[MachineSink] Use the real post dominator tree

Summary:
Fixes a FIXME in MachineSinking. Instead of using the simple heuristics in
isPostDominatedBy, use the real MachinePostDominatorTree and MachineLoopInfo.
The old heuristics caused instructions to sink unnecessarily, and might create
register pressure.

This is the second try of the fix. The first one (D4814) caused a performance
regression due to failing to sink instructions out of loops (PR21115). This
patch fixes PR21115 by sinking an instruction from a deeper loop to a shallower
one regardless of whether the target block post-dominates the source.

Thanks Alexey Volkov for reporting PR21115! 

Test Plan:
Added a NVPTX codegen test to verify that our change prevents the backend from
over-sinking. It also shows the unnecessary register pressure caused by
over-sinking.

Added an X86 test to verify we can sink instructions out of loops regardless of
the dominance relationship. This test is reduced from Alexey's test in PR21115.

Updated an affected test in X86.

Also ran SPEC CINT2006 and llvm-test-suite for compilation time and runtime
performance. Results are attached separately in the review thread.

Reviewers: Jiangning, resistor, hfinkel

Reviewed By: hfinkel

Subscribers: hfinkel, bruno, volkalexey, llvm-commits, meheff, eliben, jholewinski

Differential Revision: http://reviews.llvm.org/D5633

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219773 91177308-0d34-0410-b5e6-96231b3b80d8

lib/CodeGen/MachineSink.cpp

@@ -24,6 +24,7 @@
 #include "llvm/CodeGen/MachineBlockFrequencyInfo.h"
 #include "llvm/CodeGen/MachineDominators.h"
 #include "llvm/CodeGen/MachineLoopInfo.h"
+#include "llvm/CodeGen/MachinePostDominators.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
@@ -54,8 +55,9 @@ namespace {
   class MachineSinking : public MachineFunctionPass {
     const TargetInstrInfo *TII;
     const TargetRegisterInfo *TRI;
-    MachineRegisterInfo  *MRI;  // Machine register information
-    MachineDominatorTree *DT;   // Machine dominator tree
+    MachineRegisterInfo  *MRI;     // Machine register information
+    MachineDominatorTree *DT;      // Machine dominator tree
+    MachinePostDominatorTree *PDT; // Machine post dominator tree
     MachineLoopInfo *LI;
     const MachineBlockFrequencyInfo *MBFI;
     AliasAnalysis *AA;
@@ -81,8 +83,10 @@ namespace {
       MachineFunctionPass::getAnalysisUsage(AU);
       AU.addRequired<AliasAnalysis>();
       AU.addRequired<MachineDominatorTree>();
+      AU.addRequired<MachinePostDominatorTree>();
       AU.addRequired<MachineLoopInfo>();
       AU.addPreserved<MachineDominatorTree>();
+      AU.addPreserved<MachinePostDominatorTree>();
       AU.addPreserved<MachineLoopInfo>();
       if (UseBlockFreqInfo)
         AU.addRequired<MachineBlockFrequencyInfo>();
@@ -249,6 +253,7 @@ bool MachineSinking::runOnMachineFunction(MachineFunction &MF) {
   TRI = MF.getSubtarget().getRegisterInfo();
   MRI = &MF.getRegInfo();
   DT = &getAnalysis<MachineDominatorTree>();
+  PDT = &getAnalysis<MachinePostDominatorTree>();
   LI = &getAnalysis<MachineLoopInfo>();
   MBFI = UseBlockFreqInfo ? &getAnalysis<MachineBlockFrequencyInfo>() : nullptr;
   AA = &getAnalysis<AliasAnalysis>();
@@ -467,23 +472,6 @@ static void collectDebugValues(MachineInstr *MI,
   }
 }
 
-/// isPostDominatedBy - Return true if A is post dominated by B.
-static bool isPostDominatedBy(MachineBasicBlock *A, MachineBasicBlock *B) {
-
-  // FIXME - Use real post dominator.
-  if (A->succ_size() != 2)
-    return false;
-  MachineBasicBlock::succ_iterator I = A->succ_begin();
-  if (B == *I)
-    ++I;
-  MachineBasicBlock *OtherSuccBlock = *I;
-  if (OtherSuccBlock->succ_size() != 1 ||
-      *(OtherSuccBlock->succ_begin()) != B)
-    return false;
-
-  return true;
-}
-
 /// isProfitableToSinkTo - Return true if it is profitable to sink MI.
 bool MachineSinking::isProfitableToSinkTo(unsigned Reg, MachineInstr *MI,
                                           MachineBasicBlock *MBB,
@@ -495,8 +483,13 @@ bool MachineSinking::isProfitableToSinkTo(unsigned Reg, MachineInstr *MI,
     return false;
 
   // It is profitable if SuccToSinkTo does not post dominate current block.
-  if (!isPostDominatedBy(MBB, SuccToSinkTo))
-      return true;
+  if (!PDT->dominates(SuccToSinkTo, MBB))
+    return true;
+
+  // It is profitable to sink an instruction from a deeper loop to a shallower
+  // loop, even if the latter post-dominates the former (PR21115).
+  if (LI->getLoopDepth(MBB) > LI->getLoopDepth(SuccToSinkTo))
+    return true;
 
   // Check if only use in post dominated block is PHI instruction.
   bool NonPHIUse = false;

test/CodeGen/NVPTX/machine-sink.ll

@@ -0,0 +1,40 @@
+; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64"
+
+@scalar1 = internal addrspace(3) global float 0.000000e+00, align 4
+@scalar2 = internal addrspace(3) global float 0.000000e+00, align 4
+
+; We shouldn't sink mul.rn.f32 to BB %merge because BB %merge post-dominates
+; BB %entry. Over-sinking created more register pressure on this example. The
+; backend would sink the fmuls to BB %merge, but not the loads for being
+; conservative on sinking memory accesses. As a result, the loads and
+; the two fmuls would be separated to two basic blocks, causing two
+; cross-BB live ranges.
+define float @post_dominate(float %x, i1 %cond) {
+; CHECK-LABEL: post_dominate(
+entry:
+  %0 = load float* addrspacecast (float addrspace(3)* @scalar1 to float*), align 4
+  %1 = load float* addrspacecast (float addrspace(3)* @scalar2 to float*), align 4
+; CHECK: ld.shared.f32
+; CHECK: ld.shared.f32
+  %2 = fmul float %0, %0
+  %3 = fmul float %1, %2
+; CHECK-NOT: bra
+; CHECK: mul.rn.f32
+; CHECK: mul.rn.f32
+  br i1 %cond, label %then, label %merge
+
+then:
+  %z = fadd float %x, %x
+  br label %then2
+
+then2:
+  %z2 = fadd float %z, %z
+  br label %merge
+
+merge:
+  %y = phi float [ 0.0, %entry ], [ %z2, %then2 ]
+  %w = fadd float %y, %3
+  ret float %w
+}

test/CodeGen/X86/loop-strength-reduce8.ll

@@ -1,6 +1,9 @@
 ; RUN: llc < %s -mtriple=i386-apple-darwin | FileCheck %s
 
-; CHECK: leal 16(%eax), %edx
+; FIXME: The first two instructions, movl and addl, should have been combined to
+; "leal 16(%eax), %edx" by the backend (PR20776).
+; CHECK: movl    %eax, %edx
+; CHECK: addl    $16, %edx
 ; CHECK: align
 ; CHECK: addl    $4, %edx
 ; CHECK: decl    %ecx

test/CodeGen/X86/sink-out-of-loop.ll

@@ -5,7 +5,7 @@
 ; MOV32ri outside the loop.
 ; rdar://11980766
 define i32 @sink_succ(i32 %argc, i8** nocapture %argv) nounwind uwtable ssp {
-; CHECK: sink_succ
+; CHECK-LABEL: sink_succ
 ; CHECK: [[OUTER_LN1:LBB0_[0-9]+]]: ## %preheader
 ; CHECK: %exit
 ; CHECK-NOT: movl
@@ -52,3 +52,24 @@ for.body2:
 for.end20:
   ret i32 0
 }
+
+define i32 @sink_out_of_loop(i32 %n, i32* %output) {
+; CHECK-LABEL: sink_out_of_loop:
+entry:
+  br label %loop
+
+loop:
+  %i = phi i32 [ 0, %entry ], [ %i2, %loop ]
+  %j = mul i32 %i, %i
+  %addr = getelementptr i32* %output, i32 %i
+  store i32 %i, i32* %addr
+  %i2 = add i32 %i, 1
+  %exit_cond = icmp sge i32 %i2, %n
+  br i1 %exit_cond, label %exit, label %loop
+
+exit:
+; CHECK: BB#2
+; CHECK: imull %eax, %eax
+; CHECK: retq
+  ret i32 %j
+}