From 12b1d75de6762ee208b25bc7b4f6311b5cba3683 Mon Sep 17 00:00:00 2001
From: Trevor Gross <tmgross@umich.edu>
Date: Wed, 9 Apr 2025 01:56:14 +0000
Subject: [PATCH 1/2] Replace calls to `core::arch` intrinsics with assembly

Some backends may replace calls to `core::arch` with multiple calls to
`sqrt` [1], which becomes recursive. Help mitigate this by replacing the
call with assembly.

Results in the same assembly as the current implementation when built
with optimizations.

[1]: https://github.com/rust-lang/compiler-builtins/issues/649
---
 src/math/arch/i686.rs | 35 ++++++++++++++++++++---------------
 1 file changed, 20 insertions(+), 15 deletions(-)

diff --git a/src/math/arch/i686.rs b/src/math/arch/i686.rs
index ad54d8b61..3e1d19bfa 100644
--- a/src/math/arch/i686.rs
+++ b/src/math/arch/i686.rs
@@ -1,22 +1,27 @@
 //! Architecture-specific support for x86-32 and x86-64 with SSE2
 
-#[cfg(target_arch = "x86")]
-use core::arch::x86::*;
-#[cfg(target_arch = "x86_64")]
-use core::arch::x86_64::*;
-
-pub fn sqrtf(x: f32) -> f32 {
+pub fn sqrtf(mut x: f32) -> f32 {
+    // SAFETY: `sqrtss` is part of `sse2`, which this module is gated behind. It has no memory
+    // access or side effects.
     unsafe {
-        let m = _mm_set_ss(x);
-        let m_sqrt = _mm_sqrt_ss(m);
-        _mm_cvtss_f32(m_sqrt)
-    }
+        core::arch::asm!(
+            "sqrtss {x}, {x}",
+            x = inout(xmm_reg) x,
+            options(nostack, nomem, pure),
+        )
+    };
+    x
 }
 
-pub fn sqrt(x: f64) -> f64 {
+pub fn sqrt(mut x: f64) -> f64 {
+    // SAFETY: `sqrtsd` is part of `sse2`, which this module is gated behind. It has no memory
+    // access or side effects.
     unsafe {
-        let m = _mm_set_sd(x);
-        let m_sqrt = _mm_sqrt_pd(m);
-        _mm_cvtsd_f64(m_sqrt)
-    }
+        core::arch::asm!(
+            "sqrtsd {x}, {x}",
+            x = inout(xmm_reg) x,
+            options(nostack, nomem, pure),
+        )
+    };
+    x
 }

From e222b2700e44b181aca7f2617b064556143c2360 Mon Sep 17 00:00:00 2001
From: Trevor Gross <tmgross@umich.edu>
Date: Wed, 9 Apr 2025 02:22:15 +0000
Subject: [PATCH 2/2] Resolve small errors identified by recent clippy

---
 crates/libm-test/src/precision.rs | 1 +
 src/math/support/env.rs           | 1 -
 2 files changed, 1 insertion(+), 1 deletion(-)

diff --git a/crates/libm-test/src/precision.rs b/crates/libm-test/src/precision.rs
index 8916b43ab..f5fb5f670 100644
--- a/crates/libm-test/src/precision.rs
+++ b/crates/libm-test/src/precision.rs
@@ -13,6 +13,7 @@ use crate::{BaseName, CheckBasis, CheckCtx, Float, Identifier, Int, TestResult};
 pub struct SpecialCase;
 
 /// ULP allowed to differ from the results returned by a test basis.
+#[allow(clippy::single_match)]
 pub fn default_ulp(ctx: &CheckCtx) -> u32 {
     // ULP compared to the infinite (MPFR) result.
     let mut ulp = match ctx.base_name {
diff --git a/src/math/support/env.rs b/src/math/support/env.rs
index 7244381da..c05890d98 100644
--- a/src/math/support/env.rs
+++ b/src/math/support/env.rs
@@ -70,7 +70,6 @@ impl Status {
     /// The default result for division is +/-inf based on operand sign. For `logB`, the default
     /// result is -inf.
     /// `x / y` when `x != 0.0` and `y == 0.0`,
-
     #[cfg_attr(not(feature = "unstable-public-internals"), allow(dead_code))]
     pub const DIVIDE_BY_ZERO: Self = Self(1 << 2);