[AMDGPU] Fix undefined scc register in successor block of SI_KILL terminators (#134718)

Fix issue 131298 where an undefined $scc register causes verifier errors
when using SI_KILL_F32_COND_IMM_TERMINATOR instructions. The problem
occurs because the $scc register defined in a comparison before the kill
terminator is used in successor blocks, but was not properly marked as live-in.

This patch:
- Adds code to check if SCC is used in the successor block
- Adds SCC as a live-in to successor blocks
- Handles both explicit and implicit uses of SCC

With this patch the machine verifier no longer reports undefined $scc
errors in following kill terminator instruction.

Fixes #131298

---------

Co-authored-by: Matt Arsenault <arsenm2@gmail.com>
This commit is contained in:
mssefat
2025-04-30 09:02:45 -05:00
committed by GitHub
parent 6aafd5ef06
commit 7495f92f08
4 changed files with 235 additions and 4 deletions

View File

@@ -47,6 +47,8 @@ static std::pair<bool, bool> runImpl(MachineFunction &MF) {
const TargetInstrInfo *TII = MF.getSubtarget().getInstrInfo();
const TargetLowering *TLI = MF.getSubtarget().getTargetLowering();
TLI->finalizeLowering(MF);
// Iterate through each instruction in the function, looking for pseudos.
for (MachineFunction::iterator I = MF.begin(), E = MF.end(); I != E; ++I) {
MachineBasicBlock *MBB = &*I;
@@ -74,9 +76,6 @@ static std::pair<bool, bool> runImpl(MachineFunction &MF) {
}
}
}
TLI->finalizeLowering(MF);
return {Changed, PreserveCFG};
}

View File

@@ -4513,7 +4513,7 @@ Register SITargetLowering::getRegisterByName(const char *RegName, LLT VT,
MachineBasicBlock *
SITargetLowering::splitKillBlock(MachineInstr &MI,
MachineBasicBlock *BB) const {
MachineBasicBlock *SplitBB = BB->splitAt(MI, false /*UpdateLiveIns*/);
MachineBasicBlock *SplitBB = BB->splitAt(MI, /*UpdateLiveIns=*/true);
const SIInstrInfo *TII = getSubtarget()->getInstrInfo();
MI.setDesc(TII->getKillTerminatorFromPseudo(MI.getOpcode()));
return SplitBB;

View File

@@ -0,0 +1,73 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
# RUN: llc -mtriple=amdgcn -run-pass finalize-isel -mtriple=amdgcn -mcpu=gfx1010 -verify-machineinstrs %s -o - | FileCheck %s
---
name: phi_use_def_before_kill
tracksRegLiveness: true
body: |
; CHECK-LABEL: name: phi_use_def_before_kill
; CHECK: bb.0:
; CHECK-NEXT: successors: %bb.3(0x80000000)
; CHECK-NEXT: liveins: $sgpr0, $sgpr1
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr_32 = COPY $sgpr1
; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr_32 = COPY $sgpr0
; CHECK-NEXT: [[S_MOV_B32_:%[0-9]+]]:sgpr_32 = S_MOV_B32 1065353216
; CHECK-NEXT: [[V_ADD_F32_e64_:%[0-9]+]]:vgpr_32 = nofpexcept V_ADD_F32_e64 0, [[COPY1]], 0, killed [[S_MOV_B32_]], 0, 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[S_MOV_B32_1:%[0-9]+]]:sgpr_32 = S_MOV_B32 0
; CHECK-NEXT: [[V_CMP_GT_F32_e64_:%[0-9]+]]:sreg_32_xm0_xexec = nofpexcept V_CMP_GT_F32_e64 0, [[V_ADD_F32_e64_]], 0, [[S_MOV_B32_1]], 0, implicit $mode, implicit $exec
; CHECK-NEXT: [[S_MOV_B32_2:%[0-9]+]]:sgpr_32 = S_MOV_B32 -1082130432
; CHECK-NEXT: [[COPY2:%[0-9]+]]:vgpr_32 = COPY killed [[S_MOV_B32_2]]
; CHECK-NEXT: [[V_CNDMASK_B32_e64_:%[0-9]+]]:vgpr_32 = V_CNDMASK_B32_e64 0, [[S_MOV_B32_1]], 0, [[COPY2]], killed [[V_CMP_GT_F32_e64_]], implicit $exec
; CHECK-NEXT: [[COPY3:%[0-9]+]]:sgpr_32 = COPY [[V_CNDMASK_B32_e64_]]
; CHECK-NEXT: [[S_MOV_B32_3:%[0-9]+]]:sreg_32 = S_MOV_B32 0
; CHECK-NEXT: S_CMP_LG_U32 [[COPY]], killed [[S_MOV_B32_3]], implicit-def $scc
; CHECK-NEXT: SI_KILL_F32_COND_IMM_TERMINATOR [[V_ADD_F32_e64_]], 0, 2, implicit-def $vcc_lo, implicit $exec
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.3:
; CHECK-NEXT: successors: %bb.1(0x40000000), %bb.2(0x40000000)
; CHECK-NEXT: liveins: $vcc_lo, $scc
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: S_CBRANCH_SCC1 %bb.1, implicit $scc
; CHECK-NEXT: S_CBRANCH_VCCNZ %bb.2, implicit $vcc_lo
; CHECK-NEXT: S_BRANCH %bb.2
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.1:
; CHECK-NEXT: [[S_MOV_B32_4:%[0-9]+]]:sgpr_32 = S_MOV_B32 0
; CHECK-NEXT: [[V_CMP_EQ_F32_e64_:%[0-9]+]]:sreg_32 = nofpexcept V_CMP_EQ_F32_e64 0, [[COPY]], 0, killed [[S_MOV_B32_4]], 0, implicit $mode, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: bb.2:
; CHECK-NEXT: [[S_MOV_B32_5:%[0-9]+]]:sgpr_32 = S_MOV_B32 0
; CHECK-NEXT: [[V_CMP_EQ_F32_e64_1:%[0-9]+]]:sreg_32 = nofpexcept V_CMP_EQ_F32_e64 0, [[COPY]], 0, killed [[S_MOV_B32_5]], 0, implicit $mode, implicit $exec
; CHECK-NEXT: S_ENDPGM 0
bb.0:
liveins: $sgpr0, $sgpr1
%3:sgpr_32 = COPY $sgpr1
%2:sgpr_32 = COPY $sgpr0
%5:sgpr_32 = S_MOV_B32 1065353216
%6:vgpr_32 = nofpexcept V_ADD_F32_e64 0, %2:sgpr_32, 0, killed %5:sgpr_32, 0, 0, implicit $mode, implicit $exec
%7:sgpr_32 = S_MOV_B32 0
%8:sreg_32_xm0_xexec = nofpexcept V_CMP_GT_F32_e64 0, %6:vgpr_32, 0, %7:sgpr_32, 0, implicit $mode, implicit $exec
%9:sgpr_32 = S_MOV_B32 -1082130432
%11:vgpr_32 = COPY killed %9:sgpr_32
%10:vgpr_32 = V_CNDMASK_B32_e64 0, %7:sgpr_32, 0, %11:vgpr_32, killed %8:sreg_32_xm0_xexec, implicit $exec
%0:sgpr_32 = COPY %10:vgpr_32
%12:sreg_32 = S_MOV_B32 0
S_CMP_LG_U32 %3:sgpr_32, killed %12:sreg_32, implicit-def $scc
SI_KILL_F32_COND_IMM_PSEUDO %6:vgpr_32, 0, 2, implicit-def $vcc, implicit $exec
S_CBRANCH_SCC1 %bb.1, implicit $scc
S_CBRANCH_VCCNZ %bb.2, implicit $vcc
S_BRANCH %bb.2
bb.1:
%13:sgpr_32 = S_MOV_B32 0
%14:sreg_32 = nofpexcept V_CMP_EQ_F32_e64 0, %3:sgpr_32, 0, killed %13:sgpr_32, 0, implicit $mode, implicit $exec
S_ENDPGM 0
bb.2:
%15:sgpr_32 = S_MOV_B32 0
%16:sreg_32 = nofpexcept V_CMP_EQ_F32_e64 0, %3:sgpr_32, 0, killed %15:sgpr_32, 0, implicit $mode, implicit $exec
S_ENDPGM 0
...

View File

@@ -1956,6 +1956,165 @@ bb.1:
ret void
}
define amdgpu_ps void @scc_use_after_kill_inst(float inreg %x, i32 inreg %y) #0 {
; SI-LABEL: scc_use_after_kill_inst:
; SI: ; %bb.0: ; %bb
; SI-NEXT: v_add_f32_e64 v1, s0, 1.0
; SI-NEXT: v_cmp_lt_f32_e32 vcc, 0, v1
; SI-NEXT: s_mov_b64 s[2:3], exec
; SI-NEXT: s_cmp_lg_u32 s1, 0
; SI-NEXT: v_cndmask_b32_e64 v0, 0, -1.0, vcc
; SI-NEXT: v_cmp_nlt_f32_e32 vcc, 0, v1
; SI-NEXT: s_andn2_b64 s[2:3], s[2:3], vcc
; SI-NEXT: s_cbranch_scc0 .LBB17_6
; SI-NEXT: ; %bb.1: ; %bb
; SI-NEXT: s_andn2_b64 exec, exec, vcc
; SI-NEXT: s_cbranch_scc0 .LBB17_3
; SI-NEXT: ; %bb.2: ; %bb8
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: v_mov_b32_e32 v0, 8
; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0)
; SI-NEXT: v_mov_b32_e32 v0, 4.0
; SI-NEXT: .LBB17_3: ; %phibb
; SI-NEXT: v_cmp_eq_f32_e32 vcc, 0, v0
; SI-NEXT: s_cbranch_vccz .LBB17_5
; SI-NEXT: ; %bb.4: ; %bb10
; SI-NEXT: s_mov_b32 s3, 0xf000
; SI-NEXT: s_mov_b32 s2, -1
; SI-NEXT: v_mov_b32_e32 v0, 9
; SI-NEXT: buffer_store_dword v0, off, s[0:3], 0
; SI-NEXT: s_waitcnt vmcnt(0)
; SI-NEXT: .LBB17_5: ; %end
; SI-NEXT: s_endpgm
; SI-NEXT: .LBB17_6:
; SI-NEXT: s_mov_b64 exec, 0
; SI-NEXT: exp null off, off, off, off done vm
; SI-NEXT: s_endpgm
;
; GFX10-WAVE64-LABEL: scc_use_after_kill_inst:
; GFX10-WAVE64: ; %bb.0: ; %bb
; GFX10-WAVE64-NEXT: v_add_f32_e64 v1, s0, 1.0
; GFX10-WAVE64-NEXT: s_mov_b64 s[2:3], exec
; GFX10-WAVE64-NEXT: s_cmp_lg_u32 s1, 0
; GFX10-WAVE64-NEXT: v_cmp_lt_f32_e32 vcc, 0, v1
; GFX10-WAVE64-NEXT: v_cndmask_b32_e64 v0, 0, -1.0, vcc
; GFX10-WAVE64-NEXT: v_cmp_nlt_f32_e32 vcc, 0, v1
; GFX10-WAVE64-NEXT: s_andn2_b64 s[2:3], s[2:3], vcc
; GFX10-WAVE64-NEXT: s_cbranch_scc0 .LBB17_6
; GFX10-WAVE64-NEXT: ; %bb.1: ; %bb
; GFX10-WAVE64-NEXT: s_andn2_b64 exec, exec, vcc
; GFX10-WAVE64-NEXT: s_cbranch_scc0 .LBB17_3
; GFX10-WAVE64-NEXT: ; %bb.2: ; %bb8
; GFX10-WAVE64-NEXT: v_mov_b32_e32 v1, 8
; GFX10-WAVE64-NEXT: v_mov_b32_e32 v0, 4.0
; GFX10-WAVE64-NEXT: global_store_dword v[0:1], v1, off
; GFX10-WAVE64-NEXT: s_waitcnt_vscnt null, 0x0
; GFX10-WAVE64-NEXT: .LBB17_3: ; %phibb
; GFX10-WAVE64-NEXT: v_cmp_eq_f32_e32 vcc, 0, v0
; GFX10-WAVE64-NEXT: s_cbranch_vccz .LBB17_5
; GFX10-WAVE64-NEXT: ; %bb.4: ; %bb10
; GFX10-WAVE64-NEXT: v_mov_b32_e32 v0, 9
; GFX10-WAVE64-NEXT: global_store_dword v[0:1], v0, off
; GFX10-WAVE64-NEXT: s_waitcnt_vscnt null, 0x0
; GFX10-WAVE64-NEXT: .LBB17_5: ; %end
; GFX10-WAVE64-NEXT: s_endpgm
; GFX10-WAVE64-NEXT: .LBB17_6:
; GFX10-WAVE64-NEXT: s_mov_b64 exec, 0
; GFX10-WAVE64-NEXT: exp null off, off, off, off done vm
; GFX10-WAVE64-NEXT: s_endpgm
;
; GFX10-WAVE32-LABEL: scc_use_after_kill_inst:
; GFX10-WAVE32: ; %bb.0: ; %bb
; GFX10-WAVE32-NEXT: v_add_f32_e64 v1, s0, 1.0
; GFX10-WAVE32-NEXT: s_mov_b32 s2, exec_lo
; GFX10-WAVE32-NEXT: s_cmp_lg_u32 s1, 0
; GFX10-WAVE32-NEXT: v_cmp_lt_f32_e32 vcc_lo, 0, v1
; GFX10-WAVE32-NEXT: v_cndmask_b32_e64 v0, 0, -1.0, vcc_lo
; GFX10-WAVE32-NEXT: v_cmp_nlt_f32_e32 vcc_lo, 0, v1
; GFX10-WAVE32-NEXT: s_andn2_b32 s2, s2, vcc_lo
; GFX10-WAVE32-NEXT: s_cbranch_scc0 .LBB17_6
; GFX10-WAVE32-NEXT: ; %bb.1: ; %bb
; GFX10-WAVE32-NEXT: s_andn2_b32 exec_lo, exec_lo, vcc_lo
; GFX10-WAVE32-NEXT: s_cbranch_scc0 .LBB17_3
; GFX10-WAVE32-NEXT: ; %bb.2: ; %bb8
; GFX10-WAVE32-NEXT: v_mov_b32_e32 v1, 8
; GFX10-WAVE32-NEXT: v_mov_b32_e32 v0, 4.0
; GFX10-WAVE32-NEXT: global_store_dword v[0:1], v1, off
; GFX10-WAVE32-NEXT: s_waitcnt_vscnt null, 0x0
; GFX10-WAVE32-NEXT: .LBB17_3: ; %phibb
; GFX10-WAVE32-NEXT: v_cmp_eq_f32_e32 vcc_lo, 0, v0
; GFX10-WAVE32-NEXT: s_cbranch_vccz .LBB17_5
; GFX10-WAVE32-NEXT: ; %bb.4: ; %bb10
; GFX10-WAVE32-NEXT: v_mov_b32_e32 v0, 9
; GFX10-WAVE32-NEXT: global_store_dword v[0:1], v0, off
; GFX10-WAVE32-NEXT: s_waitcnt_vscnt null, 0x0
; GFX10-WAVE32-NEXT: .LBB17_5: ; %end
; GFX10-WAVE32-NEXT: s_endpgm
; GFX10-WAVE32-NEXT: .LBB17_6:
; GFX10-WAVE32-NEXT: s_mov_b32 exec_lo, 0
; GFX10-WAVE32-NEXT: exp null off, off, off, off done vm
; GFX10-WAVE32-NEXT: s_endpgm
;
; GFX11-LABEL: scc_use_after_kill_inst:
; GFX11: ; %bb.0: ; %bb
; GFX11-NEXT: v_add_f32_e64 v1, s0, 1.0
; GFX11-NEXT: s_mov_b64 s[2:3], exec
; GFX11-NEXT: s_cmp_lg_u32 s1, 0
; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1)
; GFX11-NEXT: v_cmp_lt_f32_e32 vcc, 0, v1
; GFX11-NEXT: v_cndmask_b32_e64 v0, 0, -1.0, vcc
; GFX11-NEXT: v_cmp_nlt_f32_e32 vcc, 0, v1
; GFX11-NEXT: s_and_not1_b64 s[2:3], s[2:3], vcc
; GFX11-NEXT: s_cbranch_scc0 .LBB17_6
; GFX11-NEXT: ; %bb.1: ; %bb
; GFX11-NEXT: s_and_not1_b64 exec, exec, vcc
; GFX11-NEXT: s_cbranch_scc0 .LBB17_3
; GFX11-NEXT: ; %bb.2: ; %bb8
; GFX11-NEXT: v_mov_b32_e32 v1, 8
; GFX11-NEXT: v_mov_b32_e32 v0, 4.0
; GFX11-NEXT: global_store_b32 v[0:1], v1, off dlc
; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
; GFX11-NEXT: .LBB17_3: ; %phibb
; GFX11-NEXT: v_cmp_eq_f32_e32 vcc, 0, v0
; GFX11-NEXT: s_cbranch_vccz .LBB17_5
; GFX11-NEXT: ; %bb.4: ; %bb10
; GFX11-NEXT: v_mov_b32_e32 v0, 9
; GFX11-NEXT: global_store_b32 v[0:1], v0, off dlc
; GFX11-NEXT: s_waitcnt_vscnt null, 0x0
; GFX11-NEXT: .LBB17_5: ; %end
; GFX11-NEXT: s_endpgm
; GFX11-NEXT: .LBB17_6:
; GFX11-NEXT: s_mov_b64 exec, 0
; GFX11-NEXT: exp mrt0 off, off, off, off done
; GFX11-NEXT: s_endpgm
bb:
%tmp = fadd float %x, 1.000000e+00
%tmp1 = fcmp olt float 0.000000e+00, %tmp
%tmp2 = select i1 %tmp1, float -1.000000e+00, float 0.000000e+00
%cmp.tmp2 = fcmp olt float %tmp2, 0.000000e+00
%uniform.cond = icmp eq i32 %y, 0
call void @llvm.amdgcn.kill(i1 %cmp.tmp2)
br i1 %uniform.cond, label %phibb, label %bb8
phibb: ; preds = %bb8, %bb
%tmp5 = phi float [ %tmp2, %bb ], [ 4.000000e+00, %bb8 ]
%tmp6 = fcmp oeq float %tmp5, 0.000000e+00
br i1 %tmp6, label %bb10, label %end
bb8: ; preds = %bb
store volatile i32 8, ptr addrspace(1) poison, align 4
br label %phibb
bb10: ; preds = %phibb
store volatile i32 9, ptr addrspace(1) poison, align 4
br label %end
end: ; preds = %bb10, %phibb
ret void
}
declare void @llvm.amdgcn.exp.f32(i32 immarg, i32 immarg, float, float, float, float, i1 immarg, i1 immarg) #3
declare float @llvm.amdgcn.image.sample.l.2darray.f32.f32(i32 immarg, float, float, float, float, <8 x i32>, <4 x i32>, i1 immarg, i32 immarg, i32 immarg) #1
declare <4 x float> @llvm.amdgcn.image.sample.c.1d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1