annotate polly/test/ScheduleOptimizer/pattern-matching-based-opts_14.ll @ 236:c4bab56944e8 llvm-original

LLVM 16
author kono
date Wed, 09 Nov 2022 17:45:10 +0900
parents 1d019706d866
children 1f2b6ac9f198
Ignore whitespace changes - Everywhere: Within whitespace: At end of lines:
rev   line source
150
anatofuz
parents:
diff changeset
1 ; RUN: opt %loadPolly -polly-import-jscop -polly-opt-isl \
anatofuz
parents:
diff changeset
2 ; RUN: -polly-target-throughput-vector-fma=1 \
anatofuz
parents:
diff changeset
3 ; RUN: -polly-target-latency-vector-fma=8 \
anatofuz
parents:
diff changeset
4 ; RUN: -polly-target-1st-cache-level-associativity=8 \
anatofuz
parents:
diff changeset
5 ; RUN: -polly-target-2nd-cache-level-associativity=8 \
anatofuz
parents:
diff changeset
6 ; RUN: -polly-target-1st-cache-level-size=32768 \
anatofuz
parents:
diff changeset
7 ; RUN: -polly-target-vector-register-bitwidth=256 \
anatofuz
parents:
diff changeset
8 ; RUN: -polly-target-2nd-cache-level-size=262144 \
anatofuz
parents:
diff changeset
9 ; RUN: -polly-import-jscop-postfix=transformed -polly-codegen -S < %s \
anatofuz
parents:
diff changeset
10 ; RUN: | FileCheck %s
anatofuz
parents:
diff changeset
11 ;
236
c4bab56944e8 LLVM 16
kono
parents: 150
diff changeset
12 ; Check that we disable the Loop Vectorizer.
150
anatofuz
parents:
diff changeset
13 ;
anatofuz
parents:
diff changeset
14 ; CHECK: !{!"llvm.loop.vectorize.enable", i1 false}
anatofuz
parents:
diff changeset
15 ;
anatofuz
parents:
diff changeset
16 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
anatofuz
parents:
diff changeset
17 target triple = "x86_64-unknown-unknown"
anatofuz
parents:
diff changeset
18
anatofuz
parents:
diff changeset
19 define void @kernel_gemm(i32 %ni, i32 %nj, i32 %nk, [1024 x double]* %A, [1024 x double]* %B, [1024 x double]* %C, double* %C1) {
anatofuz
parents:
diff changeset
20 entry:
anatofuz
parents:
diff changeset
21 br label %entry.split
anatofuz
parents:
diff changeset
22
anatofuz
parents:
diff changeset
23 entry.split: ; preds = %entry
anatofuz
parents:
diff changeset
24 br label %for.body
anatofuz
parents:
diff changeset
25
anatofuz
parents:
diff changeset
26 for.body: ; preds = %for.inc22, %entry.split
anatofuz
parents:
diff changeset
27 %indvars.iv43 = phi i64 [ 0, %entry.split ], [ %indvars.iv.next44, %for.inc22 ]
anatofuz
parents:
diff changeset
28 br label %for.body3
anatofuz
parents:
diff changeset
29
anatofuz
parents:
diff changeset
30 for.body3: ; preds = %for.inc19, %for.body
anatofuz
parents:
diff changeset
31 %indvars.iv40 = phi i64 [ 0, %for.body ], [ %indvars.iv.next41, %for.inc19 ]
anatofuz
parents:
diff changeset
32 br label %for.body6
anatofuz
parents:
diff changeset
33
anatofuz
parents:
diff changeset
34 for.body6: ; preds = %for.body6, %for.body3
anatofuz
parents:
diff changeset
35 %indvars.iv = phi i64 [ 0, %for.body3 ], [ %indvars.iv.next, %for.body6 ]
anatofuz
parents:
diff changeset
36 %tmp = load double, double* %C1, align 8
anatofuz
parents:
diff changeset
37 %arrayidx9 = getelementptr inbounds [1024 x double], [1024 x double]* %A, i64 %indvars.iv43, i64 %indvars.iv
anatofuz
parents:
diff changeset
38 %tmp1 = load double, double* %arrayidx9, align 8
anatofuz
parents:
diff changeset
39 %arrayidx13 = getelementptr inbounds [1024 x double], [1024 x double]* %B, i64 %indvars.iv, i64 %indvars.iv40
anatofuz
parents:
diff changeset
40 %tmp2 = load double, double* %arrayidx13, align 8
anatofuz
parents:
diff changeset
41 %mul = fmul double %tmp1, %tmp2
anatofuz
parents:
diff changeset
42 %add = fadd double %tmp, %mul
anatofuz
parents:
diff changeset
43 %arrayidx17 = getelementptr inbounds [1024 x double], [1024 x double]* %C, i64 %indvars.iv43, i64 %indvars.iv40
anatofuz
parents:
diff changeset
44 %tmp3 = load double, double* %arrayidx17, align 8
anatofuz
parents:
diff changeset
45 %add18 = fadd double %tmp3, %add
anatofuz
parents:
diff changeset
46 store double %add18, double* %arrayidx17, align 8
anatofuz
parents:
diff changeset
47 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
anatofuz
parents:
diff changeset
48 %exitcond = icmp ne i64 %indvars.iv.next, 1024
anatofuz
parents:
diff changeset
49 br i1 %exitcond, label %for.body6, label %for.inc19
anatofuz
parents:
diff changeset
50
anatofuz
parents:
diff changeset
51 for.inc19: ; preds = %for.body6
anatofuz
parents:
diff changeset
52 %indvars.iv.next41 = add nuw nsw i64 %indvars.iv40, 1
anatofuz
parents:
diff changeset
53 %exitcond42 = icmp ne i64 %indvars.iv.next41, 1024
anatofuz
parents:
diff changeset
54 br i1 %exitcond42, label %for.body3, label %for.inc22
anatofuz
parents:
diff changeset
55
anatofuz
parents:
diff changeset
56 for.inc22: ; preds = %for.inc19
anatofuz
parents:
diff changeset
57 %indvars.iv.next44 = add nuw nsw i64 %indvars.iv43, 1
anatofuz
parents:
diff changeset
58 %exitcond45 = icmp ne i64 %indvars.iv.next44, 1024
anatofuz
parents:
diff changeset
59 br i1 %exitcond45, label %for.body, label %for.end24
anatofuz
parents:
diff changeset
60
anatofuz
parents:
diff changeset
61 for.end24: ; preds = %for.inc22
anatofuz
parents:
diff changeset
62 ret void
anatofuz
parents:
diff changeset
63 }