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

Nanboxing #36

Open
wants to merge 4 commits into
base: main
Choose a base branch
from
Open

Nanboxing #36

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
e7db0c4
remove unnecassary retain
lenawanel May 29, 2024
acc3a5c
add single precision nanboxing helper methods
lenawanel May 29, 2024
60155c9
add nan_boxing
lenawanel May 29, 2024
8b66bae
fix tests
lenawanel May 29, 2024
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
154 changes: 100 additions & 54 deletions src/cpu.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -159,10 +159,35 @@ impl FRegisters {
self.fregs[index as usize]
}

/// read a single precision float a register,
/// expecting nanboxing according to the spec.
/// 21.2. NaN Boxing of Narrower Values
/// "[if] n<FLEN, check if the input operands are correctly NaN-boxed, i.e., all
/// upper FLEN-n bits are 1. If so, the n least-significant bits of the input are used as the input value,
/// otherwise the input value is treated as an n-bit canonical NaN."
pub fn read_nanboxed(&self, index: u64) -> f32 {
let bits = self.read(index).to_bits();
if !bits <= u32::MAX as u64 {
f32::from_bits(bits as u32)
} else {
f32::NAN
}
}

/// Write the value to a register.
pub fn write(&mut self, index: u64, value: f64) {
self.fregs[index as usize] = value;
}

/// Write the single precision number to a register.
/// nan boxing them according to the spec
/// 21.2. NaN Boxing of Narrower Values
/// "Any operation that writes a narrower result to an 'f' register must write all 1s to the uppermost
/// FLEN-n bits to yield a legal NaN-boxedvalue."
pub fn write_nanboxed(&mut self, index: u64, value: f32) {
const NAN_MASK: u64 = !(u32::MAX as u64);
self.write(index, f64::from_bits(NAN_MASK | value.to_bits() as u64));
}
}

impl fmt::Display for FRegisters {
Expand Down Expand Up @@ -1790,7 +1815,6 @@ impl Cpu {
self.write(addr, self.xregs.read(rs2), WORD)?;
self.xregs.write(rd, 0);
} else {
self.reservation_set.retain(|&x| x != addr);
self.xregs.write(rd, 1);
};
}
Expand All @@ -1811,7 +1835,6 @@ impl Cpu {
self.write(addr, self.xregs.read(rs2), DOUBLEWORD)?;
self.xregs.write(rd, 0);
} else {
self.reservation_set.retain(|&x| x != addr);
self.xregs.write(rd, 1);
}
}
Expand Down Expand Up @@ -2477,11 +2500,12 @@ impl Cpu {
inst_count!(self, "fmadd.s");
self.debug(inst, "fmadd.s");

self.fregs.write(
self.fregs.write_nanboxed(
rd,
(self.fregs.read(rs1) as f32)
.mul_add(self.fregs.read(rs2) as f32, self.fregs.read(rs3) as f32)
as f64,
(self.fregs.read_nanboxed(rs1)).mul_add(
self.fregs.read_nanboxed(rs2) ,
self.fregs.read_nanboxed(rs3) ,
),
);
}
0x1 => {
Expand Down Expand Up @@ -2512,11 +2536,12 @@ impl Cpu {
inst_count!(self, "fmsub.s");
self.debug(inst, "fmsub.s");

self.fregs.write(
self.fregs.write_nanboxed(
rd,
(self.fregs.read(rs1) as f32)
.mul_add(self.fregs.read(rs2) as f32, -self.fregs.read(rs3) as f32)
as f64,
self.fregs.read_nanboxed(rs1).mul_add(
self.fregs.read_nanboxed(rs2),
-self.fregs.read_nanboxed(rs3),
),
);
}
0x1 => {
Expand Down Expand Up @@ -2547,11 +2572,12 @@ impl Cpu {
inst_count!(self, "fnmadd.s");
self.debug(inst, "fnmadd.s");

self.fregs.write(
self.fregs.write_nanboxed(
rd,
(-self.fregs.read(rs1) as f32)
.mul_add(self.fregs.read(rs2) as f32, self.fregs.read(rs3) as f32)
as f64,
(-self.fregs.read_nanboxed(rs1)).mul_add(
self.fregs.read_nanboxed(rs2),
self.fregs.read_nanboxed(rs3),
),
);
}
0x1 => {
Expand Down Expand Up @@ -2581,11 +2607,12 @@ impl Cpu {
inst_count!(self, "fnmsub.s");
self.debug(inst, "fnmsub.s");

self.fregs.write(
self.fregs.write_nanboxed(
rd,
(-self.fregs.read(rs1) as f32)
.mul_add(self.fregs.read(rs2) as f32, -self.fregs.read(rs3) as f32)
as f64,
(-self.fregs.read_nanboxed(rs1)).mul_add(
self.fregs.read_nanboxed(rs2),
-self.fregs.read_nanboxed(rs3),
),
);
}
0x1 => {
Expand All @@ -2607,7 +2634,6 @@ impl Cpu {
0x53 => {
// RV32F and RV64F
// TODO: support the rounding mode encoding (rm).
// TODO: NaN Boxing of Narrower Values (Spec 12.2).
// TODO: set exception flags.

/*
Expand Down Expand Up @@ -2639,9 +2665,9 @@ impl Cpu {
inst_count!(self, "fadd.s");
self.debug(inst, "fadd.s");

self.fregs.write(
self.fregs.write_nanboxed(
rd,
(self.fregs.read(rs1) as f32 + self.fregs.read(rs2) as f32) as f64,
self.fregs.read_nanboxed(rs1) + self.fregs.read_nanboxed(rs2),
)
}
0x01 => {
Expand All @@ -2657,9 +2683,9 @@ impl Cpu {
inst_count!(self, "fsub.s");
self.debug(inst, "fsub.s");

self.fregs.write(
self.fregs.write_nanboxed(
rd,
(self.fregs.read(rs1) as f32 - self.fregs.read(rs2) as f32) as f64,
self.fregs.read_nanboxed(rs1) - self.fregs.read_nanboxed(rs2),
)
}
0x05 => {
Expand All @@ -2675,9 +2701,9 @@ impl Cpu {
inst_count!(self, "fmul.s");
self.debug(inst, "fmul.s");

self.fregs.write(
self.fregs.write_nanboxed(
rd,
(self.fregs.read(rs1) as f32 * self.fregs.read(rs2) as f32) as f64,
self.fregs.read_nanboxed(rs1) * self.fregs.read_nanboxed(rs2),
)
}
0x09 => {
Expand All @@ -2693,9 +2719,9 @@ impl Cpu {
inst_count!(self, "fdiv.s");
self.debug(inst, "fdiv.s");

self.fregs.write(
self.fregs.write_nanboxed(
rd,
(self.fregs.read(rs1) as f32 / self.fregs.read(rs2) as f32) as f64,
self.fregs.read_nanboxed(rs1) / self.fregs.read_nanboxed(rs2),
)
}
0x0d => {
Expand All @@ -2713,29 +2739,35 @@ impl Cpu {
inst_count!(self, "fsgnj.s");
self.debug(inst, "fsgnj.s");

self.fregs
.write(rd, self.fregs.read(rs1).copysign(self.fregs.read(rs2)));
self.fregs.write_nanboxed(
rd,
self.fregs
.read_nanboxed(rs1)
.copysign(self.fregs.read_nanboxed(rs2)),
);
}
0x1 => {
// fsgnjn.s
inst_count!(self, "fsgnjn.s");
self.debug(inst, "fsgnjn.s");

self.fregs.write(
self.fregs.write_nanboxed(
rd,
self.fregs.read(rs1).copysign(-self.fregs.read(rs2)),
self.fregs
.read_nanboxed(rs1)
.copysign(-self.fregs.read_nanboxed(rs2)),
);
}
0x2 => {
// fsgnjx.s
inst_count!(self, "fsgnjx.s");
self.debug(inst, "fsgnjx.s");

let sign1 = (self.fregs.read(rs1) as f32).to_bits() & 0x80000000;
let sign2 = (self.fregs.read(rs2) as f32).to_bits() & 0x80000000;
let other = (self.fregs.read(rs1) as f32).to_bits() & 0x7fffffff;
let sign1 = self.fregs.read_nanboxed(rs1).to_bits() & 0x80000000;
let sign2 = self.fregs.read_nanboxed(rs2).to_bits() & 0x80000000;
let other = self.fregs.read_nanboxed(rs1).to_bits() & 0x7fffffff;
self.fregs
.write(rd, f32::from_bits((sign1 ^ sign2) | other) as f64);
.write_nanboxed(rd, f32::from_bits((sign1 ^ sign2) | other));
}
_ => {
return Err(Exception::IllegalInstruction(inst));
Expand Down Expand Up @@ -2785,16 +2817,24 @@ impl Cpu {
inst_count!(self, "fmin.s");
self.debug(inst, "fmin.s");

self.fregs
.write(rd, self.fregs.read(rs1).min(self.fregs.read(rs2)));
self.fregs.write_nanboxed(
rd,
self.fregs
.read_nanboxed(rs1)
.min(self.fregs.read_nanboxed(rs2)),
);
}
0x1 => {
// fmax.s
inst_count!(self, "fmax.s");
self.debug(inst, "fmax.s");

self.fregs
.write(rd, self.fregs.read(rs1).max(self.fregs.read(rs2)));
self.fregs.write_nanboxed(
rd,
self.fregs
.read_nanboxed(rs1)
.max(self.fregs.read_nanboxed(rs2)),
);
}
_ => {
return Err(Exception::IllegalInstruction(inst));
Expand Down Expand Up @@ -2829,22 +2869,22 @@ impl Cpu {
inst_count!(self, "fcvt.s.d");
self.debug(inst, "fcvt.s.d");

self.fregs.write(rd, self.fregs.read(rs1));
self.fregs.write_nanboxed(rd, self.fregs.read(rs1) as f32);
}
0x21 => {
// fcvt.d.s
inst_count!(self, "fcvt.d.s");
self.debug(inst, "fcvt.d.s");

self.fregs.write(rd, (self.fregs.read(rs1) as f32) as f64);
self.fregs.write(rd, self.fregs.read_nanboxed(rs1) as f64);
}
0x2c => {
// fsqrt.s
inst_count!(self, "fsqrt.s");
self.debug(inst, "fsqrt.s");

self.fregs
.write(rd, (self.fregs.read(rs1) as f32).sqrt() as f64);
.write_nanboxed(rd, self.fregs.read_nanboxed(rs1).sqrt());
}
0x2d => {
// fsqrt.d
Expand All @@ -2862,7 +2902,9 @@ impl Cpu {

self.xregs.write(
rd,
if self.fregs.read(rs1) <= self.fregs.read(rs2) {
if self.fregs.read_nanboxed(rs1)
<= self.fregs.read_nanboxed(rs2)
{
1
} else {
0
Expand All @@ -2876,7 +2918,8 @@ impl Cpu {

self.xregs.write(
rd,
if self.fregs.read(rs1) < self.fregs.read(rs2) {
if self.fregs.read_nanboxed(rs1) < self.fregs.read_nanboxed(rs2)
{
1
} else {
0
Expand All @@ -2890,7 +2933,9 @@ impl Cpu {

self.xregs.write(
rd,
if self.fregs.read(rs1) == self.fregs.read(rs2) {
if self.fregs.read_nanboxed(rs1)
== self.fregs.read_nanboxed(rs2)
{
1
} else {
0
Expand Down Expand Up @@ -2960,7 +3005,7 @@ impl Cpu {

self.xregs.write(
rd,
((self.fregs.read(rs1) as f32).round() as i32) as u64,
(self.fregs.read_nanboxed(rs1).round() as i32) as u64,
);
}
0x1 => {
Expand All @@ -2970,7 +3015,7 @@ impl Cpu {

self.xregs.write(
rd,
(((self.fregs.read(rs1) as f32).round() as u32) as i32) as u64,
((self.fregs.read_nanboxed(rs1).round() as u32) as i32) as u64,
);
}
0x2 => {
Expand All @@ -2979,15 +3024,15 @@ impl Cpu {
self.debug(inst, "fcvt.l.s");

self.xregs
.write(rd, (self.fregs.read(rs1) as f32).round() as u64);
.write(rd, self.fregs.read_nanboxed(rs1).round() as u64);
}
0x3 => {
// fcvt.lu.s
inst_count!(self, "fcvt.lu.s");
self.debug(inst, "fcvt.lu.s");

self.xregs
.write(rd, (self.fregs.read(rs1) as f32).round() as u64);
.write(rd, self.fregs.read_nanboxed(rs1).round() as u64);
}
_ => {
return Err(Exception::IllegalInstruction(inst));
Expand Down Expand Up @@ -3041,30 +3086,31 @@ impl Cpu {
self.debug(inst, "fcvt.s.w");

self.fregs
.write(rd, ((self.xregs.read(rs1) as i32) as f32) as f64);
.write_nanboxed(rd, (self.xregs.read(rs1) as i32) as f32);
}
0x1 => {
// fcvt.s.wu
inst_count!(self, "fcvt.s.wu");
self.debug(inst, "fcvt.s.wu");

self.fregs
.write(rd, ((self.xregs.read(rs1) as u32) as f32) as f64);
.write_nanboxed(rd, (self.xregs.read(rs1) as u32) as f32);
}
0x2 => {
// fcvt.s.l
inst_count!(self, "fcvt.s.l");
self.debug(inst, "fcvt.s.l");

self.fregs.write(rd, (self.xregs.read(rs1) as f32) as f64);
self.fregs
.write_nanboxed(rd, self.xregs.read(rs1) as i64 as f32);
}
0x3 => {
// fcvt.s.lu
inst_count!(self, "fcvt.s.lu");
self.debug(inst, "fcvt.s.lu");

self.fregs
.write(rd, ((self.xregs.read(rs1) as u64) as f32) as f64);
.write_nanboxed(rd, (self.xregs.read(rs1) as u64) as f32);
}
_ => {
return Err(Exception::IllegalInstruction(inst));
Expand Down Expand Up @@ -3124,7 +3170,7 @@ impl Cpu {
inst_count!(self, "fclass.s");
self.debug(inst, "fclass.s");

let f = self.fregs.read(rs1);
let f = self.fregs.read_nanboxed(rs1);
match f.classify() {
FpCategory::Infinite => {
self.xregs
Expand Down
5 changes: 5 additions & 0 deletions tests/helper.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -59,3 +59,8 @@ pub fn run(
);
}
}

pub fn nan_box(num: f32) -> f64 {
let NAN_MASK = !(u32::MAX as u64);
f64::from_bits(NAN_MASK | (num.to_bits() as u64))
}
Loading