[Sparc] Do not emit nop after fcmp* instruction with V9.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192056 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/Sparc/DelaySlotFiller.cpp

@@ -14,6 +14,7 @@
 
 #define DEBUG_TYPE "delay-slot-filler"
 #include "Sparc.h"
+#include "SparcSubtarget.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
@@ -39,10 +40,13 @@ namespace {
     /// layout, etc.
     ///
     TargetMachine &TM;
+    const SparcSubtarget *Subtarget;
 
     static char ID;
     Filler(TargetMachine &tm)
-      : MachineFunctionPass(ID), TM(tm) { }
+      : MachineFunctionPass(ID), TM(tm),
+        Subtarget(&TM.getSubtarget<SparcSubtarget>()) {
+    }
 
     virtual const char *getPassName() const {
       return "SPARC Delay Slot Filler";
@@ -102,6 +106,8 @@ FunctionPass *llvm::createSparcDelaySlotFillerPass(TargetMachine &tm) {
 bool Filler::runOnMachineBasicBlock(MachineBasicBlock &MBB) {
   bool Changed = false;
 
+  const TargetInstrInfo *TII = TM.getInstrInfo();
+
   for (MachineBasicBlock::iterator I = MBB.begin(); I != MBB.end(); ) {
     MachineBasicBlock::iterator MI = I;
     ++I;
@@ -114,6 +120,14 @@ bool Filler::runOnMachineBasicBlock(MachineBasicBlock &MBB) {
       continue;
     }
 
+    if (!Subtarget->isV9() &&
+        (MI->getOpcode() == SP::FCMPS || MI->getOpcode() == SP::FCMPD
+         || MI->getOpcode() == SP::FCMPQ)) {
+      BuildMI(MBB, I, MI->getDebugLoc(), TII->get(SP::NOP));
+      Changed = true;
+      continue;
+    }
+
     // If MI has no delay slot, skip.
     if (!MI->hasDelaySlot())
       continue;
@@ -126,7 +140,6 @@ bool Filler::runOnMachineBasicBlock(MachineBasicBlock &MBB) {
     ++FilledSlots;
     Changed = true;
 
-    const TargetInstrInfo *TII = TM.getInstrInfo();
     if (D == MBB.end())
       BuildMI(MBB, I, MI->getDebugLoc(), TII->get(SP::NOP));
     else

lib/Target/Sparc/SparcInstrInfo.td

@@ -807,20 +807,22 @@ def FDIVQ  : F3_3<2, 0b110100, 0b001001111,
 // Floating-point Compare Instructions, p. 148
 // Note: the 2nd template arg is different for these guys.
 // Note 2: the result of a FCMP is not available until the 2nd cycle
-// after the instr is retired, but there is no interlock. This behavior
+// after the instr is retired, but there is no interlock in Sparc V8.
-// is modelled with a forced noop after the instruction.
+// This behavior is modeled with a forced noop after the instruction in
+// DelaySlotFiller.
+
 let Defs = [FCC] in {
   def FCMPS  : F3_3c<2, 0b110101, 0b001010001,
                    (outs), (ins FPRegs:$src1, FPRegs:$src2),
-                   "fcmps $src1, $src2\n\tnop",
+                   "fcmps $src1, $src2",
                    [(SPcmpfcc f32:$src1, f32:$src2)]>;
   def FCMPD  : F3_3c<2, 0b110101, 0b001010010,
                    (outs), (ins DFPRegs:$src1, DFPRegs:$src2),
-                   "fcmpd $src1, $src2\n\tnop",
+                   "fcmpd $src1, $src2",
                    [(SPcmpfcc f64:$src1, f64:$src2)]>;
   def FCMPQ  : F3_3c<2, 0b110101, 0b001010011,
                    (outs), (ins QFPRegs:$src1, QFPRegs:$src2),
-                   "fcmpq $src1, $src2\n\tnop",
+                   "fcmpq $src1, $src2",
                    [(SPcmpfcc f128:$src1, f128:$src2)]>,
                    Requires<[HasHardQuad]>;
 }

test/CodeGen/SPARC/2011-01-11-CC.ll

@@ -66,9 +66,11 @@ define i32 @test_select_int_fcc(float %f, i32 %a, i32 %b) nounwind readnone noin
 entry:
 ;V8-LABEL: test_select_int_fcc:
 ;V8: fcmps
+;V8-NEXT: nop
 ;V8: {{fbe|fbne}}
 ;V9-LABEL: test_select_int_fcc:
 ;V9: fcmps
+;V9-NEXT-NOT: nop
 ;V9-NOT: {{fbe|fbne}}
 ;V9: mov{{e|ne}} %fcc0
   %0 = fcmp une float %f, 0.000000e+00
@@ -95,9 +97,11 @@ define double @test_select_dfp_fcc(double %f, double %f1, double %f2) nounwind r
 entry:
 ;V8-LABEL: test_select_dfp_fcc:
 ;V8: fcmpd
+;V8-NEXT: nop
 ;V8: {{fbne|fbe}}
 ;V9-LABEL: test_select_dfp_fcc:
 ;V9: fcmpd
+;V9-NEXT-NOT: nop
 ;V9-NOT: {{fbne|fbe}}
 ;V9: fmovd{{e|ne}} %fcc0
   %0 = fcmp une double %f, 0.000000e+00

test/CodeGen/SPARC/fp128.ll

@@ -64,6 +64,7 @@ entry:
 
 ; HARD-LABEL: f128_compare
 ; HARD:       fcmpq
+; HARD-NEXT:  nop
 
 ; SOFT-LABEL: f128_compare
 ; SOFT:       _Q_cmp