[PGO] Add another interface for annotateValueSite

[xur]

Summary

Add another interface to function annotateValueSite() which directly uses the VauleData array. This is needed in indirect-call promotion where un-promoted valuedata needs to be updated in the MD.prof meta data.

[davidxl]

What is main reason/need for vp data update after the transformation?

[xur]

my indirect-call promotion implementation supports multiple targets (specified the by an option) for a single callsite and the transformation pass can be called multiple times. The promotion stops at the first non-promoted direct target (as it's sorted on the count).

For example, for a callsite with the the following valuedata {<func1, 100> <func2, 80> <func3,10>}, after we promote func1 and cannot promote func2, I will attach {<func2, 80> <func3, 10>} to the indirect-call instruction.

[davidxl]

See also discussions in review D17115.

David

[davidxl]

Aside from how transformation should be done, I think annotating the remaining indirect call with the right meta data after the transformation is the right thing to do, and the information can be used for later optimizations (such as function layout based on dyn call graph).

LGTM.

include/llvm/ProfileData/InstrProf.h

 
 /// Get the value profile data for value site \p SiteIdx from \p InstrProfR
 /// and annotate the instruction \p Inst with the value profile meta data.
 /// Annotate up to \p MaxMDCount (default 3) number of records per value site.
 void annotateValueSite(Module &M, Instruction &Inst,
                        const InstrProfRecord &InstrProfR,
                        InstrProfValueKind ValueKind, uint32_t SiteIndx,
                        uint32_t MaxMDCount = 3);
+/// Same as the above interface but using the ValueData array directly, as
+/// well as \p Sum.
+void annotateValueSite(Module &M, Instruction &Inst,
+                       const InstrProfValueData VD[], uint32_t NV,
+                       uint64_t Sum, InstrProfValueKind ValueKind,
+                       uint32_t MaxMDCount);
+
 /// Extract the value profile data from \p Inst which is annotated with
 /// value profile meta data. Return false if there is no value data annotated,
 /// otherwise  return true.
 bool getValueProfDataFromInst(const Instruction &Inst,
                               InstrProfValueKind ValueKind,
                               uint32_t MaxNumValueData,
                               InstrProfValueData ValueData[],
                               uint32_t &ActualNumValueData, uint64_t &TotalC);

lib/ProfileData/InstrProf.cpp

void annotateValueSite(Module &M, Instruction &Inst,
                        InstrProfValueKind ValueKind, uint32_t SiteIdx,
                        uint32_t MaxMDCount) {
   uint32_t NV = InstrProfR.getNumValueDataForSite(ValueKind, SiteIdx);
 
   uint64_t Sum = 0;
   std::unique_ptr<InstrProfValueData[]> VD =
       InstrProfR.getValueForSite(ValueKind, SiteIdx, &Sum);
 
+  annotateValueSite(M, Inst, VD.get(), NV, Sum, ValueKind, MaxMDCount);
+}
+
+void annotateValueSite(Module &M, Instruction &Inst,
+                       const InstrProfValueData VD[], uint32_t NV,
+                       uint64_t Sum, InstrProfValueKind ValueKind,
+                       uint32_t MaxMDCount) {
   LLVMContext &Ctx = M.getContext();
   MDBuilder MDHelper(Ctx);
   SmallVector<Metadata *, 3> Vals;
   // Tag
   Vals.push_back(MDHelper.createString("VP"));
   // Value Kind
   Vals.push_back(MDHelper.createConstant(
       ConstantInt::get(Type::getInt32Ty(Ctx), ValueKind)));

unittests/ProfileData/InstrProfTest.cpp

TEST_P(MaybeSparseInstrProfTest, annotate_vp_data) {
   ASSERT_EQ(5000U, ValueData[1].Value);
   ASSERT_EQ(5U, ValueData[1].Count);
   ASSERT_EQ(4000U, ValueData[2].Value);
   ASSERT_EQ(4U, ValueData[2].Count);
   ASSERT_EQ(3000U, ValueData[3].Value);
   ASSERT_EQ(3U, ValueData[3].Count);
   ASSERT_EQ(2000U, ValueData[4].Value);
   ASSERT_EQ(2U, ValueData[4].Count);
+
+  // Remove the MD_prof metadata 
+  Inst->setMetadata(LLVMContext::MD_prof, 0);
+  // Annotate with 4 records.
+  InstrProfValueData VD0Sorted[] = {{1000, 6}, {2000, 5}, {3000, 4}, {4000, 3},
+                              {5000, 2}, {6000, 1}};
+  annotateValueSite(*M, *Inst, &VD0Sorted[2], 4, 10, IPVK_IndirectCallTarget, 5);
+  Res = getValueProfDataFromInst(*Inst, IPVK_IndirectCallTarget, 5,
+                                      ValueData, N, T);
+  ASSERT_TRUE(Res);
+  ASSERT_EQ(4U, N);
+  ASSERT_EQ(10U, T);
+  ASSERT_EQ(3000U, ValueData[0].Value);
+  ASSERT_EQ(4U, ValueData[0].Count);
+  ASSERT_EQ(4000U, ValueData[1].Value);
+  ASSERT_EQ(3U, ValueData[1].Count);
+  ASSERT_EQ(5000U, ValueData[2].Value);
+  ASSERT_EQ(2U, ValueData[2].Count);
+  ASSERT_EQ(6000U, ValueData[3].Value);
+  ASSERT_EQ(1U, ValueData[3].Count);
 }
 
 TEST_P(MaybeSparseInstrProfTest, get_icall_data_read_write_with_weight) {
   InstrProfRecord Record1("caller", 0x1234, {1, 2});
   InstrProfRecord Record2("callee1", 0x1235, {3, 4});
   InstrProfRecord Record3("callee2", 0x1235, {3, 4});
   InstrProfRecord Record4("callee3", 0x1235, {3, 4});