1 ; RUN: llc -mcpu=pwr9 -mtriple=powerpc64le-unknown-unknown \
2 ; RUN: -enable-ppc-quad-precision -ppc-vsr-nums-as-vr \
3 ; RUN: -verify-machineinstrs -ppc-asm-full-reg-names < %s | FileCheck %s
5 ; Function Attrs: norecurse nounwind readnone
6 define fp128 @loadConstant() {
7 ; CHECK-LABEL: loadConstant:
8 ; CHECK: # %bb.0: # %entry
9 ; CHECK-NEXT: addis r[[REG0:[0-9]+]], r2, .LCPI0_0@toc@ha
10 ; CHECK-NEXT: addi r[[REG0]], r[[REG0]], .LCPI0_0@toc@l
11 ; CHECK-NEXT: lxvx v2, 0, r[[REG0]]
14 ret fp128 0xL00000000000000004001400000000000
17 ; Function Attrs: norecurse nounwind readnone
18 define fp128 @loadConstant2(fp128 %a, fp128 %b) {
19 ; CHECK-LABEL: loadConstant2:
20 ; CHECK: # %bb.0: # %entry
21 ; CHECK-NEXT: xsaddqp v2, v2, v3
22 ; CHECK-NEXT: addis r[[REG0:[0-9]+]], r2, .LCPI1_0@toc@ha
23 ; CHECK-NEXT: addi r[[REG0]], r[[REG0]], .LCPI1_0@toc@l
24 ; CHECK-NEXT: lxvx v[[REG1:[0-9]+]], 0, r[[REG0]]
25 ; CHECK-NEXT: xsaddqp v2, v2, v[[REG1]]
28 %add = fadd fp128 %a, %b
29 %add1 = fadd fp128 %add, 0xL00000000000000004001400000000000
33 ; Test passing float128 by value.
34 ; Function Attrs: norecurse nounwind readnone
35 define signext i32 @fp128Param(fp128 %a) {
36 ; CHECK-LABEL: fp128Param:
37 ; CHECK: # %bb.0: # %entry
38 ; CHECK-NEXT: xscvqpswz v2, v2
39 ; CHECK-NEXT: mfvsrwz r3, v2
40 ; CHECK-NEXT: extsw r3, r3
43 %conv = fptosi fp128 %a to i32
47 ; Test float128 as return value.
48 ; Function Attrs: norecurse nounwind readnone
49 define fp128 @fp128Return(fp128 %a, fp128 %b) {
50 ; CHECK-LABEL: fp128Return:
51 ; CHECK: # %bb.0: # %entry
52 ; CHECK-NEXT: xsaddqp v2, v2, v3
55 %add = fadd fp128 %a, %b
59 ; array of float128 types
60 ; Function Attrs: norecurse nounwind readonly
61 define fp128 @fp128Array(fp128* nocapture readonly %farray,
62 ; CHECK-LABEL: fp128Array:
63 ; CHECK: # %bb.0: # %entry
64 ; CHECK-NEXT: sldi r4, r4, 4
65 ; CHECK-NEXT: lxv v2, 0(r3)
66 ; CHECK-NEXT: add [[REG:r[0-9]+]], r3, r4
67 ; CHECK-NEXT: lxv v3, -16([[REG]])
68 ; CHECK-NEXT: xsaddqp v2, v2, v3
70 i32 signext %loopcnt, fp128* nocapture readnone %sum) {
72 %0 = load fp128, fp128* %farray, align 16
73 %sub = add nsw i32 %loopcnt, -1
74 %idxprom = sext i32 %sub to i64
75 %arrayidx1 = getelementptr inbounds fp128, fp128* %farray, i64 %idxprom
76 %1 = load fp128, fp128* %arrayidx1, align 16
77 %add = fadd fp128 %0, %1
81 ; Up to 12 qualified floating-point arguments can be passed in v2-v13.
82 ; Function to test passing 13 float128 parameters.
83 ; Function Attrs: norecurse nounwind readnone
84 define fp128 @maxVecParam(fp128 %p1, fp128 %p2, fp128 %p3, fp128 %p4, fp128 %p5,
85 ; CHECK-LABEL: maxVecParam:
86 ; CHECK: # %bb.0: # %entry
87 ; CHECK-NEXT: xsaddqp v2, v2, v3
88 ; CHECK-NEXT: lxv v[[REG0:[0-9]+]], 224(r1)
89 ; CHECK-NEXT: xsaddqp v2, v2, v4
90 ; CHECK-NEXT: xsaddqp v2, v2, v5
91 ; CHECK-NEXT: xsaddqp v2, v2, v6
92 ; CHECK-NEXT: xsaddqp v2, v2, v7
93 ; CHECK-NEXT: xsaddqp v2, v2, v8
94 ; CHECK-NEXT: xsaddqp v2, v2, v9
95 ; CHECK-NEXT: xsaddqp v2, v2, v10
96 ; CHECK-NEXT: xsaddqp v2, v2, v11
97 ; CHECK-NEXT: xsaddqp v2, v2, v12
98 ; CHECK-NEXT: xsaddqp v2, v2, v13
99 ; CHECK-NEXT: xssubqp v2, v2, v[[REG0]]
101 fp128 %p6, fp128 %p7, fp128 %p8, fp128 %p9, fp128 %p10,
102 fp128 %p11, fp128 %p12, fp128 %p13) {
104 %add = fadd fp128 %p1, %p2
105 %add1 = fadd fp128 %add, %p3
106 %add2 = fadd fp128 %add1, %p4
107 %add3 = fadd fp128 %add2, %p5
108 %add4 = fadd fp128 %add3, %p6
109 %add5 = fadd fp128 %add4, %p7
110 %add6 = fadd fp128 %add5, %p8
111 %add7 = fadd fp128 %add6, %p9
112 %add8 = fadd fp128 %add7, %p10
113 %add9 = fadd fp128 %add8, %p11
114 %add10 = fadd fp128 %add9, %p12
115 %sub = fsub fp128 %add10, %p13
119 ; Passing a mix of float128 and other type parameters.
120 ; Function Attrs: norecurse nounwind readnone
121 define fp128 @mixParam_01(fp128 %a, i32 signext %i, fp128 %b) {
122 ; CHECK-LABEL: mixParam_01:
123 ; CHECK: # %bb.0: # %entry
124 ; CHECK-DAG: mtvsrwa [[REG1:v[0-9]+]], r5
125 ; CHECK-DAG: xsaddqp v2, v2, v3
126 ; CHECK-NEXT: xscvsdqp v[[REG0:[0-9]+]], [[REG1]]
127 ; CHECK-NEXT: xsaddqp v2, v2, v[[REG0]]
130 %add = fadd fp128 %a, %b
131 %conv = sitofp i32 %i to fp128
132 %add1 = fadd fp128 %add, %conv
135 ; Function Attrs: norecurse nounwind readnone
136 define fastcc fp128 @mixParam_01f(fp128 %a, i32 signext %i, fp128 %b) {
137 ; CHECK-LABEL: mixParam_01f:
138 ; CHECK: # %bb.0: # %entry
139 ; CHECK-DAG: mtvsrwa v[[REG0:[0-9]+]], r3
140 ; CHECK-DAG: xsaddqp v2, v2, v3
141 ; CHECK-NEXT: xscvsdqp v[[REG1:[0-9]+]], v[[REG0]]
142 ; CHECK-NEXT: xsaddqp v2, v2, v[[REG1]]
145 %add = fadd fp128 %a, %b
146 %conv = sitofp i32 %i to fp128
147 %add1 = fadd fp128 %add, %conv
151 ; Function Attrs: norecurse nounwind
152 define fp128 @mixParam_02(fp128 %p1, double %p2, i64* nocapture %p3,
153 ; CHECK-LABEL: mixParam_02:
154 ; CHECK: # %bb.0: # %entry
155 ; CHECK: lwz r3, 96(r1)
156 ; CHECK: add r4, r7, r9
157 ; CHECK: add r4, r4, r10
158 ; CHECK: add r3, r4, r3
159 ; CHECK: clrldi r3, r3, 32
160 ; CHECK: std r3, 0(r6)
161 ; CHECK: lxv v[[REG1:[0-9]+]], 0(r8)
162 ; CHECK: xscpsgndp v[[REG0:[0-9]+]], f1, f1
163 ; CHECK: xscvdpqp v[[REG0]], v[[REG0]]
164 ; CHECK: xsaddqp v2, v[[REG1]], v2
165 ; CHECK: xsaddqp v2, v2, v3
167 i16 signext %p4, fp128* nocapture readonly %p5,
168 i32 signext %p6, i8 zeroext %p7, i32 zeroext %p8) {
170 %conv = sext i16 %p4 to i32
171 %add = add nsw i32 %conv, %p6
172 %conv1 = zext i8 %p7 to i32
173 %add2 = add nsw i32 %add, %conv1
174 %add3 = add i32 %add2, %p8
175 %conv4 = zext i32 %add3 to i64
176 store i64 %conv4, i64* %p3, align 8
177 %0 = load fp128, fp128* %p5, align 16
178 %add5 = fadd fp128 %0, %p1
179 %conv6 = fpext double %p2 to fp128
180 %add7 = fadd fp128 %add5, %conv6
184 ; Function Attrs: norecurse nounwind
185 define fastcc fp128 @mixParam_02f(fp128 %p1, double %p2, i64* nocapture %p3,
186 ; CHECK-LABEL: mixParam_02f:
187 ; CHECK: # %bb.0: # %entry
188 ; CHECK-NEXT: add r4, r4, r6
189 ; CHECK-NEXT: add r4, r4, r7
190 ; CHECK-NEXT: add r4, r4, r8
191 ; CHECK-NEXT: clrldi r4, r4, 32
192 ; CHECK-DAG: std r4, 0(r3)
193 ; CHECK-DAG: lxv v[[REG1:[0-9]+]], 0(r5)
194 ; CHECK-NEXT: xscpsgndp v[[REG0:[0-9]+]], f1, f1
195 ; CHECK-NEXT: xscvdpqp v[[REG0]], v[[REG0]]
196 ; CHECK-NEXT: xsaddqp v2, v[[REG1]], v2
197 ; CHECK-NEXT: xsaddqp v2, v2, v[[REG0]]
199 i16 signext %p4, fp128* nocapture readonly %p5,
200 i32 signext %p6, i8 zeroext %p7, i32 zeroext %p8) {
202 %conv = sext i16 %p4 to i32
203 %add = add nsw i32 %conv, %p6
204 %conv1 = zext i8 %p7 to i32
205 %add2 = add nsw i32 %add, %conv1
206 %add3 = add i32 %add2, %p8
207 %conv4 = zext i32 %add3 to i64
208 store i64 %conv4, i64* %p3, align 8
209 %0 = load fp128, fp128* %p5, align 16
210 %add5 = fadd fp128 %0, %p1
211 %conv6 = fpext double %p2 to fp128
212 %add7 = fadd fp128 %add5, %conv6
216 ; Passing a mix of float128 and vector parameters.
217 ; Function Attrs: norecurse nounwind
218 define void @mixParam_03(fp128 %f1, double* nocapture %d1, <4 x i32> %vec1,
219 ; CHECK-LABEL: mixParam_03:
220 ; CHECK: # %bb.0: # %entry
221 ; CHECK-DAG: ld r3, 104(r1)
222 ; CHECK-DAG: stxv v2, 0(r9)
223 ; CHECK: stxvx v3, 0, r3
224 ; CHECK: mtvsrwa v[[REG2:[0-9]+]], r10
225 ; CHECK-DAG: xscvsdqp v[[REG1:[0-9]+]], v[[REG2]]
226 ; CHECK-DAG: lxv v2, 0(r9)
227 ; CHECK-NEXT: xsaddqp v2, v2, v[[REG1]]
228 ; CHECK-NEXT: xscvqpdp v2, v2
229 ; CHECK-NEXT: stxsd v2, 0(r5)
231 fp128* nocapture %f2, i32 signext %i1, i8 zeroext %c1,
232 <4 x i32>* nocapture %vec2) {
234 store fp128 %f1, fp128* %f2, align 16
235 store <4 x i32> %vec1, <4 x i32>* %vec2, align 16
236 %0 = load fp128, fp128* %f2, align 16
237 %conv = sitofp i32 %i1 to fp128
238 %add = fadd fp128 %0, %conv
239 %conv1 = fptrunc fp128 %add to double
240 store double %conv1, double* %d1, align 8
244 ; Function Attrs: norecurse nounwind
245 define fastcc void @mixParam_03f(fp128 %f1, double* nocapture %d1, <4 x i32> %vec1,
246 ; CHECK-LABEL: mixParam_03f:
247 ; CHECK: # %bb.0: # %entry
248 ; CHECK-DAG: mtvsrwa v[[REG0:[0-9]+]], r5
249 ; CHECK-DAG: stxv v[[REG1:[0-9]+]], 0(r4)
250 ; CHECK-DAG: stxv v[[REG2:[0-9]+]], 0(r7)
251 ; CHECK-DAG: lxv v[[REG1]], 0(r4)
252 ; CHECK-NEXT: xscvsdqp v[[REG3:[0-9]+]], v[[REG0]]
253 ; CHECK-NEXT: xsaddqp v[[REG4:[0-9]+]], v[[REG1]], v[[REG3]]
254 ; CHECK-NEXT: xscvqpdp v2, v[[REG4]]
255 ; CHECK-NEXT: stxsd v2, 0(r3)
257 fp128* nocapture %f2, i32 signext %i1, i8 zeroext %c1,
258 <4 x i32>* nocapture %vec2) {
260 store fp128 %f1, fp128* %f2, align 16
261 store <4 x i32> %vec1, <4 x i32>* %vec2, align 16
262 %0 = load fp128, fp128* %f2, align 16
263 %conv = sitofp i32 %i1 to fp128
264 %add = fadd fp128 %0, %conv
265 %conv1 = fptrunc fp128 %add to double
266 store double %conv1, double* %d1, align 8