llvm/test/CodeGen/AMDGPU/operand-folding.ll
Matt Arsenault d706d030af AMDGPU: Mark all unspecified CC functions in tests as amdgpu_kernel
Currently the default C calling convention functions are treated
the same as compute kernels. Make this explicit so the default
calling convention can be changed to a non-kernel.

Converted with perl -pi -e 's/define void/define amdgpu_kernel void/'
on the relevant test directories (and undoing in one place that actually
wanted a non-kernel).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@298444 91177308-0d34-0410-b5e6-96231b3b80d8
2017-03-21 21:39:51 +00:00

130 lines
4.2 KiB
LLVM

; RUN: llc -march=amdgcn -verify-machineinstrs < %s | FileCheck %s
; CHECK-LABEL: {{^}}fold_sgpr:
; CHECK: v_add_i32_e32 v{{[0-9]+}}, vcc, s
define amdgpu_kernel void @fold_sgpr(i32 addrspace(1)* %out, i32 %fold) {
entry:
%tmp0 = icmp ne i32 %fold, 0
br i1 %tmp0, label %if, label %endif
if:
%id = call i32 @llvm.amdgcn.workitem.id.x()
%offset = add i32 %fold, %id
%tmp1 = getelementptr i32, i32 addrspace(1)* %out, i32 %offset
store i32 0, i32 addrspace(1)* %tmp1
br label %endif
endif:
ret void
}
; CHECK-LABEL: {{^}}fold_imm:
; CHECK: v_or_b32_e32 v{{[0-9]+}}, 5
define amdgpu_kernel void @fold_imm(i32 addrspace(1)* %out, i32 %cmp) {
entry:
%fold = add i32 3, 2
%tmp0 = icmp ne i32 %cmp, 0
br i1 %tmp0, label %if, label %endif
if:
%id = call i32 @llvm.amdgcn.workitem.id.x()
%val = or i32 %id, %fold
store i32 %val, i32 addrspace(1)* %out
br label %endif
endif:
ret void
}
; CHECK-LABEL: {{^}}fold_64bit_constant_add:
; CHECK-NOT: s_mov_b64
; FIXME: It would be better if we could use v_add here and drop the extra
; v_mov_b32 instructions.
; CHECK-DAG: s_add_u32 [[LO:s[0-9]+]], s{{[0-9]+}}, 1
; CHECK-DAG: s_addc_u32 [[HI:s[0-9]+]], s{{[0-9]+}}, 0
; CHECK-DAG: v_mov_b32_e32 v[[VLO:[0-9]+]], [[LO]]
; CHECK-DAG: v_mov_b32_e32 v[[VHI:[0-9]+]], [[HI]]
; CHECK: buffer_store_dwordx2 v{{\[}}[[VLO]]:[[VHI]]{{\]}},
define amdgpu_kernel void @fold_64bit_constant_add(i64 addrspace(1)* %out, i32 %cmp, i64 %val) {
entry:
%tmp0 = add i64 %val, 1
store i64 %tmp0, i64 addrspace(1)* %out
ret void
}
; Inline constants should always be folded.
; CHECK-LABEL: {{^}}vector_inline:
; CHECK: v_xor_b32_e32 v{{[0-9]+}}, 5, v{{[0-9]+}}
; CHECK: v_xor_b32_e32 v{{[0-9]+}}, 5, v{{[0-9]+}}
; CHECK: v_xor_b32_e32 v{{[0-9]+}}, 5, v{{[0-9]+}}
; CHECK: v_xor_b32_e32 v{{[0-9]+}}, 5, v{{[0-9]+}}
define amdgpu_kernel void @vector_inline(<4 x i32> addrspace(1)* %out) {
entry:
%tmp0 = call i32 @llvm.amdgcn.workitem.id.x()
%tmp1 = add i32 %tmp0, 1
%tmp2 = add i32 %tmp0, 2
%tmp3 = add i32 %tmp0, 3
%vec0 = insertelement <4 x i32> undef, i32 %tmp0, i32 0
%vec1 = insertelement <4 x i32> %vec0, i32 %tmp1, i32 1
%vec2 = insertelement <4 x i32> %vec1, i32 %tmp2, i32 2
%vec3 = insertelement <4 x i32> %vec2, i32 %tmp3, i32 3
%tmp4 = xor <4 x i32> <i32 5, i32 5, i32 5, i32 5>, %vec3
store <4 x i32> %tmp4, <4 x i32> addrspace(1)* %out
ret void
}
; Immediates with one use should be folded
; CHECK-LABEL: {{^}}imm_one_use:
; CHECK: v_xor_b32_e32 v{{[0-9]+}}, 0x64, v{{[0-9]+}}
define amdgpu_kernel void @imm_one_use(i32 addrspace(1)* %out) {
entry:
%tmp0 = call i32 @llvm.amdgcn.workitem.id.x()
%tmp1 = xor i32 %tmp0, 100
store i32 %tmp1, i32 addrspace(1)* %out
ret void
}
; CHECK-LABEL: {{^}}vector_imm:
; CHECK: s_movk_i32 [[IMM:s[0-9]+]], 0x64
; CHECK: v_xor_b32_e32 v{{[0-9]}}, [[IMM]], v{{[0-9]}}
; CHECK: v_xor_b32_e32 v{{[0-9]}}, [[IMM]], v{{[0-9]}}
; CHECK: v_xor_b32_e32 v{{[0-9]}}, [[IMM]], v{{[0-9]}}
; CHECK: v_xor_b32_e32 v{{[0-9]}}, [[IMM]], v{{[0-9]}}
define amdgpu_kernel void @vector_imm(<4 x i32> addrspace(1)* %out) {
entry:
%tmp0 = call i32 @llvm.amdgcn.workitem.id.x()
%tmp1 = add i32 %tmp0, 1
%tmp2 = add i32 %tmp0, 2
%tmp3 = add i32 %tmp0, 3
%vec0 = insertelement <4 x i32> undef, i32 %tmp0, i32 0
%vec1 = insertelement <4 x i32> %vec0, i32 %tmp1, i32 1
%vec2 = insertelement <4 x i32> %vec1, i32 %tmp2, i32 2
%vec3 = insertelement <4 x i32> %vec2, i32 %tmp3, i32 3
%tmp4 = xor <4 x i32> <i32 100, i32 100, i32 100, i32 100>, %vec3
store <4 x i32> %tmp4, <4 x i32> addrspace(1)* %out
ret void
}
; A subregister use operand should not be tied.
; CHECK-LABEL: {{^}}no_fold_tied_subregister:
; CHECK: buffer_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
; CHECK: v_mac_f32_e32 v[[LO]], 0x41200000, v[[HI]]
; CHECK: buffer_store_dword v[[LO]]
define amdgpu_kernel void @no_fold_tied_subregister() {
%tmp1 = load volatile <2 x float>, <2 x float> addrspace(1)* undef
%tmp2 = extractelement <2 x float> %tmp1, i32 0
%tmp3 = extractelement <2 x float> %tmp1, i32 1
%tmp4 = fmul float %tmp3, 10.0
%tmp5 = fadd float %tmp4, %tmp2
store volatile float %tmp5, float addrspace(1)* undef
ret void
}
declare i32 @llvm.amdgcn.workitem.id.x() #0
attributes #0 = { nounwind readnone }