YJIT: Fallback send instructions to vm_sendish (#8106)

test/ruby/test_yjit.rb

@@ -548,7 +548,7 @@ class TestYJIT < Test::Unit::TestCase
 
   def test_getblockparamproxy
     # Currently two side exits as OPTIMIZED_METHOD_TYPE_CALL is unimplemented
-    assert_compiles(<<~'RUBY', insns: [:getblockparamproxy], exits: { opt_send_without_block: 2 })
+    assert_compiles(<<~'RUBY', insns: [:getblockparamproxy])
       def foo &blk
         p blk.call
         p blk.call
@@ -607,7 +607,7 @@ class TestYJIT < Test::Unit::TestCase
 
   def test_send_kwargs
     # For now, this side-exits when calls include keyword args
-    assert_compiles(<<~'RUBY', result: "2#a:1,b:2/A", exits: {opt_send_without_block: 1})
+    assert_compiles(<<~'RUBY', result: "2#a:1,b:2/A")
       def internal_method(**kw)
         "#{kw.size}##{kw.keys.map { |k| "#{k}:#{kw[k]}" }.join(",")}"
       end
@@ -647,7 +647,7 @@ class TestYJIT < Test::Unit::TestCase
 
   def test_send_kwargs_splat
     # For now, this side-exits when calling with a splat
-    assert_compiles(<<~'RUBY', result: "2#a:1,b:2/B", exits: {opt_send_without_block: 1})
+    assert_compiles(<<~'RUBY', result: "2#a:1,b:2/B")
       def internal_method(**kw)
         "#{kw.size}##{kw.keys.map { |k| "#{k}:#{kw[k]}" }.join(",")}"
       end

vm_exec.h

@@ -169,10 +169,20 @@ default:                        \
 #define THROW_EXCEPTION(exc) return (VALUE)(exc)
 #endif
 
+// Run the interpreter from the JIT
+#define VM_EXEC(ec, val) do { \
+    if (val == Qundef) { \
+        VM_ENV_FLAGS_SET(ec->cfp->ep, VM_FRAME_FLAG_FINISH); \
+        val = vm_exec(ec); \
+    } \
+} while (0)
+
+// Run the JIT from the interpreter
 #define JIT_EXEC(ec, val) do { \
     rb_jit_func_t func; \
     if (val == Qundef && (func = jit_compile(ec))) { \
         val = func(ec, ec->cfp); \
+        RESTORE_REGS(); /* fix cfp for tailcall */ \
         if (ec->tag->state) THROW_EXCEPTION(val); \
     } \
 } while (0)

vm_insnhelper.c

@@ -5528,6 +5528,42 @@ vm_sendish(
     return val;
 }
 
+VALUE
+rb_vm_send(rb_execution_context_t *ec, rb_control_frame_t *reg_cfp, CALL_DATA cd, ISEQ blockiseq)
+{
+    VALUE bh = vm_caller_setup_arg_block(ec, GET_CFP(), cd->ci, blockiseq, false);
+    VALUE val = vm_sendish(ec, GET_CFP(), cd, bh, mexp_search_method);
+    VM_EXEC(ec, val);
+    return val;
+}
+
+VALUE
+rb_vm_opt_send_without_block(rb_execution_context_t *ec, rb_control_frame_t *reg_cfp, CALL_DATA cd)
+{
+    VALUE bh = VM_BLOCK_HANDLER_NONE;
+    VALUE val = vm_sendish(ec, GET_CFP(), cd, bh, mexp_search_method);
+    VM_EXEC(ec, val);
+    return val;
+}
+
+VALUE
+rb_vm_invokesuper(rb_execution_context_t *ec, rb_control_frame_t *reg_cfp, CALL_DATA cd, ISEQ blockiseq)
+{
+    VALUE bh = vm_caller_setup_arg_block(ec, GET_CFP(), cd->ci, blockiseq, true);
+    VALUE val = vm_sendish(ec, GET_CFP(), cd, bh, mexp_search_super);
+    VM_EXEC(ec, val);
+    return val;
+}
+
+VALUE
+rb_vm_invokeblock(rb_execution_context_t *ec, rb_control_frame_t *reg_cfp, CALL_DATA cd)
+{
+    VALUE bh = VM_BLOCK_HANDLER_NONE;
+    VALUE val = vm_sendish(ec, GET_CFP(), cd, bh, mexp_search_invokeblock);
+    VM_EXEC(ec, val);
+    return val;
+}
+
 /* object.c */
 VALUE rb_nil_to_s(VALUE);
 VALUE rb_true_to_s(VALUE);

yjit.c

@@ -1122,6 +1122,20 @@ rb_yjit_assert_holding_vm_lock(void)
     ASSERT_vm_locking();
 }
 
+// The number of stack slots that vm_sendish() pops for send and invokesuper.
+size_t
+rb_yjit_sendish_sp_pops(const struct rb_callinfo *ci)
+{
+    return 1 - sp_inc_of_sendish(ci); // + 1 to ignore return value push
+}
+
+// The number of stack slots that vm_sendish() pops for invokeblock.
+size_t
+rb_yjit_invokeblock_sp_pops(const struct rb_callinfo *ci)
+{
+    return 1 - sp_inc_of_invokeblock(ci); // + 1 to ignore return value push
+}
+
 // Primitives used by yjit.rb
 VALUE rb_yjit_stats_enabled_p(rb_execution_context_t *ec, VALUE self);
 VALUE rb_yjit_trace_exit_locations_enabled_p(rb_execution_context_t *ec, VALUE self);

yjit/bindgen/src/main.rs

@@ -325,6 +325,8 @@ fn main() {
         .allowlist_function("rb_yjit_icache_invalidate")
         .allowlist_function("rb_optimized_call")
         .allowlist_function("rb_yjit_assert_holding_vm_lock")
+        .allowlist_function("rb_yjit_sendish_sp_pops")
+        .allowlist_function("rb_yjit_invokeblock_sp_pops")
 
         // from vm_sync.h
         .allowlist_function("rb_vm_barrier")

yjit/src/codegen.rs

@@ -6428,6 +6428,38 @@ fn gen_struct_aset(
     Some(EndBlock)
 }
 
+// Generate code that calls a method with dynamic dispatch
+fn gen_send_dynamic<F: Fn(&mut Assembler) -> Opnd>(
+    jit: &mut JITState,
+    asm: &mut Assembler,
+    cd: *const rb_call_data,
+    sp_pops: usize,
+    vm_sendish: F,
+) -> Option<CodegenStatus> {
+    // Our frame handling is not compatible with tailcall
+    if unsafe { vm_ci_flag((*cd).ci) } & VM_CALL_TAILCALL != 0 {
+        return None;
+    }
+
+    // Save PC and SP to prepare for dynamic dispatch
+    jit_prepare_routine_call(jit, asm);
+
+    // Pop arguments and a receiver
+    asm.stack_pop(sp_pops);
+
+    // Dispatch a method
+    let ret = vm_sendish(asm);
+
+    // Push the return value
+    let stack_ret = asm.stack_push(Type::Unknown);
+    asm.mov(stack_ret, ret);
+
+    // Fix the interpreter SP deviated by vm_sendish
+    asm.mov(Opnd::mem(64, CFP, RUBY_OFFSET_CFP_SP), SP);
+
+    Some(KeepCompiling)
+}
+
 fn gen_send_general(
     jit: &mut JITState,
     asm: &mut Assembler,
@@ -6909,9 +6941,22 @@ fn gen_opt_send_without_block(
     asm: &mut Assembler,
     ocb: &mut OutlinedCb,
 ) -> Option<CodegenStatus> {
+    // Generate specialized code if possible
     let cd = jit.get_arg(0).as_ptr();
+    if let Some(status) = gen_send_general(jit, asm, ocb, cd, None) {
+        return Some(status);
+    }
 
-    gen_send_general(jit, asm, ocb, cd, None)
+    // Otherwise, fallback to dynamic dispatch using the interpreter's implementation of send
+    gen_send_dynamic(jit, asm, cd, unsafe { rb_yjit_sendish_sp_pops((*cd).ci) }, |asm| {
+        extern "C" {
+            fn rb_vm_opt_send_without_block(ec: EcPtr, cfp: CfpPtr, cd: VALUE) -> VALUE;
+        }
+        asm.ccall(
+            rb_vm_opt_send_without_block as *const u8,
+            vec![EC, CFP, (cd as usize).into()],
+        )
+    })
 }
 
 fn gen_send(
@@ -6919,15 +6964,54 @@ fn gen_send(
     asm: &mut Assembler,
     ocb: &mut OutlinedCb,
 ) -> Option<CodegenStatus> {
+    // Generate specialized code if possible
     let cd = jit.get_arg(0).as_ptr();
     let block = jit.get_arg(1).as_optional_ptr();
-    return gen_send_general(jit, asm, ocb, cd, block);
+    if let Some(status) = gen_send_general(jit, asm, ocb, cd, block) {
+        return Some(status);
+    }
+
+    // Otherwise, fallback to dynamic dispatch using the interpreter's implementation of send
+    let blockiseq = jit.get_arg(1).as_iseq();
+    gen_send_dynamic(jit, asm, cd, unsafe { rb_yjit_sendish_sp_pops((*cd).ci) }, |asm| {
+        extern "C" {
+            fn rb_vm_send(ec: EcPtr, cfp: CfpPtr, cd: VALUE, blockiseq: IseqPtr) -> VALUE;
+        }
+        asm.ccall(
+            rb_vm_send as *const u8,
+            vec![EC, CFP, (cd as usize).into(), VALUE(blockiseq as usize).into()],
+        )
+    })
 }
 
 fn gen_invokeblock(
     jit: &mut JITState,
     asm: &mut Assembler,
     ocb: &mut OutlinedCb,
+) -> Option<CodegenStatus> {
+    // Generate specialized code if possible
+    let cd = jit.get_arg(0).as_ptr();
+    if let Some(status) = gen_invokeblock_specialized(jit, asm, ocb, cd) {
+        return Some(status);
+    }
+
+    // Otherwise, fallback to dynamic dispatch using the interpreter's implementation of send
+    gen_send_dynamic(jit, asm, cd, unsafe { rb_yjit_invokeblock_sp_pops((*cd).ci) }, |asm| {
+        extern "C" {
+            fn rb_vm_invokeblock(ec: EcPtr, cfp: CfpPtr, cd: VALUE) -> VALUE;
+        }
+        asm.ccall(
+            rb_vm_invokeblock as *const u8,
+            vec![EC, CFP, (cd as usize).into()],
+        )
+    })
+}
+
+fn gen_invokeblock_specialized(
+    jit: &mut JITState,
+    asm: &mut Assembler,
+    ocb: &mut OutlinedCb,
+    cd: *const rb_call_data,
 ) -> Option<CodegenStatus> {
     if !jit.at_current_insn() {
         defer_compilation(jit, asm, ocb);
@@ -6935,7 +7019,6 @@ fn gen_invokeblock(
     }
 
     // Get call info
-    let cd = jit.get_arg(0).as_ptr();
     let ci = unsafe { get_call_data_ci(cd) };
     let argc: i32 = unsafe { vm_ci_argc(ci) }.try_into().unwrap();
     let flags = unsafe { vm_ci_flag(ci) };
@@ -7065,7 +7148,31 @@ fn gen_invokesuper(
     asm: &mut Assembler,
     ocb: &mut OutlinedCb,
 ) -> Option<CodegenStatus> {
-    let cd: *const rb_call_data = jit.get_arg(0).as_ptr();
+    // Generate specialized code if possible
+    let cd = jit.get_arg(0).as_ptr();
+    if let Some(status) = gen_invokesuper_specialized(jit, asm, ocb, cd) {
+        return Some(status);
+    }
+
+    // Otherwise, fallback to dynamic dispatch using the interpreter's implementation of send
+    let blockiseq = jit.get_arg(1).as_iseq();
+    gen_send_dynamic(jit, asm, cd, unsafe { rb_yjit_sendish_sp_pops((*cd).ci) }, |asm| {
+        extern "C" {
+            fn rb_vm_invokesuper(ec: EcPtr, cfp: CfpPtr, cd: VALUE, blockiseq: IseqPtr) -> VALUE;
+        }
+        asm.ccall(
+            rb_vm_invokesuper as *const u8,
+            vec![EC, CFP, (cd as usize).into(), VALUE(blockiseq as usize).into()],
+        )
+    })
+}
+
+fn gen_invokesuper_specialized(
+    jit: &mut JITState,
+    asm: &mut Assembler,
+    ocb: &mut OutlinedCb,
+    cd: *const rb_call_data,
+) -> Option<CodegenStatus> {
     let block: Option<IseqPtr> = jit.get_arg(1).as_optional_ptr();
 
     // Defer compilation so we can specialize on class of receiver

yjit/src/cruby_bindings.inc.rs

@@ -1336,4 +1336,6 @@ extern "C" {
         line: ::std::os::raw::c_int,
     );
     pub fn rb_yjit_assert_holding_vm_lock();
+    pub fn rb_yjit_sendish_sp_pops(ci: *const rb_callinfo) -> usize;
+    pub fn rb_yjit_invokeblock_sp_pops(ci: *const rb_callinfo) -> usize;
 }