[PATCH] D30242: [ExecutionDepsFix] Don't make copies of LiveReg objects when collecting operands for soft instructions

[Phabricator via Phabricator via llvm-commits]

This revision was automatically updated to reflect the committed changes.
Closed by commit rL296260: [ExecutionDepsFix] Don't make copies of LiveReg objects when collecting… (authored by ctopper).

Changed prior to commit:
  https://reviews.llvm.org/D30242?vs=89611&id=89795#toc

Repository:
  rL LLVM

https://reviews.llvm.org/D30242

Files:
  llvm/trunk/lib/CodeGen/ExecutionDepsFix.cpp
  llvm/trunk/test/CodeGen/X86/pr30284.ll


Index: llvm/trunk/test/CodeGen/X86/pr30284.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/pr30284.ll
+++ llvm/trunk/test/CodeGen/X86/pr30284.ll
@@ -0,0 +1,21 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i386-unknown-linux-gnu -mattr=avx512dq | FileCheck %s
+
+define void @f_f___un_3C_unf_3E_un_3C_unf_3E_() {
+; CHECK-LABEL: f_f___un_3C_unf_3E_un_3C_unf_3E_:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vmovapd 0, %zmm0
+; CHECK-NEXT:    vmovapd 64, %zmm1
+; CHECK-NEXT:    vmovapd {{.*#+}} zmm2 = [0,16,0,16,0,16,0,16,0,16,0,16,0,16,0,16]
+; CHECK-NEXT:    vorpd %zmm2, %zmm0, %zmm0 {%k1}
+; CHECK-NEXT:    vorpd %zmm2, %zmm1, %zmm1 {%k1}
+; CHECK-NEXT:    vmovapd %zmm1, 64
+; CHECK-NEXT:    vmovapd %zmm0, 0
+; CHECK-NEXT:    retl
+  %a_load22 = load <16 x i64>, <16 x i64>* null, align 1
+  %bitop = or <16 x i64> %a_load22, <i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736, i64 68719476736>
+  %v.i = load <16 x i64>, <16 x i64>* null
+  %v1.i41 = select <16 x i1> undef, <16 x i64> %bitop, <16 x i64> %v.i
+  store <16 x i64> %v1.i41, <16 x i64>* null
+  ret void
+}
Index: llvm/trunk/lib/CodeGen/ExecutionDepsFix.cpp
===================================================================
--- llvm/trunk/lib/CodeGen/ExecutionDepsFix.cpp
+++ llvm/trunk/lib/CodeGen/ExecutionDepsFix.cpp
@@ -721,7 +721,7 @@
 
   // Kill off any remaining uses that don't match available, and build a list of
   // incoming DomainValues that we want to merge.
-  SmallVector<LiveReg, 4> Regs;
+  SmallVector<const LiveReg *, 4> Regs;
   for (int rx : used) {
     assert(LiveRegs && "no space allocated for live registers");
     const LiveReg &LR = LiveRegs[rx];
@@ -731,31 +731,26 @@
       continue;
     }
     // Sorted insertion.
-    bool Inserted = false;
-    for (SmallVectorImpl<LiveReg>::iterator i = Regs.begin(), e = Regs.end();
-           i != e && !Inserted; ++i) {
-      if (LR.Def < i->Def) {
-        Inserted = true;
-        Regs.insert(i, LR);
-      }
-    }
-    if (!Inserted)
-      Regs.push_back(LR);
+    auto I = std::upper_bound(Regs.begin(), Regs.end(), &LR,
+                              [](const LiveReg *LHS, const LiveReg *RHS) {
+                                return LHS->Def < RHS->Def;
+                              });
+    Regs.insert(I, &LR);
   }
 
   // doms are now sorted in order of appearance. Try to merge them all, giving
   // priority to the latest ones.
   DomainValue *dv = nullptr;
   while (!Regs.empty()) {
     if (!dv) {
-      dv = Regs.pop_back_val().Value;
+      dv = Regs.pop_back_val()->Value;
       // Force the first dv to match the current instruction.
       dv->AvailableDomains = dv->getCommonDomains(available);
       assert(dv->AvailableDomains && "Domain should have been filtered");
       continue;
     }
 
-    DomainValue *Latest = Regs.pop_back_val().Value;
+    DomainValue *Latest = Regs.pop_back_val()->Value;
     // Skip already merged values.
     if (Latest == dv || Latest->Next)
       continue;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D30242.89795.patch
Type: text/x-patch
Size: 3300 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170225/c5b5fe13/attachment.bin>