annotate polly/test/CodeGen/stride_detection.ll @ 252:1f2b6ac9f198 llvm-original

LLVM16-1
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Fri, 18 Aug 2023 09:04:13 +0900
parents c4bab56944e8
children
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
150
anatofuz
parents:
diff changeset
1 ; RUN: opt %loadPolly -polly-opt-isl -polly-vectorizer=polly -polly-codegen < %s -S | FileCheck %s
anatofuz
parents:
diff changeset
2
anatofuz
parents:
diff changeset
3 ; #pragma known-parallel
anatofuz
parents:
diff changeset
4 ; for (int c0 = 0; c0 <= 31; c0 += 1)
anatofuz
parents:
diff changeset
5 ; for (int c1 = 0; c1 <= floord(nk - 1, 32); c1 += 1)
anatofuz
parents:
diff changeset
6 ; for (int c2 = 0; c2 <= 7; c2 += 1)
anatofuz
parents:
diff changeset
7 ; for (int c3 = 0; c3 <= min(31, nk - 32 * c1 - 1); c3 += 1)
anatofuz
parents:
diff changeset
8 ; #pragma simd
anatofuz
parents:
diff changeset
9 ; for (int c4 = 0; c4 <= 3; c4 += 1)
anatofuz
parents:
diff changeset
10 ; Stmt_for_body_3(32 * c0 + 4 * c2 + c4, 32 * c1 + c3);
anatofuz
parents:
diff changeset
11
anatofuz
parents:
diff changeset
12 ; CHECK: polly.stmt.for.body.3: ; preds = %polly.loop_header18
236
c4bab56944e8 LLVM 16
kono
parents: 221
diff changeset
13 ; CHECK: %_p_splat_one = load <1 x double>, <1 x double>* %_p_vec_p, align 8, !alias.scope !3, !noalias !6, !llvm.access.group !2
c4bab56944e8 LLVM 16
kono
parents: 221
diff changeset
14 ; CHECK: %_p_vec_full = load <4 x double>, <4 x double>* %vector_ptr, align 8, !alias.scope !6, !noalias !3, !llvm.access.group !2
150
anatofuz
parents:
diff changeset
15 ; CHECK: extractelement <4 x double> %addp_vec, i32 0
anatofuz
parents:
diff changeset
16 ; CHECK: extractelement <4 x double> %addp_vec, i32 1
anatofuz
parents:
diff changeset
17 ; CHECK: extractelement <4 x double> %addp_vec, i32 2
anatofuz
parents:
diff changeset
18 ; CHECK: extractelement <4 x double> %addp_vec, i32 3
236
c4bab56944e8 LLVM 16
kono
parents: 221
diff changeset
19 ; CHECK: store <4 x double> %addp_vec, <4 x double>* {{.*}}, align 8, !alias.scope !6, !noalias !3, !llvm.access.group !2
150
anatofuz
parents:
diff changeset
20
anatofuz
parents:
diff changeset
21 define void @kernel_gemm(i32 %ni, i32 %nj, i32 %nk, [1024 x double]* %C, [1024 x double]* %A) #0 {
anatofuz
parents:
diff changeset
22 entry:
anatofuz
parents:
diff changeset
23 br label %for.cond.1.preheader
anatofuz
parents:
diff changeset
24
anatofuz
parents:
diff changeset
25 for.cond.1.preheader: ; preds = %entry, %for.inc.10
anatofuz
parents:
diff changeset
26 %indvars.iv16 = phi i64 [ 0, %entry ], [ %indvars.iv.next17, %for.inc.10 ]
anatofuz
parents:
diff changeset
27 %cmp2.13 = icmp sgt i32 %nk, 0
anatofuz
parents:
diff changeset
28 br i1 %cmp2.13, label %for.body.3.lr.ph, label %for.inc.10
anatofuz
parents:
diff changeset
29
anatofuz
parents:
diff changeset
30 for.body.3.lr.ph: ; preds = %for.cond.1.preheader
anatofuz
parents:
diff changeset
31 br label %for.body.3
anatofuz
parents:
diff changeset
32
anatofuz
parents:
diff changeset
33 for.body.3: ; preds = %for.body.3.lr.ph, %for.body.3
anatofuz
parents:
diff changeset
34 %indvars.iv = phi i64 [ 0, %for.body.3.lr.ph ], [ %indvars.iv.next, %for.body.3 ]
anatofuz
parents:
diff changeset
35 %arrayidx5 = getelementptr inbounds [1024 x double], [1024 x double]* %A, i64 0, i64 %indvars.iv
anatofuz
parents:
diff changeset
36 %0 = load double, double* %arrayidx5, align 8
anatofuz
parents:
diff changeset
37 %arrayidx9 = getelementptr inbounds [1024 x double], [1024 x double]* %C, i64 0, i64 %indvars.iv16
anatofuz
parents:
diff changeset
38 %1 = load double, double* %arrayidx9, align 8
anatofuz
parents:
diff changeset
39 %add = fadd double %0, %1
anatofuz
parents:
diff changeset
40 store double %add, double* %arrayidx9, align 8
anatofuz
parents:
diff changeset
41 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
anatofuz
parents:
diff changeset
42 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
anatofuz
parents:
diff changeset
43 %exitcond = icmp ne i32 %lftr.wideiv, %nk
anatofuz
parents:
diff changeset
44 br i1 %exitcond, label %for.body.3, label %for.cond.1.for.inc.10_crit_edge
anatofuz
parents:
diff changeset
45
anatofuz
parents:
diff changeset
46 for.cond.1.for.inc.10_crit_edge: ; preds = %for.body.3
anatofuz
parents:
diff changeset
47 br label %for.inc.10
anatofuz
parents:
diff changeset
48
anatofuz
parents:
diff changeset
49 for.inc.10: ; preds = %for.cond.1.for.inc.10_crit_edge, %for.cond.1.preheader
anatofuz
parents:
diff changeset
50 %indvars.iv.next17 = add nuw nsw i64 %indvars.iv16, 1
anatofuz
parents:
diff changeset
51 %exitcond18 = icmp ne i64 %indvars.iv.next17, 1024
anatofuz
parents:
diff changeset
52 br i1 %exitcond18, label %for.cond.1.preheader, label %for.end.12
anatofuz
parents:
diff changeset
53
anatofuz
parents:
diff changeset
54 for.end.12: ; preds = %for.inc.10
anatofuz
parents:
diff changeset
55 ret void
anatofuz
parents:
diff changeset
56 }