[Inlining] Set -enable-npm-pgo-inline-deferral to false.

[kazu]

Summary

This patch turns off inline deferral by default during PGO builds
except in the pre-inlining phase.

Although this patch improves the performance of an internal benchmark,
the SPEC performance remains unchanged -- 38.1 vs 38.1.

PGO+ThinLTO bootstrapped clang becomes slightly smaller.

Section Base Exp Diff

.text.hot 5974666 5932938 -0.70%
.text 7075401 7065513 -0.14%
.text.unlikely 35931984 35931376 -0.00%

.text.startup 734704 732304 -0.33%

Top2 13050067 12998451 -0.40%
Total 49716755 49662131 -0.11%

[wenlei]

Is it intentional to turn off inline deferral for both PGO and AutoFDO? Looking at D80776, the flag covers both PGO and AutoFDO..

[davidxl]

Wenlei is right. PGOOpt covers both sample and instrumentation PGO.

[kazu]

PTAL Thanks!

llvm/lib/Passes/PassBuilder.cpp

EnableCHR("enable-chr-npm", cl::init(true), cl::Hidden,
               cl::desc("Enable control height reduction optimization (CHR)"));
 
 static cl::opt<bool> EnableCallGraphProfile(
     "enable-npm-call-graph-profile", cl::init(true), cl::Hidden,
     cl::desc("Enable call graph profile pass for the new PM (default = on)"));
 
 /// Flag to enable inline deferral during PGO.
 static cl::opt<bool>
-    EnablePGOInlineDeferral("enable-npm-pgo-inline-deferral", cl::init(true),
+    EnablePGOInlineDeferral("enable-npm-pgo-inline-deferral", cl::init(false),
                             cl::Hidden,
                             cl::desc("Enable inline deferral during PGO"));
 
 PipelineTuningOptions::PipelineTuningOptions() {
   LoopInterleaving = true;
   LoopVectorization = true;
   SLPVectorization = false;
   LoopUnrolling = true;