From e6d570241fd2e12fa143938e2b17ba6ec31001da Mon Sep 17 00:00:00 2001 From: Trevor Gross Date: Fri, 12 Jul 2024 03:54:54 -0400 Subject: [PATCH 01/28] Specify the integer type of the `powi` LLVM intrinsic Since LLVM (4c7f820b2b20, "Update @llvm.powi to handle different int sizes for the exponent"), the size of the integer can be specified for the `powi` intrinsic. Make use of this so it is more obvious that integer size is consistent across all float types. This feature is available since LLVM 13 (October 2021). Based on bootstrap we currently support >= 17.0, so there should be no support problems. --- compiler/rustc_codegen_llvm/src/context.rs | 8 ++++---- compiler/rustc_codegen_llvm/src/intrinsic.rs | 8 ++++---- 2 files changed, 8 insertions(+), 8 deletions(-) diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs index ea930421b5869..14540d41e7c62 100644 --- a/compiler/rustc_codegen_llvm/src/context.rs +++ b/compiler/rustc_codegen_llvm/src/context.rs @@ -775,10 +775,10 @@ impl<'ll> CodegenCx<'ll, '_> { ifn!("llvm.debugtrap", fn() -> void); ifn!("llvm.frameaddress", fn(t_i32) -> ptr); - ifn!("llvm.powi.f16", fn(t_f16, t_i32) -> t_f16); - ifn!("llvm.powi.f32", fn(t_f32, t_i32) -> t_f32); - ifn!("llvm.powi.f64", fn(t_f64, t_i32) -> t_f64); - ifn!("llvm.powi.f128", fn(t_f128, t_i32) -> t_f128); + ifn!("llvm.powi.f16.i32", fn(t_f16, t_i32) -> t_f16); + ifn!("llvm.powi.f32.i32", fn(t_f32, t_i32) -> t_f32); + ifn!("llvm.powi.f64.i32", fn(t_f64, t_i32) -> t_f64); + ifn!("llvm.powi.f128.i32", fn(t_f128, t_i32) -> t_f128); ifn!("llvm.pow.f16", fn(t_f16, t_f16) -> t_f16); ifn!("llvm.pow.f32", fn(t_f32, t_f32) -> t_f32); diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs index 040de1c7dd715..57d5f6fdf503f 100644 --- a/compiler/rustc_codegen_llvm/src/intrinsic.rs +++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs @@ -35,10 +35,10 @@ fn get_simple_intrinsic<'ll>( sym::sqrtf64 => "llvm.sqrt.f64", sym::sqrtf128 => "llvm.sqrt.f128", - sym::powif16 => "llvm.powi.f16", - sym::powif32 => "llvm.powi.f32", - sym::powif64 => "llvm.powi.f64", - sym::powif128 => "llvm.powi.f128", + sym::powif16 => "llvm.powi.f16.i32", + sym::powif32 => "llvm.powi.f32.i32", + sym::powif64 => "llvm.powi.f64.i32", + sym::powif128 => "llvm.powi.f128.i32", sym::sinf16 => "llvm.sin.f16", sym::sinf32 => "llvm.sin.f32", From 82b40c4d8e84ab63ff788b00c80b252887f78d86 Mon Sep 17 00:00:00 2001 From: Trevor Gross Date: Tue, 18 Jun 2024 21:39:00 -0500 Subject: [PATCH 02/28] Add math intrinsics for `f16` and `f128` These already exist in the compiler. Expose them in core so we can add their library functions. --- library/core/src/intrinsics.rs | 289 +++++++++++++++++++++++++++++++++ 1 file changed, 289 insertions(+) diff --git a/library/core/src/intrinsics.rs b/library/core/src/intrinsics.rs index 13c7f0855d8f0..f230ca612cfe8 100644 --- a/library/core/src/intrinsics.rs +++ b/library/core/src/intrinsics.rs @@ -1528,6 +1528,12 @@ extern "rust-intrinsic" { #[rustc_diagnostic_item = "intrinsics_unaligned_volatile_store"] pub fn unaligned_volatile_store(dst: *mut T, val: T); + /// Returns the square root of an `f16` + /// + /// The stabilized version of this intrinsic is + /// [`f16::sqrt`](../../std/primitive.f16.html#method.sqrt) + #[rustc_nounwind] + pub fn sqrtf16(x: f16) -> f16; /// Returns the square root of an `f32` /// /// The stabilized version of this intrinsic is @@ -1540,6 +1546,12 @@ extern "rust-intrinsic" { /// [`f64::sqrt`](../../std/primitive.f64.html#method.sqrt) #[rustc_nounwind] pub fn sqrtf64(x: f64) -> f64; + /// Returns the square root of an `f128` + /// + /// The stabilized version of this intrinsic is + /// [`f128::sqrt`](../../std/primitive.f128.html#method.sqrt) + #[rustc_nounwind] + pub fn sqrtf128(x: f128) -> f128; /// Raises an `f16` to an integer power. /// @@ -1566,6 +1578,12 @@ extern "rust-intrinsic" { #[rustc_nounwind] pub fn powif128(a: f128, x: i32) -> f128; + /// Returns the sine of an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::sin`](../../std/primitive.f16.html#method.sin) + #[rustc_nounwind] + pub fn sinf16(x: f16) -> f16; /// Returns the sine of an `f32`. /// /// The stabilized version of this intrinsic is @@ -1578,7 +1596,19 @@ extern "rust-intrinsic" { /// [`f64::sin`](../../std/primitive.f64.html#method.sin) #[rustc_nounwind] pub fn sinf64(x: f64) -> f64; + /// Returns the sine of an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::sin`](../../std/primitive.f128.html#method.sin) + #[rustc_nounwind] + pub fn sinf128(x: f128) -> f128; + /// Returns the cosine of an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::cos`](../../std/primitive.f16.html#method.cos) + #[rustc_nounwind] + pub fn cosf16(x: f16) -> f16; /// Returns the cosine of an `f32`. /// /// The stabilized version of this intrinsic is @@ -1591,7 +1621,19 @@ extern "rust-intrinsic" { /// [`f64::cos`](../../std/primitive.f64.html#method.cos) #[rustc_nounwind] pub fn cosf64(x: f64) -> f64; + /// Returns the cosine of an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::cos`](../../std/primitive.f128.html#method.cos) + #[rustc_nounwind] + pub fn cosf128(x: f128) -> f128; + /// Raises an `f16` to an `f16` power. + /// + /// The stabilized version of this intrinsic is + /// [`f16::powf`](../../std/primitive.f16.html#method.powf) + #[rustc_nounwind] + pub fn powf16(a: f16, x: f16) -> f16; /// Raises an `f32` to an `f32` power. /// /// The stabilized version of this intrinsic is @@ -1604,7 +1646,19 @@ extern "rust-intrinsic" { /// [`f64::powf`](../../std/primitive.f64.html#method.powf) #[rustc_nounwind] pub fn powf64(a: f64, x: f64) -> f64; + /// Raises an `f128` to an `f128` power. + /// + /// The stabilized version of this intrinsic is + /// [`f128::powf`](../../std/primitive.f128.html#method.powf) + #[rustc_nounwind] + pub fn powf128(a: f128, x: f128) -> f128; + /// Returns the exponential of an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::exp`](../../std/primitive.f16.html#method.exp) + #[rustc_nounwind] + pub fn expf16(x: f16) -> f16; /// Returns the exponential of an `f32`. /// /// The stabilized version of this intrinsic is @@ -1617,7 +1671,19 @@ extern "rust-intrinsic" { /// [`f64::exp`](../../std/primitive.f64.html#method.exp) #[rustc_nounwind] pub fn expf64(x: f64) -> f64; + /// Returns the exponential of an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::exp`](../../std/primitive.f128.html#method.exp) + #[rustc_nounwind] + pub fn expf128(x: f128) -> f128; + /// Returns 2 raised to the power of an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::exp2`](../../std/primitive.f16.html#method.exp2) + #[rustc_nounwind] + pub fn exp2f16(x: f16) -> f16; /// Returns 2 raised to the power of an `f32`. /// /// The stabilized version of this intrinsic is @@ -1630,7 +1696,19 @@ extern "rust-intrinsic" { /// [`f64::exp2`](../../std/primitive.f64.html#method.exp2) #[rustc_nounwind] pub fn exp2f64(x: f64) -> f64; + /// Returns 2 raised to the power of an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::exp2`](../../std/primitive.f128.html#method.exp2) + #[rustc_nounwind] + pub fn exp2f128(x: f128) -> f128; + /// Returns the natural logarithm of an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::ln`](../../std/primitive.f16.html#method.ln) + #[rustc_nounwind] + pub fn logf16(x: f16) -> f16; /// Returns the natural logarithm of an `f32`. /// /// The stabilized version of this intrinsic is @@ -1643,7 +1721,19 @@ extern "rust-intrinsic" { /// [`f64::ln`](../../std/primitive.f64.html#method.ln) #[rustc_nounwind] pub fn logf64(x: f64) -> f64; + /// Returns the natural logarithm of an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::ln`](../../std/primitive.f128.html#method.ln) + #[rustc_nounwind] + pub fn logf128(x: f128) -> f128; + /// Returns the base 10 logarithm of an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::log10`](../../std/primitive.f16.html#method.log10) + #[rustc_nounwind] + pub fn log10f16(x: f16) -> f16; /// Returns the base 10 logarithm of an `f32`. /// /// The stabilized version of this intrinsic is @@ -1656,7 +1746,19 @@ extern "rust-intrinsic" { /// [`f64::log10`](../../std/primitive.f64.html#method.log10) #[rustc_nounwind] pub fn log10f64(x: f64) -> f64; + /// Returns the base 10 logarithm of an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::log10`](../../std/primitive.f128.html#method.log10) + #[rustc_nounwind] + pub fn log10f128(x: f128) -> f128; + /// Returns the base 2 logarithm of an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::log2`](../../std/primitive.f16.html#method.log2) + #[rustc_nounwind] + pub fn log2f16(x: f16) -> f16; /// Returns the base 2 logarithm of an `f32`. /// /// The stabilized version of this intrinsic is @@ -1669,7 +1771,19 @@ extern "rust-intrinsic" { /// [`f64::log2`](../../std/primitive.f64.html#method.log2) #[rustc_nounwind] pub fn log2f64(x: f64) -> f64; + /// Returns the base 2 logarithm of an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::log2`](../../std/primitive.f128.html#method.log2) + #[rustc_nounwind] + pub fn log2f128(x: f128) -> f128; + /// Returns `a * b + c` for `f16` values. + /// + /// The stabilized version of this intrinsic is + /// [`f16::mul_add`](../../std/primitive.f16.html#method.mul_add) + #[rustc_nounwind] + pub fn fmaf16(a: f16, b: f16, c: f16) -> f16; /// Returns `a * b + c` for `f32` values. /// /// The stabilized version of this intrinsic is @@ -1682,7 +1796,19 @@ extern "rust-intrinsic" { /// [`f64::mul_add`](../../std/primitive.f64.html#method.mul_add) #[rustc_nounwind] pub fn fmaf64(a: f64, b: f64, c: f64) -> f64; + /// Returns `a * b + c` for `f128` values. + /// + /// The stabilized version of this intrinsic is + /// [`f128::mul_add`](../../std/primitive.f128.html#method.mul_add) + #[rustc_nounwind] + pub fn fmaf128(a: f128, b: f128, c: f128) -> f128; + /// Returns the absolute value of an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::abs`](../../std/primitive.f16.html#method.abs) + #[rustc_nounwind] + pub fn fabsf16(x: f16) -> f16; /// Returns the absolute value of an `f32`. /// /// The stabilized version of this intrinsic is @@ -1695,7 +1821,25 @@ extern "rust-intrinsic" { /// [`f64::abs`](../../std/primitive.f64.html#method.abs) #[rustc_nounwind] pub fn fabsf64(x: f64) -> f64; + /// Returns the absolute value of an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::abs`](../../std/primitive.f128.html#method.abs) + #[rustc_nounwind] + pub fn fabsf128(x: f128) -> f128; + /// Returns the minimum of two `f16` values. + /// + /// Note that, unlike most intrinsics, this is safe to call; + /// it does not require an `unsafe` block. + /// Therefore, implementations must not require the user to uphold + /// any safety invariants. + /// + /// The stabilized version of this intrinsic is + /// [`f16::min`] + #[rustc_safe_intrinsic] + #[rustc_nounwind] + pub fn minnumf16(x: f16, y: f16) -> f16; /// Returns the minimum of two `f32` values. /// /// Note that, unlike most intrinsics, this is safe to call; @@ -1720,6 +1864,31 @@ extern "rust-intrinsic" { #[rustc_safe_intrinsic] #[rustc_nounwind] pub fn minnumf64(x: f64, y: f64) -> f64; + /// Returns the minimum of two `f128` values. + /// + /// Note that, unlike most intrinsics, this is safe to call; + /// it does not require an `unsafe` block. + /// Therefore, implementations must not require the user to uphold + /// any safety invariants. + /// + /// The stabilized version of this intrinsic is + /// [`f128::min`] + #[rustc_safe_intrinsic] + #[rustc_nounwind] + pub fn minnumf128(x: f128, y: f128) -> f128; + + /// Returns the maximum of two `f16` values. + /// + /// Note that, unlike most intrinsics, this is safe to call; + /// it does not require an `unsafe` block. + /// Therefore, implementations must not require the user to uphold + /// any safety invariants. + /// + /// The stabilized version of this intrinsic is + /// [`f16::max`] + #[rustc_safe_intrinsic] + #[rustc_nounwind] + pub fn maxnumf16(x: f16, y: f16) -> f16; /// Returns the maximum of two `f32` values. /// /// Note that, unlike most intrinsics, this is safe to call; @@ -1744,7 +1913,25 @@ extern "rust-intrinsic" { #[rustc_safe_intrinsic] #[rustc_nounwind] pub fn maxnumf64(x: f64, y: f64) -> f64; + /// Returns the maximum of two `f128` values. + /// + /// Note that, unlike most intrinsics, this is safe to call; + /// it does not require an `unsafe` block. + /// Therefore, implementations must not require the user to uphold + /// any safety invariants. + /// + /// The stabilized version of this intrinsic is + /// [`f128::max`] + #[rustc_safe_intrinsic] + #[rustc_nounwind] + pub fn maxnumf128(x: f128, y: f128) -> f128; + /// Copies the sign from `y` to `x` for `f16` values. + /// + /// The stabilized version of this intrinsic is + /// [`f16::copysign`](../../std/primitive.f16.html#method.copysign) + #[rustc_nounwind] + pub fn copysignf16(x: f16, y: f16) -> f16; /// Copies the sign from `y` to `x` for `f32` values. /// /// The stabilized version of this intrinsic is @@ -1757,7 +1944,19 @@ extern "rust-intrinsic" { /// [`f64::copysign`](../../std/primitive.f64.html#method.copysign) #[rustc_nounwind] pub fn copysignf64(x: f64, y: f64) -> f64; + /// Copies the sign from `y` to `x` for `f128` values. + /// + /// The stabilized version of this intrinsic is + /// [`f128::copysign`](../../std/primitive.f128.html#method.copysign) + #[rustc_nounwind] + pub fn copysignf128(x: f128, y: f128) -> f128; + /// Returns the largest integer less than or equal to an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::floor`](../../std/primitive.f16.html#method.floor) + #[rustc_nounwind] + pub fn floorf16(x: f16) -> f16; /// Returns the largest integer less than or equal to an `f32`. /// /// The stabilized version of this intrinsic is @@ -1770,7 +1969,19 @@ extern "rust-intrinsic" { /// [`f64::floor`](../../std/primitive.f64.html#method.floor) #[rustc_nounwind] pub fn floorf64(x: f64) -> f64; + /// Returns the largest integer less than or equal to an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::floor`](../../std/primitive.f128.html#method.floor) + #[rustc_nounwind] + pub fn floorf128(x: f128) -> f128; + /// Returns the smallest integer greater than or equal to an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::ceil`](../../std/primitive.f16.html#method.ceil) + #[rustc_nounwind] + pub fn ceilf16(x: f16) -> f16; /// Returns the smallest integer greater than or equal to an `f32`. /// /// The stabilized version of this intrinsic is @@ -1783,7 +1994,19 @@ extern "rust-intrinsic" { /// [`f64::ceil`](../../std/primitive.f64.html#method.ceil) #[rustc_nounwind] pub fn ceilf64(x: f64) -> f64; + /// Returns the smallest integer greater than or equal to an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::ceil`](../../std/primitive.f128.html#method.ceil) + #[rustc_nounwind] + pub fn ceilf128(x: f128) -> f128; + /// Returns the integer part of an `f16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::trunc`](../../std/primitive.f16.html#method.trunc) + #[rustc_nounwind] + pub fn truncf16(x: f16) -> f16; /// Returns the integer part of an `f32`. /// /// The stabilized version of this intrinsic is @@ -1796,7 +2019,25 @@ extern "rust-intrinsic" { /// [`f64::trunc`](../../std/primitive.f64.html#method.trunc) #[rustc_nounwind] pub fn truncf64(x: f64) -> f64; + /// Returns the integer part of an `f128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::trunc`](../../std/primitive.f128.html#method.trunc) + #[rustc_nounwind] + pub fn truncf128(x: f128) -> f128; + /// Returns the nearest integer to an `f16`. Changing the rounding mode is not possible in Rust, + /// so this rounds half-way cases to the number with an even least significant digit. + /// + /// May raise an inexact floating-point exception if the argument is not an integer. + /// However, Rust assumes floating-point exceptions cannot be observed, so these exceptions + /// cannot actually be utilized from Rust code. + /// In other words, this intrinsic is equivalent in behavior to `nearbyintf16` and `roundevenf16`. + /// + /// The stabilized version of this intrinsic is + /// [`f16::round_ties_even`](../../std/primitive.f16.html#method.round_ties_even) + #[rustc_nounwind] + pub fn rintf16(x: f16) -> f16; /// Returns the nearest integer to an `f32`. Changing the rounding mode is not possible in Rust, /// so this rounds half-way cases to the number with an even least significant digit. /// @@ -1821,7 +2062,25 @@ extern "rust-intrinsic" { /// [`f64::round_ties_even`](../../std/primitive.f64.html#method.round_ties_even) #[rustc_nounwind] pub fn rintf64(x: f64) -> f64; + /// Returns the nearest integer to an `f128`. Changing the rounding mode is not possible in Rust, + /// so this rounds half-way cases to the number with an even least significant digit. + /// + /// May raise an inexact floating-point exception if the argument is not an integer. + /// However, Rust assumes floating-point exceptions cannot be observed, so these exceptions + /// cannot actually be utilized from Rust code. + /// In other words, this intrinsic is equivalent in behavior to `nearbyintf128` and `roundevenf128`. + /// + /// The stabilized version of this intrinsic is + /// [`f128::round_ties_even`](../../std/primitive.f128.html#method.round_ties_even) + #[rustc_nounwind] + pub fn rintf128(x: f128) -> f128; + /// Returns the nearest integer to an `f16`. Changing the rounding mode is not possible in Rust, + /// so this rounds half-way cases to the number with an even least significant digit. + /// + /// This intrinsic does not have a stable counterpart. + #[rustc_nounwind] + pub fn nearbyintf16(x: f16) -> f16; /// Returns the nearest integer to an `f32`. Changing the rounding mode is not possible in Rust, /// so this rounds half-way cases to the number with an even least significant digit. /// @@ -1834,7 +2093,19 @@ extern "rust-intrinsic" { /// This intrinsic does not have a stable counterpart. #[rustc_nounwind] pub fn nearbyintf64(x: f64) -> f64; + /// Returns the nearest integer to an `f128`. Changing the rounding mode is not possible in Rust, + /// so this rounds half-way cases to the number with an even least significant digit. + /// + /// This intrinsic does not have a stable counterpart. + #[rustc_nounwind] + pub fn nearbyintf128(x: f128) -> f128; + /// Returns the nearest integer to an `f16`. Rounds half-way cases away from zero. + /// + /// The stabilized version of this intrinsic is + /// [`f16::round`](../../std/primitive.f16.html#method.round) + #[rustc_nounwind] + pub fn roundf16(x: f16) -> f16; /// Returns the nearest integer to an `f32`. Rounds half-way cases away from zero. /// /// The stabilized version of this intrinsic is @@ -1847,7 +2118,19 @@ extern "rust-intrinsic" { /// [`f64::round`](../../std/primitive.f64.html#method.round) #[rustc_nounwind] pub fn roundf64(x: f64) -> f64; + /// Returns the nearest integer to an `f128`. Rounds half-way cases away from zero. + /// + /// The stabilized version of this intrinsic is + /// [`f128::round`](../../std/primitive.f128.html#method.round) + #[rustc_nounwind] + pub fn roundf128(x: f128) -> f128; + /// Returns the nearest integer to an `f16`. Rounds half-way cases to the number + /// with an even least significant digit. + /// + /// This intrinsic does not have a stable counterpart. + #[rustc_nounwind] + pub fn roundevenf16(x: f16) -> f16; /// Returns the nearest integer to an `f32`. Rounds half-way cases to the number /// with an even least significant digit. /// @@ -1860,6 +2143,12 @@ extern "rust-intrinsic" { /// This intrinsic does not have a stable counterpart. #[rustc_nounwind] pub fn roundevenf64(x: f64) -> f64; + /// Returns the nearest integer to an `f128`. Rounds half-way cases to the number + /// with an even least significant digit. + /// + /// This intrinsic does not have a stable counterpart. + #[rustc_nounwind] + pub fn roundevenf128(x: f128) -> f128; /// Float addition that allows optimizations based on algebraic rules. /// May assume inputs are finite. From fc43c01417f6351b52d8bd2dc2ba9f9fd3ede14f Mon Sep 17 00:00:00 2001 From: Trevor Gross Date: Thu, 27 Jun 2024 04:13:25 -0500 Subject: [PATCH 03/28] Add math functions for `f16` and `f128` This adds missing functions for math operations on the new float types. Platform support is pretty spotty at this point, since even platforms with generally good support can be missing math functions. `std/build.rs` is updated to reflect this. --- library/std/build.rs | 37 + library/std/src/f128.rs | 1300 ++++++++++++++++++++++++++++++++- library/std/src/f128/tests.rs | 450 +++++++++++- library/std/src/f16.rs | 1296 +++++++++++++++++++++++++++++++- library/std/src/f16/tests.rs | 444 ++++++++++- library/std/src/macros.rs | 2 +- library/std/src/sys/cmath.rs | 15 + 7 files changed, 3452 insertions(+), 92 deletions(-) diff --git a/library/std/build.rs b/library/std/build.rs index 9b58dd53ba20a..18ca7b512a9b6 100644 --- a/library/std/build.rs +++ b/library/std/build.rs @@ -85,6 +85,11 @@ fn main() { println!("cargo:rustc-check-cfg=cfg(reliable_f16)"); println!("cargo:rustc-check-cfg=cfg(reliable_f128)"); + // This is a step beyond only having the types and basic functions available. Math functions + // aren't consistently available or correct. + println!("cargo:rustc-check-cfg=cfg(reliable_f16_math)"); + println!("cargo:rustc-check-cfg=cfg(reliable_f128_math)"); + let has_reliable_f16 = match (target_arch.as_str(), target_os.as_str()) { // Selection failure until recent LLVM // FIXME(llvm19): can probably be removed at the version bump @@ -130,10 +135,42 @@ fn main() { _ => false, }; + // These are currently empty, but will fill up as some platforms move from completely + // unreliable to reliable basics but unreliable math. + + // LLVM is currenlty adding missing routines, + let has_reliable_f16_math = has_reliable_f16 + && match (target_arch.as_str(), target_os.as_str()) { + // Currently nothing special. Hooray! + // This will change as platforms gain better better support for standard ops but math + // lags behind. + _ => true, + }; + + let has_reliable_f128_math = has_reliable_f128 + && match (target_arch.as_str(), target_os.as_str()) { + // LLVM lowers `fp128` math to `long double` symbols even on platforms where + // `long double` is not IEEE binary128. See + // . + // + // This rules out anything that doesn't have `long double` = `binary128`; <= 32 bits + // (ld is `f64`), anything other than Linux (Windows and MacOS use `f64`), and `x86` + // (ld is 80-bit extended precision). + ("x86_64", _) => false, + (_, "linux") if target_pointer_width == 64 => true, + _ => false, + }; + if has_reliable_f16 { println!("cargo:rustc-cfg=reliable_f16"); } if has_reliable_f128 { println!("cargo:rustc-cfg=reliable_f128"); } + if has_reliable_f16_math { + println!("cargo:rustc-cfg=reliable_f16_math"); + } + if has_reliable_f128_math { + println!("cargo:rustc-cfg=reliable_f128_math"); + } } diff --git a/library/std/src/f128.rs b/library/std/src/f128.rs index a5b00d57cefdd..f6df6259137bf 100644 --- a/library/std/src/f128.rs +++ b/library/std/src/f128.rs @@ -12,25 +12,180 @@ pub use core::f128::consts; #[cfg(not(test))] use crate::intrinsics; +#[cfg(not(test))] +use crate::sys::cmath; #[cfg(not(test))] impl f128 { - /// Raises a number to an integer power. + /// Returns the largest integer less than or equal to `self`. /// - /// Using this function is generally faster than using `powf`. - /// It might have a different sequence of rounding operations than `powf`, - /// so the results are not guaranteed to agree. + /// This function always returns the precise result. /// - /// # Unspecified precision + /// # Examples /// - /// The precision of this function is non-deterministic. This means it varies by platform, Rust version, and - /// can even differ within the same execution from one invocation to the next. + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = 3.7_f128; + /// let g = 3.0_f128; + /// let h = -3.7_f128; + /// + /// assert_eq!(f.floor(), 3.0); + /// assert_eq!(g.floor(), 3.0); + /// assert_eq!(h.floor(), -4.0); + /// # } + /// ``` #[inline] #[rustc_allow_incoherent_impl] #[unstable(feature = "f128", issue = "116909")] #[must_use = "method returns a new number and does not mutate the original value"] - pub fn powi(self, n: i32) -> f128 { - unsafe { intrinsics::powif128(self, n) } + pub fn floor(self) -> f128 { + unsafe { intrinsics::floorf128(self) } + } + + /// Returns the smallest integer greater than or equal to `self`. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = 3.01_f128; + /// let g = 4.0_f128; + /// + /// assert_eq!(f.ceil(), 4.0); + /// assert_eq!(g.ceil(), 4.0); + /// # } + /// ``` + #[inline] + #[doc(alias = "ceiling")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn ceil(self) -> f128 { + unsafe { intrinsics::ceilf128(self) } + } + + /// Returns the nearest integer to `self`. If a value is half-way between two + /// integers, round away from `0.0`. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = 3.3_f128; + /// let g = -3.3_f128; + /// let h = -3.7_f128; + /// let i = 3.5_f128; + /// let j = 4.5_f128; + /// + /// assert_eq!(f.round(), 3.0); + /// assert_eq!(g.round(), -3.0); + /// assert_eq!(h.round(), -4.0); + /// assert_eq!(i.round(), 4.0); + /// assert_eq!(j.round(), 5.0); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn round(self) -> f128 { + unsafe { intrinsics::roundf128(self) } + } + + /// Returns the nearest integer to a number. Rounds half-way cases to the number + /// with an even least significant digit. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = 3.3_f128; + /// let g = -3.3_f128; + /// let h = 3.5_f128; + /// let i = 4.5_f128; + /// + /// assert_eq!(f.round_ties_even(), 3.0); + /// assert_eq!(g.round_ties_even(), -3.0); + /// assert_eq!(h.round_ties_even(), 4.0); + /// assert_eq!(i.round_ties_even(), 4.0); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn round_ties_even(self) -> f128 { + unsafe { intrinsics::rintf128(self) } + } + + /// Returns the integer part of `self`. + /// This means that non-integer numbers are always truncated towards zero. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = 3.7_f128; + /// let g = 3.0_f128; + /// let h = -3.7_f128; + /// + /// assert_eq!(f.trunc(), 3.0); + /// assert_eq!(g.trunc(), 3.0); + /// assert_eq!(h.trunc(), -3.0); + /// # } + /// ``` + #[inline] + #[doc(alias = "truncate")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn trunc(self) -> f128 { + unsafe { intrinsics::truncf128(self) } + } + + /// Returns the fractional part of `self`. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 3.6_f128; + /// let y = -3.6_f128; + /// let abs_difference_x = (x.fract() - 0.6).abs(); + /// let abs_difference_y = (y.fract() - (-0.6)).abs(); + /// + /// assert!(abs_difference_x <= f128::EPSILON); + /// assert!(abs_difference_y <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn fract(self) -> f128 { + self - self.trunc() } /// Computes the absolute value of `self`. @@ -41,7 +196,7 @@ impl f128 { /// /// ``` /// #![feature(f128)] - /// # #[cfg(reliable_f128)] { // FIXME(f16_f128): reliable_f128 + /// # #[cfg(reliable_f128)] { /// /// let x = 3.5_f128; /// let y = -3.5_f128; @@ -61,4 +216,1129 @@ impl f128 { // We don't do this now because LLVM has lowering bugs for f128 math. Self::from_bits(self.to_bits() & !(1 << 127)) } + + /// Returns a number that represents the sign of `self`. + /// + /// - `1.0` if the number is positive, `+0.0` or `INFINITY` + /// - `-1.0` if the number is negative, `-0.0` or `NEG_INFINITY` + /// - NaN if the number is NaN + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = 3.5_f128; + /// + /// assert_eq!(f.signum(), 1.0); + /// assert_eq!(f128::NEG_INFINITY.signum(), -1.0); + /// + /// assert!(f128::NAN.signum().is_nan()); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn signum(self) -> f128 { + if self.is_nan() { Self::NAN } else { 1.0_f128.copysign(self) } + } + + /// Returns a number composed of the magnitude of `self` and the sign of + /// `sign`. + /// + /// Equal to `self` if the sign of `self` and `sign` are the same, otherwise + /// equal to `-self`. If `self` is a NaN, then a NaN with the sign bit of + /// `sign` is returned. Note, however, that conserving the sign bit on NaN + /// across arithmetical operations is not generally guaranteed. + /// See [explanation of NaN as a special value](primitive@f128) for more info. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = 3.5_f128; + /// + /// assert_eq!(f.copysign(0.42), 3.5_f128); + /// assert_eq!(f.copysign(-0.42), -3.5_f128); + /// assert_eq!((-f).copysign(0.42), 3.5_f128); + /// assert_eq!((-f).copysign(-0.42), -3.5_f128); + /// + /// assert!(f128::NAN.copysign(1.0).is_nan()); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn copysign(self, sign: f128) -> f128 { + unsafe { intrinsics::copysignf128(self, sign) } + } + + /// Fused multiply-add. Computes `(self * a) + b` with only one rounding + /// error, yielding a more accurate result than an unfused multiply-add. + /// + /// Using `mul_add` *may* be more performant than an unfused multiply-add if + /// the target architecture has a dedicated `fma` CPU instruction. However, + /// this is not always true, and will be heavily dependant on designing + /// algorithms with specific target hardware in mind. + /// + /// # Precision + /// + /// The result of this operation is guaranteed to be the rounded + /// infinite-precision result. It is specified by IEEE 754 as + /// `fusedMultiplyAdd` and guaranteed not to change. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let m = 10.0_f128; + /// let x = 4.0_f128; + /// let b = 60.0_f128; + /// + /// assert_eq!(m.mul_add(x, b), 100.0); + /// assert_eq!(m * x + b, 100.0); + /// + /// let one_plus_eps = 1.0_f128 + f128::EPSILON; + /// let one_minus_eps = 1.0_f128 - f128::EPSILON; + /// let minus_one = -1.0_f128; + /// + /// // The exact result (1 + eps) * (1 - eps) = 1 - eps * eps. + /// assert_eq!(one_plus_eps.mul_add(one_minus_eps, minus_one), -f128::EPSILON * f128::EPSILON); + /// // Different rounding with the non-fused multiply and add. + /// assert_eq!(one_plus_eps * one_minus_eps + minus_one, 0.0); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn mul_add(self, a: f128, b: f128) -> f128 { + unsafe { intrinsics::fmaf128(self, a, b) } + } + + /// Calculates Euclidean division, the matching method for `rem_euclid`. + /// + /// This computes the integer `n` such that + /// `self = n * rhs + self.rem_euclid(rhs)`. + /// In other words, the result is `self / rhs` rounded to the integer `n` + /// such that `self >= n * rhs`. + /// + /// # Precision + /// + /// The result of this operation is guaranteed to be the rounded + /// infinite-precision result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let a: f128 = 7.0; + /// let b = 4.0; + /// assert_eq!(a.div_euclid(b), 1.0); // 7.0 > 4.0 * 1.0 + /// assert_eq!((-a).div_euclid(b), -2.0); // -7.0 >= 4.0 * -2.0 + /// assert_eq!(a.div_euclid(-b), -1.0); // 7.0 >= -4.0 * -1.0 + /// assert_eq!((-a).div_euclid(-b), 2.0); // -7.0 >= -4.0 * 2.0 + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn div_euclid(self, rhs: f128) -> f128 { + let q = (self / rhs).trunc(); + if self % rhs < 0.0 { + return if rhs > 0.0 { q - 1.0 } else { q + 1.0 }; + } + q + } + + /// Calculates the least nonnegative remainder of `self (mod rhs)`. + /// + /// In particular, the return value `r` satisfies `0.0 <= r < rhs.abs()` in + /// most cases. However, due to a floating point round-off error it can + /// result in `r == rhs.abs()`, violating the mathematical definition, if + /// `self` is much smaller than `rhs.abs()` in magnitude and `self < 0.0`. + /// This result is not an element of the function's codomain, but it is the + /// closest floating point number in the real numbers and thus fulfills the + /// property `self == self.div_euclid(rhs) * rhs + self.rem_euclid(rhs)` + /// approximately. + /// + /// # Precision + /// + /// The result of this operation is guaranteed to be the rounded + /// infinite-precision result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let a: f128 = 7.0; + /// let b = 4.0; + /// assert_eq!(a.rem_euclid(b), 3.0); + /// assert_eq!((-a).rem_euclid(b), 1.0); + /// assert_eq!(a.rem_euclid(-b), 3.0); + /// assert_eq!((-a).rem_euclid(-b), 1.0); + /// // limitation due to round-off error + /// assert!((-f128::EPSILON).rem_euclid(3.0) != 0.0); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[doc(alias = "modulo", alias = "mod")] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn rem_euclid(self, rhs: f128) -> f128 { + let r = self % rhs; + if r < 0.0 { r + rhs.abs() } else { r } + } + + /// Raises a number to an integer power. + /// + /// Using this function is generally faster than using `powf`. + /// It might have a different sequence of rounding operations than `powf`, + /// so the results are not guaranteed to agree. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn powi(self, n: i32) -> f128 { + unsafe { intrinsics::powif128(self, n) } + } + + /// Raises a number to a floating point power. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 2.0_f128; + /// let abs_difference = (x.powf(2.0) - (x * x)).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn powf(self, n: f128) -> f128 { + unsafe { intrinsics::powf128(self, n) } + } + + /// Returns the square root of a number. + /// + /// Returns NaN if `self` is a negative number other than `-0.0`. + /// + /// # Precision + /// + /// The result of this operation is guaranteed to be the rounded + /// infinite-precision result. It is specified by IEEE 754 as `squareRoot` + /// and guaranteed not to change. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let positive = 4.0_f128; + /// let negative = -4.0_f128; + /// let negative_zero = -0.0_f128; + /// + /// assert_eq!(positive.sqrt(), 2.0); + /// assert!(negative.sqrt().is_nan()); + /// assert!(negative_zero.sqrt() == negative_zero); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn sqrt(self) -> f128 { + unsafe { intrinsics::sqrtf128(self) } + } + + /// Returns `e^(self)`, (the exponential function). + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let one = 1.0f128; + /// // e^1 + /// let e = one.exp(); + /// + /// // ln(e) - 1 == 0 + /// let abs_difference = (e.ln() - 1.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn exp(self) -> f128 { + unsafe { intrinsics::expf128(self) } + } + + /// Returns `2^(self)`. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = 2.0f128; + /// + /// // 2^2 - 4 == 0 + /// let abs_difference = (f.exp2() - 4.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn exp2(self) -> f128 { + unsafe { intrinsics::exp2f128(self) } + } + + /// Returns the natural logarithm of the number. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let one = 1.0f128; + /// // e^1 + /// let e = one.exp(); + /// + /// // ln(e) - 1 == 0 + /// let abs_difference = (e.ln() - 1.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn ln(self) -> f128 { + unsafe { intrinsics::logf128(self) } + } + + /// Returns the logarithm of the number with respect to an arbitrary base. + /// + /// The result might not be correctly rounded owing to implementation details; + /// `self.log2()` can produce more accurate results for base 2, and + /// `self.log10()` can produce more accurate results for base 10. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let five = 5.0f128; + /// + /// // log5(5) - 1 == 0 + /// let abs_difference = (five.log(5.0) - 1.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn log(self, base: f128) -> f128 { + self.ln() / base.ln() + } + + /// Returns the base 2 logarithm of the number. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let two = 2.0f128; + /// + /// // log2(2) - 1 == 0 + /// let abs_difference = (two.log2() - 1.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn log2(self) -> f128 { + unsafe { intrinsics::log2f128(self) } + } + + /// Returns the base 10 logarithm of the number. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let ten = 10.0f128; + /// + /// // log10(10) - 1 == 0 + /// let abs_difference = (ten.log10() - 1.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn log10(self) -> f128 { + unsafe { intrinsics::log10f128(self) } + } + + /// Returns the cube root of a number. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// + /// This function currently corresponds to the `cbrtf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 8.0f128; + /// + /// // x^(1/3) - 2 == 0 + /// let abs_difference = (x.cbrt() - 2.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn cbrt(self) -> f128 { + unsafe { cmath::cbrtf128(self) } + } + + /// Compute the distance between the origin and a point (`x`, `y`) on the + /// Euclidean plane. Equivalently, compute the length of the hypotenuse of a + /// right-angle triangle with other sides having length `x.abs()` and + /// `y.abs()`. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// + /// This function currently corresponds to the `hypotf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 2.0f128; + /// let y = 3.0f128; + /// + /// // sqrt(x^2 + y^2) + /// let abs_difference = (x.hypot(y) - (x.powi(2) + y.powi(2)).sqrt()).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn hypot(self, other: f128) -> f128 { + unsafe { cmath::hypotf128(self, other) } + } + + /// Computes the sine of a number (in radians). + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = std::f128::consts::FRAC_PI_2; + /// + /// let abs_difference = (x.sin() - 1.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn sin(self) -> f128 { + unsafe { intrinsics::sinf128(self) } + } + + /// Computes the cosine of a number (in radians). + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 2.0 * std::f128::consts::PI; + /// + /// let abs_difference = (x.cos() - 1.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn cos(self) -> f128 { + unsafe { intrinsics::cosf128(self) } + } + + /// Computes the tangent of a number (in radians). + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `tanf128` from libc on Unix and + /// Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = std::f128::consts::FRAC_PI_4; + /// let abs_difference = (x.tan() - 1.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn tan(self) -> f128 { + unsafe { cmath::tanf128(self) } + } + + /// Computes the arcsine of a number. Return value is in radians in + /// the range [-pi/2, pi/2] or NaN if the number is outside the range + /// [-1, 1]. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `asinf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = std::f128::consts::FRAC_PI_2; + /// + /// // asin(sin(pi/2)) + /// let abs_difference = (f.sin().asin() - std::f128::consts::FRAC_PI_2).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arcsin")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn asin(self) -> f128 { + unsafe { cmath::asinf128(self) } + } + + /// Computes the arccosine of a number. Return value is in radians in + /// the range [0, pi] or NaN if the number is outside the range + /// [-1, 1]. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `acosf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = std::f128::consts::FRAC_PI_4; + /// + /// // acos(cos(pi/4)) + /// let abs_difference = (f.cos().acos() - std::f128::consts::FRAC_PI_4).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arccos")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn acos(self) -> f128 { + unsafe { cmath::acosf128(self) } + } + + /// Computes the arctangent of a number. Return value is in radians in the + /// range [-pi/2, pi/2]; + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `atanf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let f = 1.0f128; + /// + /// // atan(tan(1)) + /// let abs_difference = (f.tan().atan() - 1.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arctan")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn atan(self) -> f128 { + unsafe { cmath::atanf128(self) } + } + + /// Computes the four quadrant arctangent of `self` (`y`) and `other` (`x`) in radians. + /// + /// * `x = 0`, `y = 0`: `0` + /// * `x >= 0`: `arctan(y/x)` -> `[-pi/2, pi/2]` + /// * `y >= 0`: `arctan(y/x) + pi` -> `(pi/2, pi]` + /// * `y < 0`: `arctan(y/x) - pi` -> `(-pi, -pi/2)` + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `atan2f128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// // Positive angles measured counter-clockwise + /// // from positive x axis + /// // -pi/4 radians (45 deg clockwise) + /// let x1 = 3.0f128; + /// let y1 = -3.0f128; + /// + /// // 3pi/4 radians (135 deg counter-clockwise) + /// let x2 = -3.0f128; + /// let y2 = 3.0f128; + /// + /// let abs_difference_1 = (y1.atan2(x1) - (-std::f128::consts::FRAC_PI_4)).abs(); + /// let abs_difference_2 = (y2.atan2(x2) - (3.0 * std::f128::consts::FRAC_PI_4)).abs(); + /// + /// assert!(abs_difference_1 <= f128::EPSILON); + /// assert!(abs_difference_2 <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn atan2(self, other: f128) -> f128 { + unsafe { cmath::atan2f128(self, other) } + } + + /// Simultaneously computes the sine and cosine of the number, `x`. Returns + /// `(sin(x), cos(x))`. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `(f128::sin(x), + /// f128::cos(x))`. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = std::f128::consts::FRAC_PI_4; + /// let f = x.sin_cos(); + /// + /// let abs_difference_0 = (f.0 - x.sin()).abs(); + /// let abs_difference_1 = (f.1 - x.cos()).abs(); + /// + /// assert!(abs_difference_0 <= f128::EPSILON); + /// assert!(abs_difference_1 <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "sincos")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + pub fn sin_cos(self) -> (f128, f128) { + (self.sin(), self.cos()) + } + + /// Returns `e^(self) - 1` in a way that is accurate even if the + /// number is close to zero. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `expm1f128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 1e-8_f128; + /// + /// // for very small x, e^x is approximately 1 + x + x^2 / 2 + /// let approx = x + x * x / 2.0; + /// let abs_difference = (x.exp_m1() - approx).abs(); + /// + /// assert!(abs_difference < 1e-10); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn exp_m1(self) -> f128 { + unsafe { cmath::expm1f128(self) } + } + + /// Returns `ln(1+n)` (natural logarithm) more accurately than if + /// the operations were performed separately. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `log1pf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 1e-8_f128; + /// + /// // for very small x, ln(1 + x) is approximately x - x^2 / 2 + /// let approx = x - x * x / 2.0; + /// let abs_difference = (x.ln_1p() - approx).abs(); + /// + /// assert!(abs_difference < 1e-10); + /// # } + /// ``` + #[inline] + #[doc(alias = "log1p")] + #[must_use = "method returns a new number and does not mutate the original value"] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + pub fn ln_1p(self) -> f128 { + unsafe { cmath::log1pf128(self) } + } + + /// Hyperbolic sine function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `sinhf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let e = std::f128::consts::E; + /// let x = 1.0f128; + /// + /// let f = x.sinh(); + /// // Solving sinh() at 1 gives `(e^2-1)/(2e)` + /// let g = ((e * e) - 1.0) / (2.0 * e); + /// let abs_difference = (f - g).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn sinh(self) -> f128 { + unsafe { cmath::sinhf128(self) } + } + + /// Hyperbolic cosine function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `coshf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let e = std::f128::consts::E; + /// let x = 1.0f128; + /// let f = x.cosh(); + /// // Solving cosh() at 1 gives this result + /// let g = ((e * e) + 1.0) / (2.0 * e); + /// let abs_difference = (f - g).abs(); + /// + /// // Same result + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn cosh(self) -> f128 { + unsafe { cmath::coshf128(self) } + } + + /// Hyperbolic tangent function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `tanhf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let e = std::f128::consts::E; + /// let x = 1.0f128; + /// + /// let f = x.tanh(); + /// // Solving tanh() at 1 gives `(1 - e^(-2))/(1 + e^(-2))` + /// let g = (1.0 - e.powi(-2)) / (1.0 + e.powi(-2)); + /// let abs_difference = (f - g).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn tanh(self) -> f128 { + unsafe { cmath::tanhf128(self) } + } + + /// Inverse hyperbolic sine function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 1.0f128; + /// let f = x.sinh().asinh(); + /// + /// let abs_difference = (f - x).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arcsinh")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn asinh(self) -> f128 { + let ax = self.abs(); + let ix = 1.0 / ax; + (ax + (ax / (Self::hypot(1.0, ix) + ix))).ln_1p().copysign(self) + } + + /// Inverse hyperbolic cosine function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 1.0f128; + /// let f = x.cosh().acosh(); + /// + /// let abs_difference = (f - x).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arccosh")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn acosh(self) -> f128 { + if self < 1.0 { + Self::NAN + } else { + (self + ((self - 1.0).sqrt() * (self + 1.0).sqrt())).ln() + } + } + + /// Inverse hyperbolic tangent function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let e = std::f128::consts::E; + /// let f = e.tanh().atanh(); + /// + /// let abs_difference = (f - e).abs(); + /// + /// assert!(abs_difference <= 1e-5); + /// # } + /// ``` + #[inline] + #[doc(alias = "arctanh")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn atanh(self) -> f128 { + 0.5 * ((2.0 * self) / (1.0 - self)).ln_1p() + } + + /// Gamma function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `tgammaf128` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// #![feature(float_gamma)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 5.0f128; + /// + /// let abs_difference = (x.gamma() - 24.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn gamma(self) -> f128 { + unsafe { cmath::tgammaf128(self) } + } + + /// Natural logarithm of the absolute value of the gamma function + /// + /// The integer part of the tuple indicates the sign of the gamma function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `lgammaf128_r` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// #![feature(float_gamma)] + /// # #[cfg(reliable_f128_math)] { + /// + /// let x = 2.0f128; + /// + /// let abs_difference = (x.ln_gamma().0 - 0.0).abs(); + /// + /// assert!(abs_difference <= f128::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn ln_gamma(self) -> (f128, i32) { + let mut signgamp: i32 = 0; + let x = unsafe { cmath::lgammaf128_r(self, &mut signgamp) }; + (x, signgamp) + } } diff --git a/library/std/src/f128/tests.rs b/library/std/src/f128/tests.rs index 162c8dbad81a1..df806a639f620 100644 --- a/library/std/src/f128/tests.rs +++ b/library/std/src/f128/tests.rs @@ -4,6 +4,21 @@ use crate::f128::consts; use crate::num::{FpCategory as Fp, *}; +// Note these tolerances make sense around zero, but not for more extreme exponents. + +/// For operations that are near exact, usually not involving math of different +/// signs. +const TOL_PRECISE: f128 = 1e-28; + +/// Default tolerances. Works for values that should be near precise but not exact. Roughly +/// the precision carried by `100 * 100`. +const TOL: f128 = 1e-12; + +/// Tolerances for math that is allowed to be imprecise, usually due to multiple chained +/// operations. +#[cfg(reliable_f128_math)] +const TOL_IMPR: f128 = 1e-10; + /// Smallest number const TINY_BITS: u128 = 0x1; @@ -191,9 +206,100 @@ fn test_classify() { assert_eq!(1e-4932f128.classify(), Fp::Subnormal); } -// FIXME(f16_f128): add missing math functions when available +#[test] +#[cfg(reliable_f128_math)] +fn test_floor() { + assert_approx_eq!(1.0f128.floor(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.3f128.floor(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.5f128.floor(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.7f128.floor(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(0.0f128.floor(), 0.0f128, TOL_PRECISE); + assert_approx_eq!((-0.0f128).floor(), -0.0f128, TOL_PRECISE); + assert_approx_eq!((-1.0f128).floor(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.3f128).floor(), -2.0f128, TOL_PRECISE); + assert_approx_eq!((-1.5f128).floor(), -2.0f128, TOL_PRECISE); + assert_approx_eq!((-1.7f128).floor(), -2.0f128, TOL_PRECISE); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_ceil() { + assert_approx_eq!(1.0f128.ceil(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.3f128.ceil(), 2.0f128, TOL_PRECISE); + assert_approx_eq!(1.5f128.ceil(), 2.0f128, TOL_PRECISE); + assert_approx_eq!(1.7f128.ceil(), 2.0f128, TOL_PRECISE); + assert_approx_eq!(0.0f128.ceil(), 0.0f128, TOL_PRECISE); + assert_approx_eq!((-0.0f128).ceil(), -0.0f128, TOL_PRECISE); + assert_approx_eq!((-1.0f128).ceil(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.3f128).ceil(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.5f128).ceil(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.7f128).ceil(), -1.0f128, TOL_PRECISE); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_round() { + assert_approx_eq!(2.5f128.round(), 3.0f128, TOL_PRECISE); + assert_approx_eq!(1.0f128.round(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.3f128.round(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.5f128.round(), 2.0f128, TOL_PRECISE); + assert_approx_eq!(1.7f128.round(), 2.0f128, TOL_PRECISE); + assert_approx_eq!(0.0f128.round(), 0.0f128, TOL_PRECISE); + assert_approx_eq!((-0.0f128).round(), -0.0f128, TOL_PRECISE); + assert_approx_eq!((-1.0f128).round(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.3f128).round(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.5f128).round(), -2.0f128, TOL_PRECISE); + assert_approx_eq!((-1.7f128).round(), -2.0f128, TOL_PRECISE); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_round_ties_even() { + assert_approx_eq!(2.5f128.round_ties_even(), 2.0f128, TOL_PRECISE); + assert_approx_eq!(1.0f128.round_ties_even(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.3f128.round_ties_even(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.5f128.round_ties_even(), 2.0f128, TOL_PRECISE); + assert_approx_eq!(1.7f128.round_ties_even(), 2.0f128, TOL_PRECISE); + assert_approx_eq!(0.0f128.round_ties_even(), 0.0f128, TOL_PRECISE); + assert_approx_eq!((-0.0f128).round_ties_even(), -0.0f128, TOL_PRECISE); + assert_approx_eq!((-1.0f128).round_ties_even(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.3f128).round_ties_even(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.5f128).round_ties_even(), -2.0f128, TOL_PRECISE); + assert_approx_eq!((-1.7f128).round_ties_even(), -2.0f128, TOL_PRECISE); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_trunc() { + assert_approx_eq!(1.0f128.trunc(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.3f128.trunc(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.5f128.trunc(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(1.7f128.trunc(), 1.0f128, TOL_PRECISE); + assert_approx_eq!(0.0f128.trunc(), 0.0f128, TOL_PRECISE); + assert_approx_eq!((-0.0f128).trunc(), -0.0f128, TOL_PRECISE); + assert_approx_eq!((-1.0f128).trunc(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.3f128).trunc(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.5f128).trunc(), -1.0f128, TOL_PRECISE); + assert_approx_eq!((-1.7f128).trunc(), -1.0f128, TOL_PRECISE); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_fract() { + assert_approx_eq!(1.0f128.fract(), 0.0f128, TOL_PRECISE); + assert_approx_eq!(1.3f128.fract(), 0.3f128, TOL_PRECISE); + assert_approx_eq!(1.5f128.fract(), 0.5f128, TOL_PRECISE); + assert_approx_eq!(1.7f128.fract(), 0.7f128, TOL_PRECISE); + assert_approx_eq!(0.0f128.fract(), 0.0f128, TOL_PRECISE); + assert_approx_eq!((-0.0f128).fract(), -0.0f128, TOL_PRECISE); + assert_approx_eq!((-1.0f128).fract(), -0.0f128, TOL_PRECISE); + assert_approx_eq!((-1.3f128).fract(), -0.3f128, TOL_PRECISE); + assert_approx_eq!((-1.5f128).fract(), -0.5f128, TOL_PRECISE); + assert_approx_eq!((-1.7f128).fract(), -0.7f128, TOL_PRECISE); +} #[test] +#[cfg(reliable_f128_math)] fn test_abs() { assert_eq!(f128::INFINITY.abs(), f128::INFINITY); assert_eq!(1f128.abs(), 1f128); @@ -293,6 +399,24 @@ fn test_next_down() { } #[test] +#[cfg(reliable_f128_math)] +fn test_mul_add() { + let nan: f128 = f128::NAN; + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + assert_approx_eq!(12.3f128.mul_add(4.5, 6.7), 62.05, TOL_PRECISE); + assert_approx_eq!((-12.3f128).mul_add(-4.5, -6.7), 48.65, TOL_PRECISE); + assert_approx_eq!(0.0f128.mul_add(8.9, 1.2), 1.2, TOL_PRECISE); + assert_approx_eq!(3.4f128.mul_add(-0.0, 5.6), 5.6, TOL_PRECISE); + assert!(nan.mul_add(7.8, 9.0).is_nan()); + assert_eq!(inf.mul_add(7.8, 9.0), inf); + assert_eq!(neg_inf.mul_add(7.8, 9.0), neg_inf); + assert_eq!(8.9f128.mul_add(inf, 3.2), inf); + assert_eq!((-3.2f128).mul_add(2.4, neg_inf), neg_inf); +} + +#[test] +#[cfg(reliable_f16_math)] fn test_recip() { let nan: f128 = f128::NAN; let inf: f128 = f128::INFINITY; @@ -301,11 +425,161 @@ fn test_recip() { assert_eq!(2.0f128.recip(), 0.5); assert_eq!((-0.4f128).recip(), -2.5); assert_eq!(0.0f128.recip(), inf); + assert_approx_eq!( + f128::MAX.recip(), + 8.40525785778023376565669454330438228902076605e-4933, + 1e-4900 + ); assert!(nan.recip().is_nan()); assert_eq!(inf.recip(), 0.0); assert_eq!(neg_inf.recip(), 0.0); } +// Many math functions allow for less accurate results, so the next tolerance up is used + +#[test] +#[cfg(reliable_f128_math)] +fn test_powi() { + let nan: f128 = f128::NAN; + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + assert_eq!(1.0f128.powi(1), 1.0); + assert_approx_eq!((-3.1f128).powi(2), 9.6100000000000005506706202140776519387, TOL); + assert_approx_eq!(5.9f128.powi(-2), 0.028727377190462507313100483690639638451, TOL); + assert_eq!(8.3f128.powi(0), 1.0); + assert!(nan.powi(2).is_nan()); + assert_eq!(inf.powi(3), inf); + assert_eq!(neg_inf.powi(2), inf); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_powf() { + let nan: f128 = f128::NAN; + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + assert_eq!(1.0f128.powf(1.0), 1.0); + assert_approx_eq!(3.4f128.powf(4.5), 246.40818323761892815995637964326426756, TOL_IMPR); + assert_approx_eq!(2.7f128.powf(-3.2), 0.041652009108526178281070304373500889273, TOL_IMPR); + assert_approx_eq!((-3.1f128).powf(2.0), 9.6100000000000005506706202140776519387, TOL_IMPR); + assert_approx_eq!(5.9f128.powf(-2.0), 0.028727377190462507313100483690639638451, TOL_IMPR); + assert_eq!(8.3f128.powf(0.0), 1.0); + assert!(nan.powf(2.0).is_nan()); + assert_eq!(inf.powf(2.0), inf); + assert_eq!(neg_inf.powf(3.0), neg_inf); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_sqrt_domain() { + assert!(f128::NAN.sqrt().is_nan()); + assert!(f128::NEG_INFINITY.sqrt().is_nan()); + assert!((-1.0f128).sqrt().is_nan()); + assert_eq!((-0.0f128).sqrt(), -0.0); + assert_eq!(0.0f128.sqrt(), 0.0); + assert_eq!(1.0f128.sqrt(), 1.0); + assert_eq!(f128::INFINITY.sqrt(), f128::INFINITY); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_exp() { + assert_eq!(1.0, 0.0f128.exp()); + assert_approx_eq!(consts::E, 1.0f128.exp(), TOL); + assert_approx_eq!(148.41315910257660342111558004055227962348775, 5.0f128.exp(), TOL); + + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + let nan: f128 = f128::NAN; + assert_eq!(inf, inf.exp()); + assert_eq!(0.0, neg_inf.exp()); + assert!(nan.exp().is_nan()); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_exp2() { + assert_eq!(32.0, 5.0f128.exp2()); + assert_eq!(1.0, 0.0f128.exp2()); + + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + let nan: f128 = f128::NAN; + assert_eq!(inf, inf.exp2()); + assert_eq!(0.0, neg_inf.exp2()); + assert!(nan.exp2().is_nan()); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_ln() { + let nan: f128 = f128::NAN; + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + assert_approx_eq!(1.0f128.exp().ln(), 1.0, TOL); + assert!(nan.ln().is_nan()); + assert_eq!(inf.ln(), inf); + assert!(neg_inf.ln().is_nan()); + assert!((-2.3f128).ln().is_nan()); + assert_eq!((-0.0f128).ln(), neg_inf); + assert_eq!(0.0f128.ln(), neg_inf); + assert_approx_eq!(4.0f128.ln(), 1.3862943611198906188344642429163531366, TOL); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_log() { + let nan: f128 = f128::NAN; + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + assert_eq!(10.0f128.log(10.0), 1.0); + assert_approx_eq!(2.3f128.log(3.5), 0.66485771361478710036766645911922010272, TOL); + assert_eq!(1.0f128.exp().log(1.0f128.exp()), 1.0); + assert!(1.0f128.log(1.0).is_nan()); + assert!(1.0f128.log(-13.9).is_nan()); + assert!(nan.log(2.3).is_nan()); + assert_eq!(inf.log(10.0), inf); + assert!(neg_inf.log(8.8).is_nan()); + assert!((-2.3f128).log(0.1).is_nan()); + assert_eq!((-0.0f128).log(2.0), neg_inf); + assert_eq!(0.0f128.log(7.0), neg_inf); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_log2() { + let nan: f128 = f128::NAN; + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + assert_approx_eq!(10.0f128.log2(), 3.32192809488736234787031942948939017, TOL); + assert_approx_eq!(2.3f128.log2(), 1.2016338611696504130002982471978765921, TOL); + assert_approx_eq!(1.0f128.exp().log2(), 1.4426950408889634073599246810018921381, TOL); + assert!(nan.log2().is_nan()); + assert_eq!(inf.log2(), inf); + assert!(neg_inf.log2().is_nan()); + assert!((-2.3f128).log2().is_nan()); + assert_eq!((-0.0f128).log2(), neg_inf); + assert_eq!(0.0f128.log2(), neg_inf); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_log10() { + let nan: f128 = f128::NAN; + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + assert_eq!(10.0f128.log10(), 1.0); + assert_approx_eq!(2.3f128.log10(), 0.36172783601759284532595218865859309898, TOL); + assert_approx_eq!(1.0f128.exp().log10(), 0.43429448190325182765112891891660508222, TOL); + assert_eq!(1.0f128.log10(), 0.0); + assert!(nan.log10().is_nan()); + assert_eq!(inf.log10(), inf); + assert!(neg_inf.log10().is_nan()); + assert!((-2.3f128).log10().is_nan()); + assert_eq!((-0.0f128).log10(), neg_inf); + assert_eq!(0.0f128.log10(), neg_inf); +} + #[test] fn test_to_degrees() { let pi: f128 = consts::PI; @@ -313,8 +587,8 @@ fn test_to_degrees() { let inf: f128 = f128::INFINITY; let neg_inf: f128 = f128::NEG_INFINITY; assert_eq!(0.0f128.to_degrees(), 0.0); - assert_approx_eq!((-5.8f128).to_degrees(), -332.315521); - assert_eq!(pi.to_degrees(), 180.0); + assert_approx_eq!((-5.8f128).to_degrees(), -332.31552117587745090765431723855668471, TOL); + assert_approx_eq!(pi.to_degrees(), 180.0, TOL); assert!(nan.to_degrees().is_nan()); assert_eq!(inf.to_degrees(), inf); assert_eq!(neg_inf.to_degrees(), neg_inf); @@ -328,19 +602,122 @@ fn test_to_radians() { let inf: f128 = f128::INFINITY; let neg_inf: f128 = f128::NEG_INFINITY; assert_eq!(0.0f128.to_radians(), 0.0); - assert_approx_eq!(154.6f128.to_radians(), 2.698279); - assert_approx_eq!((-332.31f128).to_radians(), -5.799903); + assert_approx_eq!(154.6f128.to_radians(), 2.6982790235832334267135442069489767804, TOL); + assert_approx_eq!((-332.31f128).to_radians(), -5.7999036373023566567593094812182763013, TOL); // check approx rather than exact because round trip for pi doesn't fall on an exactly // representable value (unlike `f32` and `f64`). - assert_approx_eq!(180.0f128.to_radians(), pi); + assert_approx_eq!(180.0f128.to_radians(), pi, TOL_PRECISE); assert!(nan.to_radians().is_nan()); assert_eq!(inf.to_radians(), inf); assert_eq!(neg_inf.to_radians(), neg_inf); } +#[test] +#[cfg(reliable_f128_math)] +fn test_asinh() { + // Lower accuracy results are allowed, use increased tolerances + assert_eq!(0.0f128.asinh(), 0.0f128); + assert_eq!((-0.0f128).asinh(), -0.0f128); + + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + let nan: f128 = f128::NAN; + assert_eq!(inf.asinh(), inf); + assert_eq!(neg_inf.asinh(), neg_inf); + assert!(nan.asinh().is_nan()); + assert!((-0.0f128).asinh().is_sign_negative()); + + // issue 63271 + assert_approx_eq!(2.0f128.asinh(), 1.443635475178810342493276740273105f128, TOL_IMPR); + assert_approx_eq!((-2.0f128).asinh(), -1.443635475178810342493276740273105f128, TOL_IMPR); + // regression test for the catastrophic cancellation fixed in 72486 + assert_approx_eq!( + (-67452098.07139316f128).asinh(), + -18.720075426274544393985484294000831757220, + TOL_IMPR + ); + + // test for low accuracy from issue 104548 + assert_approx_eq!(60.0f128, 60.0f128.sinh().asinh(), TOL_IMPR); + // mul needed for approximate comparison to be meaningful + assert_approx_eq!(1.0f128, 1e-15f128.sinh().asinh() * 1e15f128, TOL_IMPR); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_acosh() { + assert_eq!(1.0f128.acosh(), 0.0f128); + assert!(0.999f128.acosh().is_nan()); + + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + let nan: f128 = f128::NAN; + assert_eq!(inf.acosh(), inf); + assert!(neg_inf.acosh().is_nan()); + assert!(nan.acosh().is_nan()); + assert_approx_eq!(2.0f128.acosh(), 1.31695789692481670862504634730796844f128, TOL_IMPR); + assert_approx_eq!(3.0f128.acosh(), 1.76274717403908605046521864995958461f128, TOL_IMPR); + + // test for low accuracy from issue 104548 + assert_approx_eq!(60.0f128, 60.0f128.cosh().acosh(), TOL_IMPR); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_atanh() { + assert_eq!(0.0f128.atanh(), 0.0f128); + assert_eq!((-0.0f128).atanh(), -0.0f128); + + let inf: f128 = f128::INFINITY; + let neg_inf: f128 = f128::NEG_INFINITY; + let nan: f128 = f128::NAN; + assert_eq!(1.0f128.atanh(), inf); + assert_eq!((-1.0f128).atanh(), neg_inf); + assert!(2f128.atanh().atanh().is_nan()); + assert!((-2f128).atanh().atanh().is_nan()); + assert!(inf.atanh().is_nan()); + assert!(neg_inf.atanh().is_nan()); + assert!(nan.atanh().is_nan()); + assert_approx_eq!(0.5f128.atanh(), 0.54930614433405484569762261846126285f128, TOL_IMPR); + assert_approx_eq!((-0.5f128).atanh(), -0.54930614433405484569762261846126285f128, TOL_IMPR); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_gamma() { + // precision can differ among platforms + assert_approx_eq!(1.0f128.gamma(), 1.0f128, TOL_IMPR); + assert_approx_eq!(2.0f128.gamma(), 1.0f128, TOL_IMPR); + assert_approx_eq!(3.0f128.gamma(), 2.0f128, TOL_IMPR); + assert_approx_eq!(4.0f128.gamma(), 6.0f128, TOL_IMPR); + assert_approx_eq!(5.0f128.gamma(), 24.0f128, TOL_IMPR); + assert_approx_eq!(0.5f128.gamma(), consts::PI.sqrt(), TOL_IMPR); + assert_approx_eq!((-0.5f128).gamma(), -2.0 * consts::PI.sqrt(), TOL_IMPR); + assert_eq!(0.0f128.gamma(), f128::INFINITY); + assert_eq!((-0.0f128).gamma(), f128::NEG_INFINITY); + assert!((-1.0f128).gamma().is_nan()); + assert!((-2.0f128).gamma().is_nan()); + assert!(f128::NAN.gamma().is_nan()); + assert!(f128::NEG_INFINITY.gamma().is_nan()); + assert_eq!(f128::INFINITY.gamma(), f128::INFINITY); + assert_eq!(1760.9f128.gamma(), f128::INFINITY); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_ln_gamma() { + assert_approx_eq!(1.0f128.ln_gamma().0, 0.0f128, TOL_IMPR); + assert_eq!(1.0f128.ln_gamma().1, 1); + assert_approx_eq!(2.0f128.ln_gamma().0, 0.0f128, TOL_IMPR); + assert_eq!(2.0f128.ln_gamma().1, 1); + assert_approx_eq!(3.0f128.ln_gamma().0, 2.0f128.ln(), TOL_IMPR); + assert_eq!(3.0f128.ln_gamma().1, 1); + assert_approx_eq!((-0.5f128).ln_gamma().0, (2.0 * consts::PI.sqrt()).ln(), TOL_IMPR); + assert_eq!((-0.5f128).ln_gamma().1, -1); +} + #[test] fn test_real_consts() { - // FIXME(f16_f128): add math tests when available use super::consts; let pi: f128 = consts::PI; @@ -351,29 +728,34 @@ fn test_real_consts() { let frac_pi_8: f128 = consts::FRAC_PI_8; let frac_1_pi: f128 = consts::FRAC_1_PI; let frac_2_pi: f128 = consts::FRAC_2_PI; - // let frac_2_sqrtpi: f128 = consts::FRAC_2_SQRT_PI; - // let sqrt2: f128 = consts::SQRT_2; - // let frac_1_sqrt2: f128 = consts::FRAC_1_SQRT_2; - // let e: f128 = consts::E; - // let log2_e: f128 = consts::LOG2_E; - // let log10_e: f128 = consts::LOG10_E; - // let ln_2: f128 = consts::LN_2; - // let ln_10: f128 = consts::LN_10; - - assert_approx_eq!(frac_pi_2, pi / 2f128); - assert_approx_eq!(frac_pi_3, pi / 3f128); - assert_approx_eq!(frac_pi_4, pi / 4f128); - assert_approx_eq!(frac_pi_6, pi / 6f128); - assert_approx_eq!(frac_pi_8, pi / 8f128); - assert_approx_eq!(frac_1_pi, 1f128 / pi); - assert_approx_eq!(frac_2_pi, 2f128 / pi); - // assert_approx_eq!(frac_2_sqrtpi, 2f128 / pi.sqrt()); - // assert_approx_eq!(sqrt2, 2f128.sqrt()); - // assert_approx_eq!(frac_1_sqrt2, 1f128 / 2f128.sqrt()); - // assert_approx_eq!(log2_e, e.log2()); - // assert_approx_eq!(log10_e, e.log10()); - // assert_approx_eq!(ln_2, 2f128.ln()); - // assert_approx_eq!(ln_10, 10f128.ln()); + + assert_approx_eq!(frac_pi_2, pi / 2f128, TOL_PRECISE); + assert_approx_eq!(frac_pi_3, pi / 3f128, TOL_PRECISE); + assert_approx_eq!(frac_pi_4, pi / 4f128, TOL_PRECISE); + assert_approx_eq!(frac_pi_6, pi / 6f128, TOL_PRECISE); + assert_approx_eq!(frac_pi_8, pi / 8f128, TOL_PRECISE); + assert_approx_eq!(frac_1_pi, 1f128 / pi, TOL_PRECISE); + assert_approx_eq!(frac_2_pi, 2f128 / pi, TOL_PRECISE); + + #[cfg(reliable_f128_math)] + { + let frac_2_sqrtpi: f128 = consts::FRAC_2_SQRT_PI; + let sqrt2: f128 = consts::SQRT_2; + let frac_1_sqrt2: f128 = consts::FRAC_1_SQRT_2; + let e: f128 = consts::E; + let log2_e: f128 = consts::LOG2_E; + let log10_e: f128 = consts::LOG10_E; + let ln_2: f128 = consts::LN_2; + let ln_10: f128 = consts::LN_10; + + assert_approx_eq!(frac_2_sqrtpi, 2f128 / pi.sqrt(), TOL_PRECISE); + assert_approx_eq!(sqrt2, 2f128.sqrt(), TOL_PRECISE); + assert_approx_eq!(frac_1_sqrt2, 1f128 / 2f128.sqrt(), TOL_PRECISE); + assert_approx_eq!(log2_e, e.log2(), TOL_PRECISE); + assert_approx_eq!(log10_e, e.log10(), TOL_PRECISE); + assert_approx_eq!(ln_2, 2f128.ln(), TOL_PRECISE); + assert_approx_eq!(ln_10, 10f128.ln(), TOL_PRECISE); + } } #[test] @@ -382,10 +764,10 @@ fn test_float_bits_conv() { assert_eq!((12.5f128).to_bits(), 0x40029000000000000000000000000000); assert_eq!((1337f128).to_bits(), 0x40094e40000000000000000000000000); assert_eq!((-14.25f128).to_bits(), 0xc002c800000000000000000000000000); - assert_approx_eq!(f128::from_bits(0x3fff0000000000000000000000000000), 1.0); - assert_approx_eq!(f128::from_bits(0x40029000000000000000000000000000), 12.5); - assert_approx_eq!(f128::from_bits(0x40094e40000000000000000000000000), 1337.0); - assert_approx_eq!(f128::from_bits(0xc002c800000000000000000000000000), -14.25); + assert_approx_eq!(f128::from_bits(0x3fff0000000000000000000000000000), 1.0, TOL_PRECISE); + assert_approx_eq!(f128::from_bits(0x40029000000000000000000000000000), 12.5, TOL_PRECISE); + assert_approx_eq!(f128::from_bits(0x40094e40000000000000000000000000), 1337.0, TOL_PRECISE); + assert_approx_eq!(f128::from_bits(0xc002c800000000000000000000000000), -14.25, TOL_PRECISE); // Check that NaNs roundtrip their bits regardless of signaling-ness // 0xA is 0b1010; 0x5 is 0b0101 -- so these two together clobbers all the mantissa bits diff --git a/library/std/src/f16.rs b/library/std/src/f16.rs index e3024defed734..10908332762d5 100644 --- a/library/std/src/f16.rs +++ b/library/std/src/f16.rs @@ -12,25 +12,180 @@ pub use core::f16::consts; #[cfg(not(test))] use crate::intrinsics; +#[cfg(not(test))] +use crate::sys::cmath; #[cfg(not(test))] impl f16 { - /// Raises a number to an integer power. + /// Returns the largest integer less than or equal to `self`. /// - /// Using this function is generally faster than using `powf`. - /// It might have a different sequence of rounding operations than `powf`, - /// so the results are not guaranteed to agree. + /// This function always returns the precise result. /// - /// # Unspecified precision + /// # Examples /// - /// The precision of this function is non-deterministic. This means it varies by platform, Rust version, and - /// can even differ within the same execution from one invocation to the next. + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = 3.7_f16; + /// let g = 3.0_f16; + /// let h = -3.7_f16; + /// + /// assert_eq!(f.floor(), 3.0); + /// assert_eq!(g.floor(), 3.0); + /// assert_eq!(h.floor(), -4.0); + /// # } + /// ``` #[inline] #[rustc_allow_incoherent_impl] #[unstable(feature = "f16", issue = "116909")] #[must_use = "method returns a new number and does not mutate the original value"] - pub fn powi(self, n: i32) -> f16 { - unsafe { intrinsics::powif16(self, n) } + pub fn floor(self) -> f16 { + unsafe { intrinsics::floorf16(self) } + } + + /// Returns the smallest integer greater than or equal to `self`. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = 3.01_f16; + /// let g = 4.0_f16; + /// + /// assert_eq!(f.ceil(), 4.0); + /// assert_eq!(g.ceil(), 4.0); + /// # } + /// ``` + #[inline] + #[doc(alias = "ceiling")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn ceil(self) -> f16 { + unsafe { intrinsics::ceilf16(self) } + } + + /// Returns the nearest integer to `self`. If a value is half-way between two + /// integers, round away from `0.0`. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = 3.3_f16; + /// let g = -3.3_f16; + /// let h = -3.7_f16; + /// let i = 3.5_f16; + /// let j = 4.5_f16; + /// + /// assert_eq!(f.round(), 3.0); + /// assert_eq!(g.round(), -3.0); + /// assert_eq!(h.round(), -4.0); + /// assert_eq!(i.round(), 4.0); + /// assert_eq!(j.round(), 5.0); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn round(self) -> f16 { + unsafe { intrinsics::roundf16(self) } + } + + /// Returns the nearest integer to a number. Rounds half-way cases to the number + /// with an even least significant digit. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = 3.3_f16; + /// let g = -3.3_f16; + /// let h = 3.5_f16; + /// let i = 4.5_f16; + /// + /// assert_eq!(f.round_ties_even(), 3.0); + /// assert_eq!(g.round_ties_even(), -3.0); + /// assert_eq!(h.round_ties_even(), 4.0); + /// assert_eq!(i.round_ties_even(), 4.0); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn round_ties_even(self) -> f16 { + unsafe { intrinsics::rintf16(self) } + } + + /// Returns the integer part of `self`. + /// This means that non-integer numbers are always truncated towards zero. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = 3.7_f16; + /// let g = 3.0_f16; + /// let h = -3.7_f16; + /// + /// assert_eq!(f.trunc(), 3.0); + /// assert_eq!(g.trunc(), 3.0); + /// assert_eq!(h.trunc(), -3.0); + /// # } + /// ``` + #[inline] + #[doc(alias = "truncate")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn trunc(self) -> f16 { + unsafe { intrinsics::truncf16(self) } + } + + /// Returns the fractional part of `self`. + /// + /// This function always returns the precise result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 3.6_f16; + /// let y = -3.6_f16; + /// let abs_difference_x = (x.fract() - 0.6).abs(); + /// let abs_difference_y = (y.fract() - (-0.6)).abs(); + /// + /// assert!(abs_difference_x <= f16::EPSILON); + /// assert!(abs_difference_y <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn fract(self) -> f16 { + self - self.trunc() } /// Computes the absolute value of `self`. @@ -60,4 +215,1127 @@ impl f16 { // FIXME(f16_f128): replace with `intrinsics::fabsf16` when available Self::from_bits(self.to_bits() & !(1 << 15)) } + + /// Returns a number that represents the sign of `self`. + /// + /// - `1.0` if the number is positive, `+0.0` or `INFINITY` + /// - `-1.0` if the number is negative, `-0.0` or `NEG_INFINITY` + /// - NaN if the number is NaN + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = 3.5_f16; + /// + /// assert_eq!(f.signum(), 1.0); + /// assert_eq!(f16::NEG_INFINITY.signum(), -1.0); + /// + /// assert!(f16::NAN.signum().is_nan()); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn signum(self) -> f16 { + if self.is_nan() { Self::NAN } else { 1.0_f16.copysign(self) } + } + + /// Returns a number composed of the magnitude of `self` and the sign of + /// `sign`. + /// + /// Equal to `self` if the sign of `self` and `sign` are the same, otherwise + /// equal to `-self`. If `self` is a NaN, then a NaN with the sign bit of + /// `sign` is returned. Note, however, that conserving the sign bit on NaN + /// across arithmetical operations is not generally guaranteed. + /// See [explanation of NaN as a special value](primitive@f16) for more info. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = 3.5_f16; + /// + /// assert_eq!(f.copysign(0.42), 3.5_f16); + /// assert_eq!(f.copysign(-0.42), -3.5_f16); + /// assert_eq!((-f).copysign(0.42), 3.5_f16); + /// assert_eq!((-f).copysign(-0.42), -3.5_f16); + /// + /// assert!(f16::NAN.copysign(1.0).is_nan()); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn copysign(self, sign: f16) -> f16 { + unsafe { intrinsics::copysignf16(self, sign) } + } + + /// Fused multiply-add. Computes `(self * a) + b` with only one rounding + /// error, yielding a more accurate result than an unfused multiply-add. + /// + /// Using `mul_add` *may* be more performant than an unfused multiply-add if + /// the target architecture has a dedicated `fma` CPU instruction. However, + /// this is not always true, and will be heavily dependant on designing + /// algorithms with specific target hardware in mind. + /// + /// # Precision + /// + /// The result of this operation is guaranteed to be the rounded + /// infinite-precision result. It is specified by IEEE 754 as + /// `fusedMultiplyAdd` and guaranteed not to change. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let m = 10.0_f16; + /// let x = 4.0_f16; + /// let b = 60.0_f16; + /// + /// assert_eq!(m.mul_add(x, b), 100.0); + /// assert_eq!(m * x + b, 100.0); + /// + /// let one_plus_eps = 1.0_f16 + f16::EPSILON; + /// let one_minus_eps = 1.0_f16 - f16::EPSILON; + /// let minus_one = -1.0_f16; + /// + /// // The exact result (1 + eps) * (1 - eps) = 1 - eps * eps. + /// assert_eq!(one_plus_eps.mul_add(one_minus_eps, minus_one), -f16::EPSILON * f16::EPSILON); + /// // Different rounding with the non-fused multiply and add. + /// assert_eq!(one_plus_eps * one_minus_eps + minus_one, 0.0); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn mul_add(self, a: f16, b: f16) -> f16 { + unsafe { intrinsics::fmaf16(self, a, b) } + } + + /// Calculates Euclidean division, the matching method for `rem_euclid`. + /// + /// This computes the integer `n` such that + /// `self = n * rhs + self.rem_euclid(rhs)`. + /// In other words, the result is `self / rhs` rounded to the integer `n` + /// such that `self >= n * rhs`. + /// + /// # Precision + /// + /// The result of this operation is guaranteed to be the rounded + /// infinite-precision result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let a: f16 = 7.0; + /// let b = 4.0; + /// assert_eq!(a.div_euclid(b), 1.0); // 7.0 > 4.0 * 1.0 + /// assert_eq!((-a).div_euclid(b), -2.0); // -7.0 >= 4.0 * -2.0 + /// assert_eq!(a.div_euclid(-b), -1.0); // 7.0 >= -4.0 * -1.0 + /// assert_eq!((-a).div_euclid(-b), 2.0); // -7.0 >= -4.0 * 2.0 + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn div_euclid(self, rhs: f16) -> f16 { + let q = (self / rhs).trunc(); + if self % rhs < 0.0 { + return if rhs > 0.0 { q - 1.0 } else { q + 1.0 }; + } + q + } + + /// Calculates the least nonnegative remainder of `self (mod rhs)`. + /// + /// In particular, the return value `r` satisfies `0.0 <= r < rhs.abs()` in + /// most cases. However, due to a floating point round-off error it can + /// result in `r == rhs.abs()`, violating the mathematical definition, if + /// `self` is much smaller than `rhs.abs()` in magnitude and `self < 0.0`. + /// This result is not an element of the function's codomain, but it is the + /// closest floating point number in the real numbers and thus fulfills the + /// property `self == self.div_euclid(rhs) * rhs + self.rem_euclid(rhs)` + /// approximately. + /// + /// # Precision + /// + /// The result of this operation is guaranteed to be the rounded + /// infinite-precision result. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let a: f16 = 7.0; + /// let b = 4.0; + /// assert_eq!(a.rem_euclid(b), 3.0); + /// assert_eq!((-a).rem_euclid(b), 1.0); + /// assert_eq!(a.rem_euclid(-b), 3.0); + /// assert_eq!((-a).rem_euclid(-b), 1.0); + /// // limitation due to round-off error + /// assert!((-f16::EPSILON).rem_euclid(3.0) != 0.0); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[doc(alias = "modulo", alias = "mod")] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn rem_euclid(self, rhs: f16) -> f16 { + let r = self % rhs; + if r < 0.0 { r + rhs.abs() } else { r } + } + + /// Raises a number to an integer power. + /// + /// Using this function is generally faster than using `powf`. + /// It might have a different sequence of rounding operations than `powf`, + /// so the results are not guaranteed to agree. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn powi(self, n: i32) -> f16 { + unsafe { intrinsics::powif16(self, n) } + } + + /// Raises a number to a floating point power. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 2.0_f16; + /// let abs_difference = (x.powf(2.0) - (x * x)).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn powf(self, n: f16) -> f16 { + unsafe { intrinsics::powf16(self, n) } + } + + /// Returns the square root of a number. + /// + /// Returns NaN if `self` is a negative number other than `-0.0`. + /// + /// # Precision + /// + /// The result of this operation is guaranteed to be the rounded + /// infinite-precision result. It is specified by IEEE 754 as `squareRoot` + /// and guaranteed not to change. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let positive = 4.0_f16; + /// let negative = -4.0_f16; + /// let negative_zero = -0.0_f16; + /// + /// assert_eq!(positive.sqrt(), 2.0); + /// assert!(negative.sqrt().is_nan()); + /// assert!(negative_zero.sqrt() == negative_zero); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn sqrt(self) -> f16 { + unsafe { intrinsics::sqrtf16(self) } + } + + /// Returns `e^(self)`, (the exponential function). + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let one = 1.0f16; + /// // e^1 + /// let e = one.exp(); + /// + /// // ln(e) - 1 == 0 + /// let abs_difference = (e.ln() - 1.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn exp(self) -> f16 { + unsafe { intrinsics::expf16(self) } + } + + /// Returns `2^(self)`. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = 2.0f16; + /// + /// // 2^2 - 4 == 0 + /// let abs_difference = (f.exp2() - 4.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn exp2(self) -> f16 { + unsafe { intrinsics::exp2f16(self) } + } + + /// Returns the natural logarithm of the number. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let one = 1.0f16; + /// // e^1 + /// let e = one.exp(); + /// + /// // ln(e) - 1 == 0 + /// let abs_difference = (e.ln() - 1.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn ln(self) -> f16 { + unsafe { intrinsics::logf16(self) } + } + + /// Returns the logarithm of the number with respect to an arbitrary base. + /// + /// The result might not be correctly rounded owing to implementation details; + /// `self.log2()` can produce more accurate results for base 2, and + /// `self.log10()` can produce more accurate results for base 10. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let five = 5.0f16; + /// + /// // log5(5) - 1 == 0 + /// let abs_difference = (five.log(5.0) - 1.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn log(self, base: f16) -> f16 { + self.ln() / base.ln() + } + + /// Returns the base 2 logarithm of the number. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let two = 2.0f16; + /// + /// // log2(2) - 1 == 0 + /// let abs_difference = (two.log2() - 1.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn log2(self) -> f16 { + unsafe { intrinsics::log2f16(self) } + } + + /// Returns the base 10 logarithm of the number. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let ten = 10.0f16; + /// + /// // log10(10) - 1 == 0 + /// let abs_difference = (ten.log10() - 1.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn log10(self) -> f16 { + unsafe { intrinsics::log10f16(self) } + } + + /// Returns the cube root of a number. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `cbrtf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 8.0f16; + /// + /// // x^(1/3) - 2 == 0 + /// let abs_difference = (x.cbrt() - 2.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn cbrt(self) -> f16 { + (unsafe { cmath::cbrtf(self as f32) }) as f16 + } + + /// Compute the distance between the origin and a point (`x`, `y`) on the + /// Euclidean plane. Equivalently, compute the length of the hypotenuse of a + /// right-angle triangle with other sides having length `x.abs()` and + /// `y.abs()`. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `hypotf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 2.0f16; + /// let y = 3.0f16; + /// + /// // sqrt(x^2 + y^2) + /// let abs_difference = (x.hypot(y) - (x.powi(2) + y.powi(2)).sqrt()).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn hypot(self, other: f16) -> f16 { + (unsafe { cmath::hypotf(self as f32, other as f32) }) as f16 + } + + /// Computes the sine of a number (in radians). + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = std::f16::consts::FRAC_PI_2; + /// + /// let abs_difference = (x.sin() - 1.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn sin(self) -> f16 { + unsafe { intrinsics::sinf16(self) } + } + + /// Computes the cosine of a number (in radians). + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 2.0 * std::f16::consts::PI; + /// + /// let abs_difference = (x.cos() - 1.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn cos(self) -> f16 { + unsafe { intrinsics::cosf16(self) } + } + + /// Computes the tangent of a number (in radians). + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `tanf` from libc on Unix and + /// Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = std::f16::consts::FRAC_PI_4; + /// let abs_difference = (x.tan() - 1.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn tan(self) -> f16 { + (unsafe { cmath::tanf(self as f32) }) as f16 + } + + /// Computes the arcsine of a number. Return value is in radians in + /// the range [-pi/2, pi/2] or NaN if the number is outside the range + /// [-1, 1]. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `asinf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = std::f16::consts::FRAC_PI_2; + /// + /// // asin(sin(pi/2)) + /// let abs_difference = (f.sin().asin() - std::f16::consts::FRAC_PI_2).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arcsin")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn asin(self) -> f16 { + (unsafe { cmath::asinf(self as f32) }) as f16 + } + + /// Computes the arccosine of a number. Return value is in radians in + /// the range [0, pi] or NaN if the number is outside the range + /// [-1, 1]. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `acosf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = std::f16::consts::FRAC_PI_4; + /// + /// // acos(cos(pi/4)) + /// let abs_difference = (f.cos().acos() - std::f16::consts::FRAC_PI_4).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arccos")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn acos(self) -> f16 { + (unsafe { cmath::acosf(self as f32) }) as f16 + } + + /// Computes the arctangent of a number. Return value is in radians in the + /// range [-pi/2, pi/2]; + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `atanf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let f = 1.0f16; + /// + /// // atan(tan(1)) + /// let abs_difference = (f.tan().atan() - 1.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arctan")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn atan(self) -> f16 { + (unsafe { cmath::atanf(self as f32) }) as f16 + } + + /// Computes the four quadrant arctangent of `self` (`y`) and `other` (`x`) in radians. + /// + /// * `x = 0`, `y = 0`: `0` + /// * `x >= 0`: `arctan(y/x)` -> `[-pi/2, pi/2]` + /// * `y >= 0`: `arctan(y/x) + pi` -> `(pi/2, pi]` + /// * `y < 0`: `arctan(y/x) - pi` -> `(-pi, -pi/2)` + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `atan2f` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// // Positive angles measured counter-clockwise + /// // from positive x axis + /// // -pi/4 radians (45 deg clockwise) + /// let x1 = 3.0f16; + /// let y1 = -3.0f16; + /// + /// // 3pi/4 radians (135 deg counter-clockwise) + /// let x2 = -3.0f16; + /// let y2 = 3.0f16; + /// + /// let abs_difference_1 = (y1.atan2(x1) - (-std::f16::consts::FRAC_PI_4)).abs(); + /// let abs_difference_2 = (y2.atan2(x2) - (3.0 * std::f16::consts::FRAC_PI_4)).abs(); + /// + /// assert!(abs_difference_1 <= f16::EPSILON); + /// assert!(abs_difference_2 <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn atan2(self, other: f16) -> f16 { + (unsafe { cmath::atan2f(self as f32, other as f32) }) as f16 + } + + /// Simultaneously computes the sine and cosine of the number, `x`. Returns + /// `(sin(x), cos(x))`. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `(f16::sin(x), + /// f16::cos(x))`. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = std::f16::consts::FRAC_PI_4; + /// let f = x.sin_cos(); + /// + /// let abs_difference_0 = (f.0 - x.sin()).abs(); + /// let abs_difference_1 = (f.1 - x.cos()).abs(); + /// + /// assert!(abs_difference_0 <= f16::EPSILON); + /// assert!(abs_difference_1 <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "sincos")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + pub fn sin_cos(self) -> (f16, f16) { + (self.sin(), self.cos()) + } + + /// Returns `e^(self) - 1` in a way that is accurate even if the + /// number is close to zero. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `expm1f` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 1e-4_f16; + /// + /// // for very small x, e^x is approximately 1 + x + x^2 / 2 + /// let approx = x + x * x / 2.0; + /// let abs_difference = (x.exp_m1() - approx).abs(); + /// + /// assert!(abs_difference < 1e-4); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn exp_m1(self) -> f16 { + (unsafe { cmath::expm1f(self as f32) }) as f16 + } + + /// Returns `ln(1+n)` (natural logarithm) more accurately than if + /// the operations were performed separately. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `log1pf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 1e-4_f16; + /// + /// // for very small x, ln(1 + x) is approximately x - x^2 / 2 + /// let approx = x - x * x / 2.0; + /// let abs_difference = (x.ln_1p() - approx).abs(); + /// + /// assert!(abs_difference < 1e-4); + /// # } + /// ``` + #[inline] + #[doc(alias = "log1p")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn ln_1p(self) -> f16 { + (unsafe { cmath::log1pf(self as f32) }) as f16 + } + + /// Hyperbolic sine function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `sinhf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let e = std::f16::consts::E; + /// let x = 1.0f16; + /// + /// let f = x.sinh(); + /// // Solving sinh() at 1 gives `(e^2-1)/(2e)` + /// let g = ((e * e) - 1.0) / (2.0 * e); + /// let abs_difference = (f - g).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn sinh(self) -> f16 { + (unsafe { cmath::sinhf(self as f32) }) as f16 + } + + /// Hyperbolic cosine function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `coshf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let e = std::f16::consts::E; + /// let x = 1.0f16; + /// let f = x.cosh(); + /// // Solving cosh() at 1 gives this result + /// let g = ((e * e) + 1.0) / (2.0 * e); + /// let abs_difference = (f - g).abs(); + /// + /// // Same result + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn cosh(self) -> f16 { + (unsafe { cmath::coshf(self as f32) }) as f16 + } + + /// Hyperbolic tangent function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `tanhf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let e = std::f16::consts::E; + /// let x = 1.0f16; + /// + /// let f = x.tanh(); + /// // Solving tanh() at 1 gives `(1 - e^(-2))/(1 + e^(-2))` + /// let g = (1.0 - e.powi(-2)) / (1.0 + e.powi(-2)); + /// let abs_difference = (f - g).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn tanh(self) -> f16 { + (unsafe { cmath::tanhf(self as f32) }) as f16 + } + + /// Inverse hyperbolic sine function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 1.0f16; + /// let f = x.sinh().asinh(); + /// + /// let abs_difference = (f - x).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arcsinh")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn asinh(self) -> f16 { + let ax = self.abs(); + let ix = 1.0 / ax; + (ax + (ax / (Self::hypot(1.0, ix) + ix))).ln_1p().copysign(self) + } + + /// Inverse hyperbolic cosine function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 1.0f16; + /// let f = x.cosh().acosh(); + /// + /// let abs_difference = (f - x).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[doc(alias = "arccosh")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn acosh(self) -> f16 { + if self < 1.0 { + Self::NAN + } else { + (self + ((self - 1.0).sqrt() * (self + 1.0).sqrt())).ln() + } + } + + /// Inverse hyperbolic tangent function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let e = std::f16::consts::E; + /// let f = e.tanh().atanh(); + /// + /// let abs_difference = (f - e).abs(); + /// + /// assert!(abs_difference <= 0.01); + /// # } + /// ``` + #[inline] + #[doc(alias = "arctanh")] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn atanh(self) -> f16 { + 0.5 * ((2.0 * self) / (1.0 - self)).ln_1p() + } + + /// Gamma function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `tgammaf` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// #![feature(float_gamma)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 5.0f16; + /// + /// let abs_difference = (x.gamma() - 24.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn gamma(self) -> f16 { + (unsafe { cmath::tgammaf(self as f32) }) as f16 + } + + /// Natural logarithm of the absolute value of the gamma function + /// + /// The integer part of the tuple indicates the sign of the gamma function. + /// + /// # Unspecified precision + /// + /// The precision of this function is non-deterministic. This means it varies by platform, + /// Rust version, and can even differ within the same execution from one invocation to the next. + /// + /// This function currently corresponds to the `lgamma_r` from libc on Unix + /// and Windows. Note that this might change in the future. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// #![feature(float_gamma)] + /// # #[cfg(reliable_f16_math)] { + /// + /// let x = 2.0f16; + /// + /// let abs_difference = (x.ln_gamma().0 - 0.0).abs(); + /// + /// assert!(abs_difference <= f16::EPSILON); + /// # } + /// ``` + #[inline] + #[rustc_allow_incoherent_impl] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "method returns a new number and does not mutate the original value"] + pub fn ln_gamma(self) -> (f16, i32) { + let mut signgamp: i32 = 0; + let x = (unsafe { cmath::lgammaf_r(self as f32, &mut signgamp) }) as f16; + (x, signgamp) + } } diff --git a/library/std/src/f16/tests.rs b/library/std/src/f16/tests.rs index f73bdf68e8295..f0ef807dac141 100644 --- a/library/std/src/f16/tests.rs +++ b/library/std/src/f16/tests.rs @@ -4,11 +4,21 @@ use crate::f16::consts; use crate::num::{FpCategory as Fp, *}; -// We run out of precision pretty quickly with f16 -// const F16_APPROX_L1: f16 = 0.001; -const F16_APPROX_L2: f16 = 0.01; -// const F16_APPROX_L3: f16 = 0.1; -const F16_APPROX_L4: f16 = 0.5; +/// Tolerance for results on the order of 10.0e-2; +#[cfg(reliable_f16_math)] +const TOL_N2: f16 = 0.0001; + +/// Tolerance for results on the order of 10.0e+0 +#[cfg(reliable_f16_math)] +const TOL_0: f16 = 0.01; + +/// Tolerance for results on the order of 10.0e+2 +#[cfg(reliable_f16_math)] +const TOL_P2: f16 = 0.5; + +/// Tolerance for results on the order of 10.0e+4 +#[cfg(reliable_f16_math)] +const TOL_P4: f16 = 10.0; /// Smallest number const TINY_BITS: u16 = 0x1; @@ -197,9 +207,100 @@ fn test_classify() { assert_eq!(1e-5f16.classify(), Fp::Subnormal); } -// FIXME(f16_f128): add missing math functions when available +#[test] +#[cfg(reliable_f16_math)] +fn test_floor() { + assert_approx_eq!(1.0f16.floor(), 1.0f16, TOL_0); + assert_approx_eq!(1.3f16.floor(), 1.0f16, TOL_0); + assert_approx_eq!(1.5f16.floor(), 1.0f16, TOL_0); + assert_approx_eq!(1.7f16.floor(), 1.0f16, TOL_0); + assert_approx_eq!(0.0f16.floor(), 0.0f16, TOL_0); + assert_approx_eq!((-0.0f16).floor(), -0.0f16, TOL_0); + assert_approx_eq!((-1.0f16).floor(), -1.0f16, TOL_0); + assert_approx_eq!((-1.3f16).floor(), -2.0f16, TOL_0); + assert_approx_eq!((-1.5f16).floor(), -2.0f16, TOL_0); + assert_approx_eq!((-1.7f16).floor(), -2.0f16, TOL_0); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_ceil() { + assert_approx_eq!(1.0f16.ceil(), 1.0f16, TOL_0); + assert_approx_eq!(1.3f16.ceil(), 2.0f16, TOL_0); + assert_approx_eq!(1.5f16.ceil(), 2.0f16, TOL_0); + assert_approx_eq!(1.7f16.ceil(), 2.0f16, TOL_0); + assert_approx_eq!(0.0f16.ceil(), 0.0f16, TOL_0); + assert_approx_eq!((-0.0f16).ceil(), -0.0f16, TOL_0); + assert_approx_eq!((-1.0f16).ceil(), -1.0f16, TOL_0); + assert_approx_eq!((-1.3f16).ceil(), -1.0f16, TOL_0); + assert_approx_eq!((-1.5f16).ceil(), -1.0f16, TOL_0); + assert_approx_eq!((-1.7f16).ceil(), -1.0f16, TOL_0); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_round() { + assert_approx_eq!(2.5f16.round(), 3.0f16, TOL_0); + assert_approx_eq!(1.0f16.round(), 1.0f16, TOL_0); + assert_approx_eq!(1.3f16.round(), 1.0f16, TOL_0); + assert_approx_eq!(1.5f16.round(), 2.0f16, TOL_0); + assert_approx_eq!(1.7f16.round(), 2.0f16, TOL_0); + assert_approx_eq!(0.0f16.round(), 0.0f16, TOL_0); + assert_approx_eq!((-0.0f16).round(), -0.0f16, TOL_0); + assert_approx_eq!((-1.0f16).round(), -1.0f16, TOL_0); + assert_approx_eq!((-1.3f16).round(), -1.0f16, TOL_0); + assert_approx_eq!((-1.5f16).round(), -2.0f16, TOL_0); + assert_approx_eq!((-1.7f16).round(), -2.0f16, TOL_0); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_round_ties_even() { + assert_approx_eq!(2.5f16.round_ties_even(), 2.0f16, TOL_0); + assert_approx_eq!(1.0f16.round_ties_even(), 1.0f16, TOL_0); + assert_approx_eq!(1.3f16.round_ties_even(), 1.0f16, TOL_0); + assert_approx_eq!(1.5f16.round_ties_even(), 2.0f16, TOL_0); + assert_approx_eq!(1.7f16.round_ties_even(), 2.0f16, TOL_0); + assert_approx_eq!(0.0f16.round_ties_even(), 0.0f16, TOL_0); + assert_approx_eq!((-0.0f16).round_ties_even(), -0.0f16, TOL_0); + assert_approx_eq!((-1.0f16).round_ties_even(), -1.0f16, TOL_0); + assert_approx_eq!((-1.3f16).round_ties_even(), -1.0f16, TOL_0); + assert_approx_eq!((-1.5f16).round_ties_even(), -2.0f16, TOL_0); + assert_approx_eq!((-1.7f16).round_ties_even(), -2.0f16, TOL_0); +} #[test] +#[cfg(reliable_f16_math)] +fn test_trunc() { + assert_approx_eq!(1.0f16.trunc(), 1.0f16, TOL_0); + assert_approx_eq!(1.3f16.trunc(), 1.0f16, TOL_0); + assert_approx_eq!(1.5f16.trunc(), 1.0f16, TOL_0); + assert_approx_eq!(1.7f16.trunc(), 1.0f16, TOL_0); + assert_approx_eq!(0.0f16.trunc(), 0.0f16, TOL_0); + assert_approx_eq!((-0.0f16).trunc(), -0.0f16, TOL_0); + assert_approx_eq!((-1.0f16).trunc(), -1.0f16, TOL_0); + assert_approx_eq!((-1.3f16).trunc(), -1.0f16, TOL_0); + assert_approx_eq!((-1.5f16).trunc(), -1.0f16, TOL_0); + assert_approx_eq!((-1.7f16).trunc(), -1.0f16, TOL_0); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_fract() { + assert_approx_eq!(1.0f16.fract(), 0.0f16, TOL_0); + assert_approx_eq!(1.3f16.fract(), 0.3f16, TOL_0); + assert_approx_eq!(1.5f16.fract(), 0.5f16, TOL_0); + assert_approx_eq!(1.7f16.fract(), 0.7f16, TOL_0); + assert_approx_eq!(0.0f16.fract(), 0.0f16, TOL_0); + assert_approx_eq!((-0.0f16).fract(), -0.0f16, TOL_0); + assert_approx_eq!((-1.0f16).fract(), -0.0f16, TOL_0); + assert_approx_eq!((-1.3f16).fract(), -0.3f16, TOL_0); + assert_approx_eq!((-1.5f16).fract(), -0.5f16, TOL_0); + assert_approx_eq!((-1.7f16).fract(), -0.7f16, TOL_0); +} + +#[test] +#[cfg(reliable_f16_math)] fn test_abs() { assert_eq!(f16::INFINITY.abs(), f16::INFINITY); assert_eq!(1f16.abs(), 1f16); @@ -299,6 +400,24 @@ fn test_next_down() { } #[test] +#[cfg(reliable_f16_math)] +fn test_mul_add() { + let nan: f16 = f16::NAN; + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + assert_approx_eq!(12.3f16.mul_add(4.5, 6.7), 62.05, TOL_P2); + assert_approx_eq!((-12.3f16).mul_add(-4.5, -6.7), 48.65, TOL_P2); + assert_approx_eq!(0.0f16.mul_add(8.9, 1.2), 1.2, TOL_0); + assert_approx_eq!(3.4f16.mul_add(-0.0, 5.6), 5.6, TOL_0); + assert!(nan.mul_add(7.8, 9.0).is_nan()); + assert_eq!(inf.mul_add(7.8, 9.0), inf); + assert_eq!(neg_inf.mul_add(7.8, 9.0), neg_inf); + assert_eq!(8.9f16.mul_add(inf, 3.2), inf); + assert_eq!((-3.2f16).mul_add(2.4, neg_inf), neg_inf); +} + +#[test] +#[cfg(reliable_f16_math)] fn test_recip() { let nan: f16 = f16::NAN; let inf: f16 = f16::INFINITY; @@ -307,11 +426,157 @@ fn test_recip() { assert_eq!(2.0f16.recip(), 0.5); assert_eq!((-0.4f16).recip(), -2.5); assert_eq!(0.0f16.recip(), inf); + assert_approx_eq!(f16::MAX.recip(), 1.526624e-5f16, 1e-4); assert!(nan.recip().is_nan()); assert_eq!(inf.recip(), 0.0); assert_eq!(neg_inf.recip(), 0.0); } +#[test] +#[cfg(reliable_f16_math)] +fn test_powi() { + // FIXME(llvm19): LLVM misoptimizes `powi.f16` + // + // let nan: f16 = f16::NAN; + // let inf: f16 = f16::INFINITY; + // let neg_inf: f16 = f16::NEG_INFINITY; + // assert_eq!(1.0f16.powi(1), 1.0); + // assert_approx_eq!((-3.1f16).powi(2), 9.61, TOL_0); + // assert_approx_eq!(5.9f16.powi(-2), 0.028727, TOL_N2); + // assert_eq!(8.3f16.powi(0), 1.0); + // assert!(nan.powi(2).is_nan()); + // assert_eq!(inf.powi(3), inf); + // assert_eq!(neg_inf.powi(2), inf); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_powf() { + let nan: f16 = f16::NAN; + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + assert_eq!(1.0f16.powf(1.0), 1.0); + assert_approx_eq!(3.4f16.powf(4.5), 246.408183, TOL_P2); + assert_approx_eq!(2.7f16.powf(-3.2), 0.041652, TOL_N2); + assert_approx_eq!((-3.1f16).powf(2.0), 9.61, TOL_P2); + assert_approx_eq!(5.9f16.powf(-2.0), 0.028727, TOL_N2); + assert_eq!(8.3f16.powf(0.0), 1.0); + assert!(nan.powf(2.0).is_nan()); + assert_eq!(inf.powf(2.0), inf); + assert_eq!(neg_inf.powf(3.0), neg_inf); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_sqrt_domain() { + assert!(f16::NAN.sqrt().is_nan()); + assert!(f16::NEG_INFINITY.sqrt().is_nan()); + assert!((-1.0f16).sqrt().is_nan()); + assert_eq!((-0.0f16).sqrt(), -0.0); + assert_eq!(0.0f16.sqrt(), 0.0); + assert_eq!(1.0f16.sqrt(), 1.0); + assert_eq!(f16::INFINITY.sqrt(), f16::INFINITY); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_exp() { + assert_eq!(1.0, 0.0f16.exp()); + assert_approx_eq!(2.718282, 1.0f16.exp(), TOL_0); + assert_approx_eq!(148.413159, 5.0f16.exp(), TOL_0); + + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + let nan: f16 = f16::NAN; + assert_eq!(inf, inf.exp()); + assert_eq!(0.0, neg_inf.exp()); + assert!(nan.exp().is_nan()); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_exp2() { + assert_eq!(32.0, 5.0f16.exp2()); + assert_eq!(1.0, 0.0f16.exp2()); + + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + let nan: f16 = f16::NAN; + assert_eq!(inf, inf.exp2()); + assert_eq!(0.0, neg_inf.exp2()); + assert!(nan.exp2().is_nan()); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_ln() { + let nan: f16 = f16::NAN; + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + assert_approx_eq!(1.0f16.exp().ln(), 1.0, TOL_0); + assert!(nan.ln().is_nan()); + assert_eq!(inf.ln(), inf); + assert!(neg_inf.ln().is_nan()); + assert!((-2.3f16).ln().is_nan()); + assert_eq!((-0.0f16).ln(), neg_inf); + assert_eq!(0.0f16.ln(), neg_inf); + assert_approx_eq!(4.0f16.ln(), 1.386294, TOL_0); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_log() { + let nan: f16 = f16::NAN; + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + assert_eq!(10.0f16.log(10.0), 1.0); + assert_approx_eq!(2.3f16.log(3.5), 0.664858, TOL_0); + assert_eq!(1.0f16.exp().log(1.0f16.exp()), 1.0); + assert!(1.0f16.log(1.0).is_nan()); + assert!(1.0f16.log(-13.9).is_nan()); + assert!(nan.log(2.3).is_nan()); + assert_eq!(inf.log(10.0), inf); + assert!(neg_inf.log(8.8).is_nan()); + assert!((-2.3f16).log(0.1).is_nan()); + assert_eq!((-0.0f16).log(2.0), neg_inf); + assert_eq!(0.0f16.log(7.0), neg_inf); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_log2() { + let nan: f16 = f16::NAN; + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + assert_approx_eq!(10.0f16.log2(), 3.321928, TOL_0); + assert_approx_eq!(2.3f16.log2(), 1.201634, TOL_0); + assert_approx_eq!(1.0f16.exp().log2(), 1.442695, TOL_0); + assert!(nan.log2().is_nan()); + assert_eq!(inf.log2(), inf); + assert!(neg_inf.log2().is_nan()); + assert!((-2.3f16).log2().is_nan()); + assert_eq!((-0.0f16).log2(), neg_inf); + assert_eq!(0.0f16.log2(), neg_inf); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_log10() { + let nan: f16 = f16::NAN; + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + assert_eq!(10.0f16.log10(), 1.0); + assert_approx_eq!(2.3f16.log10(), 0.361728, TOL_0); + assert_approx_eq!(1.0f16.exp().log10(), 0.434294, TOL_0); + assert_eq!(1.0f16.log10(), 0.0); + assert!(nan.log10().is_nan()); + assert_eq!(inf.log10(), inf); + assert!(neg_inf.log10().is_nan()); + assert!((-2.3f16).log10().is_nan()); + assert_eq!((-0.0f16).log10(), neg_inf); + assert_eq!(0.0f16.log10(), neg_inf); +} + #[test] fn test_to_degrees() { let pi: f16 = consts::PI; @@ -319,8 +584,8 @@ fn test_to_degrees() { let inf: f16 = f16::INFINITY; let neg_inf: f16 = f16::NEG_INFINITY; assert_eq!(0.0f16.to_degrees(), 0.0); - assert_approx_eq!((-5.8f16).to_degrees(), -332.315521); - assert_approx_eq!(pi.to_degrees(), 180.0, F16_APPROX_L4); + assert_approx_eq!((-5.8f16).to_degrees(), -332.315521, TOL_P2); + assert_approx_eq!(pi.to_degrees(), 180.0, TOL_P2); assert!(nan.to_degrees().is_nan()); assert_eq!(inf.to_degrees(), inf); assert_eq!(neg_inf.to_degrees(), neg_inf); @@ -334,14 +599,112 @@ fn test_to_radians() { let inf: f16 = f16::INFINITY; let neg_inf: f16 = f16::NEG_INFINITY; assert_eq!(0.0f16.to_radians(), 0.0); - assert_approx_eq!(154.6f16.to_radians(), 2.698279); - assert_approx_eq!((-332.31f16).to_radians(), -5.799903); - assert_approx_eq!(180.0f16.to_radians(), pi, F16_APPROX_L2); + assert_approx_eq!(154.6f16.to_radians(), 2.698279, TOL_0); + assert_approx_eq!((-332.31f16).to_radians(), -5.799903, TOL_0); + assert_approx_eq!(180.0f16.to_radians(), pi, TOL_0); assert!(nan.to_radians().is_nan()); assert_eq!(inf.to_radians(), inf); assert_eq!(neg_inf.to_radians(), neg_inf); } +#[test] +#[cfg(reliable_f16_math)] +fn test_asinh() { + assert_eq!(0.0f16.asinh(), 0.0f16); + assert_eq!((-0.0f16).asinh(), -0.0f16); + + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + let nan: f16 = f16::NAN; + assert_eq!(inf.asinh(), inf); + assert_eq!(neg_inf.asinh(), neg_inf); + assert!(nan.asinh().is_nan()); + assert!((-0.0f16).asinh().is_sign_negative()); + // issue 63271 + assert_approx_eq!(2.0f16.asinh(), 1.443635475178810342493276740273105f16, TOL_0); + assert_approx_eq!((-2.0f16).asinh(), -1.443635475178810342493276740273105f16, TOL_0); + // regression test for the catastrophic cancellation fixed in 72486 + assert_approx_eq!((-200.0f16).asinh(), -5.991470797049389, TOL_0); + + // test for low accuracy from issue 104548 + assert_approx_eq!(10.0f16, 10.0f16.sinh().asinh(), TOL_0); + // mul needed for approximate comparison to be meaningful + assert_approx_eq!(1.0f16, 1e-3f16.sinh().asinh() * 1e3f16, TOL_0); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_acosh() { + assert_eq!(1.0f16.acosh(), 0.0f16); + assert!(0.999f16.acosh().is_nan()); + + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + let nan: f16 = f16::NAN; + assert_eq!(inf.acosh(), inf); + assert!(neg_inf.acosh().is_nan()); + assert!(nan.acosh().is_nan()); + assert_approx_eq!(2.0f16.acosh(), 1.31695789692481670862504634730796844f16, TOL_0); + assert_approx_eq!(3.0f16.acosh(), 1.76274717403908605046521864995958461f16, TOL_0); + + // test for low accuracy from issue 104548 + assert_approx_eq!(10.0f16, 10.0f16.cosh().acosh(), TOL_P2); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_atanh() { + assert_eq!(0.0f16.atanh(), 0.0f16); + assert_eq!((-0.0f16).atanh(), -0.0f16); + + let inf: f16 = f16::INFINITY; + let neg_inf: f16 = f16::NEG_INFINITY; + let nan: f16 = f16::NAN; + assert_eq!(1.0f16.atanh(), inf); + assert_eq!((-1.0f16).atanh(), neg_inf); + assert!(2f16.atanh().atanh().is_nan()); + assert!((-2f16).atanh().atanh().is_nan()); + assert!(inf.atanh().is_nan()); + assert!(neg_inf.atanh().is_nan()); + assert!(nan.atanh().is_nan()); + assert_approx_eq!(0.5f16.atanh(), 0.54930614433405484569762261846126285f16, TOL_0); + assert_approx_eq!((-0.5f16).atanh(), -0.54930614433405484569762261846126285f16, TOL_0); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_gamma() { + // precision can differ among platforms + assert_approx_eq!(1.0f16.gamma(), 1.0f16, TOL_0); + assert_approx_eq!(2.0f16.gamma(), 1.0f16, TOL_0); + assert_approx_eq!(3.0f16.gamma(), 2.0f16, TOL_0); + assert_approx_eq!(4.0f16.gamma(), 6.0f16, TOL_0); + assert_approx_eq!(5.0f16.gamma(), 24.0f16, TOL_0); + assert_approx_eq!(0.5f16.gamma(), consts::PI.sqrt(), TOL_0); + assert_approx_eq!((-0.5f16).gamma(), -2.0 * consts::PI.sqrt(), TOL_0); + assert_eq!(0.0f16.gamma(), f16::INFINITY); + assert_eq!((-0.0f16).gamma(), f16::NEG_INFINITY); + assert!((-1.0f16).gamma().is_nan()); + assert!((-2.0f16).gamma().is_nan()); + assert!(f16::NAN.gamma().is_nan()); + assert!(f16::NEG_INFINITY.gamma().is_nan()); + assert_eq!(f16::INFINITY.gamma(), f16::INFINITY); + assert_eq!(171.71f16.gamma(), f16::INFINITY); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_ln_gamma() { + assert_approx_eq!(1.0f16.ln_gamma().0, 0.0f16, TOL_0); + assert_eq!(1.0f16.ln_gamma().1, 1); + assert_approx_eq!(2.0f16.ln_gamma().0, 0.0f16, TOL_0); + assert_eq!(2.0f16.ln_gamma().1, 1); + assert_approx_eq!(3.0f16.ln_gamma().0, 2.0f16.ln(), TOL_0); + assert_eq!(3.0f16.ln_gamma().1, 1); + assert_approx_eq!((-0.5f16).ln_gamma().0, (2.0 * consts::PI.sqrt()).ln(), TOL_0); + assert_eq!((-0.5f16).ln_gamma().1, -1); +} + #[test] fn test_real_consts() { // FIXME(f16_f128): add math tests when available @@ -355,29 +718,34 @@ fn test_real_consts() { let frac_pi_8: f16 = consts::FRAC_PI_8; let frac_1_pi: f16 = consts::FRAC_1_PI; let frac_2_pi: f16 = consts::FRAC_2_PI; - // let frac_2_sqrtpi: f16 = consts::FRAC_2_SQRT_PI; - // let sqrt2: f16 = consts::SQRT_2; - // let frac_1_sqrt2: f16 = consts::FRAC_1_SQRT_2; - // let e: f16 = consts::E; - // let log2_e: f16 = consts::LOG2_E; - // let log10_e: f16 = consts::LOG10_E; - // let ln_2: f16 = consts::LN_2; - // let ln_10: f16 = consts::LN_10; - - assert_approx_eq!(frac_pi_2, pi / 2f16); - assert_approx_eq!(frac_pi_3, pi / 3f16); - assert_approx_eq!(frac_pi_4, pi / 4f16); - assert_approx_eq!(frac_pi_6, pi / 6f16); - assert_approx_eq!(frac_pi_8, pi / 8f16); - assert_approx_eq!(frac_1_pi, 1f16 / pi); - assert_approx_eq!(frac_2_pi, 2f16 / pi); - // assert_approx_eq!(frac_2_sqrtpi, 2f16 / pi.sqrt()); - // assert_approx_eq!(sqrt2, 2f16.sqrt()); - // assert_approx_eq!(frac_1_sqrt2, 1f16 / 2f16.sqrt()); - // assert_approx_eq!(log2_e, e.log2()); - // assert_approx_eq!(log10_e, e.log10()); - // assert_approx_eq!(ln_2, 2f16.ln()); - // assert_approx_eq!(ln_10, 10f16.ln()); + + assert_approx_eq!(frac_pi_2, pi / 2f16, TOL_0); + assert_approx_eq!(frac_pi_3, pi / 3f16, TOL_0); + assert_approx_eq!(frac_pi_4, pi / 4f16, TOL_0); + assert_approx_eq!(frac_pi_6, pi / 6f16, TOL_0); + assert_approx_eq!(frac_pi_8, pi / 8f16, TOL_0); + assert_approx_eq!(frac_1_pi, 1f16 / pi, TOL_0); + assert_approx_eq!(frac_2_pi, 2f16 / pi, TOL_0); + + #[cfg(reliable_f16_math)] + { + let frac_2_sqrtpi: f16 = consts::FRAC_2_SQRT_PI; + let sqrt2: f16 = consts::SQRT_2; + let frac_1_sqrt2: f16 = consts::FRAC_1_SQRT_2; + let e: f16 = consts::E; + let log2_e: f16 = consts::LOG2_E; + let log10_e: f16 = consts::LOG10_E; + let ln_2: f16 = consts::LN_2; + let ln_10: f16 = consts::LN_10; + + assert_approx_eq!(frac_2_sqrtpi, 2f16 / pi.sqrt(), TOL_0); + assert_approx_eq!(sqrt2, 2f16.sqrt(), TOL_0); + assert_approx_eq!(frac_1_sqrt2, 1f16 / 2f16.sqrt(), TOL_0); + assert_approx_eq!(log2_e, e.log2(), TOL_0); + assert_approx_eq!(log10_e, e.log10(), TOL_0); + assert_approx_eq!(ln_2, 2f16.ln(), TOL_0); + assert_approx_eq!(ln_10, 10f16.ln(), TOL_0); + } } #[test] @@ -386,10 +754,10 @@ fn test_float_bits_conv() { assert_eq!((12.5f16).to_bits(), 0x4a40); assert_eq!((1337f16).to_bits(), 0x6539); assert_eq!((-14.25f16).to_bits(), 0xcb20); - assert_approx_eq!(f16::from_bits(0x3c00), 1.0); - assert_approx_eq!(f16::from_bits(0x4a40), 12.5); - assert_approx_eq!(f16::from_bits(0x6539), 1337.0); - assert_approx_eq!(f16::from_bits(0xcb20), -14.25); + assert_approx_eq!(f16::from_bits(0x3c00), 1.0, TOL_0); + assert_approx_eq!(f16::from_bits(0x4a40), 12.5, TOL_0); + assert_approx_eq!(f16::from_bits(0x6539), 1337.0, TOL_P4); + assert_approx_eq!(f16::from_bits(0xcb20), -14.25, TOL_0); // Check that NaNs roundtrip their bits regardless of signaling-ness let masked_nan1 = f16::NAN.to_bits() ^ NAN_MASK1; diff --git a/library/std/src/macros.rs b/library/std/src/macros.rs index ba519afc62b07..1b0d7f3dbf2c9 100644 --- a/library/std/src/macros.rs +++ b/library/std/src/macros.rs @@ -382,7 +382,7 @@ macro_rules! assert_approx_eq { let diff = (*a - *b).abs(); assert!( diff < $lim, - "{a:?} is not approximately equal to {b:?} (threshold {lim:?}, actual {diff:?})", + "{a:?} is not approximately equal to {b:?} (threshold {lim:?}, difference {diff:?})", lim = $lim ); }}; diff --git a/library/std/src/sys/cmath.rs b/library/std/src/sys/cmath.rs index 99df503b82de2..2997e908fa1b2 100644 --- a/library/std/src/sys/cmath.rs +++ b/library/std/src/sys/cmath.rs @@ -28,6 +28,21 @@ extern "C" { pub fn lgamma_r(n: f64, s: &mut i32) -> f64; pub fn lgammaf_r(n: f32, s: &mut i32) -> f32; + pub fn acosf128(n: f128) -> f128; + pub fn asinf128(n: f128) -> f128; + pub fn atanf128(n: f128) -> f128; + pub fn atan2f128(a: f128, b: f128) -> f128; + pub fn cbrtf128(n: f128) -> f128; + pub fn coshf128(n: f128) -> f128; + pub fn expm1f128(n: f128) -> f128; + pub fn hypotf128(x: f128, y: f128) -> f128; + pub fn log1pf128(n: f128) -> f128; + pub fn sinhf128(n: f128) -> f128; + pub fn tanf128(n: f128) -> f128; + pub fn tanhf128(n: f128) -> f128; + pub fn tgammaf128(n: f128) -> f128; + pub fn lgammaf128_r(n: f128, s: &mut i32) -> f128; + cfg_if::cfg_if! { if #[cfg(not(all(target_os = "windows", target_env = "msvc", target_arch = "x86")))] { pub fn acosf(n: f32) -> f32; From e18036c7694df9824a1cd97ce8e1b18be576e569 Mon Sep 17 00:00:00 2001 From: Trevor Gross Date: Wed, 17 Jul 2024 16:17:09 -0400 Subject: [PATCH 04/28] Add `core` functions for `f16` and `f128` that require math routines `min`, `max`, and similar functions require external math routines. Add these under the same gates as `std` math functions (`reliable_f16_math` and `reliable_f128_math`). --- library/core/src/num/f128.rs | 176 ++++++++++++++++++++++++++++++++++ library/core/src/num/f16.rs | 171 +++++++++++++++++++++++++++++++++ library/std/src/f128/tests.rs | 28 +++++- library/std/src/f16/tests.rs | 28 +++++- 4 files changed, 401 insertions(+), 2 deletions(-) diff --git a/library/core/src/num/f128.rs b/library/core/src/num/f128.rs index 6a24748fd9e87..cfe3a20df532e 100644 --- a/library/core/src/num/f128.rs +++ b/library/core/src/num/f128.rs @@ -686,6 +686,182 @@ impl f128 { self * RADS_PER_DEG } + /// Returns the maximum of the two numbers, ignoring NaN. + /// + /// If one of the arguments is NaN, then the other argument is returned. + /// This follows the IEEE 754-2008 semantics for maxNum, except for handling of signaling NaNs; + /// this function handles all NaNs the same way and avoids maxNum's problems with associativity. + /// This also matches the behavior of libm’s fmax. + /// + /// ``` + /// #![feature(f128)] + /// # // Using aarch64 because `reliable_f128_math` is needed + /// # #[cfg(all(target_arch = "aarch64", target_os = "linux"))] { + /// + /// let x = 1.0f128; + /// let y = 2.0f128; + /// + /// assert_eq!(x.max(y), y); + /// # } + /// ``` + #[inline] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "this returns the result of the comparison, without modifying either input"] + pub fn max(self, other: f128) -> f128 { + intrinsics::maxnumf128(self, other) + } + + /// Returns the minimum of the two numbers, ignoring NaN. + /// + /// If one of the arguments is NaN, then the other argument is returned. + /// This follows the IEEE 754-2008 semantics for minNum, except for handling of signaling NaNs; + /// this function handles all NaNs the same way and avoids minNum's problems with associativity. + /// This also matches the behavior of libm’s fmin. + /// + /// ``` + /// #![feature(f128)] + /// # // Using aarch64 because `reliable_f128_math` is needed + /// # #[cfg(all(target_arch = "aarch64", target_os = "linux"))] { + /// + /// let x = 1.0f128; + /// let y = 2.0f128; + /// + /// assert_eq!(x.min(y), x); + /// # } + /// ``` + #[inline] + #[unstable(feature = "f128", issue = "116909")] + #[must_use = "this returns the result of the comparison, without modifying either input"] + pub fn min(self, other: f128) -> f128 { + intrinsics::minnumf128(self, other) + } + + /// Returns the maximum of the two numbers, propagating NaN. + /// + /// This returns NaN when *either* argument is NaN, as opposed to + /// [`f128::max`] which only returns NaN when *both* arguments are NaN. + /// + /// ``` + /// #![feature(f128)] + /// #![feature(float_minimum_maximum)] + /// # // Using aarch64 because `reliable_f128_math` is needed + /// # #[cfg(all(target_arch = "aarch64", target_os = "linux"))] { + /// + /// let x = 1.0f128; + /// let y = 2.0f128; + /// + /// assert_eq!(x.maximum(y), y); + /// assert!(x.maximum(f128::NAN).is_nan()); + /// # } + /// ``` + /// + /// If one of the arguments is NaN, then NaN is returned. Otherwise this returns the greater + /// of the two numbers. For this operation, -0.0 is considered to be less than +0.0. + /// Note that this follows the semantics specified in IEEE 754-2019. + /// + /// Also note that "propagation" of NaNs here doesn't necessarily mean that the bitpattern of a NaN + /// operand is conserved; see [explanation of NaN as a special value](f128) for more info. + #[inline] + #[unstable(feature = "f128", issue = "116909")] + // #[unstable(feature = "float_minimum_maximum", issue = "91079")] + #[must_use = "this returns the result of the comparison, without modifying either input"] + pub fn maximum(self, other: f128) -> f128 { + if self > other { + self + } else if other > self { + other + } else if self == other { + if self.is_sign_positive() && other.is_sign_negative() { self } else { other } + } else { + self + other + } + } + + /// Returns the minimum of the two numbers, propagating NaN. + /// + /// This returns NaN when *either* argument is NaN, as opposed to + /// [`f128::min`] which only returns NaN when *both* arguments are NaN. + /// + /// ``` + /// #![feature(f128)] + /// #![feature(float_minimum_maximum)] + /// # // Using aarch64 because `reliable_f128_math` is needed + /// # #[cfg(all(target_arch = "aarch64", target_os = "linux"))] { + /// + /// let x = 1.0f128; + /// let y = 2.0f128; + /// + /// assert_eq!(x.minimum(y), x); + /// assert!(x.minimum(f128::NAN).is_nan()); + /// # } + /// ``` + /// + /// If one of the arguments is NaN, then NaN is returned. Otherwise this returns the lesser + /// of the two numbers. For this operation, -0.0 is considered to be less than +0.0. + /// Note that this follows the semantics specified in IEEE 754-2019. + /// + /// Also note that "propagation" of NaNs here doesn't necessarily mean that the bitpattern of a NaN + /// operand is conserved; see [explanation of NaN as a special value](f128) for more info. + #[inline] + #[unstable(feature = "f128", issue = "116909")] + // #[unstable(feature = "float_minimum_maximum", issue = "91079")] + #[must_use = "this returns the result of the comparison, without modifying either input"] + pub fn minimum(self, other: f128) -> f128 { + if self < other { + self + } else if other < self { + other + } else if self == other { + if self.is_sign_negative() && other.is_sign_positive() { self } else { other } + } else { + // At least one input is NaN. Use `+` to perform NaN propagation and quieting. + self + other + } + } + + /// Calculates the middle point of `self` and `rhs`. + /// + /// This returns NaN when *either* argument is NaN or if a combination of + /// +inf and -inf is provided as arguments. + /// + /// # Examples + /// + /// ``` + /// #![feature(f128)] + /// #![feature(num_midpoint)] + /// # // Using aarch64 because `reliable_f128_math` is needed + /// # #[cfg(all(target_arch = "aarch64", target_os = "linux"))] { + /// + /// assert_eq!(1f128.midpoint(4.0), 2.5); + /// assert_eq!((-5.5f128).midpoint(8.0), 1.25); + /// # } + /// ``` + #[inline] + #[unstable(feature = "f128", issue = "116909")] + // #[unstable(feature = "num_midpoint", issue = "110840")] + pub fn midpoint(self, other: f128) -> f128 { + const LO: f128 = f128::MIN_POSITIVE * 2.; + const HI: f128 = f128::MAX / 2.; + + let (a, b) = (self, other); + let abs_a = a.abs_private(); + let abs_b = b.abs_private(); + + if abs_a <= HI && abs_b <= HI { + // Overflow is impossible + (a + b) / 2. + } else if abs_a < LO { + // Not safe to halve a + a + (b / 2.) + } else if abs_b < LO { + // Not safe to halve b + (a / 2.) + b + } else { + // Not safe to halve a and b + (a / 2.) + (b / 2.) + } + } + /// Rounds toward zero and converts to any primitive integer type, /// assuming that the value is finite and fits in that type. /// diff --git a/library/core/src/num/f16.rs b/library/core/src/num/f16.rs index 054897b3c96bc..8947bf0d84826 100644 --- a/library/core/src/num/f16.rs +++ b/library/core/src/num/f16.rs @@ -720,6 +720,177 @@ impl f16 { self * RADS_PER_DEG } + /// Returns the maximum of the two numbers, ignoring NaN. + /// + /// If one of the arguments is NaN, then the other argument is returned. + /// This follows the IEEE 754-2008 semantics for maxNum, except for handling of signaling NaNs; + /// this function handles all NaNs the same way and avoids maxNum's problems with associativity. + /// This also matches the behavior of libm’s fmax. + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885 + /// + /// let x = 1.0f16; + /// let y = 2.0f16; + /// + /// assert_eq!(x.max(y), y); + /// # } + /// ``` + #[inline] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "this returns the result of the comparison, without modifying either input"] + pub fn max(self, other: f16) -> f16 { + intrinsics::maxnumf16(self, other) + } + + /// Returns the minimum of the two numbers, ignoring NaN. + /// + /// If one of the arguments is NaN, then the other argument is returned. + /// This follows the IEEE 754-2008 semantics for minNum, except for handling of signaling NaNs; + /// this function handles all NaNs the same way and avoids minNum's problems with associativity. + /// This also matches the behavior of libm’s fmin. + /// + /// ``` + /// #![feature(f16)] + /// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885 + /// + /// let x = 1.0f16; + /// let y = 2.0f16; + /// + /// assert_eq!(x.min(y), x); + /// # } + /// ``` + #[inline] + #[unstable(feature = "f16", issue = "116909")] + #[must_use = "this returns the result of the comparison, without modifying either input"] + pub fn min(self, other: f16) -> f16 { + intrinsics::minnumf16(self, other) + } + + /// Returns the maximum of the two numbers, propagating NaN. + /// + /// This returns NaN when *either* argument is NaN, as opposed to + /// [`f16::max`] which only returns NaN when *both* arguments are NaN. + /// + /// ``` + /// #![feature(f16)] + /// #![feature(float_minimum_maximum)] + /// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885 + /// + /// let x = 1.0f16; + /// let y = 2.0f16; + /// + /// assert_eq!(x.maximum(y), y); + /// assert!(x.maximum(f16::NAN).is_nan()); + /// # } + /// ``` + /// + /// If one of the arguments is NaN, then NaN is returned. Otherwise this returns the greater + /// of the two numbers. For this operation, -0.0 is considered to be less than +0.0. + /// Note that this follows the semantics specified in IEEE 754-2019. + /// + /// Also note that "propagation" of NaNs here doesn't necessarily mean that the bitpattern of a NaN + /// operand is conserved; see [explanation of NaN as a special value](f16) for more info. + #[inline] + #[unstable(feature = "f16", issue = "116909")] + // #[unstable(feature = "float_minimum_maximum", issue = "91079")] + #[must_use = "this returns the result of the comparison, without modifying either input"] + pub fn maximum(self, other: f16) -> f16 { + if self > other { + self + } else if other > self { + other + } else if self == other { + if self.is_sign_positive() && other.is_sign_negative() { self } else { other } + } else { + self + other + } + } + + /// Returns the minimum of the two numbers, propagating NaN. + /// + /// This returns NaN when *either* argument is NaN, as opposed to + /// [`f16::min`] which only returns NaN when *both* arguments are NaN. + /// + /// ``` + /// #![feature(f16)] + /// #![feature(float_minimum_maximum)] + /// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885 + /// + /// let x = 1.0f16; + /// let y = 2.0f16; + /// + /// assert_eq!(x.minimum(y), x); + /// assert!(x.minimum(f16::NAN).is_nan()); + /// # } + /// ``` + /// + /// If one of the arguments is NaN, then NaN is returned. Otherwise this returns the lesser + /// of the two numbers. For this operation, -0.0 is considered to be less than +0.0. + /// Note that this follows the semantics specified in IEEE 754-2019. + /// + /// Also note that "propagation" of NaNs here doesn't necessarily mean that the bitpattern of a NaN + /// operand is conserved; see [explanation of NaN as a special value](f16) for more info. + #[inline] + #[unstable(feature = "f16", issue = "116909")] + // #[unstable(feature = "float_minimum_maximum", issue = "91079")] + #[must_use = "this returns the result of the comparison, without modifying either input"] + pub fn minimum(self, other: f16) -> f16 { + if self < other { + self + } else if other < self { + other + } else if self == other { + if self.is_sign_negative() && other.is_sign_positive() { self } else { other } + } else { + // At least one input is NaN. Use `+` to perform NaN propagation and quieting. + self + other + } + } + + /// Calculates the middle point of `self` and `rhs`. + /// + /// This returns NaN when *either* argument is NaN or if a combination of + /// +inf and -inf is provided as arguments. + /// + /// # Examples + /// + /// ``` + /// #![feature(f16)] + /// #![feature(num_midpoint)] + /// # #[cfg(target_arch = "aarch64")] { // FIXME(f16_F128): rust-lang/rust#123885 + /// + /// assert_eq!(1f16.midpoint(4.0), 2.5); + /// assert_eq!((-5.5f16).midpoint(8.0), 1.25); + /// # } + /// ``` + #[inline] + #[unstable(feature = "f16", issue = "116909")] + // #[unstable(feature = "num_midpoint", issue = "110840")] + pub fn midpoint(self, other: f16) -> f16 { + const LO: f16 = f16::MIN_POSITIVE * 2.; + const HI: f16 = f16::MAX / 2.; + + let (a, b) = (self, other); + let abs_a = a.abs_private(); + let abs_b = b.abs_private(); + + if abs_a <= HI && abs_b <= HI { + // Overflow is impossible + (a + b) / 2. + } else if abs_a < LO { + // Not safe to halve a + a + (b / 2.) + } else if abs_b < LO { + // Not safe to halve b + (a / 2.) + b + } else { + // Not safe to halve a and b + (a / 2.) + (b / 2.) + } + } + /// Rounds toward zero and converts to any primitive integer type, /// assuming that the value is finite and fits in that type. /// diff --git a/library/std/src/f128/tests.rs b/library/std/src/f128/tests.rs index df806a639f620..7051c051bf723 100644 --- a/library/std/src/f128/tests.rs +++ b/library/std/src/f128/tests.rs @@ -56,7 +56,33 @@ fn test_num_f128() { test_num(10f128, 2f128); } -// FIXME(f16_f128): add min and max tests when available +#[test] +#[cfg(reliable_f128_math)] +fn test_min_nan() { + assert_eq!(f128::NAN.min(2.0), 2.0); + assert_eq!(2.0f128.min(f128::NAN), 2.0); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_max_nan() { + assert_eq!(f128::NAN.max(2.0), 2.0); + assert_eq!(2.0f128.max(f128::NAN), 2.0); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_minimum() { + assert!(f128::NAN.minimum(2.0).is_nan()); + assert!(2.0f128.minimum(f128::NAN).is_nan()); +} + +#[test] +#[cfg(reliable_f128_math)] +fn test_maximum() { + assert!(f128::NAN.maximum(2.0).is_nan()); + assert!(2.0f128.maximum(f128::NAN).is_nan()); +} #[test] fn test_nan() { diff --git a/library/std/src/f16/tests.rs b/library/std/src/f16/tests.rs index f0ef807dac141..50504e7ffd94f 100644 --- a/library/std/src/f16/tests.rs +++ b/library/std/src/f16/tests.rs @@ -57,7 +57,33 @@ fn test_num_f16() { test_num(10f16, 2f16); } -// FIXME(f16_f128): add min and max tests when available +#[test] +#[cfg(reliable_f16_math)] +fn test_min_nan() { + assert_eq!(f16::NAN.min(2.0), 2.0); + assert_eq!(2.0f16.min(f16::NAN), 2.0); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_max_nan() { + assert_eq!(f16::NAN.max(2.0), 2.0); + assert_eq!(2.0f16.max(f16::NAN), 2.0); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_minimum() { + assert!(f16::NAN.minimum(2.0).is_nan()); + assert!(2.0f16.minimum(f16::NAN).is_nan()); +} + +#[test] +#[cfg(reliable_f16_math)] +fn test_maximum() { + assert!(f16::NAN.maximum(2.0).is_nan()); + assert!(2.0f16.maximum(f16::NAN).is_nan()); +} #[test] fn test_nan() { From 43836421f8ddb872976f50e05f164b27159966c5 Mon Sep 17 00:00:00 2001 From: Trevor Gross Date: Wed, 17 Jul 2024 20:45:13 -0400 Subject: [PATCH 05/28] Update comments for `{f16, f32, f64, f128}::midpoint` Clarify what makes some operations not safe, and correct comment in the default branch ("not safe" -> "safe"). --- library/core/src/num/f128.rs | 6 +++--- library/core/src/num/f16.rs | 6 +++--- library/core/src/num/f32.rs | 6 +++--- library/core/src/num/f64.rs | 6 +++--- 4 files changed, 12 insertions(+), 12 deletions(-) diff --git a/library/core/src/num/f128.rs b/library/core/src/num/f128.rs index cfe3a20df532e..0c04f47fe7df1 100644 --- a/library/core/src/num/f128.rs +++ b/library/core/src/num/f128.rs @@ -851,13 +851,13 @@ impl f128 { // Overflow is impossible (a + b) / 2. } else if abs_a < LO { - // Not safe to halve a + // Not safe to halve `a` (would underflow) a + (b / 2.) } else if abs_b < LO { - // Not safe to halve b + // Not safe to halve `b` (would underflow) (a / 2.) + b } else { - // Not safe to halve a and b + // Safe to halve `a` and `b` (a / 2.) + (b / 2.) } } diff --git a/library/core/src/num/f16.rs b/library/core/src/num/f16.rs index 8947bf0d84826..e5b1148e19215 100644 --- a/library/core/src/num/f16.rs +++ b/library/core/src/num/f16.rs @@ -880,13 +880,13 @@ impl f16 { // Overflow is impossible (a + b) / 2. } else if abs_a < LO { - // Not safe to halve a + // Not safe to halve `a` (would underflow) a + (b / 2.) } else if abs_b < LO { - // Not safe to halve b + // Not safe to halve `b` (would underflow) (a / 2.) + b } else { - // Not safe to halve a and b + // Safe to halve `a` and `b` (a / 2.) + (b / 2.) } } diff --git a/library/core/src/num/f32.rs b/library/core/src/num/f32.rs index 08d863f17caf7..e65c982b17227 100644 --- a/library/core/src/num/f32.rs +++ b/library/core/src/num/f32.rs @@ -1070,13 +1070,13 @@ impl f32 { // Overflow is impossible (a + b) / 2. } else if abs_a < LO { - // Not safe to halve a + // Not safe to halve `a` (would underflow) a + (b / 2.) } else if abs_b < LO { - // Not safe to halve b + // Not safe to halve `b` (would underflow) (a / 2.) + b } else { - // Not safe to halve a and b + // Safe to halve `a` and `b` (a / 2.) + (b / 2.) } } diff --git a/library/core/src/num/f64.rs b/library/core/src/num/f64.rs index 5d33eea6d011f..b27d47b07d544 100644 --- a/library/core/src/num/f64.rs +++ b/library/core/src/num/f64.rs @@ -1064,13 +1064,13 @@ impl f64 { // Overflow is impossible (a + b) / 2. } else if abs_a < LO { - // Not safe to halve a + // Not safe to halve `a` (would underflow) a + (b / 2.) } else if abs_b < LO { - // Not safe to halve b + // Not safe to halve `b` (would underflow) (a / 2.) + b } else { - // Not safe to halve a and b + // Safe to halve `a` and `b` (a / 2.) + (b / 2.) } } From 8e2ca0c9d500ea435fbf9858a4d8cec8ee7dfde3 Mon Sep 17 00:00:00 2001 From: Trevor Gross Date: Tue, 30 Jul 2024 18:11:46 -0400 Subject: [PATCH 06/28] Add a disclaimer about x86 `f128` math functions Due to a LLVM bug, `f128` math functions link successfully but LLVM chooses the wrong symbols (`long double` symbols rather than those for binary128). Since this is a notable problem that may surprise a number of users, add a note about it. Link: https://github.com/llvm/llvm-project/issues/44744 --- library/core/src/primitive_docs.rs | 3 +++ 1 file changed, 3 insertions(+) diff --git a/library/core/src/primitive_docs.rs b/library/core/src/primitive_docs.rs index 5989bcbcc5201..09ebef89fb0c2 100644 --- a/library/core/src/primitive_docs.rs +++ b/library/core/src/primitive_docs.rs @@ -1244,6 +1244,9 @@ mod prim_f64 {} /// actually implement it. For x86-64 and AArch64, ISA support is not even specified, /// so it will always be a software implementation significantly slower than `f64`. /// +/// _Note: `f128` support is incomplete. Many platforms will not be able to link math functions. On +/// x86 in particular, these functions do link but their results are always incorrect._ +/// /// *[See also the `std::f128::consts` module](crate::f128::consts).* /// /// [wikipedia]: https://en.wikipedia.org/wiki/Quadruple-precision_floating-point_format From 3268b2e18d33146f7df17b29421f8c6df3737afe Mon Sep 17 00:00:00 2001 From: Chris Denton Date: Sun, 4 Aug 2024 14:59:50 +0000 Subject: [PATCH 07/28] Enable msvc for link-args-order --- tests/run-make/link-args-order/rmake.rs | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-) diff --git a/tests/run-make/link-args-order/rmake.rs b/tests/run-make/link-args-order/rmake.rs index d238ad23f27c7..b7ef8333267f2 100644 --- a/tests/run-make/link-args-order/rmake.rs +++ b/tests/run-make/link-args-order/rmake.rs @@ -3,15 +3,14 @@ // checks that linker arguments remain intact and in the order they were originally passed in. // See https://github.com/rust-lang/rust/pull/70665 -//@ ignore-msvc -// Reason: the ld linker does not exist on Windows. - -use run_make_support::rustc; +use run_make_support::{is_msvc, rustc}; fn main() { + let linker = if is_msvc() { "msvc" } else { "ld" }; + rustc() .input("empty.rs") - .linker_flavor("ld") + .linker_flavor(linker) .link_arg("a") .link_args("b c") .link_args("d e") @@ -20,7 +19,7 @@ fn main() { .assert_stderr_contains(r#""a" "b" "c" "d" "e" "f""#); rustc() .input("empty.rs") - .linker_flavor("ld") + .linker_flavor(linker) .arg("-Zpre-link-arg=a") .arg("-Zpre-link-args=b c") .arg("-Zpre-link-args=d e") From 1737845cb41c3eefe5318fd6bb8db4dbabe03f69 Mon Sep 17 00:00:00 2001 From: Chris Denton Date: Sun, 4 Aug 2024 16:19:00 +0000 Subject: [PATCH 08/28] Enable msvc for run-make/rust-lld This is simply a matter of using the right argument for lld-link. --- tests/run-make/rust-lld/rmake.rs | 14 ++++++++------ 1 file changed, 8 insertions(+), 6 deletions(-) diff --git a/tests/run-make/rust-lld/rmake.rs b/tests/run-make/rust-lld/rmake.rs index 87477c122301d..d0bc19130d7cb 100644 --- a/tests/run-make/rust-lld/rmake.rs +++ b/tests/run-make/rust-lld/rmake.rs @@ -2,15 +2,17 @@ // see https://github.com/rust-lang/compiler-team/issues/510 for more info //@ needs-rust-lld -//@ ignore-msvc //@ ignore-s390x lld does not yet support s390x as target use std::process::Output; use run_make_support::regex::Regex; -use run_make_support::rustc; +use run_make_support::{is_msvc, rustc}; fn main() { + // lld-link is used if msvc, otherwise a gnu-compatible lld is used. + let linker_version_flag = if is_msvc() { "--version" } else { "-Wl,-v" }; + // Opt-in to lld and the self-contained linker, to link with rust-lld. We'll check that by // asking the linker to display its version number with a link-arg. let output = rustc() @@ -18,7 +20,7 @@ fn main() { .arg("-Zlinker-features=+lld") .arg("-Clink-self-contained=+linker") .arg("-Zunstable-options") - .link_arg("-Wl,-v") + .link_arg(linker_version_flag) .input("main.rs") .run(); assert!( @@ -27,10 +29,10 @@ fn main() { output.stderr_utf8() ); - // It should not be used when we explictly opt-out of lld. + // It should not be used when we explicitly opt-out of lld. let output = rustc() .env("RUSTC_LOG", "rustc_codegen_ssa::back::link=info") - .link_arg("-Wl,-v") + .link_arg(linker_version_flag) .arg("-Zlinker-features=-lld") .input("main.rs") .run(); @@ -44,7 +46,7 @@ fn main() { // times to rustc. let output = rustc() .env("RUSTC_LOG", "rustc_codegen_ssa::back::link=info") - .link_arg("-Wl,-v") + .link_arg(linker_version_flag) .arg("-Clink-self-contained=+linker") .arg("-Zunstable-options") .arg("-Zlinker-features=-lld") From 131d453248b4ff7e7f6d91137d924c8216e94bf9 Mon Sep 17 00:00:00 2001 From: Oneirical Date: Tue, 23 Jul 2024 13:02:48 -0400 Subject: [PATCH 09/28] rewrite raw-dylib-alt-calling-conventions to rmake --- .../tidy/src/allowed_run_make_makefiles.txt | 1 - .../raw-dylib-alt-calling-convention/Makefile | 24 ---------------- .../raw-dylib-alt-calling-convention/rmake.rs | 28 +++++++++++++++++++ 3 files changed, 28 insertions(+), 25 deletions(-) delete mode 100644 tests/run-make/raw-dylib-alt-calling-convention/Makefile create mode 100644 tests/run-make/raw-dylib-alt-calling-convention/rmake.rs diff --git a/src/tools/tidy/src/allowed_run_make_makefiles.txt b/src/tools/tidy/src/allowed_run_make_makefiles.txt index 088d5ba0c2b37..ed20a8300754a 100644 --- a/src/tools/tidy/src/allowed_run_make_makefiles.txt +++ b/src/tools/tidy/src/allowed_run_make_makefiles.txt @@ -23,7 +23,6 @@ run-make/no-alloc-shim/Makefile run-make/pdb-buildinfo-cl-cmd/Makefile run-make/pgo-gen-lto/Makefile run-make/pgo-indirect-call-promotion/Makefile -run-make/raw-dylib-alt-calling-convention/Makefile run-make/raw-dylib-c/Makefile run-make/redundant-libs/Makefile run-make/remap-path-prefix-dwarf/Makefile diff --git a/tests/run-make/raw-dylib-alt-calling-convention/Makefile b/tests/run-make/raw-dylib-alt-calling-convention/Makefile deleted file mode 100644 index 14d23a5d20106..0000000000000 --- a/tests/run-make/raw-dylib-alt-calling-convention/Makefile +++ /dev/null @@ -1,24 +0,0 @@ -# Test the behavior of #[link(.., kind = "raw-dylib")] with alternative calling conventions. - -# only-x86 -# only-windows - -include ../tools.mk - -all: - $(RUSTC) --crate-type lib --crate-name raw_dylib_alt_calling_convention_test lib.rs - $(RUSTC) --crate-type bin driver.rs -L "$(TMPDIR)" - $(call COMPILE_OBJ,"$(TMPDIR)"/extern.obj,extern.c) -ifdef IS_MSVC - $(CC) "$(TMPDIR)"/extern.obj -link -dll -out:"$(TMPDIR)"/extern.dll -noimplib -else - $(CC) "$(TMPDIR)"/extern.obj -shared -o "$(TMPDIR)"/extern.dll -endif - - "$(TMPDIR)"/driver > "$(TMPDIR)"/output.txt - $(RUSTC_TEST_OP) "$(TMPDIR)"/output.txt output.txt - -ifdef IS_MSVC - "$(TMPDIR)"/driver true > "$(TMPDIR)"/output.msvc.txt - $(RUSTC_TEST_OP) "$(TMPDIR)"/output.msvc.txt output.msvc.txt -endif diff --git a/tests/run-make/raw-dylib-alt-calling-convention/rmake.rs b/tests/run-make/raw-dylib-alt-calling-convention/rmake.rs new file mode 100644 index 0000000000000..736b3fee2aef9 --- /dev/null +++ b/tests/run-make/raw-dylib-alt-calling-convention/rmake.rs @@ -0,0 +1,28 @@ +// `raw-dylib` is a Windows-specific attribute which emits idata sections for the items in the +// attached extern block, +// so they may be linked against without linking against an import library. +// To learn more, read https://github.com/rust-lang/rfcs/blob/master/text/2627-raw-dylib-kind.md +// This test uses this feature alongside alternative calling conventions, checking that both +// features are compatible and result in the expected output upon execution of the binary. +// See https://github.com/rust-lang/rust/pull/84171 + +//@ only-x86 +//@ only-windows + +use run_make_support::{build_native_dynamic_lib, diff, is_msvc, run, run_with_args, rustc}; + +fn main() { + rustc() + .crate_type("lib") + .crate_name("raw_dylib_alt_calling_convention_test") + .input("lib.rs") + .run(); + rustc().crate_type("bin").input("driver.rs").run(); + build_native_dynamic_lib("extern"); + let out = run("driver").stdout_utf8(); + diff().expected_file("output.txt").actual_text("actual", out).run(); + if is_msvc() { + let out_msvc = run_with_args("driver", &["true"]).stdout_utf8(); + diff().expected_file("output.msvc.txt").actual_text("actual", out_msvc).run(); + } +} From f31f8c488aa2fa22d595ff64091008da39bfe347 Mon Sep 17 00:00:00 2001 From: Oneirical Date: Tue, 23 Jul 2024 13:21:41 -0400 Subject: [PATCH 10/28] rewrite raw-dylib-c to rmake --- .../tidy/src/allowed_run_make_makefiles.txt | 1 - tests/run-make/raw-dylib-c/Makefile | 28 ------------------ tests/run-make/raw-dylib-c/rmake.rs | 29 +++++++++++++++++++ 3 files changed, 29 insertions(+), 29 deletions(-) delete mode 100644 tests/run-make/raw-dylib-c/Makefile create mode 100644 tests/run-make/raw-dylib-c/rmake.rs diff --git a/src/tools/tidy/src/allowed_run_make_makefiles.txt b/src/tools/tidy/src/allowed_run_make_makefiles.txt index ed20a8300754a..46c8b9aff9506 100644 --- a/src/tools/tidy/src/allowed_run_make_makefiles.txt +++ b/src/tools/tidy/src/allowed_run_make_makefiles.txt @@ -23,7 +23,6 @@ run-make/no-alloc-shim/Makefile run-make/pdb-buildinfo-cl-cmd/Makefile run-make/pgo-gen-lto/Makefile run-make/pgo-indirect-call-promotion/Makefile -run-make/raw-dylib-c/Makefile run-make/redundant-libs/Makefile run-make/remap-path-prefix-dwarf/Makefile run-make/reproducible-build-2/Makefile diff --git a/tests/run-make/raw-dylib-c/Makefile b/tests/run-make/raw-dylib-c/Makefile deleted file mode 100644 index af5c4a6edd7ba..0000000000000 --- a/tests/run-make/raw-dylib-c/Makefile +++ /dev/null @@ -1,28 +0,0 @@ -# Test the behavior of #[link(.., kind = "raw-dylib")] on windows-msvc - -# only-windows - -include ../tools.mk - -all: - $(RUSTC) --crate-type lib --crate-name raw_dylib_test lib.rs - $(RUSTC) --crate-type bin driver.rs -L "$(TMPDIR)" - $(RUSTC) --crate-type bin --crate-name raw_dylib_test_bin lib.rs - $(call COMPILE_OBJ,"$(TMPDIR)"/extern_1.obj,extern_1.c) - $(call COMPILE_OBJ,"$(TMPDIR)"/extern_2.obj,extern_2.c) -ifdef IS_MSVC - $(CC) "$(TMPDIR)"/extern_1.obj -link -dll -out:"$(TMPDIR)"/extern_1.dll -noimplib - $(CC) "$(TMPDIR)"/extern_2.obj -link -dll -out:"$(TMPDIR)"/extern_2.dll -noimplib -else - $(CC) "$(TMPDIR)"/extern_1.obj -shared -o "$(TMPDIR)"/extern_1.dll - $(CC) "$(TMPDIR)"/extern_2.obj -shared -o "$(TMPDIR)"/extern_2.dll -endif - "$(TMPDIR)"/driver | tr -d '\r' > "$(TMPDIR)"/output.txt - "$(TMPDIR)"/raw_dylib_test_bin > "$(TMPDIR)"/output_bin.txt - -ifdef RUSTC_BLESS_TEST - cp "$(TMPDIR)"/output.txt output.txt -else - $(DIFF) output.txt "$(TMPDIR)"/output.txt - $(DIFF) output.txt "$(TMPDIR)"/output_bin.txt -endif diff --git a/tests/run-make/raw-dylib-c/rmake.rs b/tests/run-make/raw-dylib-c/rmake.rs new file mode 100644 index 0000000000000..e4e390b2b6482 --- /dev/null +++ b/tests/run-make/raw-dylib-c/rmake.rs @@ -0,0 +1,29 @@ +// `raw-dylib` is a Windows-specific attribute which emits idata sections for the items in the +// attached extern block, +// so they may be linked against without linking against an import library. +// To learn more, read https://github.com/rust-lang/rfcs/blob/master/text/2627-raw-dylib-kind.md +// This test is the simplest of the raw-dylib tests, simply smoke-testing that the feature +// can be used to build an executable binary with an expected output with native C files +// compiling into dynamic libraries. +// See https://github.com/rust-lang/rust/pull/86419 + +//@ only-windows + +use run_make_support::{build_native_dynamic_lib, diff, run, rustc}; + +fn main() { + rustc().crate_type("lib").crate_name("raw_dylib_test").input("lib.rs").run(); + rustc().crate_type("bin").input("driver.rs").run(); + rustc().crate_type("bin").crate_name("raw_dylib_test_bin").input("lib.rs").run(); + build_native_dynamic_lib("extern_1"); + build_native_dynamic_lib("extern_2"); + let out_driver = run("driver").stdout_utf8(); + let out_raw = run("raw_dylib_test_bin").stdout_utf8(); + + diff() + .expected_file("output.txt") + .actual_text("actual", out_driver) + .normalize(r#"\r"#, "") + .run(); + diff().expected_file("output.txt").actual_text("actual", out_raw).run(); +} From 011727f14ebf66a52f282cd865f5e42d140debe0 Mon Sep 17 00:00:00 2001 From: Oneirical Date: Tue, 23 Jul 2024 13:29:14 -0400 Subject: [PATCH 11/28] rewrite redundant-libs to rmake --- .../tidy/src/allowed_run_make_makefiles.txt | 1 - .../raw-dylib-alt-calling-convention/rmake.rs | 8 +++-- tests/run-make/raw-dylib-c/rmake.rs | 2 +- tests/run-make/redundant-libs/Makefile | 24 ------------- tests/run-make/redundant-libs/rmake.rs | 34 +++++++++++++++++++ 5 files changed, 41 insertions(+), 28 deletions(-) delete mode 100644 tests/run-make/redundant-libs/Makefile create mode 100644 tests/run-make/redundant-libs/rmake.rs diff --git a/src/tools/tidy/src/allowed_run_make_makefiles.txt b/src/tools/tidy/src/allowed_run_make_makefiles.txt index 46c8b9aff9506..892f3d62402bb 100644 --- a/src/tools/tidy/src/allowed_run_make_makefiles.txt +++ b/src/tools/tidy/src/allowed_run_make_makefiles.txt @@ -23,7 +23,6 @@ run-make/no-alloc-shim/Makefile run-make/pdb-buildinfo-cl-cmd/Makefile run-make/pgo-gen-lto/Makefile run-make/pgo-indirect-call-promotion/Makefile -run-make/redundant-libs/Makefile run-make/remap-path-prefix-dwarf/Makefile run-make/reproducible-build-2/Makefile run-make/reproducible-build/Makefile diff --git a/tests/run-make/raw-dylib-alt-calling-convention/rmake.rs b/tests/run-make/raw-dylib-alt-calling-convention/rmake.rs index 736b3fee2aef9..1a1622f275421 100644 --- a/tests/run-make/raw-dylib-alt-calling-convention/rmake.rs +++ b/tests/run-make/raw-dylib-alt-calling-convention/rmake.rs @@ -20,9 +20,13 @@ fn main() { rustc().crate_type("bin").input("driver.rs").run(); build_native_dynamic_lib("extern"); let out = run("driver").stdout_utf8(); - diff().expected_file("output.txt").actual_text("actual", out).run(); + diff().expected_file("output.txt").actual_text("actual", out).normalize(r#"\r"#, "").run(); if is_msvc() { let out_msvc = run_with_args("driver", &["true"]).stdout_utf8(); - diff().expected_file("output.msvc.txt").actual_text("actual", out_msvc).run(); + diff() + .expected_file("output.msvc.txt") + .actual_text("actual", out_msvc) + .normalize(r#"\r"#, "") + .run(); } } diff --git a/tests/run-make/raw-dylib-c/rmake.rs b/tests/run-make/raw-dylib-c/rmake.rs index e4e390b2b6482..3cfd8cb400bbf 100644 --- a/tests/run-make/raw-dylib-c/rmake.rs +++ b/tests/run-make/raw-dylib-c/rmake.rs @@ -25,5 +25,5 @@ fn main() { .actual_text("actual", out_driver) .normalize(r#"\r"#, "") .run(); - diff().expected_file("output.txt").actual_text("actual", out_raw).run(); + diff().expected_file("output.txt").actual_text("actual", out_raw).normalize(r#"\r"#, "").run(); } diff --git a/tests/run-make/redundant-libs/Makefile b/tests/run-make/redundant-libs/Makefile deleted file mode 100644 index 0a48b2b280136..0000000000000 --- a/tests/run-make/redundant-libs/Makefile +++ /dev/null @@ -1,24 +0,0 @@ -# ignore-cross-compile -include ../tools.mk - -# ignore-windows-msvc - -# rustc will remove one of the two redundant references to foo below. Depending -# on which one gets removed, we'll get a linker error on SOME platforms (like -# Linux). On these platforms, when a library is referenced, the linker will -# only pull in the symbols needed _at that point in time_. If a later library -# depends on additional symbols from the library, they will not have been pulled -# in, and you'll get undefined symbols errors. -# -# So in this example, we need to ensure that rustc keeps the _later_ reference -# to foo, and not the former one. -RUSTC_FLAGS = \ - -l static=bar \ - -l foo \ - -l static=baz \ - -l foo \ - --print link-args - -all: $(call DYLIB,foo) $(call STATICLIB,bar) $(call STATICLIB,baz) - $(RUSTC) $(RUSTC_FLAGS) main.rs - $(call RUN,main) diff --git a/tests/run-make/redundant-libs/rmake.rs b/tests/run-make/redundant-libs/rmake.rs new file mode 100644 index 0000000000000..fb1b3bca8ade3 --- /dev/null +++ b/tests/run-make/redundant-libs/rmake.rs @@ -0,0 +1,34 @@ +// rustc will remove one of the two redundant references to foo below. Depending +// on which one gets removed, we'll get a linker error on SOME platforms (like +// Linux). On these platforms, when a library is referenced, the linker will +// only pull in the symbols needed _at that point in time_. If a later library +// depends on additional symbols from the library, they will not have been pulled +// in, and you'll get undefined symbols errors. +// +// So in this example, we need to ensure that rustc keeps the _later_ reference +// to foo, and not the former one. + +//@ ignore-cross-compile +// Reason: the compiled binary is executed +//@ ignore-windows-msvc +// Reason: this test links libraries via link.exe, which only accepts the import library +// for the dynamic library, i.e. `foo.dll.lib`. However, build_native_dynamic_lib only +// produces `foo.dll` - the dynamic library itself. To make this test work on MSVC, one +// would need to derive the import library from the dynamic library. +// See https://stackoverflow.com/questions/9360280/ + +use run_make_support::{ + build_native_dynamic_lib, build_native_static_lib, cwd, is_msvc, rfs, run, rustc, +}; + +fn main() { + build_native_dynamic_lib("foo"); + build_native_static_lib("bar"); + build_native_static_lib("baz"); + rustc() + .args(&["-lstatic=bar", "-lfoo", "-lstatic=baz", "-lfoo"]) + .input("main.rs") + .print("link-args") + .run(); + run("main"); +} From 0ac3d2e39a6566249526545fa506ef37fbc1839e Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Fri, 26 Jul 2024 00:05:20 -0400 Subject: [PATCH 12/28] Add implied target features to target_feature attribute --- compiler/rustc_codegen_llvm/src/llvm_util.rs | 16 ----- .../rustc_codegen_ssa/src/target_features.rs | 40 ++++++++++--- compiler/rustc_middle/src/query/mod.rs | 6 ++ compiler/rustc_target/src/target_features.rs | 58 +++++++++++++++++-- tests/ui/target-feature/implied-features.rs | 24 ++++++++ 5 files changed, 113 insertions(+), 31 deletions(-) create mode 100644 tests/ui/target-feature/implied-features.rs diff --git a/compiler/rustc_codegen_llvm/src/llvm_util.rs b/compiler/rustc_codegen_llvm/src/llvm_util.rs index af8a9be1ccbfd..dc21b92a95f76 100644 --- a/compiler/rustc_codegen_llvm/src/llvm_util.rs +++ b/compiler/rustc_codegen_llvm/src/llvm_util.rs @@ -646,22 +646,6 @@ pub(crate) fn global_llvm_features(sess: &Session, diagnostics: bool) -> Vec, which didn't make - // it into a released version of LLVM yet. - // - // This doesn't use the "implicit target feature" system because it is only - // used for function attributes in other targets, which fixes this bug as - // well on the function attribute level. - if sess.target.families.contains(&"wasm".into()) { - if features.iter().any(|f| f == "+relaxed-simd") - && !features.iter().any(|f| f == "+simd128") - { - features.push("+simd128".into()); - } - } - if diagnostics && let Some(f) = check_tied_features(sess, &featsmap) { sess.dcx().emit_err(TargetFeatureDisableOrEnable { features: f, diff --git a/compiler/rustc_codegen_ssa/src/target_features.rs b/compiler/rustc_codegen_ssa/src/target_features.rs index 127244a34f8f0..1bf842b53a3c5 100644 --- a/compiler/rustc_codegen_ssa/src/target_features.rs +++ b/compiler/rustc_codegen_ssa/src/target_features.rs @@ -1,7 +1,7 @@ use rustc_ast::ast; use rustc_attr::InstructionSetAttr; -use rustc_data_structures::fx::FxIndexSet; -use rustc_data_structures::unord::UnordMap; +use rustc_data_structures::fx::{FxHashMap, FxIndexSet}; +use rustc_data_structures::unord::{UnordMap, UnordSet}; use rustc_errors::Applicability; use rustc_hir::def::DefKind; use rustc_hir::def_id::{DefId, LocalDefId, LOCAL_CRATE}; @@ -30,6 +30,7 @@ pub fn from_target_feature( .emit(); }; let rust_features = tcx.features(); + let mut added_target_features = Vec::new(); for item in list { // Only `enable = ...` is accepted in the meta-item list. if !item.has_name(sym::enable) { @@ -44,7 +45,7 @@ pub fn from_target_feature( }; // We allow comma separation to enable multiple features. - target_features.extend(value.as_str().split(',').filter_map(|feature| { + added_target_features.extend(value.as_str().split(',').filter_map(|feature| { let Some(feature_gate) = supported_target_features.get(feature) else { let msg = format!("the feature named `{feature}` is not valid for this target"); let mut err = tcx.dcx().struct_span_err(item.span(), msg); @@ -98,13 +99,12 @@ pub fn from_target_feature( })); } - for (feature, requires) in tcx.sess.target.implicit_target_features() { - if target_features.iter().any(|f| f.as_str() == *feature) - && !target_features.iter().any(|f| f.as_str() == *requires) - { - target_features.push(Symbol::intern(requires)); - } + // Add implied features + for feature in added_target_features.iter() { + target_features + .extend(tcx.implied_target_features(*feature).clone().into_sorted_stable_ord()); } + target_features.extend(added_target_features) } /// Computes the set of target features used in a function for the purposes of @@ -162,6 +162,28 @@ pub(crate) fn provide(providers: &mut Providers) { .collect() } }, + implied_target_features: |tcx, feature| { + let implied_features = tcx + .sess + .target + .implied_target_features() + .iter() + .map(|(f, i)| (Symbol::intern(f), i)) + .collect::>(); + + // implied target features have their own implied target features, so we traverse the + // map until there are no more features to add + let mut features = UnordSet::new(); + let mut new_features = vec![feature]; + while let Some(new_feature) = new_features.pop() { + if features.insert(new_feature) { + if let Some(implied_features) = implied_features.get(&new_feature) { + new_features.extend(implied_features.iter().copied().map(Symbol::intern)) + } + } + } + features + }, asm_target_features, ..*providers } diff --git a/compiler/rustc_middle/src/query/mod.rs b/compiler/rustc_middle/src/query/mod.rs index c22c2e985abba..b6a2943265034 100644 --- a/compiler/rustc_middle/src/query/mod.rs +++ b/compiler/rustc_middle/src/query/mod.rs @@ -2183,6 +2183,12 @@ rustc_queries! { desc { "looking up supported target features" } } + query implied_target_features(feature: Symbol) -> &'tcx UnordSet { + arena_cache + eval_always + desc { "looking up implied target features" } + } + query features_query(_: ()) -> &'tcx rustc_feature::Features { feedable desc { "looking up enabled feature gates" } diff --git a/compiler/rustc_target/src/target_features.rs b/compiler/rustc_target/src/target_features.rs index 4e2617c467949..5b79495831a34 100644 --- a/compiler/rustc_target/src/target_features.rs +++ b/compiler/rustc_target/src/target_features.rs @@ -339,8 +339,6 @@ const WASM_ALLOWED_FEATURES: &[(&str, Stability)] = &[ // tidy-alphabetical-end ]; -const WASM_IMPLICIT_FEATURES: &[(&str, &str)] = &[("relaxed-simd", "simd128")]; - const BPF_ALLOWED_FEATURES: &[(&str, Stability)] = &[("alu32", Unstable(sym::bpf_target_feature))]; const CSKY_ALLOWED_FEATURES: &[(&str, Stability)] = &[ @@ -411,6 +409,54 @@ const IBMZ_ALLOWED_FEATURES: &[(&str, Stability)] = &[ // tidy-alphabetical-end ]; +const X86_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ + // tidy-alphabetical-start + ("aes", &["sse2"]), + ("avx", &["sse4.2"]), + ("avx2", &["avx"]), + ("f16c", &["avx"]), + ("fma", &["avx"]), + ("pclmulqdq", &["sse2"]), + ("sha", &["sse2"]), + ("sse2", &["sse"]), + ("sse3", &["sse2"]), + ("sse4.1", &["ssse3"]), + ("sse4.2", &["sse4.1"]), + ("ssse3", &["sse3"]), + // tidy-alphabetical-end +]; + +const AARCH64_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ + // tidy-alphabetical-start + ("aes", &["neon"]), + ("f32mm", &["sve"]), + ("f64mm", &["sve"]), + ("fcma", &["neon"]), + ("fhm", &["fp16"]), + ("fp16", &["neon"]), + ("jsconv", &["neon"]), + ("rcpc2", &["rcpc"]), + ("sha2", &["neon"]), + ("sha3", &["sha2"]), + ("sm4", &["neon"]), + ("sve", &["fp16"]), + ("sve2", &["sve"]), + ("sve2-aes", &["sve2", "aes"]), + ("sve2-bitperm", &["sve2"]), + ("sve2-sha3", &["sve2", "sha3"]), + ("sve2-sm4", &["sve2", "sm4"]), + // tidy-alphabetical-end +]; + +const RISCV_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ + // tidy-alphabetical-start + ("zb", &["zba", "zbc", "zbs"]), + ("zk", &["zkn", "zkr", "zks", "zkt", "zbkb", "zbkc", "zkbx"]), + ("zkn", &["zknd", "zkne", "zknh", "zbkb", "zbkc", "zkbx"]), + ("zks", &["zksed", "zksh", "zbkb", "zbkc", "zkbx"]), + // tidy-alphabetical-end +]; + /// When rustdoc is running, provide a list of all known features so that all their respective /// primitives may be documented. /// @@ -458,11 +504,11 @@ impl super::spec::Target { } } - /// Returns a list of target features. Each items first target feature - /// implicitly enables the second one. - pub fn implicit_target_features(&self) -> &'static [(&'static str, &'static str)] { + pub fn implied_target_features(&self) -> &'static [(&'static str, &'static [&'static str])] { match &*self.arch { - "wasm32" | "wasm64" => WASM_IMPLICIT_FEATURES, + "aarch4" => AARCH64_IMPLIED_FEATURES, + "riscv32" | "riscv64" => RISCV_IMPLIED_FEATURES, + "x86" | "x86_64" => X86_IMPLIED_FEATURES, _ => &[], } } diff --git a/tests/ui/target-feature/implied-features.rs b/tests/ui/target-feature/implied-features.rs new file mode 100644 index 0000000000000..c6d9ba78c21e6 --- /dev/null +++ b/tests/ui/target-feature/implied-features.rs @@ -0,0 +1,24 @@ +//@ only-x86_64 +//@ run-pass +#![feature(target_feature_11)] +#![allow(dead_code)] + +#[target_feature(enable = "ssse3")] +fn call_ssse3() {} + +#[target_feature(enable = "avx")] +fn call_avx() {} + +#[target_feature(enable = "avx2")] +fn test_avx2() { + call_ssse3(); + call_avx(); +} + +#[target_feature(enable = "fma")] +fn test_fma() { + call_ssse3(); + call_avx(); +} + +fn main() {} From a42ec982f78ac66e404c8b01248ae100e79e6192 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Fri, 26 Jul 2024 10:04:27 -0400 Subject: [PATCH 13/28] Add test to ensure implied target features work with asm, and fix failing tests --- .../rustc_codegen_ssa/src/target_features.rs | 12 ++-- .../const-eval/const_fn_target_feature.stderr | 2 +- .../rfc-2396-target_feature-11/safe-calls.rs | 7 +-- .../safe-calls.stderr | 60 +++++++------------ .../asm-implied-features-issue-128125.rs | 10 ++++ tests/ui/target-feature/implied-features.rs | 2 +- 6 files changed, 44 insertions(+), 49 deletions(-) create mode 100644 tests/ui/target-feature/asm-implied-features-issue-128125.rs diff --git a/compiler/rustc_codegen_ssa/src/target_features.rs b/compiler/rustc_codegen_ssa/src/target_features.rs index 1bf842b53a3c5..24b2c9c51c6e1 100644 --- a/compiler/rustc_codegen_ssa/src/target_features.rs +++ b/compiler/rustc_codegen_ssa/src/target_features.rs @@ -1,7 +1,7 @@ use rustc_ast::ast; use rustc_attr::InstructionSetAttr; use rustc_data_structures::fx::{FxHashMap, FxIndexSet}; -use rustc_data_structures::unord::{UnordMap, UnordSet}; +use rustc_data_structures::unord::{ExtendUnord, UnordMap, UnordSet}; use rustc_errors::Applicability; use rustc_hir::def::DefKind; use rustc_hir::def_id::{DefId, LocalDefId, LOCAL_CRATE}; @@ -99,12 +99,14 @@ pub fn from_target_feature( })); } - // Add implied features + // Add both explicit and implied target features, using a set to deduplicate + let mut target_features_set = UnordSet::new(); for feature in added_target_features.iter() { - target_features - .extend(tcx.implied_target_features(*feature).clone().into_sorted_stable_ord()); + target_features_set + .extend_unord(tcx.implied_target_features(*feature).clone().into_items()); } - target_features.extend(added_target_features) + target_features_set.extend(added_target_features); + target_features.extend(target_features_set.into_sorted_stable_ord()) } /// Computes the set of target features used in a function for the purposes of diff --git a/tests/ui/consts/const-eval/const_fn_target_feature.stderr b/tests/ui/consts/const-eval/const_fn_target_feature.stderr index d3a00b57ebb50..ad40d733546af 100644 --- a/tests/ui/consts/const-eval/const_fn_target_feature.stderr +++ b/tests/ui/consts/const-eval/const_fn_target_feature.stderr @@ -2,7 +2,7 @@ error[E0080]: evaluation of constant value failed --> $DIR/const_fn_target_feature.rs:11:24 | LL | const B: () = unsafe { avx2_fn() }; - | ^^^^^^^^^ calling a function that requires unavailable target features: avx2 + | ^^^^^^^^^ calling a function that requires unavailable target features: avx, avx2, sse4.1, sse4.2 error: aborting due to 1 previous error diff --git a/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.rs b/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.rs index de002ef71d7d3..fec4e75290fc8 100644 --- a/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.rs +++ b/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.rs @@ -34,6 +34,7 @@ fn foo() { #[target_feature(enable = "sse2")] fn bar() { + sse2(); avx_bmi2(); //~^ ERROR call to function `avx_bmi2` with `#[target_feature]` is unsafe Quux.avx_bmi2(); @@ -43,7 +44,6 @@ fn bar() { #[target_feature(enable = "avx")] fn baz() { sse2(); - //~^ ERROR call to function `sse2` with `#[target_feature]` is unsafe avx_bmi2(); //~^ ERROR call to function `avx_bmi2` with `#[target_feature]` is unsafe Quux.avx_bmi2(); @@ -54,7 +54,8 @@ fn baz() { #[target_feature(enable = "bmi2")] fn qux() { sse2(); - //~^ ERROR call to function `sse2` with `#[target_feature]` is unsafe + avx_bmi2(); + Quux.avx_bmi2(); } const _: () = sse2(); @@ -64,8 +65,6 @@ const _: () = sse2_and_fxsr(); //~^ ERROR call to function `sse2_and_fxsr` with `#[target_feature]` is unsafe #[deny(unsafe_op_in_unsafe_fn)] -#[target_feature(enable = "avx")] -#[target_feature(enable = "bmi2")] unsafe fn needs_unsafe_block() { sse2(); //~^ ERROR call to function `sse2` with `#[target_feature]` is unsafe diff --git a/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.stderr b/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.stderr index 537819ab8595c..c2227f8e84783 100644 --- a/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.stderr +++ b/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.stderr @@ -4,8 +4,8 @@ error[E0133]: call to function `sse2` with `#[target_feature]` is unsafe and req LL | sse2(); | ^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target feature: sse2 - = note: the sse2 target feature being enabled in the build configuration does not remove the requirement to list it in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target features: sse and sse2 + = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` error[E0133]: call to function `avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block --> $DIR/safe-calls.rs:29:5 @@ -13,7 +13,8 @@ error[E0133]: call to function `avx_bmi2` with `#[target_feature]` is unsafe and LL | avx_bmi2(); | ^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: avx and bmi2 + = help: in order for the call to be safe, the context requires the following additional target features: avx, sse, sse2, sse3, sse4.1, sse4.2, ssse3, and bmi2 + = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` error[E0133]: call to function `Quux::avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block --> $DIR/safe-calls.rs:31:5 @@ -21,32 +22,24 @@ error[E0133]: call to function `Quux::avx_bmi2` with `#[target_feature]` is unsa LL | Quux.avx_bmi2(); | ^^^^^^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: avx and bmi2 + = help: in order for the call to be safe, the context requires the following additional target features: avx, sse, sse2, sse3, sse4.1, sse4.2, ssse3, and bmi2 + = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` error[E0133]: call to function `avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block - --> $DIR/safe-calls.rs:37:5 + --> $DIR/safe-calls.rs:38:5 | LL | avx_bmi2(); | ^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: avx and bmi2 + = help: in order for the call to be safe, the context requires the following additional target features: avx, sse3, sse4.1, sse4.2, ssse3, and bmi2 error[E0133]: call to function `Quux::avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block - --> $DIR/safe-calls.rs:39:5 + --> $DIR/safe-calls.rs:40:5 | LL | Quux.avx_bmi2(); | ^^^^^^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: avx and bmi2 - -error[E0133]: call to function `sse2` with `#[target_feature]` is unsafe and requires unsafe function or block - --> $DIR/safe-calls.rs:45:5 - | -LL | sse2(); - | ^^^^^^ call to function with `#[target_feature]` - | - = help: in order for the call to be safe, the context requires the following additional target feature: sse2 - = note: the sse2 target feature being enabled in the build configuration does not remove the requirement to list it in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target features: avx, sse3, sse4.1, sse4.2, ssse3, and bmi2 error[E0133]: call to function `avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block --> $DIR/safe-calls.rs:47:5 @@ -65,52 +58,43 @@ LL | Quux.avx_bmi2(); = help: in order for the call to be safe, the context requires the following additional target feature: bmi2 error[E0133]: call to function `sse2` with `#[target_feature]` is unsafe and requires unsafe function or block - --> $DIR/safe-calls.rs:56:5 - | -LL | sse2(); - | ^^^^^^ call to function with `#[target_feature]` - | - = help: in order for the call to be safe, the context requires the following additional target feature: sse2 - = note: the sse2 target feature being enabled in the build configuration does not remove the requirement to list it in `#[target_feature]` - -error[E0133]: call to function `sse2` with `#[target_feature]` is unsafe and requires unsafe function or block - --> $DIR/safe-calls.rs:60:15 + --> $DIR/safe-calls.rs:61:15 | LL | const _: () = sse2(); | ^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target feature: sse2 - = note: the sse2 target feature being enabled in the build configuration does not remove the requirement to list it in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target features: sse and sse2 + = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` error[E0133]: call to function `sse2_and_fxsr` with `#[target_feature]` is unsafe and requires unsafe function or block - --> $DIR/safe-calls.rs:63:15 + --> $DIR/safe-calls.rs:64:15 | LL | const _: () = sse2_and_fxsr(); | ^^^^^^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: sse2 and fxsr - = note: the fxsr and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target features: sse, sse2, and fxsr + = note: the fxsr, sse, and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` error[E0133]: call to function `sse2` with `#[target_feature]` is unsafe and requires unsafe block - --> $DIR/safe-calls.rs:70:5 + --> $DIR/safe-calls.rs:69:5 | LL | sse2(); | ^^^^^^ call to function with `#[target_feature]` | = note: for more information, see issue #71668 - = help: in order for the call to be safe, the context requires the following additional target feature: sse2 - = note: the sse2 target feature being enabled in the build configuration does not remove the requirement to list it in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target features: sse and sse2 + = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` note: an unsafe function restricts its caller, but its body is safe by default - --> $DIR/safe-calls.rs:69:1 + --> $DIR/safe-calls.rs:68:1 | LL | unsafe fn needs_unsafe_block() { | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ note: the lint level is defined here - --> $DIR/safe-calls.rs:66:8 + --> $DIR/safe-calls.rs:67:8 | LL | #[deny(unsafe_op_in_unsafe_fn)] | ^^^^^^^^^^^^^^^^^^^^^^ -error: aborting due to 12 previous errors +error: aborting due to 10 previous errors For more information about this error, try `rustc --explain E0133`. diff --git a/tests/ui/target-feature/asm-implied-features-issue-128125.rs b/tests/ui/target-feature/asm-implied-features-issue-128125.rs new file mode 100644 index 0000000000000..2b4f1d7df8563 --- /dev/null +++ b/tests/ui/target-feature/asm-implied-features-issue-128125.rs @@ -0,0 +1,10 @@ +//@ only-x86_64 +//@ build-pass +#![allow(dead_code)] + +#[target_feature(enable = "avx2")] +unsafe fn demo(v: std::arch::x86_64::__m256i) { + std::arch::asm!("/* {v} */", v = in(ymm_reg) v); +} + +fn main() {} diff --git a/tests/ui/target-feature/implied-features.rs b/tests/ui/target-feature/implied-features.rs index c6d9ba78c21e6..4fdd843e6c289 100644 --- a/tests/ui/target-feature/implied-features.rs +++ b/tests/ui/target-feature/implied-features.rs @@ -1,5 +1,5 @@ //@ only-x86_64 -//@ run-pass +//@ build-pass #![feature(target_feature_11)] #![allow(dead_code)] From 82fd5e982994c309d9886a78374c789094a422ff Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Fri, 26 Jul 2024 11:27:21 -0400 Subject: [PATCH 14/28] Fix codegen tests --- tests/codegen/sse42-implies-crc32.rs | 2 +- tests/codegen/target-feature-overrides.rs | 8 ++++---- 2 files changed, 5 insertions(+), 5 deletions(-) diff --git a/tests/codegen/sse42-implies-crc32.rs b/tests/codegen/sse42-implies-crc32.rs index 94fcd77bc8842..8a9c496a3a541 100644 --- a/tests/codegen/sse42-implies-crc32.rs +++ b/tests/codegen/sse42-implies-crc32.rs @@ -12,4 +12,4 @@ pub unsafe fn crc32sse(v: u8) -> u32 { _mm_crc32_u8(out, v) } -// CHECK: attributes #0 {{.*"target-features"=".*\+sse4.2,\+crc32"}} +// CHECK: attributes #0 {{.*"target-features"=".*\+sse4.2,\+crc32.*"}} diff --git a/tests/codegen/target-feature-overrides.rs b/tests/codegen/target-feature-overrides.rs index 1e2c364dbbc9a..87783706d9535 100644 --- a/tests/codegen/target-feature-overrides.rs +++ b/tests/codegen/target-feature-overrides.rs @@ -1,7 +1,7 @@ //@ revisions: COMPAT INCOMPAT //@ needs-llvm-components: x86 //@ compile-flags: --target=x86_64-unknown-linux-gnu -Copt-level=3 -//@ [COMPAT] compile-flags: -Ctarget-feature=+avx2,+avx +//@ [COMPAT] compile-flags: -Ctarget-feature=+avx2,+avx,+sse4.2,+sse4.1,+ssse3,+sse3 //@ [INCOMPAT] compile-flags: -Ctarget-feature=-avx2,-avx // See also tests/assembly/target-feature-multiple.rs @@ -39,8 +39,8 @@ pub unsafe fn banana() -> u32 { } // CHECK: attributes [[APPLEATTRS]] -// COMPAT-SAME: "target-features"="+avx2,+avx,+avx" -// INCOMPAT-SAME: "target-features"="-avx2,-avx,+avx" +// COMPAT-SAME: "target-features"="+avx2,+avx,{{.*}}" +// INCOMPAT-SAME: "target-features"="-avx2,-avx,+avx,{{.*}}" // CHECK: attributes [[BANANAATTRS]] -// COMPAT-SAME: "target-features"="+avx2,+avx" +// COMPAT-SAME: "target-features"="+avx2,+avx,{{.*}}" // INCOMPAT-SAME: "target-features"="-avx2,-avx" From 5077f3372457c4b12a94992a2ab2bd89051d9a56 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Mon, 29 Jul 2024 00:09:47 -0400 Subject: [PATCH 15/28] Add missing features --- compiler/rustc_target/src/target_features.rs | 26 ++++++++++++++++++++ 1 file changed, 26 insertions(+) diff --git a/compiler/rustc_target/src/target_features.rs b/compiler/rustc_target/src/target_features.rs index 5b79495831a34..64f966d7a305e 100644 --- a/compiler/rustc_target/src/target_features.rs +++ b/compiler/rustc_target/src/target_features.rs @@ -414,8 +414,22 @@ const X86_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ ("aes", &["sse2"]), ("avx", &["sse4.2"]), ("avx2", &["avx"]), + ("avx512bf16", &["avx512bw"]), + ("avx512bitalg", &["avx512bw"]), + ("avx512bw", &["avx512f"]), + ("avx512cd", &["avx512f"]), + ("avx512dq", &["avx512f"]), + ("avx512f", &["avx2"]), + ("avx512fp16", &["avx512bw", "avx512vl", "avx512dq"]), + ("avx512vbmi", &["avx512bw"]), + ("avx512vbmi2", &["avx512bw"]), + ("avx512vl", &["avx512f"]), + ("avx512vnni", &["avx512f"]), + ("avx512vp2intersect", &["avx512f"]), + ("avx512vpopcntdq", &["avx512f"]), ("f16c", &["avx"]), ("fma", &["avx"]), + ("gfni", &["sse2"]), ("pclmulqdq", &["sse2"]), ("sha", &["sse2"]), ("sse2", &["sse"]), @@ -423,6 +437,11 @@ const X86_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ ("sse4.1", &["ssse3"]), ("sse4.2", &["sse4.1"]), ("ssse3", &["sse3"]), + ("vaes", &["avx", "aes"]), + ("vpclmulqdq", &["avx", "pclmulqdq"]), + ("xsavec", &["xsave"]), + ("xsaveopt", &["xsave"]), + ("xsaves", &["xsave"]), // tidy-alphabetical-end ]; @@ -457,6 +476,12 @@ const RISCV_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ // tidy-alphabetical-end ]; +const WASM_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ + // tidy-alphabetical-start + ("relaxed-simd", &["simd128"]), + // tidy-alphabetical-end +]; + /// When rustdoc is running, provide a list of all known features so that all their respective /// primitives may be documented. /// @@ -509,6 +534,7 @@ impl super::spec::Target { "aarch4" => AARCH64_IMPLIED_FEATURES, "riscv32" | "riscv64" => RISCV_IMPLIED_FEATURES, "x86" | "x86_64" => X86_IMPLIED_FEATURES, + "wasm32" | "wasm64" => WASM_IMPLIED_FEATURES, _ => &[], } } From a3ffd8dd80a8f47b5c96dec1c0ac23923c0e0295 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Tue, 30 Jul 2024 23:58:51 -0400 Subject: [PATCH 16/28] Add +sse4.2 due to #128426 --- src/tools/miri/tests/pass/shims/x86/intrinsics-x86-aes-vaes.rs | 2 +- src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx.rs | 2 +- src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx2.rs | 2 +- src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx512.rs | 2 +- 4 files changed, 4 insertions(+), 4 deletions(-) diff --git a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-aes-vaes.rs b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-aes-vaes.rs index 7363c75361779..13ebe56146381 100644 --- a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-aes-vaes.rs +++ b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-aes-vaes.rs @@ -7,7 +7,7 @@ //@ignore-target-s390x //@ignore-target-thumbv7em //@ignore-target-wasm32 -//@compile-flags: -C target-feature=+aes,+vaes,+avx512f +//@compile-flags: -C target-feature=+aes,+vaes,+avx512f,+sse4.2 #![feature(avx512_target_feature, stdarch_x86_avx512)] diff --git a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx.rs b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx.rs index 728f57d48f17e..18693a2a97d11 100644 --- a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx.rs +++ b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx.rs @@ -7,7 +7,7 @@ //@ignore-target-s390x //@ignore-target-thumbv7em //@ignore-target-wasm32 -//@compile-flags: -C target-feature=+avx +//@compile-flags: -C target-feature=+avx,+sse4.2 #[cfg(target_arch = "x86")] use std::arch::x86::*; diff --git a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx2.rs b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx2.rs index 80d125bb85650..f42ff6dfb51b2 100644 --- a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx2.rs +++ b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx2.rs @@ -7,7 +7,7 @@ //@ignore-target-s390x //@ignore-target-thumbv7em //@ignore-target-wasm32 -//@compile-flags: -C target-feature=+avx2 +//@compile-flags: -C target-feature=+avx2,+sse4.2 #[cfg(target_arch = "x86")] use std::arch::x86::*; diff --git a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx512.rs b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx512.rs index 66bfcb20f1c99..7d25854ccc075 100644 --- a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx512.rs +++ b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx512.rs @@ -7,7 +7,7 @@ //@ignore-target-s390x //@ignore-target-thumbv7em //@ignore-target-wasm32 -//@compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bitalg,+avx512vpopcntdq +//@compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bitalg,+avx512vpopcntdq,+sse4.2 #![feature(avx512_target_feature)] #![feature(stdarch_x86_avx512)] From d00d42f60e9f1996415da257dfb30ecd8e8c2073 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Wed, 31 Jul 2024 01:50:02 -0400 Subject: [PATCH 17/28] Bless tests --- .../fail/function_calls/simd_feature_flag_difference.stderr | 4 ++-- src/tools/miri/tests/fail/function_calls/target_feature.rs | 2 +- .../miri/tests/fail/function_calls/target_feature.stderr | 4 ++-- 3 files changed, 5 insertions(+), 5 deletions(-) diff --git a/src/tools/miri/tests/fail/function_calls/simd_feature_flag_difference.stderr b/src/tools/miri/tests/fail/function_calls/simd_feature_flag_difference.stderr index 2544421c7e8c7..8ebd5d2726ffd 100644 --- a/src/tools/miri/tests/fail/function_calls/simd_feature_flag_difference.stderr +++ b/src/tools/miri/tests/fail/function_calls/simd_feature_flag_difference.stderr @@ -1,8 +1,8 @@ -error: Undefined Behavior: calling a function that requires unavailable target features: avx +error: Undefined Behavior: calling a function that requires unavailable target features: avx, sse3, sse4.1, sse4.2, ssse3 --> $DIR/simd_feature_flag_difference.rs:LL:CC | LL | unsafe { foo(0.0, x) } - | ^^^^^^^^^^^ calling a function that requires unavailable target features: avx + | ^^^^^^^^^^^ calling a function that requires unavailable target features: avx, sse3, sse4.1, sse4.2, ssse3 | = help: this indicates a bug in the program: it performed an invalid operation, and caused Undefined Behavior = help: see https://doc.rust-lang.org/nightly/reference/behavior-considered-undefined.html for further information diff --git a/src/tools/miri/tests/fail/function_calls/target_feature.rs b/src/tools/miri/tests/fail/function_calls/target_feature.rs index 84e01eb4803fe..713d7c16f4c25 100644 --- a/src/tools/miri/tests/fail/function_calls/target_feature.rs +++ b/src/tools/miri/tests/fail/function_calls/target_feature.rs @@ -4,7 +4,7 @@ fn main() { assert!(!is_x86_feature_detected!("ssse3")); unsafe { - ssse3_fn(); //~ ERROR: calling a function that requires unavailable target features: ssse3 + ssse3_fn(); //~ ERROR: calling a function that requires unavailable target features: sse3, ssse3 } } diff --git a/src/tools/miri/tests/fail/function_calls/target_feature.stderr b/src/tools/miri/tests/fail/function_calls/target_feature.stderr index 4d3cf6e9d3bc0..74a862c7b681c 100644 --- a/src/tools/miri/tests/fail/function_calls/target_feature.stderr +++ b/src/tools/miri/tests/fail/function_calls/target_feature.stderr @@ -1,8 +1,8 @@ -error: Undefined Behavior: calling a function that requires unavailable target features: ssse3 +error: Undefined Behavior: calling a function that requires unavailable target features: sse3, ssse3 --> $DIR/target_feature.rs:LL:CC | LL | ssse3_fn(); - | ^^^^^^^^^^ calling a function that requires unavailable target features: ssse3 + | ^^^^^^^^^^ calling a function that requires unavailable target features: sse3, ssse3 | = help: this indicates a bug in the program: it performed an invalid operation, and caused Undefined Behavior = help: see https://doc.rust-lang.org/nightly/reference/behavior-considered-undefined.html for further information From 0cdc5088ac37da875270a5cf8c85c36c6fae01a1 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Fri, 2 Aug 2024 00:20:49 -0400 Subject: [PATCH 18/28] Refactor and fill out target feature lists --- compiler/rustc_codegen_gcc/src/gcc_util.rs | 4 +- compiler/rustc_codegen_gcc/src/lib.rs | 2 +- compiler/rustc_codegen_llvm/src/llvm_util.rs | 28 +- .../rustc_codegen_ssa/src/target_features.rs | 6 +- compiler/rustc_target/src/target_features.rs | 613 ++++++++---------- 5 files changed, 293 insertions(+), 360 deletions(-) diff --git a/compiler/rustc_codegen_gcc/src/gcc_util.rs b/compiler/rustc_codegen_gcc/src/gcc_util.rs index 8bb90efe6fb7c..5308ccdb61469 100644 --- a/compiler/rustc_codegen_gcc/src/gcc_util.rs +++ b/compiler/rustc_codegen_gcc/src/gcc_util.rs @@ -65,8 +65,8 @@ pub(crate) fn global_gcc_features(sess: &Session, diagnostics: bool) -> Vec Vec { sess.target .supported_target_features() .iter() - .filter_map(|&(feature, gate)| { + .filter_map(|&(feature, gate, _)| { if sess.is_nightly_build() || allow_unstable || gate.is_stable() { Some(feature) } else { @@ -386,7 +386,7 @@ fn print_target_features(out: &mut String, sess: &Session, tm: &llvm::TargetMach .target .supported_target_features() .iter() - .map(|(feature, _gate)| { + .map(|(feature, _gate, _implied)| { // LLVM asserts that these are sorted. LLVM and Rust both use byte comparison for these strings. let llvm_feature = to_llvm_features(sess, *feature).llvm_feature_name; let desc = @@ -571,17 +571,19 @@ pub(crate) fn global_llvm_features(sess: &Session, diagnostics: bool) -> Vec Vec>(); // implied target features have their own implied target features, so we traverse the diff --git a/compiler/rustc_target/src/target_features.rs b/compiler/rustc_target/src/target_features.rs index 64f966d7a305e..f408db1dcbad2 100644 --- a/compiler/rustc_target/src/target_features.rs +++ b/compiler/rustc_target/src/target_features.rs @@ -53,136 +53,146 @@ impl Stability { // // Stabilizing a target feature requires t-lang approval. -const ARM_ALLOWED_FEATURES: &[(&str, Stability)] = &[ +type ImpliedFeatures = &'static [&'static str]; + +const ARM_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("aclass", Unstable(sym::arm_target_feature)), - ("aes", Unstable(sym::arm_target_feature)), - ("crc", Unstable(sym::arm_target_feature)), - ("d32", Unstable(sym::arm_target_feature)), - ("dotprod", Unstable(sym::arm_target_feature)), - ("dsp", Unstable(sym::arm_target_feature)), - ("fp-armv8", Unstable(sym::arm_target_feature)), - ("i8mm", Unstable(sym::arm_target_feature)), - ("mclass", Unstable(sym::arm_target_feature)), - ("neon", Unstable(sym::arm_target_feature)), - ("rclass", Unstable(sym::arm_target_feature)), - ("sha2", Unstable(sym::arm_target_feature)), + ("aclass", Unstable(sym::arm_target_feature), &[]), + ("aes", Unstable(sym::arm_target_feature), &["neon"]), + ("crc", Unstable(sym::arm_target_feature), &[]), + ("d32", Unstable(sym::arm_target_feature), &[]), + ("dotprod", Unstable(sym::arm_target_feature), &["neon"]), + ("dsp", Unstable(sym::arm_target_feature), &[]), + ("fp-armv8", Unstable(sym::arm_target_feature), &["vfp4"]), + ("i8mm", Unstable(sym::arm_target_feature), &["neon"]), + ("mclass", Unstable(sym::arm_target_feature), &[]), + ("neon", Unstable(sym::arm_target_feature), &["vfp3"]), + ("rclass", Unstable(sym::arm_target_feature), &[]), + ("sha2", Unstable(sym::arm_target_feature), &["neon"]), // This is needed for inline assembly, but shouldn't be stabilized as-is // since it should be enabled per-function using #[instruction_set], not // #[target_feature]. - ("thumb-mode", Unstable(sym::arm_target_feature)), - ("thumb2", Unstable(sym::arm_target_feature)), - ("trustzone", Unstable(sym::arm_target_feature)), - ("v5te", Unstable(sym::arm_target_feature)), - ("v6", Unstable(sym::arm_target_feature)), - ("v6k", Unstable(sym::arm_target_feature)), - ("v6t2", Unstable(sym::arm_target_feature)), - ("v7", Unstable(sym::arm_target_feature)), - ("v8", Unstable(sym::arm_target_feature)), - ("vfp2", Unstable(sym::arm_target_feature)), - ("vfp3", Unstable(sym::arm_target_feature)), - ("vfp4", Unstable(sym::arm_target_feature)), - ("virtualization", Unstable(sym::arm_target_feature)), + ("thumb-mode", Unstable(sym::arm_target_feature), &[]), + ("thumb2", Unstable(sym::arm_target_feature), &[]), + ("trustzone", Unstable(sym::arm_target_feature), &[]), + ("v5te", Unstable(sym::arm_target_feature), &[]), + ("v6", Unstable(sym::arm_target_feature), &["v5te"]), + ("v6k", Unstable(sym::arm_target_feature), &["v6"]), + ("v6t2", Unstable(sym::arm_target_feature), &["v6k", "thumb2"]), + ("v7", Unstable(sym::arm_target_feature), &["v6t2"]), + ("v8", Unstable(sym::arm_target_feature), &["v7"]), + ("vfp2", Unstable(sym::arm_target_feature), &[]), + ("vfp3", Unstable(sym::arm_target_feature), &["vfp2", "d32"]), + ("vfp4", Unstable(sym::arm_target_feature), &["vfp3"]), + ("virtualization", Unstable(sym::arm_target_feature), &[]), // tidy-alphabetical-end ]; -const AARCH64_ALLOWED_FEATURES: &[(&str, Stability)] = &[ +const AARCH64_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start // FEAT_AES & FEAT_PMULL - ("aes", Stable), + ("aes", Stable, &["neon"]), // FEAT_BF16 - ("bf16", Stable), + ("bf16", Stable, &[]), // FEAT_BTI - ("bti", Stable), + ("bti", Stable, &[]), // FEAT_CRC - ("crc", Stable), + ("crc", Stable, &[]), // FEAT_DIT - ("dit", Stable), + ("dit", Stable, &[]), // FEAT_DotProd - ("dotprod", Stable), + ("dotprod", Stable, &["neon"]), // FEAT_DPB - ("dpb", Stable), + ("dpb", Stable, &[]), // FEAT_DPB2 - ("dpb2", Stable), + ("dpb2", Stable, &["dpb"]), // FEAT_F32MM - ("f32mm", Stable), + ("f32mm", Stable, &["sve"]), // FEAT_F64MM - ("f64mm", Stable), + ("f64mm", Stable, &["sve"]), // FEAT_FCMA - ("fcma", Stable), + ("fcma", Stable, &["neon"]), // FEAT_FHM - ("fhm", Stable), + ("fhm", Stable, &["fp16"]), // FEAT_FLAGM - ("flagm", Stable), + ("flagm", Stable, &[]), // FEAT_FP16 - ("fp16", Stable), + ("fp16", Stable, &[]), // FEAT_FRINTTS - ("frintts", Stable), + ("frintts", Stable, &[]), // FEAT_I8MM - ("i8mm", Stable), + ("i8mm", Stable, &[]), // FEAT_JSCVT - ("jsconv", Stable), + ("jsconv", Stable, &[]), // FEAT_LOR - ("lor", Stable), + ("lor", Stable, &[]), // FEAT_LSE - ("lse", Stable), + ("lse", Stable, &[]), // FEAT_MTE & FEAT_MTE2 - ("mte", Stable), + ("mte", Stable, &[]), // FEAT_AdvSimd & FEAT_FP - ("neon", Stable), + ("neon", Stable, &[]), // FEAT_PAUTH (address authentication) - ("paca", Stable), + ("paca", Stable, &[]), // FEAT_PAUTH (generic authentication) - ("pacg", Stable), + ("pacg", Stable, &[]), // FEAT_PAN - ("pan", Stable), + ("pan", Stable, &[]), // FEAT_PMUv3 - ("pmuv3", Stable), + ("pmuv3", Stable, &[]), // FEAT_RAND - ("rand", Stable), + ("rand", Stable, &[]), // FEAT_RAS & FEAT_RASv1p1 - ("ras", Stable), + ("ras", Stable, &[]), // FEAT_RCPC - ("rcpc", Stable), + ("rcpc", Stable, &[]), // FEAT_RCPC2 - ("rcpc2", Stable), + ("rcpc2", Stable, &["rcpc"]), // FEAT_RDM - ("rdm", Stable), + ("rdm", Stable, &["neon"]), // FEAT_SB - ("sb", Stable), + ("sb", Stable, &[]), // FEAT_SHA1 & FEAT_SHA256 - ("sha2", Stable), + ("sha2", Stable, &["neon"]), // FEAT_SHA512 & FEAT_SHA3 - ("sha3", Stable), + ("sha3", Stable, &["sha2"]), // FEAT_SM3 & FEAT_SM4 - ("sm4", Stable), + ("sm4", Stable, &["neon"]), // FEAT_SPE - ("spe", Stable), + ("spe", Stable, &[]), // FEAT_SSBS & FEAT_SSBS2 - ("ssbs", Stable), + ("ssbs", Stable, &[]), // FEAT_SVE - ("sve", Stable), + ("sve", Stable, &[]), // FEAT_SVE2 - ("sve2", Stable), + ("sve2", Stable, &["sve"]), // FEAT_SVE2_AES - ("sve2-aes", Stable), + ("sve2-aes", Stable, &["sve2", "aes"]), // FEAT_SVE2_BitPerm - ("sve2-bitperm", Stable), + ("sve2-bitperm", Stable, &["sve2"]), // FEAT_SVE2_SHA3 - ("sve2-sha3", Stable), + ("sve2-sha3", Stable, &["sve2", "sha3"]), // FEAT_SVE2_SM4 - ("sve2-sm4", Stable), + ("sve2-sm4", Stable, &["sve2", "sm4"]), // FEAT_TME - ("tme", Stable), - ("v8.1a", Unstable(sym::aarch64_ver_target_feature)), - ("v8.2a", Unstable(sym::aarch64_ver_target_feature)), - ("v8.3a", Unstable(sym::aarch64_ver_target_feature)), - ("v8.4a", Unstable(sym::aarch64_ver_target_feature)), - ("v8.5a", Unstable(sym::aarch64_ver_target_feature)), - ("v8.6a", Unstable(sym::aarch64_ver_target_feature)), - ("v8.7a", Unstable(sym::aarch64_ver_target_feature)), + ("tme", Stable, &[]), + ( + "v8.1a", + Unstable(sym::aarch64_ver_target_feature), + &["crc", "lse", "rdm", "pan", "lor", "vh"], + ), + ("v8.2a", Unstable(sym::aarch64_ver_target_feature), &["v8.1a", "ras", "dpb"]), + ( + "v8.3a", + Unstable(sym::aarch64_ver_target_feature), + &["v8.2a", "rcpc", "paca", "pacg", "jsconv"], + ), + ("v8.4a", Unstable(sym::aarch64_ver_target_feature), &["v8.3a", "dotprod", "dit", "flagm"]), + ("v8.5a", Unstable(sym::aarch64_ver_target_feature), &["v8.4a", "ssbs", "sb", "dpb2", "bti"]), + ("v8.6a", Unstable(sym::aarch64_ver_target_feature), &["v8.5a", "bf16", "i8mm"]), + ("v8.7a", Unstable(sym::aarch64_ver_target_feature), &[]), // FEAT_VHE - ("vh", Stable), + ("vh", Stable, &[]), // tidy-alphabetical-end ]; @@ -190,295 +200,223 @@ const AARCH64_TIED_FEATURES: &[&[&str]] = &[ &["paca", "pacg"], // Together these represent `pauth` in LLVM ]; -const X86_ALLOWED_FEATURES: &[(&str, Stability)] = &[ +const X86_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("adx", Stable), - ("aes", Stable), - ("amx-bf16", Unstable(sym::x86_amx_intrinsics)), - ("amx-complex", Unstable(sym::x86_amx_intrinsics)), - ("amx-fp16", Unstable(sym::x86_amx_intrinsics)), - ("amx-int8", Unstable(sym::x86_amx_intrinsics)), - ("amx-tile", Unstable(sym::x86_amx_intrinsics)), - ("avx", Stable), - ("avx2", Stable), - ("avx512bf16", Unstable(sym::avx512_target_feature)), - ("avx512bitalg", Unstable(sym::avx512_target_feature)), - ("avx512bw", Unstable(sym::avx512_target_feature)), - ("avx512cd", Unstable(sym::avx512_target_feature)), - ("avx512dq", Unstable(sym::avx512_target_feature)), - ("avx512f", Unstable(sym::avx512_target_feature)), - ("avx512fp16", Unstable(sym::avx512_target_feature)), - ("avx512ifma", Unstable(sym::avx512_target_feature)), - ("avx512vbmi", Unstable(sym::avx512_target_feature)), - ("avx512vbmi2", Unstable(sym::avx512_target_feature)), - ("avx512vl", Unstable(sym::avx512_target_feature)), - ("avx512vnni", Unstable(sym::avx512_target_feature)), - ("avx512vp2intersect", Unstable(sym::avx512_target_feature)), - ("avx512vpopcntdq", Unstable(sym::avx512_target_feature)), - ("avxifma", Unstable(sym::avx512_target_feature)), - ("avxneconvert", Unstable(sym::avx512_target_feature)), - ("avxvnni", Unstable(sym::avx512_target_feature)), - ("avxvnniint16", Unstable(sym::avx512_target_feature)), - ("avxvnniint8", Unstable(sym::avx512_target_feature)), - ("bmi1", Stable), - ("bmi2", Stable), - ("cmpxchg16b", Stable), - ("ermsb", Unstable(sym::ermsb_target_feature)), - ("f16c", Stable), - ("fma", Stable), - ("fxsr", Stable), - ("gfni", Unstable(sym::avx512_target_feature)), - ("lahfsahf", Unstable(sym::lahfsahf_target_feature)), - ("lzcnt", Stable), - ("movbe", Stable), - ("pclmulqdq", Stable), - ("popcnt", Stable), - ("prfchw", Unstable(sym::prfchw_target_feature)), - ("rdrand", Stable), - ("rdseed", Stable), - ("rtm", Unstable(sym::rtm_target_feature)), - ("sha", Stable), - ("sha512", Unstable(sym::sha512_sm_x86)), - ("sm3", Unstable(sym::sha512_sm_x86)), - ("sm4", Unstable(sym::sha512_sm_x86)), - ("sse", Stable), - ("sse2", Stable), - ("sse3", Stable), - ("sse4.1", Stable), - ("sse4.2", Stable), - ("sse4a", Unstable(sym::sse4a_target_feature)), - ("ssse3", Stable), - ("tbm", Unstable(sym::tbm_target_feature)), - ("vaes", Unstable(sym::avx512_target_feature)), - ("vpclmulqdq", Unstable(sym::avx512_target_feature)), - ("xop", Unstable(sym::xop_target_feature)), - ("xsave", Stable), - ("xsavec", Stable), - ("xsaveopt", Stable), - ("xsaves", Stable), + ("adx", Stable, &[]), + ("aes", Stable, &["sse2"]), + ("amx-bf16", Unstable(sym::x86_amx_intrinsics), &["amx-tile"]), + ("amx-complex", Unstable(sym::x86_amx_intrinsics), &["amx-tile"]), + ("amx-fp16", Unstable(sym::x86_amx_intrinsics), &["amx-tile"]), + ("amx-int8", Unstable(sym::x86_amx_intrinsics), &["amx-tile"]), + ("amx-tile", Unstable(sym::x86_amx_intrinsics), &[]), + ("avx", Stable, &["sse4.2"]), + ("avx2", Stable, &["avx"]), + ("avx512bf16", Unstable(sym::avx512_target_feature), &["avx512bw"]), + ("avx512bitalg", Unstable(sym::avx512_target_feature), &["avx512bw"]), + ("avx512bw", Unstable(sym::avx512_target_feature), &["avx512f"]), + ("avx512cd", Unstable(sym::avx512_target_feature), &["avx512f"]), + ("avx512dq", Unstable(sym::avx512_target_feature), &["avx512f"]), + ("avx512f", Unstable(sym::avx512_target_feature), &["avx2", "fma", "f16c"]), + ("avx512fp16", Unstable(sym::avx512_target_feature), &["avx512bw", "avx512vl", "avx512dq"]), + ("avx512ifma", Unstable(sym::avx512_target_feature), &["avx512f"]), + ("avx512vbmi", Unstable(sym::avx512_target_feature), &["avx512bw"]), + ("avx512vbmi2", Unstable(sym::avx512_target_feature), &["avx512bw"]), + ("avx512vl", Unstable(sym::avx512_target_feature), &["avx512f"]), + ("avx512vnni", Unstable(sym::avx512_target_feature), &["avx512f"]), + ("avx512vp2intersect", Unstable(sym::avx512_target_feature), &["avx512f"]), + ("avx512vpopcntdq", Unstable(sym::avx512_target_feature), &["avx512f"]), + ("avxifma", Unstable(sym::avx512_target_feature), &["avx2"]), + ("avxneconvert", Unstable(sym::avx512_target_feature), &["avx2"]), + ("avxvnni", Unstable(sym::avx512_target_feature), &["avx2"]), + ("avxvnniint16", Unstable(sym::avx512_target_feature), &["avx2"]), + ("avxvnniint8", Unstable(sym::avx512_target_feature), &["avx2"]), + ("bmi1", Stable, &[]), + ("bmi2", Stable, &[]), + ("cmpxchg16b", Stable, &[]), + ("ermsb", Unstable(sym::ermsb_target_feature), &[]), + ("f16c", Stable, &["avx"]), + ("fma", Stable, &["avx"]), + ("fxsr", Stable, &[]), + ("gfni", Unstable(sym::avx512_target_feature), &["sse2"]), + ("lahfsahf", Unstable(sym::lahfsahf_target_feature), &[]), + ("lzcnt", Stable, &[]), + ("movbe", Stable, &[]), + ("pclmulqdq", Stable, &[]), + ("popcnt", Stable, &[]), + ("prfchw", Unstable(sym::prfchw_target_feature), &[]), + ("rdrand", Stable, &[]), + ("rdseed", Stable, &[]), + ("rtm", Unstable(sym::rtm_target_feature), &[]), + ("sha", Stable, &["sse2"]), + ("sha512", Unstable(sym::sha512_sm_x86), &["avx2"]), + ("sm3", Unstable(sym::sha512_sm_x86), &["avx"]), + ("sm4", Unstable(sym::sha512_sm_x86), &["avx2"]), + ("sse", Stable, &[]), + ("sse2", Stable, &["sse"]), + ("sse3", Stable, &["sse2"]), + ("sse4.1", Stable, &["ssse3"]), + ("sse4.2", Stable, &["sse4.1"]), + ("sse4a", Unstable(sym::sse4a_target_feature), &["sse3"]), + ("ssse3", Stable, &["sse3"]), + ("tbm", Unstable(sym::tbm_target_feature), &[]), + ("vaes", Unstable(sym::avx512_target_feature), &["avx2", "aes"]), + ("vpclmulqdq", Unstable(sym::avx512_target_feature), &["avx", "pclmulqdq"]), + ("xop", Unstable(sym::xop_target_feature), &[/*"fma4", */ "avx", "sse4a"]), + ("xsave", Stable, &[]), + ("xsavec", Stable, &["xsave"]), + ("xsaveopt", Stable, &["xsave"]), + ("xsaves", Stable, &["xsave"]), // tidy-alphabetical-end ]; -const HEXAGON_ALLOWED_FEATURES: &[(&str, Stability)] = &[ +const HEXAGON_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("hvx", Unstable(sym::hexagon_target_feature)), - ("hvx-length128b", Unstable(sym::hexagon_target_feature)), + ("hvx", Unstable(sym::hexagon_target_feature), &[]), + ("hvx-length128b", Unstable(sym::hexagon_target_feature), &["hvx"]), // tidy-alphabetical-end ]; -const POWERPC_ALLOWED_FEATURES: &[(&str, Stability)] = &[ +const POWERPC_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("altivec", Unstable(sym::powerpc_target_feature)), - ("power10-vector", Unstable(sym::powerpc_target_feature)), - ("power8-altivec", Unstable(sym::powerpc_target_feature)), - ("power8-vector", Unstable(sym::powerpc_target_feature)), - ("power9-altivec", Unstable(sym::powerpc_target_feature)), - ("power9-vector", Unstable(sym::powerpc_target_feature)), - ("vsx", Unstable(sym::powerpc_target_feature)), + ("altivec", Unstable(sym::powerpc_target_feature), &[]), + ("power10-vector", Unstable(sym::powerpc_target_feature), &["power9-vector"]), + ("power8-altivec", Unstable(sym::powerpc_target_feature), &["altivec"]), + ("power8-vector", Unstable(sym::powerpc_target_feature), &["vsx", "power8-altivec"]), + ("power9-altivec", Unstable(sym::powerpc_target_feature), &["power8-altivec"]), + ("power9-vector", Unstable(sym::powerpc_target_feature), &["power8-vector", "power9-altivec"]), + ("vsx", Unstable(sym::powerpc_target_feature), &["altivec"]), // tidy-alphabetical-end ]; -const MIPS_ALLOWED_FEATURES: &[(&str, Stability)] = &[ +const MIPS_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("fp64", Unstable(sym::mips_target_feature)), - ("msa", Unstable(sym::mips_target_feature)), - ("virt", Unstable(sym::mips_target_feature)), + ("fp64", Unstable(sym::mips_target_feature), &[]), + ("msa", Unstable(sym::mips_target_feature), &[]), + ("virt", Unstable(sym::mips_target_feature), &[]), // tidy-alphabetical-end ]; -const RISCV_ALLOWED_FEATURES: &[(&str, Stability)] = &[ +const RISCV_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("a", Stable), - ("c", Stable), - ("d", Unstable(sym::riscv_target_feature)), - ("e", Unstable(sym::riscv_target_feature)), - ("f", Unstable(sym::riscv_target_feature)), - ("m", Stable), - ("relax", Unstable(sym::riscv_target_feature)), - ("unaligned-scalar-mem", Unstable(sym::riscv_target_feature)), - ("v", Unstable(sym::riscv_target_feature)), - ("zba", Stable), - ("zbb", Stable), - ("zbc", Stable), - ("zbkb", Stable), - ("zbkc", Stable), - ("zbkx", Stable), - ("zbs", Stable), - ("zdinx", Unstable(sym::riscv_target_feature)), - ("zfh", Unstable(sym::riscv_target_feature)), - ("zfhmin", Unstable(sym::riscv_target_feature)), - ("zfinx", Unstable(sym::riscv_target_feature)), - ("zhinx", Unstable(sym::riscv_target_feature)), - ("zhinxmin", Unstable(sym::riscv_target_feature)), - ("zk", Stable), - ("zkn", Stable), - ("zknd", Stable), - ("zkne", Stable), - ("zknh", Stable), - ("zkr", Stable), - ("zks", Stable), - ("zksed", Stable), - ("zksh", Stable), - ("zkt", Stable), + ("a", Stable, &[]), + ("c", Stable, &[]), + ("d", Unstable(sym::riscv_target_feature), &["f"]), + ("e", Unstable(sym::riscv_target_feature), &[]), + ("f", Unstable(sym::riscv_target_feature), &[]), + ("m", Stable, &[]), + ("relax", Unstable(sym::riscv_target_feature), &[]), + ("unaligned-scalar-mem", Unstable(sym::riscv_target_feature), &[]), + ("v", Unstable(sym::riscv_target_feature), &[]), + ("zba", Stable, &[]), + ("zbb", Stable, &[]), + ("zbc", Stable, &[]), + ("zbkb", Stable, &[]), + ("zbkc", Stable, &[]), + ("zbkx", Stable, &[]), + ("zbs", Stable, &[]), + ("zdinx", Unstable(sym::riscv_target_feature), &["zfinx"]), + ("zfh", Unstable(sym::riscv_target_feature), &["zfhmin"]), + ("zfhmin", Unstable(sym::riscv_target_feature), &["f"]), + ("zfinx", Unstable(sym::riscv_target_feature), &[]), + ("zhinx", Unstable(sym::riscv_target_feature), &["zhinxmin"]), + ("zhinxmin", Unstable(sym::riscv_target_feature), &["zfinx"]), + ("zk", Stable, &["zkn", "zkr", "zkt"]), + ("zkn", Stable, &["zbkb", "zbkc", "zbkx", "zkne", "zknd", "zknh"]), + ("zknd", Stable, &[]), + ("zkne", Stable, &[]), + ("zknh", Stable, &[]), + ("zkr", Stable, &[]), + ("zks", Stable, &["zbkb", "bzkc", "zbkx", "zksed", "zksh"]), + ("zksed", Stable, &[]), + ("zksh", Stable, &[]), + ("zkt", Stable, &[]), // tidy-alphabetical-end ]; -const WASM_ALLOWED_FEATURES: &[(&str, Stability)] = &[ +const WASM_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("atomics", Unstable(sym::wasm_target_feature)), - ("bulk-memory", Stable), - ("exception-handling", Unstable(sym::wasm_target_feature)), - ("extended-const", Stable), - ("multivalue", Unstable(sym::wasm_target_feature)), - ("mutable-globals", Stable), - ("nontrapping-fptoint", Stable), - ("reference-types", Unstable(sym::wasm_target_feature)), - ("relaxed-simd", Stable), - ("sign-ext", Stable), - ("simd128", Stable), + ("atomics", Unstable(sym::wasm_target_feature), &[]), + ("bulk-memory", Stable, &[]), + ("exception-handling", Unstable(sym::wasm_target_feature), &[]), + ("extended-const", Stable, &[]), + ("multivalue", Unstable(sym::wasm_target_feature), &[]), + ("mutable-globals", Stable, &[]), + ("nontrapping-fptoint", Stable, &[]), + ("reference-types", Unstable(sym::wasm_target_feature), &[]), + ("relaxed-simd", Stable, &["simd128"]), + ("sign-ext", Stable, &[]), + ("simd128", Stable, &[]), // tidy-alphabetical-end ]; -const BPF_ALLOWED_FEATURES: &[(&str, Stability)] = &[("alu32", Unstable(sym::bpf_target_feature))]; +const BPF_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = + &[("alu32", Unstable(sym::bpf_target_feature), &[])]; -const CSKY_ALLOWED_FEATURES: &[(&str, Stability)] = &[ +const CSKY_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("10e60", Unstable(sym::csky_target_feature)), - ("2e3", Unstable(sym::csky_target_feature)), - ("3e3r1", Unstable(sym::csky_target_feature)), - ("3e3r2", Unstable(sym::csky_target_feature)), - ("3e3r3", Unstable(sym::csky_target_feature)), - ("3e7", Unstable(sym::csky_target_feature)), - ("7e10", Unstable(sym::csky_target_feature)), - ("cache", Unstable(sym::csky_target_feature)), - ("doloop", Unstable(sym::csky_target_feature)), - ("dsp1e2", Unstable(sym::csky_target_feature)), - ("dspe60", Unstable(sym::csky_target_feature)), - ("e1", Unstable(sym::csky_target_feature)), - ("e2", Unstable(sym::csky_target_feature)), - ("edsp", Unstable(sym::csky_target_feature)), - ("elrw", Unstable(sym::csky_target_feature)), - ("float1e2", Unstable(sym::csky_target_feature)), - ("float1e3", Unstable(sym::csky_target_feature)), - ("float3e4", Unstable(sym::csky_target_feature)), - ("float7e60", Unstable(sym::csky_target_feature)), - ("floate1", Unstable(sym::csky_target_feature)), - ("hard-tp", Unstable(sym::csky_target_feature)), - ("high-registers", Unstable(sym::csky_target_feature)), - ("hwdiv", Unstable(sym::csky_target_feature)), - ("mp", Unstable(sym::csky_target_feature)), - ("mp1e2", Unstable(sym::csky_target_feature)), - ("nvic", Unstable(sym::csky_target_feature)), - ("trust", Unstable(sym::csky_target_feature)), - ("vdsp2e60f", Unstable(sym::csky_target_feature)), - ("vdspv1", Unstable(sym::csky_target_feature)), - ("vdspv2", Unstable(sym::csky_target_feature)), + ("10e60", Unstable(sym::csky_target_feature), &["7e10"]), + ("2e3", Unstable(sym::csky_target_feature), &["e2"]), + ("3e3r1", Unstable(sym::csky_target_feature), &[]), + ("3e3r2", Unstable(sym::csky_target_feature), &["3e3r1", "doloop"]), + ("3e3r3", Unstable(sym::csky_target_feature), &["doloop"]), + ("3e7", Unstable(sym::csky_target_feature), &["2e3"]), + ("7e10", Unstable(sym::csky_target_feature), &["3e7"]), + ("cache", Unstable(sym::csky_target_feature), &[]), + ("doloop", Unstable(sym::csky_target_feature), &[]), + ("dsp1e2", Unstable(sym::csky_target_feature), &[]), + ("dspe60", Unstable(sym::csky_target_feature), &[]), + ("e1", Unstable(sym::csky_target_feature), &["elrw"]), + ("e2", Unstable(sym::csky_target_feature), &["e2"]), + ("edsp", Unstable(sym::csky_target_feature), &[]), + ("elrw", Unstable(sym::csky_target_feature), &[]), + ("float1e2", Unstable(sym::csky_target_feature), &[]), + ("float1e3", Unstable(sym::csky_target_feature), &[]), + ("float3e4", Unstable(sym::csky_target_feature), &[]), + ("float7e60", Unstable(sym::csky_target_feature), &[]), + ("floate1", Unstable(sym::csky_target_feature), &[]), + ("hard-tp", Unstable(sym::csky_target_feature), &[]), + ("high-registers", Unstable(sym::csky_target_feature), &[]), + ("hwdiv", Unstable(sym::csky_target_feature), &[]), + ("mp", Unstable(sym::csky_target_feature), &["2e3"]), + ("mp1e2", Unstable(sym::csky_target_feature), &["3e7"]), + ("nvic", Unstable(sym::csky_target_feature), &[]), + ("trust", Unstable(sym::csky_target_feature), &[]), + ("vdsp2e60f", Unstable(sym::csky_target_feature), &[]), + ("vdspv1", Unstable(sym::csky_target_feature), &[]), + ("vdspv2", Unstable(sym::csky_target_feature), &[]), // tidy-alphabetical-end //fpu // tidy-alphabetical-start - ("fdivdu", Unstable(sym::csky_target_feature)), - ("fpuv2_df", Unstable(sym::csky_target_feature)), - ("fpuv2_sf", Unstable(sym::csky_target_feature)), - ("fpuv3_df", Unstable(sym::csky_target_feature)), - ("fpuv3_hf", Unstable(sym::csky_target_feature)), - ("fpuv3_hi", Unstable(sym::csky_target_feature)), - ("fpuv3_sf", Unstable(sym::csky_target_feature)), - ("hard-float", Unstable(sym::csky_target_feature)), - ("hard-float-abi", Unstable(sym::csky_target_feature)), - // tidy-alphabetical-end -]; - -const LOONGARCH_ALLOWED_FEATURES: &[(&str, Stability)] = &[ - // tidy-alphabetical-start - ("d", Unstable(sym::loongarch_target_feature)), - ("f", Unstable(sym::loongarch_target_feature)), - ("frecipe", Unstable(sym::loongarch_target_feature)), - ("lasx", Unstable(sym::loongarch_target_feature)), - ("lbt", Unstable(sym::loongarch_target_feature)), - ("lsx", Unstable(sym::loongarch_target_feature)), - ("lvz", Unstable(sym::loongarch_target_feature)), - ("relax", Unstable(sym::loongarch_target_feature)), - ("ual", Unstable(sym::loongarch_target_feature)), - // tidy-alphabetical-end -]; - -const IBMZ_ALLOWED_FEATURES: &[(&str, Stability)] = &[ - // tidy-alphabetical-start - ("backchain", Unstable(sym::s390x_target_feature)), - ("vector", Unstable(sym::s390x_target_feature)), - // tidy-alphabetical-end -]; - -const X86_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ - // tidy-alphabetical-start - ("aes", &["sse2"]), - ("avx", &["sse4.2"]), - ("avx2", &["avx"]), - ("avx512bf16", &["avx512bw"]), - ("avx512bitalg", &["avx512bw"]), - ("avx512bw", &["avx512f"]), - ("avx512cd", &["avx512f"]), - ("avx512dq", &["avx512f"]), - ("avx512f", &["avx2"]), - ("avx512fp16", &["avx512bw", "avx512vl", "avx512dq"]), - ("avx512vbmi", &["avx512bw"]), - ("avx512vbmi2", &["avx512bw"]), - ("avx512vl", &["avx512f"]), - ("avx512vnni", &["avx512f"]), - ("avx512vp2intersect", &["avx512f"]), - ("avx512vpopcntdq", &["avx512f"]), - ("f16c", &["avx"]), - ("fma", &["avx"]), - ("gfni", &["sse2"]), - ("pclmulqdq", &["sse2"]), - ("sha", &["sse2"]), - ("sse2", &["sse"]), - ("sse3", &["sse2"]), - ("sse4.1", &["ssse3"]), - ("sse4.2", &["sse4.1"]), - ("ssse3", &["sse3"]), - ("vaes", &["avx", "aes"]), - ("vpclmulqdq", &["avx", "pclmulqdq"]), - ("xsavec", &["xsave"]), - ("xsaveopt", &["xsave"]), - ("xsaves", &["xsave"]), - // tidy-alphabetical-end -]; - -const AARCH64_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ - // tidy-alphabetical-start - ("aes", &["neon"]), - ("f32mm", &["sve"]), - ("f64mm", &["sve"]), - ("fcma", &["neon"]), - ("fhm", &["fp16"]), - ("fp16", &["neon"]), - ("jsconv", &["neon"]), - ("rcpc2", &["rcpc"]), - ("sha2", &["neon"]), - ("sha3", &["sha2"]), - ("sm4", &["neon"]), - ("sve", &["fp16"]), - ("sve2", &["sve"]), - ("sve2-aes", &["sve2", "aes"]), - ("sve2-bitperm", &["sve2"]), - ("sve2-sha3", &["sve2", "sha3"]), - ("sve2-sm4", &["sve2", "sm4"]), + ("fdivdu", Unstable(sym::csky_target_feature), &[]), + ("fpuv2_df", Unstable(sym::csky_target_feature), &[]), + ("fpuv2_sf", Unstable(sym::csky_target_feature), &[]), + ("fpuv3_df", Unstable(sym::csky_target_feature), &[]), + ("fpuv3_hf", Unstable(sym::csky_target_feature), &[]), + ("fpuv3_hi", Unstable(sym::csky_target_feature), &[]), + ("fpuv3_sf", Unstable(sym::csky_target_feature), &[]), + ("hard-float", Unstable(sym::csky_target_feature), &[]), + ("hard-float-abi", Unstable(sym::csky_target_feature), &[]), // tidy-alphabetical-end ]; -const RISCV_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ +const LOONGARCH_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("zb", &["zba", "zbc", "zbs"]), - ("zk", &["zkn", "zkr", "zks", "zkt", "zbkb", "zbkc", "zkbx"]), - ("zkn", &["zknd", "zkne", "zknh", "zbkb", "zbkc", "zkbx"]), - ("zks", &["zksed", "zksh", "zbkb", "zbkc", "zkbx"]), + ("d", Unstable(sym::loongarch_target_feature), &["f"]), + ("f", Unstable(sym::loongarch_target_feature), &[]), + ("frecipe", Unstable(sym::loongarch_target_feature), &[]), + ("lasx", Unstable(sym::loongarch_target_feature), &["lsx"]), + ("lbt", Unstable(sym::loongarch_target_feature), &[]), + ("lsx", Unstable(sym::loongarch_target_feature), &["d"]), + ("lvz", Unstable(sym::loongarch_target_feature), &[]), + ("relax", Unstable(sym::loongarch_target_feature), &[]), + ("ual", Unstable(sym::loongarch_target_feature), &[]), // tidy-alphabetical-end ]; -const WASM_IMPLIED_FEATURES: &[(&str, &[&str])] = &[ +const IBMZ_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // tidy-alphabetical-start - ("relaxed-simd", &["simd128"]), + ("backchain", Unstable(sym::s390x_target_feature), &[]), + ("vector", Unstable(sym::s390x_target_feature), &[]), // tidy-alphabetical-end ]; @@ -501,10 +439,13 @@ pub fn all_known_features() -> impl Iterator { .chain(LOONGARCH_ALLOWED_FEATURES) .chain(IBMZ_ALLOWED_FEATURES) .cloned() + .map(|(f, s, _)| (f, s)) } impl super::spec::Target { - pub fn supported_target_features(&self) -> &'static [(&'static str, Stability)] { + pub fn supported_target_features( + &self, + ) -> &'static [(&'static str, Stability, ImpliedFeatures)] { match &*self.arch { "arm" => ARM_ALLOWED_FEATURES, "aarch64" | "arm64ec" => AARCH64_ALLOWED_FEATURES, @@ -528,14 +469,4 @@ impl super::spec::Target { _ => &[], } } - - pub fn implied_target_features(&self) -> &'static [(&'static str, &'static [&'static str])] { - match &*self.arch { - "aarch4" => AARCH64_IMPLIED_FEATURES, - "riscv32" | "riscv64" => RISCV_IMPLIED_FEATURES, - "x86" | "x86_64" => X86_IMPLIED_FEATURES, - "wasm32" | "wasm64" => WASM_IMPLIED_FEATURES, - _ => &[], - } - } } From d327fdbc098ebc54d37d1c35984be7005cdb7784 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Sat, 3 Aug 2024 04:45:48 -0400 Subject: [PATCH 19/28] Don't use LLVM's target features --- compiler/rustc_codegen_llvm/src/back/write.rs | 8 ++- compiler/rustc_codegen_llvm/src/context.rs | 2 +- compiler/rustc_codegen_llvm/src/lib.rs | 2 +- compiler/rustc_codegen_llvm/src/llvm_util.rs | 63 ++++++++++++++----- .../rustc_codegen_ssa/src/target_features.rs | 23 +------ compiler/rustc_target/src/target_features.rs | 25 ++++++++ 6 files changed, 84 insertions(+), 39 deletions(-) diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 5a7909d151139..890fcf508a8bd 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -95,11 +95,15 @@ pub fn write_output_file<'ll>( } } -pub fn create_informational_target_machine(sess: &Session) -> OwnedTargetMachine { +pub fn create_informational_target_machine( + sess: &Session, + extra_features: bool, +) -> OwnedTargetMachine { let config = TargetMachineFactoryConfig { split_dwarf_file: None, output_obj_file: None }; // Can't use query system here quite yet because this function is invoked before the query // system/tcx is set up. - let features = llvm_util::global_llvm_features(sess, false); + let features = + if extra_features { llvm_util::global_llvm_features(sess, false) } else { Vec::new() }; target_machine_factory(sess, config::OptLevel::No, &features)(config) .unwrap_or_else(|err| llvm_err(sess.dcx(), err).raise()) } diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs index ea930421b5869..1dc3fbfc7b303 100644 --- a/compiler/rustc_codegen_llvm/src/context.rs +++ b/compiler/rustc_codegen_llvm/src/context.rs @@ -149,7 +149,7 @@ pub unsafe fn create_module<'ll>( // Ensure the data-layout values hardcoded remain the defaults. { - let tm = crate::back::write::create_informational_target_machine(tcx.sess); + let tm = crate::back::write::create_informational_target_machine(tcx.sess, true); unsafe { llvm::LLVMRustSetDataLayoutFromTargetMachine(llmod, &tm); } diff --git a/compiler/rustc_codegen_llvm/src/lib.rs b/compiler/rustc_codegen_llvm/src/lib.rs index 41e9cfd1066b9..333f1fdf6e03a 100644 --- a/compiler/rustc_codegen_llvm/src/lib.rs +++ b/compiler/rustc_codegen_llvm/src/lib.rs @@ -434,7 +434,7 @@ impl ModuleLlvm { ModuleLlvm { llmod_raw, llcx, - tm: ManuallyDrop::new(create_informational_target_machine(tcx.sess)), + tm: ManuallyDrop::new(create_informational_target_machine(tcx.sess, true)), } } } diff --git a/compiler/rustc_codegen_llvm/src/llvm_util.rs b/compiler/rustc_codegen_llvm/src/llvm_util.rs index c70f6dd8180bf..e85974b7cc1ec 100644 --- a/compiler/rustc_codegen_llvm/src/llvm_util.rs +++ b/compiler/rustc_codegen_llvm/src/llvm_util.rs @@ -308,7 +308,53 @@ pub fn check_tied_features( /// Used to generate cfg variables and apply features /// Must express features in the way Rust understands them pub fn target_features(sess: &Session, allow_unstable: bool) -> Vec { - let target_machine = create_informational_target_machine(sess); + let rust_features = sess + .target + .supported_target_features() + .iter() + .map(|(feature, _, _)| { + (to_llvm_features(sess, feature).llvm_feature_name, Symbol::intern(feature)) + }) + .collect::>(); + + let mut features = FxHashSet::default(); + + // Add base features for the target + let target_machine = create_informational_target_machine(sess, false); + features.extend( + sess.target + .supported_target_features() + .iter() + .filter(|(feature, _, _)| { + // skip checking special features, as LLVM may not understands them + if RUSTC_SPECIAL_FEATURES.contains(feature) { + return true; + } + // check that all features in a given smallvec are enabled + for llvm_feature in to_llvm_features(sess, feature) { + let cstr = SmallCStr::new(llvm_feature); + if !unsafe { llvm::LLVMRustHasFeature(&target_machine, cstr.as_ptr()) } { + return false; + } + } + true + }) + .map(|(feature, _, _)| Symbol::intern(feature)), + ); + + // Add enabled features + for llvm_feature in global_llvm_features(sess, false) { + let (add, llvm_feature) = llvm_feature.split_at(1); + let feature = + rust_features.get(llvm_feature).cloned().unwrap_or(Symbol::intern(llvm_feature)); + if add == "+" { + features.extend(sess.target.implied_target_features(std::iter::once(feature))); + } else if add == "-" { + features.remove(&feature); + } + } + + // Filter enabled features based on feature gates sess.target .supported_target_features() .iter() @@ -320,18 +366,7 @@ pub fn target_features(sess: &Session, allow_unstable: bool) -> Vec { } }) .filter(|feature| { - // skip checking special features, as LLVM may not understands them - if RUSTC_SPECIAL_FEATURES.contains(feature) { - return true; - } - // check that all features in a given smallvec are enabled - for llvm_feature in to_llvm_features(sess, feature) { - let cstr = SmallCStr::new(llvm_feature); - if !unsafe { llvm::LLVMRustHasFeature(&target_machine, cstr.as_ptr()) } { - return false; - } - } - true + RUSTC_SPECIAL_FEATURES.contains(feature) || features.contains(&Symbol::intern(feature)) }) .map(|feature| Symbol::intern(feature)) .collect() @@ -440,7 +475,7 @@ fn print_target_features(out: &mut String, sess: &Session, tm: &llvm::TargetMach pub(crate) fn print(req: &PrintRequest, mut out: &mut String, sess: &Session) { require_inited(); - let tm = create_informational_target_machine(sess); + let tm = create_informational_target_machine(sess, true); match req.kind { PrintKind::TargetCPUs => { // SAFETY generate a C compatible string from a byte slice to pass diff --git a/compiler/rustc_codegen_ssa/src/target_features.rs b/compiler/rustc_codegen_ssa/src/target_features.rs index 536f39375bb90..c84b844cd0401 100644 --- a/compiler/rustc_codegen_ssa/src/target_features.rs +++ b/compiler/rustc_codegen_ssa/src/target_features.rs @@ -1,6 +1,6 @@ use rustc_ast::ast; use rustc_attr::InstructionSetAttr; -use rustc_data_structures::fx::{FxHashMap, FxIndexSet}; +use rustc_data_structures::fx::FxIndexSet; use rustc_data_structures::unord::{ExtendUnord, UnordMap, UnordSet}; use rustc_errors::Applicability; use rustc_hir::def::DefKind; @@ -165,26 +165,7 @@ pub(crate) fn provide(providers: &mut Providers) { } }, implied_target_features: |tcx, feature| { - let implied_features = tcx - .sess - .target - .supported_target_features() - .iter() - .map(|(f, _, i)| (Symbol::intern(f), i)) - .collect::>(); - - // implied target features have their own implied target features, so we traverse the - // map until there are no more features to add - let mut features = UnordSet::new(); - let mut new_features = vec![feature]; - while let Some(new_feature) = new_features.pop() { - if features.insert(new_feature) { - if let Some(implied_features) = implied_features.get(&new_feature) { - new_features.extend(implied_features.iter().copied().map(Symbol::intern)) - } - } - } - features + tcx.sess.target.implied_target_features(std::iter::once(feature)).into() }, asm_target_features, ..*providers diff --git a/compiler/rustc_target/src/target_features.rs b/compiler/rustc_target/src/target_features.rs index f408db1dcbad2..0e9979e1bb7f6 100644 --- a/compiler/rustc_target/src/target_features.rs +++ b/compiler/rustc_target/src/target_features.rs @@ -1,3 +1,4 @@ +use rustc_data_structures::fx::{FxHashMap, FxHashSet}; use rustc_span::symbol::{sym, Symbol}; /// Features that control behaviour of rustc, rather than the codegen. @@ -469,4 +470,28 @@ impl super::spec::Target { _ => &[], } } + + pub fn implied_target_features( + &self, + base_features: impl Iterator, + ) -> FxHashSet { + let implied_features = self + .supported_target_features() + .iter() + .map(|(f, _, i)| (Symbol::intern(f), i)) + .collect::>(); + + // implied target features have their own implied target features, so we traverse the + // map until there are no more features to add + let mut features = FxHashSet::default(); + let mut new_features = base_features.collect::>(); + while let Some(new_feature) = new_features.pop() { + if features.insert(new_feature) { + if let Some(implied_features) = implied_features.get(&new_feature) { + new_features.extend(implied_features.iter().copied().map(Symbol::intern)) + } + } + } + features + } } From 9869a9942fd87a1553c8dfc0d5a612c1a3d8e6b2 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Sat, 3 Aug 2024 23:51:37 -0400 Subject: [PATCH 20/28] Don't use LLVM to compute -Ctarget-feature --- compiler/rustc_codegen_llvm/src/back/write.rs | 5 +- compiler/rustc_codegen_llvm/src/context.rs | 2 +- compiler/rustc_codegen_llvm/src/lib.rs | 4 +- compiler/rustc_codegen_llvm/src/llvm_util.rs | 211 +++++++++--------- compiler/rustc_target/src/target_features.rs | 8 +- tests/codegen/target-feature-overrides.rs | 2 +- 6 files changed, 120 insertions(+), 112 deletions(-) diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs index 890fcf508a8bd..a1f2433ab6f3b 100644 --- a/compiler/rustc_codegen_llvm/src/back/write.rs +++ b/compiler/rustc_codegen_llvm/src/back/write.rs @@ -97,13 +97,12 @@ pub fn write_output_file<'ll>( pub fn create_informational_target_machine( sess: &Session, - extra_features: bool, + only_base_features: bool, ) -> OwnedTargetMachine { let config = TargetMachineFactoryConfig { split_dwarf_file: None, output_obj_file: None }; // Can't use query system here quite yet because this function is invoked before the query // system/tcx is set up. - let features = - if extra_features { llvm_util::global_llvm_features(sess, false) } else { Vec::new() }; + let features = llvm_util::global_llvm_features(sess, false, only_base_features); target_machine_factory(sess, config::OptLevel::No, &features)(config) .unwrap_or_else(|err| llvm_err(sess.dcx(), err).raise()) } diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs index 1dc3fbfc7b303..173b8a479efc5 100644 --- a/compiler/rustc_codegen_llvm/src/context.rs +++ b/compiler/rustc_codegen_llvm/src/context.rs @@ -149,7 +149,7 @@ pub unsafe fn create_module<'ll>( // Ensure the data-layout values hardcoded remain the defaults. { - let tm = crate::back::write::create_informational_target_machine(tcx.sess, true); + let tm = crate::back::write::create_informational_target_machine(tcx.sess, false); unsafe { llvm::LLVMRustSetDataLayoutFromTargetMachine(llmod, &tm); } diff --git a/compiler/rustc_codegen_llvm/src/lib.rs b/compiler/rustc_codegen_llvm/src/lib.rs index 333f1fdf6e03a..518a86e0cb06d 100644 --- a/compiler/rustc_codegen_llvm/src/lib.rs +++ b/compiler/rustc_codegen_llvm/src/lib.rs @@ -269,7 +269,7 @@ impl CodegenBackend for LlvmCodegenBackend { fn provide(&self, providers: &mut Providers) { providers.global_backend_features = - |tcx, ()| llvm_util::global_llvm_features(tcx.sess, true) + |tcx, ()| llvm_util::global_llvm_features(tcx.sess, true, false) } fn print(&self, req: &PrintRequest, out: &mut String, sess: &Session) { @@ -434,7 +434,7 @@ impl ModuleLlvm { ModuleLlvm { llmod_raw, llcx, - tm: ManuallyDrop::new(create_informational_target_machine(tcx.sess, true)), + tm: ManuallyDrop::new(create_informational_target_machine(tcx.sess, false)), } } } diff --git a/compiler/rustc_codegen_llvm/src/llvm_util.rs b/compiler/rustc_codegen_llvm/src/llvm_util.rs index e85974b7cc1ec..3190853a84c90 100644 --- a/compiler/rustc_codegen_llvm/src/llvm_util.rs +++ b/compiler/rustc_codegen_llvm/src/llvm_util.rs @@ -308,19 +308,10 @@ pub fn check_tied_features( /// Used to generate cfg variables and apply features /// Must express features in the way Rust understands them pub fn target_features(sess: &Session, allow_unstable: bool) -> Vec { - let rust_features = sess - .target - .supported_target_features() - .iter() - .map(|(feature, _, _)| { - (to_llvm_features(sess, feature).llvm_feature_name, Symbol::intern(feature)) - }) - .collect::>(); - let mut features = FxHashSet::default(); // Add base features for the target - let target_machine = create_informational_target_machine(sess, false); + let target_machine = create_informational_target_machine(sess, true); features.extend( sess.target .supported_target_features() @@ -343,13 +334,16 @@ pub fn target_features(sess: &Session, allow_unstable: bool) -> Vec { ); // Add enabled features - for llvm_feature in global_llvm_features(sess, false) { - let (add, llvm_feature) = llvm_feature.split_at(1); - let feature = - rust_features.get(llvm_feature).cloned().unwrap_or(Symbol::intern(llvm_feature)); - if add == "+" { + for (enabled, feature) in + sess.opts.cg.target_feature.split(',').filter_map(|s| match s.chars().next() { + Some('+') => Some((true, Symbol::intern(&s[1..]))), + Some('-') => Some((false, Symbol::intern(&s[1..]))), + _ => None, + }) + { + if enabled { features.extend(sess.target.implied_target_features(std::iter::once(feature))); - } else if add == "-" { + } else { features.remove(&feature); } } @@ -475,7 +469,7 @@ fn print_target_features(out: &mut String, sess: &Session, tm: &llvm::TargetMach pub(crate) fn print(req: &PrintRequest, mut out: &mut String, sess: &Session) { require_inited(); - let tm = create_informational_target_machine(sess, true); + let tm = create_informational_target_machine(sess, false); match req.kind { PrintKind::TargetCPUs => { // SAFETY generate a C compatible string from a byte slice to pass @@ -523,7 +517,11 @@ pub fn target_cpu(sess: &Session) -> &str { /// The list of LLVM features computed from CLI flags (`-Ctarget-cpu`, `-Ctarget-feature`, /// `--target` and similar). -pub(crate) fn global_llvm_features(sess: &Session, diagnostics: bool) -> Vec { +pub(crate) fn global_llvm_features( + sess: &Session, + diagnostics: bool, + only_base_features: bool, +) -> Vec { // Features that come earlier are overridden by conflicting features later in the string. // Typically we'll want more explicit settings to override the implicit ones, so: // @@ -583,96 +581,109 @@ pub(crate) fn global_llvm_features(sess: &Session, diagnostics: bool) -> Vec return None, - Some(c @ ('+' | '-')) => c, - Some(_) => { - if diagnostics { - sess.dcx().emit_warn(UnknownCTargetFeaturePrefix { feature: s }); + if !only_base_features { + let supported_features = sess.target.supported_target_features(); + let (llvm_major, _, _) = get_version(); + let mut featsmap = FxHashMap::default(); + let feats = sess + .opts + .cg + .target_feature + .split(',') + .filter_map(|s| { + let enable_disable = match s.chars().next() { + None => return None, + Some(c @ ('+' | '-')) => c, + Some(_) => { + if diagnostics { + sess.dcx().emit_warn(UnknownCTargetFeaturePrefix { feature: s }); + } + return None; } - return None; - } - }; + }; - let feature = backend_feature_name(sess, s)?; - // Warn against use of LLVM specific feature names and unstable features on the CLI. - if diagnostics { - let feature_state = supported_features.iter().find(|&&(v, _, _)| v == feature); - if feature_state.is_none() { - let rust_feature = - supported_features.iter().find_map(|&(rust_feature, _, _)| { - let llvm_features = to_llvm_features(sess, rust_feature); - if llvm_features.contains(feature) - && !llvm_features.contains(rust_feature) - { - Some(rust_feature) - } else { - None + let feature = backend_feature_name(sess, s)?; + // Warn against use of LLVM specific feature names and unstable features on the CLI. + if diagnostics { + let feature_state = supported_features.iter().find(|&&(v, _, _)| v == feature); + if feature_state.is_none() { + let rust_feature = + supported_features.iter().find_map(|&(rust_feature, _, _)| { + let llvm_features = to_llvm_features(sess, rust_feature); + if llvm_features.contains(feature) + && !llvm_features.contains(rust_feature) + { + Some(rust_feature) + } else { + None + } + }); + let unknown_feature = if let Some(rust_feature) = rust_feature { + UnknownCTargetFeature { + feature, + rust_feature: PossibleFeature::Some { rust_feature }, } - }); - let unknown_feature = if let Some(rust_feature) = rust_feature { - UnknownCTargetFeature { - feature, - rust_feature: PossibleFeature::Some { rust_feature }, - } - } else { - UnknownCTargetFeature { feature, rust_feature: PossibleFeature::None } - }; - sess.dcx().emit_warn(unknown_feature); - } else if feature_state - .is_some_and(|(_name, feature_gate, _implied)| !feature_gate.is_stable()) - { - // An unstable feature. Warn about using it. - sess.dcx().emit_warn(UnstableCTargetFeature { feature }); + } else { + UnknownCTargetFeature { feature, rust_feature: PossibleFeature::None } + }; + sess.dcx().emit_warn(unknown_feature); + } else if feature_state + .is_some_and(|(_name, feature_gate, _implied)| !feature_gate.is_stable()) + { + // An unstable feature. Warn about using it. + sess.dcx().emit_warn(UnstableCTargetFeature { feature }); + } } - } - if diagnostics { - // FIXME(nagisa): figure out how to not allocate a full hashset here. - featsmap.insert(feature, enable_disable == '+'); - } + if diagnostics { + // FIXME(nagisa): figure out how to not allocate a full hashset here. + featsmap.insert(feature, enable_disable == '+'); + } - // rustc-specific features do not get passed down to LLVM… - if RUSTC_SPECIFIC_FEATURES.contains(&feature) { - return None; - } + // rustc-specific features do not get passed down to LLVM… + if RUSTC_SPECIFIC_FEATURES.contains(&feature) { + return None; + } - // if the target-feature is "backchain" and LLVM version is greater than 18 - // then we also need to add "+backchain" to the target-features attribute. - // otherwise, we will only add the naked `backchain` attribute to the attribute-group. - if feature == "backchain" && llvm_major < 18 { - return None; - } - // ... otherwise though we run through `to_llvm_features` when - // passing requests down to LLVM. This means that all in-language - // features also work on the command line instead of having two - // different names when the LLVM name and the Rust name differ. - let llvm_feature = to_llvm_features(sess, feature); - - Some( - std::iter::once(format!("{}{}", enable_disable, llvm_feature.llvm_feature_name)) - .chain(llvm_feature.dependency.into_iter().filter_map(move |feat| { - match (enable_disable, feat) { + // if the target-feature is "backchain" and LLVM version is greater than 18 + // then we also need to add "+backchain" to the target-features attribute. + // otherwise, we will only add the naked `backchain` attribute to the attribute-group. + if feature == "backchain" && llvm_major < 18 { + return None; + } + // ... otherwise though we run through `to_llvm_features` when + // passing requests down to LLVM. This means that all in-language + // features also work on the command line instead of having two + // different names when the LLVM name and the Rust name differ. + let llvm_feature = to_llvm_features(sess, feature); + + Some( + std::iter::once(format!( + "{}{}", + enable_disable, llvm_feature.llvm_feature_name + )) + .chain(llvm_feature.dependency.into_iter().filter_map( + move |feat| match (enable_disable, feat) { ('-' | '+', TargetFeatureFoldStrength::Both(f)) | ('+', TargetFeatureFoldStrength::EnableOnly(f)) => { Some(format!("{enable_disable}{f}")) } _ => None, - } - })), - ) - }) - .flatten(); - features.extend(feats); + }, + )), + ) + }) + .flatten(); + features.extend(feats); + + if diagnostics && let Some(f) = check_tied_features(sess, &featsmap) { + sess.dcx().emit_err(TargetFeatureDisableOrEnable { + features: f, + span: None, + missing_features: None, + }); + } + } // -Zfixed-x18 if sess.opts.unstable_opts.fixed_x18 { @@ -683,14 +694,6 @@ pub(crate) fn global_llvm_features(sess: &Session, diagnostics: bool) -> Vec Date: Sun, 4 Aug 2024 01:02:53 -0400 Subject: [PATCH 21/28] Remove redundant implied features --- compiler/rustc_codegen_llvm/src/llvm_util.rs | 36 ++------------------ compiler/rustc_target/src/target_features.rs | 6 ++-- 2 files changed, 6 insertions(+), 36 deletions(-) diff --git a/compiler/rustc_codegen_llvm/src/llvm_util.rs b/compiler/rustc_codegen_llvm/src/llvm_util.rs index 3190853a84c90..10cba179c7564 100644 --- a/compiler/rustc_codegen_llvm/src/llvm_util.rs +++ b/compiler/rustc_codegen_llvm/src/llvm_util.rs @@ -239,40 +239,8 @@ pub fn to_llvm_features<'a>(sess: &Session, s: &'a str) -> LLVMFeature<'a> { } // In LLVM neon implicitly enables fp, but we manually enable // neon when a feature only implicitly enables fp - ("aarch64", "f32mm") => { - LLVMFeature::with_dependency("f32mm", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "f64mm") => { - LLVMFeature::with_dependency("f64mm", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "fhm") => { - LLVMFeature::with_dependency("fp16fml", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "fp16") => { - LLVMFeature::with_dependency("fullfp16", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "jsconv") => { - LLVMFeature::with_dependency("jsconv", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "sve") => { - LLVMFeature::with_dependency("sve", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "sve2") => { - LLVMFeature::with_dependency("sve2", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "sve2-aes") => { - LLVMFeature::with_dependency("sve2-aes", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "sve2-sm4") => { - LLVMFeature::with_dependency("sve2-sm4", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "sve2-sha3") => { - LLVMFeature::with_dependency("sve2-sha3", TargetFeatureFoldStrength::EnableOnly("neon")) - } - ("aarch64", "sve2-bitperm") => LLVMFeature::with_dependency( - "sve2-bitperm", - TargetFeatureFoldStrength::EnableOnly("neon"), - ), + ("aarch64", "fhm") => LLVMFeature::new("fp16fml"), + ("aarch64", "fp16") => LLVMFeature::new("fullfp16"), // In LLVM 18, `unaligned-scalar-mem` was merged with `unaligned-vector-mem` into a single feature called // `fast-unaligned-access`. In LLVM 19, it was split back out. ("riscv32" | "riscv64", "unaligned-scalar-mem") if get_version().0 == 18 => { diff --git a/compiler/rustc_target/src/target_features.rs b/compiler/rustc_target/src/target_features.rs index 63bdcfddfc5f7..746926308b6b4 100644 --- a/compiler/rustc_target/src/target_features.rs +++ b/compiler/rustc_target/src/target_features.rs @@ -118,13 +118,15 @@ const AARCH64_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ // FEAT_FLAGM ("flagm", Stable, &[]), // FEAT_FP16 - ("fp16", Stable, &[]), + // Rust ties FP and Neon: https://github.com/rust-lang/rust/pull/91608 + ("fp16", Stable, &["neon"]), // FEAT_FRINTTS ("frintts", Stable, &[]), // FEAT_I8MM ("i8mm", Stable, &[]), // FEAT_JSCVT - ("jsconv", Stable, &[]), + // Rust ties FP and Neon: https://github.com/rust-lang/rust/pull/91608 + ("jsconv", Stable, &["neon"]), // FEAT_LOR ("lor", Stable, &[]), // FEAT_LSE From 9c00b345b2ca8ec2d4f9d47dd9a0b5098acf4313 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Sun, 4 Aug 2024 15:32:17 -0400 Subject: [PATCH 22/28] Add implied features to non-target-feature functions --- compiler/rustc_codegen_llvm/src/llvm_util.rs | 26 ++++++++++++++++---- tests/codegen/target-feature-overrides.rs | 6 ++--- tests/codegen/tied-features-strength.rs | 2 +- 3 files changed, 25 insertions(+), 9 deletions(-) diff --git a/compiler/rustc_codegen_llvm/src/llvm_util.rs b/compiler/rustc_codegen_llvm/src/llvm_util.rs index 10cba179c7564..1a80824a3b70d 100644 --- a/compiler/rustc_codegen_llvm/src/llvm_util.rs +++ b/compiler/rustc_codegen_llvm/src/llvm_util.rs @@ -8,6 +8,7 @@ use libc::c_int; use rustc_codegen_ssa::base::wants_wasm_eh; use rustc_data_structures::fx::{FxHashMap, FxHashSet}; use rustc_data_structures::small_c_str::SmallCStr; +use rustc_data_structures::unord::UnordSet; use rustc_fs_util::path_to_c_string; use rustc_middle::bug; use rustc_session::config::{PrintKind, PrintRequest}; @@ -553,11 +554,26 @@ pub(crate) fn global_llvm_features( let supported_features = sess.target.supported_target_features(); let (llvm_major, _, _) = get_version(); let mut featsmap = FxHashMap::default(); - let feats = sess - .opts - .cg - .target_feature - .split(',') + + // insert implied features + let mut all_rust_features = vec![]; + for feature in sess.opts.cg.target_feature.split(',') { + match feature.strip_prefix('+') { + Some(feature) => all_rust_features.extend( + UnordSet::from( + sess.target + .implied_target_features(std::iter::once(Symbol::intern(feature))), + ) + .to_sorted_stable_ord() + .iter() + .map(|s| format!("+{}", s.as_str())), + ), + _ => all_rust_features.push(feature.to_string()), + } + } + + let feats = all_rust_features + .iter() .filter_map(|s| { let enable_disable = match s.chars().next() { None => return None, diff --git a/tests/codegen/target-feature-overrides.rs b/tests/codegen/target-feature-overrides.rs index 5b7489c96cf97..f38a1ae72de5d 100644 --- a/tests/codegen/target-feature-overrides.rs +++ b/tests/codegen/target-feature-overrides.rs @@ -1,7 +1,7 @@ //@ revisions: COMPAT INCOMPAT //@ needs-llvm-components: x86 //@ compile-flags: --target=x86_64-unknown-linux-gnu -Copt-level=3 -//@ [COMPAT] compile-flags: -Ctarget-feature=+avx2,+avx +//@ [COMPAT] compile-flags: -Ctarget-feature=+avx2 //@ [INCOMPAT] compile-flags: -Ctarget-feature=-avx2,-avx // See also tests/assembly/target-feature-multiple.rs @@ -39,8 +39,8 @@ pub unsafe fn banana() -> u32 { } // CHECK: attributes [[APPLEATTRS]] -// COMPAT-SAME: "target-features"="+avx2,+avx,{{.*}}" +// COMPAT-SAME: "target-features"="+avx,+avx2,{{.*}}" // INCOMPAT-SAME: "target-features"="-avx2,-avx,+avx,{{.*}}" // CHECK: attributes [[BANANAATTRS]] -// COMPAT-SAME: "target-features"="+avx2,+avx,{{.*}}" +// COMPAT-SAME: "target-features"="+avx,+avx2,{{.*}}" // INCOMPAT-SAME: "target-features"="-avx2,-avx" diff --git a/tests/codegen/tied-features-strength.rs b/tests/codegen/tied-features-strength.rs index 7f0805bc1b435..1b4596ae2cb57 100644 --- a/tests/codegen/tied-features-strength.rs +++ b/tests/codegen/tied-features-strength.rs @@ -8,7 +8,7 @@ // is LLVM-14 we can remove the optional regex matching for this feature. //@ [ENABLE_SVE] compile-flags: -C target-feature=+sve -Copt-level=0 -// ENABLE_SVE: attributes #0 = { {{.*}} "target-features"="{{((\+outline-atomics,?)|(\+v8a,?)?|(\+sve,?)|(\+neon,?))*}}" } +// ENABLE_SVE: attributes #0 = { {{.*}} "target-features"="{{((\+outline-atomics,?)|(\+v8a,?)?|(\+sve,?)|(\+neon,?)|(\+fp-armv8,?))*}}" } //@ [DISABLE_SVE] compile-flags: -C target-feature=-sve -Copt-level=0 // DISABLE_SVE: attributes #0 = { {{.*}} "target-features"="{{((\+outline-atomics,?)|(\+v8a,?)?|(-sve,?)|(\+neon,?))*}}" } From a141d768bd984c9728d213009040c92610d53217 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Sun, 4 Aug 2024 23:51:59 -0400 Subject: [PATCH 23/28] Hide implicit target features from diagnostics when possible --- compiler/rustc_codegen_gcc/src/attributes.rs | 2 +- compiler/rustc_codegen_llvm/src/attributes.rs | 2 +- .../rustc_codegen_ssa/src/target_features.rs | 28 ++++++++++++++----- .../src/interpret/terminator.rs | 13 +++++++-- .../src/middle/codegen_fn_attrs.rs | 11 +++++++- .../rustc_mir_build/src/check_unsafety.rs | 26 +++++++++++++---- compiler/rustc_mir_transform/src/inline.rs | 4 ++- .../simd_feature_flag_difference.stderr | 4 +-- .../fail/function_calls/target_feature.rs | 2 +- .../fail/function_calls/target_feature.stderr | 4 +-- .../pass/shims/x86/intrinsics-x86-aes-vaes.rs | 2 +- .../pass/shims/x86/intrinsics-x86-avx.rs | 2 +- .../pass/shims/x86/intrinsics-x86-avx2.rs | 2 +- .../pass/shims/x86/intrinsics-x86-avx512.rs | 2 +- .../const-eval/const_fn_target_feature.stderr | 2 +- .../safe-calls.stderr | 26 ++++++++--------- 16 files changed, 89 insertions(+), 43 deletions(-) diff --git a/compiler/rustc_codegen_gcc/src/attributes.rs b/compiler/rustc_codegen_gcc/src/attributes.rs index e521551304ef8..5fdf2680aac88 100644 --- a/compiler/rustc_codegen_gcc/src/attributes.rs +++ b/compiler/rustc_codegen_gcc/src/attributes.rs @@ -75,7 +75,7 @@ pub fn from_fn_attrs<'gcc, 'tcx>( let function_features = codegen_fn_attrs .target_features .iter() - .map(|features| features.as_str()) + .map(|features| features.name.as_str()) .collect::>(); if let Some(features) = check_tied_features( diff --git a/compiler/rustc_codegen_llvm/src/attributes.rs b/compiler/rustc_codegen_llvm/src/attributes.rs index ad38814a68b6d..876f05c6277c1 100644 --- a/compiler/rustc_codegen_llvm/src/attributes.rs +++ b/compiler/rustc_codegen_llvm/src/attributes.rs @@ -496,7 +496,7 @@ pub fn from_fn_attrs<'ll, 'tcx>( to_add.extend(tune_cpu_attr(cx)); let function_features = - codegen_fn_attrs.target_features.iter().map(|f| f.as_str()).collect::>(); + codegen_fn_attrs.target_features.iter().map(|f| f.name.as_str()).collect::>(); if let Some(f) = llvm_util::check_tied_features( cx.tcx.sess, diff --git a/compiler/rustc_codegen_ssa/src/target_features.rs b/compiler/rustc_codegen_ssa/src/target_features.rs index c84b844cd0401..145b1ece23007 100644 --- a/compiler/rustc_codegen_ssa/src/target_features.rs +++ b/compiler/rustc_codegen_ssa/src/target_features.rs @@ -6,6 +6,7 @@ use rustc_errors::Applicability; use rustc_hir::def::DefKind; use rustc_hir::def_id::{DefId, LocalDefId, LOCAL_CRATE}; use rustc_middle::bug; +use rustc_middle::middle::codegen_fn_attrs::TargetFeature; use rustc_middle::query::Providers; use rustc_middle::ty::TyCtxt; use rustc_session::parse::feature_err; @@ -18,7 +19,7 @@ pub fn from_target_feature( tcx: TyCtxt<'_>, attr: &ast::Attribute, supported_target_features: &UnordMap>, - target_features: &mut Vec, + target_features: &mut Vec, ) { let Some(list) = attr.meta_item_list() else { return }; let bad_item = |span| { @@ -99,14 +100,27 @@ pub fn from_target_feature( })); } - // Add both explicit and implied target features, using a set to deduplicate - let mut target_features_set = UnordSet::new(); + // Add explicit features + target_features.extend( + added_target_features.iter().copied().map(|name| TargetFeature { name, implied: false }), + ); + + // Add implied features + let mut implied_target_features = UnordSet::new(); for feature in added_target_features.iter() { - target_features_set + implied_target_features .extend_unord(tcx.implied_target_features(*feature).clone().into_items()); } - target_features_set.extend(added_target_features); - target_features.extend(target_features_set.into_sorted_stable_ord()) + for feature in added_target_features.iter() { + implied_target_features.remove(feature); + } + target_features.extend( + implied_target_features + .into_sorted_stable_ord() + .iter() + .copied() + .map(|name| TargetFeature { name, implied: true }), + ) } /// Computes the set of target features used in a function for the purposes of @@ -115,7 +129,7 @@ fn asm_target_features(tcx: TyCtxt<'_>, did: DefId) -> &FxIndexSet { let mut target_features = tcx.sess.unstable_target_features.clone(); if tcx.def_kind(did).has_codegen_attrs() { let attrs = tcx.codegen_fn_attrs(did); - target_features.extend(&attrs.target_features); + target_features.extend(attrs.target_features.iter().map(|feature| feature.name)); match attrs.instruction_set { None => {} Some(InstructionSetAttr::ArmA32) => { diff --git a/compiler/rustc_const_eval/src/interpret/terminator.rs b/compiler/rustc_const_eval/src/interpret/terminator.rs index 47d0e22b527d3..63cb4e633f401 100644 --- a/compiler/rustc_const_eval/src/interpret/terminator.rs +++ b/compiler/rustc_const_eval/src/interpret/terminator.rs @@ -998,19 +998,26 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> { && attrs .target_features .iter() - .any(|feature| !self.tcx.sess.target_features.contains(feature)) + .any(|feature| !self.tcx.sess.target_features.contains(&feature.name)) { + // Don't include implicit features in the error, unless only implicit features are + // missing. This should be rare, because it can only happen when an implicit feature + // is disabled, e.g. `+avx2,-avx` + let missing_explicit_features = attrs.target_features.iter().any(|feature| { + !feature.implied && !self.tcx.sess.target_features.contains(&feature.name) + }); throw_ub_custom!( fluent::const_eval_unavailable_target_features_for_fn, unavailable_feats = attrs .target_features .iter() - .filter(|&feature| !self.tcx.sess.target_features.contains(feature)) + .filter(|&feature| !(missing_explicit_features && feature.implied) + && !self.tcx.sess.target_features.contains(&feature.name)) .fold(String::new(), |mut s, feature| { if !s.is_empty() { s.push_str(", "); } - s.push_str(feature.as_str()); + s.push_str(feature.name.as_str()); s }), ); diff --git a/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs b/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs index ff6a3a9c12d36..b7d290e58d22b 100644 --- a/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs +++ b/compiler/rustc_middle/src/middle/codegen_fn_attrs.rs @@ -28,7 +28,7 @@ pub struct CodegenFnAttrs { pub link_ordinal: Option, /// The `#[target_feature(enable = "...")]` attribute and the enabled /// features (only enabled features are supported right now). - pub target_features: Vec, + pub target_features: Vec, /// The `#[linkage = "..."]` attribute on Rust-defined items and the value we found. pub linkage: Option, /// The `#[linkage = "..."]` attribute on foreign items and the value we found. @@ -51,6 +51,15 @@ pub struct CodegenFnAttrs { pub patchable_function_entry: Option, } +#[derive(Copy, Clone, Debug, TyEncodable, TyDecodable, HashStable)] +pub struct TargetFeature { + /// The name of the target feature (e.g. "avx") + pub name: Symbol, + /// The feature is implied by another feature, rather than explicitly added by the + /// `#[target_feature]` attribute + pub implied: bool, +} + #[derive(Copy, Clone, Debug, TyEncodable, TyDecodable, HashStable)] pub struct PatchableFunctionEntry { /// Nops to prepend to the function diff --git a/compiler/rustc_mir_build/src/check_unsafety.rs b/compiler/rustc_mir_build/src/check_unsafety.rs index 48018fcaa36df..0b4f0632f2b36 100644 --- a/compiler/rustc_mir_build/src/check_unsafety.rs +++ b/compiler/rustc_mir_build/src/check_unsafety.rs @@ -5,6 +5,7 @@ use std::ops::Bound; use rustc_errors::DiagArgValue; use rustc_hir::def::DefKind; use rustc_hir::{self as hir, BindingMode, ByRef, HirId, Mutability}; +use rustc_middle::middle::codegen_fn_attrs::TargetFeature; use rustc_middle::mir::BorrowKind; use rustc_middle::span_bug; use rustc_middle::thir::visit::Visitor; @@ -31,7 +32,7 @@ struct UnsafetyVisitor<'a, 'tcx> { safety_context: SafetyContext, /// The `#[target_feature]` attributes of the body. Used for checking /// calls to functions with `#[target_feature]` (RFC 2396). - body_target_features: &'tcx [Symbol], + body_target_features: &'tcx [TargetFeature], /// When inside the LHS of an assignment to a field, this is the type /// of the LHS and the span of the assignment expression. assignment_info: Option>, @@ -442,14 +443,29 @@ impl<'a, 'tcx> Visitor<'a, 'tcx> for UnsafetyVisitor<'a, 'tcx> { // is_like_wasm check in hir_analysis/src/collect.rs let callee_features = &self.tcx.codegen_fn_attrs(func_did).target_features; if !self.tcx.sess.target.options.is_like_wasm - && !callee_features - .iter() - .all(|feature| self.body_target_features.contains(feature)) + && !callee_features.iter().all(|feature| { + self.body_target_features.iter().any(|f| f.name == feature.name) + }) { + // Don't include implicit features in the error, unless only implicit + // features are missing. + let missing_explicit_features = callee_features.iter().any(|feature| { + !feature.implied + && !self.body_target_features.iter().any(|body_feature| { + !feature.implied && body_feature.name == feature.name + }) + }); let missing: Vec<_> = callee_features .iter() .copied() - .filter(|feature| !self.body_target_features.contains(feature)) + .filter(|feature| { + !(missing_explicit_features && feature.implied) + && !self + .body_target_features + .iter() + .any(|body_feature| body_feature.name == feature.name) + }) + .map(|feature| feature.name) .collect(); let build_enabled = self .tcx diff --git a/compiler/rustc_mir_transform/src/inline.rs b/compiler/rustc_mir_transform/src/inline.rs index f30732e6aaf3b..0f012242c3738 100644 --- a/compiler/rustc_mir_transform/src/inline.rs +++ b/compiler/rustc_mir_transform/src/inline.rs @@ -479,7 +479,9 @@ impl<'tcx> Inliner<'tcx> { return Err("incompatible instruction set"); } - if callee_attrs.target_features != self.codegen_fn_attrs.target_features { + let callee_feature_names = callee_attrs.target_features.iter().map(|f| f.name); + let this_feature_names = self.codegen_fn_attrs.target_features.iter().map(|f| f.name); + if callee_feature_names.ne(this_feature_names) { // In general it is not correct to inline a callee with target features that are a // subset of the caller. This is because the callee might contain calls, and the ABI of // those calls depends on the target features of the surrounding function. By moving a diff --git a/src/tools/miri/tests/fail/function_calls/simd_feature_flag_difference.stderr b/src/tools/miri/tests/fail/function_calls/simd_feature_flag_difference.stderr index 8ebd5d2726ffd..2544421c7e8c7 100644 --- a/src/tools/miri/tests/fail/function_calls/simd_feature_flag_difference.stderr +++ b/src/tools/miri/tests/fail/function_calls/simd_feature_flag_difference.stderr @@ -1,8 +1,8 @@ -error: Undefined Behavior: calling a function that requires unavailable target features: avx, sse3, sse4.1, sse4.2, ssse3 +error: Undefined Behavior: calling a function that requires unavailable target features: avx --> $DIR/simd_feature_flag_difference.rs:LL:CC | LL | unsafe { foo(0.0, x) } - | ^^^^^^^^^^^ calling a function that requires unavailable target features: avx, sse3, sse4.1, sse4.2, ssse3 + | ^^^^^^^^^^^ calling a function that requires unavailable target features: avx | = help: this indicates a bug in the program: it performed an invalid operation, and caused Undefined Behavior = help: see https://doc.rust-lang.org/nightly/reference/behavior-considered-undefined.html for further information diff --git a/src/tools/miri/tests/fail/function_calls/target_feature.rs b/src/tools/miri/tests/fail/function_calls/target_feature.rs index 713d7c16f4c25..84e01eb4803fe 100644 --- a/src/tools/miri/tests/fail/function_calls/target_feature.rs +++ b/src/tools/miri/tests/fail/function_calls/target_feature.rs @@ -4,7 +4,7 @@ fn main() { assert!(!is_x86_feature_detected!("ssse3")); unsafe { - ssse3_fn(); //~ ERROR: calling a function that requires unavailable target features: sse3, ssse3 + ssse3_fn(); //~ ERROR: calling a function that requires unavailable target features: ssse3 } } diff --git a/src/tools/miri/tests/fail/function_calls/target_feature.stderr b/src/tools/miri/tests/fail/function_calls/target_feature.stderr index 74a862c7b681c..4d3cf6e9d3bc0 100644 --- a/src/tools/miri/tests/fail/function_calls/target_feature.stderr +++ b/src/tools/miri/tests/fail/function_calls/target_feature.stderr @@ -1,8 +1,8 @@ -error: Undefined Behavior: calling a function that requires unavailable target features: sse3, ssse3 +error: Undefined Behavior: calling a function that requires unavailable target features: ssse3 --> $DIR/target_feature.rs:LL:CC | LL | ssse3_fn(); - | ^^^^^^^^^^ calling a function that requires unavailable target features: sse3, ssse3 + | ^^^^^^^^^^ calling a function that requires unavailable target features: ssse3 | = help: this indicates a bug in the program: it performed an invalid operation, and caused Undefined Behavior = help: see https://doc.rust-lang.org/nightly/reference/behavior-considered-undefined.html for further information diff --git a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-aes-vaes.rs b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-aes-vaes.rs index 13ebe56146381..7363c75361779 100644 --- a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-aes-vaes.rs +++ b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-aes-vaes.rs @@ -7,7 +7,7 @@ //@ignore-target-s390x //@ignore-target-thumbv7em //@ignore-target-wasm32 -//@compile-flags: -C target-feature=+aes,+vaes,+avx512f,+sse4.2 +//@compile-flags: -C target-feature=+aes,+vaes,+avx512f #![feature(avx512_target_feature, stdarch_x86_avx512)] diff --git a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx.rs b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx.rs index 18693a2a97d11..728f57d48f17e 100644 --- a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx.rs +++ b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx.rs @@ -7,7 +7,7 @@ //@ignore-target-s390x //@ignore-target-thumbv7em //@ignore-target-wasm32 -//@compile-flags: -C target-feature=+avx,+sse4.2 +//@compile-flags: -C target-feature=+avx #[cfg(target_arch = "x86")] use std::arch::x86::*; diff --git a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx2.rs b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx2.rs index f42ff6dfb51b2..80d125bb85650 100644 --- a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx2.rs +++ b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx2.rs @@ -7,7 +7,7 @@ //@ignore-target-s390x //@ignore-target-thumbv7em //@ignore-target-wasm32 -//@compile-flags: -C target-feature=+avx2,+sse4.2 +//@compile-flags: -C target-feature=+avx2 #[cfg(target_arch = "x86")] use std::arch::x86::*; diff --git a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx512.rs b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx512.rs index 7d25854ccc075..66bfcb20f1c99 100644 --- a/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx512.rs +++ b/src/tools/miri/tests/pass/shims/x86/intrinsics-x86-avx512.rs @@ -7,7 +7,7 @@ //@ignore-target-s390x //@ignore-target-thumbv7em //@ignore-target-wasm32 -//@compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bitalg,+avx512vpopcntdq,+sse4.2 +//@compile-flags: -C target-feature=+avx512f,+avx512vl,+avx512bitalg,+avx512vpopcntdq #![feature(avx512_target_feature)] #![feature(stdarch_x86_avx512)] diff --git a/tests/ui/consts/const-eval/const_fn_target_feature.stderr b/tests/ui/consts/const-eval/const_fn_target_feature.stderr index ad40d733546af..d3a00b57ebb50 100644 --- a/tests/ui/consts/const-eval/const_fn_target_feature.stderr +++ b/tests/ui/consts/const-eval/const_fn_target_feature.stderr @@ -2,7 +2,7 @@ error[E0080]: evaluation of constant value failed --> $DIR/const_fn_target_feature.rs:11:24 | LL | const B: () = unsafe { avx2_fn() }; - | ^^^^^^^^^ calling a function that requires unavailable target features: avx, avx2, sse4.1, sse4.2 + | ^^^^^^^^^ calling a function that requires unavailable target features: avx2 error: aborting due to 1 previous error diff --git a/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.stderr b/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.stderr index c2227f8e84783..1ddf05b40a606 100644 --- a/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.stderr +++ b/tests/ui/rfcs/rfc-2396-target_feature-11/safe-calls.stderr @@ -4,8 +4,8 @@ error[E0133]: call to function `sse2` with `#[target_feature]` is unsafe and req LL | sse2(); | ^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: sse and sse2 - = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target feature: sse2 + = note: the sse2 target feature being enabled in the build configuration does not remove the requirement to list it in `#[target_feature]` error[E0133]: call to function `avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block --> $DIR/safe-calls.rs:29:5 @@ -13,8 +13,7 @@ error[E0133]: call to function `avx_bmi2` with `#[target_feature]` is unsafe and LL | avx_bmi2(); | ^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: avx, sse, sse2, sse3, sse4.1, sse4.2, ssse3, and bmi2 - = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target features: avx and bmi2 error[E0133]: call to function `Quux::avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block --> $DIR/safe-calls.rs:31:5 @@ -22,8 +21,7 @@ error[E0133]: call to function `Quux::avx_bmi2` with `#[target_feature]` is unsa LL | Quux.avx_bmi2(); | ^^^^^^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: avx, sse, sse2, sse3, sse4.1, sse4.2, ssse3, and bmi2 - = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target features: avx and bmi2 error[E0133]: call to function `avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block --> $DIR/safe-calls.rs:38:5 @@ -31,7 +29,7 @@ error[E0133]: call to function `avx_bmi2` with `#[target_feature]` is unsafe and LL | avx_bmi2(); | ^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: avx, sse3, sse4.1, sse4.2, ssse3, and bmi2 + = help: in order for the call to be safe, the context requires the following additional target features: avx and bmi2 error[E0133]: call to function `Quux::avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block --> $DIR/safe-calls.rs:40:5 @@ -39,7 +37,7 @@ error[E0133]: call to function `Quux::avx_bmi2` with `#[target_feature]` is unsa LL | Quux.avx_bmi2(); | ^^^^^^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: avx, sse3, sse4.1, sse4.2, ssse3, and bmi2 + = help: in order for the call to be safe, the context requires the following additional target features: avx and bmi2 error[E0133]: call to function `avx_bmi2` with `#[target_feature]` is unsafe and requires unsafe function or block --> $DIR/safe-calls.rs:47:5 @@ -63,8 +61,8 @@ error[E0133]: call to function `sse2` with `#[target_feature]` is unsafe and req LL | const _: () = sse2(); | ^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: sse and sse2 - = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target feature: sse2 + = note: the sse2 target feature being enabled in the build configuration does not remove the requirement to list it in `#[target_feature]` error[E0133]: call to function `sse2_and_fxsr` with `#[target_feature]` is unsafe and requires unsafe function or block --> $DIR/safe-calls.rs:64:15 @@ -72,8 +70,8 @@ error[E0133]: call to function `sse2_and_fxsr` with `#[target_feature]` is unsaf LL | const _: () = sse2_and_fxsr(); | ^^^^^^^^^^^^^^^ call to function with `#[target_feature]` | - = help: in order for the call to be safe, the context requires the following additional target features: sse, sse2, and fxsr - = note: the fxsr, sse, and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target features: sse2 and fxsr + = note: the fxsr and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` error[E0133]: call to function `sse2` with `#[target_feature]` is unsafe and requires unsafe block --> $DIR/safe-calls.rs:69:5 @@ -82,8 +80,8 @@ LL | sse2(); | ^^^^^^ call to function with `#[target_feature]` | = note: for more information, see issue #71668 - = help: in order for the call to be safe, the context requires the following additional target features: sse and sse2 - = note: the sse and sse2 target features being enabled in the build configuration does not remove the requirement to list them in `#[target_feature]` + = help: in order for the call to be safe, the context requires the following additional target feature: sse2 + = note: the sse2 target feature being enabled in the build configuration does not remove the requirement to list it in `#[target_feature]` note: an unsafe function restricts its caller, but its body is safe by default --> $DIR/safe-calls.rs:68:1 | From 7a78383a5e8cb93561afbbc21aa5b9f07d9c4534 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Mon, 5 Aug 2024 10:31:03 -0400 Subject: [PATCH 24/28] Fix typo Co-authored-by: Amanieu d'Antras --- compiler/rustc_target/src/target_features.rs | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/compiler/rustc_target/src/target_features.rs b/compiler/rustc_target/src/target_features.rs index 746926308b6b4..da66ba270b33c 100644 --- a/compiler/rustc_target/src/target_features.rs +++ b/compiler/rustc_target/src/target_features.rs @@ -335,7 +335,7 @@ const RISCV_ALLOWED_FEATURES: &[(&str, Stability, ImpliedFeatures)] = &[ ("zkne", Stable, &[]), ("zknh", Stable, &[]), ("zkr", Stable, &[]), - ("zks", Stable, &["zbkb", "bzkc", "zbkx", "zksed", "zksh"]), + ("zks", Stable, &["zbkb", "zbkc", "zbkx", "zksed", "zksh"]), ("zksed", Stable, &[]), ("zksh", Stable, &[]), ("zkt", Stable, &[]), From 4fad881d4172309e2a604684a392c88d2ac6e2a9 Mon Sep 17 00:00:00 2001 From: Caleb Zulawski Date: Tue, 6 Aug 2024 00:35:32 -0400 Subject: [PATCH 25/28] Disallow enabling features without their implied features --- compiler/rustc_codegen_llvm/src/llvm_util.rs | 6 ++++-- compiler/rustc_codegen_ssa/src/target_features.rs | 8 ++++---- compiler/rustc_const_eval/src/interpret/terminator.rs | 8 +------- compiler/rustc_middle/src/query/mod.rs | 2 +- compiler/rustc_mir_build/src/check_unsafety.rs | 10 +--------- 5 files changed, 11 insertions(+), 23 deletions(-) diff --git a/compiler/rustc_codegen_llvm/src/llvm_util.rs b/compiler/rustc_codegen_llvm/src/llvm_util.rs index 1a80824a3b70d..9fd8ca43789dd 100644 --- a/compiler/rustc_codegen_llvm/src/llvm_util.rs +++ b/compiler/rustc_codegen_llvm/src/llvm_util.rs @@ -277,7 +277,7 @@ pub fn check_tied_features( /// Used to generate cfg variables and apply features /// Must express features in the way Rust understands them pub fn target_features(sess: &Session, allow_unstable: bool) -> Vec { - let mut features = FxHashSet::default(); + let mut features = vec![]; // Add base features for the target let target_machine = create_informational_target_machine(sess, true); @@ -313,7 +313,9 @@ pub fn target_features(sess: &Session, allow_unstable: bool) -> Vec { if enabled { features.extend(sess.target.implied_target_features(std::iter::once(feature))); } else { - features.remove(&feature); + features.retain(|f| { + !sess.target.implied_target_features(std::iter::once(*f)).contains(&feature) + }); } } diff --git a/compiler/rustc_codegen_ssa/src/target_features.rs b/compiler/rustc_codegen_ssa/src/target_features.rs index 145b1ece23007..cf8f7fa25d856 100644 --- a/compiler/rustc_codegen_ssa/src/target_features.rs +++ b/compiler/rustc_codegen_ssa/src/target_features.rs @@ -1,7 +1,7 @@ use rustc_ast::ast; use rustc_attr::InstructionSetAttr; use rustc_data_structures::fx::FxIndexSet; -use rustc_data_structures::unord::{ExtendUnord, UnordMap, UnordSet}; +use rustc_data_structures::unord::{UnordMap, UnordSet}; use rustc_errors::Applicability; use rustc_hir::def::DefKind; use rustc_hir::def_id::{DefId, LocalDefId, LOCAL_CRATE}; @@ -108,8 +108,7 @@ pub fn from_target_feature( // Add implied features let mut implied_target_features = UnordSet::new(); for feature in added_target_features.iter() { - implied_target_features - .extend_unord(tcx.implied_target_features(*feature).clone().into_items()); + implied_target_features.extend(tcx.implied_target_features(*feature).clone()); } for feature in added_target_features.iter() { implied_target_features.remove(feature); @@ -179,7 +178,8 @@ pub(crate) fn provide(providers: &mut Providers) { } }, implied_target_features: |tcx, feature| { - tcx.sess.target.implied_target_features(std::iter::once(feature)).into() + UnordSet::from(tcx.sess.target.implied_target_features(std::iter::once(feature))) + .into_sorted_stable_ord() }, asm_target_features, ..*providers diff --git a/compiler/rustc_const_eval/src/interpret/terminator.rs b/compiler/rustc_const_eval/src/interpret/terminator.rs index 63cb4e633f401..22cc9c65bc987 100644 --- a/compiler/rustc_const_eval/src/interpret/terminator.rs +++ b/compiler/rustc_const_eval/src/interpret/terminator.rs @@ -1000,18 +1000,12 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> { .iter() .any(|feature| !self.tcx.sess.target_features.contains(&feature.name)) { - // Don't include implicit features in the error, unless only implicit features are - // missing. This should be rare, because it can only happen when an implicit feature - // is disabled, e.g. `+avx2,-avx` - let missing_explicit_features = attrs.target_features.iter().any(|feature| { - !feature.implied && !self.tcx.sess.target_features.contains(&feature.name) - }); throw_ub_custom!( fluent::const_eval_unavailable_target_features_for_fn, unavailable_feats = attrs .target_features .iter() - .filter(|&feature| !(missing_explicit_features && feature.implied) + .filter(|&feature| !feature.implied && !self.tcx.sess.target_features.contains(&feature.name)) .fold(String::new(), |mut s, feature| { if !s.is_empty() { diff --git a/compiler/rustc_middle/src/query/mod.rs b/compiler/rustc_middle/src/query/mod.rs index b6a2943265034..5b114c9515c19 100644 --- a/compiler/rustc_middle/src/query/mod.rs +++ b/compiler/rustc_middle/src/query/mod.rs @@ -2183,7 +2183,7 @@ rustc_queries! { desc { "looking up supported target features" } } - query implied_target_features(feature: Symbol) -> &'tcx UnordSet { + query implied_target_features(feature: Symbol) -> &'tcx Vec { arena_cache eval_always desc { "looking up implied target features" } diff --git a/compiler/rustc_mir_build/src/check_unsafety.rs b/compiler/rustc_mir_build/src/check_unsafety.rs index 0b4f0632f2b36..54a4204da71e8 100644 --- a/compiler/rustc_mir_build/src/check_unsafety.rs +++ b/compiler/rustc_mir_build/src/check_unsafety.rs @@ -447,19 +447,11 @@ impl<'a, 'tcx> Visitor<'a, 'tcx> for UnsafetyVisitor<'a, 'tcx> { self.body_target_features.iter().any(|f| f.name == feature.name) }) { - // Don't include implicit features in the error, unless only implicit - // features are missing. - let missing_explicit_features = callee_features.iter().any(|feature| { - !feature.implied - && !self.body_target_features.iter().any(|body_feature| { - !feature.implied && body_feature.name == feature.name - }) - }); let missing: Vec<_> = callee_features .iter() .copied() .filter(|feature| { - !(missing_explicit_features && feature.implied) + !feature.implied && !self .body_target_features .iter() From 342b807e1a344c01a10e5e591ad22cdc56911b39 Mon Sep 17 00:00:00 2001 From: Oneirical Date: Thu, 25 Jul 2024 11:43:52 -0400 Subject: [PATCH 26/28] rewrite cross-lang-lto-upstream-rlibs to rmake --- .../tidy/src/allowed_run_make_makefiles.txt | 1 - .../cross-lang-lto-upstream-rlibs/Makefile | 32 ---------- .../cross-lang-lto-upstream-rlibs/rmake.rs | 61 +++++++++++++++++++ 3 files changed, 61 insertions(+), 33 deletions(-) delete mode 100644 tests/run-make/cross-lang-lto-upstream-rlibs/Makefile create mode 100644 tests/run-make/cross-lang-lto-upstream-rlibs/rmake.rs diff --git a/src/tools/tidy/src/allowed_run_make_makefiles.txt b/src/tools/tidy/src/allowed_run_make_makefiles.txt index a2cfdea712e7c..b248b96d8e173 100644 --- a/src/tools/tidy/src/allowed_run_make_makefiles.txt +++ b/src/tools/tidy/src/allowed_run_make_makefiles.txt @@ -1,6 +1,5 @@ run-make/branch-protection-check-IBT/Makefile run-make/cat-and-grep-sanity-check/Makefile -run-make/cross-lang-lto-upstream-rlibs/Makefile run-make/dep-info-doesnt-run-much/Makefile run-make/dep-info-spaces/Makefile run-make/dep-info/Makefile diff --git a/tests/run-make/cross-lang-lto-upstream-rlibs/Makefile b/tests/run-make/cross-lang-lto-upstream-rlibs/Makefile deleted file mode 100644 index 6f1caa31a8061..0000000000000 --- a/tests/run-make/cross-lang-lto-upstream-rlibs/Makefile +++ /dev/null @@ -1,32 +0,0 @@ -include ../tools.mk - -# ignore windows due to libLLVM being present in PATH and the PATH and library path being the same -# (so fixing it is harder). See #57765 for context -ifndef IS_WINDOWS - -# This test makes sure that we don't loose upstream object files when compiling -# staticlibs with -C linker-plugin-lto - -all: staticlib.rs upstream.rs - $(RUSTC) upstream.rs -C linker-plugin-lto -Ccodegen-units=1 - - # Check No LTO - $(RUSTC) staticlib.rs -C linker-plugin-lto -Ccodegen-units=1 -L. -o $(TMPDIR)/staticlib.a - (cd $(TMPDIR); "$(LLVM_BIN_DIR)"/llvm-ar x ./staticlib.a) - # Make sure the upstream object file was included - ls $(TMPDIR)/upstream.*.rcgu.o - - # Cleanup - rm $(TMPDIR)/* - - # Check ThinLTO - $(RUSTC) upstream.rs -C linker-plugin-lto -Ccodegen-units=1 -Clto=thin - $(RUSTC) staticlib.rs -C linker-plugin-lto -Ccodegen-units=1 -Clto=thin -L. -o $(TMPDIR)/staticlib.a - (cd $(TMPDIR); "$(LLVM_BIN_DIR)"/llvm-ar x ./staticlib.a) - ls $(TMPDIR)/upstream.*.rcgu.o - -else - -all: - -endif diff --git a/tests/run-make/cross-lang-lto-upstream-rlibs/rmake.rs b/tests/run-make/cross-lang-lto-upstream-rlibs/rmake.rs new file mode 100644 index 0000000000000..964be01bc051a --- /dev/null +++ b/tests/run-make/cross-lang-lto-upstream-rlibs/rmake.rs @@ -0,0 +1,61 @@ +// When using the flag -C linker-plugin-lto, static libraries could lose their upstream object +// files during compilation. This bug was fixed in #53031, and this test compiles a staticlib +// dependent on upstream, checking that the upstream object file still exists after no LTO and +// thin LTO. +// See https://github.com/rust-lang/rust/pull/53031 + +// ignore windows due to libLLVM being present in PATH and the PATH and library path being the same +// (so fixing it is harder). See #57765 for context +//FIXME(Oneirical): ignore-windows + +use run_make_support::{ + cwd, has_extension, has_prefix, has_suffix, llvm_ar, rfs, rustc, shallow_find_files, + static_lib_name, +}; + +fn main() { + // The test starts with no LTO enabled. + rustc().input("upstream.rs").arg("-Clinker-plugin-lto").codegen_units(1).run(); + rustc() + .input("staticlib.rs") + .arg("-Clinker-plugin-lto") + .codegen_units(1) + .output(static_lib_name("staticlib")) + .run(); + llvm_ar().arg("x").arg(static_lib_name("staticlib")).run(); + // Ensure the upstream object file was included. + assert_eq!( + shallow_find_files(cwd(), |path| { + has_prefix(path, "upstream.") && has_suffix(path, ".rcgu.o") + }) + .len(), + 1 + ); + // Remove all output files that are not source Rust code for cleanup. + for file in shallow_find_files(cwd(), |path| !has_extension(path, "rs")) { + rfs::remove_file(file) + } + + // Check it again, with Thin LTO. + rustc() + .input("upstream.rs") + .arg("-Clinker-plugin-lto") + .codegen_units(1) + .arg("-Clto=thin") + .run(); + rustc() + .input("staticlib.rs") + .arg("-Clinker-plugin-lto") + .codegen_units(1) + .arg("-Clto=thin") + .output(static_lib_name("staticlib")) + .run(); + llvm_ar().arg("x").arg(static_lib_name("staticlib")).run(); + assert_eq!( + shallow_find_files(cwd(), |path| { + has_prefix(path, "upstream.") && has_suffix(path, ".rcgu.o") + }) + .len(), + 1 + ); +} From fe6feb8c6e55667816dde1ba6fc658a48b77f134 Mon Sep 17 00:00:00 2001 From: Oneirical Date: Thu, 25 Jul 2024 13:07:26 -0400 Subject: [PATCH 27/28] rewrite long-linker-command-lines to rmake --- .../tidy/src/allowed_run_make_makefiles.txt | 1 - .../long-linker-command-lines/Makefile | 8 -------- .../run-make/long-linker-command-lines/foo.rs | 16 ++-------------- .../long-linker-command-lines/rmake.rs | 19 +++++++++++++++++++ 4 files changed, 21 insertions(+), 23 deletions(-) delete mode 100644 tests/run-make/long-linker-command-lines/Makefile create mode 100644 tests/run-make/long-linker-command-lines/rmake.rs diff --git a/src/tools/tidy/src/allowed_run_make_makefiles.txt b/src/tools/tidy/src/allowed_run_make_makefiles.txt index b248b96d8e173..a67474d998fe6 100644 --- a/src/tools/tidy/src/allowed_run_make_makefiles.txt +++ b/src/tools/tidy/src/allowed_run_make_makefiles.txt @@ -13,7 +13,6 @@ run-make/libtest-json/Makefile run-make/libtest-junit/Makefile run-make/libtest-thread-limit/Makefile run-make/long-linker-command-lines-cmd-exe/Makefile -run-make/long-linker-command-lines/Makefile run-make/macos-deployment-target/Makefile run-make/min-global-align/Makefile run-make/native-link-modifier-bundle/Makefile diff --git a/tests/run-make/long-linker-command-lines/Makefile b/tests/run-make/long-linker-command-lines/Makefile deleted file mode 100644 index b573038e344aa..0000000000000 --- a/tests/run-make/long-linker-command-lines/Makefile +++ /dev/null @@ -1,8 +0,0 @@ -# ignore-cross-compile -include ../tools.mk - -export LD_LIBRARY_PATH := $(HOST_RPATH_DIR) - -all: - $(RUSTC) foo.rs -g -O - RUSTC="$(RUSTC_ORIGINAL)" $(call RUN,foo) diff --git a/tests/run-make/long-linker-command-lines/foo.rs b/tests/run-make/long-linker-command-lines/foo.rs index 9d4a701ad8769..4f0ff30e3c073 100644 --- a/tests/run-make/long-linker-command-lines/foo.rs +++ b/tests/run-make/long-linker-command-lines/foo.rs @@ -1,12 +1,3 @@ -// This is a test which attempts to blow out the system limit with how many -// arguments can be passed to a process. This'll successively call rustc with -// larger and larger argument lists in an attempt to find one that's way too -// big for the system at hand. This file itself is then used as a "linker" to -// detect when the process creation succeeds. -// -// Eventually we should see an argument that looks like `@` as we switch from -// passing literal arguments to passing everything in the file. - use std::collections::HashSet; use std::env; use std::fs::{self, File}; @@ -43,8 +34,7 @@ fn read_linker_args(path: &Path) -> String { } fn main() { - let tmpdir = PathBuf::from(env::var_os("TMPDIR").unwrap()); - let ok = tmpdir.join("ok"); + let ok = PathBuf::from("ok"); if env::var("YOU_ARE_A_LINKER").is_ok() { if let Some(file) = env::args_os().find(|a| a.to_string_lossy().contains("@")) { let file = file.to_str().expect("non-utf8 file argument"); @@ -57,7 +47,7 @@ fn main() { let me_as_linker = format!("linker={}", env::current_exe().unwrap().display()); for i in (1..).map(|i| i * 100) { println!("attempt: {}", i); - let file = tmpdir.join("bar.rs"); + let file = PathBuf::from("bar.rs"); let mut expected_libs = write_test_case(&file, i); drop(fs::remove_file(&ok)); @@ -65,8 +55,6 @@ fn main() { .arg(&file) .arg("-C") .arg(&me_as_linker) - .arg("--out-dir") - .arg(&tmpdir) .env("YOU_ARE_A_LINKER", "1") .output() .unwrap(); diff --git a/tests/run-make/long-linker-command-lines/rmake.rs b/tests/run-make/long-linker-command-lines/rmake.rs new file mode 100644 index 0000000000000..e832d7f03e27e --- /dev/null +++ b/tests/run-make/long-linker-command-lines/rmake.rs @@ -0,0 +1,19 @@ +// This is a test which attempts to blow out the system limit with how many +// arguments can be passed to a process. This'll successively call rustc with +// larger and larger argument lists in an attempt to find one that's way too +// big for the system at hand. This file itself is then used as a "linker" to +// detect when the process creation succeeds. +// +// Eventually we should see an argument that looks like `@` as we switch from +// passing literal arguments to passing everything in the file. +// See https://github.com/rust-lang/rust/issues/41190 + +//@ ignore-cross-compile +// Reason: the compiled binary is executed + +use run_make_support::{run, rustc}; + +fn main() { + rustc().input("foo.rs").arg("-g").opt().run(); + run("foo"); +} From fe4cd9aa8debdf202ca869b32958b2dfc38f6f92 Mon Sep 17 00:00:00 2001 From: Oneirical Date: Thu, 25 Jul 2024 13:11:36 -0400 Subject: [PATCH 28/28] rewrite long-linker-command-lines-cmd-exe to rmake --- .../tidy/src/allowed_run_make_makefiles.txt | 1 - .../cross-lang-lto-upstream-rlibs/rmake.rs | 8 ++---- .../Makefile | 7 ----- .../long-linker-command-lines-cmd-exe/foo.rs | 26 +++---------------- .../rmake.rs | 26 +++++++++++++++++++ .../run-make/long-linker-command-lines/foo.rs | 2 +- 6 files changed, 32 insertions(+), 38 deletions(-) delete mode 100644 tests/run-make/long-linker-command-lines-cmd-exe/Makefile create mode 100644 tests/run-make/long-linker-command-lines-cmd-exe/rmake.rs diff --git a/src/tools/tidy/src/allowed_run_make_makefiles.txt b/src/tools/tidy/src/allowed_run_make_makefiles.txt index a67474d998fe6..202113658a84e 100644 --- a/src/tools/tidy/src/allowed_run_make_makefiles.txt +++ b/src/tools/tidy/src/allowed_run_make_makefiles.txt @@ -12,7 +12,6 @@ run-make/libs-through-symlinks/Makefile run-make/libtest-json/Makefile run-make/libtest-junit/Makefile run-make/libtest-thread-limit/Makefile -run-make/long-linker-command-lines-cmd-exe/Makefile run-make/macos-deployment-target/Makefile run-make/min-global-align/Makefile run-make/native-link-modifier-bundle/Makefile diff --git a/tests/run-make/cross-lang-lto-upstream-rlibs/rmake.rs b/tests/run-make/cross-lang-lto-upstream-rlibs/rmake.rs index 964be01bc051a..f0b8fa75bee3e 100644 --- a/tests/run-make/cross-lang-lto-upstream-rlibs/rmake.rs +++ b/tests/run-make/cross-lang-lto-upstream-rlibs/rmake.rs @@ -4,10 +4,6 @@ // thin LTO. // See https://github.com/rust-lang/rust/pull/53031 -// ignore windows due to libLLVM being present in PATH and the PATH and library path being the same -// (so fixing it is harder). See #57765 for context -//FIXME(Oneirical): ignore-windows - use run_make_support::{ cwd, has_extension, has_prefix, has_suffix, llvm_ar, rfs, rustc, shallow_find_files, static_lib_name, @@ -22,7 +18,7 @@ fn main() { .codegen_units(1) .output(static_lib_name("staticlib")) .run(); - llvm_ar().arg("x").arg(static_lib_name("staticlib")).run(); + llvm_ar().extract().arg(static_lib_name("staticlib")).run(); // Ensure the upstream object file was included. assert_eq!( shallow_find_files(cwd(), |path| { @@ -50,7 +46,7 @@ fn main() { .arg("-Clto=thin") .output(static_lib_name("staticlib")) .run(); - llvm_ar().arg("x").arg(static_lib_name("staticlib")).run(); + llvm_ar().extract().arg(static_lib_name("staticlib")).run(); assert_eq!( shallow_find_files(cwd(), |path| { has_prefix(path, "upstream.") && has_suffix(path, ".rcgu.o") diff --git a/tests/run-make/long-linker-command-lines-cmd-exe/Makefile b/tests/run-make/long-linker-command-lines-cmd-exe/Makefile deleted file mode 100644 index e43aab7f8e0fd..0000000000000 --- a/tests/run-make/long-linker-command-lines-cmd-exe/Makefile +++ /dev/null @@ -1,7 +0,0 @@ -# ignore-cross-compile -include ../tools.mk - -all: - $(RUSTC) foo.rs -g - cp foo.bat $(TMPDIR)/ - OUT_DIR="$(TMPDIR)" RUSTC="$(RUSTC_ORIGINAL)" $(call RUN,foo) diff --git a/tests/run-make/long-linker-command-lines-cmd-exe/foo.rs b/tests/run-make/long-linker-command-lines-cmd-exe/foo.rs index 1d5202dcdb493..a28cc7909fefb 100644 --- a/tests/run-make/long-linker-command-lines-cmd-exe/foo.rs +++ b/tests/run-make/long-linker-command-lines-cmd-exe/foo.rs @@ -1,16 +1,3 @@ -// Like the `long-linker-command-lines` test this test attempts to blow -// a command line limit for running the linker. Unlike that test, however, -// this test is testing `cmd.exe` specifically rather than the OS. -// -// Unfortunately `cmd.exe` has a 8192 limit which is relatively small -// in the grand scheme of things and anyone sripting rustc's linker -// is probably using a `*.bat` script and is likely to hit this limit. -// -// This test uses a `foo.bat` script as the linker which just simply -// delegates back to this program. The compiler should use a lower -// limit for arguments before passing everything via `@`, which -// means that everything should still succeed here. - use std::env; use std::fs::{self, File}; use std::io::{BufWriter, Read, Write}; @@ -18,13 +5,8 @@ use std::path::PathBuf; use std::process::Command; fn main() { - if !cfg!(windows) { - return; - } - - let tmpdir = PathBuf::from(env::var_os("OUT_DIR").unwrap()); - let ok = tmpdir.join("ok"); - let not_ok = tmpdir.join("not_ok"); + let ok = PathBuf::from("ok"); + let not_ok = PathBuf::from("not_ok"); if env::var("YOU_ARE_A_LINKER").is_ok() { match env::args_os().find(|a| a.to_string_lossy().contains("@")) { Some(file) => { @@ -45,7 +27,7 @@ fn main() { for i in (1..).map(|i| i * 10) { println!("attempt: {}", i); - let file = tmpdir.join("bar.rs"); + let file = PathBuf::from("bar.rs"); let mut f = BufWriter::new(File::create(&file).unwrap()); let mut lib_name = String::new(); for _ in 0..i { @@ -63,8 +45,6 @@ fn main() { .arg(&file) .arg("-C") .arg(&bat_linker) - .arg("--out-dir") - .arg(&tmpdir) .env("YOU_ARE_A_LINKER", "1") .env("MY_LINKER", &me) .status() diff --git a/tests/run-make/long-linker-command-lines-cmd-exe/rmake.rs b/tests/run-make/long-linker-command-lines-cmd-exe/rmake.rs new file mode 100644 index 0000000000000..60ed2c5bcd203 --- /dev/null +++ b/tests/run-make/long-linker-command-lines-cmd-exe/rmake.rs @@ -0,0 +1,26 @@ +// Like the `long-linker-command-lines` test this test attempts to blow +// a command line limit for running the linker. Unlike that test, however, +// this test is testing `cmd.exe` specifically rather than the OS. +// +// Unfortunately, the maximum length of the string that you can use at the +// command prompt (`cmd.exe`) is 8191 characters. +// Anyone scripting rustc's linker +// is probably using a `*.bat` script and is likely to hit this limit. +// +// This test uses a `foo.bat` script as the linker which just simply +// delegates back to this program. The compiler should use a lower +// limit for arguments before passing everything via `@`, which +// means that everything should still succeed here. +// See https://github.com/rust-lang/rust/pull/47507 + +//@ ignore-cross-compile +// Reason: the compiled binary is executed +//@ only-windows +// Reason: this test is specific to Windows executables + +use run_make_support::{run, rustc}; + +fn main() { + rustc().input("foo.rs").arg("-g").run(); + run("foo"); +} diff --git a/tests/run-make/long-linker-command-lines/foo.rs b/tests/run-make/long-linker-command-lines/foo.rs index 4f0ff30e3c073..5b30c06fac9e0 100644 --- a/tests/run-make/long-linker-command-lines/foo.rs +++ b/tests/run-make/long-linker-command-lines/foo.rs @@ -43,7 +43,7 @@ fn main() { return; } - let rustc = env::var_os("RUSTC").unwrap_or("rustc".into()); + let rustc = env::var_os("RUSTC").unwrap(); let me_as_linker = format!("linker={}", env::current_exe().unwrap().display()); for i in (1..).map(|i| i * 100) { println!("attempt: {}", i);