1; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck %s 2 3; CHECK: {{^}}main1: 4; CHECK: MOV * T{{[0-9]+\.[XYZW], KC0}} 5define amdgpu_kernel void @main1() #0 { 6main_body: 7 %tmp = load <4 x float>, <4 x float> addrspace(8)* null 8 %tmp7 = extractelement <4 x float> %tmp, i32 0 9 %tmp8 = 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) 10 %tmp9 = extractelement <4 x float> %tmp8, i32 0 11 %tmp10 = 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) 12 %tmp11 = extractelement <4 x float> %tmp10, i32 0 13 %tmp12 = fcmp ogt float %tmp7, 0.000000e+00 14 %tmp13 = select i1 %tmp12, float %tmp9, float %tmp11 15 %tmp14 = load <4 x float>, <4 x float> addrspace(8)* null 16 %tmp15 = extractelement <4 x float> %tmp14, i32 1 17 %tmp16 = 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) 18 %tmp17 = extractelement <4 x float> %tmp16, i32 1 19 %tmp18 = 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) 20 %tmp19 = extractelement <4 x float> %tmp18, i32 1 21 %tmp20 = fcmp ogt float %tmp15, 0.000000e+00 22 %tmp21 = select i1 %tmp20, float %tmp17, float %tmp19 23 %tmp22 = load <4 x float>, <4 x float> addrspace(8)* null 24 %tmp23 = extractelement <4 x float> %tmp22, i32 2 25 %tmp24 = 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) 26 %tmp25 = extractelement <4 x float> %tmp24, i32 2 27 %tmp26 = 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) 28 %tmp27 = extractelement <4 x float> %tmp26, i32 2 29 %tmp28 = fcmp ogt float %tmp23, 0.000000e+00 30 %tmp29 = select i1 %tmp28, float %tmp25, float %tmp27 31 %tmp30 = load <4 x float>, <4 x float> addrspace(8)* null 32 %tmp31 = extractelement <4 x float> %tmp30, i32 3 33 %tmp32 = 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) 34 %tmp33 = extractelement <4 x float> %tmp32, i32 3 35 %tmp34 = 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) 36 %tmp35 = extractelement <4 x float> %tmp34, i32 3 37 %tmp36 = fcmp ogt float %tmp31, 0.000000e+00 38 %tmp37 = select i1 %tmp36, float %tmp33, float %tmp35 39 %max.0.i = call float @llvm.maxnum.f32(float %tmp13, float 0.000000e+00) 40 %clamp.i = call float @llvm.minnum.f32(float %max.0.i, float 1.000000e+00) 41 %max.0.i5 = call float @llvm.maxnum.f32(float %tmp21, float 0.000000e+00) 42 %clamp.i6 = call float @llvm.minnum.f32(float %max.0.i5, float 1.000000e+00) 43 %max.0.i3 = call float @llvm.maxnum.f32(float %tmp29, float 0.000000e+00) 44 %clamp.i4 = call float @llvm.minnum.f32(float %max.0.i3, float 1.000000e+00) 45 %max.0.i1 = call float @llvm.maxnum.f32(float %tmp37, float 0.000000e+00) 46 %clamp.i2 = call float @llvm.minnum.f32(float %max.0.i1, float 1.000000e+00) 47 %tmp38 = insertelement <4 x float> undef, float %clamp.i, i32 0 48 %tmp39 = insertelement <4 x float> %tmp38, float %clamp.i6, i32 1 49 %tmp40 = insertelement <4 x float> %tmp39, float %clamp.i4, i32 2 50 %tmp41 = insertelement <4 x float> %tmp40, float %clamp.i2, i32 3 51 call void @llvm.r600.store.swizzle(<4 x float> %tmp41, i32 0, i32 0) 52 ret void 53} 54 55; CHECK: {{^}}main2: 56; CHECK-NOT: MOV 57define amdgpu_kernel void @main2() #0 { 58main_body: 59 %tmp = load <4 x float>, <4 x float> addrspace(8)* null 60 %tmp7 = extractelement <4 x float> %tmp, i32 0 61 %tmp8 = 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) 62 %tmp9 = extractelement <4 x float> %tmp8, i32 0 63 %tmp10 = 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) 64 %tmp11 = extractelement <4 x float> %tmp10, i32 1 65 %tmp12 = fcmp ogt float %tmp7, 0.000000e+00 66 %tmp13 = select i1 %tmp12, float %tmp9, float %tmp11 67 %tmp14 = load <4 x float>, <4 x float> addrspace(8)* null 68 %tmp15 = extractelement <4 x float> %tmp14, i32 1 69 %tmp16 = 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) 70 %tmp17 = extractelement <4 x float> %tmp16, i32 0 71 %tmp18 = 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 %tmp19 = extractelement <4 x float> %tmp18, i32 1 73 %tmp20 = fcmp ogt float %tmp15, 0.000000e+00 74 %tmp21 = select i1 %tmp20, float %tmp17, float %tmp19 75 %tmp22 = load <4 x float>, <4 x float> addrspace(8)* null 76 %tmp23 = extractelement <4 x float> %tmp22, i32 2 77 %tmp24 = 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) 78 %tmp25 = extractelement <4 x float> %tmp24, i32 3 79 %tmp26 = 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) 80 %tmp27 = extractelement <4 x float> %tmp26, i32 2 81 %tmp28 = fcmp ogt float %tmp23, 0.000000e+00 82 %tmp29 = select i1 %tmp28, float %tmp25, float %tmp27 83 %tmp30 = load <4 x float>, <4 x float> addrspace(8)* null 84 %tmp31 = extractelement <4 x float> %tmp30, i32 3 85 %tmp32 = 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) 86 %tmp33 = extractelement <4 x float> %tmp32, i32 3 87 %tmp34 = 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) 88 %tmp35 = extractelement <4 x float> %tmp34, i32 2 89 %tmp36 = fcmp ogt float %tmp31, 0.000000e+00 90 %tmp37 = select i1 %tmp36, float %tmp33, float %tmp35 91 %max.0.i = call float @llvm.maxnum.f32(float %tmp13, float 0.000000e+00) 92 %clamp.i = call float @llvm.minnum.f32(float %max.0.i, float 1.000000e+00) 93 %max.0.i5 = call float @llvm.maxnum.f32(float %tmp21, float 0.000000e+00) 94 %clamp.i6 = call float @llvm.minnum.f32(float %max.0.i5, float 1.000000e+00) 95 %max.0.i3 = call float @llvm.maxnum.f32(float %tmp29, float 0.000000e+00) 96 %clamp.i4 = call float @llvm.minnum.f32(float %max.0.i3, float 1.000000e+00) 97 %max.0.i1 = call float @llvm.maxnum.f32(float %tmp37, float 0.000000e+00) 98 %clamp.i2 = call float @llvm.minnum.f32(float %max.0.i1, float 1.000000e+00) 99 %tmp38 = insertelement <4 x float> undef, float %clamp.i, i32 0 100 %tmp39 = insertelement <4 x float> %tmp38, float %clamp.i6, i32 1 101 %tmp40 = insertelement <4 x float> %tmp39, float %clamp.i4, i32 2 102 %tmp41 = insertelement <4 x float> %tmp40, float %clamp.i2, i32 3 103 call void @llvm.r600.store.swizzle(<4 x float> %tmp41, i32 0, i32 0) 104 ret void 105} 106 107declare void @llvm.r600.store.swizzle(<4 x float>, i32, i32) #0 108declare float @llvm.minnum.f32(float, float) #1 109declare float @llvm.maxnum.f32(float, float) #1 110 111attributes #0 = { nounwind } 112attributes #1 = { nounwind readnone } 113