1 ; RUN: llc -march=r600 < %s | FileCheck %s 2 3 ; CHECK: DOT4 * T{{[0-9]\.W}} (MASKED) 4 ; CHECK: MAX T{{[0-9].[XYZW]}}, 0.0, PV.X 5 define amdgpu_vs void @main(<4 x float> inreg %reg0, <4 x float> inreg %reg1, <4 x float> inreg %reg2, <4 x float> inreg %reg3, <4 x float> inreg %reg4, <4 x float> inreg %reg5, <4 x float> inreg %reg6, <4 x float> inreg %reg7) { 6 main_body: 7 %tmp = extractelement <4 x float> %reg1, i32 0 8 %tmp13 = extractelement <4 x float> %reg1, i32 1 9 %tmp14 = extractelement <4 x float> %reg1, i32 2 10 %tmp15 = extractelement <4 x float> %reg1, i32 3 11 %tmp16 = extractelement <4 x float> %reg2, i32 0 12 %tmp17 = extractelement <4 x float> %reg2, i32 1 13 %tmp18 = extractelement <4 x float> %reg2, i32 2 14 %tmp19 = extractelement <4 x float> %reg2, i32 3 15 %tmp20 = extractelement <4 x float> %reg3, i32 0 16 %tmp21 = extractelement <4 x float> %reg3, i32 1 17 %tmp22 = extractelement <4 x float> %reg3, i32 2 18 %tmp23 = extractelement <4 x float> %reg3, i32 3 19 %tmp24 = extractelement <4 x float> %reg4, i32 0 20 %tmp25 = extractelement <4 x float> %reg4, i32 1 21 %tmp26 = extractelement <4 x float> %reg4, i32 2 22 %tmp27 = extractelement <4 x float> %reg4, i32 3 23 %tmp28 = extractelement <4 x float> %reg5, i32 0 24 %tmp29 = extractelement <4 x float> %reg5, i32 1 25 %tmp30 = extractelement <4 x float> %reg5, i32 2 26 %tmp31 = extractelement <4 x float> %reg5, i32 3 27 %tmp32 = extractelement <4 x float> %reg6, i32 0 28 %tmp33 = extractelement <4 x float> %reg6, i32 1 29 %tmp34 = extractelement <4 x float> %reg6, i32 2 30 %tmp35 = extractelement <4 x float> %reg6, i32 3 31 %tmp36 = extractelement <4 x float> %reg7, i32 0 32 %tmp37 = extractelement <4 x float> %reg7, i32 1 33 %tmp38 = extractelement <4 x float> %reg7, i32 2 34 %tmp39 = extractelement <4 x float> %reg7, i32 3 35 %tmp40 = load <4 x float>, <4 x float> addrspace(8)* null 36 %tmp41 = extractelement <4 x float> %tmp40, i32 0 37 %tmp42 = fmul float %tmp, %tmp41 38 %tmp43 = load <4 x float>, <4 x float> addrspace(8)* null 39 %tmp44 = extractelement <4 x float> %tmp43, i32 1 40 %tmp45 = fmul float %tmp, %tmp44 41 %tmp46 = load <4 x float>, <4 x float> addrspace(8)* null 42 %tmp47 = extractelement <4 x float> %tmp46, i32 2 43 %tmp48 = fmul float %tmp, %tmp47 44 %tmp49 = load <4 x float>, <4 x float> addrspace(8)* null 45 %tmp50 = extractelement <4 x float> %tmp49, i32 3 46 %tmp51 = fmul float %tmp, %tmp50 47 %tmp52 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 1) 48 %tmp53 = extractelement <4 x float> %tmp52, i32 0 49 %tmp54 = fmul float %tmp13, %tmp53 50 %tmp55 = fadd float %tmp54, %tmp42 51 %tmp56 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 1) 52 %tmp57 = extractelement <4 x float> %tmp56, i32 1 53 %tmp58 = fmul float %tmp13, %tmp57 54 %tmp59 = fadd float %tmp58, %tmp45 55 %tmp60 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 1) 56 %tmp61 = extractelement <4 x float> %tmp60, i32 2 57 %tmp62 = fmul float %tmp13, %tmp61 58 %tmp63 = fadd float %tmp62, %tmp48 59 %tmp64 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 1) 60 %tmp65 = extractelement <4 x float> %tmp64, i32 3 61 %tmp66 = fmul float %tmp13, %tmp65 62 %tmp67 = fadd float %tmp66, %tmp51 63 %tmp68 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 2) 64 %tmp69 = extractelement <4 x float> %tmp68, i32 0 65 %tmp70 = fmul float %tmp14, %tmp69 66 %tmp71 = fadd float %tmp70, %tmp55 67 %tmp72 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 2) 68 %tmp73 = extractelement <4 x float> %tmp72, i32 1 69 %tmp74 = fmul float %tmp14, %tmp73 70 %tmp75 = fadd float %tmp74, %tmp59 71 %tmp76 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 2) 72 %tmp77 = extractelement <4 x float> %tmp76, i32 2 73 %tmp78 = fmul float %tmp14, %tmp77 74 %tmp79 = fadd float %tmp78, %tmp63 75 %tmp80 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 2) 76 %tmp81 = extractelement <4 x float> %tmp80, i32 3 77 %tmp82 = fmul float %tmp14, %tmp81 78 %tmp83 = fadd float %tmp82, %tmp67 79 %tmp84 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 3) 80 %tmp85 = extractelement <4 x float> %tmp84, i32 0 81 %tmp86 = fmul float %tmp15, %tmp85 82 %tmp87 = fadd float %tmp86, %tmp71 83 %tmp88 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 3) 84 %tmp89 = extractelement <4 x float> %tmp88, i32 1 85 %tmp90 = fmul float %tmp15, %tmp89 86 %tmp91 = fadd float %tmp90, %tmp75 87 %tmp92 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 3) 88 %tmp93 = extractelement <4 x float> %tmp92, i32 2 89 %tmp94 = fmul float %tmp15, %tmp93 90 %tmp95 = fadd float %tmp94, %tmp79 91 %tmp96 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 3) 92 %tmp97 = extractelement <4 x float> %tmp96, i32 3 93 %tmp98 = fmul float %tmp15, %tmp97 94 %tmp99 = fadd float %tmp98, %tmp83 95 %tmp100 = insertelement <4 x float> undef, float %tmp16, i32 0 96 %tmp101 = insertelement <4 x float> %tmp100, float %tmp17, i32 1 97 %tmp102 = insertelement <4 x float> %tmp101, float %tmp18, i32 2 98 %tmp103 = insertelement <4 x float> %tmp102, float 0.000000e+00, i32 3 99 %tmp104 = insertelement <4 x float> undef, float %tmp16, i32 0 100 %tmp105 = insertelement <4 x float> %tmp104, float %tmp17, i32 1 101 %tmp106 = insertelement <4 x float> %tmp105, float %tmp18, i32 2 102 %tmp107 = insertelement <4 x float> %tmp106, float 0.000000e+00, i32 3 103 %tmp108 = call float @llvm.r600.dot4(<4 x float> %tmp103, <4 x float> %tmp107) 104 %tmp109 = call float @llvm.fabs.f32(float %tmp108) 105 %tmp110 = call float @llvm.r600.recipsqrt.clamped.f32(float %tmp109) 106 %tmp111 = fmul float %tmp16, %tmp110 107 %tmp112 = fmul float %tmp17, %tmp110 108 %tmp113 = fmul float %tmp18, %tmp110 109 %tmp114 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 4) 110 %tmp115 = extractelement <4 x float> %tmp114, i32 0 111 %tmp116 = fmul float %tmp115, %tmp20 112 %tmp117 = fadd float %tmp116, %tmp32 113 %tmp118 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 4) 114 %tmp119 = extractelement <4 x float> %tmp118, i32 1 115 %tmp120 = fmul float %tmp119, %tmp21 116 %tmp121 = fadd float %tmp120, %tmp33 117 %tmp122 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 4) 118 %tmp123 = extractelement <4 x float> %tmp122, i32 2 119 %tmp124 = fmul float %tmp123, %tmp22 120 %tmp125 = fadd float %tmp124, %tmp34 121 %max.0.i = call float @llvm.maxnum.f32(float %tmp117, float 0.000000e+00) 122 %clamp.i = call float @llvm.minnum.f32(float %max.0.i, float 1.000000e+00) 123 %max.0.i11 = call float @llvm.maxnum.f32(float %tmp121, float 0.000000e+00) 124 %clamp.i12 = call float @llvm.minnum.f32(float %max.0.i11, float 1.000000e+00) 125 %max.0.i9 = call float @llvm.maxnum.f32(float %tmp125, float 0.000000e+00) 126 %clamp.i10 = call float @llvm.minnum.f32(float %max.0.i9, float 1.000000e+00) 127 %max.0.i7 = call float @llvm.maxnum.f32(float %tmp27, float 0.000000e+00) 128 %clamp.i8 = call float @llvm.minnum.f32(float %max.0.i7, float 1.000000e+00) 129 %tmp126 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 5) 130 %tmp127 = extractelement <4 x float> %tmp126, i32 0 131 %tmp128 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 5) 132 %tmp129 = extractelement <4 x float> %tmp128, i32 1 133 %tmp130 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 5) 134 %tmp131 = extractelement <4 x float> %tmp130, i32 2 135 %tmp132 = insertelement <4 x float> undef, float %tmp111, i32 0 136 %tmp133 = insertelement <4 x float> %tmp132, float %tmp112, i32 1 137 %tmp134 = insertelement <4 x float> %tmp133, float %tmp113, i32 2 138 %tmp135 = insertelement <4 x float> %tmp134, float 0.000000e+00, i32 3 139 %tmp136 = insertelement <4 x float> undef, float %tmp127, i32 0 140 %tmp137 = insertelement <4 x float> %tmp136, float %tmp129, i32 1 141 %tmp138 = insertelement <4 x float> %tmp137, float %tmp131, i32 2 142 %tmp139 = insertelement <4 x float> %tmp138, float 0.000000e+00, i32 3 143 %tmp140 = call float @llvm.r600.dot4(<4 x float> %tmp135, <4 x float> %tmp139) 144 %tmp141 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 7) 145 %tmp142 = extractelement <4 x float> %tmp141, i32 0 146 %tmp143 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 7) 147 %tmp144 = extractelement <4 x float> %tmp143, i32 1 148 %tmp145 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 7) 149 %tmp146 = extractelement <4 x float> %tmp145, i32 2 150 %tmp147 = insertelement <4 x float> undef, float %tmp111, i32 0 151 %tmp148 = insertelement <4 x float> %tmp147, float %tmp112, i32 1 152 %tmp149 = insertelement <4 x float> %tmp148, float %tmp113, i32 2 153 %tmp150 = insertelement <4 x float> %tmp149, float 0.000000e+00, i32 3 154 %tmp151 = insertelement <4 x float> undef, float %tmp142, i32 0 155 %tmp152 = insertelement <4 x float> %tmp151, float %tmp144, i32 1 156 %tmp153 = insertelement <4 x float> %tmp152, float %tmp146, i32 2 157 %tmp154 = insertelement <4 x float> %tmp153, float 0.000000e+00, i32 3 158 %tmp155 = call float @llvm.r600.dot4(<4 x float> %tmp150, <4 x float> %tmp154) 159 %tmp156 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 8) 160 %tmp157 = extractelement <4 x float> %tmp156, i32 0 161 %tmp158 = fmul float %tmp157, %tmp20 162 %tmp159 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 8) 163 %tmp160 = extractelement <4 x float> %tmp159, i32 1 164 %tmp161 = fmul float %tmp160, %tmp21 165 %tmp162 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 8) 166 %tmp163 = extractelement <4 x float> %tmp162, i32 2 167 %tmp164 = fmul float %tmp163, %tmp22 168 %tmp165 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 9) 169 %tmp166 = extractelement <4 x float> %tmp165, i32 0 170 %tmp167 = fmul float %tmp166, %tmp24 171 %tmp168 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 9) 172 %tmp169 = extractelement <4 x float> %tmp168, i32 1 173 %tmp170 = fmul float %tmp169, %tmp25 174 %tmp171 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 9) 175 %tmp172 = extractelement <4 x float> %tmp171, i32 2 176 %tmp173 = fmul float %tmp172, %tmp26 177 %tmp174 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 10) 178 %tmp175 = extractelement <4 x float> %tmp174, i32 0 179 %tmp176 = fmul float %tmp175, %tmp28 180 %tmp177 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 10) 181 %tmp178 = extractelement <4 x float> %tmp177, i32 1 182 %tmp179 = fmul float %tmp178, %tmp29 183 %tmp180 = load <4 x float>, <4 x float> addrspace(8)* getelementptr ([1024 x <4 x float>], [1024 x <4 x float>] addrspace(8)* null, i64 0, i32 10) 184 %tmp181 = extractelement <4 x float> %tmp180, i32 2 185 %tmp182 = fmul float %tmp181, %tmp30 186 %tmp183 = fcmp uge float %tmp140, 0.000000e+00 187 %tmp184 = select i1 %tmp183, float %tmp140, float 0.000000e+00 188 %tmp185 = fcmp uge float %tmp155, 0.000000e+00 189 %tmp186 = select i1 %tmp185, float %tmp155, float 0.000000e+00 190 %tmp187 = call float @llvm.pow.f32(float %tmp186, float %tmp36) 191 %tmp188 = fcmp ult float %tmp140, 0.000000e+00 192 %tmp189 = select i1 %tmp188, float 0.000000e+00, float %tmp187 193 %tmp190 = fadd float %tmp158, %tmp117 194 %tmp191 = fadd float %tmp161, %tmp121 195 %tmp192 = fadd float %tmp164, %tmp125 196 %tmp193 = fmul float %tmp184, %tmp167 197 %tmp194 = fadd float %tmp193, %tmp190 198 %tmp195 = fmul float %tmp184, %tmp170 199 %tmp196 = fadd float %tmp195, %tmp191 200 %tmp197 = fmul float %tmp184, %tmp173 201 %tmp198 = fadd float %tmp197, %tmp192 202 %tmp199 = fmul float %tmp189, %tmp176 203 %tmp200 = fadd float %tmp199, %tmp194 204 %tmp201 = fmul float %tmp189, %tmp179 205 %tmp202 = fadd float %tmp201, %tmp196 206 %tmp203 = fmul float %tmp189, %tmp182 207 %tmp204 = fadd float %tmp203, %tmp198 208 %max.0.i5 = call float @llvm.maxnum.f32(float %tmp200, float 0.000000e+00) 209 %clamp.i6 = call float @llvm.minnum.f32(float %max.0.i5, float 1.000000e+00) 210 %max.0.i3 = call float @llvm.maxnum.f32(float %tmp202, float 0.000000e+00) 211 %clamp.i4 = call float @llvm.minnum.f32(float %max.0.i3, float 1.000000e+00) 212 %max.0.i1 = call float @llvm.maxnum.f32(float %tmp204, float 0.000000e+00) 213 %clamp.i2 = call float @llvm.minnum.f32(float %max.0.i1, float 1.000000e+00) 214 %tmp205 = insertelement <4 x float> undef, float %tmp87, i32 0 215 %tmp206 = insertelement <4 x float> %tmp205, float %tmp91, i32 1 216 %tmp207 = insertelement <4 x float> %tmp206, float %tmp95, i32 2 217 %tmp208 = insertelement <4 x float> %tmp207, float %tmp99, i32 3 218 call void @llvm.r600.store.swizzle(<4 x float> %tmp208, i32 60, i32 1) 219 %tmp209 = insertelement <4 x float> undef, float %clamp.i6, i32 0 220 %tmp210 = insertelement <4 x float> %tmp209, float %clamp.i4, i32 1 221 %tmp211 = insertelement <4 x float> %tmp210, float %clamp.i2, i32 2 222 %tmp212 = insertelement <4 x float> %tmp211, float %clamp.i8, i32 3 223 call void @llvm.r600.store.swizzle(<4 x float> %tmp212, i32 0, i32 2) 224 ret void 225 } 226 227 declare float @llvm.minnum.f32(float, float) #0 228 declare float @llvm.maxnum.f32(float, float) #0 229 declare float @llvm.r600.dot4(<4 x float>, <4 x float>) #0 230 declare float @llvm.fabs.f32(float) #0 231 declare float @llvm.r600.recipsqrt.clamped.f32(float) #0 232 declare float @llvm.pow.f32(float, float) #0 233 declare void @llvm.r600.store.swizzle(<4 x float>, i32, i32) #1 234 235 attributes #0 = { nounwind readnone } 236 attributes #1 = { nounwind } 237
