1 ; RUN: opt %loadPolly -polly-opt-isl -polly-invariant-load-hoisting=true \
2 ; RUN: -polly-pattern-matching-based-opts=true \
3 ; RUN: -polly-target-throughput-vector-fma=1 \
4 ; RUN: -polly-target-latency-vector-fma=1 \
5 ; RUN: -polly-codegen -polly-target-1st-cache-level-associativity=8 \
6 ; RUN: -polly-target-2nd-cache-level-associativity=8 \
7 ; RUN: -polly-target-1st-cache-level-size=32768 \
8 ; RUN: -polly-target-vector-register-bitwidth=256 \
9 ; RUN: -polly-target-2nd-cache-level-size=262144 -S < %s \
12 ; This test case checks whether Polly generates second level alias metadata
13 ; to distinguish the specific accesses in case of the ublas gemm kernel.
15 ; CHECK: %tmp22_p_scalar_{{[0-9]*}} = load double, double* %scevgep[[N0:[a-z_0-9]*]], align 8
16 ; CHECK: store double %p_tmp23{{[0-9]*}}, double* %scevgep[[N0]], align 8
17 ; CHECK: %tmp22_p_scalar_{{[0-9]*}} = load double, double* %scevgep[[N1:[a-z_0-9]*]], align 8
18 ; CHECK: store double %p_tmp23{{[0-9]*}}, double* %scevgep[[N1]], align 8
19 ; CHECK: %tmp22_p_scalar_{{[0-9]*}} = load double, double* %scevgep[[N2:[a-z_0-9]*]], align 8
20 ; CHECK: store double %p_tmp23{{[0-9]*}}, double* %scevgep[[N2]], align 8
21 ; CHECK: %tmp22_p_scalar_{{[0-9]*}} = load double, double* %scevgep[[N3:[a-z_0-9]*]], align 8
22 ; CHECK: store double %p_tmp23{{[0-9]*}}, double* %scevgep[[N3]], align 8
24 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
25 target triple = "x86_64-unknown-unknown"
27 define internal void @kernel_gemm(i32 %arg, i32 %arg1, i32 %arg2, double %arg3, double %arg4, [1056 x double]* %arg5, [1024 x double]* %arg6, [1056 x double]* %arg7) {
31 bb8: ; preds = %bb29, %bb
32 %tmp = phi i64 [ 0, %bb ], [ %tmp30, %bb29 ]
35 bb9: ; preds = %bb26, %bb8
36 %tmp10 = phi i64 [ 0, %bb8 ], [ %tmp27, %bb26 ]
37 %tmp11 = getelementptr inbounds [1056 x double], [1056 x double]* %arg5, i64 %tmp, i64 %tmp10
38 %tmp12 = load double, double* %tmp11, align 8
39 %tmp13 = fmul double %tmp12, %arg4
40 store double %tmp13, double* %tmp11, align 8
43 Copy_0: ; preds = %Copy_0, %bb9
44 %tmp15 = phi i64 [ 0, %bb9 ], [ %tmp24, %Copy_0 ]
45 %tmp16 = getelementptr inbounds [1024 x double], [1024 x double]* %arg6, i64 %tmp, i64 %tmp15
46 %tmp17 = load double, double* %tmp16, align 8
47 %tmp18 = fmul double %tmp17, %arg3
48 %tmp19 = getelementptr inbounds [1056 x double], [1056 x double]* %arg7, i64 %tmp15, i64 %tmp10
49 %tmp20 = load double, double* %tmp19, align 8
50 %tmp21 = fmul double %tmp18, %tmp20
51 %tmp22 = load double, double* %tmp11, align 8
52 %tmp23 = fadd double %tmp22, %tmp21
53 store double %tmp23, double* %tmp11, align 8
54 %tmp24 = add nuw nsw i64 %tmp15, 1
55 %tmp25 = icmp ne i64 %tmp24, 1024
56 br i1 %tmp25, label %Copy_0, label %bb26
58 bb26: ; preds = %Copy_0
59 %tmp27 = add nuw nsw i64 %tmp10, 1
60 %tmp28 = icmp ne i64 %tmp27, 1056
61 br i1 %tmp28, label %bb9, label %bb29
64 %tmp30 = add nuw nsw i64 %tmp, 1
65 %tmp31 = icmp ne i64 %tmp30, 1056
66 br i1 %tmp31, label %bb8, label %bb32