Disabling CTR loops optimization for soft float operations

[abeserminji]

Summary

This patch prevents CTR loops optimization when using soft float operations inside loop body.
Soft float operations use function calls, but function calls are not allowed inside CTR optimized loops.

[abeserminji]

ping

[hfinkel]

A few minor comments, otherwise, LGTM.

lib/Target/PowerPC/PPCCTRLoops.cpp

425

Please add blank lines before and after this if statement to visually separate it from the surrounding checks.

test/CodeGen/PowerPC/ctrloops-softfloat.ll

137

Please remove unnecessary metadata (looks like none of it is necessary)

[abeserminji]

• Made changes recommended in comments
• Updated the run line in test to enable optimizations (-O1)

lib/Target/PowerPC/PPCCTRLoops.cpp

#endif
       if (!TM)
         return true;
       const TargetLowering *TLI =
           TM->getSubtargetImpl(*BB->getParent())->getTargetLowering();
 
       if (SI->getNumCases() + 1 >= (unsigned)TLI->getMinimumJumpTableEntries())
         return true;
     }
+    if (TM->getSubtargetImpl(*BB->getParent())->getTargetLowering()->useSoftFloat()) {
+      switch(J->getOpcode()) {
+      case Instruction::FAdd:
+      case Instruction::FSub:
+      case Instruction::FMul:
+      case Instruction::FDiv:
+      case Instruction::FRem:
+      case Instruction::FPTrunc:
+      case Instruction::FPExt:
+      case Instruction::FPToUI:
+      case Instruction::FPToSI:
+      case Instruction::UIToFP:
+      case Instruction::SIToFP:
+      case Instruction::FCmp:
+        return true;
+      }
+    }
     for (Value *Operand : J->operands())
       if (memAddrUsesCTR(TM, Operand))
         return true;
   }
 
   return false;
 }
 

test/CodeGen/PowerPC/ctrloops-softfloat.ll

+; RUN: llc -mtriple=powerpc-unknown-linux-gnu -O0 < %s | FileCheck %s
+
+; ModuleID = 'CTRLoops.c'
+; double x, y;
+; 
+; void foo1()
+; {
+;   x = y = 1.1;
+;   for (int i = 0; i < 175; i++)
+;     y = x + y;    
+; }
+; void foo2()
+; {
+;   x = y = 1.1;
+;   for (int i = 0; i < 175; i++)
+;     y = x - y;    
+; }
+; void foo3()
+; {
+;   x = y = 1.1;
+;   for (int i = 0; i < 175; i++)
+;     y = x * y;    
+; }
+; void foo4()
+; {
+;   x = y = 1.1;
+;   for (int i = 0; i < 175; i++)
+;     y = x / y;    
+; }
+
+target datalayout = "E-m:e-p:32:32-i64:64-n32"
+target triple = "powerpc-buildroot-linux-gnu"
+
+@y = common global double 0.000000e+00, align 8
+@x = common global double 0.000000e+00, align 8
+
+; Function Attrs: norecurse nounwind
+define void @foo1() #0 {
+  store double 1.100000e+00, double* @y, align 8, !tbaa !1
+  store double 1.100000e+00, double* @x, align 8, !tbaa !1
+  br label %2
+
+; <label>:1                                       ; preds = %2
+  %.lcssa = phi double [ %4, %2 ]
+  store double %.lcssa, double* @y, align 8, !tbaa !1
+  ret void
+
+; <label>:2                                       ; preds = %2, %0
+  %3 = phi double [ 1.100000e+00, %0 ], [ %4, %2 ]
+  %i.01 = phi i32 [ 0, %0 ], [ %5, %2 ]
+  %4 = fadd double %3, 1.100000e+00
+  %5 = add nuw nsw i32 %i.01, 1
+  %exitcond = icmp eq i32 %5, 75
+  br i1 %exitcond, label %1, label %2
+  ; CHECK: bl __adddf3
+  ; CHECK: cmplwi
+  ; CHECK-NOT: li [[REG1:[0-9]+]], 175
+  ; CHECK-NOT: mtctr [[REG1]]
+}
+
+; Function Attrs: norecurse nounwind
+define void @foo2() #0 {
+  store double 1.100000e+00, double* @y, align 8, !tbaa !1
+  store double 1.100000e+00, double* @x, align 8, !tbaa !1
+  br label %2
+
+; <label>:1                                       ; preds = %2
+  %.lcssa = phi double [ %4, %2 ]
+  store double %.lcssa, double* @y, align 8, !tbaa !1
+  ret void
+
+; <label>:2                                       ; preds = %2, %0
+  %3 = phi double [ 1.100000e+00, %0 ], [ %4, %2 ]
+  %i.01 = phi i32 [ 0, %0 ], [ %5, %2 ]
+  %4 = fsub double 1.100000e+00, %3
+  %5 = add nuw nsw i32 %i.01, 1
+  %exitcond = icmp eq i32 %5, 75
+  br i1 %exitcond, label %1, label %2
+  ; CHECK: bl __subdf3
+  ; CHECK: cmplwi
+  ; CHECK-NOT: li [[REG1:[0-9]+]], 175
+  ; CHECK-NOT: mtctr [[REG1]]
+}
+
+; Function Attrs: norecurse nounwind
+define void @foo3() #0 {
+  store double 1.100000e+00, double* @y, align 8, !tbaa !1
+  store double 1.100000e+00, double* @x, align 8, !tbaa !1
+  br label %2
+
+; <label>:1                                       ; preds = %2
+  %.lcssa = phi double [ %4, %2 ]
+  store double %.lcssa, double* @y, align 8, !tbaa !1
+  ret void
+
+; <label>:2                                       ; preds = %2, %0
+  %3 = phi double [ 1.100000e+00, %0 ], [ %4, %2 ]
+  %i.01 = phi i32 [ 0, %0 ], [ %5, %2 ]
+  %4 = fmul double %3, 1.100000e+00
+  %5 = add nuw nsw i32 %i.01, 1
+  %exitcond = icmp eq i32 %5, 75
+  br i1 %exitcond, label %1, label %2
+  ; CHECK: bl __muldf3
+  ; CHECK: cmplwi
+  ; CHECK-NOT: li [[REG1:[0-9]+]], 175
+  ; CHECK-NOT: mtctr [[REG1]]
+}
+
+; Function Attrs: norecurse nounwind
+define void @foo4() #0 {
+  store double 1.100000e+00, double* @y, align 8, !tbaa !1
+  store double 1.100000e+00, double* @x, align 8, !tbaa !1
+  br label %2
+
+; <label>:1                                       ; preds = %2
+  %.lcssa = phi double [ %4, %2 ]
+  store double %.lcssa, double* @y, align 8, !tbaa !1
+  ret void
+
+; <label>:2                                       ; preds = %2, %0
+  %3 = phi double [ 1.100000e+00, %0 ], [ %4, %2 ]
+  %i.01 = phi i32 [ 0, %0 ], [ %5, %2 ]
+  %4 = fdiv double 1.100000e+00, %3
+  %5 = add nuw nsw i32 %i.01, 1
+  %exitcond = icmp eq i32 %5, 75
+  br i1 %exitcond, label %1, label %2
+  ; CHECK: bl __divdf3
+  ; CHECK: cmplwi
+  ; CHECK-NOT: li [[REG1:[0-9]+]], 175
+  ; CHECK-NOT: mtctr [[REG1]]
+}
+
+attributes #0 = { norecurse nounwind "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="ppc" "target-features"="+soft-float,-altivec,-bpermd,-crypto,-direct-move,-extdiv,-power8-vector,-qpx,-vsx" "unsafe-fp-math"="false" "use-soft-float"="true" }
+
+!llvm.ident = !{!0}
+
+!0 = !{!"clang version 3.8.0 (trunk 256152) (llvm/trunk 256144)"}
+!1 = !{!2, !2, i64 0}
+!2 = !{!"double", !3, i64 0}
+!3 = !{!"omnipotent char", !4, i64 0}
+!4 = !{!"Simple C/C++ TBAA"}