Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Cranelift: update to latest regalloc2 for better clobber and splitting handling. #4324

Merged
merged 1 commit into from
Jun 28, 2022
Merged

Cranelift: update to latest regalloc2 for better clobber and splitting handling. #4324

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions Cargo.lock
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

2 changes: 1 addition & 1 deletion cranelift/codegen/Cargo.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,7 +23,7 @@ serde = { version = "1.0.94", features = ["derive"], optional = true }
bincode = { version = "1.2.1", optional = true }
gimli = { version = "0.26.0", default-features = false, features = ["write"], optional = true }
smallvec = { version = "1.6.1" }
regalloc2 = { version = "0.2.3", features = ["checker"] }
regalloc2 = { version = "0.3.0", features = ["checker"] }
souper-ir = { version = "2.1.0", optional = true }
# It is a goal of the cranelift-codegen crate to have minimal external dependencies.
# Please don't add any unless they are essential to the task of creating binary
Expand Down
152 changes: 91 additions & 61 deletions cranelift/codegen/src/isa/aarch64/abi.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ use crate::settings;
use crate::{CodegenError, CodegenResult};
use alloc::boxed::Box;
use alloc::vec::Vec;
use regalloc2::VReg;
use regalloc2::{PRegSet, VReg};
use smallvec::{smallvec, SmallVec};

// We use a generic implementation that factors out AArch64 and x64 ABI commonalities, because
Expand Down Expand Up @@ -1062,8 +1062,9 @@ impl ABIMachineSpec for AArch64MachineDeps {

fn gen_call(
dest: &CallDest,
uses: Vec<Reg>,
defs: Vec<Writable<Reg>>,
uses: SmallVec<[Reg; 8]>,
defs: SmallVec<[Writable<Reg>; 8]>,
clobbers: PRegSet,
opcode: ir::Opcode,
tmp: Writable<Reg>,
callee_conv: isa::CallConv,
Expand All @@ -1076,6 +1077,7 @@ impl ABIMachineSpec for AArch64MachineDeps {
dest: name.clone(),
uses,
defs,
clobbers,
opcode,
caller_callconv: caller_conv,
callee_callconv: callee_conv,
Expand All @@ -1092,6 +1094,7 @@ impl ABIMachineSpec for AArch64MachineDeps {
rn: tmp.to_reg(),
uses,
defs,
clobbers,
opcode,
caller_callconv: caller_conv,
callee_callconv: callee_conv,
Expand All @@ -1103,6 +1106,7 @@ impl ABIMachineSpec for AArch64MachineDeps {
rn: *reg,
uses,
defs,
clobbers,
opcode,
caller_callconv: caller_conv,
callee_callconv: callee_conv,
Expand Down Expand Up @@ -1131,8 +1135,9 @@ impl ABIMachineSpec for AArch64MachineDeps {
insts.push(Inst::Call {
info: Box::new(CallInfo {
dest: ExternalName::LibCall(LibCall::Memcpy),
uses: vec![arg0.to_reg(), arg1.to_reg(), arg2.to_reg()],
defs: Self::get_regs_clobbered_by_call(call_conv),
uses: smallvec![arg0.to_reg(), arg1.to_reg(), arg2.to_reg()],
defs: smallvec![],
clobbers: Self::get_regs_clobbered_by_call(call_conv),
opcode: Opcode::Call,
caller_callconv: call_conv,
callee_callconv: call_conv,
Expand All @@ -1159,21 +1164,19 @@ impl ABIMachineSpec for AArch64MachineDeps {
s.nominal_sp_to_fp
}

fn get_regs_clobbered_by_call(call_conv_of_callee: isa::CallConv) -> Vec<Writable<Reg>> {
let mut caller_saved = Vec::new();
for i in 0..29 {
let x = writable_xreg(i);
if is_reg_clobbered_by_call(call_conv_of_callee, x.to_reg().to_real_reg().unwrap()) {
caller_saved.push(x);
}
}
for i in 0..32 {
let v = writable_vreg(i);
if is_reg_clobbered_by_call(call_conv_of_callee, v.to_reg().to_real_reg().unwrap()) {
caller_saved.push(v);
fn get_regs_clobbered_by_call(call_conv_of_callee: isa::CallConv) -> PRegSet {
let mut clobbers = DEFAULT_AAPCS_CLOBBERS;

if call_conv_of_callee.extends_baldrdash() {
// Every X-register except for x16, x17, x18 is
// caller-saved (clobbered by a call).
for i in 19..=28 {
clobbers.add(xreg_preg(i));
}
clobbers.add(vreg_preg(31));
}
caller_saved

clobbers
}

fn get_ext_mode(
Expand Down Expand Up @@ -1290,47 +1293,74 @@ fn get_regs_restored_in_epilogue(
(int_saves, vec_saves)
}

fn is_reg_clobbered_by_call(call_conv_of_callee: isa::CallConv, r: RealReg) -> bool {
if call_conv_of_callee.extends_baldrdash() {
match r.class() {
RegClass::Int => {
let enc = r.hw_enc() & 31;
if !BALDRDASH_JIT_CALLEE_SAVED_GPR[enc as usize] {
return true;
}
// Otherwise, fall through to preserve native's ABI caller-saved.
}
RegClass::Float => {
let enc = r.hw_enc() & 31;
if !BALDRDASH_JIT_CALLEE_SAVED_FPU[enc as usize] {
return true;
}
// Otherwise, fall through to preserve native's ABI caller-saved.
}
};
}

match r.class() {
RegClass::Int => {
// x0 - x17 inclusive are caller-saves.
r.hw_enc() <= 17
}
RegClass::Float => {
// v0 - v7 inclusive and v16 - v31 inclusive are caller-saves. The
// upper 64 bits of v8 - v15 inclusive are also caller-saves.
// However, because we cannot currently represent partial registers
// to regalloc.rs, we indicate here that every vector register is
// caller-save. Because this function is used at *callsites*,
// approximating in this direction (save more than necessary) is
// conservative and thus safe.
//
// Note that we set the 'not included in clobber set' flag in the
// regalloc.rs API when a call instruction's callee has the same ABI
// as the caller (the current function body); this is safe (anything
// clobbered by callee can be clobbered by caller as well) and
// avoids unnecessary saves of v8-v15 in the prologue even though we
// include them as defs here.
true
}
}
const fn default_aapcs_clobbers() -> PRegSet {
PRegSet::empty()
// x0 - x17 inclusive are caller-saves.
.with(xreg_preg(0))
.with(xreg_preg(1))
.with(xreg_preg(2))
.with(xreg_preg(3))
.with(xreg_preg(4))
.with(xreg_preg(5))
.with(xreg_preg(6))
.with(xreg_preg(7))
.with(xreg_preg(8))
.with(xreg_preg(9))
.with(xreg_preg(10))
.with(xreg_preg(11))
.with(xreg_preg(12))
.with(xreg_preg(13))
.with(xreg_preg(14))
.with(xreg_preg(15))
.with(xreg_preg(16))
.with(xreg_preg(17))
// v0 - v7 inclusive and v16 - v31 inclusive are
// caller-saves. The upper 64 bits of v8 - v15 inclusive are
// also caller-saves. However, because we cannot currently
// represent partial registers to regalloc2, we indicate here
// that every vector register is caller-save. Because this
// function is used at *callsites*, approximating in this
// direction (save more than necessary) is conservative and
// thus safe.
//
// Note that we exclude clobbers from a call instruction when
// a call instruction's callee has the same ABI as the caller
// (the current function body); this is safe (anything
// clobbered by callee can be clobbered by caller as well) and
// avoids unnecessary saves of v8-v15 in the prologue even
// though we include them as defs here.
.with(vreg_preg(0))
.with(vreg_preg(1))
.with(vreg_preg(2))
.with(vreg_preg(3))
.with(vreg_preg(4))
.with(vreg_preg(5))
.with(vreg_preg(6))
.with(vreg_preg(7))
.with(vreg_preg(8))
.with(vreg_preg(9))
.with(vreg_preg(10))
.with(vreg_preg(11))
.with(vreg_preg(12))
.with(vreg_preg(13))
.with(vreg_preg(14))
.with(vreg_preg(15))
.with(vreg_preg(16))
.with(vreg_preg(17))
.with(vreg_preg(18))
.with(vreg_preg(19))
.with(vreg_preg(20))
.with(vreg_preg(21))
.with(vreg_preg(22))
.with(vreg_preg(23))
.with(vreg_preg(24))
.with(vreg_preg(25))
.with(vreg_preg(26))
.with(vreg_preg(27))
.with(vreg_preg(28))
.with(vreg_preg(29))
.with(vreg_preg(30))
.with(vreg_preg(31))
}

const DEFAULT_AAPCS_CLOBBERS: PRegSet = default_aapcs_clobbers();
10 changes: 6 additions & 4 deletions cranelift/codegen/src/isa/aarch64/inst/emit_tests.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5286,8 +5286,9 @@ fn test_aarch64_binemit() {
Inst::Call {
info: Box::new(CallInfo {
dest: ExternalName::testcase("test0"),
uses: Vec::new(),
defs: Vec::new(),
uses: smallvec![],
defs: smallvec![],
clobbers: PRegSet::empty(),
opcode: Opcode::Call,
caller_callconv: CallConv::SystemV,
callee_callconv: CallConv::SystemV,
Expand All @@ -5301,8 +5302,9 @@ fn test_aarch64_binemit() {
Inst::CallInd {
info: Box::new(CallIndInfo {
rn: xreg(10),
uses: Vec::new(),
defs: Vec::new(),
uses: smallvec![],
defs: smallvec![],
clobbers: PRegSet::empty(),
opcode: Opcode::CallIndirect,
caller_callconv: CallConv::SystemV,
callee_callconv: CallConv::SystemV,
Expand Down
20 changes: 12 additions & 8 deletions cranelift/codegen/src/isa/aarch64/inst/mod.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,7 @@ use crate::machinst::{PrettyPrint, Reg, RegClass, Writable};

use alloc::vec::Vec;
use core::convert::TryFrom;
use regalloc2::VReg;
use regalloc2::{PRegSet, VReg};
use smallvec::{smallvec, SmallVec};
use std::string::{String, ToString};

Expand Down Expand Up @@ -70,8 +70,9 @@ impl BitOp {
#[derive(Clone, Debug)]
pub struct CallInfo {
pub dest: ExternalName,
pub uses: Vec<Reg>,
pub defs: Vec<Writable<Reg>>,
pub uses: SmallVec<[Reg; 8]>,
pub defs: SmallVec<[Writable<Reg>; 8]>,
pub clobbers: PRegSet,
pub opcode: Opcode,
pub caller_callconv: CallConv,
pub callee_callconv: CallConv,
Expand All @@ -82,8 +83,9 @@ pub struct CallInfo {
#[derive(Clone, Debug)]
pub struct CallIndInfo {
pub rn: Reg,
pub uses: Vec<Reg>,
pub defs: Vec<Writable<Reg>>,
pub uses: SmallVec<[Reg; 8]>,
pub defs: SmallVec<[Writable<Reg>; 8]>,
pub clobbers: PRegSet,
pub opcode: Opcode,
pub caller_callconv: CallConv,
pub callee_callconv: CallConv,
Expand Down Expand Up @@ -983,11 +985,13 @@ fn aarch64_get_operands<F: Fn(VReg) -> VReg>(inst: &Inst, collector: &mut Operan
&Inst::Call { ref info, .. } => {
collector.reg_uses(&info.uses[..]);
collector.reg_defs(&info.defs[..]);
collector.reg_clobbers(info.clobbers);
}
&Inst::CallInd { ref info, .. } => {
collector.reg_use(info.rn);
collector.reg_uses(&info.uses[..]);
collector.reg_defs(&info.defs[..]);
collector.reg_clobbers(info.clobbers);
}
&Inst::CondBr { ref kind, .. } => match kind {
CondBrKind::Zero(rt) | CondBrKind::NotZero(rt) => {
Expand Down Expand Up @@ -1028,9 +1032,9 @@ fn aarch64_get_operands<F: Fn(VReg) -> VReg>(inst: &Inst, collector: &mut Operan
&Inst::VirtualSPOffsetAdj { .. } => {}

&Inst::ElfTlsGetAddr { .. } => {
for reg in AArch64MachineDeps::get_regs_clobbered_by_call(CallConv::SystemV) {
collector.reg_def(reg);
}
collector.reg_clobbers(AArch64MachineDeps::get_regs_clobbered_by_call(
CallConv::SystemV,
));
}
&Inst::Unwind { .. } => {}
&Inst::EmitIsland { .. } => {}
Expand Down
16 changes: 12 additions & 4 deletions cranelift/codegen/src/isa/aarch64/inst/regs.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -24,9 +24,13 @@ pub const PINNED_REG: u8 = 21;
/// Get a reference to an X-register (integer register). Do not use
/// this for xsp / xzr; we have two special registers for those.
pub fn xreg(num: u8) -> Reg {
Reg::from(xreg_preg(num))
}

/// Get the given X-register as a PReg.
pub(crate) const fn xreg_preg(num: u8) -> PReg {
assert!(num < 31);
let preg = PReg::new(num as usize, RegClass::Int);
Reg::from(VReg::new(preg.index(), RegClass::Int))
PReg::new(num as usize, RegClass::Int)
}

/// Get a writable reference to an X-register.
Expand All @@ -36,9 +40,13 @@ pub fn writable_xreg(num: u8) -> Writable<Reg> {

/// Get a reference to a V-register (vector/FP register).
pub fn vreg(num: u8) -> Reg {
Reg::from(vreg_preg(num))
}

/// Get the given V-register as a PReg.
pub(crate) const fn vreg_preg(num: u8) -> PReg {
assert!(num < 32);
let preg = PReg::new(num as usize, RegClass::Float);
Reg::from(VReg::new(preg.index(), RegClass::Float))
PReg::new(num as usize, RegClass::Float)
}

/// Get a writable reference to a V-register.
Expand Down
Loading