1; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py 2; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck %s 3; RUN: llc -global-isel -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx810 -stop-after=instruction-select -verify-machineinstrs -o - %s | FileCheck %s 4 5; Natural mapping 6define amdgpu_ps float @struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 7 ; CHECK-LABEL: name: struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 8 ; CHECK: bb.1 (%ir-block.0): 9 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 10 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 11 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 12 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 13 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 14 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 15 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 16 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 17 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 18 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 19 ; CHECK: [[BUFFER_LOAD_DWORD_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4) 20 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_BOTHEN]] 21 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 22 %val = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 23 ret float %val 24} 25 26; Natural mapping 27define amdgpu_ps <2 x float> @struct_buffer_load_v2f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 28 ; CHECK-LABEL: name: struct_buffer_load_v2f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 29 ; CHECK: bb.1 (%ir-block.0): 30 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 31 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 32 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 33 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 34 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 35 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 36 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 37 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 38 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 39 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 40 ; CHECK: [[BUFFER_LOAD_DWORDX2_BOTHEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_DWORDX2_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8 from custom "TargetCustom7", align 1, addrspace 4) 41 ; CHECK: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_BOTHEN]].sub0 42 ; CHECK: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_BOTHEN]].sub1 43 ; CHECK: $vgpr0 = COPY [[COPY7]] 44 ; CHECK: $vgpr1 = COPY [[COPY8]] 45 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1 46 %val = call <2 x float> @llvm.amdgcn.struct.buffer.load.v2f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 47 ret <2 x float> %val 48} 49 50; Natural mapping 51define amdgpu_ps <3 x float> @struct_buffer_load_v3f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 52 ; CHECK-LABEL: name: struct_buffer_load_v3f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 53 ; CHECK: bb.1 (%ir-block.0): 54 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 55 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 56 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 57 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 58 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 59 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 60 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 61 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 62 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 63 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 64 ; CHECK: [[BUFFER_LOAD_DWORDX3_BOTHEN:%[0-9]+]]:vreg_96 = BUFFER_LOAD_DWORDX3_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 12 from custom "TargetCustom7", align 1, addrspace 4) 65 ; CHECK: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_BOTHEN]].sub0 66 ; CHECK: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_BOTHEN]].sub1 67 ; CHECK: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX3_BOTHEN]].sub2 68 ; CHECK: $vgpr0 = COPY [[COPY7]] 69 ; CHECK: $vgpr1 = COPY [[COPY8]] 70 ; CHECK: $vgpr2 = COPY [[COPY9]] 71 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2 72 %val = call <3 x float> @llvm.amdgcn.struct.buffer.load.v3f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 73 ret <3 x float> %val 74} 75 76; Natural mapping 77define amdgpu_ps <4 x float> @struct_buffer_load_v4f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 78 ; CHECK-LABEL: name: struct_buffer_load_v4f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 79 ; CHECK: bb.1 (%ir-block.0): 80 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 81 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 82 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 83 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 84 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 85 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 86 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 87 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 88 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 89 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 90 ; CHECK: [[BUFFER_LOAD_DWORDX4_BOTHEN:%[0-9]+]]:vreg_128 = BUFFER_LOAD_DWORDX4_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 16 from custom "TargetCustom7", align 1, addrspace 4) 91 ; CHECK: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_BOTHEN]].sub0 92 ; CHECK: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_BOTHEN]].sub1 93 ; CHECK: [[COPY9:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_BOTHEN]].sub2 94 ; CHECK: [[COPY10:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX4_BOTHEN]].sub3 95 ; CHECK: $vgpr0 = COPY [[COPY7]] 96 ; CHECK: $vgpr1 = COPY [[COPY8]] 97 ; CHECK: $vgpr2 = COPY [[COPY9]] 98 ; CHECK: $vgpr3 = COPY [[COPY10]] 99 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1, implicit $vgpr2, implicit $vgpr3 100 %val = call <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 101 ret <4 x float> %val 102} 103 104; Natural mapping 105define amdgpu_ps float @struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_vindex0(<4 x i32> inreg %rsrc, i32 %voffset, i32 inreg %soffset) { 106 ; CHECK-LABEL: name: struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_vindex0 107 ; CHECK: bb.1 (%ir-block.0): 108 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0 109 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 110 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 111 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 112 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 113 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 114 ; CHECK: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr6 115 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 116 ; CHECK: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 0 117 ; CHECK: [[COPY6:%[0-9]+]]:vgpr_32 = COPY [[S_MOV_B32_]] 118 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY6]], %subreg.sub0, [[COPY4]], %subreg.sub1 119 ; CHECK: [[BUFFER_LOAD_DWORD_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY5]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4) 120 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_BOTHEN]] 121 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 122 %val = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> %rsrc, i32 0, i32 %voffset, i32 %soffset, i32 0) 123 ret float %val 124} 125 126; Natural mapping 127define amdgpu_ps float @struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_voffset_add4095(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset.base, i32 inreg %soffset) { 128 ; CHECK-LABEL: name: struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_voffset_add4095 129 ; CHECK: bb.1 (%ir-block.0): 130 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 131 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 132 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 133 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 134 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 135 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 136 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 137 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 138 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 139 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 140 ; CHECK: [[BUFFER_LOAD_DWORD_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 4095, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7" + 4095, align 1, addrspace 4) 141 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_BOTHEN]] 142 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 143 %voffset = add i32 %voffset.base, 4095 144 %val = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 145 ret float %val 146} 147 148define amdgpu_ps float @struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_soffset_64(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset) { 149 ; CHECK-LABEL: name: struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_soffset_64 150 ; CHECK: bb.1 (%ir-block.0): 151 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $vgpr0, $vgpr1 152 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 153 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 154 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 155 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 156 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 157 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 158 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 159 ; CHECK: [[S_MOV_B32_:%[0-9]+]]:sreg_32 = S_MOV_B32 64 160 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 161 ; CHECK: [[BUFFER_LOAD_DWORD_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[S_MOV_B32_]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4) 162 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_BOTHEN]] 163 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 164 %val = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 64, i32 0) 165 ret float %val 166} 167 168; Need to legalize all reg operands 169define amdgpu_ps float @struct_buffer_load_f32__vgpr_rsrc__sgpr_vindex__sgpr_voffset__vgpr_soffset(<4 x i32> %rsrc, i32 inreg %vindex, i32 inreg %voffset, i32 %soffset) { 170 ; CHECK-LABEL: name: struct_buffer_load_f32__vgpr_rsrc__sgpr_vindex__sgpr_voffset__vgpr_soffset 171 ; CHECK: bb.1 (%ir-block.0): 172 ; CHECK: successors: %bb.2(0x80000000) 173 ; CHECK: liveins: $sgpr2, $sgpr3, $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4 174 ; CHECK: [[COPY:%[0-9]+]]:vgpr_32 = COPY $vgpr0 175 ; CHECK: [[COPY1:%[0-9]+]]:vgpr_32 = COPY $vgpr1 176 ; CHECK: [[COPY2:%[0-9]+]]:vgpr_32 = COPY $vgpr2 177 ; CHECK: [[COPY3:%[0-9]+]]:vgpr_32 = COPY $vgpr3 178 ; CHECK: [[COPY4:%[0-9]+]]:sreg_32 = COPY $sgpr2 179 ; CHECK: [[COPY5:%[0-9]+]]:sreg_32 = COPY $sgpr3 180 ; CHECK: [[COPY6:%[0-9]+]]:vgpr_32 = COPY $vgpr4 181 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:vreg_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 182 ; CHECK: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[COPY4]] 183 ; CHECK: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[COPY5]] 184 ; CHECK: [[COPY9:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub0_sub1 185 ; CHECK: [[COPY10:%[0-9]+]]:vreg_64 = COPY [[REG_SEQUENCE]].sub2_sub3 186 ; CHECK: [[S_MOV_B64_term:%[0-9]+]]:sreg_64_xexec = S_MOV_B64_term $exec 187 ; CHECK: bb.2: 188 ; CHECK: successors: %bb.3(0x40000000), %bb.2(0x40000000) 189 ; CHECK: [[V_READFIRSTLANE_B32_:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub0, implicit $exec 190 ; CHECK: [[V_READFIRSTLANE_B32_1:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY9]].sub1, implicit $exec 191 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1 192 ; CHECK: [[V_CMP_EQ_U64_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE1]], [[COPY9]], implicit $exec 193 ; CHECK: [[V_READFIRSTLANE_B32_2:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY10]].sub0, implicit $exec 194 ; CHECK: [[V_READFIRSTLANE_B32_3:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY10]].sub1, implicit $exec 195 ; CHECK: [[REG_SEQUENCE2:%[0-9]+]]:sreg_64_xexec = REG_SEQUENCE [[V_READFIRSTLANE_B32_2]], %subreg.sub0, [[V_READFIRSTLANE_B32_3]], %subreg.sub1 196 ; CHECK: [[V_CMP_EQ_U64_e64_1:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U64_e64 [[REG_SEQUENCE2]], [[COPY10]], implicit $exec 197 ; CHECK: [[S_AND_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U64_e64_1]], [[V_CMP_EQ_U64_e64_]], implicit-def $scc 198 ; CHECK: [[REG_SEQUENCE3:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[V_READFIRSTLANE_B32_]], %subreg.sub0, [[V_READFIRSTLANE_B32_1]], %subreg.sub1, [[V_READFIRSTLANE_B32_2]], %subreg.sub2, [[V_READFIRSTLANE_B32_3]], %subreg.sub3 199 ; CHECK: [[V_READFIRSTLANE_B32_4:%[0-9]+]]:sreg_32_xm0 = V_READFIRSTLANE_B32 [[COPY6]], implicit $exec 200 ; CHECK: [[V_CMP_EQ_U32_e64_:%[0-9]+]]:sreg_64_xexec = V_CMP_EQ_U32_e64 [[V_READFIRSTLANE_B32_4]], [[COPY6]], implicit $exec 201 ; CHECK: [[S_AND_B64_1:%[0-9]+]]:sreg_64_xexec = S_AND_B64 [[V_CMP_EQ_U32_e64_]], [[S_AND_B64_]], implicit-def $scc 202 ; CHECK: [[REG_SEQUENCE4:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY7]], %subreg.sub0, [[COPY8]], %subreg.sub1 203 ; CHECK: [[BUFFER_LOAD_DWORD_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_BOTHEN [[REG_SEQUENCE4]], [[REG_SEQUENCE3]], [[V_READFIRSTLANE_B32_4]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4) 204 ; CHECK: [[S_AND_SAVEEXEC_B64_:%[0-9]+]]:sreg_64_xexec = S_AND_SAVEEXEC_B64 killed [[S_AND_B64_1]], implicit-def $exec, implicit-def $scc, implicit $exec 205 ; CHECK: $exec = S_XOR_B64_term $exec, [[S_AND_SAVEEXEC_B64_]], implicit-def $scc 206 ; CHECK: S_CBRANCH_EXECNZ %bb.2, implicit $exec 207 ; CHECK: bb.3: 208 ; CHECK: successors: %bb.4(0x80000000) 209 ; CHECK: $exec = S_MOV_B64_term [[S_MOV_B64_term]] 210 ; CHECK: bb.4: 211 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_BOTHEN]] 212 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 213 %val = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 214 ret float %val 215} 216 217define amdgpu_ps float @struct_buffer_load_i8_zext__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 218 ; CHECK-LABEL: name: struct_buffer_load_i8_zext__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 219 ; CHECK: bb.1 (%ir-block.0): 220 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 221 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 222 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 223 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 224 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 225 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 226 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 227 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 228 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 229 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 230 ; CHECK: [[BUFFER_LOAD_UBYTE_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 1 from custom "TargetCustom7", addrspace 4) 231 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_UBYTE_BOTHEN]] 232 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 233 %val = call i8 @llvm.amdgcn.struct.buffer.load.i8(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 234 %ext = zext i8 %val to i32 235 %cast = bitcast i32 %ext to float 236 ret float %cast 237} 238 239define amdgpu_ps float @struct_buffer_load_i8_sext__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 240 ; CHECK-LABEL: name: struct_buffer_load_i8_sext__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 241 ; CHECK: bb.1 (%ir-block.0): 242 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 243 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 244 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 245 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 246 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 247 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 248 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 249 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 250 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 251 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 252 ; CHECK: [[BUFFER_LOAD_UBYTE_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_UBYTE_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 1 from custom "TargetCustom7", addrspace 4) 253 ; CHECK: [[V_BFE_I32_:%[0-9]+]]:vgpr_32 = V_BFE_I32 [[BUFFER_LOAD_UBYTE_BOTHEN]], 0, 8, implicit $exec 254 ; CHECK: $vgpr0 = COPY [[V_BFE_I32_]] 255 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 256 %val = call i8 @llvm.amdgcn.struct.buffer.load.i8(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 257 %ext = sext i8 %val to i32 258 %cast = bitcast i32 %ext to float 259 ret float %cast 260} 261 262define amdgpu_ps float @struct_buffer_load_i16_zext__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 263 ; CHECK-LABEL: name: struct_buffer_load_i16_zext__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 264 ; CHECK: bb.1 (%ir-block.0): 265 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 266 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 267 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 268 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 269 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 270 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 271 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 272 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 273 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 274 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 275 ; CHECK: [[BUFFER_LOAD_USHORT_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_USHORT_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4) 276 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_USHORT_BOTHEN]] 277 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 278 %val = call i16 @llvm.amdgcn.struct.buffer.load.i16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 279 %ext = zext i16 %val to i32 280 %cast = bitcast i32 %ext to float 281 ret float %cast 282} 283 284define amdgpu_ps float @struct_buffer_load_i16_sext__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 285 ; CHECK-LABEL: name: struct_buffer_load_i16_sext__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 286 ; CHECK: bb.1 (%ir-block.0): 287 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 288 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 289 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 290 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 291 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 292 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 293 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 294 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 295 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 296 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 297 ; CHECK: [[BUFFER_LOAD_USHORT_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_USHORT_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4) 298 ; CHECK: [[V_BFE_I32_:%[0-9]+]]:vgpr_32 = V_BFE_I32 [[BUFFER_LOAD_USHORT_BOTHEN]], 0, 16, implicit $exec 299 ; CHECK: $vgpr0 = COPY [[V_BFE_I32_]] 300 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 301 %val = call i16 @llvm.amdgcn.struct.buffer.load.i16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 302 %ext = sext i16 %val to i32 303 %cast = bitcast i32 %ext to float 304 ret float %cast 305} 306 307; Natural mapping 308define amdgpu_ps half @struct_buffer_load_f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 309 ; CHECK-LABEL: name: struct_buffer_load_f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 310 ; CHECK: bb.1 (%ir-block.0): 311 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 312 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 313 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 314 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 315 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 316 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 317 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 318 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 319 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 320 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 321 ; CHECK: [[BUFFER_LOAD_USHORT_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_USHORT_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 2 from custom "TargetCustom7", align 1, addrspace 4) 322 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_USHORT_BOTHEN]] 323 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 324 %val = call half @llvm.amdgcn.struct.buffer.load.f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 325 ret half %val 326} 327 328; Natural mapping 329define amdgpu_ps <2 x half> @struct_buffer_load_v2f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 330 ; CHECK-LABEL: name: struct_buffer_load_v2f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 331 ; CHECK: bb.1 (%ir-block.0): 332 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 333 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 334 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 335 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 336 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 337 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 338 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 339 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 340 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 341 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 342 ; CHECK: [[BUFFER_LOAD_DWORD_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4) 343 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_BOTHEN]] 344 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 345 %val = call <2 x half> @llvm.amdgcn.struct.buffer.load.v2f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 346 ret <2 x half> %val 347} 348 349; FIXME: Crashes 350; define amdgpu_ps <3 x half> @struct_buffer_load_v3f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 351; %val = call <3 x half> @llvm.amdgcn.struct.buffer.load.v3f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 352; ret <3 x half> %val 353; } 354 355; Natural mapping 356define amdgpu_ps <4 x half> @struct_buffer_load_v4f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 357 ; CHECK-LABEL: name: struct_buffer_load_v4f16__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset 358 ; CHECK: bb.1 (%ir-block.0): 359 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 360 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 361 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 362 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 363 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 364 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 365 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 366 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 367 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 368 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 369 ; CHECK: [[BUFFER_LOAD_DWORDX2_BOTHEN:%[0-9]+]]:vreg_64 = BUFFER_LOAD_DWORDX2_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8 from custom "TargetCustom7", align 1, addrspace 4) 370 ; CHECK: [[COPY7:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_BOTHEN]].sub0 371 ; CHECK: [[COPY8:%[0-9]+]]:vgpr_32 = COPY [[BUFFER_LOAD_DWORDX2_BOTHEN]].sub1 372 ; CHECK: $vgpr0 = COPY [[COPY7]] 373 ; CHECK: $vgpr1 = COPY [[COPY8]] 374 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0, implicit $vgpr1 375 %val = call <4 x half> @llvm.amdgcn.struct.buffer.load.v4f16(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 0) 376 ret <4 x half> %val 377} 378 379; Natural mapping + glc 380define amdgpu_ps float @struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_glc(<4 x i32> inreg %rsrc, i32 %vindex, i32 %voffset, i32 inreg %soffset) { 381 ; CHECK-LABEL: name: struct_buffer_load_f32__sgpr_rsrc__vgpr_vindex__vgpr_voffset__sgpr_soffset_glc 382 ; CHECK: bb.1 (%ir-block.0): 383 ; CHECK: liveins: $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $vgpr0, $vgpr1 384 ; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY $sgpr2 385 ; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY $sgpr3 386 ; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY $sgpr4 387 ; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY $sgpr5 388 ; CHECK: [[COPY4:%[0-9]+]]:vgpr_32 = COPY $vgpr0 389 ; CHECK: [[COPY5:%[0-9]+]]:vgpr_32 = COPY $vgpr1 390 ; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY $sgpr6 391 ; CHECK: [[REG_SEQUENCE:%[0-9]+]]:sgpr_128 = REG_SEQUENCE [[COPY]], %subreg.sub0, [[COPY1]], %subreg.sub1, [[COPY2]], %subreg.sub2, [[COPY3]], %subreg.sub3 392 ; CHECK: [[REG_SEQUENCE1:%[0-9]+]]:vreg_64 = REG_SEQUENCE [[COPY4]], %subreg.sub0, [[COPY5]], %subreg.sub1 393 ; CHECK: [[BUFFER_LOAD_DWORD_BOTHEN:%[0-9]+]]:vgpr_32 = BUFFER_LOAD_DWORD_BOTHEN [[REG_SEQUENCE1]], [[REG_SEQUENCE]], [[COPY6]], 0, 1, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from custom "TargetCustom7", align 1, addrspace 4) 394 ; CHECK: $vgpr0 = COPY [[BUFFER_LOAD_DWORD_BOTHEN]] 395 ; CHECK: SI_RETURN_TO_EPILOG implicit $vgpr0 396 %val = call float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32> %rsrc, i32 %vindex, i32 %voffset, i32 %soffset, i32 1) 397 ret float %val 398} 399 400declare i8 @llvm.amdgcn.struct.buffer.load.i8(<4 x i32>, i32, i32, i32, i32 immarg) #0 401declare i16 @llvm.amdgcn.struct.buffer.load.i16(<4 x i32>, i32, i32, i32, i32 immarg) #0 402declare float @llvm.amdgcn.struct.buffer.load.f32(<4 x i32>, i32, i32, i32, i32 immarg) #0 403declare <2 x float> @llvm.amdgcn.struct.buffer.load.v2f32(<4 x i32>, i32, i32, i32, i32 immarg) #0 404declare <3 x float> @llvm.amdgcn.struct.buffer.load.v3f32(<4 x i32>, i32, i32, i32, i32 immarg) #0 405declare <4 x float> @llvm.amdgcn.struct.buffer.load.v4f32(<4 x i32>, i32, i32, i32, i32 immarg) #0 406declare half @llvm.amdgcn.struct.buffer.load.f16(<4 x i32>, i32, i32, i32, i32 immarg) #0 407declare <2 x half> @llvm.amdgcn.struct.buffer.load.v2f16(<4 x i32>, i32, i32, i32, i32 immarg) #0 408declare <3 x half> @llvm.amdgcn.struct.buffer.load.v3f16(<4 x i32>, i32, i32, i32, i32 immarg) #0 409declare <4 x half> @llvm.amdgcn.struct.buffer.load.v4f16(<4 x i32>, i32, i32, i32, i32 immarg) #0 410 411attributes #0 = { nounwind readonly } 412