1 ; RUN: opt %loadPolly -polly-codegen-ppcg -polly-acc-dump-code \
2 ; RUN: -disable-output < %s | \
3 ; RUN: FileCheck -check-prefix=CODE %s
5 ; RUN: opt %loadPolly -polly-codegen-ppcg -polly-acc-dump-kernel-ir \
6 ; RUN: -disable-output < %s | \
7 ; RUN: FileCheck %s -check-prefix=KERNEL-IR
13 ; float foo(float A[]) {
16 ; for (long i = 0; i < 32; i++)
19 ; for (long i = 0; i < 32; i++)
22 ; for (long i = 0; i < 32; i++)
31 ; printf("%f\n", sum);
39 ; CODE-NEXT: dim3 k0_dimBlock(32);
40 ; CODE-NEXT: dim3 k0_dimGrid(1);
41 ; CODE-NEXT: kernel0 <<<k0_dimGrid, k0_dimBlock>>> (dev_MemRef_A);
42 ; CODE-NEXT: cudaCheckKernel();
46 ; CODE-NEXT: dim3 k1_dimBlock;
47 ; CODE-NEXT: dim3 k1_dimGrid;
48 ; CODE-NEXT: kernel1 <<<k1_dimGrid, k1_dimBlock>>> (dev_MemRef_sum_0__phi);
49 ; CODE-NEXT: cudaCheckKernel();
52 ; CODE: for (int c0 = 0; c0 <= 32; c0 += 1) {
54 ; CODE-NEXT: dim3 k2_dimBlock;
55 ; CODE-NEXT: dim3 k2_dimGrid;
56 ; CODE-NEXT: kernel2 <<<k2_dimGrid, k2_dimBlock>>> (dev_MemRef_sum_0__phi, dev_MemRef_sum_0, c0);
57 ; CODE-NEXT: cudaCheckKernel();
62 ; CODE-NEXT: dim3 k3_dimBlock;
63 ; CODE-NEXT: dim3 k3_dimGrid;
64 ; CODE-NEXT: kernel3 <<<k3_dimGrid, k3_dimBlock>>> (dev_MemRef_A, dev_MemRef_sum_0__phi, dev_MemRef_sum_0, c0);
65 ; CODE-NEXT: cudaCheckKernel();
69 ; CODE-NEXT: cudaCheckReturn(cudaMemcpy(MemRef_A, dev_MemRef_A, (32) * sizeof(float), cudaMemcpyDeviceToHost));
70 ; CODE-NEXT: cudaCheckReturn(cudaMemcpy(&MemRef_sum_0__phi, dev_MemRef_sum_0__phi, sizeof(float), cudaMemcpyDeviceToHost));
71 ; CODE-NEXT: cudaCheckReturn(cudaMemcpy(&MemRef_sum_0, dev_MemRef_sum_0, sizeof(float), cudaMemcpyDeviceToHost));
76 ; CODE-NEXT: Stmt_bb4(t0);
77 ; CODE-NEXT: Stmt_bb10(t0);
81 ; CODE-NEXT: Stmt_bb17();
84 ; CODE-NEXT: Stmt_bb18(c0);
87 ; CODE-NEXT: Stmt_bb20(c0);
89 ; KERNEL-IR: store float %p_tmp23, float* %sum.0.phiops
90 ; KERNEL-IR-NEXT: [[REGA:%.+]] = addrspacecast i8 addrspace(1)* %MemRef_sum_0__phi to float*
91 ; KERNEL-IR-NEXT: [[REGB:%.+]] = load float, float* %sum.0.phiops
92 ; KERNEL-IR-NEXT: store float [[REGB]], float* [[REGA]]
93 ; KERNEL-IR-NEXT: [[REGC:%.+]] = addrspacecast i8 addrspace(1)* %MemRef_sum_0 to float*
94 ; KERNEL-IR-NEXT: [[REGD:%.+]] = load float, float* %sum.0.s2a
95 ; KERNEL-IR-NEXT: store float [[REGD]], float* [[REGC]]
96 ; KERNEL-IR-NEXT: ret void
98 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
100 @.str = private unnamed_addr constant [4 x i8] c"%f\0A\00", align 1
102 define float @foo(float* %A) {
106 bb3: ; preds = %bb6, %bb
107 %i.0 = phi i64 [ 0, %bb ], [ %tmp7, %bb6 ]
108 %exitcond2 = icmp ne i64 %i.0, 32
109 br i1 %exitcond2, label %bb4, label %bb8
112 %tmp = sitofp i64 %i.0 to float
113 %tmp5 = getelementptr inbounds float, float* %A, i64 %i.0
114 store float %tmp, float* %tmp5, align 4
118 %tmp7 = add nuw nsw i64 %i.0, 1
124 bb9: ; preds = %bb15, %bb8
125 %i1.0 = phi i64 [ 0, %bb8 ], [ %tmp16, %bb15 ]
126 %exitcond1 = icmp ne i64 %i1.0, 32
127 br i1 %exitcond1, label %bb10, label %bb17
130 %tmp11 = sitofp i64 %i1.0 to float
131 %tmp12 = getelementptr inbounds float, float* %A, i64 %i1.0
132 %tmp13 = load float, float* %tmp12, align 4
133 %tmp14 = fadd float %tmp13, %tmp11
134 store float %tmp14, float* %tmp12, align 4
137 bb15: ; preds = %bb10
138 %tmp16 = add nuw nsw i64 %i1.0, 1
144 bb18: ; preds = %bb20, %bb17
145 %sum.0 = phi float [ 0.000000e+00, %bb17 ], [ %tmp23, %bb20 ]
146 %i2.0 = phi i64 [ 0, %bb17 ], [ %tmp24, %bb20 ]
147 %exitcond = icmp ne i64 %i2.0, 32
148 br i1 %exitcond, label %bb19, label %bb25
150 bb19: ; preds = %bb18
153 bb20: ; preds = %bb19
154 %tmp21 = getelementptr inbounds float, float* %A, i64 %i2.0
155 %tmp22 = load float, float* %tmp21, align 4
156 %tmp23 = fadd float %sum.0, %tmp22
157 %tmp24 = add nuw nsw i64 %i2.0, 1
160 bb25: ; preds = %bb18
161 %sum.0.lcssa = phi float [ %sum.0, %bb18 ]
162 ret float %sum.0.lcssa
167 %A = alloca [32 x float], align 16
168 %tmp = getelementptr inbounds [32 x float], [32 x float]* %A, i64 0, i64 0
169 %tmp1 = call float @foo(float* %tmp)
170 %tmp2 = fpext float %tmp1 to double
171 %tmp3 = call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str, i64 0, i64 0), double %tmp2) #2
175 declare i32 @printf(i8*, ...) #1