C++/C#: Fix getIRTypeForPRValue join order

jbj · jbj · commit 6385528d5fe9 · 2019-11-07T11:48:16.000+01:00
This predicate was taking 39s on a snapshot of Facebook Fizz because it
had disjuncts like this:

    43685     ~0%     {1} r34 = JOIN Type::FunctionPointerIshType#f AS L WITH Type::Type::getUnspecifiedType_dispred#ff_10#join_rhs AS R ON FIRST 1 OUTPUT R.&lt;1&gt;
    43685     ~1%     {2} r35 = JOIN r34 WITH CppType::getTypeSize#ff AS R ON FIRST 1 OUTPUT R.&lt;1&gt;, r34.&lt;0&gt;
    170371500 ~2%     {2} r36 = JOIN r35 WITH IRType::IRSizedType#ff_10#join_rhs AS R ON FIRST 1 OUTPUT R.&lt;1&gt;, r35.&lt;1&gt;
    43685     ~6%     {2} r37 = JOIN r36 WITH IRType::IRFunctionAddressType#class#ff AS R ON FIRST 1 OUTPUT r36.&lt;1&gt;, r36.&lt;0&gt;

Instead of fixing the joins in `getIRTypeForPRValue` itself, I've
changed the `IRType::getByteSize` predicate such that the optimiser
knows how to join with it efficiently.

The disjunct shown above now looks like this instead:

    43685  ~0%     {1} r26 = JOIN Type::FunctionPointerIshType#f AS L WITH Type::Type::getUnspecifiedType_dispred#ff_10#join_rhs AS R ON FIRST 1 OUTPUT R.&lt;1&gt;
    43685  ~1%     {2} r27 = JOIN r26 WITH CppType::getTypeSize#ff AS R ON FIRST 1 OUTPUT R.&lt;1&gt;, r26.&lt;0&gt;
    43685  ~6%     {2} r28 = JOIN r27 WITH IRType::IRFunctionAddressType::getByteSize#ff_10#join_rhs AS R ON FIRST 1 OUTPUT r27.&lt;1&gt;, R.&lt;1&gt;
diff --git a/cpp/ql/src/semmle/code/cpp/ir/implementation/IRType.qll b/cpp/ql/src/semmle/code/cpp/ir/implementation/IRType.qll
@@ -42,6 +42,10 @@ class IRType extends TIRType {
    *
    * This will hold for all `IRType` objects except `IRUnknownType`.
    */
+  // This predicate is overridden with `pragma[noinline]` in every leaf subclass.
+  // This allows callers to ask for things like _the_ floating-point type of
+  // size 4 without getting a join that first finds all types of size 4 and
+  // _then_ restricts them to floating-point types.
   int getByteSize() { none() }
 
   /**
@@ -104,8 +108,6 @@ private class IRSizedType extends IRType {
     this = TIRFunctionAddressType(byteSize) or
     this = TIROpaqueType(_, byteSize)
   }
-
-  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -117,6 +119,9 @@ class IRBooleanType extends IRSizedType, TIRBooleanType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalBooleanType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -141,6 +146,9 @@ class IRSignedIntegerType extends IRNumericType, TIRSignedIntegerType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalSignedIntegerType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -153,6 +161,9 @@ class IRUnsignedIntegerType extends IRNumericType, TIRUnsignedIntegerType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalUnsignedIntegerType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -164,6 +175,9 @@ class IRFloatingPointType extends IRNumericType, TIRFloatingPointType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalFloatingPointType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -178,6 +192,9 @@ class IRAddressType extends IRSizedType, TIRAddressType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalAddressType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -190,6 +207,9 @@ class IRFunctionAddressType extends IRSizedType, TIRFunctionAddressType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalFunctionAddressType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -218,6 +238,9 @@ class IROpaqueType extends IRSizedType, TIROpaqueType {
    * same size.
    */
   final Language::OpaqueTypeTag getTag() { result = tag }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 module IRTypeSanity {
diff --git a/csharp/ql/src/semmle/code/csharp/ir/implementation/IRType.qll b/csharp/ql/src/semmle/code/csharp/ir/implementation/IRType.qll
@@ -42,6 +42,10 @@ class IRType extends TIRType {
    *
    * This will hold for all `IRType` objects except `IRUnknownType`.
    */
+  // This predicate is overridden with `pragma[noinline]` in every leaf subclass.
+  // This allows callers to ask for things like _the_ floating-point type of
+  // size 4 without getting a join that first finds all types of size 4 and
+  // _then_ restricts them to floating-point types.
   int getByteSize() { none() }
 
   /**
@@ -104,8 +108,6 @@ private class IRSizedType extends IRType {
     this = TIRFunctionAddressType(byteSize) or
     this = TIROpaqueType(_, byteSize)
   }
-
-  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -117,6 +119,9 @@ class IRBooleanType extends IRSizedType, TIRBooleanType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalBooleanType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -141,6 +146,9 @@ class IRSignedIntegerType extends IRNumericType, TIRSignedIntegerType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalSignedIntegerType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -153,6 +161,9 @@ class IRUnsignedIntegerType extends IRNumericType, TIRUnsignedIntegerType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalUnsignedIntegerType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -164,6 +175,9 @@ class IRFloatingPointType extends IRNumericType, TIRFloatingPointType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalFloatingPointType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -178,6 +192,9 @@ class IRAddressType extends IRSizedType, TIRAddressType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalAddressType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -190,6 +207,9 @@ class IRFunctionAddressType extends IRSizedType, TIRFunctionAddressType {
   final override Language::LanguageType getCanonicalLanguageType() {
     result = Language::getCanonicalFunctionAddressType(byteSize)
   }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 /**
@@ -218,6 +238,9 @@ class IROpaqueType extends IRSizedType, TIROpaqueType {
    * same size.
    */
   final Language::OpaqueTypeTag getTag() { result = tag }
+
+  pragma[noinline]
+  final override int getByteSize() { result = byteSize }
 }
 
 module IRTypeSanity {

Original file line number	Diff line number	Diff line change
`@@ -42,6 +42,10 @@ class IRType extends TIRType {`
`42`	`42`	`*`
`43`	`43`	* This will hold for all `IRType` objects except `IRUnknownType`.
`44`	`44`	`*/`
	`45`	+ // This predicate is overridden with `pragma[noinline]` in every leaf subclass.
	`46`	`+ // This allows callers to ask for things like _the_ floating-point type of`
	`47`	`+ // size 4 without getting a join that first finds all types of size 4 and`
	`48`	`+ // _then_ restricts them to floating-point types.`
`45`	`49`	`int getByteSize() { none() }`
`46`	`50`
`47`	`51`	`/**`
`@@ -104,8 +108,6 @@ private class IRSizedType extends IRType {`
`104`	`108`	`this = TIRFunctionAddressType(byteSize) or`
`105`	`109`	`this = TIROpaqueType(_, byteSize)`
`106`	`110`	`}`
`107`		`-`
`108`		`- final override int getByteSize() { result = byteSize }`
`109`	`111`	`}`
`110`	`112`
`111`	`113`	`/**`
`@@ -117,6 +119,9 @@ class IRBooleanType extends IRSizedType, TIRBooleanType {`
`117`	`119`	`final override Language::LanguageType getCanonicalLanguageType() {`
`118`	`120`	`result = Language::getCanonicalBooleanType(byteSize)`
`119`	`121`	`}`
	`122`	`+`
	`123`	`+ pragma[noinline]`
	`124`	`+ final override int getByteSize() { result = byteSize }`
`120`	`125`	`}`
`121`	`126`
`122`	`127`	`/**`
`@@ -141,6 +146,9 @@ class IRSignedIntegerType extends IRNumericType, TIRSignedIntegerType {`
`141`	`146`	`final override Language::LanguageType getCanonicalLanguageType() {`
`142`	`147`	`result = Language::getCanonicalSignedIntegerType(byteSize)`
`143`	`148`	`}`
	`149`	`+`
	`150`	`+ pragma[noinline]`
	`151`	`+ final override int getByteSize() { result = byteSize }`
`144`	`152`	`}`
`145`	`153`
`146`	`154`	`/**`
`@@ -153,6 +161,9 @@ class IRUnsignedIntegerType extends IRNumericType, TIRUnsignedIntegerType {`
`153`	`161`	`final override Language::LanguageType getCanonicalLanguageType() {`
`154`	`162`	`result = Language::getCanonicalUnsignedIntegerType(byteSize)`
`155`	`163`	`}`
	`164`	`+`
	`165`	`+ pragma[noinline]`
	`166`	`+ final override int getByteSize() { result = byteSize }`
`156`	`167`	`}`
`157`	`168`
`158`	`169`	`/**`
`@@ -164,6 +175,9 @@ class IRFloatingPointType extends IRNumericType, TIRFloatingPointType {`
`164`	`175`	`final override Language::LanguageType getCanonicalLanguageType() {`
`165`	`176`	`result = Language::getCanonicalFloatingPointType(byteSize)`
`166`	`177`	`}`
	`178`	`+`
	`179`	`+ pragma[noinline]`
	`180`	`+ final override int getByteSize() { result = byteSize }`
`167`	`181`	`}`
`168`	`182`
`169`	`183`	`/**`
`@@ -178,6 +192,9 @@ class IRAddressType extends IRSizedType, TIRAddressType {`
`178`	`192`	`final override Language::LanguageType getCanonicalLanguageType() {`
`179`	`193`	`result = Language::getCanonicalAddressType(byteSize)`
`180`	`194`	`}`
	`195`	`+`
	`196`	`+ pragma[noinline]`
	`197`	`+ final override int getByteSize() { result = byteSize }`
`181`	`198`	`}`
`182`	`199`
`183`	`200`	`/**`
`@@ -190,6 +207,9 @@ class IRFunctionAddressType extends IRSizedType, TIRFunctionAddressType {`
`190`	`207`	`final override Language::LanguageType getCanonicalLanguageType() {`
`191`	`208`	`result = Language::getCanonicalFunctionAddressType(byteSize)`
`192`	`209`	`}`
	`210`	`+`
	`211`	`+ pragma[noinline]`
	`212`	`+ final override int getByteSize() { result = byteSize }`
`193`	`213`	`}`
`194`	`214`
`195`	`215`	`/**`
`@@ -218,6 +238,9 @@ class IROpaqueType extends IRSizedType, TIROpaqueType {`
`218`	`238`	`* same size.`
`219`	`239`	`*/`
`220`	`240`	`final Language::OpaqueTypeTag getTag() { result = tag }`
	`241`	`+`
	`242`	`+ pragma[noinline]`
	`243`	`+ final override int getByteSize() { result = byteSize }`
`221`	`244`	`}`
`222`	`245`
`223`	`246`	`module IRTypeSanity {`