1; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI %s 2; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=VI %s 3 4; GCN-LABEL: {{^}}s_sext_i1_to_i32: 5; GCN: v_cndmask_b32_e64 6; GCN: s_endpgm 7define void @s_sext_i1_to_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 8 %cmp = icmp eq i32 %a, %b 9 %sext = sext i1 %cmp to i32 10 store i32 %sext, i32 addrspace(1)* %out, align 4 11 ret void 12} 13 14; GCN-LABEL: {{^}}test_s_sext_i32_to_i64: 15; GCN: s_ashr_i32 16; GCN: s_endpg 17define void @test_s_sext_i32_to_i64(i64 addrspace(1)* %out, i32 %a, i32 %b, i32 %c) nounwind { 18entry: 19 %mul = mul i32 %a, %b 20 %add = add i32 %mul, %c 21 %sext = sext i32 %add to i64 22 store i64 %sext, i64 addrspace(1)* %out, align 8 23 ret void 24} 25 26; GCN-LABEL: {{^}}s_sext_i1_to_i64: 27; GCN: v_cndmask_b32_e64 v[[LOREG:[0-9]+]], 0, -1, vcc 28; GCN: v_mov_b32_e32 v[[HIREG:[0-9]+]], v[[LOREG]] 29; GCN: buffer_store_dwordx2 v{{\[}}[[LOREG]]:[[HIREG]]{{\]}} 30; GCN: s_endpgm 31define void @s_sext_i1_to_i64(i64 addrspace(1)* %out, i32 %a, i32 %b) nounwind { 32 %cmp = icmp eq i32 %a, %b 33 %sext = sext i1 %cmp to i64 34 store i64 %sext, i64 addrspace(1)* %out, align 8 35 ret void 36} 37 38; GCN-LABEL: {{^}}s_sext_i32_to_i64: 39; GCN: s_ashr_i32 40; GCN: s_endpgm 41define void @s_sext_i32_to_i64(i64 addrspace(1)* %out, i32 %a) nounwind { 42 %sext = sext i32 %a to i64 43 store i64 %sext, i64 addrspace(1)* %out, align 8 44 ret void 45} 46 47; GCN-LABEL: {{^}}v_sext_i32_to_i64: 48; GCN: v_ashr 49; GCN: s_endpgm 50define void @v_sext_i32_to_i64(i64 addrspace(1)* %out, i32 addrspace(1)* %in) nounwind { 51 %val = load i32, i32 addrspace(1)* %in, align 4 52 %sext = sext i32 %val to i64 53 store i64 %sext, i64 addrspace(1)* %out, align 8 54 ret void 55} 56 57; GCN-LABEL: {{^}}s_sext_i16_to_i64: 58; GCN: s_endpgm 59define void @s_sext_i16_to_i64(i64 addrspace(1)* %out, i16 %a) nounwind { 60 %sext = sext i16 %a to i64 61 store i64 %sext, i64 addrspace(1)* %out, align 8 62 ret void 63} 64 65; GCN-LABEL: {{^}}s_sext_v4i8_to_v4i32: 66; GCN: s_load_dword [[VAL:s[0-9]+]] 67; GCN-DAG: s_sext_i32_i8 [[EXT0:s[0-9]+]], [[VAL]] 68; GCN-DAG: s_bfe_i32 [[EXT1:s[0-9]+]], [[VAL]], 0x80008 69; GCN-DAG: s_bfe_i32 [[EXT2:s[0-9]+]], [[VAL]], 0x80010 70; GCN-DAG: s_ashr_i32 [[EXT3:s[0-9]+]], [[VAL]], 24 71 72; GCN-DAG: v_mov_b32_e32 [[VEXT0:v[0-9]+]], [[EXT0]] 73; GCN-DAG: v_mov_b32_e32 [[VEXT1:v[0-9]+]], [[EXT1]] 74; GCN-DAG: v_mov_b32_e32 [[VEXT2:v[0-9]+]], [[EXT2]] 75; GCN-DAG: v_mov_b32_e32 [[VEXT3:v[0-9]+]], [[EXT3]] 76 77; GCN-DAG: buffer_store_dword [[VEXT0]] 78; GCN-DAG: buffer_store_dword [[VEXT1]] 79; GCN-DAG: buffer_store_dword [[VEXT2]] 80; GCN-DAG: buffer_store_dword [[VEXT3]] 81 82; GCN: s_endpgm 83define void @s_sext_v4i8_to_v4i32(i32 addrspace(1)* %out, i32 %a) nounwind { 84 %cast = bitcast i32 %a to <4 x i8> 85 %ext = sext <4 x i8> %cast to <4 x i32> 86 %elt0 = extractelement <4 x i32> %ext, i32 0 87 %elt1 = extractelement <4 x i32> %ext, i32 1 88 %elt2 = extractelement <4 x i32> %ext, i32 2 89 %elt3 = extractelement <4 x i32> %ext, i32 3 90 store volatile i32 %elt0, i32 addrspace(1)* %out 91 store volatile i32 %elt1, i32 addrspace(1)* %out 92 store volatile i32 %elt2, i32 addrspace(1)* %out 93 store volatile i32 %elt3, i32 addrspace(1)* %out 94 ret void 95} 96 97; GCN-LABEL: {{^}}v_sext_v4i8_to_v4i32: 98; GCN: buffer_load_dword [[VAL:v[0-9]+]] 99; GCN-DAG: v_bfe_i32 [[EXT0:v[0-9]+]], [[VAL]], 0, 8 100; GCN-DAG: v_bfe_i32 [[EXT1:v[0-9]+]], [[VAL]], 8, 8 101; GCN-DAG: v_bfe_i32 [[EXT2:v[0-9]+]], [[VAL]], 16, 8 102; GCN-DAG: v_ashrrev_i32_e32 [[EXT3:v[0-9]+]], 24, [[VAL]] 103 104; GCN: buffer_store_dword [[EXT0]] 105; GCN: buffer_store_dword [[EXT1]] 106; GCN: buffer_store_dword [[EXT2]] 107; GCN: buffer_store_dword [[EXT3]] 108define void @v_sext_v4i8_to_v4i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) nounwind { 109 %a = load i32, i32 addrspace(1)* %in 110 %cast = bitcast i32 %a to <4 x i8> 111 %ext = sext <4 x i8> %cast to <4 x i32> 112 %elt0 = extractelement <4 x i32> %ext, i32 0 113 %elt1 = extractelement <4 x i32> %ext, i32 1 114 %elt2 = extractelement <4 x i32> %ext, i32 2 115 %elt3 = extractelement <4 x i32> %ext, i32 3 116 store volatile i32 %elt0, i32 addrspace(1)* %out 117 store volatile i32 %elt1, i32 addrspace(1)* %out 118 store volatile i32 %elt2, i32 addrspace(1)* %out 119 store volatile i32 %elt3, i32 addrspace(1)* %out 120 ret void 121} 122 123; FIXME: s_bfe_i64 124; GCN-LABEL: {{^}}s_sext_v4i16_to_v4i32: 125; GCN-DAG: s_ashr_i64 s{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, 48 126; GCN-DAG: s_ashr_i32 s{{[0-9]+}}, s{{[0-9]+}}, 16 127; GCN-DAG: s_sext_i32_i16 128; GCN-DAG: s_sext_i32_i16 129; GCN: s_endpgm 130define void @s_sext_v4i16_to_v4i32(i32 addrspace(1)* %out, i64 %a) nounwind { 131 %cast = bitcast i64 %a to <4 x i16> 132 %ext = sext <4 x i16> %cast to <4 x i32> 133 %elt0 = extractelement <4 x i32> %ext, i32 0 134 %elt1 = extractelement <4 x i32> %ext, i32 1 135 %elt2 = extractelement <4 x i32> %ext, i32 2 136 %elt3 = extractelement <4 x i32> %ext, i32 3 137 store volatile i32 %elt0, i32 addrspace(1)* %out 138 store volatile i32 %elt1, i32 addrspace(1)* %out 139 store volatile i32 %elt2, i32 addrspace(1)* %out 140 store volatile i32 %elt3, i32 addrspace(1)* %out 141 ret void 142} 143 144; GCN-LABEL: {{^}}v_sext_v4i16_to_v4i32: 145; SI-DAG: v_ashr_i64 v{{\[[0-9]+:[0-9]+\]}}, v{{\[[0-9]+:[0-9]+\]}}, 48 146; VI-DAG: v_ashrrev_i64 v{{\[[0-9]+:[0-9]+\]}}, 48, v{{\[[0-9]+:[0-9]+\]}} 147; GCN-DAG: v_ashrrev_i32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}} 148; GCN-DAG: v_ashrrev_i32_e32 v{{[0-9]+}}, 16, v{{[0-9]+}} 149; GCN-DAG: v_bfe_i32 v{{[0-9]+}}, v{{[0-9]+}}, 0, 16 150; GCN-DAG: v_bfe_i32 v{{[0-9]+}}, v{{[0-9]+}}, 0, 16 151; GCN: s_endpgm 152define void @v_sext_v4i16_to_v4i32(i32 addrspace(1)* %out, i64 addrspace(1)* %in) nounwind { 153 %a = load i64, i64 addrspace(1)* %in 154 %cast = bitcast i64 %a to <4 x i16> 155 %ext = sext <4 x i16> %cast to <4 x i32> 156 %elt0 = extractelement <4 x i32> %ext, i32 0 157 %elt1 = extractelement <4 x i32> %ext, i32 1 158 %elt2 = extractelement <4 x i32> %ext, i32 2 159 %elt3 = extractelement <4 x i32> %ext, i32 3 160 store volatile i32 %elt0, i32 addrspace(1)* %out 161 store volatile i32 %elt1, i32 addrspace(1)* %out 162 store volatile i32 %elt2, i32 addrspace(1)* %out 163 store volatile i32 %elt3, i32 addrspace(1)* %out 164 ret void 165} 166