[InlineCost] Add flag to allow changing the default inline cost

[Holman]

Summary

It can be useful to tune the default inline threshold without overriding other inlining thresholds (e.g. in code compiled for size).

The existing -inline-threshold flag overrides other thresholds, so it is insufficient in codebases where there is a mix of code compiled for size and speed.

[tejohnson]

@davidxl and @mtrofin for thoughts.

Seems reasonable to decouple the setting of the default from the overriding of the opt size thresholds. However, with your change the only use of InlineThreshold is essentially as a bool flag (was it specified or not). How about leave the InlineThreshold option alone except add a new bool flag to control whether it applies to the opt size thresholds?

Also, needs a test.

[davidxl]

This change won't work. See

if (InlineThreshold.getNumOccurrences() == 0) {
  Params.OptMinSizeThreshold = InlineConstants::OptMinSizeThreshold;
  Params.OptSizeThreshold = InlineConstants::OptSizeThreshold;
  Params.ColdThreshold = ColdThreshold;
} else if (ColdThreshold.getNumOccurrences() > 0) {
  Params.ColdThreshold = ColdThreshold;
}

[davidxl]

ok. I see the intention. As long as the behavior of --inline-threshold option is not changed (it still overrides the new option), the new option seems fine to me. Adding individually controlled option for size opt seems a good idea too.

[davidxl]

there are checks of explicit occurrences of inline-threshold option later, so the behavior seems unchanged.

[Holman]

Added test

[tejohnson]

lgtm

[Holman]

Thanks! I don't have commit access, so can someone help me with that?

llvm/lib/Analysis/InlineCost.cpp

 #include "llvm/Support/raw_ostream.h"
 
 using namespace llvm;
 
 #define DEBUG_TYPE "inline-cost"
 
 STATISTIC(NumCallsAnalyzed, "Number of call sites analyzed");
 
+static cl::opt<int>
+    DefaultThreshold("inlinedefault-threshold", cl::Hidden, cl::init(225),
+                     cl::ZeroOrMore,
+                     cl::desc("Default amount of inlining to perform"));
+
 static cl::opt<int> InlineThreshold(
     "inline-threshold", cl::Hidden, cl::init(225), cl::ZeroOrMore,
     cl::desc("Control the amount of inlining to perform (default = 225)"));
 
 static cl::opt<int> HintThreshold(
     "inlinehint-threshold", cl::Hidden, cl::init(325), cl::ZeroOrMore,
     cl::desc("Threshold for inlining functions with inline hint"));
 
if (InlineThreshold.getNumOccurrences() == 0) {
     Params.ColdThreshold = ColdThreshold;
   } else if (ColdThreshold.getNumOccurrences() > 0) {
     Params.ColdThreshold = ColdThreshold;
   }
   return Params;
 }
 
 InlineParams llvm::getInlineParams() {
-  return getInlineParams(InlineThreshold);
+  return getInlineParams(DefaultThreshold);
 }
 
 // Compute the default threshold for inlining based on the opt level and the
 // size opt level.
 static int computeThresholdFromOptLevels(unsigned OptLevel,
                                          unsigned SizeOptLevel) {
   if (OptLevel > 2)
     return InlineConstants::OptAggressiveThreshold;
   if (SizeOptLevel == 1) // -Os
     return InlineConstants::OptSizeThreshold;
   if (SizeOptLevel == 2) // -Oz
     return InlineConstants::OptMinSizeThreshold;
-  return InlineThreshold;
+  return DefaultThreshold;
 }
 
 InlineParams llvm::getInlineParams(unsigned OptLevel, unsigned SizeOptLevel) {
   auto Params =
       getInlineParams(computeThresholdFromOptLevels(OptLevel, SizeOptLevel));
   // At O3, use the value of -locally-hot-callsite-threshold option to populate
   // Params.LocallyHotCallSiteThreshold. Below O3, this flag has effect only
   // when it is specified explicitly.
   if (OptLevel > 2)
     Params.LocallyHotCallSiteThreshold = LocallyHotCallSiteThreshold;
   return Params;
 }

llvm/test/Transforms/Inline/inlinedefault-threshold.ll

+; RUN: opt < %s -O2 -inlinedefault-threshold=100 -S | FileCheck %s
+
+; Check that the inlinedefault-threshold does not alter the inline threshold
+; for optsize or minsize functions
+
+@a = global i32 4
+
+define i32 @inner() {
+  call void @extern()
+  %a1 = load volatile i32, i32* @a
+  %x1 = add i32 %a1,  %a1
+  ret i32 %x1
+}
+
+define i32 @inner2() {
+  call void @extern()
+  %a1 = load volatile i32, i32* @a
+  %x1 = add i32 %a1,  %a1
+  %a2 = load volatile i32, i32* @a
+  %x2 = add i32 %x1, %a2
+  %a3 = load volatile i32, i32* @a
+  %x3 = add i32 %x2, %a3
+  %a4 = load volatile i32, i32* @a
+  %x4 = add i32 %x3, %a4
+  %a5 = load volatile i32, i32* @a
+  %x5 = add i32 %x3, %a5
+  %a6 = load volatile i32, i32* @a
+  %x6 = add i32 %x5, %a6
+  %a7 = load volatile i32, i32* @a
+  %x7 = add i32 %x6, %a7
+  %a8 = load volatile i32, i32* @a
+  %x8 = add i32 %x7, %a8
+  ret i32 %x8
+}
+
+define i32 @inner3() {
+  call void @extern()
+  %a1 = load volatile i32, i32* @a
+  %x1 = add i32 %a1,  %a1
+  %a2 = load volatile i32, i32* @a
+  %x2 = add i32 %x1, %a2
+  %a3 = load volatile i32, i32* @a
+  %x3 = add i32 %x2, %a3
+  %a4 = load volatile i32, i32* @a
+  %x4 = add i32 %x3, %a4
+  %a5 = load volatile i32, i32* @a
+  %x5 = add i32 %x4, %a5
+  %a6 = load volatile i32, i32* @a
+  %x6 = add i32 %x5, %a6
+  %a7 = load volatile i32, i32* @a
+  %x7 = add i32 %x6, %a7
+  %a8 = load volatile i32, i32* @a
+  %x8 = add i32 %x7, %a8
+  %a9 = load volatile i32, i32* @a
+  %x9 = add i32 %x8, %a9
+  %a10 = load volatile i32, i32* @a
+  %x10 = add i32 %x9, %a10
+  %a11 = load volatile i32, i32* @a
+  %x11 = add i32 %x10, %a11
+  %a12 = load volatile i32, i32* @a
+  %x12 = add i32 %x11, %a12
+  %a13 = load volatile i32, i32* @a
+  %x13 = add i32 %x12, %a13
+  %a14 = load volatile i32, i32* @a
+  %x14 = add i32 %x13, %a14
+  %a15 = load volatile i32, i32* @a
+  %x15 = add i32 %x14, %a15
+  ret i32 %x15
+}
+
+define i32 @outer() optsize {
+; CHECK-LABEL: @outer
+; CHECK-NOT: call i32 @inner()
+   %r = call i32 @inner()
+   ret i32 %r
+}
+
+define i32 @outer2() optsize {
+; CHECK-LABEL: @outer2
+; CHECK: call i32 @inner2()
+   %r = call i32 @inner2()
+   ret i32 %r
+}
+
+define i32 @outer3() minsize {
+; CHECK-LABEL: @outer3
+; CHECK: call i32 @inner()
+   %r = call i32 @inner()
+   ret i32 %r
+}
+
+define i32 @outer4() {
+; CHECK-LABEL: @outer4
+; CHECK-NOT: call i32 @inner()
+   %r = call i32 @inner()
+   ret i32 %r
+}
+
+define i32 @outer5() {
+; CHECK-LABEL: @outer5
+; CHECK-NOT: call i32 @inner2()
+   %r = call i32 @inner2()
+   ret i32 %r
+}
+
+define i32 @outer6() {
+; CHECK-LABEL: @outer6
+; CHECK: call i32 @inner3()
+   %r = call i32 @inner3()
+   ret i32 %r
+}
+
+declare void @extern()