1 ; RUN: llc -mtriple=amdgcn--amdhsa -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
3 ; If the workgroup id range is restricted, we should be able to use
4 ; mad24 for the usual indexing pattern.
6 declare i32 @llvm.amdgcn.workgroup.id.x() #0
7 declare i32 @llvm.amdgcn.workitem.id.x() #0
8 declare ptr addrspace(4) @llvm.amdgcn.dispatch.ptr() #0
10 ; GCN-LABEL: {{^}}get_global_id_0:
11 ; GCN: s_and_b32 [[WGSIZEX:s[0-9]+]], {{s[0-9]+}}, 0xffff
12 ; GCN: s_mul_i32 [[MUL:s[0-9]+]], s8, [[WGSIZEX]]
13 ; GCN: v_add_i32_e32 v{{[0-9]+}}, vcc, [[MUL]], v0
14 define amdgpu_kernel void @get_global_id_0(ptr addrspace(1) %out) #1 {
15 %dispatch.ptr = call ptr addrspace(4) @llvm.amdgcn.dispatch.ptr()
16 %gep = getelementptr inbounds i32, ptr addrspace(4) %dispatch.ptr, i64 1
17 %workgroup.size.xy = load i32, ptr addrspace(4) %gep, align 4, !invariant.load !0
18 %workgroup.size.x = and i32 %workgroup.size.xy, 65535
20 %workitem.id.x = call i32 @llvm.amdgcn.workitem.id.x(), !range !1
21 %workgroup.id.x = call i32 @llvm.amdgcn.workgroup.id.x(), !range !2
23 %mul = mul i32 %workgroup.id.x, %workgroup.size.x
24 %add = add i32 %mul, %workitem.id.x
26 store i32 %add, ptr addrspace(1) %out, align 4
30 attributes #0 = { nounwind readnone }
31 attributes #1 = { nounwind }
34 !1 = !{i32 0, i32 1024}
35 !2 = !{i32 0, i32 16777216}