T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Switch Cranelift over to regalloc2. (#3989)

Browse Source 
This PR switches Cranelift over to the new register allocator, regalloc2.

See [this document](https://gist.github.com/cfallin/08553421a91f150254fe878f67301801)
for a summary of the design changes. This switchover has implications for
core VCode/MachInst types and the lowering pass.

Overall, this change brings improvements to both compile time and speed of
generated code (runtime), as reported in #3942:

```
Benchmark       Compilation (wallclock)     Execution (wallclock)
blake3-scalar   25% faster                  28% faster
blake3-simd     no diff                     no diff
meshoptimizer   19% faster                  17% faster
pulldown-cmark  17% faster                  no diff
bz2             15% faster                  no diff
SpiderMonkey,   21% faster                  2% faster
  fib(30)
clang.wasm      42% faster                  N/A
```
This commit is contained in:
Chris Fallin
2022-04-14 10:28:21 -07:00
committed by GitHub
parent bfae6384aa
commit a0318f36f0
181 changed files with 16887 additions and 21587 deletions
Download Patch File Download Diff File Expand all files Collapse all files

68
cranelift/codegen/src/isa/aarch64/lower_inst.rs
View File

@@ -1,27 +1,22 @@
//! Lower a single Cranelift instruction into vcode.
use super::lower::*;
use crate::binemit::CodeOffset;
use crate::ir::condcodes::FloatCC;
use crate::ir::types::*;
use crate::ir::Inst as IRInst;
use crate::ir::{InstructionData, Opcode, TrapCode};
use crate::isa::aarch64::abi::*;
use crate::isa::aarch64::inst::*;
use crate::isa::aarch64::settings as aarch64_settings;
use crate::machinst::lower::*;
use crate::machinst::*;
use crate::settings::{Flags, TlsModel};
use crate::{CodegenError, CodegenResult};
use crate::isa::aarch64::abi::*;
use crate::isa::aarch64::inst::*;
use regalloc::Writable;
use alloc::boxed::Box;
use alloc::vec::Vec;
use core::convert::TryFrom;
use super::lower::*;
/// Actually codegen an instruction's results into registers.
pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
ctx: &mut C,
@@ -766,7 +761,7 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
Opcode::Trap | Opcode::ResumableTrap => {
let trap_code = ctx.data(insn).trap_code().unwrap();
ctx.emit_safepoint(Inst::Udf { trap_code });
ctx.emit(Inst::Udf { trap_code });
}
Opcode::Trapif | Opcode::Trapff => {
@@ -797,7 +792,7 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
cond
};
ctx.emit_safepoint(Inst::TrapIf {
ctx.emit(Inst::TrapIf {
trap_code,
kind: CondBrKind::Cond(cond),
});
@@ -1507,35 +1502,34 @@ pub(crate) fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
let lane_type = ty.lane_type();
let rd = get_output_reg(ctx, outputs[0]).only_reg().unwrap();
let mut match_long_pair =
|ext_low_op, ext_high_op| -> Option<(VecRRPairLongOp, regalloc::Reg)> {
if let Some(lhs) = maybe_input_insn(ctx, inputs[0], ext_low_op) {
if let Some(rhs) = maybe_input_insn(ctx, inputs[1], ext_high_op) {
let lhs_inputs = insn_inputs(ctx, lhs);
let rhs_inputs = insn_inputs(ctx, rhs);
let low = put_input_in_reg(ctx, lhs_inputs[0], NarrowValueMode::None);
let high = put_input_in_reg(ctx, rhs_inputs[0], NarrowValueMode::None);
if low == high {
match (lane_type, ext_low_op) {
(I16, Opcode::SwidenLow) => {
return Some((VecRRPairLongOp::Saddlp8, low))
}
(I32, Opcode::SwidenLow) => {
return Some((VecRRPairLongOp::Saddlp16, low))
}
(I16, Opcode::UwidenLow) => {
return Some((VecRRPairLongOp::Uaddlp8, low))
}
(I32, Opcode::UwidenLow) => {
return Some((VecRRPairLongOp::Uaddlp16, low))
}
_ => (),
};
}
let mut match_long_pair = |ext_low_op, ext_high_op| -> Option<(VecRRPairLongOp, Reg)> {
if let Some(lhs) = maybe_input_insn(ctx, inputs[0], ext_low_op) {
if let Some(rhs) = maybe_input_insn(ctx, inputs[1], ext_high_op) {
let lhs_inputs = insn_inputs(ctx, lhs);
let rhs_inputs = insn_inputs(ctx, rhs);
let low = put_input_in_reg(ctx, lhs_inputs[0], NarrowValueMode::None);
let high = put_input_in_reg(ctx, rhs_inputs[0], NarrowValueMode::None);
if low == high {
match (lane_type, ext_low_op) {
(I16, Opcode::SwidenLow) => {
return Some((VecRRPairLongOp::Saddlp8, low))
}
(I32, Opcode::SwidenLow) => {
return Some((VecRRPairLongOp::Saddlp16, low))
}
(I16, Opcode::UwidenLow) => {
return Some((VecRRPairLongOp::Uaddlp8, low))
}
(I32, Opcode::UwidenLow) => {
return Some((VecRRPairLongOp::Uaddlp16, low))
}
_ => (),
};
}
}
None
};
}
None
};
if let Some((op, rn)) = match_long_pair(Opcode::SwidenLow, Opcode::SwidenHigh) {
ctx.emit(Inst::VecRRPairLong { op, rd, rn });