1 ; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s 2 ; XUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s 3 4 5 @a = internal addrspace(2) constant [1 x i8] [ i8 7 ], align 1 6 7 ; FUNC-LABEL: {{^}}test_i8: 8 ; EG: CF_END 9 ; SI: buffer_store_byte 10 ; SI: s_endpgm 11 define void @test_i8( i32 %s, i8 addrspace(1)* %out) #3 { 12 %arrayidx = getelementptr inbounds [1 x i8], [1 x i8] addrspace(2)* @a, i32 0, i32 %s 13 %1 = load i8, i8 addrspace(2)* %arrayidx, align 1 14 store i8 %1, i8 addrspace(1)* %out 15 ret void 16 } 17 18 @b = internal addrspace(2) constant [1 x i16] [ i16 7 ], align 2 19 20 ; FUNC-LABEL: {{^}}test_i16: 21 ; EG: CF_END 22 ; SI: buffer_store_short 23 ; SI: s_endpgm 24 define void @test_i16( i32 %s, i16 addrspace(1)* %out) #3 { 25 %arrayidx = getelementptr inbounds [1 x i16], [1 x i16] addrspace(2)* @b, i32 0, i32 %s 26 %1 = load i16, i16 addrspace(2)* %arrayidx, align 2 27 store i16 %1, i16 addrspace(1)* %out 28 ret void 29 } 30 31 %struct.bar = type { float, [5 x i8] } 32 33 ; The illegal i8s aren't handled 34 @struct_bar_gv = internal addrspace(2) constant [1 x %struct.bar] [ %struct.bar { float 16.0, [5 x i8] [i8 0, i8 1, i8 2, i8 3, i8 4] } ] 35 36 ; FUNC-LABEL: {{^}}struct_bar_gv_load: 37 define void @struct_bar_gv_load(i8 addrspace(1)* %out, i32 %index) { 38 %gep = getelementptr inbounds [1 x %struct.bar], [1 x %struct.bar] addrspace(2)* @struct_bar_gv, i32 0, i32 0, i32 1, i32 %index 39 %load = load i8, i8 addrspace(2)* %gep, align 1 40 store i8 %load, i8 addrspace(1)* %out, align 1 41 ret void 42 } 43 44 45 ; The private load isn't scalarzied. 46 @array_vector_gv = internal addrspace(2) constant [4 x <4 x i32>] [ <4 x i32> <i32 1, i32 2, i32 3, i32 4>, 47 <4 x i32> <i32 5, i32 6, i32 7, i32 8>, 48 <4 x i32> <i32 9, i32 10, i32 11, i32 12>, 49 <4 x i32> <i32 13, i32 14, i32 15, i32 16> ] 50 51 ; FUNC-LABEL: {{^}}array_vector_gv_load: 52 define void @array_vector_gv_load(<4 x i32> addrspace(1)* %out, i32 %index) { 53 %gep = getelementptr inbounds [4 x <4 x i32>], [4 x <4 x i32>] addrspace(2)* @array_vector_gv, i32 0, i32 %index 54 %load = load <4 x i32>, <4 x i32> addrspace(2)* %gep, align 16 55 store <4 x i32> %load, <4 x i32> addrspace(1)* %out, align 16 56 ret void 57 } 58
