[PATCH] D138272: [MachineTraceMetrics] Pick the trace successor for an entry block

[Anton Sidorenko via Phabricator via llvm-commits]

asi-sc created this revision.
Herald added subscribers: pengfei, hiraditya.
Herald added a project: All.
asi-sc requested review of this revision.
Herald added a project: LLVM.
Herald added a subscriber: llvm-commits.

We generate erroneous trace for a basic block if it does not have at least one
predecessor when MinInstr strategy is used. Currently only this strategy is
implemented, so we always have a wrong trace for any entry block. This results in
wrong instructions heights calculation and also leads to wrong critical path.

The described behavior is demonstrated on a simple test. It shows that early
if-conv pass makes wrong decisions due to incorrectly calculated critical path
lenght.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D138272

Files:
  llvm/lib/CodeGen/MachineTraceMetrics.cpp
  llvm/test/CodeGen/X86/machine-trace-metrics-entryBB-critpath.ll


Index: llvm/test/CodeGen/X86/machine-trace-metrics-entryBB-critpath.ll
===================================================================
--- llvm/test/CodeGen/X86/machine-trace-metrics-entryBB-critpath.ll
+++ llvm/test/CodeGen/X86/machine-trace-metrics-entryBB-critpath.ll
@@ -6,27 +6,27 @@
 ; MinInstr strategy is used. The behavior is demonstrated on early if conversion
 ; pass.
 
-; CHECK: TBB: MinInstr trace %bb.0 --> %bb.0 --> %bb.0: 4 instrs. 1 cycles.
+; CHECK: TBB: MinInstr trace %bb.0 --> %bb.0 --> %bb.2: 8 instrs. 30 cycles.
 ; CHECK: %bb.0
+; CHECK:     -> %bb.2
 
 ; CHECK: FBB: MinInstr trace %bb.0 --> %bb.1 --> %bb.2: 10 instrs. 32 cycles.
 ; CHECK: %bb.1 <- %bb.0
 ; CHECK:     -> %bb.2
 
-; CHECK: Resource length 10, minimal critical path 1
-; CHECK: Not enough available ILP.
+; CHECK: Resource length 10, minimal critical path 30
+; CHECK: If-converting
 
 define i32 @_Z3fooiidd(i32 %a, i32 %b, double %d, double %e) #0 {
 ; CHECK-LABEL: _Z3fooiidd:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    movl %esi, %eax
-; CHECK-NEXT:    addl %edi, %eax
-; CHECK-NEXT:    cmpl $3, %edi
-; CHECK-NEXT:    jl .LBB0_2
-; CHECK-NEXT:  # %bb.1: # %if.then
-; CHECK-NEXT:    cvttsd2si %xmm0, %ecx
+; CHECK-NEXT:    # kill: def $esi killed $esi def $rsi
+; CHECK-NEXT:    # kill: def $edi killed $edi def $rdi
+; CHECK-NEXT:    leal (%rsi,%rdi), %ecx
+; CHECK-NEXT:    cvttsd2si %xmm0, %eax
 ; CHECK-NEXT:    addl %ecx, %eax
-; CHECK-NEXT:  .LBB0_2: # %if.end
+; CHECK-NEXT:    cmpl $3, %edi
+; CHECK-NEXT:    cmovll %ecx, %eax
 ; CHECK-NEXT:    cvttsd2si %xmm1, %ecx
 ; CHECK-NEXT:    cltd
 ; CHECK-NEXT:    idivl %ecx
Index: llvm/lib/CodeGen/MachineTraceMetrics.cpp
===================================================================
--- llvm/lib/CodeGen/MachineTraceMetrics.cpp
+++ llvm/lib/CodeGen/MachineTraceMetrics.cpp
@@ -352,7 +352,7 @@
 // Select the preferred successor for MBB.
 const MachineBasicBlock*
 MinInstrCountEnsemble::pickTraceSucc(const MachineBasicBlock *MBB) {
-  if (MBB->pred_empty())
+  if (MBB->succ_empty())
     return nullptr;
   const MachineLoop *CurLoop = getLoopFor(MBB);
   const MachineBasicBlock *Best = nullptr;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D138272.476385.patch
Type: text/x-patch
Size: 2170 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20221118/05813c38/attachment.bin>