annotate polly/test/ScheduleOptimizer/mat_mul_pattern_data_layout.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 5f17cb93ff66
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-pattern-matching-based-opts=true \
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-2nd-cache-level-size=262144 \
anatofuz
parents:
diff changeset
8 ; RUN: -polly-optimized-scops \
anatofuz
parents:
diff changeset
9 ; RUN: -polly-target-vector-register-bitwidth=256 \
223
5f17cb93ff66 LLVM13 (2021/7/18)
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 150
diff changeset
10 ; RUN: -disable-output < %s
150
anatofuz
parents:
diff changeset
11 ;
anatofuz
parents:
diff changeset
12 ; /* C := alpha*A*B + beta*C */
anatofuz
parents:
diff changeset
13 ; for (i = 0; i < _PB_NI; i++)
anatofuz
parents:
diff changeset
14 ; for (j = 0; j < _PB_NJ; j++)
anatofuz
parents:
diff changeset
15 ; {
anatofuz
parents:
diff changeset
16 ; C[i][j] *= beta;
anatofuz
parents:
diff changeset
17 ; for (k = 0; k < _PB_NK; ++k)
anatofuz
parents:
diff changeset
18 ; C[i][j] += alpha * A[i][k] * B[k][j];
anatofuz
parents:
diff changeset
19 ; }
anatofuz
parents:
diff changeset
20 ;
anatofuz
parents:
diff changeset
21 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
anatofuz
parents:
diff changeset
22 target triple = "x86_64-unknown-unknown"
anatofuz
parents:
diff changeset
23
252
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
24 define internal void @kernel_gemm(i32 %arg, i32 %arg1, i32 %arg2, double %arg3, double %arg4, ptr %arg5, ptr %arg6, ptr %arg7) #0 {
150
anatofuz
parents:
diff changeset
25 bb:
anatofuz
parents:
diff changeset
26 br label %bb8
anatofuz
parents:
diff changeset
27
anatofuz
parents:
diff changeset
28 bb8: ; preds = %bb29, %bb
anatofuz
parents:
diff changeset
29 %tmp = phi i64 [ 0, %bb ], [ %tmp30, %bb29 ]
anatofuz
parents:
diff changeset
30 br label %bb9
anatofuz
parents:
diff changeset
31
anatofuz
parents:
diff changeset
32 bb9: ; preds = %bb26, %bb8
anatofuz
parents:
diff changeset
33 %tmp10 = phi i64 [ 0, %bb8 ], [ %tmp27, %bb26 ]
252
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
34 %tmp11 = getelementptr inbounds [1056 x double], ptr %arg5, i64 %tmp, i64 %tmp10
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
35 %tmp12 = load double, ptr %tmp11, align 8
150
anatofuz
parents:
diff changeset
36 %tmp13 = fmul double %tmp12, %arg4
252
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
37 store double %tmp13, ptr %tmp11, align 8
150
anatofuz
parents:
diff changeset
38 br label %Copy_0
anatofuz
parents:
diff changeset
39
anatofuz
parents:
diff changeset
40 Copy_0: ; preds = %Copy_0, %bb9
anatofuz
parents:
diff changeset
41 %tmp15 = phi i64 [ 0, %bb9 ], [ %tmp24, %Copy_0 ]
252
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
42 %tmp16 = getelementptr inbounds [1024 x double], ptr %arg6, i64 %tmp, i64 %tmp15
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
43 %tmp17 = load double, ptr %tmp16, align 8
150
anatofuz
parents:
diff changeset
44 %tmp18 = fmul double %tmp17, %arg3
252
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
45 %tmp19 = getelementptr inbounds [1056 x double], ptr %arg7, i64 %tmp15, i64 %tmp10
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
46 %tmp20 = load double, ptr %tmp19, align 8
150
anatofuz
parents:
diff changeset
47 %tmp21 = fmul double %tmp18, %tmp20
252
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
48 %tmp22 = load double, ptr %tmp11, align 8
150
anatofuz
parents:
diff changeset
49 %tmp23 = fadd double %tmp22, %tmp21
252
1f2b6ac9f198 LLVM16-1
Shinji KONO <kono@ie.u-ryukyu.ac.jp>
parents: 223
diff changeset
50 store double %tmp23, ptr %tmp11, align 8
150
anatofuz
parents:
diff changeset
51 %tmp24 = add nuw nsw i64 %tmp15, 1
anatofuz
parents:
diff changeset
52 %tmp25 = icmp ne i64 %tmp24, 1024
anatofuz
parents:
diff changeset
53 br i1 %tmp25, label %Copy_0, label %bb26
anatofuz
parents:
diff changeset
54
anatofuz
parents:
diff changeset
55 bb26: ; preds = %Copy_0
anatofuz
parents:
diff changeset
56 %tmp27 = add nuw nsw i64 %tmp10, 1
anatofuz
parents:
diff changeset
57 %tmp28 = icmp ne i64 %tmp27, 1056
anatofuz
parents:
diff changeset
58 br i1 %tmp28, label %bb9, label %bb29
anatofuz
parents:
diff changeset
59
anatofuz
parents:
diff changeset
60 bb29: ; preds = %bb26
anatofuz
parents:
diff changeset
61 %tmp30 = add nuw nsw i64 %tmp, 1
anatofuz
parents:
diff changeset
62 %tmp31 = icmp ne i64 %tmp30, 1056
anatofuz
parents:
diff changeset
63 br i1 %tmp31, label %bb8, label %bb32
anatofuz
parents:
diff changeset
64
anatofuz
parents:
diff changeset
65 bb32: ; preds = %bb29
anatofuz
parents:
diff changeset
66 ret void
anatofuz
parents:
diff changeset
67 }
anatofuz
parents:
diff changeset
68
anatofuz
parents:
diff changeset
69 attributes #0 = { nounwind uwtable "target-cpu"="x86-64" "target-features"="+aes,+avx,+cmov,+cx16,+fxsr,+mmx,+pclmul,+popcnt,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsaveopt" }