1 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=-fp64-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-STRICT,GFX9-FLUSH,GFX9 %s
2 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=-fp64-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-STRICT,GFX9-FLUSH,GFX9 %s
3 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=-fp64-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-CONTRACT,GFX9-FLUSH,GFX9 %s
4 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=-fp64-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-CONTRACT,GFX9-FLUSH,GFX9 %s
6 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=+fp64-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-STRICT,GFX9-DENORM-STRICT,GFX9-DENORM,GFX9 %s
7 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=+fp64-fp16-denormals -fp-contract=on -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-STRICT,GFX9-DENORM-STRICT,GFX9-DENORM,GFX9 %s
8 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=+fp64-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-CONTRACT,GFX9-DENORM-CONTRACT,GFX9-DENORM,GFX9 %s
9 ; RUN: llc -amdgpu-scalarize-global-loads=false -march=amdgcn -mcpu=gfx900 -mattr=+fp64-fp16-denormals -fp-contract=fast -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-CONTRACT,GFX9-DENORM-CONTRACT,GFX9-DENORM,GFX9 %s
11 declare i32 @llvm.amdgcn.workitem.id.x() #1
12 declare <2 x half> @llvm.fmuladd.v2f16(<2 x half>, <2 x half>, <2 x half>) #1
13 declare <2 x half> @llvm.fabs.v2f16(<2 x half>) #1
15 ; GCN-LABEL: {{^}}fmuladd_v2f16:
16 ; GFX9-FLUSH: v_pk_mul_f16 {{v[0-9]+, v[0-9]+, v[0-9]+}}
17 ; GFX9-FLUSH: v_pk_add_f16 {{v[0-9]+, v[0-9]+, v[0-9]+}}
19 ; GFX9-DENORM: v_pk_fma_f16 {{v[0-9]+, v[0-9]+, v[0-9]+}}
20 define amdgpu_kernel void @fmuladd_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in1,
21 <2 x half> addrspace(1)* %in2, <2 x half> addrspace(1)* %in3) #0 {
22 %r0 = load <2 x half>, <2 x half> addrspace(1)* %in1
23 %r1 = load <2 x half>, <2 x half> addrspace(1)* %in2
24 %r2 = load <2 x half>, <2 x half> addrspace(1)* %in3
25 %r3 = tail call <2 x half> @llvm.fmuladd.v2f16(<2 x half> %r0, <2 x half> %r1, <2 x half> %r2)
26 store <2 x half> %r3, <2 x half> addrspace(1)* %out
30 ; GCN-LABEL: {{^}}fmuladd_2.0_a_b_v2f16:
31 ; GCN: {{buffer|flat|global}}_load_dword [[R1:v[0-9]+]],
32 ; GCN: {{buffer|flat|global}}_load_dword [[R2:v[0-9]+]],
33 ; GFX9-FLUSH: v_pk_add_f16 [[ADD0:v[0-9]+]], [[R1]], [[R1]]
34 ; GFX9-FLUSH: v_pk_add_f16 [[RESULT:v[0-9]+]], [[ADD0]], [[R2]]
36 ; GFX9-FLUSH: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
38 ; GFX9-DENORM: v_pk_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
39 ; GFX9-DENORM: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
40 define amdgpu_kernel void @fmuladd_2.0_a_b_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
41 %tid = call i32 @llvm.amdgcn.workitem.id.x()
42 %gep.0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
43 %gep.1 = getelementptr <2 x half>, <2 x half> addrspace(1)* %gep.0, i32 1
44 %gep.out = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
46 %r1 = load volatile <2 x half>, <2 x half> addrspace(1)* %gep.0
47 %r2 = load volatile <2 x half>, <2 x half> addrspace(1)* %gep.1
49 %r3 = tail call <2 x half> @llvm.fmuladd.v2f16(<2 x half> <half 2.0, half 2.0>, <2 x half> %r1, <2 x half> %r2)
50 store <2 x half> %r3, <2 x half> addrspace(1)* %gep.out
54 ; GCN-LABEL: {{^}}fmuladd_a_2.0_b_v2f16:
55 ; GCN: {{buffer|flat|global}}_load_dword [[R1:v[0-9]+]],
56 ; GCN: {{buffer|flat|global}}_load_dword [[R2:v[0-9]+]],
57 ; GFX9-FLUSH: v_pk_add_f16 [[ADD0:v[0-9]+]], [[R1]], [[R1]]
58 ; GFX9-FLUSH: v_pk_add_f16 [[RESULT:v[0-9]+]], [[ADD0]], [[R2]]
60 ; GFX9-FLUSH: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
62 ; GFX9-DENORM: v_pk_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
63 ; GFX9-DENORM: global_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
64 define amdgpu_kernel void @fmuladd_a_2.0_b_v2f16(<2 x half> addrspace(1)* %out, <2 x half> addrspace(1)* %in) #0 {
65 %tid = call i32 @llvm.amdgcn.workitem.id.x()
66 %gep.0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
67 %gep.1 = getelementptr <2 x half>, <2 x half> addrspace(1)* %gep.0, i32 1
68 %gep.out = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
70 %r1 = load volatile <2 x half>, <2 x half> addrspace(1)* %gep.0
71 %r2 = load volatile <2 x half>, <2 x half> addrspace(1)* %gep.1
73 %r3 = tail call <2 x half> @llvm.fmuladd.v2f16(<2 x half> %r1, <2 x half> <half 2.0, half 2.0>, <2 x half> %r2)
74 store <2 x half> %r3, <2 x half> addrspace(1)* %gep.out
78 ; GCN-LABEL: {{^}}fadd_a_a_b_v2f16:
79 ; GCN: {{buffer|flat|global}}_load_dword [[R1:v[0-9]+]],
80 ; GCN: {{buffer|flat|global}}_load_dword [[R2:v[0-9]+]],
81 ; GFX9-FLUSH: v_pk_add_f16 [[ADD0:v[0-9]+]], [[R1]], [[R1]]
82 ; GFX9-FLUSH: v_pk_add_f16 [[RESULT:v[0-9]+]], [[ADD0]], [[R2]]
84 ; GFX9-DENORM-STRICT: v_pk_add_f16 [[ADD0:v[0-9]+]], [[R1]], [[R1]]
85 ; GFX9-DENORM-STRICT: v_pk_add_f16 [[RESULT:v[0-9]+]], [[ADD0]], [[R2]]
87 ; GFX9-DENORM-CONTRACT: v_pk_fma_f16 [[RESULT:v[0-9]+]], [[R1]], 2.0, [[R2]]
88 ; GCN: {{flat|global}}_store_dword v{{\[[0-9]+:[0-9]+\]}}, [[RESULT]]
89 define amdgpu_kernel void @fadd_a_a_b_v2f16(<2 x half> addrspace(1)* %out,
90 <2 x half> addrspace(1)* %in1,
91 <2 x half> addrspace(1)* %in2) #0 {
92 %tid = call i32 @llvm.amdgcn.workitem.id.x()
93 %gep.0 = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
94 %gep.1 = getelementptr <2 x half>, <2 x half> addrspace(1)* %gep.0, i32 1
95 %gep.out = getelementptr <2 x half>, <2 x half> addrspace(1)* %out, i32 %tid
97 %r0 = load volatile <2 x half>, <2 x half> addrspace(1)* %gep.0
98 %r1 = load volatile <2 x half>, <2 x half> addrspace(1)* %gep.1
100 %add.0 = fadd <2 x half> %r0, %r0
101 %add.1 = fadd <2 x half> %add.0, %r1
102 store <2 x half> %add.1, <2 x half> addrspace(1)* %gep.out
106 attributes #0 = { nounwind }
107 attributes #1 = { nounwind readnone }