1 ; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=SI -check-prefix=FUNC %s 2 ; RUN: llc -march=amdgcn -mcpu=bonaire -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=FUNC %s 3 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCN -check-prefix=CI -check-prefix=FUNC %s 4 5 ; RUN: llc -march=amdgcn -enable-unsafe-fp-math -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-UNSAFE -check-prefix=SI-UNSAFE -check-prefix=FUNC %s 6 ; RUN: llc -march=amdgcn -mcpu=tonga -enable-unsafe-fp-math -verify-machineinstrs < %s | FileCheck -check-prefix=GCN-UNSAFE -check-prefix=VI-UNSAFE -check-prefix=FUNC %s 7 8 declare double @llvm.fabs.f64(double) #0 9 declare double @llvm.floor.f64(double) #0 10 11 ; FUNC-LABEL: {{^}}fract_f64: 12 ; SI-DAG: v_fract_f64_e32 [[FRC:v\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]] 13 ; SI-DAG: v_mov_b32_e32 v[[UPLO:[0-9]+]], -1 14 ; SI-DAG: v_mov_b32_e32 v[[UPHI:[0-9]+]], 0x3fefffff 15 ; SI-DAG: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], v{{\[}}[[UPLO]]:[[UPHI]]], [[FRC]] 16 ; SI-DAG: v_cmp_class_f64_e64 [[COND:s\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO]]:[[HI]]], 3 17 ; SI: v_cndmask_b32_e64 v[[RESLO:[0-9]+]], v[[MINLO]], v[[LO]], [[COND]] 18 ; SI: v_cndmask_b32_e64 v[[RESHI:[0-9]+]], v[[MINHI]], v[[HI]], [[COND]] 19 ; SI: v_add_f64 [[SUB0:v\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO]]:[[HI]]{{\]}}, -v{{\[}}[[RESLO]]:[[RESHI]]{{\]}} 20 ; SI: v_add_f64 [[FRACT:v\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO]]:[[HI]]{{\]}}, -[[SUB0]] 21 22 ; CI: buffer_load_dwordx2 [[X:v\[[0-9]+:[0-9]+\]]] 23 ; CI: v_floor_f64_e32 [[FLOORX:v\[[0-9]+:[0-9]+\]]], [[X]] 24 ; CI: v_add_f64 [[FRACT:v\[[0-9]+:[0-9]+\]]], [[X]], -[[FLOORX]] 25 26 ; GCN-UNSAFE: buffer_load_dwordx2 [[X:v\[[0-9]+:[0-9]+\]]] 27 ; GCN-UNSAFE: v_fract_f64_e32 [[FRACT:v\[[0-9]+:[0-9]+\]]], [[X]] 28 29 ; GCN: buffer_store_dwordx2 [[FRACT]] 30 define void @fract_f64(double addrspace(1)* %out, double addrspace(1)* %src) #1 { 31 %x = load double, double addrspace(1)* %src 32 %floor.x = call double @llvm.floor.f64(double %x) 33 %fract = fsub double %x, %floor.x 34 store double %fract, double addrspace(1)* %out 35 ret void 36 } 37 38 ; FUNC-LABEL: {{^}}fract_f64_neg: 39 ; SI-DAG: v_fract_f64_e64 [[FRC:v\[[0-9]+:[0-9]+\]]], -v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]] 40 ; SI-DAG: v_mov_b32_e32 v[[UPLO:[0-9]+]], -1 41 ; SI-DAG: v_mov_b32_e32 v[[UPHI:[0-9]+]], 0x3fefffff 42 ; SI-DAG: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], v{{\[}}[[UPLO]]:[[UPHI]]], [[FRC]] 43 ; SI-DAG: v_cmp_class_f64_e64 [[COND:s\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO]]:[[HI]]], 3 44 ; SI: v_cndmask_b32_e64 v[[RESLO:[0-9]+]], v[[MINLO]], v[[LO]], [[COND]] 45 ; SI: v_cndmask_b32_e64 v[[RESHI:[0-9]+]], v[[MINHI]], v[[HI]], [[COND]] 46 ; SI: v_add_f64 [[SUB0:v\[[0-9]+:[0-9]+\]]], -v{{\[}}[[LO]]:[[HI]]{{\]}}, -v{{\[}}[[RESLO]]:[[RESHI]]{{\]}} 47 ; SI: v_add_f64 [[FRACT:v\[[0-9]+:[0-9]+\]]], -v{{\[}}[[LO]]:[[HI]]{{\]}}, -[[SUB0]] 48 49 ; CI: buffer_load_dwordx2 [[X:v\[[0-9]+:[0-9]+\]]] 50 ; CI: v_floor_f64_e64 [[FLOORX:v\[[0-9]+:[0-9]+\]]], -[[X]] 51 ; CI: v_add_f64 [[FRACT:v\[[0-9]+:[0-9]+\]]], -[[X]], -[[FLOORX]] 52 53 ; GCN-UNSAFE: buffer_load_dwordx2 [[X:v\[[0-9]+:[0-9]+\]]] 54 ; GCN-UNSAFE: v_fract_f64_e64 [[FRACT:v\[[0-9]+:[0-9]+\]]], -[[X]] 55 56 ; GCN: buffer_store_dwordx2 [[FRACT]] 57 define void @fract_f64_neg(double addrspace(1)* %out, double addrspace(1)* %src) #1 { 58 %x = load double, double addrspace(1)* %src 59 %neg.x = fsub double -0.0, %x 60 %floor.neg.x = call double @llvm.floor.f64(double %neg.x) 61 %fract = fsub double %neg.x, %floor.neg.x 62 store double %fract, double addrspace(1)* %out 63 ret void 64 } 65 66 ; FUNC-LABEL: {{^}}fract_f64_neg_abs: 67 ; SI-DAG: v_fract_f64_e64 [[FRC:v\[[0-9]+:[0-9]+\]]], -|v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]]| 68 ; SI-DAG: v_mov_b32_e32 v[[UPLO:[0-9]+]], -1 69 ; SI-DAG: v_mov_b32_e32 v[[UPHI:[0-9]+]], 0x3fefffff 70 ; SI-DAG: v_min_f64 v{{\[}}[[MINLO:[0-9]+]]:[[MINHI:[0-9]+]]], v{{\[}}[[UPLO]]:[[UPHI]]], [[FRC]] 71 ; SI-DAG: v_cmp_class_f64_e64 [[COND:s\[[0-9]+:[0-9]+\]]], v{{\[}}[[LO]]:[[HI]]], 3 72 ; SI: v_cndmask_b32_e64 v[[RESLO:[0-9]+]], v[[MINLO]], v[[LO]], [[COND]] 73 ; SI: v_cndmask_b32_e64 v[[RESHI:[0-9]+]], v[[MINHI]], v[[HI]], [[COND]] 74 ; SI: v_add_f64 [[SUB0:v\[[0-9]+:[0-9]+\]]], -|v{{\[}}[[LO]]:[[HI]]{{\]}}|, -v{{\[}}[[RESLO]]:[[RESHI]]{{\]}} 75 ; SI: v_add_f64 [[FRACT:v\[[0-9]+:[0-9]+\]]], -|v{{\[}}[[LO]]:[[HI]]{{\]}}|, -[[SUB0]] 76 77 ; CI: buffer_load_dwordx2 [[X:v\[[0-9]+:[0-9]+\]]] 78 ; CI: v_floor_f64_e64 [[FLOORX:v\[[0-9]+:[0-9]+\]]], -|[[X]]| 79 ; CI: v_add_f64 [[FRACT:v\[[0-9]+:[0-9]+\]]], -|[[X]]|, -[[FLOORX]] 80 81 ; GCN-UNSAFE: buffer_load_dwordx2 [[X:v\[[0-9]+:[0-9]+\]]] 82 ; GCN-UNSAFE: v_fract_f64_e64 [[FRACT:v\[[0-9]+:[0-9]+\]]], -|[[X]]| 83 84 ; GCN: buffer_store_dwordx2 [[FRACT]] 85 define void @fract_f64_neg_abs(double addrspace(1)* %out, double addrspace(1)* %src) #1 { 86 %x = load double, double addrspace(1)* %src 87 %abs.x = call double @llvm.fabs.f64(double %x) 88 %neg.abs.x = fsub double -0.0, %abs.x 89 %floor.neg.abs.x = call double @llvm.floor.f64(double %neg.abs.x) 90 %fract = fsub double %neg.abs.x, %floor.neg.abs.x 91 store double %fract, double addrspace(1)* %out 92 ret void 93 } 94 95 ; FUNC-LABEL: {{^}}multi_use_floor_fract_f64: 96 ; VI-UNSAFE: buffer_load_dwordx2 [[X:v\[[0-9]+:[0-9]+\]]] 97 ; VI-UNSAFE-DAG: v_floor_f64_e32 [[FLOOR:v\[[0-9]+:[0-9]+\]]], [[X]] 98 ; VI-UNSAFE-DAG: v_fract_f64_e32 [[FRACT:v\[[0-9]+:[0-9]+\]]], [[X]] 99 ; VI-UNSAFE: buffer_store_dwordx2 [[FLOOR]] 100 ; VI-UNSAFE: buffer_store_dwordx2 [[FRACT]] 101 define void @multi_use_floor_fract_f64(double addrspace(1)* %out, double addrspace(1)* %src) #1 { 102 %x = load double, double addrspace(1)* %src 103 %floor.x = call double @llvm.floor.f64(double %x) 104 %fract = fsub double %x, %floor.x 105 store volatile double %floor.x, double addrspace(1)* %out 106 store volatile double %fract, double addrspace(1)* %out 107 ret void 108 } 109 110 attributes #0 = { nounwind readnone } 111 attributes #1 = { nounwind } 112
