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-2nd-cache-level-size=262144 \
8 ; RUN: -polly-target-vector-register-bitwidth=256 \
9 ; RUN: '-passes=polly-opt-isl,print<polly-ast>' -disable-output < %s | FileCheck %s
11 ; /* C := alpha*A*B + beta*C */
12 ; /* _PB_NK % Kc != 0 */
13 ; for (i = 0; i < _PB_NI; i++)
14 ; for (j = 0; j < _PB_NJ; j++)
17 ; for (k = 0; k < _PB_NK; ++k)
18 ; C[i][j] += alpha * A[i][k] * B[k][j];
21 ; CHECK-LABEL: :: isl ast :: kernel_gemm :: %bb8---%bb32
23 ; CHECK-NEXT: // 1st level tiling - Tiles
24 ; CHECK-NEXT: for (int c0 = 0; c0 <= 32; c0 += 1)
25 ; CHECK-NEXT: for (int c1 = 0; c1 <= 32; c1 += 1) {
26 ; CHECK-NEXT: // 1st level tiling - Points
27 ; CHECK-NEXT: for (int c2 = 0; c2 <= 31; c2 += 1)
28 ; CHECK-NEXT: for (int c3 = 0; c3 <= 31; c3 += 1)
29 ; CHECK-NEXT: Stmt_bb9(32 * c0 + c2, 32 * c1 + c3);
31 ; CHECK-NEXT: // 1st level tiling - Tiles
32 ; CHECK-NEXT: for (int c1 = 0; c1 <= 3; c1 += 1) {
33 ; CHECK-NEXT: for (int c3 = 0; c3 <= 1055; c3 += 1)
34 ; CHECK-NEXT: for (int c4 = 256 * c1; c4 <= min(1022, 256 * c1 + 255); c4 += 1)
35 ; CHECK-NEXT: CopyStmt_0(0, c3, c4);
36 ; CHECK-NEXT: for (int c2 = 0; c2 <= 10; c2 += 1) {
37 ; CHECK-NEXT: for (int c6 = 96 * c2; c6 <= 96 * c2 + 95; c6 += 1)
38 ; CHECK-NEXT: for (int c7 = 256 * c1; c7 <= min(1022, 256 * c1 + 255); c7 += 1)
39 ; CHECK-NEXT: CopyStmt_1(0, c1, c2, c6, c7);
40 ; CHECK-NEXT: // 1st level tiling - Points
41 ; CHECK-NEXT: // Register tiling - Tiles
42 ; CHECK-NEXT: for (int c3 = 0; c3 <= 131; c3 += 1)
43 ; CHECK-NEXT: for (int c4 = 0; c4 <= 23; c4 += 1)
44 ; CHECK-NEXT: for (int c5 = 0; c5 <= min(255, -256 * c1 + 1022); c5 += 1) {
45 ; CHECK-NEXT: // Loop Vectorizer Disabled
46 ; CHECK-NEXT: // Register tiling - Points
48 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4, 8 * c3, 256 * c1 + c5);
49 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4, 8 * c3 + 1, 256 * c1 + c5);
50 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4, 8 * c3 + 2, 256 * c1 + c5);
51 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4, 8 * c3 + 3, 256 * c1 + c5);
52 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4, 8 * c3 + 4, 256 * c1 + c5);
53 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4, 8 * c3 + 5, 256 * c1 + c5);
54 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4, 8 * c3 + 6, 256 * c1 + c5);
55 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4, 8 * c3 + 7, 256 * c1 + c5);
56 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 1, 8 * c3, 256 * c1 + c5);
57 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 1, 8 * c3 + 1, 256 * c1 + c5);
58 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 1, 8 * c3 + 2, 256 * c1 + c5);
59 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 1, 8 * c3 + 3, 256 * c1 + c5);
60 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 1, 8 * c3 + 4, 256 * c1 + c5);
61 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 1, 8 * c3 + 5, 256 * c1 + c5);
62 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 1, 8 * c3 + 6, 256 * c1 + c5);
63 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 1, 8 * c3 + 7, 256 * c1 + c5);
64 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 2, 8 * c3, 256 * c1 + c5);
65 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 2, 8 * c3 + 1, 256 * c1 + c5);
66 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 2, 8 * c3 + 2, 256 * c1 + c5);
67 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 2, 8 * c3 + 3, 256 * c1 + c5);
68 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 2, 8 * c3 + 4, 256 * c1 + c5);
69 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 2, 8 * c3 + 5, 256 * c1 + c5);
70 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 2, 8 * c3 + 6, 256 * c1 + c5);
71 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 2, 8 * c3 + 7, 256 * c1 + c5);
72 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 3, 8 * c3, 256 * c1 + c5);
73 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 3, 8 * c3 + 1, 256 * c1 + c5);
74 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 3, 8 * c3 + 2, 256 * c1 + c5);
75 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 3, 8 * c3 + 3, 256 * c1 + c5);
76 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 3, 8 * c3 + 4, 256 * c1 + c5);
77 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 3, 8 * c3 + 5, 256 * c1 + c5);
78 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 3, 8 * c3 + 6, 256 * c1 + c5);
79 ; CHECK-NEXT: Stmt_Copy_0(96 * c2 + 4 * c4 + 3, 8 * c3 + 7, 256 * c1 + c5);
86 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
87 target triple = "x86_64-unknown-unknown"
89 define internal void @kernel_gemm(i32 %arg, i32 %arg1, i32 %arg2, double %arg3, double %arg4, ptr %arg5, ptr %arg6, ptr %arg7) #0 {
93 bb8: ; preds = %bb29, %bb
94 %tmp = phi i64 [ 0, %bb ], [ %tmp30, %bb29 ]
97 bb9: ; preds = %bb26, %bb8
98 %tmp10 = phi i64 [ 0, %bb8 ], [ %tmp27, %bb26 ]
99 %tmp11 = getelementptr inbounds [1056 x double], ptr %arg5, i64 %tmp, i64 %tmp10
100 %tmp12 = load double, ptr %tmp11, align 8
101 %tmp13 = fmul double %tmp12, %arg4
102 store double %tmp13, ptr %tmp11, align 8
105 Copy_0: ; preds = %Copy_0, %bb9
106 %tmp15 = phi i64 [ 0, %bb9 ], [ %tmp24, %Copy_0 ]
107 %tmp16 = getelementptr inbounds [1023 x double], ptr %arg6, i64 %tmp, i64 %tmp15
108 %tmp17 = load double, ptr %tmp16, align 8
109 %tmp18 = fmul double %tmp17, %arg3
110 %tmp19 = getelementptr inbounds [1056 x double], ptr %arg7, i64 %tmp15, i64 %tmp10
111 %tmp20 = load double, ptr %tmp19, align 8
112 %tmp21 = fmul double %tmp18, %tmp20
113 %tmp22 = load double, ptr %tmp11, align 8
114 %tmp23 = fadd double %tmp22, %tmp21
115 store double %tmp23, ptr %tmp11, align 8
116 %tmp24 = add nuw nsw i64 %tmp15, 1
117 %tmp25 = icmp ne i64 %tmp24, 1023
118 br i1 %tmp25, label %Copy_0, label %bb26
120 bb26: ; preds = %Copy_0
121 %tmp27 = add nuw nsw i64 %tmp10, 1
122 %tmp28 = icmp ne i64 %tmp27, 1056
123 br i1 %tmp28, label %bb9, label %bb29
125 bb29: ; preds = %bb26
126 %tmp30 = add nuw nsw i64 %tmp, 1
127 %tmp31 = icmp ne i64 %tmp30, 1056
128 br i1 %tmp31, label %bb8, label %bb32
130 bb32: ; preds = %bb29
134 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" }