150
|
1 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck -check-prefixes=CHECK,CI,CIGFX9 %s
|
|
2 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx802 -verify-machineinstrs < %s | FileCheck -check-prefixes=CHECK,GFX9,CIGFX9 %s
|
|
3 ; RUN: llc -mtriple=amdgcn--amdhsa -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=CHECK,GFX10 %s
|
|
4
|
|
5 declare i32 @llvm.amdgcn.writelane(i32, i32, i32) #0
|
|
6
|
|
7 ; CHECK-LABEL: {{^}}test_writelane_sreg:
|
|
8 ; CIGFX9: v_writelane_b32 v{{[0-9]+}}, s{{[0-9]+}}, m0
|
|
9 ; GFX10: v_writelane_b32 v{{[0-9]+}}, s{{[0-9]+}}, s{{[0-9]+}}
|
|
10 define amdgpu_kernel void @test_writelane_sreg(i32 addrspace(1)* %out, i32 %src0, i32 %src1) #1 {
|
|
11 %oldval = load i32, i32 addrspace(1)* %out
|
|
12 %writelane = call i32 @llvm.amdgcn.writelane(i32 %src0, i32 %src1, i32 %oldval)
|
|
13 store i32 %writelane, i32 addrspace(1)* %out, align 4
|
|
14 ret void
|
|
15 }
|
|
16
|
|
17 ; CHECK-LABEL: {{^}}test_writelane_imm_sreg:
|
|
18 ; CHECK: v_writelane_b32 v{{[0-9]+}}, 32, s{{[0-9]+}}
|
|
19 define amdgpu_kernel void @test_writelane_imm_sreg(i32 addrspace(1)* %out, i32 %src1) #1 {
|
|
20 %oldval = load i32, i32 addrspace(1)* %out
|
|
21 %writelane = call i32 @llvm.amdgcn.writelane(i32 32, i32 %src1, i32 %oldval)
|
|
22 store i32 %writelane, i32 addrspace(1)* %out, align 4
|
|
23 ret void
|
|
24 }
|
|
25
|
|
26 ; CHECK-LABEL: {{^}}test_writelane_vreg_lane:
|
|
27 ; CHECK: v_readfirstlane_b32 [[LANE:s[0-9]+]], v{{[0-9]+}}
|
|
28 ; CHECK: v_writelane_b32 v{{[0-9]+}}, 12, [[LANE]]
|
|
29 define amdgpu_kernel void @test_writelane_vreg_lane(i32 addrspace(1)* %out, <2 x i32> addrspace(1)* %in) #1 {
|
|
30 %tid = call i32 @llvm.amdgcn.workitem.id.x()
|
|
31 %gep.in = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 %tid
|
|
32 %args = load <2 x i32>, <2 x i32> addrspace(1)* %gep.in
|
|
33 %oldval = load i32, i32 addrspace(1)* %out
|
|
34 %lane = extractelement <2 x i32> %args, i32 1
|
|
35 %writelane = call i32 @llvm.amdgcn.writelane(i32 12, i32 %lane, i32 %oldval)
|
|
36 store i32 %writelane, i32 addrspace(1)* %out, align 4
|
|
37 ret void
|
|
38 }
|
|
39
|
|
40 ; CHECK-LABEL: {{^}}test_writelane_m0_sreg:
|
|
41 ; CHECK: s_mov_b32 m0, -1
|
|
42 ; CIGFX9: s_mov_b32 [[COPY_M0:s[0-9]+]], m0
|
|
43 ; CIGFX9: v_writelane_b32 v{{[0-9]+}}, [[COPY_M0]], m0
|
|
44 ; GFX10: v_writelane_b32 v{{[0-9]+}}, m0, s{{[0-9]+}}
|
|
45 define amdgpu_kernel void @test_writelane_m0_sreg(i32 addrspace(1)* %out, i32 %src1) #1 {
|
|
46 %oldval = load i32, i32 addrspace(1)* %out
|
|
47 %m0 = call i32 asm "s_mov_b32 m0, -1", "={m0}"()
|
|
48 %writelane = call i32 @llvm.amdgcn.writelane(i32 %m0, i32 %src1, i32 %oldval)
|
|
49 store i32 %writelane, i32 addrspace(1)* %out, align 4
|
|
50 ret void
|
|
51 }
|
|
52
|
|
53 ; CHECK-LABEL: {{^}}test_writelane_imm:
|
|
54 ; CHECK: v_writelane_b32 v{{[0-9]+}}, s{{[0-9]+}}, 32
|
|
55 define amdgpu_kernel void @test_writelane_imm(i32 addrspace(1)* %out, i32 %src0) #1 {
|
|
56 %oldval = load i32, i32 addrspace(1)* %out
|
|
57 %writelane = call i32 @llvm.amdgcn.writelane(i32 %src0, i32 32, i32 %oldval) #0
|
|
58 store i32 %writelane, i32 addrspace(1)* %out, align 4
|
|
59 ret void
|
|
60 }
|
|
61
|
|
62 ; CHECK-LABEL: {{^}}test_writelane_sreg_oldval:
|
|
63 ; CHECK: v_mov_b32_e32 [[OLDVAL:v[0-9]+]], s{{[0-9]+}}
|
|
64 ; CIGFX9: v_writelane_b32 [[OLDVAL]], s{{[0-9]+}}, m0
|
|
65 ; GFX10: v_writelane_b32 [[OLDVAL]], s{{[0-9]+}}, s{{[0-9]+}}
|
|
66 define amdgpu_kernel void @test_writelane_sreg_oldval(i32 inreg %oldval, i32 addrspace(1)* %out, i32 %src0, i32 %src1) #1 {
|
|
67 %writelane = call i32 @llvm.amdgcn.writelane(i32 %src0, i32 %src1, i32 %oldval)
|
|
68 store i32 %writelane, i32 addrspace(1)* %out, align 4
|
|
69 ret void
|
|
70 }
|
|
71
|
|
72 ; CHECK-LABEL: {{^}}test_writelane_imm_oldval:
|
|
73 ; CHECK: v_mov_b32_e32 [[OLDVAL:v[0-9]+]], 42
|
|
74 ; CIGFX9: v_writelane_b32 [[OLDVAL]], s{{[0-9]+}}, m0
|
|
75 ; GFX10: v_writelane_b32 [[OLDVAL]], s{{[0-9]+}}, s{{[0-9]+}}
|
|
76 define amdgpu_kernel void @test_writelane_imm_oldval(i32 addrspace(1)* %out, i32 %src0, i32 %src1) #1 {
|
|
77 %writelane = call i32 @llvm.amdgcn.writelane(i32 %src0, i32 %src1, i32 42)
|
|
78 store i32 %writelane, i32 addrspace(1)* %out, align 4
|
|
79 ret void
|
|
80 }
|
|
81
|
|
82 declare i32 @llvm.amdgcn.workitem.id.x() #2
|
|
83
|
|
84 attributes #0 = { nounwind readnone convergent }
|
|
85 attributes #1 = { nounwind }
|
|
86 attributes #2 = { nounwind readnone }
|