Auto merge of #95971 - workingjubilee:no-weird-fp-in-const, r=oli-obk

No "weird" floats in const fn {from,to}_bits

I suspect this code is subtly incorrect and that we don't even e.g. use x87-style floats in CTFE, so I don't have to guard against that case. A future PR will be hopefully removing them from concern entirely, anyways. But at the moment I wanted to get this rolling because small questions like that one seem best answered by review.

r? `@oli-obk`
cc `@eddyb` `@thomcc`

Author: bors

library/core/src/num/f32.rs

@@ -449,7 +449,8 @@ impl f32 {
     #[inline]
     #[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
     pub(crate) const fn abs_private(self) -> f32 {
-        f32::from_bits(self.to_bits() & 0x7fff_ffff)
+        // SAFETY: This transmutation is fine. Probably. For the reasons std is using it.
+        unsafe { mem::transmute::<u32, f32>(mem::transmute::<f32, u32>(self) & 0x7fff_ffff) }
     }
 
     /// Returns `true` if this value is positive infinity or negative infinity, and
@@ -472,7 +473,10 @@ impl f32 {
     #[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
     #[inline]
     pub const fn is_infinite(self) -> bool {
-        self.abs_private() == Self::INFINITY
+        // Getting clever with transmutation can result in incorrect answers on some FPUs
+        // FIXME: alter the Rust <-> Rust calling convention to prevent this problem.
+        // See https://github.com/rust-lang/rust/issues/72327
+        (self == f32::INFINITY) | (self == f32::NEG_INFINITY)
     }
 
     /// Returns `true` if this number is neither infinite nor `NaN`.
@@ -568,15 +572,76 @@ impl f32 {
     #[stable(feature = "rust1", since = "1.0.0")]
     #[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
     pub const fn classify(self) -> FpCategory {
+        // A previous implementation tried to only use bitmask-based checks,
+        // using f32::to_bits to transmute the float to its bit repr and match on that.
+        // Unfortunately, floating point numbers can be much worse than that.
+        // This also needs to not result in recursive evaluations of f64::to_bits.
+        //
+        // On some processors, in some cases, LLVM will "helpfully" lower floating point ops,
+        // in spite of a request for them using f32 and f64, to things like x87 operations.
+        // These have an f64's mantissa, but can have a larger than normal exponent.
+        // FIXME(jubilee): Using x87 operations is never necessary in order to function
+        // on x86 processors for Rust-to-Rust calls, so this issue should not happen.
+        // Code generation should be adjusted to use non-C calling conventions, avoiding this.
+        //
+        if self.is_infinite() {
+            // Thus, a value may compare unequal to infinity, despite having a "full" exponent mask.
+            FpCategory::Infinite
+        } else if self.is_nan() {
+            // And it may not be NaN, as it can simply be an "overextended" finite value.
+            FpCategory::Nan
+        } else {
+            // However, std can't simply compare to zero to check for zero, either,
+            // as correctness requires avoiding equality tests that may be Subnormal == -0.0
+            // because it may be wrong under "denormals are zero" and "flush to zero" modes.
+            // Most of std's targets don't use those, but they are used for thumbv7neon.
+            // So, this does use bitpattern matching for the rest.
+
+            // SAFETY: f32 to u32 is fine. Usually.
+            // If classify has gotten this far, the value is definitely in one of these categories.
+            unsafe { f32::partial_classify(self) }
+        }
+    }
+
+    // This doesn't actually return a right answer for NaN on purpose,
+    // seeing as how it cannot correctly discern between a floating point NaN,
+    // and some normal floating point numbers truncated from an x87 FPU.
+    // FIXME(jubilee): This probably could at least answer things correctly for Infinity,
+    // like the f64 version does, but I need to run more checks on how things go on x86.
+    // I fear losing mantissa data that would have answered that differently.
+    //
+    // # Safety
+    // This requires making sure you call this function for values it answers correctly on,
+    // otherwise it returns a wrong answer. This is not important for memory safety per se,
+    // but getting floats correct is important for not accidentally leaking const eval
+    // runtime-deviating logic which may or may not be acceptable.
+    #[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
+    const unsafe fn partial_classify(self) -> FpCategory {
         const EXP_MASK: u32 = 0x7f800000;
         const MAN_MASK: u32 = 0x007fffff;
 
-        let bits = self.to_bits();
-        match (bits & MAN_MASK, bits & EXP_MASK) {
+        // SAFETY: The caller is not asking questions for which this will tell lies.
+        let b = unsafe { mem::transmute::<f32, u32>(self) };
+        match (b & MAN_MASK, b & EXP_MASK) {
             (0, 0) => FpCategory::Zero,
             (_, 0) => FpCategory::Subnormal,
+            _ => FpCategory::Normal,
+        }
+    }
+
+    // This operates on bits, and only bits, so it can ignore concerns about weird FPUs.
+    // FIXME(jubilee): In a just world, this would be the entire impl for classify,
+    // plus a transmute. We do not live in a just world, but we can make it more so.
+    #[rustc_const_unstable(feature = "const_float_classify", issue = "72505")]
+    const fn classify_bits(b: u32) -> FpCategory {
+        const EXP_MASK: u32 = 0x7f800000;
+        const MAN_MASK: u32 = 0x007fffff;
+
+        match (b & MAN_MASK, b & EXP_MASK) {
             (0, EXP_MASK) => FpCategory::Infinite,
             (_, EXP_MASK) => FpCategory::Nan,
+            (0, 0) => FpCategory::Zero,
+            (_, 0) => FpCategory::Subnormal,
             _ => FpCategory::Normal,
         }
     }
@@ -616,7 +681,8 @@ impl f32 {
     pub const fn is_sign_negative(self) -> bool {
         // IEEE754 says: isSignMinus(x) is true if and only if x has negative sign. isSignMinus
         // applies to zeros and NaNs as well.
-        self.to_bits() & 0x8000_0000 != 0
+        // SAFETY: This is just transmuting to get the sign bit, it's fine.
+        unsafe { mem::transmute::<f32, u32>(self) & 0x8000_0000 != 0 }
     }
 
     /// Takes the reciprocal (inverse) of a number, `1/x`.
@@ -831,8 +897,49 @@ impl f32 {
     #[rustc_const_unstable(feature = "const_float_bits_conv", issue = "72447")]
     #[inline]
     pub const fn to_bits(self) -> u32 {
-        // SAFETY: `u32` is a plain old datatype so we can always transmute to it
-        unsafe { mem::transmute(self) }
+        // SAFETY: `u32` is a plain old datatype so we can always transmute to it.
+        // ...sorta.
+        //
+        // It turns out that at runtime, it is possible for a floating point number
+        // to be subject to a floating point mode that alters nonzero subnormal numbers
+        // to zero on reads and writes, aka "denormals are zero" and "flush to zero".
+        // This is not a problem per se, but at least one tier2 platform for Rust
+        // actually exhibits this behavior by default.
+        //
+        // In addition, on x86 targets with SSE or SSE2 disabled and the x87 FPU enabled,
+        // i.e. not soft-float, the way Rust does parameter passing can actually alter
+        // a number that is "not infinity" to have the same exponent as infinity,
+        // in a slightly unpredictable manner.
+        //
+        // And, of course evaluating to a NaN value is fairly nondeterministic.
+        // More precisely: when NaN should be returned is knowable, but which NaN?
+        // So far that's defined by a combination of LLVM and the CPU, not Rust.
+        // This function, however, allows observing the bitstring of a NaN,
+        // thus introspection on CTFE.
+        //
+        // In order to preserve, at least for the moment, const-to-runtime equivalence,
+        // we reject any of these possible situations from happening.
+        #[rustc_const_unstable(feature = "const_float_bits_conv", issue = "72447")]
+        const fn ct_f32_to_u32(ct: f32) -> u32 {
+            match ct.classify() {
+                FpCategory::Nan => {
+                    panic!("const-eval error: cannot use f32::to_bits on a NaN")
+                }
+                FpCategory::Subnormal => {
+                    panic!("const-eval error: cannot use f32::to_bits on a subnormal number")
+                }
+                FpCategory::Infinite | FpCategory::Normal | FpCategory::Zero => {
+                    // SAFETY: We have a normal floating point number. Now we transmute, i.e. do a bitcopy.
+                    unsafe { mem::transmute::<f32, u32>(ct) }
+                }
+            }
+        }
+        // SAFETY: `u32` is a plain old datatype so we can always... uh...
+        // ...look, just pretend you forgot what you just read.
+        // Stability concerns.
+        let rt_f32_to_u32 = |rt| unsafe { mem::transmute::<f32, u32>(rt) };
+        // SAFETY: We use internal implementations that either always work or fail at compile time.
+        unsafe { intrinsics::const_eval_select((self,), ct_f32_to_u32, rt_f32_to_u32) }
     }
 
     /// Raw transmutation from `u32`.
@@ -876,9 +983,51 @@ impl f32 {
     #[must_use]
     #[inline]
     pub const fn from_bits(v: u32) -> Self {
-        // SAFETY: `u32` is a plain old datatype so we can always transmute from it
         // It turns out the safety issues with sNaN were overblown! Hooray!
-        unsafe { mem::transmute(v) }
+        // SAFETY: `u32` is a plain old datatype so we can always transmute from it
+        // ...sorta.
+        //
+        // It turns out that at runtime, it is possible for a floating point number
+        // to be subject to floating point modes that alter nonzero subnormal numbers
+        // to zero on reads and writes, aka "denormals are zero" and "flush to zero".
+        // This is not a problem usually, but at least one tier2 platform for Rust
+        // actually exhibits this behavior by default: thumbv7neon
+        // aka "the Neon FPU in AArch32 state"
+        //
+        // In addition, on x86 targets with SSE or SSE2 disabled and the x87 FPU enabled,
+        // i.e. not soft-float, the way Rust does parameter passing can actually alter
+        // a number that is "not infinity" to have the same exponent as infinity,
+        // in a slightly unpredictable manner.
+        //
+        // And, of course evaluating to a NaN value is fairly nondeterministic.
+        // More precisely: when NaN should be returned is knowable, but which NaN?
+        // So far that's defined by a combination of LLVM and the CPU, not Rust.
+        // This function, however, allows observing the bitstring of a NaN,
+        // thus introspection on CTFE.
+        //
+        // In order to preserve, at least for the moment, const-to-runtime equivalence,
+        // reject any of these possible situations from happening.
+        #[rustc_const_unstable(feature = "const_float_bits_conv", issue = "72447")]
+        const fn ct_u32_to_f32(ct: u32) -> f32 {
+            match f32::classify_bits(ct) {
+                FpCategory::Subnormal => {
+                    panic!("const-eval error: cannot use f32::from_bits on a subnormal number")
+                }
+                FpCategory::Nan => {
+                    panic!("const-eval error: cannot use f32::from_bits on NaN")
+                }
+                FpCategory::Infinite | FpCategory::Normal | FpCategory::Zero => {
+                    // SAFETY: It's not a frumious number
+                    unsafe { mem::transmute::<u32, f32>(ct) }
+                }
+            }
+        }
+        // SAFETY: `u32` is a plain old datatype so we can always... uh...
+        // ...look, just pretend you forgot what you just read.
+        // Stability concerns.
+        let rt_u32_to_f32 = |rt| unsafe { mem::transmute::<u32, f32>(rt) };
+        // SAFETY: We use internal implementations that either always work or fail at compile time.
+        unsafe { intrinsics::const_eval_select((v,), ct_u32_to_f32, rt_u32_to_f32) }
     }
 
     /// Return the memory representation of this floating point number as a byte array in