Only select to a VGPR if it's trivally used in VGPR only contexts. This fixes mishandling frame indexes used in SGPR only contexts, like inline assembly constraints. This is suboptimal in the common case where the frame index is transitively used by only VALU ops. We make up for this by later folding the copy to VALU plus scalar op in SIFoldOperands.
345 lines
12 KiB
LLVM
345 lines
12 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
|
|
; RUN: llc -mtriple=amdgcn -mcpu=gfx1150 -amdgpu-enable-vopd=0 < %s | FileCheck -check-prefix=GCN %s
|
|
|
|
define amdgpu_ps void @test_export_zeroes_f32() #0 {
|
|
; GCN-LABEL: test_export_zeroes_f32:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: v_mov_b32_e32 v0, 0
|
|
; GCN-NEXT: exp mrt0 off, off, off, off
|
|
; GCN-NEXT: exp mrt0 off, off, off, off done
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_endpgm
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float 0.0, float 0.0, float 0.0, float 0.0, i1 false, i1 false)
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 0, float 0.0, float 0.0, float 0.0, float 0.0, i1 true, i1 false)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_export_en_src0_f32() #0 {
|
|
; GCN-LABEL: test_export_en_src0_f32:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: v_mov_b32_e32 v0, 4.0
|
|
; GCN-NEXT: v_mov_b32_e32 v1, 0.5
|
|
; GCN-NEXT: v_mov_b32_e32 v2, 2.0
|
|
; GCN-NEXT: v_mov_b32_e32 v3, 1.0
|
|
; GCN-NEXT: exp mrt0 v3, off, off, off done
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_endpgm
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 1, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_gs void @test_export_gs() #0 {
|
|
; GCN-LABEL: test_export_gs:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: v_mov_b32_e32 v0, 4.0
|
|
; GCN-NEXT: v_mov_b32_e32 v1, 0.5
|
|
; GCN-NEXT: v_mov_b32_e32 v2, 2.0
|
|
; GCN-NEXT: v_mov_b32_e32 v3, 1.0
|
|
; GCN-NEXT: exp mrt0 off, v2, off, off done
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_endpgm
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_hs void @test_export_hs() #0 {
|
|
; GCN-LABEL: test_export_hs:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: v_mov_b32_e32 v0, 4.0
|
|
; GCN-NEXT: v_mov_b32_e32 v1, 0.5
|
|
; GCN-NEXT: v_mov_b32_e32 v2, 2.0
|
|
; GCN-NEXT: v_mov_b32_e32 v3, 1.0
|
|
; GCN-NEXT: exp mrt0 off, v2, off, off done
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_endpgm
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_gfx void @test_export_gfx(float %v) #0 {
|
|
; GCN-LABEL: test_export_gfx:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: v_mov_b32_e32 v1, 4.0
|
|
; GCN-NEXT: v_mov_b32_e32 v2, 0.5
|
|
; GCN-NEXT: v_mov_b32_e32 v3, 2.0
|
|
; GCN-NEXT: exp mrt0 off, v3, off, off done
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_waitcnt_expcnt null, 0x0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: s_waitcnt expcnt(0)
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float %v, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_cs void @test_export_cs() #0 {
|
|
; GCN-LABEL: test_export_cs:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: v_mov_b32_e32 v0, 4.0
|
|
; GCN-NEXT: v_mov_b32_e32 v1, 0.5
|
|
; GCN-NEXT: v_mov_b32_e32 v2, 2.0
|
|
; GCN-NEXT: v_mov_b32_e32 v3, 1.0
|
|
; GCN-NEXT: exp mrt0 off, v2, off, off done
|
|
; GCN-NEXT: s_endpgm
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_kernel void @test_export_kernel() #0 {
|
|
; GCN-LABEL: test_export_kernel:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: v_mov_b32_e32 v0, 4.0
|
|
; GCN-NEXT: v_mov_b32_e32 v1, 0.5
|
|
; GCN-NEXT: v_mov_b32_e32 v2, 2.0
|
|
; GCN-NEXT: v_mov_b32_e32 v3, 1.0
|
|
; GCN-NEXT: exp mrt0 off, v2, off, off done
|
|
; GCN-NEXT: s_endpgm
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 2, float 1.0, float 2.0, float 0.5, float 4.0, i1 true, i1 false)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_gfx void @test_no_export_gfx(float %v) #0 {
|
|
; GCN-LABEL: test_no_export_gfx:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
|
|
; GCN-NEXT: s_setpc_b64 s[30:31]
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_no_export_ps(float %v) #0 {
|
|
; GCN-LABEL: test_no_export_ps:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_endpgm
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_if_export_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
|
|
; GCN-LABEL: test_if_export_f32:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: s_mov_b32 s0, exec_lo
|
|
; GCN-NEXT: v_cmpx_ne_u32_e32 0, v0
|
|
; GCN-NEXT: s_cbranch_execz .LBB9_2
|
|
; GCN-NEXT: ; %bb.1: ; %exp
|
|
; GCN-NEXT: exp mrt0 v1, v2, v3, v4
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_waitcnt_expcnt null, 0x0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: .LBB9_2: ; %end
|
|
; GCN-NEXT: s_endpgm
|
|
%cc = icmp eq i32 %flag, 0
|
|
br i1 %cc, label %end, label %exp
|
|
|
|
exp:
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 false, i1 false)
|
|
br label %end
|
|
|
|
end:
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_if_export_vm_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
|
|
; GCN-LABEL: test_if_export_vm_f32:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: s_mov_b32 s0, exec_lo
|
|
; GCN-NEXT: v_cmpx_ne_u32_e32 0, v0
|
|
; GCN-NEXT: s_cbranch_execz .LBB10_2
|
|
; GCN-NEXT: ; %bb.1: ; %exp
|
|
; GCN-NEXT: exp mrt0 v1, v2, v3, v4
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_waitcnt_expcnt null, 0x0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: .LBB10_2: ; %end
|
|
; GCN-NEXT: s_endpgm
|
|
%cc = icmp eq i32 %flag, 0
|
|
br i1 %cc, label %end, label %exp
|
|
|
|
exp:
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 false, i1 true)
|
|
br label %end
|
|
|
|
end:
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_if_export_done_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
|
|
; GCN-LABEL: test_if_export_done_f32:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: s_mov_b32 s0, exec_lo
|
|
; GCN-NEXT: v_cmpx_ne_u32_e32 0, v0
|
|
; GCN-NEXT: s_cbranch_execz .LBB11_2
|
|
; GCN-NEXT: ; %bb.1: ; %exp
|
|
; GCN-NEXT: exp mrt0 v1, v2, v3, v4 done
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_waitcnt_expcnt null, 0x0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: .LBB11_2: ; %end
|
|
; GCN-NEXT: s_endpgm
|
|
%cc = icmp eq i32 %flag, 0
|
|
br i1 %cc, label %end, label %exp
|
|
|
|
exp:
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 true, i1 false)
|
|
br label %end
|
|
|
|
end:
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_if_export_vm_done_f32(i32 %flag, float %x, float %y, float %z, float %w) #0 {
|
|
; GCN-LABEL: test_if_export_vm_done_f32:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: s_mov_b32 s0, exec_lo
|
|
; GCN-NEXT: v_cmpx_ne_u32_e32 0, v0
|
|
; GCN-NEXT: s_cbranch_execz .LBB12_2
|
|
; GCN-NEXT: ; %bb.1: ; %exp
|
|
; GCN-NEXT: exp mrt0 v1, v2, v3, v4 done
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_waitcnt_expcnt null, 0x0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: .LBB12_2: ; %end
|
|
; GCN-NEXT: s_endpgm
|
|
%cc = icmp eq i32 %flag, 0
|
|
br i1 %cc, label %end, label %exp
|
|
|
|
exp:
|
|
call void @llvm.amdgcn.exp.f32(i32 0, i32 15, float %x, float %y, float %z, float %w, i1 true, i1 true)
|
|
br label %end
|
|
|
|
end:
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_export_pos_before_param_across_load(i32 %idx) #0 {
|
|
; GCN-LABEL: test_export_pos_before_param_across_load:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: buffer_load_b32 v0, v0, s[0:3], 0 offen
|
|
; GCN-NEXT: v_mov_b32_e32 v1, 0
|
|
; GCN-NEXT: v_mov_b32_e32 v2, 1.0
|
|
; GCN-NEXT: v_mov_b32_e32 v3, 0.5
|
|
; GCN-NEXT: s_waitcnt vmcnt(0)
|
|
; GCN-NEXT: exp pos0 v1, v1, v1, v0 done
|
|
; GCN-NEXT: exp invalid_target_32 v2, v2, v2, v2
|
|
; GCN-NEXT: exp invalid_target_33 v2, v2, v2, v3
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_endpgm
|
|
call void @llvm.amdgcn.exp.f32(i32 32, i32 15, float 1.0, float 1.0, float 1.0, float 1.0, i1 false, i1 false)
|
|
call void @llvm.amdgcn.exp.f32(i32 33, i32 15, float 1.0, float 1.0, float 1.0, float 0.5, i1 false, i1 false)
|
|
%load = call float @llvm.amdgcn.raw.ptr.buffer.load.f32(ptr addrspace(8) undef, i32 %idx, i32 0, i32 0)
|
|
call void @llvm.amdgcn.exp.f32(i32 12, i32 15, float 0.0, float 0.0, float 0.0, float %load, i1 true, i1 false)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_export_across_store_load(i32 %idx, float %v) #0 {
|
|
; GCN-LABEL: test_export_across_store_load:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: v_mov_b32_e32 v2, 0
|
|
; GCN-NEXT: v_cmp_eq_u32_e32 vcc_lo, 1, v0
|
|
; GCN-NEXT: s_delay_alu instid0(VALU_DEP_2)
|
|
; GCN-NEXT: v_cndmask_b32_e32 v0, 16, v2, vcc_lo
|
|
; GCN-NEXT: v_mov_b32_e32 v2, 0
|
|
; GCN-NEXT: scratch_store_b32 v0, v1, off
|
|
; GCN-NEXT: scratch_load_b32 v0, off, off
|
|
; GCN-NEXT: v_mov_b32_e32 v1, 1.0
|
|
; GCN-NEXT: exp pos0 v2, v2, v2, v1 done
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_waitcnt_expcnt null, 0x0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: s_waitcnt vmcnt(0)
|
|
; GCN-NEXT: exp invalid_target_32 v0, v2, v1, v2
|
|
; GCN-NEXT: exp invalid_target_33 v0, v2, v1, v2
|
|
; GCN-NEXT: s_setprio 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_nop 0
|
|
; GCN-NEXT: s_endpgm
|
|
%data0 = alloca <4 x float>, align 8, addrspace(5)
|
|
%data1 = alloca <4 x float>, align 8, addrspace(5)
|
|
%cmp = icmp eq i32 %idx, 1
|
|
%data = select i1 %cmp, ptr addrspace(5) %data0, ptr addrspace(5) %data1
|
|
store float %v, ptr addrspace(5) %data, align 8
|
|
call void @llvm.amdgcn.exp.f32(i32 12, i32 15, float 0.0, float 0.0, float 0.0, float 1.0, i1 true, i1 false)
|
|
%load0 = load float, ptr addrspace(5) %data0, align 8
|
|
call void @llvm.amdgcn.exp.f32(i32 32, i32 15, float %load0, float 0.0, float 1.0, float 0.0, i1 false, i1 false)
|
|
call void @llvm.amdgcn.exp.f32(i32 33, i32 15, float %load0, float 0.0, float 1.0, float 0.0, i1 false, i1 false)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_export_in_callee(float %v) #0 {
|
|
; GCN-LABEL: test_export_in_callee:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: s_getpc_b64 s[0:1]
|
|
; GCN-NEXT: s_add_u32 s0, s0, test_export_gfx@gotpcrel32@lo+4
|
|
; GCN-NEXT: s_addc_u32 s1, s1, test_export_gfx@gotpcrel32@hi+12
|
|
; GCN-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; GCN-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
|
|
; GCN-NEXT: s_mov_b32 s32, 0
|
|
; GCN-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GCN-NEXT: s_swappc_b64 s[30:31], s[0:1]
|
|
; GCN-NEXT: s_endpgm
|
|
%x = fadd float %v, 1.0
|
|
call void @test_export_gfx(float %x)
|
|
ret void
|
|
}
|
|
|
|
define amdgpu_ps void @test_export_in_callee_prio(float %v) #0 {
|
|
; GCN-LABEL: test_export_in_callee_prio:
|
|
; GCN: ; %bb.0:
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: s_mov_b32 s32, 0
|
|
; GCN-NEXT: v_add_f32_e32 v0, 1.0, v0
|
|
; GCN-NEXT: s_setprio 2
|
|
; GCN-NEXT: s_getpc_b64 s[0:1]
|
|
; GCN-NEXT: s_add_u32 s0, s0, test_export_gfx@gotpcrel32@lo+4
|
|
; GCN-NEXT: s_addc_u32 s1, s1, test_export_gfx@gotpcrel32@hi+12
|
|
; GCN-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
|
|
; GCN-NEXT: s_waitcnt lgkmcnt(0)
|
|
; GCN-NEXT: s_swappc_b64 s[30:31], s[0:1]
|
|
; GCN-NEXT: s_endpgm
|
|
%x = fadd float %v, 1.0
|
|
call void @llvm.amdgcn.s.setprio(i16 0)
|
|
call void @test_export_gfx(float %x)
|
|
ret void
|
|
}
|
|
|
|
declare void @llvm.amdgcn.exp.f32(i32, i32, float, float, float, float, i1, i1) #1
|
|
declare void @llvm.amdgcn.exp.i32(i32, i32, i32, i32, i32, i32, i1, i1) #1
|
|
declare float @llvm.amdgcn.raw.ptr.buffer.load.f32(ptr addrspace(8), i32, i32, i32) #2
|
|
declare void @llvm.amdgcn.s.setprio(i16)
|
|
|
|
attributes #0 = { nounwind }
|
|
attributes #1 = { nounwind inaccessiblememonly }
|
|
attributes #2 = { nounwind readnone }
|