1 ; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=hawaii -mattr=+half-rate-64-ops < %s | FileCheck -check-prefix=ALL -check-prefix=CIFASTF64 %s 2 ; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=kaveri -mattr=-half-rate-64-ops < %s | FileCheck -check-prefix=ALL -check-prefix=CISLOWF64 %s 3 ; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=tahiti -mattr=+half-rate-64-ops < %s | FileCheck -check-prefix=ALL -check-prefix=SIFASTF64 %s 4 ; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=verde -mattr=-half-rate-64-ops < %s | FileCheck -check-prefix=ALL -check-prefix=SISLOWF64 %s 5 6 ; CHECK: 'fdiv_f32' 7 ; ALL: estimated cost of 10 for {{.*}} fdiv float 8 define void @fdiv_f32(float addrspace(1)* %out, float addrspace(1)* %vaddr, float %b) #0 { 9 %vec = load float, float addrspace(1)* %vaddr 10 %add = fdiv float %vec, %b 11 store float %add, float addrspace(1)* %out 12 ret void 13 } 14 15 ; ALL: 'fdiv_v2f32' 16 ; ALL: estimated cost of 20 for {{.*}} fdiv <2 x float> 17 define void @fdiv_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %vaddr, <2 x float> %b) #0 { 18 %vec = load <2 x float>, <2 x float> addrspace(1)* %vaddr 19 %add = fdiv <2 x float> %vec, %b 20 store <2 x float> %add, <2 x float> addrspace(1)* %out 21 ret void 22 } 23 24 ; ALL: 'fdiv_v3f32' 25 ; ALL: estimated cost of 30 for {{.*}} fdiv <3 x float> 26 define void @fdiv_v3f32(<3 x float> addrspace(1)* %out, <3 x float> addrspace(1)* %vaddr, <3 x float> %b) #0 { 27 %vec = load <3 x float>, <3 x float> addrspace(1)* %vaddr 28 %add = fdiv <3 x float> %vec, %b 29 store <3 x float> %add, <3 x float> addrspace(1)* %out 30 ret void 31 } 32 33 ; ALL: 'fdiv_f64' 34 ; CIFASTF64: estimated cost of 29 for {{.*}} fdiv double 35 ; CISLOWF64: estimated cost of 33 for {{.*}} fdiv double 36 ; SIFASTF64: estimated cost of 32 for {{.*}} fdiv double 37 ; SISLOWF64: estimated cost of 36 for {{.*}} fdiv double 38 define void @fdiv_f64(double addrspace(1)* %out, double addrspace(1)* %vaddr, double %b) #0 { 39 %vec = load double, double addrspace(1)* %vaddr 40 %add = fdiv double %vec, %b 41 store double %add, double addrspace(1)* %out 42 ret void 43 } 44 45 ; ALL: 'fdiv_v2f64' 46 ; CIFASTF64: estimated cost of 58 for {{.*}} fdiv <2 x double> 47 ; CISLOWF64: estimated cost of 66 for {{.*}} fdiv <2 x double> 48 ; SIFASTF64: estimated cost of 64 for {{.*}} fdiv <2 x double> 49 ; SISLOWF64: estimated cost of 72 for {{.*}} fdiv <2 x double> 50 define void @fdiv_v2f64(<2 x double> addrspace(1)* %out, <2 x double> addrspace(1)* %vaddr, <2 x double> %b) #0 { 51 %vec = load <2 x double>, <2 x double> addrspace(1)* %vaddr 52 %add = fdiv <2 x double> %vec, %b 53 store <2 x double> %add, <2 x double> addrspace(1)* %out 54 ret void 55 } 56 57 ; ALL: 'fdiv_v3f64' 58 ; CIFASTF64: estimated cost of 87 for {{.*}} fdiv <3 x double> 59 ; CISLOWF64: estimated cost of 99 for {{.*}} fdiv <3 x double> 60 ; SIFASTF64: estimated cost of 96 for {{.*}} fdiv <3 x double> 61 ; SISLOWF64: estimated cost of 108 for {{.*}} fdiv <3 x double> 62 define void @fdiv_v3f64(<3 x double> addrspace(1)* %out, <3 x double> addrspace(1)* %vaddr, <3 x double> %b) #0 { 63 %vec = load <3 x double>, <3 x double> addrspace(1)* %vaddr 64 %add = fdiv <3 x double> %vec, %b 65 store <3 x double> %add, <3 x double> addrspace(1)* %out 66 ret void 67 } 68 69 ; ALL: 'fdiv_f16' 70 ; ALL: estimated cost of 10 for {{.*}} fdiv half 71 define void @fdiv_f16(half addrspace(1)* %out, half addrspace(1)* %vaddr, half %b) #0 { 72 %vec = load half, half addrspace(1)* %vaddr 73 %add = fdiv half %vec, %b 74 store half %add, half addrspace(1)* %out 75 ret void 76 } 77 78 ; ALL: 'fdiv_v2f16' 79 ; ALL: estimated cost of 20 for {{.*}} fdiv <2 x half> 80 define void @fdiv_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %vaddr, <2 x half> %b) #0 { 81 %vec = load <2 x half>, <2 x half> addrspace(1)* %vaddr 82 %add = fdiv <2 x half> %vec, %b 83 store <2 x half> %add, <2 x half> addrspace(1)* %out 84 ret void 85 } 86 87 ; ALL: 'fdiv_v4f16' 88 ; ALL: estimated cost of 40 for {{.*}} fdiv <4 x half> 89 define void @fdiv_v4f16(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %vaddr, <4 x half> %b) #0 { 90 %vec = load <4 x half>, <4 x half> addrspace(1)* %vaddr 91 %add = fdiv <4 x half> %vec, %b 92 store <4 x half> %add, <4 x half> addrspace(1)* %out 93 ret void 94 } 95 96 attributes #0 = { nounwind } 97
