1; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mattr=+half-rate-64-ops < %s | FileCheck -check-prefix=FASTF64 -check-prefix=ALL %s 2; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mattr=-half-rate-64-ops < %s | FileCheck -check-prefix=SLOWF64 -check-prefix=ALL %s 3 4; ALL: 'fadd_f32' 5; ALL: estimated cost of 1 for {{.*}} fadd float 6define amdgpu_kernel void @fadd_f32(float addrspace(1)* %out, float addrspace(1)* %vaddr, float %b) #0 { 7 %vec = load float, float addrspace(1)* %vaddr 8 %add = fadd float %vec, %b 9 store float %add, float addrspace(1)* %out 10 ret void 11} 12 13; ALL: 'fadd_v2f32' 14; ALL: estimated cost of 2 for {{.*}} fadd <2 x float> 15define amdgpu_kernel void @fadd_v2f32(<2 x float> addrspace(1)* %out, <2 x float> addrspace(1)* %vaddr, <2 x float> %b) #0 { 16 %vec = load <2 x float>, <2 x float> addrspace(1)* %vaddr 17 %add = fadd <2 x float> %vec, %b 18 store <2 x float> %add, <2 x float> addrspace(1)* %out 19 ret void 20} 21 22; ALL: 'fadd_v3f32' 23; ALL: estimated cost of 3 for {{.*}} fadd <3 x float> 24define amdgpu_kernel void @fadd_v3f32(<3 x float> addrspace(1)* %out, <3 x float> addrspace(1)* %vaddr, <3 x float> %b) #0 { 25 %vec = load <3 x float>, <3 x float> addrspace(1)* %vaddr 26 %add = fadd <3 x float> %vec, %b 27 store <3 x float> %add, <3 x float> addrspace(1)* %out 28 ret void 29} 30 31; ALL: 'fadd_f64' 32; FASTF64: estimated cost of 2 for {{.*}} fadd double 33; SLOWF64: estimated cost of 3 for {{.*}} fadd double 34define amdgpu_kernel void @fadd_f64(double addrspace(1)* %out, double addrspace(1)* %vaddr, double %b) #0 { 35 %vec = load double, double addrspace(1)* %vaddr 36 %add = fadd double %vec, %b 37 store double %add, double addrspace(1)* %out 38 ret void 39} 40 41; ALL: 'fadd_v2f64' 42; FASTF64: estimated cost of 4 for {{.*}} fadd <2 x double> 43; SLOWF64: estimated cost of 6 for {{.*}} fadd <2 x double> 44define amdgpu_kernel void @fadd_v2f64(<2 x double> addrspace(1)* %out, <2 x double> addrspace(1)* %vaddr, <2 x double> %b) #0 { 45 %vec = load <2 x double>, <2 x double> addrspace(1)* %vaddr 46 %add = fadd <2 x double> %vec, %b 47 store <2 x double> %add, <2 x double> addrspace(1)* %out 48 ret void 49} 50 51; ALL: 'fadd_v3f64' 52; FASTF64: estimated cost of 6 for {{.*}} fadd <3 x double> 53; SLOWF64: estimated cost of 9 for {{.*}} fadd <3 x double> 54define amdgpu_kernel void @fadd_v3f64(<3 x double> addrspace(1)* %out, <3 x double> addrspace(1)* %vaddr, <3 x double> %b) #0 { 55 %vec = load <3 x double>, <3 x double> addrspace(1)* %vaddr 56 %add = fadd <3 x double> %vec, %b 57 store <3 x double> %add, <3 x double> addrspace(1)* %out 58 ret void 59} 60 61; ALL 'fadd_f16' 62; ALL estimated cost of 1 for {{.*}} fadd half 63define amdgpu_kernel void @fadd_f16(half addrspace(1)* %out, half addrspace(1)* %vaddr, half %b) #0 { 64 %vec = load half, half addrspace(1)* %vaddr 65 %add = fadd half %vec, %b 66 store half %add, half addrspace(1)* %out 67 ret void 68} 69 70; ALL 'fadd_v2f16' 71; ALL estimated cost of 2 for {{.*}} fadd <2 x half> 72define amdgpu_kernel void @fadd_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %vaddr, <2 x half> %b) #0 { 73 %vec = load <2 x half>, <2 x half> addrspace(1)* %vaddr 74 %add = fadd <2 x half> %vec, %b 75 store <2 x half> %add, <2 x half> addrspace(1)* %out 76 ret void 77} 78 79; ALL 'fadd_v4f16' 80; ALL estimated cost of 4 for {{.*}} fadd <4 x half> 81define amdgpu_kernel void @fadd_v4f16(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %vaddr, <4 x half> %b) #0 { 82 %vec = load <4 x half>, <4 x half> addrspace(1)* %vaddr 83 %add = fadd <4 x half> %vec, %b 84 store <4 x half> %add, <4 x half> addrspace(1)* %out 85 ret void 86} 87 88attributes #0 = { nounwind } 89