Files
archived-llvm/test/CodeGen/AMDGPU/basic-branch.ll
Matt Arsenault 4404d0d6e3 AMDGPU: Implement SGPR spilling with scalar stores
nThis avoids the nasty problems caused by using
memory instructions that read the exec mask while
spilling / restoring registers used for control flow
masking, but only for VI when these were added.

This always uses the scalar stores when enabled currently,
but it may be better to still try to spill to a VGPR
and use this on the fallback memory path.

The cache also needs to be flushed before wave termination
if a scalar store is used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@286766 91177308-0d34-0410-b5e6-96231b3b80d8
2016-11-13 18:20:54 +00:00

55 lines
1.5 KiB
LLVM

; RUN: llc -O0 -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCNNOOPT -check-prefix=GCN %s
; RUN: llc -O0 -march=amdgcn -mcpu=tonga -amdgpu-spill-sgpr-to-smem=0 -verify-machineinstrs < %s | FileCheck -check-prefix=GCNNOOPT -check-prefix=GCN %s
; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck -check-prefix=GCNOPT -check-prefix=GCN %s
; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=GCNOPT -check-prefix=GCN %s
; GCN-LABEL: {{^}}test_branch:
; GCNNOOPT: v_writelane_b32
; GCNNOOPT: v_writelane_b32
; GCN: s_cbranch_scc1 [[END:BB[0-9]+_[0-9]+]]
; GCN: ; BB#1
; GCNNOOPT: v_readlane_b32
; GCNNOOPT: v_readlane_b32
; GCN: buffer_store_dword
; GCN: s_endpgm
; GCN: {{^}}[[END]]
; GCN: s_endpgm
define void @test_branch(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in, i32 %val) #0 {
%cmp = icmp ne i32 %val, 0
br i1 %cmp, label %store, label %end
store:
store i32 222, i32 addrspace(1)* %out
ret void
end:
ret void
}
; GCN-LABEL: {{^}}test_brcc_i1:
; GCN: buffer_load_ubyte
; GCN: v_and_b32_e32 v{{[0-9]+}}, 1,
; GCN: v_cmp_eq_u32_e32 vcc,
; GCN: s_cbranch_vccnz [[END:BB[0-9]+_[0-9]+]]
; GCN: buffer_store_dword
; GCN: s_endpgm
; GCN: {{^}}[[END]]
; GCN: s_endpgm
define void @test_brcc_i1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %in, i1 %val) #0 {
%cmp0 = icmp ne i1 %val, 0
br i1 %cmp0, label %store, label %end
store:
store i32 222, i32 addrspace(1)* %out
ret void
end:
ret void
}
attributes #0 = { nounwind }