1 ; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx900 -mattr=+half-rate-64-ops < %s | FileCheck -check-prefixes=FASTF64,FASTF16,ALL %s
2 ; RUN: opt -cost-model -analyze -mtriple=amdgcn-unknown-amdhsa -mattr=-half-rate-64-ops < %s | FileCheck -check-prefixes=SLOWF64,SLOWF16,ALL %s
3 ; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mcpu=gfx900 -mattr=+half-rate-64-ops < %s | FileCheck -check-prefixes=FASTF64,FASTF16,ALL %s
4 ; RUN: opt -cost-model -cost-kind=code-size -analyze -mtriple=amdgcn-unknown-amdhsa -mattr=-half-rate-64-ops < %s | FileCheck -check-prefixes=SLOWF64,SLOWF16,ALL %s
7 ; ALL: estimated cost of 1 for {{.*}} fadd float
8 define amdgpu_kernel void @fadd_f32(float addrspace(1)* %out, float addrspace(1)* %vaddr, float %b) #0 {
9 %vec = load float, float addrspace(1)* %vaddr
10 %add = fadd float %vec, %b
11 store float %add, float addrspace(1)* %out
16 ; ALL: estimated cost of 2 for {{.*}} fadd <2 x float>
17 define amdgpu_kernel void @fadd_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 = fadd <2 x float> %vec, %b
20 store <2 x float> %add, <2 x float> addrspace(1)* %out
25 ; Allow for 4 when v3f32 is illegal and TargetLowering thinks it needs widening,
26 ; and 3 when it is legal.
27 ; ALL: estimated cost of {{[34]}} for {{.*}} fadd <3 x float>
28 define amdgpu_kernel void @fadd_v3f32(<3 x float> addrspace(1)* %out, <3 x float> addrspace(1)* %vaddr, <3 x float> %b) #0 {
29 %vec = load <3 x float>, <3 x float> addrspace(1)* %vaddr
30 %add = fadd <3 x float> %vec, %b
31 store <3 x float> %add, <3 x float> addrspace(1)* %out
36 ; Allow for 8 when v5f32 is illegal and TargetLowering thinks it needs widening,
37 ; and 5 when it is legal.
38 ; ALL: estimated cost of {{[58]}} for {{.*}} fadd <5 x float>
39 define amdgpu_kernel void @fadd_v5f32(<5 x float> addrspace(1)* %out, <5 x float> addrspace(1)* %vaddr, <5 x float> %b) #0 {
40 %vec = load <5 x float>, <5 x float> addrspace(1)* %vaddr
41 %add = fadd <5 x float> %vec, %b
42 store <5 x float> %add, <5 x float> addrspace(1)* %out
47 ; FASTF64: estimated cost of 2 for {{.*}} fadd double
48 ; SLOWF64: estimated cost of 3 for {{.*}} fadd double
49 define amdgpu_kernel void @fadd_f64(double addrspace(1)* %out, double addrspace(1)* %vaddr, double %b) #0 {
50 %vec = load double, double addrspace(1)* %vaddr
51 %add = fadd double %vec, %b
52 store double %add, double addrspace(1)* %out
57 ; FASTF64: estimated cost of 4 for {{.*}} fadd <2 x double>
58 ; SLOWF64: estimated cost of 6 for {{.*}} fadd <2 x double>
59 define amdgpu_kernel void @fadd_v2f64(<2 x double> addrspace(1)* %out, <2 x double> addrspace(1)* %vaddr, <2 x double> %b) #0 {
60 %vec = load <2 x double>, <2 x double> addrspace(1)* %vaddr
61 %add = fadd <2 x double> %vec, %b
62 store <2 x double> %add, <2 x double> addrspace(1)* %out
67 ; FASTF64: estimated cost of 6 for {{.*}} fadd <3 x double>
68 ; SLOWF64: estimated cost of 9 for {{.*}} fadd <3 x double>
69 define amdgpu_kernel void @fadd_v3f64(<3 x double> addrspace(1)* %out, <3 x double> addrspace(1)* %vaddr, <3 x double> %b) #0 {
70 %vec = load <3 x double>, <3 x double> addrspace(1)* %vaddr
71 %add = fadd <3 x double> %vec, %b
72 store <3 x double> %add, <3 x double> addrspace(1)* %out
77 ; ALL: estimated cost of 1 for {{.*}} fadd half
78 define amdgpu_kernel void @fadd_f16(half addrspace(1)* %out, half addrspace(1)* %vaddr, half %b) #0 {
79 %vec = load half, half addrspace(1)* %vaddr
80 %add = fadd half %vec, %b
81 store half %add, half addrspace(1)* %out
86 ; SLOWF16: estimated cost of 2 for {{.*}} fadd <2 x half>
87 ; FASTF16: estimated cost of 1 for {{.*}} fadd <2 x half>
88 define amdgpu_kernel void @fadd_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %vaddr, <2 x half> %b) #0 {
89 %vec = load <2 x half>, <2 x half> addrspace(1)* %vaddr
90 %add = fadd <2 x half> %vec, %b
91 store <2 x half> %add, <2 x half> addrspace(1)* %out
96 ; SLOWF16: estimated cost of 4 for {{.*}} fadd <3 x half>
97 ; FASTF16: estimated cost of 2 for {{.*}} fadd <3 x half>
98 define amdgpu_kernel void @fadd_v3f16(<3 x half> addrspace(1)* %out, <3 x half> addrspace(1)* %vaddr, <3 x half> %b) #0 {
99 %vec = load <3 x half>, <3 x half> addrspace(1)* %vaddr
100 %add = fadd <3 x half> %vec, %b
101 store <3 x half> %add, <3 x half> addrspace(1)* %out
106 ; SLOWF16: estimated cost of 4 for {{.*}} fadd <4 x half>
107 ; FASTF16: estimated cost of 2 for {{.*}} fadd <4 x half>
108 define amdgpu_kernel void @fadd_v4f16(<4 x half> addrspace(1)* %out, <4 x half> addrspace(1)* %vaddr, <4 x half> %b) #0 {
109 %vec = load <4 x half>, <4 x half> addrspace(1)* %vaddr
110 %add = fadd <4 x half> %vec, %b
111 store <4 x half> %add, <4 x half> addrspace(1)* %out
115 attributes #0 = { nounwind }