x64: handle ISA features more completely (#11272)

* asm: generate boolean terms of CPU features

As discussed [here], we will soon need the ability to express more
complex combinations of CPU features. These are best expressed as
boolean terms: e.g., `(32-bit OR 64-bit) AND ...`, `(32-bit OR 64-bit)
AND ((AVX512VL AND AVX512F) OR AVX10.1)`. This change modifies the
generated code to have a `Inst::is_available` method which contains a
Rust-ified version of the instruction's boolean term. To do this, we now
pass in a `Features` trait, which the instruction can query to see if
its desired feature set is available.

[here]: https://bytecodealliance.zulipchat.com/#narrow/channel/217117-cranelift/topic/boolean.20terms.20for.20x64.20features

* x64: wire up `Inst::is_available` in Cranelift

This change makes us of the assembler's new generated
`Inst::is_available` methods to check an instruction's feature set in a
more succinct (and likely quicker) way. Unfortunately, this does not
allow us to print the missing ISA requirements on failure--something to
address later.

* Rename `Features` to `AvailableFeatures`

* Remove unused `InstructionSet`

* asm: fix all feature definitions

This is a mechanical transformation converting all instruction
definitions. Now, instructions should have a correct boolean term
describe the features required: e.g., `(_64b | compat) & avx`.

* x64: replace `use_*` with `has_*` when checking ISA features

In Cranelift, the `has_*` flags of `isa::x64::settings::Flags` indicate
that the CPU _has_ some capability; the `use_*` flags indicate that
Cranelift _should emit_ instructions using those capabilities. Further,
the `use_*` flags may turned on by the presence of more than one `has_*`
flags; e.g., when `has_avx` and `has_avx2` are available, `use_avx2` is
enabled.

Now that Cranelift's new x64 assembler understands boolean terms, we no
longer need the `use_*` flags for checking if an instruction can be
emitted: instead, we should use the `has_*` flags and rely on the logic
encoded in `Inst::is_available`.

* asm: materialize `Features` via `Inst::features`

For better error messages (and just for general use of CPU features, see
discussion [here]), this change adds `Inst::features`--a way to
explicitly examine the boolean term for an instruction. This function
returns a `&'static Features` that contains the `AND` and `OR` branches
defining when an instruction is available. This is all generated into
something that looks like:

```rust
pub fn features(&self) -> &'static Features {
    const F1: &'static Features = &Features::Feature(Feature::_64b);
    const F2: &'static Features = &Features::Feature(Feature::compat);
    const F0: &'static Features = &Features::Or(F1, F2);
    F0
}
```

This change makes use of `for_each_feature` more: we build up the
`AvailableFeatures` trait and the `Feature` enum from it. This should be
a bit more direct than searching through the generated code (?).

[here]: bytecodealliance/wasmtime#11272 (comment)

Author: abrown

cranelift/assembler-x64/meta/src/dsl.rs

@@ -127,9 +127,7 @@ impl core::fmt::Display for Inst {
             custom,
         } = self;
         write!(f, "{name}: {format} => {encoding}")?;
-        if !features.is_empty() {
-            write!(f, " [{features}]")?;
-        }
+        write!(f, " [{features}]")?;
         if let Some(alternate) = alternate {
             write!(f, " (alternate: {alternate})")?;
         }

cranelift/assembler-x64/meta/src/dsl/features.rs

@@ -1,61 +1,65 @@
 //! A DSL for describing x64 CPU features.
 
 use core::fmt;
-use std::ops::BitOr;
+use std::ops::{BitAnd, BitOr};
 
-/// A collection of CPU features.
+/// A boolean term of CPU features.
 ///
-/// An instruction is valid when _any_ of the features in the collection are
-/// enabled; i.e., the collection is an `OR` expression.
+/// An instruction is valid when the boolean term (a recursive tree of `AND` and
+/// `OR` terms) is satisfied.
 ///
 /// ```
 /// # use cranelift_assembler_x64_meta::dsl::{Features, Feature};
 /// let fs = Feature::_64b | Feature::compat;
-/// assert_eq!(fs.to_string(), "_64b | compat");
-/// ```
-///
-/// Duplicate features are not allowed and will cause a panic.
-///
-/// ```should_panic
-/// # use cranelift_assembler_x64_meta::dsl::Feature;
-/// let fs = Feature::_64b | Feature::_64b;
+/// assert_eq!(fs.to_string(), "(_64b | compat)");
 /// ```
 #[derive(PartialEq)]
-pub struct Features(Vec<Feature>);
+pub enum Features {
+    And(Box<Features>, Box<Features>),
+    Or(Box<Features>, Box<Features>),
+    Feature(Feature),
+}
 
 impl Features {
-    #[must_use]
-    pub fn is_empty(&self) -> bool {
-        self.0.is_empty()
-    }
-
-    pub fn iter(&self) -> impl Iterator<Item = &Feature> {
-        self.0.iter()
+    pub(crate) fn is_sse(&self) -> bool {
+        use Feature::*;
+        match self {
+            Features::And(lhs, rhs) => lhs.is_sse() || rhs.is_sse(),
+            Features::Or(lhs, rhs) => lhs.is_sse() || rhs.is_sse(),
+            Features::Feature(feature) => {
+                matches!(feature, sse | sse2 | sse3 | ssse3 | sse41 | sse42)
+            }
+        }
     }
+}
 
-    pub fn contains(&self, feature: Feature) -> bool {
-        self.0.contains(&feature)
+impl fmt::Display for Features {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        match self {
+            Features::And(lhs, rhs) => write!(f, "({lhs} & {rhs})"),
+            Features::Or(lhs, rhs) => write!(f, "({lhs} | {rhs})"),
+            Features::Feature(feature) => write!(f, "{feature:#?}"),
+        }
     }
+}
 
-    pub(crate) fn is_sse(&self) -> bool {
-        use Feature::*;
-        self.0
-            .iter()
-            .any(|f| matches!(f, sse | sse2 | sse3 | ssse3 | sse41 | sse42))
+impl<T> BitOr<T> for Features
+where
+    T: Into<Features>,
+{
+    type Output = Features;
+    fn bitor(self, rhs: T) -> Self::Output {
+        Features::Or(Box::new(self), Box::new(rhs.into()))
     }
 }
 
-impl fmt::Display for Features {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        write!(
-            f,
-            "{}",
-            self.0
-                .iter()
-                .map(ToString::to_string)
-                .collect::<Vec<_>>()
-                .join(" | ")
-        )
+impl<T> BitAnd<T> for Features
+where
+    T: Into<Features>,
+{
+    type Output = Features;
+    fn bitand(self, rhs: T) -> Self::Output {
+        Features::And(Box::new(self), Box::new(rhs.into()))
     }
 }
 
@@ -132,30 +136,27 @@ impl fmt::Display for Feature {
 }
 
 impl From<Feature> for Features {
-    fn from(flag: Feature) -> Self {
-        Features(vec![flag])
-    }
-}
-
-impl From<Option<Feature>> for Features {
-    fn from(flag: Option<Feature>) -> Self {
-        Features(flag.into_iter().collect())
+    fn from(f: Feature) -> Self {
+        Features::Feature(f)
     }
 }
 
-impl BitOr for Feature {
+impl<T> BitAnd<T> for Feature
+where
+    T: Into<Features>,
+{
     type Output = Features;
-    fn bitor(self, rhs: Self) -> Self::Output {
-        assert_ne!(self, rhs, "duplicate feature: {self:?}");
-        Features(vec![self, rhs])
+    fn bitand(self, rhs: T) -> Self::Output {
+        Features::from(self) & rhs.into()
     }
 }
 
-impl BitOr<Feature> for Features {
+impl<T> BitOr<T> for Feature
+where
+    T: Into<Features>,
+{
     type Output = Features;
-    fn bitor(mut self, rhs: Feature) -> Self::Output {
-        assert!(!self.0.contains(&rhs), "duplicate feature: {rhs:?}");
-        self.0.push(rhs);
-        self
+    fn bitor(self, rhs: T) -> Self::Output {
+        Features::from(self) | rhs.into()
     }
 }

cranelift/assembler-x64/meta/src/generate.rs

@@ -15,6 +15,7 @@ pub fn rust_assembler(f: &mut Formatter, insts: &[dsl::Inst]) {
     generate_inst_display_impl(f, insts);
     generate_inst_encode_impl(f, insts);
     generate_inst_visit_impl(f, insts);
+    generate_inst_is_available_impl(f, insts);
     generate_inst_features_impl(f, insts);
 
     // Generate per-instruction structs.
@@ -27,8 +28,8 @@ pub fn rust_assembler(f: &mut Formatter, insts: &[dsl::Inst]) {
         f.empty_line();
     }
 
-    // Generate the `Feature` enum.
-    dsl::Feature::generate_enum(f);
+    // Generate the `Feature` trait.
+    dsl::Feature::generate_macro(f);
 }
 
 /// `enum Inst { ... }`
@@ -117,10 +118,27 @@ fn generate_inst_visit_impl(f: &mut Formatter, insts: &[dsl::Inst]) {
     fmtln!(f, "}}");
 }
 
+/// `impl Inst { fn is_available... }`
+fn generate_inst_is_available_impl(f: &mut Formatter, insts: &[dsl::Inst]) {
+    f.add_block("impl<R: Registers> Inst<R>", |f| {
+        f.add_block(
+            "pub fn is_available(&self, f: &impl AvailableFeatures) -> bool",
+            |f| {
+                f.add_block("match self", |f| {
+                    for inst in insts {
+                        let variant_name = inst.name();
+                        fmtln!(f, "Self::{variant_name}(i) => i.is_available(f),");
+                    }
+                });
+            },
+        );
+    });
+}
+
 /// `impl Inst { fn features... }`
 fn generate_inst_features_impl(f: &mut Formatter, insts: &[dsl::Inst]) {
     f.add_block("impl<R: Registers> Inst<R>", |f| {
-        f.add_block("pub fn features(&self) -> Vec<Feature>", |f| {
+        f.add_block("pub fn features(&self) -> &'static Features", |f| {
             f.add_block("match self", |f| {
                 for inst in insts {
                     let variant_name = inst.name();

cranelift/assembler-x64/meta/src/generate/features.rs

@@ -1,21 +1,101 @@
 //! Generate feature-related Rust code.
 
 use super::{Formatter, fmtln};
-use crate::{dsl, generate::generate_derive};
+use crate::dsl;
 
 impl dsl::Feature {
-    /// `pub enum Feature { ... }`
+    /// `macro_rules! for_each_feature { ... }`
     ///
-    /// This function recreates the `Feature` struct itself in the generated
-    /// code.
-    pub fn generate_enum(f: &mut Formatter) {
+    /// This function generates a macro to allow generating code for each CPU
+    /// feature.
+    pub(crate) fn generate_macro(f: &mut Formatter) {
         fmtln!(f, "#[doc(hidden)]");
-        generate_derive(f);
-        fmtln!(f, "#[derive(PartialEq)]"); // Add more helpful derives.
-        f.add_block("pub enum Feature", |f| {
-            for feature in dsl::ALL_FEATURES {
-                fmtln!(f, "{feature},");
-            }
+        fmtln!(f, "#[macro_export]");
+        f.add_block("macro_rules! for_each_feature", |f| {
+            f.add_block("($m:ident) =>", |f| {
+                f.add_block("$m!", |f| {
+                    for feature in dsl::ALL_FEATURES {
+                        fmtln!(f, "{feature}");
+                    }
+                });
+            });
         });
     }
 }
+
+impl dsl::Features {
+    /// E.g., `features.is_sse2() && features.is_64b()`
+    ///
+    /// Generate a boolean expression that checks if the features are available.
+    pub(crate) fn generate_boolean_expr(&self, name: &str) -> String {
+        use dsl::Features::*;
+        match self {
+            And(lhs, rhs) => {
+                let lhs = lhs.generate_inner_boolean_expr(name);
+                let rhs = rhs.generate_inner_boolean_expr(name);
+                format!("{lhs} && {rhs}")
+            }
+            Or(lhs, rhs) => {
+                let lhs = lhs.generate_inner_boolean_expr(name);
+                let rhs = rhs.generate_inner_boolean_expr(name);
+                format!("{lhs} || {rhs}")
+            }
+            Feature(feature) => {
+                format!("{name}.{feature}()")
+            }
+        }
+    }
+
+    // This adds parentheses for inner terms.
+    fn generate_inner_boolean_expr(&self, name: &str) -> String {
+        use dsl::Features::*;
+        match self {
+            And(lhs, rhs) => {
+                let lhs = lhs.generate_inner_boolean_expr(name);
+                let rhs = rhs.generate_inner_boolean_expr(name);
+                format!("({lhs} && {rhs})")
+            }
+            Or(lhs, rhs) => {
+                let lhs = lhs.generate_inner_boolean_expr(name);
+                let rhs = rhs.generate_inner_boolean_expr(name);
+                format!("({lhs} || {rhs})")
+            }
+            Feature(feature) => format!("{name}.{feature}()"),
+        }
+    }
+
+    /// E.g., `Features::Or(Features::Feature(compat), Features::Feature(64b))`
+    ///
+    /// Generate a Rust constructor expression that contains the feature
+    /// boolean term.
+    pub(crate) fn generate_constructor_expr(&self, f: &mut Formatter) {
+        let mut index = 0;
+        let name = self.generate_inner_constructor_expr(f, &mut index);
+        fmtln!(f, "{name}");
+    }
+
+    fn generate_inner_constructor_expr(&self, f: &mut Formatter, index: &mut u32) -> String {
+        use dsl::Features::*;
+
+        let name = format!("F{index}");
+        *index += 1;
+
+        let const_expr = format!("const {name}: &'static Features");
+        match self {
+            And(lhs, rhs) => {
+                let lhs = lhs.generate_inner_constructor_expr(f, index);
+                let rhs = rhs.generate_inner_constructor_expr(f, index);
+                fmtln!(f, "{const_expr} = &Features::And({lhs}, {rhs});");
+            }
+            Or(lhs, rhs) => {
+                let lhs = lhs.generate_inner_constructor_expr(f, index);
+                let rhs = rhs.generate_inner_constructor_expr(f, index);
+                fmtln!(f, "{const_expr} = &Features::Or({lhs}, {rhs});");
+            }
+            Feature(feature) => {
+                fmtln!(f, "{const_expr} = &Features::Feature(Feature::{feature});");
+            }
+        }
+        name
+    }
+}

cranelift/assembler-x64/meta/src/generate/inst.rs

@@ -65,6 +65,8 @@ impl dsl::Inst {
             f.empty_line();
             self.generate_visit_function(f);
             f.empty_line();
+            self.generate_is_available_function(f);
+            f.empty_line();
             self.generate_features_function(f);
         });
     }
@@ -222,16 +224,23 @@ impl dsl::Inst {
         });
     }
 
-    /// `fn features(&self) -> Vec<Flag> { ... }`
+    /// `fn is_available(&self, ...) -> bool { ... }`
+    fn generate_is_available_function(&self, f: &mut Formatter) {
+        fmtln!(f, "#[must_use]");
+        f.add_block(
+            "pub fn is_available(&self, features: &impl AvailableFeatures) -> bool",
+            |f| {
+                let expr = self.features.generate_boolean_expr("features");
+                fmtln!(f, "{expr}");
+            },
+        );
+    }
+
+    /// `fn features(&self) -> Features { ... }`
     fn generate_features_function(&self, f: &mut Formatter) {
         fmtln!(f, "#[must_use]");
-        f.add_block("pub fn features(&self) -> Vec<Feature>", |f| {
-            let flags = self
-                .features
-                .iter()
-                .map(|f| format!("Feature::{f}"))
-                .collect::<Vec<_>>();
-            fmtln!(f, "vec![{}]", flags.join(", "));
+        f.add_block("pub fn features(&self) -> &'static Features", |f| {
+            self.features.generate_constructor_expr(f);
         });
     }
 

cranelift/assembler-x64/meta/src/instructions/abs.rs

@@ -4,17 +4,17 @@ use crate::dsl::{align, evex, fmt, inst, r, rex, vex, w};
 #[rustfmt::skip] // Keeps instructions on a single line.
 pub fn list() -> Vec<Inst> {
     vec![
-        inst("pabsb", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x38, 0x1C]), _64b | compat | ssse3).alt(avx, "vpabsb_a"),
-        inst("vpabsb", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._66()._0f38().op(0x1C), _64b | compat | avx),
+        inst("pabsb", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x38, 0x1C]), (_64b | compat) & ssse3).alt(avx, "vpabsb_a"),
+        inst("vpabsb", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._66()._0f38().op(0x1C), (_64b | compat) & avx),
         // FIXME: uncomment once the avx512bw feature is bound
-        // inst("vpabsb", fmt("B", [w(xmm1), r(xmm_m128)]), evex(L128, FullMem)._66()._0f38().wig().op(0x1C).r(), _64b | compat | avx512vl | avx512bw),
-        inst("pabsw", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x38, 0x1D]), _64b | compat | ssse3).alt(avx, "vpabsw_a"),
-        inst("vpabsw", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._66()._0f38().op(0x1D), _64b | compat | avx),
+        // inst("vpabsb", fmt("B", [w(xmm1), r(xmm_m128)]), evex(L128, FullMem)._66()._0f38().wig().op(0x1C).r(), (_64b | compat) & avx512vl & avx512bw),
+        inst("pabsw", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x38, 0x1D]), (_64b | compat) & ssse3).alt(avx, "vpabsw_a"),
+        inst("vpabsw", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._66()._0f38().op(0x1D), (_64b | compat) & avx),
         // FIXME: uncomment once the avx512bw feature is bound
-        // inst("vpabsw", fmt("B", [w(xmm1), r(xmm_m128)]), evex(L128, FullMem)._66()._0f38().wig().op(0x1D).r(), _64b | compat | avx512vl | avx512bw),
-        inst("pabsd", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x38, 0x1E]), _64b | compat | ssse3).alt(avx, "vpabsd_a"),
-        inst("vpabsd", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._66()._0f38().op(0x1E), _64b | compat | avx),
-        inst("vpabsd", fmt("C", [w(xmm1), r(xmm_m128)]), evex(L128, Full)._66()._0f38().w0().op(0x1E).r(), _64b | compat | avx512vl | avx512f),
-        inst("vpabsq", fmt("C", [w(xmm1), r(xmm_m128)]), evex(L128, Full)._66()._0f38().w1().op(0x1F).r(), _64b | compat | avx512vl | avx512f),
+        // inst("vpabsw", fmt("B", [w(xmm1), r(xmm_m128)]), evex(L128, FullMem)._66()._0f38().wig().op(0x1D).r(), (_64b | compat) & avx512vl & avx512bw),
+        inst("pabsd", fmt("A", [w(xmm1), r(align(xmm_m128))]), rex([0x66, 0x0F, 0x38, 0x1E]), (_64b | compat) & ssse3).alt(avx, "vpabsd_a"),
+        inst("vpabsd", fmt("A", [w(xmm1), r(xmm_m128)]), vex(L128)._66()._0f38().op(0x1E), (_64b | compat) & avx),
+        inst("vpabsd", fmt("C", [w(xmm1), r(xmm_m128)]), evex(L128, Full)._66()._0f38().w0().op(0x1E).r(), (_64b | compat) & avx512vl & avx512f),
+        inst("vpabsq", fmt("C", [w(xmm1), r(xmm_m128)]), evex(L128, Full)._66()._0f38().w1().op(0x1F).r(), (_64b | compat) & avx512vl & avx512f),
     ]
 }