[clang] e0092ea - [RISCV][clang] Optimize memory usage of intrinsic lookup table (#77487)

[via cfe-commits]

Author: Brandon Wu
Date: 2024-01-26T11:16:28+08:00
New Revision: e0092eae431956a2fd17f7ea88e7ba26d5e44f7e

URL: https://github.com/llvm/llvm-project/commit/e0092eae431956a2fd17f7ea88e7ba26d5e44f7e
DIFF: https://github.com/llvm/llvm-project/commit/e0092eae431956a2fd17f7ea88e7ba26d5e44f7e.diff

LOG: [RISCV][clang] Optimize memory usage of intrinsic lookup table (#77487)

This patch optimize:
  1. Reduce string size of RVVIntrinsicDef.
  2. Reduce the type size of the index of intrinsics.

I use valgrind --tool=massif to analyze a simple program:
```
#include <riscv_vector.h>
vint32m1_t test(vint32m1_t v1, vint32m1_t v2, size_t vl) {
  return __riscv_vadd(v1, v2, vl);
}
```
and before optimization, the peak memory usage is 15.68MB,
after optimization, the peak memory usage is 13.69MB.

Added: 
    

Modified: 
    clang/include/clang/Support/RISCVVIntrinsicUtils.h
    clang/lib/Sema/SemaRISCVVectorLookup.cpp
    clang/lib/Support/RISCVVIntrinsicUtils.cpp

Removed: 
    


################################################################################
diff  --git a/clang/include/clang/Support/RISCVVIntrinsicUtils.h b/clang/include/clang/Support/RISCVVIntrinsicUtils.h
index 05a5e02e139005..15416a33934410 100644
--- a/clang/include/clang/Support/RISCVVIntrinsicUtils.h
+++ b/clang/include/clang/Support/RISCVVIntrinsicUtils.h
@@ -416,8 +416,6 @@ class RVVIntrinsic {
   RVVTypePtr getOutputType() const { return OutputType; }
   const RVVTypes &getInputTypes() const { return InputTypes; }
   llvm::StringRef getBuiltinName() const { return BuiltinName; }
-  llvm::StringRef getName() const { return Name; }
-  llvm::StringRef getOverloadedName() const { return OverloadedName; }
   bool hasMaskedOffOperand() const { return HasMaskedOffOperand; }
   bool hasVL() const { return HasVL; }
   bool hasPolicy() const { return Scheme != PolicyScheme::SchemeNone; }

diff  --git a/clang/lib/Sema/SemaRISCVVectorLookup.cpp b/clang/lib/Sema/SemaRISCVVectorLookup.cpp
index 00a5ea65f3f4e7..25d5d20be7c36c 100644
--- a/clang/lib/Sema/SemaRISCVVectorLookup.cpp
+++ b/clang/lib/Sema/SemaRISCVVectorLookup.cpp
@@ -43,7 +43,7 @@ struct RVVIntrinsicDef {
 
 struct RVVOverloadIntrinsicDef {
   // Indexes of RISCVIntrinsicManagerImpl::IntrinsicList.
-  SmallVector<uint32_t, 8> Indexes;
+  SmallVector<uint16_t, 8> Indexes;
 };
 
 } // namespace
@@ -162,7 +162,7 @@ class RISCVIntrinsicManagerImpl : public sema::RISCVIntrinsicManager {
   // List of all RVV intrinsic.
   std::vector<RVVIntrinsicDef> IntrinsicList;
   // Mapping function name to index of IntrinsicList.
-  StringMap<uint32_t> Intrinsics;
+  StringMap<uint16_t> Intrinsics;
   // Mapping function name to RVVOverloadIntrinsicDef.
   StringMap<RVVOverloadIntrinsicDef> OverloadIntrinsics;
 
@@ -379,14 +379,16 @@ void RISCVIntrinsicManagerImpl::InitRVVIntrinsic(
     OverloadedName += "_" + OverloadedSuffixStr.str();
 
   // clang built-in function name, e.g. __builtin_rvv_vadd.
-  std::string BuiltinName = "__builtin_rvv_" + std::string(Record.Name);
+  std::string BuiltinName = std::string(Record.Name);
 
   RVVIntrinsic::updateNamesAndPolicy(IsMasked, HasPolicy, Name, BuiltinName,
                                      OverloadedName, PolicyAttrs,
                                      Record.HasFRMRoundModeOp);
 
   // Put into IntrinsicList.
-  uint32_t Index = IntrinsicList.size();
+  uint16_t Index = IntrinsicList.size();
+  assert(IntrinsicList.size() == (size_t)Index &&
+         "Intrinsics indices overflow.");
   IntrinsicList.push_back({BuiltinName, Signature});
 
   // Creating mapping to Intrinsics.
@@ -451,7 +453,8 @@ void RISCVIntrinsicManagerImpl::CreateRVVIntrinsicDecl(LookupResult &LR,
     RVVIntrinsicDecl->addAttr(OverloadableAttr::CreateImplicit(Context));
 
   // Setup alias to __builtin_rvv_*
-  IdentifierInfo &IntrinsicII = PP.getIdentifierTable().get(IDef.BuiltinName);
+  IdentifierInfo &IntrinsicII =
+      PP.getIdentifierTable().get("__builtin_rvv_" + IDef.BuiltinName);
   RVVIntrinsicDecl->addAttr(
       BuiltinAliasAttr::CreateImplicit(S.Context, &IntrinsicII));
 
@@ -463,6 +466,8 @@ bool RISCVIntrinsicManagerImpl::CreateIntrinsicIfFound(LookupResult &LR,
                                                        IdentifierInfo *II,
                                                        Preprocessor &PP) {
   StringRef Name = II->getName();
+  if (!Name.consume_front("__riscv_"))
+    return false;
 
   // Lookup the function name from the overload intrinsics first.
   auto OvIItr = OverloadIntrinsics.find(Name);

diff  --git a/clang/lib/Support/RISCVVIntrinsicUtils.cpp b/clang/lib/Support/RISCVVIntrinsicUtils.cpp
index 2de977a3dc720b..7d2a2d7e826f9c 100644
--- a/clang/lib/Support/RISCVVIntrinsicUtils.cpp
+++ b/clang/lib/Support/RISCVVIntrinsicUtils.cpp
@@ -1150,11 +1150,6 @@ void RVVIntrinsic::updateNamesAndPolicy(
     OverloadedName += suffix;
   };
 
-  // This follows the naming guideline under riscv-c-api-doc to add the
-  // `__riscv_` suffix for all RVV intrinsics.
-  Name = "__riscv_" + Name;
-  OverloadedName = "__riscv_" + OverloadedName;
-
   if (HasFRMRoundModeOp) {
     Name += "_rm";
     BuiltinName += "_rm";