType(), Op.getOperand(2), Op.getOperand(3),
6642+
Op.getOperand(1));
66236643
case Intrinsic::aarch64_sve_ucvtf:
66246644
return DAG.getNode(AArch64ISD::UINT_TO_FP_MERGE_PASSTHRU, DL,
66256645
Op.getValueType(), Op.getOperand(2), Op.getOperand(3),

‎llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td‎

Lines changed: 12 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -290,9 +290,13 @@ def AArch64frintp_mt : SDNode<"AArch64ISD::FCEIL_MERGE_PASSTHRU", SDT_AArch64Ari
290290
def AArch64frintm_mt : SDNode<"AArch64ISD::FFLOOR_MERGE_PASSTHRU", SDT_AArch64Arith>;
291291
def AArch64frinti_mt : SDNode<"AArch64ISD::FNEARBYINT_MERGE_PASSTHRU", SDT_AArch64Arith>;
292292
def AArch64frintx_mt : SDNode<"AArch64ISD::FRINT_MERGE_PASSTHRU", SDT_AArch64Arith>;
293+
def AArch64frint32x_mt : SDNode<"AArch64ISD::FRINT32_MERGE_PASSTHRU", SDT_AArch64Arith>;
294+
def AArch64frint64x_mt : SDNode<"AArch64ISD::FRINT64_MERGE_PASSTHRU", SDT_AArch64Arith>;
293295
def AArch64frinta_mt : SDNode<"AArch64ISD::FROUND_MERGE_PASSTHRU", SDT_AArch64Arith>;
294296
def AArch64frintn_mt : SDNode<"AArch64ISD::FROUNDEVEN_MERGE_PASSTHRU", SDT_AArch64Arith>;
295297
def AArch64frintz_mt : SDNode<"AArch64ISD::FTRUNC_MERGE_PASSTHRU", SDT_AArch64Arith>;
298+
def AArch64frint32z_mt : SDNode<"AArch64ISD::FTRUNC32_MERGE_PASSTHRU", SDT_AArch64Arith>;
299+
def AArch64frint64z_mt : SDNode<"AArch64ISD::FTRUNC64_MERGE_PASSTHRU", SDT_AArch64Arith>;
296300
def AArch64fsqrt_mt : SDNode<"AArch64ISD::FSQRT_MERGE_PASSTHRU", SDT_AArch64Arith>;
297301
def AArch64frecpx_mt : SDNode<"AArch64ISD::FRECPX_MERGE_PASSTHRU", SDT_AArch64Arith>;
298302
def AArch64rbit_mt : SDNode<"AArch64ISD::BITREVERSE_MERGE_PASSTHRU", SDT_AArch64Arith>;
@@ -4611,15 +4615,15 @@ let Predicates = [HasSVE2p2_or_SME2p2] in {
46114615

46124616
// Floating point round to integral fp value in integer size range
46134617
// Merging
4614-
defm FRINT32Z_ZPmZ : sve_fp_2op_p_zd_frint<0b00, "frint32z">;
4615-
defm FRINT32X_ZPmZ : sve_fp_2op_p_zd_frint<0b01, "frint32x">;
4616-
defm FRINT64X_ZPmZ : sve_fp_2op_p_zd_frint<0b10, "frint64z">;
4617-
defm FRINT64Z_ZPmZ : sve_fp_2op_p_zd_frint<0b11, "frint64x">;
4618+
defm FRINT32Z_ZPmZ : sve_fp_2op_p_zd_frint<0b00, "frint32z", AArch64frint32z_mt>;
4619+
defm FRINT32X_ZPmZ : sve_fp_2op_p_zd_frint<0b01, "frint32x", AArch64frint32x_mt>;
4620+
defm FRINT64Z_ZPmZ : sve_fp_2op_p_zd_frint<0b10, "frint64z", AArch64frint64z_mt>;
4621+
defm FRINT64X_ZPmZ : sve_fp_2op_p_zd_frint<0b11, "frint64x", AArch64frint64x_mt>;
46184622
// Zeroing
4619-
defm FRINT32Z_ZPzZ : sve_fp_z2op_p_zd_frint<0b00, "frint32z">;
4620-
defm FRINT32X_ZPzZ : sve_fp_z2op_p_zd_frint<0b01, "frint32x">;
4621-
defm FRINT64Z_ZPzZ : sve_fp_z2op_p_zd_frint<0b10, "frint64z">;
4622-
defm FRINT64X_ZPzZ : sve_fp_z2op_p_zd_frint<0b11, "frint64x">;
4623+
defm FRINT32Z_ZPzZ : sve_fp_z2op_p_zd_frint<0b00, "frint32z", AArch64frint32z_mt>;
4624+
defm FRINT32X_ZPzZ : sve_fp_z2op_p_zd_frint<0b01, "frint32x", AArch64frint32x_mt>;
4625+
defm FRINT64Z_ZPzZ : sve_fp_z2op_p_zd_frint<0b10, "frint64z", AArch64frint64z_mt>;
4626+
defm FRINT64X_ZPzZ : sve_fp_z2op_p_zd_frint<0b11, "frint64x", AArch64frint64x_mt>;
46234627

46244628
// Floating-point round to integral fp value, zeroing predicate
46254629
defm FRINTN_ZPzZ : sve_fp_z2op_p_zd_hsd<0b00000, "frintn", AArch64frintn_mt>;

‎llvm/lib/Target/AArch64/SVEInstrFormats.td‎

Lines changed: 17 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -3249,9 +3249,20 @@ multiclass sve2_fp_convert_down_odd_rounding<string asm, string op, SDPatternOpe
32493249
def : SVE_1_Op_Passthru_Pat<nxv2f32, ir_op, nxv2i1, nxv2f64, !cast<Instruction>(NAME # _DtoS)>;
32503250
}
32513251

3252-
multiclass sve_fp_2op_p_zd_frint<bits<2> opc, string asm> {
3252+
multiclass sve_fp_2op_p_zd_frint<bits<2> opc, string asm, SDPatternOperator op = null_frag> {
32533253
def _S : sve_fp_2op_p_zd<{ 0b0010, opc{1}, 0, opc{0} }, asm, ZPR32, ZPR32, ElementSizeS>;
32543254
def _D : sve_fp_2op_p_zd<{ 0b0010, opc{1}, 1, opc{0} }, asm, ZPR64, ZPR64, ElementSizeD>;
3255+
3256+
def : SVE_1_Op_Passthru_Pat<nxv4f32, op, nxv4i1, nxv4f32, !cast<Instruction>(NAME # _S)>;
3257+
def : SVE_1_Op_Passthru_Pat<nxv2f32, op, nxv2i1, nxv2f32, !cast<Instruction>(NAME # _S)>;
3258+
def : SVE_1_Op_Passthru_Pat<nxv2f64, op, nxv2i1, nxv2f64, !cast<Instruction>(NAME # _D)>;
3259+
3260+
def _S_UNDEF : PredOneOpPassthruPseudo<NAME # _S, ZPR32>;
3261+
def _D_UNDEF : PredOneOpPassthruPseudo<NAME # _D, ZPR64>;
3262+
3263+
defm : SVE_1_Op_PassthruUndef_Pat<nxv4f32, op, nxv4i1, nxv4f32, !cast<Instruction>(NAME # _S_UNDEF)>;
3264+
defm : SVE_1_Op_PassthruUndef_Pat<nxv2f32, op, nxv2i1, nxv2f32, !cast<Instruction>(NAME # _S_UNDEF)>;
3265+
defm : SVE_1_Op_PassthruUndef_Pat<nxv2f64, op, nxv2i1, nxv2f64, !cast<Instruction>(NAME # _D_UNDEF)>;
32553266
}
32563267

32573268
//===----------------------------------------------------------------------===//
@@ -3334,9 +3345,13 @@ multiclass sve_fp_z2op_p_zd_hsd<bits<5> opc, string asm, SDPatternOperator op> {
33343345
defm : SVE_1_Op_PassthruUndefZero_Pat<nxv2f64, op, nxv2i1, nxv2f64, !cast<Instruction>(NAME # _D)>;
33353346
}
33363347

3337-
multiclass sve_fp_z2op_p_zd_frint<bits<2> opc, string asm> {
3348+
multiclass sve_fp_z2op_p_zd_frint<bits<2> opc, string asm, SDPatternOperator op = null_frag> {
33383349
def _S : sve_fp_z2op_p_zd<{ 0b0010, opc{1}, 0, opc{0} }, asm, ZPR32, ZPR32>;
33393350
def _D : sve_fp_z2op_p_zd<{ 0b0010, opc{1}, 1, opc{0} }, asm, ZPR64, ZPR64>;
3351+
3352+
defm : SVE_1_Op_PassthruUndefZero_Pat<nxv4f32, op, nxv4i1, nxv4f32, !cast<Instruction>(NAME # _S)>;
3353+
defm : SVE_1_Op_PassthruUndefZero_Pat<nxv2f32, op, nxv2i1, nxv2f32, !cast<Instruction>(NAME # _S)>;
3354+
defm : SVE_1_Op_PassthruUndefZero_Pat<nxv2f64, op, nxv2i1, nxv2f64, !cast<Instruction>(NAME # _D)>;
33403355
}
33413356

33423357
multiclass sve_fp_z2op_p_zd_bfcvt<string asm, SDPatternOperator op> {

0 commit comments

Comments
 (0)