1 ; RUN: llc < %s -march=r600 --mcpu=redwood | FileCheck %s --check-prefix=R600 2 ; RUN: llc < %s -march=amdgcn --mcpu=SI -verify-machineinstrs| FileCheck %s --check-prefix=SI 3 ; RUN: llc < %s -march=amdgcn --mcpu=tonga -verify-machineinstrs| FileCheck %s --check-prefix=SI 4 5 ; R600-LABEL: {{^}}sqrt_f32: 6 ; R600: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[2].Z 7 ; R600: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[2].Z, PS 8 ; SI-LABEL: {{^}}sqrt_f32: 9 ; SI: v_sqrt_f32_e32 10 define void @sqrt_f32(float addrspace(1)* %out, float %in) { 11 entry: 12 %0 = call float @llvm.sqrt.f32(float %in) 13 store float %0, float addrspace(1)* %out 14 ret void 15 } 16 17 ; R600-LABEL: {{^}}sqrt_v2f32: 18 ; R600-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[2].W 19 ; R600-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[2].W, PS 20 ; R600-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[3].X 21 ; R600-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[3].X, PS 22 ; SI-LABEL: {{^}}sqrt_v2f32: 23 ; SI: v_sqrt_f32_e32 24 ; SI: v_sqrt_f32_e32 25 define void @sqrt_v2f32(<2 x float> addrspace(1)* %out, <2 x float> %in) { 26 entry: 27 %0 = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %in) 28 store <2 x float> %0, <2 x float> addrspace(1)* %out 29 ret void 30 } 31 32 ; R600-LABEL: {{^}}sqrt_v4f32: 33 ; R600-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[3].Y 34 ; R600-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[3].Y, PS 35 ; R600-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[3].Z 36 ; R600-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[3].Z, PS 37 ; R600-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[3].W 38 ; R600-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[3].W, PS 39 ; R600-DAG: RECIPSQRT_CLAMPED * T{{[0-9]\.[XYZW]}}, KC0[4].X 40 ; R600-DAG: MUL NON-IEEE T{{[0-9]\.[XYZW]}}, KC0[4].X, PS 41 ; SI-LABEL: {{^}}sqrt_v4f32: 42 ; SI: v_sqrt_f32_e32 43 ; SI: v_sqrt_f32_e32 44 ; SI: v_sqrt_f32_e32 45 ; SI: v_sqrt_f32_e32 46 define void @sqrt_v4f32(<4 x float> addrspace(1)* %out, <4 x float> %in) { 47 entry: 48 %0 = call <4 x float> @llvm.sqrt.v4f32(<4 x float> %in) 49 store <4 x float> %0, <4 x float> addrspace(1)* %out 50 ret void 51 } 52 53 ; SI-LABEL: {{^}}elim_redun_check: 54 ; SI: v_sqrt_f32_e32 55 ; SI-NOT: v_cndmask 56 define void @elim_redun_check(float addrspace(1)* %out, float %in) { 57 entry: 58 %sqrt = call float @llvm.sqrt.f32(float %in) 59 %cmp = fcmp olt float %in, -0.000000e+00 60 %res = select i1 %cmp, float 0x7FF8000000000000, float %sqrt 61 store float %res, float addrspace(1)* %out 62 ret void 63 } 64 65 ; SI-LABEL: {{^}}elim_redun_check_ult: 66 ; SI: v_sqrt_f32_e32 67 ; SI-NOT: v_cndmask 68 define void @elim_redun_check_ult(float addrspace(1)* %out, float %in) { 69 entry: 70 %sqrt = call float @llvm.sqrt.f32(float %in) 71 %cmp = fcmp ult float %in, -0.000000e+00 72 %res = select i1 %cmp, float 0x7FF8000000000000, float %sqrt 73 store float %res, float addrspace(1)* %out 74 ret void 75 } 76 77 ; SI-LABEL: {{^}}elim_redun_check_v2: 78 ; SI: v_sqrt_f32_e32 79 ; SI: v_sqrt_f32_e32 80 ; SI-NOT: v_cndmask 81 define void @elim_redun_check_v2(<2 x float> addrspace(1)* %out, <2 x float> %in) { 82 entry: 83 %sqrt = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %in) 84 %cmp = fcmp olt <2 x float> %in, <float -0.000000e+00, float -0.000000e+00> 85 %res = select <2 x i1> %cmp, <2 x float> <float 0x7FF8000000000000, float 0x7FF8000000000000>, <2 x float> %sqrt 86 store <2 x float> %res, <2 x float> addrspace(1)* %out 87 ret void 88 } 89 90 ; SI-LABEL: {{^}}elim_redun_check_v2_ult 91 ; SI: v_sqrt_f32_e32 92 ; SI: v_sqrt_f32_e32 93 ; SI-NOT: v_cndmask 94 define void @elim_redun_check_v2_ult(<2 x float> addrspace(1)* %out, <2 x float> %in) { 95 entry: 96 %sqrt = call <2 x float> @llvm.sqrt.v2f32(<2 x float> %in) 97 %cmp = fcmp ult <2 x float> %in, <float -0.000000e+00, float -0.000000e+00> 98 %res = select <2 x i1> %cmp, <2 x float> <float 0x7FF8000000000000, float 0x7FF8000000000000>, <2 x float> %sqrt 99 store <2 x float> %res, <2 x float> addrspace(1)* %out 100 ret void 101 } 102 103 declare float @llvm.sqrt.f32(float %in) 104 declare <2 x float> @llvm.sqrt.v2f32(<2 x float> %in) 105 declare <4 x float> @llvm.sqrt.v4f32(<4 x float> %in) 106
