[PATCH] D14178: Alternative to long nops for X86 CPUs

[Alexey Bataev via llvm-commits]

This revision was automatically updated to reflect the committed changes.
Closed by commit rL253557: Alternative to long nops for X86 CPUs, by Andrey Turetsky (authored by ABataev).

Changed prior to commit:
  http://reviews.llvm.org/D14178?vs=38742&id=40625#toc

Repository:
  rL LLVM

http://reviews.llvm.org/D14178

Files:
  llvm/trunk/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
  llvm/trunk/test/MC/X86/x86_nop.s

Index: llvm/trunk/test/MC/X86/x86_nop.s
===================================================================
--- llvm/trunk/test/MC/X86/x86_nop.s
+++ llvm/trunk/test/MC/X86/x86_nop.s
@@ -22,13 +22,7 @@
 inc %eax
 
 // CHECK: 0:	40                                           	incl	%eax
-// CHECK: 1:	90                                           	nop
-// CHECK: 2:	90                                           	nop
-// CHECK: 3:	90                                           	nop
-// CHECK: 4:	90                                           	nop
-// CHECK: 5:	90                                           	nop
-// CHECK: 6:	90                                           	nop
-// CHECK: 7:	90                                           	nop
+// CHECK: 1:	8d b4 26 00 00 00 00                            leal (%esi), %esi
 // CHECK: 8:	40                                           	incl	%eax
 
 
Index: llvm/trunk/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
===================================================================
--- llvm/trunk/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
+++ llvm/trunk/lib/Target/X86/MCTargetDesc/X86AsmBackend.cpp
@@ -69,15 +69,19 @@
 class X86AsmBackend : public MCAsmBackend {
   const StringRef CPU;
   bool HasNopl;
-  const uint64_t MaxNopLength;
+  uint64_t MaxNopLength;
 public:
-  X86AsmBackend(const Target &T, StringRef CPU)
-      : MCAsmBackend(), CPU(CPU), MaxNopLength(CPU == "slm" ? 7 : 15) {
+  X86AsmBackend(const Target &T, StringRef CPU) : MCAsmBackend(), CPU(CPU) {
     HasNopl = CPU != "generic" && CPU != "i386" && CPU != "i486" &&
               CPU != "i586" && CPU != "pentium" && CPU != "pentium-mmx" &&
               CPU != "i686" && CPU != "k6" && CPU != "k6-2" && CPU != "k6-3" &&
               CPU != "geode" && CPU != "winchip-c6" && CPU != "winchip2" &&
               CPU != "c3" && CPU != "c3-2";
+    // Max length of true long nop instruction is 15 bytes.
+    // Max length of long nop replacement instruction is 7 bytes.
+    // Taking into account SilverMont architecture features max length of nops
+    // is reduced for it to achieve better performance.
+    MaxNopLength = (!HasNopl || CPU == "slm") ? 7 : 15;
   }
 
   unsigned getNumFixupKinds() const override {
@@ -279,7 +283,7 @@
 /// bytes.
 /// \return - true on success, false on failure
 bool X86AsmBackend::writeNopData(uint64_t Count, MCObjectWriter *OW) const {
-  static const uint8_t Nops[10][10] = {
+  static const uint8_t TrueNops[10][10] = {
     // nop
     {0x90},
     // xchg %ax,%ax
@@ -302,17 +306,31 @@
     {0x66, 0x2e, 0x0f, 0x1f, 0x84, 0x00, 0x00, 0x00, 0x00, 0x00},
   };
 
-  // This CPU doesn't support long nops. If needed add more.
-  // FIXME: Can we get this from the subtarget somehow?
-  // FIXME: We could generated something better than plain 0x90.
-  if (!HasNopl) {
-    for (uint64_t i = 0; i < Count; ++i)
-      OW->write8(0x90);
-    return true;
-  }
+  // Alternative nop instructions for CPUs which don't support long nops.
+  static const uint8_t AltNops[7][10] = {
+      // nop
+      {0x90},
+      // xchg %ax,%ax
+      {0x66, 0x90},
+      // lea 0x0(%esi),%esi
+      {0x8d, 0x76, 0x00},
+      // lea 0x0(%esi),%esi
+      {0x8d, 0x74, 0x26, 0x00},
+      // nop + lea 0x0(%esi),%esi
+      {0x90, 0x8d, 0x74, 0x26, 0x00},
+      // lea 0x0(%esi),%esi
+      {0x8d, 0xb6, 0x00, 0x00, 0x00, 0x00 },
+      // lea 0x0(%esi),%esi
+      {0x8d, 0xb4, 0x26, 0x00, 0x00, 0x00, 0x00},
+  };
+
+  // Select the right NOP table.
+  // FIXME: Can we get if CPU supports long nops from the subtarget somehow?
+  const uint8_t (*Nops)[10] = HasNopl ? TrueNops : AltNops;
+  assert(HasNopl || MaxNopLength <= 7);
 
-  // 15 is the longest single nop instruction.  Emit as many 15-byte nops as
-  // needed, then emit a nop of the remaining length.
+  // Emit as many largest nops as needed, then emit a nop of the remaining
+  // length.
   do {
     const uint8_t ThisNopLength = (uint8_t) std::min(Count, MaxNopLength);
     const uint8_t Prefixes = ThisNopLength <= 10 ? 0 : ThisNopLength - 10;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D14178.40625.patch
Type: text/x-patch
Size: 4064 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151119/3adc0282/attachment.bin>