| blob | e6188e8b43059e6fe59fcde7e6bc2d7b580cff04 |
1 ; RUN: opt %loadPolly -O3 -polly -polly-target=gpu \
2 ; RUN: -polly-gpu-arch=spir32 \
3 ; RUN: -polly-acc-dump-kernel-ir -polly-process-unprofitable -disable-output < %s | \
4 ; RUN: FileCheck %s
6 ; REQUIRES: pollyacc
8 ; CHECK: target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-i128:128:128-f32:32:32-f64:64:64-v16:16:16-v24:32:32-v32:32:32-v48:64:64-v64:64:64-v96:128:128-v128:128:128-v192:256:256-v256:256:256-v512:512:512-v1024:1024:1024"
9 ; CHECK-NEXT: target triple = "spir-unknown-unknown"
11 ; CHECK-LABEL: define spir_kernel void @FUNC_double_parallel_loop_SCOP_0_KERNEL_0(i8 addrspace(1)* %MemRef_A) #0 !kernel_arg_addr_space !0 !kernel_arg_name !1 !kernel_arg_access_qual !1 !kernel_arg_type !1 !kernel_arg_type_qual !1 !kernel_arg_base_type !1 {
12 ; CHECK-NEXT: entry:
13 ; CHECK-NEXT: %0 = call i32 @__gen_ocl_get_group_id0()
14 ; CHECK-NEXT: %__gen_ocl_get_group_id0 = zext i32 %0 to i64
15 ; CHECK-NEXT: %1 = call i32 @__gen_ocl_get_group_id1()
16 ; CHECK-NEXT: %__gen_ocl_get_group_id1 = zext i32 %1 to i64
17 ; CHECK-NEXT: %2 = call i32 @__gen_ocl_get_local_id0()
18 ; CHECK-NEXT: %__gen_ocl_get_local_id0 = zext i32 %2 to i64
19 ; CHECK-NEXT: %3 = call i32 @__gen_ocl_get_local_id1()
20 ; CHECK-NEXT: %__gen_ocl_get_local_id1 = zext i32 %3 to i64
21 ; CHECK-NEXT: br label %polly.loop_preheader
23 ; CHECK-LABEL: polly.loop_exit: ; preds = %polly.stmt.bb5
24 ; CHECK-NEXT: ret void
26 ; CHECK-LABEL: polly.loop_header: ; preds = %polly.stmt.bb5, %polly.loop_preheader
27 ; CHECK-NEXT: %polly.indvar = phi i64 [ 0, %polly.loop_preheader ], [ %polly.indvar_next, %polly.stmt.bb5 ]
28 ; CHECK-NEXT: %4 = mul nsw i64 32, %__gen_ocl_get_group_id0
29 ; CHECK-NEXT: %5 = add nsw i64 %4, %__gen_ocl_get_local_id0
30 ; CHECK-NEXT: %6 = mul nsw i64 32, %__gen_ocl_get_group_id1
31 ; CHECK-NEXT: %7 = add nsw i64 %6, %__gen_ocl_get_local_id1
32 ; CHECK-NEXT: %8 = mul nsw i64 16, %polly.indvar
33 ; CHECK-NEXT: %9 = add nsw i64 %7, %8
34 ; CHECK-NEXT: br label %polly.stmt.bb5
36 ; CHECK-LABEL: polly.stmt.bb5: ; preds = %polly.loop_header
37 ; CHECK-NEXT: %10 = mul i64 %5, %9
38 ; CHECK-NEXT: %p_tmp6 = sitofp i64 %10 to float
39 ; CHECK-NEXT: %polly.access.cast.MemRef_A = bitcast i8 addrspace(1)* %MemRef_A to float addrspace(1)*
40 ; CHECK-NEXT: %11 = mul nsw i64 32, %__gen_ocl_get_group_id0
41 ; CHECK-NEXT: %12 = add nsw i64 %11, %__gen_ocl_get_local_id0
42 ; CHECK-NEXT: %polly.access.mul.MemRef_A = mul nsw i64 %12, 1024
43 ; CHECK-NEXT: %13 = mul nsw i64 32, %__gen_ocl_get_group_id1
44 ; CHECK-NEXT: %14 = add nsw i64 %13, %__gen_ocl_get_local_id1
45 ; CHECK-NEXT: %15 = mul nsw i64 16, %polly.indvar
46 ; CHECK-NEXT: %16 = add nsw i64 %14, %15
47 ; CHECK-NEXT: %polly.access.add.MemRef_A = add nsw i64 %polly.access.mul.MemRef_A, %16
48 ; CHECK-NEXT: %polly.access.MemRef_A = getelementptr float, float addrspace(1)* %polly.access.cast.MemRef_A, i64 %polly.access.add.MemRef_A
49 ; CHECK-NEXT: %tmp8_p_scalar_ = load float, float addrspace(1)* %polly.access.MemRef_A, align 4
50 ; CHECK-NEXT: %p_tmp9 = fadd float %tmp8_p_scalar_, %p_tmp6
51 ; CHECK-NEXT: %polly.access.cast.MemRef_A1 = bitcast i8 addrspace(1)* %MemRef_A to float addrspace(1)*
52 ; CHECK-NEXT: %17 = mul nsw i64 32, %__gen_ocl_get_group_id0
53 ; CHECK-NEXT: %18 = add nsw i64 %17, %__gen_ocl_get_local_id0
54 ; CHECK-NEXT: %polly.access.mul.MemRef_A2 = mul nsw i64 %18, 1024
55 ; CHECK-NEXT: %19 = mul nsw i64 32, %__gen_ocl_get_group_id1
56 ; CHECK-NEXT: %20 = add nsw i64 %19, %__gen_ocl_get_local_id1
57 ; CHECK-NEXT: %21 = mul nsw i64 16, %polly.indvar
58 ; CHECK-NEXT: %22 = add nsw i64 %20, %21
59 ; CHECK-NEXT: %polly.access.add.MemRef_A3 = add nsw i64 %polly.access.mul.MemRef_A2, %22
60 ; CHECK-NEXT: %polly.access.MemRef_A4 = getelementptr float, float addrspace(1)* %polly.access.cast.MemRef_A1, i64 %polly.access.add.MemRef_A3
61 ; CHECK-NEXT: store float %p_tmp9, float addrspace(1)* %polly.access.MemRef_A4, align 4
62 ; CHECK-NEXT: %polly.indvar_next = add nsw i64 %polly.indvar, 1
63 ; CHECK-NEXT: %polly.loop_cond = icmp sle i64 %polly.indvar_next, 1
64 ; CHECK-NEXT: br i1 %polly.loop_cond, label %polly.loop_header, label %polly.loop_exit
66 ; CHECK-LABEL: polly.loop_preheader: ; preds = %entry
67 ; CHECK-NEXT: br label %polly.loop_header
69 ; CHECK: attributes #0 = { "polly.skip.fn" }
71 ; void double_parallel_loop(float A[][1024]) {
72 ; for (long i = 0; i < 1024; i++)
73 ; for (long j = 0; j < 1024; j++)
74 ; A[i][j] += i * j;
75 ; }
76 ;
77 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
79 define void @double_parallel_loop([1024 x float]* %A) {
80 bb:
81 br label %bb2
83 bb2: ; preds = %bb13, %bb
84 %i.0 = phi i64 [ 0, %bb ], [ %tmp14, %bb13 ]
85 %exitcond1 = icmp ne i64 %i.0, 1024
86 br i1 %exitcond1, label %bb3, label %bb15
88 bb3: ; preds = %bb2
89 br label %bb4
91 bb4: ; preds = %bb10, %bb3
92 %j.0 = phi i64 [ 0, %bb3 ], [ %tmp11, %bb10 ]
93 %exitcond = icmp ne i64 %j.0, 1024
94 br i1 %exitcond, label %bb5, label %bb12
96 bb5: ; preds = %bb4
97 %tmp = mul nuw nsw i64 %i.0, %j.0
98 %tmp6 = sitofp i64 %tmp to float
99 %tmp7 = getelementptr inbounds [1024 x float], [1024 x float]* %A, i64 %i.0, i64 %j.0
100 %tmp8 = load float, float* %tmp7, align 4
101 %tmp9 = fadd float %tmp8, %tmp6
102 store float %tmp9, float* %tmp7, align 4
103 br label %bb10
105 bb10: ; preds = %bb5
106 %tmp11 = add nuw nsw i64 %j.0, 1
107 br label %bb4
109 bb12: ; preds = %bb4
110 br label %bb13
112 bb13: ; preds = %bb12
113 %tmp14 = add nuw nsw i64 %i.0, 1
114 br label %bb2
116 bb15: ; preds = %bb2
117 ret void
118 }