Adding Matching and Inference Functionality to Propeller-PR3: Read basic block hashes from propeller profile. (llvm#164223)

Adding Matching and Inference Functionality to Propeller. For detailed
information, please refer to the following RFC:
https://discourse.llvm.org/t/rfc-adding-matching-and-inference-functionality-to-propeller/86238.
This is the third PR, which is used to read basic block hashes from the
propeller profile. The associated PRs are:
PR1: llvm#160706
PR2: llvm#162963

co-authors: lifengxiang1025
[[email protected]](mailto:[email protected]); zcfh
[[email protected]](mailto:[email protected])

Co-authored-by: lifengxiang1025 <[email protected]>
Co-authored-by: zcfh <[email protected]>

Author: 3 people

llvm/include/llvm/CodeGen/BasicBlockSectionsProfileReader.h

@@ -54,6 +54,10 @@ struct FunctionPathAndClusterInfo {
   DenseMap<UniqueBBID, uint64_t> NodeCounts;
   // Edge counts for each edge, stored as a nested map.
   DenseMap<UniqueBBID, DenseMap<UniqueBBID, uint64_t>> EdgeCounts;
+  // Hash for each basic block. The Hashes are stored for every original block
+  // (not cloned blocks), hence the map key being unsigned instead of
+  // UniqueBBID.
+  DenseMap<unsigned, uint64_t> BBHashes;
 };
 
 class BasicBlockSectionsProfileReader {

llvm/lib/CodeGen/BasicBlockSectionsProfileReader.cpp

@@ -287,6 +287,25 @@ Error BasicBlockSectionsProfileReader::ReadV1Profile() {
       }
       continue;
     }
+    case 'h': { // Basic block hash secifier.
+      // Skip the profile when the profile iterator (FI) refers to the
+      // past-the-end element.
+      if (FI == ProgramPathAndClusterInfo.end())
+        continue;
+      for (auto BBIDHashStr : Values) {
+        auto [BBIDStr, HashStr] = BBIDHashStr.split(':');
+        unsigned long long BBID = 0, Hash = 0;
+        if (getAsUnsignedInteger(BBIDStr, 10, BBID))
+          return createProfileParseError(Twine("unsigned integer expected: '") +
+                                         BBIDStr + "'");
+        if (getAsUnsignedInteger(HashStr, 16, Hash))
+          return createProfileParseError(
+              Twine("unsigned integer expected in hex format: '") + HashStr +
+              "'");
+        FI->second.BBHashes[BBID] = Hash;
+      }
+      continue;
+    }
     default:
       return createProfileParseError(Twine("invalid specifier: '") +
                                      Twine(Specifier) + "'");

llvm/test/CodeGen/X86/basic-block-sections-bb-hash.ll

@@ -0,0 +1,51 @@
+; BB section test with basic block hashes.
+;
+; RUN: llc %s -O0 -mtriple=x86_64-pc-linux -function-sections -filetype=obj -basic-block-address-map -emit-bb-hash -o %t.o
+; RUN: obj2yaml %t.o -o %t.yaml
+;
+;; Profile for version 1:
+; RUN: echo 'v1' > %t
+; RUN: echo 'f foo' >> %t
+; RUN: echo 'g 0:10,1:9,2:1 1:8,3:8 2:2,3:2 3:11' >> %t
+; RUN: echo 'c 0 2 3' >> %t
+
+; These commands read BB hashes from SHT_LLVM_BB_ADDR_MAP 
+; and put them into the basic blocks sections profile.
+; RUN: grep -E '^\s+(- ID:|Hash:)' %t.yaml | \
+; RUN: grep -B1 'Hash:' | \
+; RUN: sed 's/^\s*//; s/^- ID: *//; s/Hash: *0x//' | \
+; RUN: paste -d: - - | \
+; RUN: tr '\n' ' ' | \
+; RUN: sed 's/ $/\n/; s/^/h /' >> %t
+;
+; RUN: llc < %s -O0 -mtriple=x86_64-pc-linux -function-sections -basic-block-sections=%t | FileCheck %s
+;
+define void @foo(i1 zeroext) nounwind {
+  %2 = alloca i8, align 1
+  %3 = zext i1 %0 to i8
+  store i8 %3, ptr %2, align 1
+  %4 = load i8, ptr %2, align 1
+  %5 = trunc i8 %4 to i1
+  br i1 %5, label %6, label %8
+
+6:                                                ; preds = %1
+  %7 = call i32 @bar()
+  br label %10
+
+8:                                                ; preds = %1
+  %9 = call i32 @baz()
+  br label %10
+
+10:                                               ; preds = %8, %6
+  ret void
+}
+
+declare i32 @bar() #1
+
+declare i32 @baz() #1
+
+; CHECK: .section	.text.foo,"ax",@progbits
+; CHECK: callq baz
+; CHECK: retq
+; CHECK: .section	.text.split.foo,"ax",@progbits
+; CHECK: callq bar

llvm/test/CodeGen/X86/basic-block-sections-clusters-error.ll

@@ -69,6 +69,20 @@
 ; RUN: echo 'g 0:4,1:2:3' >> %t15
 ; RUN: not --crash llc < %s -O0 -mtriple=x86_64 -function-sections -basic-block-sections=%t15 2>&1 | FileCheck %s --check-prefix=CHECK-ERROR15
 ; CHECK-ERROR15: LLVM ERROR: invalid profile {{.*}} at line 4: unsigned integer expected: '2:3'
+; RUN: echo 'v1' > %t16
+; RUN: echo 'f dummy1' >> %t16
+; RUN: echo 'c 0 1' >> %t16
+; RUN: echo 'g 0:4,1:2' >> %t16
+; RUN: echo 'h a:1111111111111111 1:ffffffffffffffff' >> %t16
+; RUN: not --crash llc < %s -O0 -mtriple=x86_64 -function-sections -basic-block-sections=%t16 2>&1 | FileCheck %s --check-prefix=CHECK-ERROR16
+; CHECK-ERROR16: LLVM ERROR: invalid profile {{.*}} at line 5: unsigned integer expected: 'a'
+; RUN: echo 'v1' > %t17
+; RUN: echo 'f dummy1' >> %t17
+; RUN: echo 'c 0 1' >> %t17
+; RUN: echo 'g 0:4,1:2' >> %t17
+; RUN: echo 'h 0:111111111111111g 1:ffffffffffffffff' >> %t17
+; RUN: not --crash llc < %s -O0 -mtriple=x86_64 -function-sections -basic-block-sections=%t17 2>&1 | FileCheck %s --check-prefix=CHECK-ERROR17
+; CHECK-ERROR17: LLVM ERROR: invalid profile {{.*}} at line 5: unsigned integer expected in hex format: '111111111111111g'
 
 
 define i32 @dummy1(i32 %x, i32 %y, i32 %z) {