Skip to content

Vectorize [un]pack4x{I, U}8[Clamp] on spv and msl #7664

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

Open
wants to merge 7 commits into
base: trunk
Choose a base branch
from
+927 −489
Open

Vectorize [un]pack4x{I, U}8[Clamp] on spv and msl #7664

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
1f695b1
[naga] Vectorize `[un]pack4x{I, U}8[Clamp]` on spv
robamler May 3, 2025
ee63be4
[naga] Vectorize `[un]pack4x{I, U}8[Clamp]` on msl
robamler May 3, 2025
c22e8fa
Add changelog entry for #7664
robamler May 3, 2025
62c32b3
Require `Capability::Int8` for vectorized [un]pack
robamler May 3, 2025
e7ebc1b
[wgpu-hal] separate 2 float16-related vk features
robamler May 4, 2025
c197de1
Rename a field to follow convention set by `ash`
robamler May 4, 2025
659df18
[wgpu-hal] Add `PrivateCapabilities::shader_int8` on Vulkan
robamler May 4, 2025
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
1 change: 1 addition & 0 deletions CHANGELOG.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,7 @@ Naga now infers the correct binding layout when a resource appears only in an as
- Apply necessary automatic conversions to the `value` argument of `textureStore`. By @jimblandy in [#7567](https://github.com/gfx-rs/wgpu/pull/7567).
- Properly apply WGSL's automatic conversions to the arguments to texture sampling functions. By @jimblandy in [#7548](https://github.com/gfx-rs/wgpu/pull/7548).
- Properly evaluate `abs(most negative abstract int)`. By @jimblandy in [#7507](https://github.com/gfx-rs/wgpu/pull/7507).
- Generate vectorized code for `[un]pack4x{I,U}8[Clamp]` on SPIR-V and MSL 2.1+. By @robamler in [#7664](https://github.com/gfx-rs/wgpu/pull/7664).

#### DX12

Expand Down
145 changes: 98 additions & 47 deletions naga/src/back/msl/writer.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1497,6 +1497,63 @@ impl<W: Write> Writer<W> {
Ok(())
}

/// Emit code for the WGSL functions `pack4x{I, U}8[Clamp]`.
fn put_pack4x8(
&mut self,
arg: Handle<crate::Expression>,
context: &ExpressionContext<'_>,
was_signed: bool,
clamp_bounds: Option<(&str, &str)>,
) -> Result<(), Error> {
if context.lang_version >= (2, 1) {
let packed_type = if was_signed {
"packed_char4"
} else {
"packed_uchar4"
};
// Metal uses little endian byte order, which matches what WGSL expects here.
write!(self.out, "as_type<uint>({packed_type}(")?;
if let Some((min, max)) = clamp_bounds {
// Clamping a vector to scalar bounds works and operates component-wise.
write!(self.out, "{NAMESPACE}::clamp(")?;
self.put_expression(arg, context, true)?;
write!(self.out, ", {min}, {max})")?;
} else {
self.put_expression(arg, context, true)?;
}
write!(self.out, "))")?;
} else {
// MSL < 2.1 doesn't support `as_type` casting between packed chars and scalars.
if was_signed {
write!(self.out, "uint(")?;
}
let write_arg = |this: &mut Self| -> BackendResult {
if let Some((min, max)) = clamp_bounds {
write!(this.out, "{NAMESPACE}::clamp(")?;
this.put_expression(arg, context, true)?;
write!(this.out, ", {min}, {max})")?;
} else {
this.put_expression(arg, context, true)?;
}
Ok(())
};
write!(self.out, "(")?;
write_arg(self)?;
write!(self.out, "[0] & 0xFF) | ((")?;
write_arg(self)?;
write!(self.out, "[1] & 0xFF) << 8) | ((")?;
write_arg(self)?;
write!(self.out, "[2] & 0xFF) << 16) | ((")?;
write_arg(self)?;
write!(self.out, "[3] & 0xFF) << 24)")?;
if was_signed {
write!(self.out, ")")?;
}
}

Ok(())
}

/// Emit code for the isign expression.
///
fn put_isign(
Expand Down Expand Up @@ -2437,53 +2494,41 @@ impl<W: Write> Writer<W> {
write!(self.out, "{fun_name}")?;
self.put_call_parameters(iter::once(arg), context)?;
}
fun @ (Mf::Pack4xI8 | Mf::Pack4xU8 | Mf::Pack4xI8Clamp | Mf::Pack4xU8Clamp) => {
let was_signed = matches!(fun, Mf::Pack4xI8 | Mf::Pack4xI8Clamp);
let clamp_bounds = match fun {
Mf::Pack4xI8Clamp => Some(("-128", "127")),
Mf::Pack4xU8Clamp => Some(("0", "255")),
_ => None,
};
if was_signed {
write!(self.out, "uint(")?;
}
let write_arg = |this: &mut Self| -> BackendResult {
if let Some((min, max)) = clamp_bounds {
write!(this.out, "{NAMESPACE}::clamp(")?;
this.put_expression(arg, context, true)?;
write!(this.out, ", {min}, {max})")?;
} else {
this.put_expression(arg, context, true)?;
}
Ok(())
};
write!(self.out, "(")?;
write_arg(self)?;
write!(self.out, "[0] & 0xFF) | ((")?;
write_arg(self)?;
write!(self.out, "[1] & 0xFF) << 8) | ((")?;
write_arg(self)?;
write!(self.out, "[2] & 0xFF) << 16) | ((")?;
write_arg(self)?;
write!(self.out, "[3] & 0xFF) << 24)")?;
if was_signed {
write!(self.out, ")")?;
}
Mf::Pack4xI8 => self.put_pack4x8(arg, context, true, None)?,
Mf::Pack4xU8 => self.put_pack4x8(arg, context, false, None)?,
Mf::Pack4xI8Clamp => {
self.put_pack4x8(arg, context, true, Some(("-128", "127")))?
}
Mf::Pack4xU8Clamp => {
self.put_pack4x8(arg, context, false, Some(("0", "255")))?
}
fun @ (Mf::Unpack4xI8 | Mf::Unpack4xU8) => {
write!(self.out, "(")?;
if matches!(fun, Mf::Unpack4xU8) {
write!(self.out, "u")?;
let sign_prefix = if matches!(fun, Mf::Unpack4xU8) {
"u"
} else {
""
};

if context.lang_version >= (2, 1) {
// Metal uses little endian byte order, which matches what WGSL expects here.
write!(
self.out,
"{sign_prefix}int4(as_type<packed_{sign_prefix}char4>("
)?;
self.put_expression(arg, context, true)?;
write!(self.out, "))")?;
} else {
// MSL < 2.1 doesn't support `as_type` casting between packed chars and scalars.
write!(self.out, "({sign_prefix}int4(")?;
self.put_expression(arg, context, true)?;
write!(self.out, ", ")?;
self.put_expression(arg, context, true)?;
write!(self.out, " >> 8, ")?;
self.put_expression(arg, context, true)?;
write!(self.out, " >> 16, ")?;
self.put_expression(arg, context, true)?;
write!(self.out, " >> 24) << 24 >> 24)")?;
}
write!(self.out, "int4(")?;
self.put_expression(arg, context, true)?;
write!(self.out, ", ")?;
self.put_expression(arg, context, true)?;
write!(self.out, " >> 8, ")?;
self.put_expression(arg, context, true)?;
write!(self.out, " >> 16, ")?;
self.put_expression(arg, context, true)?;
write!(self.out, " >> 24) << 24 >> 24)")?;
}
Mf::QuantizeToF16 => {
match *context.resolve_type(arg) {
Expand Down Expand Up @@ -3226,14 +3271,20 @@ impl<W: Write> Writer<W> {
self.need_bake_expressions.insert(arg);
self.need_bake_expressions.insert(arg1.unwrap());
}
crate::MathFunction::FirstLeadingBit
| crate::MathFunction::Pack4xI8
crate::MathFunction::FirstLeadingBit => {
self.need_bake_expressions.insert(arg);
}
crate::MathFunction::Pack4xI8
| crate::MathFunction::Pack4xU8
| crate::MathFunction::Pack4xI8Clamp
| crate::MathFunction::Pack4xU8Clamp
| crate::MathFunction::Unpack4xI8
| crate::MathFunction::Unpack4xU8 => {
self.need_bake_expressions.insert(arg);
// On MSL < 2.1, we emit a polyfill for these functions that uses the
// argument multiple times. This is no longer necessary on MSL >= 2.1.
if context.lang_version < (2, 1) {
self.need_bake_expressions.insert(arg);
}
}
crate::MathFunction::ExtractBits => {
// Only argument 1 is re-used.
Expand Down
Loading