Auto merge of rust-lang#2989 - eduardosm:x86-intrinsics, r=RalfJung

miri: implement some `llvm.x86.sse.*` intrinsics and add tests

PR moved from rust-lang#113932.

Implements LLVM intrisics needed to run most SSE functions from `core::arch::x86{,_64}`.

Also adds miri tests for those functions (mostly copied from core_arch tests).

r? `@RalfJung`

The first commit is the same that the commit in the PR I had opened in the Rust repository. I addressed review comments in additional commits to make it easier to review. I also fixed formatting and clippy warnings.

Author: bors

Diff for: src/tools/miri/src/lib.rs

@@ -21,6 +21,7 @@
     clippy::enum_variant_names,
     clippy::field_reassign_with_default,
     clippy::manual_map,
+    clippy::neg_cmp_op_on_partial_ord,
     clippy::new_without_default,
     clippy::single_match,
     clippy::useless_format,

Diff for: src/tools/miri/src/shims/foreign_items.rs

@@ -942,6 +942,33 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                 this.write_scalar(Scalar::from_u64(res.to_bits()), dest)?;
             }
 
+            "llvm.prefetch" => {
+                let [p, rw, loc, ty] =
+                    this.check_shim(abi, Abi::C { unwind: false }, link_name, args)?;
+
+                let _ = this.read_pointer(p)?;
+                let rw = this.read_scalar(rw)?.to_i32()?;
+                let loc = this.read_scalar(loc)?.to_i32()?;
+                let ty = this.read_scalar(ty)?.to_i32()?;
+
+                if ty == 1 {
+                    // Data cache prefetch.
+                    // Notably, we do not have to check the pointer, this operation is never UB!
+
+                    if !matches!(rw, 0 | 1) {
+                        throw_unsup_format!("invalid `rw` value passed to `llvm.prefetch`: {}", rw);
+                    }
+                    if !matches!(loc, 0..=3) {
+                        throw_unsup_format!(
+                            "invalid `loc` value passed to `llvm.prefetch`: {}",
+                            loc
+                        );
+                    }
+                } else {
+                    throw_unsup_format!("unsupported `llvm.prefetch` type argument: {}", ty);
+                }
+            }
+
             // Architecture-specific shims
             "llvm.x86.addcarry.64" if this.tcx.sess.target.arch == "x86_64" => {
                 // Computes u8+u64+u64, returning tuple (u8,u64) comprising the output carry and truncated sum.
@@ -994,6 +1021,12 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                 }
             }
 
+            name if name.starts_with("llvm.x86.sse.") => {
+                return shims::x86::sse::EvalContextExt::emulate_x86_sse_intrinsic(
+                    this, link_name, abi, args, dest,
+                );
+            }
+
             // Platform-specific shims
             _ =>
                 return match this.tcx.sess.target.os.as_ref() {

Diff for: src/tools/miri/src/shims/mod.rs

@@ -7,6 +7,7 @@ pub mod foreign_items;
 pub mod intrinsics;
 pub mod unix;
 pub mod windows;
+mod x86;
 
 pub mod dlsym;
 pub mod env;

Diff for: src/tools/miri/src/shims/x86/mod.rs

@@ -0,0 +1 @@
+pub(super) mod sse;