Differentiate T_ARRAY and array subclasses (#7091)

* Differentiate T_ARRAY and array subclasses

This commit teaches the YJIT context the difference between Arrays
(objects with type T_ARRAY and class rb_cArray) vs Array subclasses
(objects with type T_ARRAY but _not_ class rb_cArray).  It uses this
information to reduce the number of guards emitted when using
`jit_guard_known_klass` with rb_cArray, notably opt_aref

* Update yjit/src/core.rs

Co-authored-by: Maxime Chevalier-Boisvert <maximechevalierb@gmail.com>

2 files changed, 22 insertions, 8 deletions

diff --git a/yjit/src/codegen.rs b/yjit/src/codegen.rs
index a7c2b37eeb..3dbc589466 100644
--- a/yjit/src/codegen.rs
+++ b/yjit/src/codegen.rs
@@ -1161,7 +1161,7 @@ fn gen_newarray(
     );
 
     ctx.stack_pop(n.as_usize());
-    let stack_ret = ctx.stack_push(Type::Array);
+    let stack_ret = ctx.stack_push(Type::CArray);
     asm.mov(stack_ret, new_ary);
 
     KeepCompiling
@@ -1185,7 +1185,7 @@ fn gen_duparray(
         vec![ary.into()],
     );
 
-    let stack_ret = ctx.stack_push(Type::Array);
+    let stack_ret = ctx.stack_push(Type::CArray);
     asm.mov(stack_ret, new_ary);
 
     KeepCompiling
@@ -1231,7 +1231,7 @@ fn gen_splatarray(
     // Call rb_vm_splat_array(flag, ary)
     let ary = asm.ccall(rb_vm_splat_array as *const u8, vec![flag.into(), ary_opnd]);
 
-    let stack_ret = ctx.stack_push(Type::Array);
+    let stack_ret = ctx.stack_push(Type::TArray);
     asm.mov(stack_ret, ary);
 
     KeepCompiling
@@ -1255,7 +1255,7 @@ fn gen_concatarray(
     // Call rb_vm_concat_array(ary1, ary2st)
     let ary = asm.ccall(rb_vm_concat_array as *const u8, vec![ary1_opnd, ary2st_opnd]);
 
-    let stack_ret = ctx.stack_push(Type::Array);
+    let stack_ret = ctx.stack_push(Type::TArray);
     asm.mov(stack_ret, ary);
 
     KeepCompiling
@@ -3886,6 +3886,8 @@ fn jit_guard_known_klass(
 
         if known_klass == unsafe { rb_cString } {
             ctx.upgrade_opnd_type(insn_opnd, Type::CString);
+        } else if known_klass == unsafe { rb_cArray } {
+            ctx.upgrade_opnd_type(insn_opnd, Type::CArray);
         }
     }
 }
diff --git a/yjit/src/core.rs b/yjit/src/core.rs
index 3b471e782e..71c9ae5a9e 100644
--- a/yjit/src/core.rs
+++ b/yjit/src/core.rs
@@ -35,7 +35,6 @@ pub enum Type {
     False,
     Fixnum,
     Flonum,
-    Array,
     Hash,
     ImmSymbol,
 
@@ -44,6 +43,8 @@ pub enum Type {
 
     TString, // An object with the T_STRING flag set, possibly an rb_cString
     CString, // An un-subclassed string of type rb_cString (can have instance vars in some cases)
+    TArray, // An object with the T_ARRAY flag set, possibly an rb_cArray
+    CArray, // An un-subclassed string of type rb_cArray (can have instance vars in some cases)
 
     BlockParamProxy, // A special sentinel value indicating the block parameter should be read from
                      // the current surrounding cfp
@@ -82,6 +83,10 @@ impl Type {
             if val.class_of() == unsafe { rb_cString } {
                 return Type::CString;
             }
+            #[cfg(not(test))]
+            if val.class_of() == unsafe { rb_cArray } {
+                return Type::CArray;
+            }
             // We likewise can't reference rb_block_param_proxy, but it's again an optimisation;
             // we can just treat it as a normal Object.
             #[cfg(not(test))]
@@ -89,7 +94,7 @@ impl Type {
                 return Type::BlockParamProxy;
             }
             match val.builtin_type() {
-                RUBY_T_ARRAY => Type::Array,
+                RUBY_T_ARRAY => Type::TArray,
                 RUBY_T_HASH => Type::Hash,
                 RUBY_T_STRING => Type::TString,
                 _ => Type::UnknownHeap,
@@ -130,7 +135,8 @@ impl Type {
     pub fn is_heap(&self) -> bool {
         match self {
             Type::UnknownHeap => true,
-            Type::Array => true,
+            Type::TArray => true,
+            Type::CArray => true,
             Type::Hash => true,
             Type::HeapSymbol => true,
             Type::TString => true,
@@ -147,7 +153,7 @@ impl Type {
             Type::False => Some(RUBY_T_FALSE),
             Type::Fixnum => Some(RUBY_T_FIXNUM),
             Type::Flonum => Some(RUBY_T_FLOAT),
-            Type::Array => Some(RUBY_T_ARRAY),
+            Type::TArray | Type::CArray => Some(RUBY_T_ARRAY),
             Type::Hash => Some(RUBY_T_HASH),
             Type::ImmSymbol | Type::HeapSymbol => Some(RUBY_T_SYMBOL),
             Type::TString | Type::CString => Some(RUBY_T_STRING),
@@ -167,6 +173,7 @@ impl Type {
                 Type::Flonum => Some(rb_cFloat),
                 Type::ImmSymbol | Type::HeapSymbol => Some(rb_cSymbol),
                 Type::CString => Some(rb_cString),
+                Type::CArray => Some(rb_cArray),
                 _ => None,
             }
         }
@@ -224,6 +231,11 @@ impl Type {
             return 1;
         }
 
+        // A CArray is also a TArray.
+        if self == Type::CArray && dst == Type::TArray {
+            return 1;
+        }
+
         // Specific heap type into unknown heap type is imperfect but valid
         if self.is_heap() && dst == Type::UnknownHeap {
             return 1;