ZJIT: Standardize C call related insn fields

- Add `recv` field to `CCall` and `CCallWithFrame` so now all method dispatch
  related instructions have `recv` field, separate from `args` field.
  This ensures consistent pointer arithmetic when generating code for these
  instructions.
- Standardize `recv` field's display position in send related instructions.

Author: st0012

zjit/src/codegen.rs

@@ -425,12 +425,13 @@ fn gen_insn(cb: &mut CodeBlock, jit: &mut JITState, asm: &mut Assembler, functio
         &Insn::GuardLess { left, right, state } => gen_guard_less(jit, asm, opnd!(left), opnd!(right), &function.frame_state(state)),
         &Insn::GuardGreaterEq { left, right, state } => gen_guard_greater_eq(jit, asm, opnd!(left), opnd!(right), &function.frame_state(state)),
         Insn::PatchPoint { invariant, state } => no_output!(gen_patch_point(jit, asm, invariant, &function.frame_state(*state))),
-        Insn::CCall { cfunc, args, name, return_type: _, elidable: _ } => gen_ccall(asm, *cfunc, *name, opnds!(args)),
-        // Give up CCallWithFrame for 7+ args since asm.ccall() doesn't support it.
-        Insn::CCallWithFrame { cd, state, args, .. } if args.len() > C_ARG_OPNDS.len() =>
+        Insn::CCall { cfunc, recv, args, name, return_type: _, elidable: _ } => gen_ccall(asm, *cfunc, *name, opnd!(recv), opnds!(args)),
+        // Give up CCallWithFrame for 7+ args since asm.ccall() supports at most 6 args (recv + args).
+        // There's no test case for this because no core cfuncs have this many parameters. But C extensions could have such methods.
+        Insn::CCallWithFrame { cd, state, args, .. } if args.len() + 1 > C_ARG_OPNDS.len() =>
             gen_send_without_block(jit, asm, *cd, &function.frame_state(*state), SendFallbackReason::CCallWithFrameTooManyArgs),
-        Insn::CCallWithFrame { cfunc, name, args, cme, state, blockiseq, .. } =>
-            gen_ccall_with_frame(jit, asm, *cfunc, *name, opnds!(args), *cme, *blockiseq, &function.frame_state(*state)),
+        Insn::CCallWithFrame { cfunc, recv, name, args, cme, state, blockiseq, .. } =>
+            gen_ccall_with_frame(jit, asm, *cfunc, *name, opnd!(recv), opnds!(args), *cme, *blockiseq, &function.frame_state(*state)),
         Insn::CCallVariadic { cfunc, recv, args, name, cme, state, return_type: _, elidable: _ } => {
             gen_ccall_variadic(jit, asm, *cfunc, *name, opnd!(recv), opnds!(args), *cme, &function.frame_state(*state))
         }
@@ -766,6 +767,7 @@ fn gen_ccall_with_frame(
     asm: &mut Assembler,
     cfunc: *const u8,
     name: ID,
+    recv: Opnd,
     args: Vec<Opnd>,
     cme: *const rb_callable_method_entry_t,
     blockiseq: Option<IseqPtr>,
@@ -774,7 +776,8 @@ fn gen_ccall_with_frame(
     gen_incr_counter(asm, Counter::non_variadic_cfunc_optimized_send_count);
     gen_stack_overflow_check(jit, asm, state, state.stack_size());
 
-    let caller_stack_size = state.stack_size() - args.len();
+    let args_with_recv_len = args.len() + 1;
+    let caller_stack_size = state.stack().len() - args_with_recv_len;
 
     // Can't use gen_prepare_non_leaf_call() because we need to adjust the SP
     // to account for the receiver and arguments (and block arguments if any)
@@ -794,8 +797,8 @@ fn gen_ccall_with_frame(
         VM_BLOCK_HANDLER_NONE.into()
     };
 
-    gen_push_frame(asm, args.len(), state, ControlFrame {
-        recv: args[0],
+    gen_push_frame(asm, args_with_recv_len, state, ControlFrame {
+        recv,
         iseq: None,
         cme,
         frame_type: VM_FRAME_MAGIC_CFUNC | VM_FRAME_FLAG_CFRAME | VM_ENV_FLAG_LOCAL,
@@ -813,8 +816,10 @@ fn gen_ccall_with_frame(
     asm.mov(CFP, new_cfp);
     asm.store(Opnd::mem(64, EC, RUBY_OFFSET_EC_CFP), CFP);
 
+    let mut cfunc_args = vec![recv];
+    cfunc_args.extend(args);
     asm.count_call_to(&name.contents_lossy());
-    let result = asm.ccall(cfunc, args);
+    let result = asm.ccall(cfunc, cfunc_args);
 
     asm_comment!(asm, "pop C frame");
     let new_cfp = asm.add(CFP, RUBY_SIZEOF_CONTROL_FRAME.into());
@@ -830,9 +835,11 @@ fn gen_ccall_with_frame(
 
 /// Lowering for [`Insn::CCall`]. This is a low-level raw call that doesn't know
 /// anything about the callee, so handling for e.g. GC safety is dealt with elsewhere.
-fn gen_ccall(asm: &mut Assembler, cfunc: *const u8, name: ID, args: Vec<Opnd>) -> lir::Opnd {
+fn gen_ccall(asm: &mut Assembler, cfunc: *const u8, name: ID, recv: Opnd, args: Vec<Opnd>) -> lir::Opnd {
+    let mut cfunc_args = vec![recv];
+    cfunc_args.extend(args);
     asm.count_call_to(&name.contents_lossy());
-    asm.ccall(cfunc, args)
+    asm.ccall(cfunc, cfunc_args)
 }
 
 /// Generate code for a variadic C function call

zjit/src/cruby_methods.rs

@@ -440,7 +440,8 @@ fn inline_string_eq(fun: &mut hir::Function, block: hir::BlockId, recv: hir::Ins
         // TODO(max): Make StringEqual its own opcode so that we can later constant-fold StringEqual(a, a) => true
         let result = fun.push_insn(block, hir::Insn::CCall {
             cfunc: rb_yarv_str_eql_internal as *const u8,
-            args: vec![recv, other],
+            recv,
+            args: vec![other],
             name: ID!(string_eq),
             return_type,
             elidable,

zjit/src/hir.rs

@@ -778,12 +778,13 @@ pub enum Insn {
 
     /// Call a C function without pushing a frame
     /// `name` is for printing purposes only
-    CCall { cfunc: *const u8, args: Vec<InsnId>, name: ID, return_type: Type, elidable: bool },
+    CCall { cfunc: *const u8, recv: InsnId, args: Vec<InsnId>, name: ID, return_type: Type, elidable: bool },
 
     /// Call a C function that pushes a frame
     CCallWithFrame {
         cd: *const rb_call_data, // cd for falling back to SendWithoutBlock
         cfunc: *const u8,
+        recv: InsnId,
         args: Vec<InsnId>,
         cme: *const rb_callable_method_entry_t,
         name: ID,
@@ -1179,8 +1180,8 @@ impl<'a> std::fmt::Display for InsnPrinter<'a> {
                 }
                 Ok(())
             }
-            Insn::SendForward { cd, args, blockiseq, .. } => {
-                write!(f, "SendForward {:p}, :{}", self.ptr_map.map_ptr(blockiseq), ruby_call_method_name(*cd))?;
+            Insn::SendForward { recv, cd, args, blockiseq, .. } => {
+                write!(f, "SendForward {recv}, {:p}, :{}", self.ptr_map.map_ptr(blockiseq), ruby_call_method_name(*cd))?;
                 for arg in args {
                     write!(f, ", {arg}")?;
                 }
@@ -1239,15 +1240,15 @@ impl<'a> std::fmt::Display for InsnPrinter<'a> {
             Insn::GetConstantPath { ic, .. } => { write!(f, "GetConstantPath {:p}", self.ptr_map.map_ptr(ic)) },
             Insn::IsBlockGiven => { write!(f, "IsBlockGiven") },
             Insn::FixnumBitCheck {val, index} => { write!(f, "FixnumBitCheck {val}, {index}") },
-            Insn::CCall { cfunc, args, name, return_type: _, elidable: _ } => {
-                write!(f, "CCall {}@{:p}", name.contents_lossy(), self.ptr_map.map_ptr(cfunc))?;
+            Insn::CCall { cfunc, recv, args, name, return_type: _, elidable: _ } => {
+                write!(f, "CCall {recv}, :{}@{:p}", name.contents_lossy(), self.ptr_map.map_ptr(cfunc))?;
                 for arg in args {
                     write!(f, ", {arg}")?;
                 }
                 Ok(())
             },
-            Insn::CCallWithFrame { cfunc, args, name, blockiseq, .. } => {
-                write!(f, "CCallWithFrame {}@{:p}", name.contents_lossy(), self.ptr_map.map_ptr(cfunc))?;
+            Insn::CCallWithFrame { cfunc, recv, args, name, blockiseq, .. } => {
+                write!(f, "CCallWithFrame {recv}, :{}@{:p}", name.contents_lossy(), self.ptr_map.map_ptr(cfunc))?;
                 for arg in args {
                     write!(f, ", {arg}")?;
                 }
@@ -1256,8 +1257,8 @@ impl<'a> std::fmt::Display for InsnPrinter<'a> {
                 }
                 Ok(())
             },
-            Insn::CCallVariadic { cfunc,  recv, args, name, .. } => {
-                write!(f, "CCallVariadic {}@{:p}, {recv}", name.contents_lossy(), self.ptr_map.map_ptr(cfunc))?;
+            Insn::CCallVariadic { cfunc, recv, args, name, .. } => {
+                write!(f, "CCallVariadic {recv}, :{}@{:p}", name.contents_lossy(), self.ptr_map.map_ptr(cfunc))?;
                 for arg in args {
                     write!(f, ", {arg}")?;
                 }
@@ -1908,10 +1909,11 @@ impl Function {
             &HashAref { hash, key, state } => HashAref { hash: find!(hash), key: find!(key), state },
             &ObjectAlloc { val, state } => ObjectAlloc { val: find!(val), state },
             &ObjectAllocClass { class, state } => ObjectAllocClass { class, state: find!(state) },
-            &CCall { cfunc, ref args, name, return_type, elidable } => CCall { cfunc, args: find_vec!(args), name, return_type, elidable },
-            &CCallWithFrame { cd, cfunc, ref args, cme, name, state, return_type, elidable, blockiseq } => CCallWithFrame {
+            &CCall { cfunc, recv, ref args, name, return_type, elidable } => CCall { cfunc, recv: find!(recv), args: find_vec!(args), name, return_type, elidable },
+            &CCallWithFrame { cd, cfunc, recv, ref args, cme, name, state, return_type, elidable, blockiseq } => CCallWithFrame {
                 cd,
                 cfunc,
+                recv: find!(recv),
                 args: find_vec!(args),
                 cme,
                 name,
@@ -2958,7 +2960,7 @@ impl Function {
             send: Insn,
             send_insn_id: InsnId,
         ) -> Result<(), ()> {
-            let Insn::Send { mut recv, cd, blockiseq, mut args, state, .. } = send else {
+            let Insn::Send { mut recv, cd, blockiseq, args, state, .. } = send else {
                 return Err(());
             };
 
@@ -3027,14 +3029,13 @@ impl Function {
 
                     // Emit a call
                     let cfunc = unsafe { get_mct_func(cfunc) }.cast();
-                    let mut cfunc_args = vec![recv];
-                    cfunc_args.append(&mut args);
 
                     let name = rust_str_to_id(&qualified_method_name(unsafe { (*cme).owner }, unsafe { (*cme).called_id }));
                     let ccall = fun.push_insn(block, Insn::CCallWithFrame {
                         cd,
                         cfunc,
-                        args: cfunc_args,
+                        recv,
+                        args,
                         cme,
                         name,
                         state,
@@ -3068,7 +3069,7 @@ impl Function {
             send: Insn,
             send_insn_id: InsnId,
         ) -> Result<(), ()> {
-            let Insn::SendWithoutBlock { mut recv, cd, mut args, state, .. } = send else {
+            let Insn::SendWithoutBlock { mut recv, cd, args, state, .. } = send else {
                 return Err(());
             };
 
@@ -3161,14 +3162,12 @@ impl Function {
                     // No inlining; emit a call
                     let cfunc = unsafe { get_mct_func(cfunc) }.cast();
                     let name = rust_str_to_id(&qualified_method_name(unsafe { (*cme).owner }, unsafe { (*cme).called_id }));
-                    let mut cfunc_args = vec![recv];
-                    cfunc_args.append(&mut args);
                     let return_type = props.return_type;
                     let elidable = props.elidable;
                     // Filter for a leaf and GC free function
                     if props.leaf && props.no_gc {
                         fun.push_insn(block, Insn::IncrCounter(Counter::inline_cfunc_optimized_send_count));
-                        let ccall = fun.push_insn(block, Insn::CCall { cfunc, args: cfunc_args, name, return_type, elidable });
+                        let ccall = fun.push_insn(block, Insn::CCall { cfunc, recv, args, name, return_type, elidable });
                         fun.make_equal_to(send_insn_id, ccall);
                     } else {
                         if get_option!(stats) {
@@ -3177,7 +3176,8 @@ impl Function {
                         let ccall = fun.push_insn(block, Insn::CCallWithFrame {
                             cd,
                             cfunc,
-                            args: cfunc_args,
+                            recv,
+                            args,
                             cme,
                             name,
                             state,
@@ -3631,19 +3631,22 @@ impl Function {
             | &Insn::SendForward { recv, ref args, state, .. }
             | &Insn::SendWithoutBlock { recv, ref args, state, .. }
             | &Insn::CCallVariadic { recv, ref args, state, .. }
+            | &Insn::CCallWithFrame { recv, ref args, state, .. }
             | &Insn::SendWithoutBlockDirect { recv, ref args, state, .. }
             | &Insn::InvokeSuper { recv, ref args, state, .. } => {
                 worklist.push_back(recv);
                 worklist.extend(args);
                 worklist.push_back(state);
             }
-            &Insn::CCallWithFrame { ref args, state, .. }
-            | &Insn::InvokeBuiltin { ref args, state, .. }
+            &Insn::InvokeBuiltin { ref args, state, .. }
             | &Insn::InvokeBlock { ref args, state, .. } => {
                 worklist.extend(args);
                 worklist.push_back(state)
             }
-            Insn::CCall { args, .. } => worklist.extend(args),
+            &Insn::CCall { recv, ref args, .. } => {
+                worklist.push_back(recv);
+                worklist.extend(args);
+            }
             &Insn::GetIvar { self_val, state, .. } | &Insn::DefinedIvar { self_val, state, .. } => {
                 worklist.push_back(self_val);
                 worklist.push_back(state);
@@ -4197,7 +4200,6 @@ impl Function {
             | Insn::IncrCounter { .. }
             | Insn::IncrCounterPtr { .. }
             | Insn::CheckInterrupts { .. }
-            | Insn::CCall { .. }
             | Insn::GetClassVar { .. }
             | Insn::GetSpecialNumber { .. }
             | Insn::GetSpecialSymbol { .. }
@@ -4245,6 +4247,8 @@ impl Function {
             | Insn::Send { recv, ref args, .. }
             | Insn::SendForward { recv, ref args, .. }
             | Insn::InvokeSuper { recv, ref args, .. }
+            | Insn::CCall { recv, ref args, .. }
+            | Insn::CCallWithFrame { recv, ref args, .. }
             | Insn::CCallVariadic { recv, ref args, .. }
             | Insn::ArrayInclude { target: recv, elements: ref args, .. } => {
                 self.assert_subtype(insn_id, recv, types::BasicObject)?;
@@ -4254,8 +4258,7 @@ impl Function {
                 Ok(())
             }
             // Instructions with a Vec of Ruby objects
-            Insn::CCallWithFrame { ref args, .. }
-            | Insn::InvokeBuiltin { ref args, .. }
+            Insn::InvokeBuiltin { ref args, .. }
             | Insn::InvokeBlock { ref args, .. }
             | Insn::NewArray { elements: ref args, .. }
             | Insn::ArrayHash { elements: ref args, .. }