[MCP] Remove dead copies from basic blocks with successors. (#86973)

topperc · vladimirradosavljevic · commit b4327a0b6430 · 2024-10-09T16:56:14.000+02:00
Previously we wouldn't remove dead copies from basic blocks with
successors. The comment said we didn't want to trust the live-in lists.
The comment is very old so I'm not sure if that's still a concern today.

This patch checks the live-in lists and removes copies from
MaybeDeadCopies if they are referenced by any live-ins in any
successors. We only do this if the tracksLiveness property is set. If
that property is not set, we retain the old behavior.
diff --git a/llvm/lib/CodeGen/MachineCopyPropagation.cpp b/llvm/lib/CodeGen/MachineCopyPropagation.cpp
@@ -370,6 +370,7 @@ class MachineCopyPropagation : public MachineFunctionPass {
   typedef enum { DebugUse = false, RegularUse = true } DebugType;
 
   void ReadRegister(MCRegister Reg, MachineInstr &Reader, DebugType DT);
+  void readSuccessorLiveIns(const MachineBasicBlock &MBB);
   void ForwardCopyPropagateBlock(MachineBasicBlock &MBB);
   void BackwardCopyPropagateBlock(MachineBasicBlock &MBB);
   void EliminateSpillageCopies(MachineBasicBlock &MBB);
@@ -422,6 +423,22 @@ void MachineCopyPropagation::ReadRegister(MCRegister Reg, MachineInstr &Reader,
   }
 }
 
+void MachineCopyPropagation::readSuccessorLiveIns(
+    const MachineBasicBlock &MBB) {
+  if (MaybeDeadCopies.empty())
+    return;
+
+  // If a copy result is livein to a successor, it is not dead.
+  for (const MachineBasicBlock *Succ : MBB.successors()) {
+    for (const auto &LI : Succ->liveins()) {
+      for (MCRegUnit Unit : TRI->regunits(LI.PhysReg)) {
+        if (MachineInstr *Copy = Tracker.findCopyForUnit(Unit, *TRI))
+          MaybeDeadCopies.remove(Copy);
+      }
+    }
+  }
+}
+
 /// Return true if \p PreviousCopy did copy register \p Src to register \p Def.
 /// This fact may have been obscured by sub register usage or may not be true at
 /// all even though Src and Def are subregisters of the registers used in
@@ -873,10 +890,17 @@ void MachineCopyPropagation::ForwardCopyPropagateBlock(MachineBasicBlock &MBB) {
       Tracker.clobberRegister(Reg, *TRI, *TII, UseCopyInstr);
   }
 
-  // If MBB doesn't have successors, delete the copies whose defs are not used.
-  // If MBB does have successors, then conservative assume the defs are live-out
-  // since we don't want to trust live-in lists.
-  if (MBB.succ_empty()) {
+  bool TracksLiveness = MRI->tracksLiveness();
+
+  // If liveness is tracked, we can use the live-in lists to know which
+  // copies aren't dead.
+  if (TracksLiveness)
+    readSuccessorLiveIns(MBB);
+
+  // If MBB doesn't have succesor, delete copies whose defs are not used.
+  // If MBB does have successors, we can only delete copies if we are able to
+  // use liveness information from successors to confirm they are really dead.
+  if (MBB.succ_empty() || TracksLiveness) {
     for (MachineInstr *MaybeDead : MaybeDeadCopies) {
       LLVM_DEBUG(dbgs() << "MCP: Removing copy due to no live-out succ: ";
                  MaybeDead->dump());
diff --git a/llvm/test/CodeGen/AMDGPU/atomic_optimizations_pixelshader.ll b/llvm/test/CodeGen/AMDGPU/atomic_optimizations_pixelshader.ll
@@ -468,7 +468,6 @@ define amdgpu_ps void @add_i32_varying(ptr addrspace(8) inreg %out, ptr addrspac
 ; GFX1032-NEXT:    s_cbranch_execz .LBB1_3
 ; GFX1032-NEXT:  ; %bb.2:
 ; GFX1032-NEXT:    v_mov_b32_e32 v0, s11
-; GFX1032-NEXT:    s_mov_b32 s10, s11
 ; GFX1032-NEXT:    buffer_atomic_add v0, off, s[4:7], 0 glc
 ; GFX1032-NEXT:  .LBB1_3:
 ; GFX1032-NEXT:    s_waitcnt_depctr 0xffe3
@@ -604,7 +603,6 @@ define amdgpu_ps void @add_i32_varying(ptr addrspace(8) inreg %out, ptr addrspac
 ; GFX1132-NEXT:    s_cbranch_execz .LBB1_3
 ; GFX1132-NEXT:  ; %bb.2:
 ; GFX1132-NEXT:    v_mov_b32_e32 v0, s11
-; GFX1132-NEXT:    s_mov_b32 s10, s11
 ; GFX1132-NEXT:    buffer_atomic_add_u32 v0, off, s[4:7], 0 glc
 ; GFX1132-NEXT:  .LBB1_3:
 ; GFX1132-NEXT:    s_or_b32 exec_lo, exec_lo, s9
diff --git a/llvm/test/CodeGen/PowerPC/stack-restore-with-setjmp.ll b/llvm/test/CodeGen/PowerPC/stack-restore-with-setjmp.ll
@@ -30,7 +30,6 @@ define dso_local signext i32 @main(i32 signext %argc, ptr nocapture readnone %ar
 ; CHECK-NEXT:    # kill: def $r3 killed $r3 killed $x3
 ; CHECK-NEXT:    cmpwi 3, 0
 ; CHECK-NEXT:    crorc 20, 10, 2
-; CHECK-NEXT:    crmove 21, 2
 ; CHECK-NEXT:    bc 4, 20, .LBB0_4
 ; CHECK-NEXT:  # %bb.2: # %if.end5
 ; CHECK-NEXT:    addis 3, 2, .L.str@toc@ha
@@ -75,11 +74,9 @@ define dso_local signext i32 @main(i32 signext %argc, ptr nocapture readnone %ar
 ; BE-NEXT:    addi 3, 31, 128
 ; BE-NEXT:    bl _setjmp
 ; BE-NEXT:    nop
-; BE-NEXT:    crmove 20, 10
 ; BE-NEXT:    # kill: def $r3 killed $r3 killed $x3
 ; BE-NEXT:    cmpwi 3, 0
 ; BE-NEXT:    crorc 20, 10, 2
-; BE-NEXT:    crmove 21, 2
 ; BE-NEXT:    bc 4, 20, .LBB0_4
 ; BE-NEXT:  # %bb.2: # %if.end5
 ; BE-NEXT:    addis 3, 2, .L.str@toc@ha
diff --git a/llvm/test/CodeGen/Thumb2/aligned-nonfallthrough.ll b/llvm/test/CodeGen/Thumb2/aligned-nonfallthrough.ll
@@ -42,9 +42,8 @@ define i64 @loopif(ptr nocapture readonly %x, i32 %y, i32 %n) {
 ; CHECK-NEXT:    cmp r2, #1
 ; CHECK-NEXT:    blt .LBB1_4
 ; CHECK-NEXT:  @ %bb.1: @ %for.body.lr.ph
-; CHECK-NEXT:    mov lr, r2
-; CHECK-NEXT:    mov r12, r0
 ; CHECK-NEXT:    dls lr, r2
+; CHECK-NEXT:    mov r12, r0
 ; CHECK-NEXT:    movs r0, #0
 ; CHECK-NEXT:    movs r3, #0
 ; CHECK-NEXT:    .p2align 2
diff --git a/llvm/test/CodeGen/Thumb2/mve-gather-increment.ll b/llvm/test/CodeGen/Thumb2/mve-gather-increment.ll
@@ -542,9 +542,7 @@ define arm_aapcs_vfpcc void @gather_inc_v8i16_simple(ptr noalias nocapture reado
 ; CHECK-NEXT:    .pad #28
 ; CHECK-NEXT:    sub sp, #28
 ; CHECK-NEXT:    cmp r2, #1
-; CHECK-NEXT:    str r1, [sp, #4] @ 4-byte Spill
-; CHECK-NEXT:    mov r1, r2
-; CHECK-NEXT:    str r2, [sp, #8] @ 4-byte Spill
+; CHECK-NEXT:    strd r1, r2, [sp, #4] @ 8-byte Folded Spill
 ; CHECK-NEXT:    blt .LBB11_5
 ; CHECK-NEXT:  @ %bb.1: @ %vector.ph.preheader
 ; CHECK-NEXT:    ldr r1, [sp, #8] @ 4-byte Reload
@@ -661,9 +659,7 @@ define arm_aapcs_vfpcc void @gather_inc_v8i16_complex(ptr noalias nocapture read
 ; CHECK-NEXT:    .pad #136
 ; CHECK-NEXT:    sub sp, #136
 ; CHECK-NEXT:    cmp r2, #1
-; CHECK-NEXT:    str r1, [sp, #64] @ 4-byte Spill
-; CHECK-NEXT:    mov r1, r2
-; CHECK-NEXT:    str r2, [sp, #68] @ 4-byte Spill
+; CHECK-NEXT:    strd r1, r2, [sp, #64] @ 8-byte Folded Spill
 ; CHECK-NEXT:    blt.w .LBB12_5
 ; CHECK-NEXT:  @ %bb.1: @ %vector.ph.preheader
 ; CHECK-NEXT:    ldr r1, [sp, #68] @ 4-byte Reload
@@ -952,11 +948,9 @@ define arm_aapcs_vfpcc void @gather_inc_v16i8_complex(ptr noalias nocapture read
 ; CHECK-NEXT:    vstrw.32 q1, [sp, #152] @ 16-byte Spill
 ; CHECK-NEXT:    vldrw.u32 q1, [sp, #296] @ 16-byte Reload
 ; CHECK-NEXT:    vstrw.32 q0, [sp, #168] @ 16-byte Spill
-; CHECK-NEXT:    vmov q0, q2
-; CHECK-NEXT:    vmov q3, q5
-; CHECK-NEXT:    vadd.i32 q1, q1, r0
 ; CHECK-NEXT:    vldrw.u32 q0, [sp, #248] @ 16-byte Reload
 ; CHECK-NEXT:    vldrw.u32 q3, [sp, #216] @ 16-byte Reload
+; CHECK-NEXT:    vadd.i32 q1, q1, r0
 ; CHECK-NEXT:    vstrw.32 q5, [sp, #120] @ 16-byte Spill
 ; CHECK-NEXT:    vadd.i32 q0, q0, r0
 ; CHECK-NEXT:    subs.w r11, r11, #16
@@ -1243,9 +1237,7 @@ define arm_aapcs_vfpcc void @gather_inc_v16i8_simple(ptr noalias nocapture reado
 ; CHECK-NEXT:    .pad #64
 ; CHECK-NEXT:    sub sp, #64
 ; CHECK-NEXT:    cmp r2, #1
-; CHECK-NEXT:    str r1, [sp, #56] @ 4-byte Spill
-; CHECK-NEXT:    mov r1, r2
-; CHECK-NEXT:    str r2, [sp, #60] @ 4-byte Spill
+; CHECK-NEXT:    strd r1, r2, [sp, #56] @ 8-byte Folded Spill
 ; CHECK-NEXT:    blt.w .LBB14_5
 ; CHECK-NEXT:  @ %bb.1: @ %vector.ph.preheader
 ; CHECK-NEXT:    adr r5, .LCPI14_3
diff --git a/llvm/test/CodeGen/Thumb2/mve-gather-scatter-optimisation.ll b/llvm/test/CodeGen/Thumb2/mve-gather-scatter-optimisation.ll
@@ -608,7 +608,6 @@ define dso_local void @arm_mat_mult_q15(i16* noalias nocapture readonly %A, i16*
 ; CHECK-NEXT:    strd r0, r2, [sp, #24] @ 8-byte Folded Spill
 ; CHECK-NEXT:    cmp r3, #0
 ; CHECK-NEXT:    str r3, [sp, #8] @ 4-byte Spill
-; CHECK-NEXT:    mov r0, r3
 ; CHECK-NEXT:    itt ne
 ; CHECK-NEXT:    ldrne r0, [sp, #136]
 ; CHECK-NEXT:    cmpne r0, #0
diff --git a/llvm/test/CodeGen/Thumb2/mve-laneinterleaving-reduct.ll b/llvm/test/CodeGen/Thumb2/mve-laneinterleaving-reduct.ll
@@ -118,9 +118,7 @@ define void @correlate(ptr nocapture noundef readonly %ID, ptr nocapture noundef
 ; CHECK-NEXT:    .pad #12
 ; CHECK-NEXT:    sub sp, #12
 ; CHECK-NEXT:    cmp r3, #1
-; CHECK-NEXT:    strd r0, r1, [sp] @ 8-byte Folded Spill
-; CHECK-NEXT:    mov r1, r3
-; CHECK-NEXT:    str r3, [sp, #8] @ 4-byte Spill
+; CHECK-NEXT:    stm.w sp, {r0, r1, r3} @ 12-byte Folded Spill
 ; CHECK-NEXT:    blt .LBB4_12
 ; CHECK-NEXT:  @ %bb.1: @ %for.body.lr.ph
 ; CHECK-NEXT:    ldr r1, [sp, #48]
diff --git a/llvm/test/CodeGen/Thumb2/mve-postinc-lsr.ll b/llvm/test/CodeGen/Thumb2/mve-postinc-lsr.ll
@@ -1062,9 +1062,8 @@ define arm_aapcs_vfpcc void @_Z37_arm_radix4_butterfly_inverse_f32_mvePK21arm_cf
 ; CHECK-NEXT:    .pad #40
 ; CHECK-NEXT:    sub sp, #40
 ; CHECK-NEXT:    cmp r2, #8
-; CHECK-NEXT:    str r1, [sp, #16] @ 4-byte Spill
 ; CHECK-NEXT:    vstr s0, [sp] @ 4-byte Spill
-; CHECK-NEXT:    mov r1, r2
+; CHECK-NEXT:    str r1, [sp, #16] @ 4-byte Spill
 ; CHECK-NEXT:    str r2, [sp, #4] @ 4-byte Spill
 ; CHECK-NEXT:    blo .LBB7_9
 ; CHECK-NEXT:  @ %bb.1:
diff --git a/llvm/test/CodeGen/Thumb2/mve-vldst4.ll b/llvm/test/CodeGen/Thumb2/mve-vldst4.ll
@@ -95,14 +95,13 @@ define void @vldst4(ptr nocapture readonly %pIn, ptr nocapture %pOut, i32 %numRo
 ; CHECK-NEXT:    vmovx.f16 s8, s27
 ; CHECK-NEXT:    vins.f16 s12, s24
 ; CHECK-NEXT:    vins.f16 s13, s25
+; CHECK-NEXT:    vins.f16 s2, s10
 ; CHECK-NEXT:    vins.f16 s3, s11
 ; CHECK-NEXT:    vins.f16 s1, s9
-; CHECK-NEXT:    vins.f16 s2, s10
 ; CHECK-NEXT:    vins.f16 s22, s8
 ; CHECK-NEXT:    vmov q2, q3
-; CHECK-NEXT:    vmov.f32 s17, s0
-; CHECK-NEXT:    vmov.f32 s10, s4
 ; CHECK-NEXT:    vmov q6, q0
+; CHECK-NEXT:    vmov.f32 s10, s4
 ; CHECK-NEXT:    vmov.f32 s11, s7
 ; CHECK-NEXT:    vmov.f32 s9, s0
 ; CHECK-NEXT:    vmov.f32 s17, s2
diff --git a/llvm/test/CodeGen/X86/optimize-max-0.ll b/llvm/test/CodeGen/X86/optimize-max-0.ll
@@ -492,7 +492,6 @@ define void @bar(ptr %r, i32 %s, i32 %w, i32 %x, ptr %j, i32 %d) nounwind {
 ; CHECK-NEXT:    jb LBB1_4
 ; CHECK-NEXT:  ## %bb.5: ## %bb9
 ; CHECK-NEXT:    ## in Loop: Header=BB1_4 Depth=1
-; CHECK-NEXT:    movl %edi, %ebx
 ; CHECK-NEXT:    incl %ecx
 ; CHECK-NEXT:    addl {{[0-9]+}}(%esp), %eax
 ; CHECK-NEXT:    addl %edi, %edx
diff --git a/llvm/test/CodeGen/X86/tls-loads-control3.ll b/llvm/test/CodeGen/X86/tls-loads-control3.ll
@@ -183,7 +183,6 @@ define i32 @_Z2f2i(i32 %c) local_unnamed_addr #0 {
 ; HOIST0-NEXT:  # %bb.1: # %while.body.preheader
 ; HOIST0-NEXT:    leaq _ZZ2f2iE2st.0@TLSLD(%rip), %rdi
 ; HOIST0-NEXT:    callq __tls_get_addr@PLT
-; HOIST0-NEXT:    movq %rax, %rcx
 ; HOIST0-NEXT:    leaq _ZZ2f2iE2st.0@DTPOFF(%rax), %r15
 ; HOIST0-NEXT:    leaq _ZZ2f2iE2st.1@DTPOFF(%rax), %r12
 ; HOIST0-NEXT:    .p2align 4, 0x90
@@ -245,9 +244,7 @@ define i32 @_Z2f2i(i32 %c) local_unnamed_addr #0 {
 ; HOIST2-NEXT:    movq %rax, %r14
 ; HOIST2-NEXT:    addb %bpl, _ZZ2f2iE2st.0@DTPOFF(%rax)
 ; HOIST2-NEXT:    callq _Z5gfuncv@PLT
-; HOIST2-NEXT:    movl %eax, %ecx
-; HOIST2-NEXT:    movq %r14, %rax
-; HOIST2-NEXT:    addl %ecx, _ZZ2f2iE2st.1@DTPOFF(%r14)
+; HOIST2-NEXT:    addl %eax, _ZZ2f2iE2st.1@DTPOFF(%r14)
 ; HOIST2-NEXT:    decl %ebx
 ; HOIST2-NEXT:    jne .LBB1_2
 ; HOIST2-NEXT:  .LBB1_3: # %while.end
