[InstrProf] Add option to avoid instrumenting small functions

[ellis]

Summary

If a function only has a few instructions, instrumentation can significantly increase the size and performance overhead of that function. Add the -pgo-function-size-threshold option to select a size threshold so these small functions are not instrumented.

A similar option -fxray-instruction-threshold=<N> is used for XRay to reduce binary size overhead [1].

[1] https://www.llvm.org/docs/XRay.html

[ellis]

I originally wanted to check this threshold in clang and add the skipprofile attribute for small functions. Unfortunately, clang doesn't have a good idea of how small functions are, i.e., how many instructions they have, before codegen and before optimizations. Of course, the number of IR instructions does not map directly to the size of the assembly function, but this is the best we can do since this is when we are injecting instrumentation.

[MaskRay]

LGTM. Please wait a bit for others' opinions.

llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp

297

The convention is to omit the period for help messages. Just ignore a few options in this file which do not stick with the convention.

llvm/test/Transforms/PGOProfile/function_size_threshold.ll

8

CHECK-LABEL

15

CHECK-LABEL

[ellis]

llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp

297

Thanks!

I think you’ve helped me with opt<> before, e.g., to omit cl::init(0). Is there some guide that describes the defaults and the conversions for this? Maybe we could add it to the programmers manual?

https://www.llvm.org/docs/ProgrammersManual.html

[ellis]

Resolve comments

llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp

static cl::opt<unsigned> PGOVerifyBFICutoff(
     cl::desc("Set the threshold for pgo-verify-bfi: skip the counts whose "
              "profile count value is below."));
 
 static cl::opt<std::string> PGOTraceFuncHash(
     "pgo-trace-func-hash", cl::init("-"), cl::Hidden,
     cl::value_desc("function name"),
     cl::desc("Trace the hash of the function with this name."));
 
+static cl::opt<unsigned> PGOFunctionSizeThreshold(
+    "pgo-function-size-threshold", cl::Hidden,
+    cl::desc("Do not instrument functions smaller than this threshold"));
+
 namespace llvm {
 // Command line option to turn on CFG dot dump after profile annotation.
 // Defined in Analysis/BlockFrequencyInfo.cpp:  -pgo-view-counts
 extern cl::opt<PGOViewCountsType> PGOViewCounts;
 
 // Command line option to specify the name of the function for CFG dump
 // Defined in Analysis/BlockFrequencyInfo.cpp:  -view-bfi-func-name=
 extern cl::opt<std::string> ViewBlockFreqFuncName;
static bool InstrumentAllFunctions(
 
   for (auto &F : M) {
     if (F.isDeclaration())
       continue;
     if (F.hasFnAttribute(llvm::Attribute::NoProfile))
       continue;
     if (F.hasFnAttribute(llvm::Attribute::SkipProfile))
       continue;
+    if (F.getInstructionCount() < PGOFunctionSizeThreshold)
+      continue;
     auto &TLI = LookupTLI(F);
     auto *BPI = LookupBPI(F);
     auto *BFI = LookupBFI(F);
     instrumentOneFunc(F, &M, TLI, BPI, BFI, ComdatMembers, IsCS);
   }
   return true;
 }
 

llvm/test/Transforms/PGOProfile/function_size_threshold.ll

+; RUN: opt < %s -passes=pgo-instr-gen -S | FileCheck %s --implicit-check-not='llvm.instrprof.' --check-prefixes=CHECK,ALL
+; RUN: opt < %s -passes=pgo-instr-gen -pgo-function-size-threshold=2 -S | FileCheck %s --implicit-check-not='llvm.instrprof.' --check-prefixes=CHECK,ALL
+; RUN: opt < %s -passes=pgo-instr-gen -pgo-function-size-threshold=3 -S | FileCheck %s --implicit-check-not='llvm.instrprof.'
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; CHECK-LABEL: define i32 @small(
+define i32 @small(i32 %i) {
+  ; ALL: call void @llvm.instrprof.increment({{.*}})
+  %add = add i32 %i, 4
+  ret i32 %add
+}
+
+; CHECK-LABEL: define i32 @large(
+define i32 @large(i32 %0) {
+  ; CHECK: call void @llvm.instrprof.increment({{.*}})
+  %2 = shl nsw i32 %0, 3
+  %3 = or i32 %2, 4
+  %4 = mul i32 %3, %0
+  %5 = or i32 %4, 3
+  %6 = sdiv i32 2, %0
+  %7 = add nsw i32 %5, %6
+  %8 = mul nsw i32 %0, %0
+  %9 = udiv i32 5, %8
+  %10 = add nsw i32 %7, %9
+  ret i32 %10
+}
+
+; CHECK: declare void @llvm.instrprof.increment({{.*}})