Auto merge of #140792 - Urgau:minimum-maximum-intrinsics, r=<try>

Use intrinsics for `{f16,f32,f64,f128}::{minimum,maximum}` operations

This PR creates intrinsics for `{f16,f32,f64,f64}::{minimum,maximum}` operations.

This wasn't done when those operations were added as the LLVM support was too weak but now that LLVM has libcalls for unsupported platforms we can finally use them.

Cranelift and GCC[^1] support are partial, Cranelift doesn't support `f16` and `f128`, while GCC doesn't support `f16`.

r? `@tgross35`

try-job: aarch64-gnu
try-job: dist-various-1
try-job: dist-various-2

[^1]: https://www.gnu.org/software///gnulib/manual/html_node/Functions-in-_003cmath_002eh_003e.html

Author: bors

compiler/rustc_codegen_cranelift/src/intrinsics/mod.rs

@@ -1109,6 +1109,43 @@ fn codegen_regular_intrinsic_call<'tcx>(
             ret.write_cvalue(fx, old);
         }
 
+        sym::minimumf32 => {
+            intrinsic_args!(fx, args => (a, b); intrinsic);
+            let a = a.load_scalar(fx);
+            let b = b.load_scalar(fx);
+
+            let val = fx.bcx.ins().fmin(a, b);
+            let val = CValue::by_val(val, fx.layout_of(fx.tcx.types.f32));
+            ret.write_cvalue(fx, val);
+        }
+        sym::minimumf64 => {
+            intrinsic_args!(fx, args => (a, b); intrinsic);
+            let a = a.load_scalar(fx);
+            let b = b.load_scalar(fx);
+
+            let val = fx.bcx.ins().fmin(a, b);
+            let val = CValue::by_val(val, fx.layout_of(fx.tcx.types.f64));
+            ret.write_cvalue(fx, val);
+        }
+        sym::maximumf32 => {
+            intrinsic_args!(fx, args => (a, b); intrinsic);
+            let a = a.load_scalar(fx);
+            let b = b.load_scalar(fx);
+
+            let val = fx.bcx.ins().fmax(a, b);
+            let val = CValue::by_val(val, fx.layout_of(fx.tcx.types.f32));
+            ret.write_cvalue(fx, val);
+        }
+        sym::maximumf64 => {
+            intrinsic_args!(fx, args => (a, b); intrinsic);
+            let a = a.load_scalar(fx);
+            let b = b.load_scalar(fx);
+
+            let val = fx.bcx.ins().fmax(a, b);
+            let val = CValue::by_val(val, fx.layout_of(fx.tcx.types.f64));
+            ret.write_cvalue(fx, val);
+        }
+
         sym::minnumf32 => {
             intrinsic_args!(fx, args => (a, b); intrinsic);
             let a = a.load_scalar(fx);

compiler/rustc_codegen_gcc/src/intrinsic/mod.rs

@@ -74,8 +74,44 @@ fn get_simple_intrinsic<'gcc, 'tcx>(
         sym::fabsf64 => "fabs",
         sym::minnumf32 => "fminf",
         sym::minnumf64 => "fmin",
+        sym::minimumf32 => "fminimumf",
+        sym::minimumf64 => "fminimum",
+        sym::minimumf128 => {
+            // GCC doesn't have the intrinsic we want so we use the compiler-builtins one
+            // https://docs.rs/compiler_builtins/latest/compiler_builtins/math/full_availability/fn.fminimumf128.html
+            let f128_type = cx.type_f128();
+            return Some(cx.context.new_function(
+                None,
+                FunctionType::Extern,
+                f128_type,
+                &[
+                    cx.context.new_parameter(None, f128_type, "a"),
+                    cx.context.new_parameter(None, f128_type, "b"),
+                ],
+                "fminimumf128",
+                false,
+            ));
+        }
         sym::maxnumf32 => "fmaxf",
         sym::maxnumf64 => "fmax",
+        sym::maximumf32 => "fmaximumf",
+        sym::maximumf64 => "fmaximum",
+        sym::maximumf128 => {
+            // GCC doesn't have the intrinsic we want so we use the compiler-builtins one
+            // https://docs.rs/compiler_builtins/latest/compiler_builtins/math/full_availability/fn.fmaximumf128.html
+            let f128_type = cx.type_f128();
+            return Some(cx.context.new_function(
+                None,
+                FunctionType::Extern,
+                f128_type,
+                &[
+                    cx.context.new_parameter(None, f128_type, "a"),
+                    cx.context.new_parameter(None, f128_type, "b"),
+                ],
+                "fmaximumf128",
+                false,
+            ));
+        }
         sym::copysignf32 => "copysignf",
         sym::copysignf64 => "copysign",
         sym::copysignf128 => "copysignl",

compiler/rustc_codegen_llvm/src/context.rs

@@ -1009,11 +1009,27 @@ impl<'ll> CodegenCx<'ll, '_> {
         ifn!("llvm.minnum.f64", fn(t_f64, t_f64) -> t_f64);
         ifn!("llvm.minnum.f128", fn(t_f128, t_f128) -> t_f128);
 
+        ifn!("llvm.minimum.f16", fn(t_f16, t_f16) -> t_f16);
+        ifn!("llvm.minimum.f32", fn(t_f32, t_f32) -> t_f32);
+        ifn!("llvm.minimum.f64", fn(t_f64, t_f64) -> t_f64);
+        // There are issues on x86_64 and aarch64 with the f128 variant.
+        //  - https://github.com/llvm/llvm-project/issues/139380
+        //  - https://github.com/llvm/llvm-project/issues/139381
+        // ifn!("llvm.minimum.f128", fn(t_f128, t_f128) -> t_f128);
+
         ifn!("llvm.maxnum.f16", fn(t_f16, t_f16) -> t_f16);
         ifn!("llvm.maxnum.f32", fn(t_f32, t_f32) -> t_f32);
         ifn!("llvm.maxnum.f64", fn(t_f64, t_f64) -> t_f64);
         ifn!("llvm.maxnum.f128", fn(t_f128, t_f128) -> t_f128);
 
+        ifn!("llvm.maximum.f16", fn(t_f16, t_f16) -> t_f16);
+        ifn!("llvm.maximum.f32", fn(t_f32, t_f32) -> t_f32);
+        ifn!("llvm.maximum.f64", fn(t_f64, t_f64) -> t_f64);
+        // There are issues on x86_64 and aarch64 with the f128 variant.
+        //  - https://github.com/llvm/llvm-project/issues/139380
+        //  - https://github.com/llvm/llvm-project/issues/139381
+        // ifn!("llvm.maximum.f128", fn(t_f128, t_f128) -> t_f128);
+
         ifn!("llvm.floor.f16", fn(t_f16) -> t_f16);
         ifn!("llvm.floor.f32", fn(t_f32) -> t_f32);
         ifn!("llvm.floor.f64", fn(t_f64) -> t_f64);

compiler/rustc_codegen_llvm/src/intrinsic.rs

@@ -103,11 +103,23 @@ fn get_simple_intrinsic<'ll>(
         sym::minnumf64 => "llvm.minnum.f64",
         sym::minnumf128 => "llvm.minnum.f128",
 
+        sym::minimumf16 => "llvm.minimum.f16",
+        sym::minimumf32 => "llvm.minimum.f32",
+        sym::minimumf64 => "llvm.minimum.f64",
+        // There are issues on x86_64 and aarch64 with the f128 variant,
+        // let's instead use the instrinsic fallback body.
+        // sym::minimumf128 => "llvm.minimum.f128",
         sym::maxnumf16 => "llvm.maxnum.f16",
         sym::maxnumf32 => "llvm.maxnum.f32",
         sym::maxnumf64 => "llvm.maxnum.f64",
         sym::maxnumf128 => "llvm.maxnum.f128",
 
+        sym::maximumf16 => "llvm.maximum.f16",
+        sym::maximumf32 => "llvm.maximum.f32",
+        sym::maximumf64 => "llvm.maximum.f64",
+        // There are issues on x86_64 and aarch64 with the f128 variant,
+        // let's instead use the instrinsic fallback body.
+        // sym::maximumf128 => "llvm.maximum.f128",
         sym::copysignf16 => "llvm.copysign.f16",
         sym::copysignf32 => "llvm.copysign.f32",
         sym::copysignf64 => "llvm.copysign.f64",

compiler/rustc_const_eval/src/interpret/intrinsics.rs

@@ -493,11 +493,21 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
             sym::minnumf64 => self.float_min_intrinsic::<Double>(args, dest)?,
             sym::minnumf128 => self.float_min_intrinsic::<Quad>(args, dest)?,
 
+            sym::minimumf16 => self.float_minimum_intrinsic::<Half>(args, dest)?,
+            sym::minimumf32 => self.float_minimum_intrinsic::<Single>(args, dest)?,
+            sym::minimumf64 => self.float_minimum_intrinsic::<Double>(args, dest)?,
+            sym::minimumf128 => self.float_minimum_intrinsic::<Quad>(args, dest)?,
+
             sym::maxnumf16 => self.float_max_intrinsic::<Half>(args, dest)?,
             sym::maxnumf32 => self.float_max_intrinsic::<Single>(args, dest)?,
             sym::maxnumf64 => self.float_max_intrinsic::<Double>(args, dest)?,
             sym::maxnumf128 => self.float_max_intrinsic::<Quad>(args, dest)?,
 
+            sym::maximumf16 => self.float_maximum_intrinsic::<Half>(args, dest)?,
+            sym::maximumf32 => self.float_maximum_intrinsic::<Single>(args, dest)?,
+            sym::maximumf64 => self.float_maximum_intrinsic::<Double>(args, dest)?,
+            sym::maximumf128 => self.float_maximum_intrinsic::<Quad>(args, dest)?,
+
             sym::copysignf16 => self.float_copysign_intrinsic::<Half>(args, dest)?,
             sym::copysignf32 => self.float_copysign_intrinsic::<Single>(args, dest)?,
             sym::copysignf64 => self.float_copysign_intrinsic::<Double>(args, dest)?,
@@ -830,6 +840,38 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         interp_ok(())
     }
 
+    fn float_minimum_intrinsic<F>(
+        &mut self,
+        args: &[OpTy<'tcx, M::Provenance>],
+        dest: &MPlaceTy<'tcx, M::Provenance>,
+    ) -> InterpResult<'tcx, ()>
+    where
+        F: rustc_apfloat::Float + rustc_apfloat::FloatConvert<F> + Into<Scalar<M::Provenance>>,
+    {
+        let a: F = self.read_scalar(&args[0])?.to_float()?;
+        let b: F = self.read_scalar(&args[1])?.to_float()?;
+        let res = a.minimum(b);
+        let res = self.adjust_nan(res, &[a, b]);
+        self.write_scalar(res, dest)?;
+        interp_ok(())
+    }
+
+    fn float_maximum_intrinsic<F>(
+        &mut self,
+        args: &[OpTy<'tcx, M::Provenance>],
+        dest: &MPlaceTy<'tcx, M::Provenance>,
+    ) -> InterpResult<'tcx, ()>
+    where
+        F: rustc_apfloat::Float + rustc_apfloat::FloatConvert<F> + Into<Scalar<M::Provenance>>,
+    {
+        let a: F = self.read_scalar(&args[0])?.to_float()?;
+        let b: F = self.read_scalar(&args[1])?.to_float()?;
+        let res = a.maximum(b);
+        let res = self.adjust_nan(res, &[a, b]);
+        self.write_scalar(res, dest)?;
+        interp_ok(())
+    }
+
     fn float_copysign_intrinsic<F>(
         &mut self,
         args: &[OpTy<'tcx, M::Provenance>],

compiler/rustc_hir_analysis/src/check/intrinsic.rs

@@ -103,10 +103,18 @@ fn intrinsic_operation_unsafety(tcx: TyCtxt<'_>, intrinsic_id: LocalDefId) -> hi
         | sym::minnumf32
         | sym::minnumf64
         | sym::minnumf128
+        | sym::minimumf16
+        | sym::minimumf32
+        | sym::minimumf64
+        | sym::minimumf128
         | sym::maxnumf16
         | sym::maxnumf32
         | sym::maxnumf64
         | sym::maxnumf128
+        | sym::maximumf16
+        | sym::maximumf32
+        | sym::maximumf64
+        | sym::maximumf128
         | sym::rustc_peek
         | sym::type_name
         | sym::forget
@@ -374,11 +382,21 @@ pub(crate) fn check_intrinsic_type(
             sym::minnumf64 => (0, 0, vec![tcx.types.f64, tcx.types.f64], tcx.types.f64),
             sym::minnumf128 => (0, 0, vec![tcx.types.f128, tcx.types.f128], tcx.types.f128),
 
+            sym::minimumf16 => (0, 0, vec![tcx.types.f16, tcx.types.f16], tcx.types.f16),
+            sym::minimumf32 => (0, 0, vec![tcx.types.f32, tcx.types.f32], tcx.types.f32),
+            sym::minimumf64 => (0, 0, vec![tcx.types.f64, tcx.types.f64], tcx.types.f64),
+            sym::minimumf128 => (0, 0, vec![tcx.types.f128, tcx.types.f128], tcx.types.f128),
+
             sym::maxnumf16 => (0, 0, vec![tcx.types.f16, tcx.types.f16], tcx.types.f16),
             sym::maxnumf32 => (0, 0, vec![tcx.types.f32, tcx.types.f32], tcx.types.f32),
             sym::maxnumf64 => (0, 0, vec![tcx.types.f64, tcx.types.f64], tcx.types.f64),
             sym::maxnumf128 => (0, 0, vec![tcx.types.f128, tcx.types.f128], tcx.types.f128),
 
+            sym::maximumf16 => (0, 0, vec![tcx.types.f16, tcx.types.f16], tcx.types.f16),
+            sym::maximumf32 => (0, 0, vec![tcx.types.f32, tcx.types.f32], tcx.types.f32),
+            sym::maximumf64 => (0, 0, vec![tcx.types.f64, tcx.types.f64], tcx.types.f64),
+            sym::maximumf128 => (0, 0, vec![tcx.types.f128, tcx.types.f128], tcx.types.f128),
+
             sym::copysignf16 => (0, 0, vec![tcx.types.f16, tcx.types.f16], tcx.types.f16),
             sym::copysignf32 => (0, 0, vec![tcx.types.f32, tcx.types.f32], tcx.types.f32),
             sym::copysignf64 => (0, 0, vec![tcx.types.f64, tcx.types.f64], tcx.types.f64),

compiler/rustc_span/src/symbol.rs

@@ -1301,6 +1301,10 @@ symbols! {
         match_beginning_vert,
         match_default_bindings,
         matches_macro,
+        maximumf128,
+        maximumf16,
+        maximumf32,
+        maximumf64,
         maxnumf128,
         maxnumf16,
         maxnumf32,
@@ -1335,6 +1339,10 @@ symbols! {
         min_generic_const_args,
         min_specialization,
         min_type_alias_impl_trait,
+        minimumf128,
+        minimumf16,
+        minimumf32,
+        minimumf64,
         minnumf128,
         minnumf16,
         minnumf32,