[PATCH] Value soft float calls as more expensive in the inliner.

[Tim Northover]

Hi Cameron,

We could probably do with some tests for this, and I think the ARM logic needs a bit of work.


================
Comment at: lib/Target/ARM/ARMSubtarget.h:316-318
@@ -314,1 +315,5 @@
+
+  bool hasFloatingPoint() const {
+    return HasVFPv2 || HasVFPv3 || HasVFPv4 || HasFPARMv8;
+  }
 
----------------
I don't think this is quite right. The relevant code in ARMISelLowering is:

    if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
        !Subtarget->isThumb1Only())

(notably: any CPU with float has VFP2; SoftFloat can be specified even if the hardware supports it; and Thumb1 didn't have access to VFP).

================
Comment at: lib/Target/ARM/ARMTargetTransformInfo.cpp:455
@@ +454,3 @@
+unsigned ARMTTI::getFPOpCost(Type *Ty) const {
+  if (ST->hasFloatingPoint()) {
+    if (Ty->isFloatTy()) {
----------------
I don't think this is quite right. The relevant code in ARMISelLowering is:

if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
    !Subtarget->isThumb1Only())
(notably: any CPU with float has VFP2; SoftFloat can be specified even if the hardware supports it; and Thumb1 didn't have access to VFP).

http://reviews.llvm.org/D6936

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/