* YJIT: Support invokeblock

* Update yjit/src/backend/arm64/mod.rs

* Update yjit/src/codegen.rs

Co-authored-by: Maxime Chevalier-Boisvert <maximechevalierb@gmail.com>
This commit is contained in:
Takashi Kokubun 2022-11-02 09:30:48 -07:00 коммит произвёл GitHub
Родитель ee7c031dc4
Коммит 81e84e0a4d
Не найден ключ, соответствующий данной подписи
Идентификатор ключа GPG: 4AEE18F83AFDEB23
10 изменённых файлов: 198 добавлений и 30 удалений

14
yjit.c
Просмотреть файл

@ -626,6 +626,12 @@ rb_get_iseq_body_stack_max(const rb_iseq_t *iseq)
return iseq->body->stack_max;
}
bool
rb_get_iseq_flags_has_lead(const rb_iseq_t *iseq)
{
return iseq->body->param.flags.has_lead;
}
bool
rb_get_iseq_flags_has_opt(const rb_iseq_t *iseq)
{
@ -669,7 +675,13 @@ rb_get_iseq_flags_has_block(const rb_iseq_t *iseq)
}
bool
rb_get_iseq_flags_has_accepts_no_kwarg(const rb_iseq_t *iseq)
rb_get_iseq_flags_ambiguous_param0(const rb_iseq_t *iseq)
{
return iseq->body->param.flags.ambiguous_param0;
}
bool
rb_get_iseq_flags_accepts_no_kwarg(const rb_iseq_t *iseq)
{
return iseq->body->param.flags.accepts_no_kwarg;
}

Просмотреть файл

@ -187,6 +187,7 @@ module RubyVM::YJIT
$stderr.puts("***YJIT: Printing YJIT statistics on exit***")
print_counters(stats, prefix: 'send_', prompt: 'method call exit reasons: ')
print_counters(stats, prefix: 'invokeblock_', prompt: 'invokeblock exit reasons: ')
print_counters(stats, prefix: 'invokesuper_', prompt: 'invokesuper exit reasons: ')
print_counters(stats, prefix: 'leave_', prompt: 'leave exit reasons: ')
print_counters(stats, prefix: 'gbpp_', prompt: 'getblockparamproxy exit reasons: ')

Просмотреть файл

@ -250,6 +250,7 @@ fn main() {
.blocklist_type("rb_control_frame_struct")
.opaque_type("rb_control_frame_struct")
.allowlist_function("rb_vm_bh_to_procval")
.allowlist_function("rb_vm_ep_local_ep")
.allowlist_type("vm_special_object_type")
.allowlist_var("VM_ENV_DATA_INDEX_SPECVAL")
.allowlist_var("VM_ENV_DATA_INDEX_FLAGS")
@ -353,13 +354,15 @@ fn main() {
.allowlist_function("rb_get_iseq_body_parent_iseq")
.allowlist_function("rb_get_iseq_body_iseq_encoded")
.allowlist_function("rb_get_iseq_body_stack_max")
.allowlist_function("rb_get_iseq_flags_has_lead")
.allowlist_function("rb_get_iseq_flags_has_opt")
.allowlist_function("rb_get_iseq_flags_has_kw")
.allowlist_function("rb_get_iseq_flags_has_rest")
.allowlist_function("rb_get_iseq_flags_has_post")
.allowlist_function("rb_get_iseq_flags_has_kwrest")
.allowlist_function("rb_get_iseq_flags_has_block")
.allowlist_function("rb_get_iseq_flags_has_accepts_no_kwarg")
.allowlist_function("rb_get_iseq_flags_ambiguous_param0")
.allowlist_function("rb_get_iseq_flags_accepts_no_kwarg")
.allowlist_function("rb_get_iseq_flags_ruby2_keywords")
.allowlist_function("rb_get_iseq_body_local_table_size")
.allowlist_function("rb_get_iseq_body_param_keyword")

Просмотреть файл

@ -165,8 +165,8 @@ impl Assembler
Opnd::Reg(_) | Opnd::InsnOut { .. } => opnd,
Opnd::Mem(_) => split_load_operand(asm, opnd),
Opnd::Imm(imm) => {
if imm <= 0 {
asm.load(opnd)
if imm == 0 {
Opnd::Reg(XZR_REG)
} else if (dest_num_bits == 64 &&
BitmaskImmediate::try_from(imm as u64).is_ok()) ||
(dest_num_bits == 32 &&
@ -1352,8 +1352,8 @@ mod tests {
asm.test(Opnd::Reg(X0_REG), Opnd::Imm(-7));
asm.compile_with_num_regs(&mut cb, 1);
// Assert that a load and a test instruction were written.
assert_eq!(8, cb.get_write_pos());
// Assert that a test instruction is written.
assert_eq!(4, cb.get_write_pos());
}
#[test]

Просмотреть файл

@ -93,6 +93,7 @@ impl Assembler
vec![
RAX_REG,
RCX_REG,
RDX_REG,
]
}

Просмотреть файл

@ -4011,6 +4011,7 @@ enum SpecVal {
BlockISeq(IseqPtr),
BlockParamProxy,
PrevEP(*const VALUE),
PrevEPOpnd(Opnd),
}
struct ControlFrame {
@ -4050,17 +4051,6 @@ fn gen_push_frame(
let sp = frame.sp;
let num_locals = frame.local_size;
if num_locals > 0 {
asm.comment("initialize locals");
// Initialize local variables to Qnil
for i in 0..num_locals {
let offs = (SIZEOF_VALUE as i32) * (i - num_locals - 3);
asm.store(Opnd::mem(64, sp, offs), Qnil.into());
}
}
asm.comment("push cme, specval, frame type");
// Write method entry at sp[-3]
@ -4099,9 +4089,25 @@ fn gen_push_frame(
let tagged_prev_ep = (prev_ep as usize) | 1;
VALUE(tagged_prev_ep).into()
}
SpecVal::PrevEPOpnd(ep_opnd) => {
asm.or(ep_opnd, 1.into())
},
};
asm.store(Opnd::mem(64, sp, SIZEOF_VALUE_I32 * -2), specval);
// Arm requires another register to load the immediate value of Qnil before storing it.
// So doing this after releasing the register for specval to avoid register spill.
let num_locals = frame.local_size;
if num_locals > 0 {
asm.comment("initialize locals");
// Initialize local variables to Qnil
for i in 0..num_locals {
let offs = (SIZEOF_VALUE as i32) * (i - num_locals - 3);
asm.store(Opnd::mem(64, sp, offs), Qnil.into());
}
}
// Write env flags at sp[-1]
// sp[-1] = frame_type;
asm.store(Opnd::mem(64, sp, SIZEOF_VALUE_I32 * -1), frame.frame_type.into());
@ -4522,7 +4528,7 @@ fn gen_send_bmethod(
}
let frame_type = VM_FRAME_MAGIC_BLOCK | VM_FRAME_FLAG_BMETHOD | VM_FRAME_FLAG_LAMBDA;
gen_send_iseq(jit, ctx, asm, ocb, iseq, ci, frame_type, Some(capture.ep), cme, block, flags, argc)
gen_send_iseq(jit, ctx, asm, ocb, iseq, ci, frame_type, Some(capture.ep), cme, block, flags, argc, None)
}
fn gen_send_iseq(
@ -4538,10 +4544,10 @@ fn gen_send_iseq(
block: Option<IseqPtr>,
flags: u32,
argc: i32,
captured_opnd: Option<Opnd>,
) -> CodegenStatus {
let mut argc = argc;
// Create a side-exit to fall back to the interpreter
let side_exit = get_side_exit(jit, ocb, ctx);
@ -4592,7 +4598,7 @@ fn gen_send_iseq(
// If we have a method accepting no kwargs (**nil), exit if we have passed
// it any kwargs.
if supplying_kws && unsafe { get_iseq_flags_has_accepts_no_kwarg(iseq) } {
if supplying_kws && unsafe { get_iseq_flags_accepts_no_kwarg(iseq) } {
gen_counter_incr!(asm, send_iseq_complex_callee);
return CantCompile;
}
@ -4997,12 +5003,17 @@ fn gen_send_iseq(
asm.mov(ctx.stack_opnd(-1), unspec_opnd.into());
}
// Points to the receiver operand on the stack
let recv = ctx.stack_opnd(argc);
// Points to the receiver operand on the stack unless a captured environment is used
let recv = match captured_opnd {
Some(captured_opnd) => asm.load(Opnd::mem(64, captured_opnd, 0)), // captured->self
_ => ctx.stack_opnd(argc),
};
let captured_self = captured_opnd.is_some();
let sp_offset = (argc as isize) + if captured_self { 0 } else { 1 };
// Store the updated SP on the current frame (pop arguments and receiver)
asm.comment("store caller sp");
let caller_sp = asm.lea(ctx.sp_opnd((SIZEOF_VALUE as isize) * -((argc as isize) + 1)));
let caller_sp = asm.lea(ctx.sp_opnd((SIZEOF_VALUE as isize) * -sp_offset));
asm.store(Opnd::mem(64, CFP, RUBY_OFFSET_CFP_SP), caller_sp);
// Store the next PC in the current frame
@ -5017,6 +5028,9 @@ fn gen_send_iseq(
// We've already side-exited if the callee expects a block, so we
// ignore any supplied block here
SpecVal::PrevEP(prev_ep)
} else if let Some(captured_opnd) = captured_opnd {
let ep_opnd = asm.load(Opnd::mem(64, captured_opnd, SIZEOF_VALUE_I32)); // captured->ep
SpecVal::PrevEPOpnd(ep_opnd)
} else if block_arg_type == Some(Type::BlockParamProxy) {
SpecVal::BlockParamProxy
} else if let Some(block_val) = block {
@ -5058,7 +5072,11 @@ fn gen_send_iseq(
callee_ctx.set_local_type(arg_idx.try_into().unwrap(), arg_type);
}
let recv_type = ctx.get_opnd_type(StackOpnd(argc.try_into().unwrap()));
let recv_type = if captured_self {
ctx.get_opnd_type(CapturedSelfOpnd)
} else {
ctx.get_opnd_type(StackOpnd(argc.try_into().unwrap()))
};
callee_ctx.upgrade_opnd_type(SelfOpnd, recv_type);
// The callee might change locals through Kernel#binding and other means.
@ -5068,7 +5086,7 @@ fn gen_send_iseq(
// After the return, sp_offset will be 1. The codegen for leave writes
// the return value in case of JIT-to-JIT return.
let mut return_ctx = *ctx;
return_ctx.stack_pop((argc + 1).try_into().unwrap());
return_ctx.stack_pop(sp_offset.try_into().unwrap());
return_ctx.stack_push(Type::Unknown);
return_ctx.set_sp_offset(1);
return_ctx.reset_chain_depth();
@ -5317,7 +5335,7 @@ fn gen_send_general(
VM_METHOD_TYPE_ISEQ => {
let iseq = unsafe { get_def_iseq_ptr((*cme).def) };
let frame_type = VM_FRAME_MAGIC_METHOD | VM_ENV_FLAG_LOCAL;
return gen_send_iseq(jit, ctx, asm, ocb, iseq, ci, frame_type, None, cme, block, flags, argc);
return gen_send_iseq(jit, ctx, asm, ocb, iseq, ci, frame_type, None, cme, block, flags, argc, None);
}
VM_METHOD_TYPE_CFUNC => {
return gen_send_cfunc(
@ -5642,6 +5660,95 @@ fn gen_send(
return gen_send_general(jit, ctx, asm, ocb, cd, block);
}
fn gen_invokeblock(
jit: &mut JITState,
ctx: &mut Context,
asm: &mut Assembler,
ocb: &mut OutlinedCb,
) -> CodegenStatus {
if !jit_at_current_insn(jit) {
defer_compilation(jit, ctx, asm, ocb);
return EndBlock;
}
// Get call info
let cd = jit_get_arg(jit, 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) };
// Get block_handler
let cfp = unsafe { get_ec_cfp(jit.ec.unwrap()) };
let lep = unsafe { rb_vm_ep_local_ep(get_cfp_ep(cfp)) };
let comptime_handler = unsafe { *lep.offset(VM_ENV_DATA_INDEX_SPECVAL.try_into().unwrap()) };
// Handle each block_handler type
if comptime_handler.0 == VM_BLOCK_HANDLER_NONE as usize { // no block given
gen_counter_incr!(asm, invokeblock_none);
CantCompile
} else if comptime_handler.0 & 0x3 == 0x1 { // VM_BH_ISEQ_BLOCK_P
asm.comment("get local EP");
let ep_opnd = gen_get_lep(jit, asm);
let block_handler_opnd = asm.load(
Opnd::mem(64, ep_opnd, (SIZEOF_VALUE as i32) * (VM_ENV_DATA_INDEX_SPECVAL as i32))
);
asm.comment("guard block_handler type");
let side_exit = get_side_exit(jit, ocb, ctx);
let tag_opnd = asm.and(block_handler_opnd, 0x3.into()); // block_handler is a tagged pointer
asm.cmp(tag_opnd, 0x1.into()); // VM_BH_ISEQ_BLOCK_P
asm.jne(counted_exit!(ocb, side_exit, invokeblock_iseq_tag_changed).into());
// Not supporting vm_callee_setup_block_arg_arg0_splat for now
let comptime_captured = unsafe { ((comptime_handler.0 & !0x3) as *const rb_captured_block).as_ref().unwrap() };
let comptime_iseq = unsafe { *comptime_captured.code.iseq.as_ref() };
if argc == 1 && unsafe { get_iseq_flags_has_lead(comptime_iseq) && !get_iseq_flags_ambiguous_param0(comptime_iseq) } {
gen_counter_incr!(asm, invokeblock_iseq_arg0_splat);
return CantCompile;
}
asm.comment("guard known ISEQ");
let captured_opnd = asm.and(block_handler_opnd, Opnd::Imm(!0x3));
let iseq_opnd = asm.load(Opnd::mem(64, captured_opnd, SIZEOF_VALUE_I32 * 2));
asm.cmp(iseq_opnd, (comptime_iseq as usize).into());
let block_changed_exit = counted_exit!(ocb, side_exit, invokeblock_iseq_block_changed);
jit_chain_guard(
JCC_JNE,
jit,
ctx,
asm,
ocb,
SEND_MAX_CHAIN_DEPTH,
block_changed_exit,
);
gen_send_iseq(
jit,
ctx,
asm,
ocb,
comptime_iseq,
ci,
VM_FRAME_MAGIC_BLOCK,
None,
0 as _,
None,
flags,
argc,
Some(captured_opnd),
)
} else if comptime_handler.0 & 0x3 == 0x3 { // VM_BH_IFUNC_P
gen_counter_incr!(asm, invokeblock_ifunc);
CantCompile
} else if comptime_handler.symbol_p() {
gen_counter_incr!(asm, invokeblock_symbol);
CantCompile
} else { // Proc
gen_counter_incr!(asm, invokeblock_proc);
CantCompile
}
}
fn gen_invokesuper(
jit: &mut JITState,
ctx: &mut Context,
@ -5781,7 +5888,7 @@ fn gen_invokesuper(
VM_METHOD_TYPE_ISEQ => {
let iseq = unsafe { get_def_iseq_ptr((*cme).def) };
let frame_type = VM_FRAME_MAGIC_METHOD | VM_ENV_FLAG_LOCAL;
gen_send_iseq(jit, ctx, asm, ocb, iseq, ci, frame_type, None, cme, block, ci_flags, argc)
gen_send_iseq(jit, ctx, asm, ocb, iseq, ci, frame_type, None, cme, block, ci_flags, argc, None)
}
VM_METHOD_TYPE_CFUNC => {
gen_send_cfunc(jit, ctx, asm, ocb, ci, cme, block, ptr::null(), ci_flags, argc)
@ -6548,6 +6655,7 @@ fn get_gen_fn(opcode: VALUE) -> Option<InsnGenFn> {
YARVINSN_getblockparam => Some(gen_getblockparam),
YARVINSN_opt_send_without_block => Some(gen_opt_send_without_block),
YARVINSN_send => Some(gen_send),
YARVINSN_invokeblock => Some(gen_invokeblock),
YARVINSN_invokesuper => Some(gen_invokesuper),
YARVINSN_leave => Some(gen_leave),

Просмотреть файл

@ -269,6 +269,9 @@ pub enum InsnOpnd {
// The value is self
SelfOpnd,
// Captured block's self
CapturedSelfOpnd,
// Temporary stack operand with stack index
StackOpnd(u16),
}
@ -297,6 +300,9 @@ pub struct Context {
// Type we track for self
self_type: Type,
// Type we track for captured block's self
captured_self_type: Type,
// Mapping of temp stack entries to types we track
temp_mapping: [TempMapping; MAX_TEMP_TYPES],
}
@ -1160,6 +1166,7 @@ impl Context {
pub fn get_opnd_type(&self, opnd: InsnOpnd) -> Type {
match opnd {
SelfOpnd => self.self_type,
CapturedSelfOpnd => self.captured_self_type,
StackOpnd(idx) => {
let idx = idx as u16;
assert!(idx < self.stack_size);
@ -1201,6 +1208,7 @@ impl Context {
match opnd {
SelfOpnd => self.self_type.upgrade(opnd_type),
CapturedSelfOpnd => self.self_type.upgrade(opnd_type),
StackOpnd(idx) => {
let idx = idx as u16;
assert!(idx < self.stack_size);
@ -1236,6 +1244,7 @@ impl Context {
match opnd {
SelfOpnd => (MapToSelf, opnd_type),
CapturedSelfOpnd => unreachable!("not used for captured self"),
StackOpnd(idx) => {
let idx = idx as u16;
assert!(idx < self.stack_size);
@ -1257,6 +1266,7 @@ impl Context {
pub fn set_opnd_mapping(&mut self, opnd: InsnOpnd, (mapping, opnd_type): (TempMapping, Type)) {
match opnd {
SelfOpnd => unreachable!("self always maps to self"),
CapturedSelfOpnd => unreachable!("not used for captured self"),
StackOpnd(idx) => {
assert!(idx < self.stack_size);
let stack_idx = (self.stack_size - 1 - idx) as usize;

Просмотреть файл

@ -162,6 +162,7 @@ pub use rb_iseq_encoded_size as get_iseq_encoded_size;
pub use rb_get_iseq_body_local_iseq as get_iseq_body_local_iseq;
pub use rb_get_iseq_body_iseq_encoded as get_iseq_body_iseq_encoded;
pub use rb_get_iseq_body_stack_max as get_iseq_body_stack_max;
pub use rb_get_iseq_flags_has_lead as get_iseq_flags_has_lead;
pub use rb_get_iseq_flags_has_opt as get_iseq_flags_has_opt;
pub use rb_get_iseq_flags_has_kw as get_iseq_flags_has_kw;
pub use rb_get_iseq_flags_has_rest as get_iseq_flags_has_rest;
@ -169,7 +170,8 @@ pub use rb_get_iseq_flags_ruby2_keywords as get_iseq_flags_ruby2_keywords;
pub use rb_get_iseq_flags_has_post as get_iseq_flags_has_post;
pub use rb_get_iseq_flags_has_kwrest as get_iseq_flags_has_kwrest;
pub use rb_get_iseq_flags_has_block as get_iseq_flags_has_block;
pub use rb_get_iseq_flags_has_accepts_no_kwarg as get_iseq_flags_has_accepts_no_kwarg;
pub use rb_get_iseq_flags_ambiguous_param0 as get_iseq_flags_ambiguous_param0;
pub use rb_get_iseq_flags_accepts_no_kwarg as get_iseq_flags_accepts_no_kwarg;
pub use rb_get_iseq_body_local_table_size as get_iseq_body_local_table_size;
pub use rb_get_iseq_body_param_keyword as get_iseq_body_param_keyword;
pub use rb_get_iseq_body_param_size as get_iseq_body_param_size;
@ -346,13 +348,27 @@ impl VALUE {
(cval & mask) == flag
}
/// Return true for a static (non-heap) Ruby symbol
/// Return true if the value is a Ruby symbol (RB_SYMBOL_P)
pub fn symbol_p(self) -> bool {
self.static_sym_p() || self.dynamic_sym_p()
}
/// Return true for a static (non-heap) Ruby symbol (RB_STATIC_SYM_P)
pub fn static_sym_p(self) -> bool {
let VALUE(cval) = self;
let flag = RUBY_SYMBOL_FLAG as usize;
(cval & 0xff) == flag
}
/// Return true for a dynamic Ruby symbol (RB_DYNAMIC_SYM_P)
fn dynamic_sym_p(self) -> bool {
return if self.special_const_p() {
false
} else {
self.builtin_type() == RUBY_T_SYMBOL
}
}
/// Returns true or false depending on whether the value is nil
pub fn nil_p(self) -> bool {
self == Qnil

Просмотреть файл

@ -965,6 +965,9 @@ pub const VM_ENV_FLAG_ESCAPED: vm_frame_env_flags = 4;
pub const VM_ENV_FLAG_WB_REQUIRED: vm_frame_env_flags = 8;
pub const VM_ENV_FLAG_ISOLATED: vm_frame_env_flags = 16;
pub type vm_frame_env_flags = u32;
extern "C" {
pub fn rb_vm_ep_local_ep(ep: *const VALUE) -> *const VALUE;
}
extern "C" {
pub fn rb_iseq_path(iseq: *const rb_iseq_t) -> VALUE;
}
@ -1430,6 +1433,9 @@ extern "C" {
extern "C" {
pub fn rb_get_iseq_body_stack_max(iseq: *const rb_iseq_t) -> ::std::os::raw::c_uint;
}
extern "C" {
pub fn rb_get_iseq_flags_has_lead(iseq: *const rb_iseq_t) -> bool;
}
extern "C" {
pub fn rb_get_iseq_flags_has_opt(iseq: *const rb_iseq_t) -> bool;
}
@ -1452,7 +1458,10 @@ extern "C" {
pub fn rb_get_iseq_flags_has_block(iseq: *const rb_iseq_t) -> bool;
}
extern "C" {
pub fn rb_get_iseq_flags_has_accepts_no_kwarg(iseq: *const rb_iseq_t) -> bool;
pub fn rb_get_iseq_flags_ambiguous_param0(iseq: *const rb_iseq_t) -> bool;
}
extern "C" {
pub fn rb_get_iseq_flags_accepts_no_kwarg(iseq: *const rb_iseq_t) -> bool;
}
extern "C" {
pub fn rb_get_iseq_body_param_keyword(

Просмотреть файл

@ -217,6 +217,14 @@ make_counters! {
invokesuper_me_changed,
invokesuper_block,
invokeblock_none,
invokeblock_iseq_arg0_splat,
invokeblock_iseq_block_changed,
invokeblock_iseq_tag_changed,
invokeblock_ifunc,
invokeblock_proc,
invokeblock_symbol,
leave_se_interrupt,
leave_interp_return,
leave_start_pc_non_zero,