1 ; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI %s
3 declare void @llvm.AMDGPU.barrier.local() noduplicate nounwind
5 ; SI-LABEL: @private_access_f64_alloca:
8 define void @private_access_f64_alloca(double addrspace(1)* noalias %out, double addrspace(1)* noalias %in, i32 %b) nounwind {
9 %val = load double addrspace(1)* %in, align 8
10 %array = alloca double, i32 16, align 8
11 %ptr = getelementptr double* %array, i32 %b
12 store double %val, double* %ptr, align 8
13 call void @llvm.AMDGPU.barrier.local() noduplicate nounwind
14 %result = load double* %ptr, align 8
15 store double %result, double addrspace(1)* %out, align 8
19 ; SI-LABEL: @private_access_v2f64_alloca:
24 define void @private_access_v2f64_alloca(<2 x double> addrspace(1)* noalias %out, <2 x double> addrspace(1)* noalias %in, i32 %b) nounwind {
25 %val = load <2 x double> addrspace(1)* %in, align 16
26 %array = alloca <2 x double>, i32 16, align 16
27 %ptr = getelementptr <2 x double>* %array, i32 %b
28 store <2 x double> %val, <2 x double>* %ptr, align 16
29 call void @llvm.AMDGPU.barrier.local() noduplicate nounwind
30 %result = load <2 x double>* %ptr, align 16
31 store <2 x double> %result, <2 x double> addrspace(1)* %out, align 16
35 ; SI-LABEL: @private_access_i64_alloca:
38 define void @private_access_i64_alloca(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in, i32 %b) nounwind {
39 %val = load i64 addrspace(1)* %in, align 8
40 %array = alloca i64, i32 16, align 8
41 %ptr = getelementptr i64* %array, i32 %b
42 store i64 %val, i64* %ptr, align 8
43 call void @llvm.AMDGPU.barrier.local() noduplicate nounwind
44 %result = load i64* %ptr, align 8
45 store i64 %result, i64 addrspace(1)* %out, align 8
49 ; SI-LABEL: @private_access_v2i64_alloca:
54 define void @private_access_v2i64_alloca(<2 x i64> addrspace(1)* noalias %out, <2 x i64> addrspace(1)* noalias %in, i32 %b) nounwind {
55 %val = load <2 x i64> addrspace(1)* %in, align 16
56 %array = alloca <2 x i64>, i32 16, align 16
57 %ptr = getelementptr <2 x i64>* %array, i32 %b
58 store <2 x i64> %val, <2 x i64>* %ptr, align 16
59 call void @llvm.AMDGPU.barrier.local() noduplicate nounwind
60 %result = load <2 x i64>* %ptr, align 16
61 store <2 x i64> %result, <2 x i64> addrspace(1)* %out, align 16