diff options
Diffstat (limited to 'yjit')
-rw-r--r-- | yjit/bindgen/src/main.rs | 2 | ||||
-rw-r--r-- | yjit/src/codegen.rs | 94 | ||||
-rw-r--r-- | yjit/src/cruby_bindings.inc.rs | 246 | ||||
-rw-r--r-- | yjit/src/stats.rs | 3 |
4 files changed, 324 insertions, 21 deletions
diff --git a/yjit/bindgen/src/main.rs b/yjit/bindgen/src/main.rs index c3d4a39a2b..ebeeab14b1 100644 --- a/yjit/bindgen/src/main.rs +++ b/yjit/bindgen/src/main.rs @@ -273,6 +273,7 @@ fn main() { .allowlist_function("rb_RSTRING_PTR") .allowlist_function("rb_RSTRING_LEN") .allowlist_function("rb_ENCODING_GET") + .allowlist_function("rb_yjit_get_proc_ptr") .allowlist_function("rb_yjit_exit_locations_dict") .allowlist_function("rb_yjit_icache_invalidate") @@ -332,6 +333,7 @@ fn main() { .allowlist_function("rb_get_mct_argc") .allowlist_function("rb_get_mct_func") .allowlist_function("rb_get_def_iseq_ptr") + .allowlist_function("rb_get_def_bmethod_proc") .allowlist_function("rb_iseq_encoded_size") .allowlist_function("rb_get_iseq_body_local_iseq") .allowlist_function("rb_get_iseq_body_iseq_encoded") diff --git a/yjit/src/codegen.rs b/yjit/src/codegen.rs index 42ec05aafc..22634a7de0 100644 --- a/yjit/src/codegen.rs +++ b/yjit/src/codegen.rs @@ -4035,6 +4035,7 @@ struct ControlFrame { pc: Option<u64>, frame_type: u32, block_handler: BlockHandler, + prev_ep: Option<*const VALUE>, cme: *const rb_callable_method_entry_t, local_size: i32 } @@ -4051,14 +4052,14 @@ struct ControlFrame { // * Provided sp should point to the new frame's sp, immediately following locals and the environment // * At entry, CFP points to the caller (not callee) frame // * At exit, ec->cfp is updated to the pushed CFP -// * CFP and SP registers are updated only if switch_in_jit is set +// * CFP and SP registers are updated only if set_sp_cfp is set // * Stack overflow is not checked (should be done by the caller) // * Interrupts are not checked (should be done by the caller) fn gen_push_frame( _jit: &mut JITState, _ctx: &mut Context, asm: &mut Assembler, - set_pc_cfp: bool, // if true CFP and SP will be switched to the callee + set_sp_cfp: bool, // if true CFP and SP will be switched to the callee frame: ControlFrame, ) { assert!(frame.local_size >= 0); @@ -4076,7 +4077,7 @@ fn gen_push_frame( } } - asm.comment("push cme, block handler, frame type"); + asm.comment("push cme, specval, frame type"); // Write method entry at sp[-3] // sp[-3] = me; @@ -4084,18 +4085,24 @@ fn gen_push_frame( // any cme we depend on become outdated. See yjit_method_lookup_change(). asm.store(Opnd::mem(64, sp, SIZEOF_VALUE_I32 * -3), VALUE::from(frame.cme).into()); - // Write block handler at sp[-2] - // sp[-2] = block_handler; - let block_handler: Opnd = match frame.block_handler { - BlockHandler::None => { + // Write special value at sp[-2]. It's either a block handler or a pointer to + // the outer environment depending on the frame type. + // sp[-2] = specval; + let specval: Opnd = match (frame.prev_ep, frame.block_handler) { + (None, BlockHandler::None) => { VM_BLOCK_HANDLER_NONE.into() - }, - BlockHandler::CurrentFrame => { + } + (None, BlockHandler::CurrentFrame) => { let cfp_self = asm.lea(Opnd::mem(64, CFP, RUBY_OFFSET_CFP_SELF)); asm.or(cfp_self, Opnd::Imm(1)) - }, + } + (Some(prev_ep), BlockHandler::None) => { + let tagged_prev_ep = (prev_ep as usize) | 1; + VALUE(tagged_prev_ep).into() + } + (_, _) => panic!("specval can only be one of prev_ep or block_handler") }; - asm.store(Opnd::mem(64, sp, SIZEOF_VALUE_I32 * -2), block_handler); + asm.store(Opnd::mem(64, sp, SIZEOF_VALUE_I32 * -2), specval); // Write env flags at sp[-1] // sp[-1] = frame_type; @@ -4134,7 +4141,7 @@ fn gen_push_frame( asm.mov(cfp_opnd(RUBY_OFFSET_CFP_SELF), frame.recv); asm.mov(cfp_opnd(RUBY_OFFSET_CFP_BLOCK_CODE), 0.into()); - if set_pc_cfp { + if set_sp_cfp { // Saving SP before calculating ep avoids a dependency on a register // However this must be done after referencing frame.recv, which may be SP-relative asm.mov(SP, sp); @@ -4144,7 +4151,7 @@ fn gen_push_frame( asm.comment("switch to new CFP"); let new_cfp = asm.lea(cfp_opnd(0)); - if set_pc_cfp { + if set_sp_cfp { asm.mov(CFP, new_cfp); asm.store(Opnd::mem(64, EC, RUBY_OFFSET_EC_CFP), CFP); } else { @@ -4274,6 +4281,7 @@ fn gen_send_cfunc( cme, recv, sp, + prev_ep: None, pc: Some(0), iseq: None, local_size: 0, @@ -4443,17 +4451,59 @@ fn push_splat_args(required_args: i32, ctx: &mut Context, asm: &mut Assembler, o } } +fn gen_send_bmethod( + jit: &mut JITState, + ctx: &mut Context, + asm: &mut Assembler, + ocb: &mut OutlinedCb, + ci: *const rb_callinfo, + cme: *const rb_callable_method_entry_t, + block: Option<IseqPtr>, + argc: i32, +) -> CodegenStatus { + let procv = unsafe { rb_get_def_bmethod_proc((*cme).def) }; + + let proc = unsafe { rb_yjit_get_proc_ptr(procv) }; + let proc_block = unsafe { &(*proc).block }; + + if proc_block.type_ != block_type_iseq { + return CantCompile; + } + + let capture = unsafe { proc_block.as_.captured.as_ref() }; + let iseq = unsafe { *capture.code.iseq.as_ref() }; + + // Optimize for single ractor mode and avoid runtime check for + // "defined with an un-shareable Proc in a different Ractor" + if !assume_single_ractor_mode(jit, ocb) { + gen_counter_incr!(asm, send_bmethod_ractor); + return CantCompile; + } + + // Passing a block to a block needs logic different from passing + // a block to a method and sometimes requires allocation. Bail for now. + if block.is_some() { + gen_counter_incr!(asm, send_bmethod_block_arg); + return CantCompile; + } + + 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, argc) +} + fn gen_send_iseq( jit: &mut JITState, ctx: &mut Context, asm: &mut Assembler, ocb: &mut OutlinedCb, + iseq: *const rb_iseq_t, ci: *const rb_callinfo, + frame_type: u32, + prev_ep: Option<*const VALUE>, cme: *const rb_callable_method_entry_t, block: Option<IseqPtr>, argc: i32, ) -> CodegenStatus { - let iseq = unsafe { get_def_iseq_ptr((*cme).def) }; let mut argc = argc; let flags = unsafe { vm_ci_flag(ci) }; @@ -4893,8 +4943,6 @@ fn gen_send_iseq( BlockHandler::None }; - let frame_type = VM_FRAME_MAGIC_METHOD | VM_ENV_FLAG_LOCAL; - // Setup the new frame gen_push_frame(jit, ctx, asm, true, ControlFrame { frame_type, @@ -4902,6 +4950,7 @@ fn gen_send_iseq( cme, recv, sp: callee_sp, + prev_ep, iseq: Some(iseq), pc: None, // We are calling into jitted code, which will set the PC as necessary local_size: num_locals @@ -5173,7 +5222,9 @@ fn gen_send_general( match def_type { VM_METHOD_TYPE_ISEQ => { - return gen_send_iseq(jit, ctx, asm, ocb, ci, cme, block, argc); + 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, argc); } VM_METHOD_TYPE_CFUNC => { return gen_send_cfunc( @@ -5243,8 +5294,7 @@ fn gen_send_general( } // Block method, e.g. define_method(:foo) { :my_block } VM_METHOD_TYPE_BMETHOD => { - gen_counter_incr!(asm, send_bmethod); - return CantCompile; + return gen_send_bmethod(jit, ctx, asm, ocb, ci, cme, block, argc); } VM_METHOD_TYPE_ZSUPER => { gen_counter_incr!(asm, send_zsuper_method); @@ -5481,7 +5531,11 @@ fn gen_invokesuper( ctx.clear_local_types(); match cme_def_type { - VM_METHOD_TYPE_ISEQ => gen_send_iseq(jit, ctx, asm, ocb, ci, cme, block, argc), + 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, argc) + } VM_METHOD_TYPE_CFUNC => { gen_send_cfunc(jit, ctx, asm, ocb, ci, cme, block, argc, ptr::null()) } diff --git a/yjit/src/cruby_bindings.inc.rs b/yjit/src/cruby_bindings.inc.rs index f58bf1ca05..3bffc30731 100644 --- a/yjit/src/cruby_bindings.inc.rs +++ b/yjit/src/cruby_bindings.inc.rs @@ -1,5 +1,128 @@ /* automatically generated by rust-bindgen 0.60.1 */ +#[repr(C)] +#[derive(Copy, Clone, Debug, Default, Eq, Hash, Ord, PartialEq, PartialOrd)] +pub struct __BindgenBitfieldUnit<Storage> { + storage: Storage, +} +impl<Storage> __BindgenBitfieldUnit<Storage> { + #[inline] + pub const fn new(storage: Storage) -> Self { + Self { storage } + } +} +impl<Storage> __BindgenBitfieldUnit<Storage> +where + Storage: AsRef<[u8]> + AsMut<[u8]>, +{ + #[inline] + pub fn get_bit(&self, index: usize) -> bool { + debug_assert!(index / 8 < self.storage.as_ref().len()); + let byte_index = index / 8; + let byte = self.storage.as_ref()[byte_index]; + let bit_index = if cfg!(target_endian = "big") { + 7 - (index % 8) + } else { + index % 8 + }; + let mask = 1 << bit_index; + byte & mask == mask + } + #[inline] + pub fn set_bit(&mut self, index: usize, val: bool) { + debug_assert!(index / 8 < self.storage.as_ref().len()); + let byte_index = index / 8; + let byte = &mut self.storage.as_mut()[byte_index]; + let bit_index = if cfg!(target_endian = "big") { + 7 - (index % 8) + } else { + index % 8 + }; + let mask = 1 << bit_index; + if val { + *byte |= mask; + } else { + *byte &= !mask; + } + } + #[inline] + pub fn get(&self, bit_offset: usize, bit_width: u8) -> u64 { + debug_assert!(bit_width <= 64); + debug_assert!(bit_offset / 8 < self.storage.as_ref().len()); + debug_assert!((bit_offset + (bit_width as usize)) / 8 <= self.storage.as_ref().len()); + let mut val = 0; + for i in 0..(bit_width as usize) { + if self.get_bit(i + bit_offset) { + let index = if cfg!(target_endian = "big") { + bit_width as usize - 1 - i + } else { + i + }; + val |= 1 << index; + } + } + val + } + #[inline] + pub fn set(&mut self, bit_offset: usize, bit_width: u8, val: u64) { + debug_assert!(bit_width <= 64); + debug_assert!(bit_offset / 8 < self.storage.as_ref().len()); + debug_assert!((bit_offset + (bit_width as usize)) / 8 <= self.storage.as_ref().len()); + for i in 0..(bit_width as usize) { + let mask = 1 << i; + let val_bit_is_set = val & mask == mask; + let index = if cfg!(target_endian = "big") { + bit_width as usize - 1 - i + } else { + i + }; + self.set_bit(index + bit_offset, val_bit_is_set); + } + } +} +#[repr(C)] +pub struct __BindgenUnionField<T>(::std::marker::PhantomData<T>); +impl<T> __BindgenUnionField<T> { + #[inline] + pub const fn new() -> Self { + __BindgenUnionField(::std::marker::PhantomData) + } + #[inline] + pub unsafe fn as_ref(&self) -> &T { + ::std::mem::transmute(self) + } + #[inline] + pub unsafe fn as_mut(&mut self) -> &mut T { + ::std::mem::transmute(self) + } +} +impl<T> ::std::default::Default for __BindgenUnionField<T> { + #[inline] + fn default() -> Self { + Self::new() + } +} +impl<T> ::std::clone::Clone for __BindgenUnionField<T> { + #[inline] + fn clone(&self) -> Self { + Self::new() + } +} +impl<T> ::std::marker::Copy for __BindgenUnionField<T> {} +impl<T> ::std::fmt::Debug for __BindgenUnionField<T> { + fn fmt(&self, fmt: &mut ::std::fmt::Formatter<'_>) -> ::std::fmt::Result { + fmt.write_str("__BindgenUnionField") + } +} +impl<T> ::std::hash::Hash for __BindgenUnionField<T> { + fn hash<H: ::std::hash::Hasher>(&self, _state: &mut H) {} +} +impl<T> ::std::cmp::PartialEq for __BindgenUnionField<T> { + fn eq(&self, _other: &__BindgenUnionField<T>) -> bool { + true + } +} +impl<T> ::std::cmp::Eq for __BindgenUnionField<T> {} pub const USE_RVARGC: u32 = 1; pub const INTEGER_REDEFINED_OP_FLAG: u32 = 1; pub const FLOAT_REDEFINED_OP_FLAG: u32 = 2; @@ -157,6 +280,16 @@ extern "C" { extern "C" { pub static mut rb_cTrueClass: VALUE; } +pub type rb_block_call_func = ::std::option::Option< + unsafe extern "C" fn( + yielded_arg: VALUE, + callback_arg: VALUE, + argc: ::std::os::raw::c_int, + argv: *const VALUE, + blockarg: VALUE, + ) -> VALUE, +>; +pub type rb_block_call_func_t = rb_block_call_func; extern "C" { pub fn rb_ary_new_capa(capa: ::std::os::raw::c_long) -> VALUE; } @@ -514,6 +647,20 @@ pub const imemo_callinfo: imemo_type = 11; pub const imemo_callcache: imemo_type = 12; pub const imemo_constcache: imemo_type = 13; pub type imemo_type = u32; +#[repr(C)] +#[derive(Debug, Copy, Clone)] +pub struct vm_ifunc_argc { + pub min: ::std::os::raw::c_int, + pub max: ::std::os::raw::c_int, +} +#[repr(C)] +pub struct vm_ifunc { + pub flags: VALUE, + pub reserved: VALUE, + pub func: rb_block_call_func_t, + pub data: *const ::std::os::raw::c_void, + pub argc: vm_ifunc_argc, +} pub const METHOD_VISI_UNDEF: rb_method_visibility_t = 0; pub const METHOD_VISI_PUBLIC: rb_method_visibility_t = 1; pub const METHOD_VISI_PRIVATE: rb_method_visibility_t = 2; @@ -650,6 +797,36 @@ pub const BOP_AND: ruby_basic_operators = 27; pub const BOP_OR: ruby_basic_operators = 28; pub const BOP_LAST_: ruby_basic_operators = 29; pub type ruby_basic_operators = u32; +#[repr(C)] +pub struct rb_captured_block { + pub self_: VALUE, + pub ep: *const VALUE, + pub code: rb_captured_block__bindgen_ty_1, +} +#[repr(C)] +pub struct rb_captured_block__bindgen_ty_1 { + pub iseq: __BindgenUnionField<*const rb_iseq_t>, + pub ifunc: __BindgenUnionField<*const vm_ifunc>, + pub val: __BindgenUnionField<VALUE>, + pub bindgen_union_field: u64, +} +pub const block_type_iseq: rb_block_type = 0; +pub const block_type_ifunc: rb_block_type = 1; +pub const block_type_symbol: rb_block_type = 2; +pub const block_type_proc: rb_block_type = 3; +pub type rb_block_type = u32; +#[repr(C)] +pub struct rb_block { + pub as_: rb_block__bindgen_ty_1, + pub type_: rb_block_type, +} +#[repr(C)] +pub struct rb_block__bindgen_ty_1 { + pub captured: __BindgenUnionField<rb_captured_block>, + pub symbol: __BindgenUnionField<VALUE>, + pub proc_: __BindgenUnionField<VALUE>, + pub bindgen_union_field: [u64; 3usize], +} pub type rb_control_frame_t = rb_control_frame_struct; extern "C" { pub static mut rb_mRubyVMFrozenCore: VALUE; @@ -657,6 +834,69 @@ extern "C" { extern "C" { pub static mut rb_block_param_proxy: VALUE; } +#[repr(C)] +pub struct rb_proc_t { + pub block: rb_block, + pub _bitfield_align_1: [u8; 0], + pub _bitfield_1: __BindgenBitfieldUnit<[u8; 1usize]>, + pub __bindgen_padding_0: [u8; 7usize], +} +impl rb_proc_t { + #[inline] + pub fn is_from_method(&self) -> ::std::os::raw::c_uint { + unsafe { ::std::mem::transmute(self._bitfield_1.get(0usize, 1u8) as u32) } + } + #[inline] + pub fn set_is_from_method(&mut self, val: ::std::os::raw::c_uint) { + unsafe { + let val: u32 = ::std::mem::transmute(val); + self._bitfield_1.set(0usize, 1u8, val as u64) + } + } + #[inline] + pub fn is_lambda(&self) -> ::std::os::raw::c_uint { + unsafe { ::std::mem::transmute(self._bitfield_1.get(1usize, 1u8) as u32) } + } + #[inline] + pub fn set_is_lambda(&mut self, val: ::std::os::raw::c_uint) { + unsafe { + let val: u32 = ::std::mem::transmute(val); + self._bitfield_1.set(1usize, 1u8, val as u64) + } + } + #[inline] + pub fn is_isolated(&self) -> ::std::os::raw::c_uint { + unsafe { ::std::mem::transmute(self._bitfield_1.get(2usize, 1u8) as u32) } + } + #[inline] + pub fn set_is_isolated(&mut self, val: ::std::os::raw::c_uint) { + unsafe { + let val: u32 = ::std::mem::transmute(val); + self._bitfield_1.set(2usize, 1u8, val as u64) + } + } + #[inline] + pub fn new_bitfield_1( + is_from_method: ::std::os::raw::c_uint, + is_lambda: ::std::os::raw::c_uint, + is_isolated: ::std::os::raw::c_uint, + ) -> __BindgenBitfieldUnit<[u8; 1usize]> { + let mut __bindgen_bitfield_unit: __BindgenBitfieldUnit<[u8; 1usize]> = Default::default(); + __bindgen_bitfield_unit.set(0usize, 1u8, { + let is_from_method: u32 = unsafe { ::std::mem::transmute(is_from_method) }; + is_from_method as u64 + }); + __bindgen_bitfield_unit.set(1usize, 1u8, { + let is_lambda: u32 = unsafe { ::std::mem::transmute(is_lambda) }; + is_lambda as u64 + }); + __bindgen_bitfield_unit.set(2usize, 1u8, { + let is_isolated: u32 = unsafe { ::std::mem::transmute(is_isolated) }; + is_isolated as u64 + }); + __bindgen_bitfield_unit + } +} pub const VM_SPECIAL_OBJECT_VMCORE: vm_special_object_type = 1; pub const VM_SPECIAL_OBJECT_CBASE: vm_special_object_type = 2; pub const VM_SPECIAL_OBJECT_CONST_BASE: vm_special_object_type = 3; @@ -1060,6 +1300,9 @@ extern "C" { extern "C" { pub fn rb_RSTRING_PTR(str_: VALUE) -> *mut ::std::os::raw::c_char; } +extern "C" { + pub fn rb_yjit_get_proc_ptr(procv: VALUE) -> *mut rb_proc_t; +} pub type rb_seq_param_keyword_struct = rb_iseq_constant_body__bindgen_ty_1_rb_iseq_param_keyword; extern "C" { pub fn rb_insn_name(insn: VALUE) -> *const ::std::os::raw::c_char; @@ -1128,6 +1371,9 @@ extern "C" { pub fn rb_get_def_iseq_ptr(def: *mut rb_method_definition_t) -> *const rb_iseq_t; } extern "C" { + pub fn rb_get_def_bmethod_proc(def: *mut rb_method_definition_t) -> VALUE; +} +extern "C" { pub fn rb_get_iseq_body_local_iseq(iseq: *const rb_iseq_t) -> *const rb_iseq_t; } extern "C" { diff --git a/yjit/src/stats.rs b/yjit/src/stats.rs index 7669308f9b..74319ec4ed 100644 --- a/yjit/src/stats.rs +++ b/yjit/src/stats.rs @@ -175,7 +175,6 @@ make_counters! { send_optimized_method_call, send_optimized_method_block_call, send_missing_method, - send_bmethod, send_refined_method, send_cfunc_ruby_array_varg, send_cfunc_argc_mismatch, @@ -198,6 +197,8 @@ make_counters! { send_args_splat_non_iseq, send_args_splat_cfunc, send_iseq_ruby2_keywords, + send_bmethod_ractor, + send_bmethod_block_arg, traced_cfunc_return, |