1 ; RUN: opt %loadNPMPolly -polly-pattern-matching-based-opts=true \
2 ; RUN: -polly-target-throughput-vector-fma=1 \
3 ; RUN: -polly-target-latency-vector-fma=8 \
4 ; RUN: -polly-target-1st-cache-level-associativity=8 \
5 ; RUN: -polly-target-2nd-cache-level-associativity=8 \
6 ; RUN: -polly-target-1st-cache-level-size=32768 \
7 ; RUN: -polly-target-vector-register-bitwidth=256 \
8 ; RUN: -polly-target-2nd-cache-level-size=262144 \
9 ; RUN: '-passes=polly-opt-isl,print<polly-ast>' -disable-output < %s | FileCheck %s
11 ; /* C := A * B + C */
12 ; /* Elements of the matrices B, C have the double type. */
13 ; /* Elements of the matrix A have the float type. */
14 ; /* The type size of elements of the matrix multiplication operands is used
15 ; to determine the parameters of the code produced by the optimization
16 ; of the matrix multiplication (e.g. bounds of the loops of the loop
17 ; nest, the innermost loop body). This test checks the form of
18 ; the generated loop nest. See getMicroKernelParams and
19 ; getMacroKernelParams from lib/Transform/ScheduleOptimizer.cpp
21 ; for (i = 0; i < _PB_NI; i++)
22 ; for (j = 0; j < _PB_NJ; j++)
23 ; for (k = 0; k < _PB_NK; ++k)
24 ; C[i][j] += A[i][k] * B[k][j];
26 ; CHECK: // 1st level tiling - Tiles
27 ; CHECK-NEXT: for (int c1 = 0; c1 <= 3; c1 += 1) {
28 ; CHECK-NEXT: for (int c3 = 0; c3 <= 1023; c3 += 1)
29 ; CHECK-NEXT: for (int c4 = 256 * c1; c4 <= 256 * c1 + 255; c4 += 1)
30 ; CHECK-NEXT: CopyStmt_0(0, c3, c4);
31 ; CHECK-NEXT: for (int c2 = 0; c2 <= 10; c2 += 1) {
32 ; CHECK-NEXT: for (int c6 = 96 * c2; c6 <= min(1023, 96 * c2 + 95); c6 += 1)
33 ; CHECK-NEXT: for (int c7 = 256 * c1; c7 <= 256 * c1 + 255; c7 += 1)
34 ; CHECK-NEXT: CopyStmt_1(0, c1, c2, c6, c7);
35 ; CHECK-NEXT: // 1st level tiling - Points
36 ; CHECK-NEXT: // Register tiling - Tiles
37 ; CHECK-NEXT: for (int c3 = 0; c3 <= 127; c3 += 1)
38 ; CHECK-NEXT: for (int c4 = 0; c4 <= min(23, -24 * c2 + 255); c4 += 1)
39 ; CHECK-NEXT: for (int c5 = 0; c5 <= 255; c5 += 1) {
40 ; CHECK-NEXT: // Loop Vectorizer Disabled
41 ; CHECK-NEXT: // Register tiling - Points
43 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4, 8 * c3, 256 * c1 + c5);
44 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4, 8 * c3 + 1, 256 * c1 + c5);
45 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4, 8 * c3 + 2, 256 * c1 + c5);
46 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4, 8 * c3 + 3, 256 * c1 + c5);
47 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4, 8 * c3 + 4, 256 * c1 + c5);
48 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4, 8 * c3 + 5, 256 * c1 + c5);
49 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4, 8 * c3 + 6, 256 * c1 + c5);
50 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4, 8 * c3 + 7, 256 * c1 + c5);
51 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 1, 8 * c3, 256 * c1 + c5);
52 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 1, 8 * c3 + 1, 256 * c1 + c5);
53 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 1, 8 * c3 + 2, 256 * c1 + c5);
54 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 1, 8 * c3 + 3, 256 * c1 + c5);
55 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 1, 8 * c3 + 4, 256 * c1 + c5);
56 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 1, 8 * c3 + 5, 256 * c1 + c5);
57 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 1, 8 * c3 + 6, 256 * c1 + c5);
58 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 1, 8 * c3 + 7, 256 * c1 + c5);
59 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 2, 8 * c3, 256 * c1 + c5);
60 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 2, 8 * c3 + 1, 256 * c1 + c5);
61 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 2, 8 * c3 + 2, 256 * c1 + c5);
62 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 2, 8 * c3 + 3, 256 * c1 + c5);
63 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 2, 8 * c3 + 4, 256 * c1 + c5);
64 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 2, 8 * c3 + 5, 256 * c1 + c5);
65 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 2, 8 * c3 + 6, 256 * c1 + c5);
66 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 2, 8 * c3 + 7, 256 * c1 + c5);
67 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 3, 8 * c3, 256 * c1 + c5);
68 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 3, 8 * c3 + 1, 256 * c1 + c5);
69 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 3, 8 * c3 + 2, 256 * c1 + c5);
70 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 3, 8 * c3 + 3, 256 * c1 + c5);
71 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 3, 8 * c3 + 4, 256 * c1 + c5);
72 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 3, 8 * c3 + 5, 256 * c1 + c5);
73 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 3, 8 * c3 + 6, 256 * c1 + c5);
74 ; CHECK-NEXT: Stmt_for_body6(96 * c2 + 4 * c4 + 3, 8 * c3 + 7, 256 * c1 + c5);
80 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
81 target triple = "x86_64-unknown-unknown"
83 ; Function Attrs: noinline nounwind uwtable
84 define internal void @kernel_gemm(i32 %ni, i32 %nj, i32 %nk, double %alpha, double %beta, ptr %C, ptr %A, ptr %B) #0 {
88 entry.split: ; preds = %entry
89 br label %for.cond1.preheader
91 for.cond1.preheader: ; preds = %for.inc20, %entry.split
92 %indvars.iv41 = phi i64 [ 0, %entry.split ], [ %indvars.iv.next42, %for.inc20 ]
93 br label %for.cond4.preheader
95 for.cond4.preheader: ; preds = %for.inc17, %for.cond1.preheader
96 %indvars.iv38 = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next39, %for.inc17 ]
99 for.body6: ; preds = %for.body6, %for.cond4.preheader
100 %indvars.iv = phi i64 [ 0, %for.cond4.preheader ], [ %indvars.iv.next, %for.body6 ]
101 %arrayidx8 = getelementptr inbounds [1024 x float], ptr %A, i64 %indvars.iv41, i64 %indvars.iv
102 %tmp = load float, ptr %arrayidx8, align 4
103 %conv = fpext float %tmp to double
104 %arrayidx12 = getelementptr inbounds [1024 x double], ptr %B, i64 %indvars.iv, i64 %indvars.iv38
105 %tmp1 = load double, ptr %arrayidx12, align 8
106 %mul = fmul double %conv, %tmp1
107 %arrayidx16 = getelementptr inbounds [1024 x double], ptr %C, i64 %indvars.iv41, i64 %indvars.iv38
108 %tmp2 = load double, ptr %arrayidx16, align 8
109 %add = fadd double %tmp2, %mul
110 store double %add, ptr %arrayidx16, align 8
111 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
112 %exitcond = icmp ne i64 %indvars.iv.next, 1024
113 br i1 %exitcond, label %for.body6, label %for.inc17
115 for.inc17: ; preds = %for.body6
116 %indvars.iv.next39 = add nuw nsw i64 %indvars.iv38, 1
117 %exitcond40 = icmp ne i64 %indvars.iv.next39, 1024
118 br i1 %exitcond40, label %for.cond4.preheader, label %for.inc20
120 for.inc20: ; preds = %for.inc17
121 %indvars.iv.next42 = add nuw nsw i64 %indvars.iv41, 1
122 %exitcond43 = icmp ne i64 %indvars.iv.next42, 1024
123 br i1 %exitcond43, label %for.cond1.preheader, label %for.end22
125 for.end22: ; preds = %for.inc20