Miri subtree update

Cargo.lock

@@ -2136,6 +2136,16 @@ dependencies = [
  "winapi",
 ]
 
+[[package]]
+name = "libloading"
+version = "0.8.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c571b676ddfc9a8c12f1f3d3085a7b163966a8fd8098a90640953ce5f6170161"
+dependencies = [
+ "cfg-if",
+ "windows-sys 0.48.0",
+]
+
 [[package]]
 name = "libm"
 version = "0.1.4"
@@ -2411,7 +2421,7 @@ dependencies = [
  "lazy_static",
  "libc",
  "libffi",
- "libloading",
+ "libloading 0.8.1",
  "log",
  "measureme",
  "rand",
@@ -3914,7 +3924,7 @@ dependencies = [
 name = "rustc_interface"
 version = "0.0.0"
 dependencies = [
- "libloading",
+ "libloading 0.7.4",
  "rustc-rayon",
  "rustc-rayon-core",
  "rustc_ast",
@@ -4045,7 +4055,7 @@ name = "rustc_metadata"
 version = "0.0.0"
 dependencies = [
  "bitflags 1.3.2",
- "libloading",
+ "libloading 0.7.4",
  "odht",
  "rustc_ast",
  "rustc_attr",
@@ -4260,7 +4270,7 @@ dependencies = [
 name = "rustc_plugin_impl"
 version = "0.0.0"
 dependencies = [
- "libloading",
+ "libloading 0.7.4",
  "rustc_ast",
  "rustc_errors",
  "rustc_fluent_macro",

src/tools/miri/Cargo.lock

@@ -409,12 +409,12 @@ dependencies = [
 
 [[package]]
 name = "libloading"
-version = "0.7.4"
+version = "0.8.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b67380fd3b2fbe7527a606e18729d21c6f3951633d0500574c4dc22d2d638b9f"
+checksum = "d580318f95776505201b28cf98eb1fa5e4be3b689633ba6a3e6cd880ff22d8cb"
 dependencies = [
  "cfg-if",
- "winapi",
+ "windows-sys 0.48.0",
 ]
 
 [[package]]
@@ -947,9 +947,9 @@ dependencies = [
 
 [[package]]
 name = "ui_test"
-version = "0.21.1"
+version = "0.21.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "accffe020b57a6dd50014d457b5842c5a2ca73cd84f07d86d0a19c460a6509ae"
+checksum = "aaf4bf7c184b8dfc7a4d3b90df789b1eb992ee42811cd115f32a7a1eb781058d"
 dependencies = [
  "annotate-snippets",
  "anyhow",

src/tools/miri/Cargo.toml

@@ -32,7 +32,7 @@ libc = "0.2"
 
 [target.'cfg(target_os = "linux")'.dependencies]
 libffi = "3.2.0"
-libloading = "0.7"
+libloading = "0.8"
 
 [dev-dependencies]
 colored = "2"

src/tools/miri/rust-version

@@ -1 +1 @@
-2ba4eb2d49e774b5fbc2a06258ac7b0f60b92b7e
+bb6c66be3793ac5c738eeac91ecdc4b99388d0b4

src/tools/miri/src/shims/intrinsics/mod.rs

@@ -325,7 +325,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
 
             "fmaf32" => {
                 let [a, b, c] = check_arg_count(args)?;
-                // FIXME: Using host floats, to work around https://github.com/rust-lang/miri/issues/2468.
+                // FIXME: Using host floats, to work around https://github.com/rust-lang/rustc_apfloat/issues/11
                 let a = f32::from_bits(this.read_scalar(a)?.to_u32()?);
                 let b = f32::from_bits(this.read_scalar(b)?.to_u32()?);
                 let c = f32::from_bits(this.read_scalar(c)?.to_u32()?);
@@ -335,7 +335,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
 
             "fmaf64" => {
                 let [a, b, c] = check_arg_count(args)?;
-                // FIXME: Using host floats, to work around https://github.com/rust-lang/miri/issues/2468.
+                // FIXME: Using host floats, to work around https://github.com/rust-lang/rustc_apfloat/issues/11
                 let a = f64::from_bits(this.read_scalar(a)?.to_u64()?);
                 let b = f64::from_bits(this.read_scalar(b)?.to_u64()?);
                 let c = f64::from_bits(this.read_scalar(c)?.to_u64()?);

src/tools/miri/src/shims/intrinsics/simd.rs

@@ -495,15 +495,21 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                 let (right, right_len) = this.operand_to_simd(right)?;
                 let (dest, dest_len) = this.place_to_simd(dest)?;
 
-                let index = generic_args[2].expect_const().eval(*this.tcx, this.param_env(), Some(this.tcx.span)).unwrap().unwrap_branch();
+                let index = generic_args[2]
+                    .expect_const()
+                    .eval(*this.tcx, this.param_env(), Some(this.tcx.span))
+                    .unwrap()
+                    .unwrap_branch();
                 let index_len = index.len();
 
                 assert_eq!(left_len, right_len);
                 assert_eq!(index_len as u64, dest_len);
 
                 for i in 0..dest_len {
-                    let src_index: u64 = index[i as usize].unwrap_leaf()
-                        .try_to_u32().unwrap()
+                    let src_index: u64 = index[usize::try_from(i).unwrap()]
+                        .unwrap_leaf()
+                        .try_to_u32()
+                        .unwrap()
                         .into();
                     let dest = this.project_index(&dest, i)?;
 

src/tools/miri/src/shims/x86/mod.rs

@@ -9,6 +9,8 @@ use shims::foreign_items::EmulateByNameResult;
 
 mod sse;
 mod sse2;
+mod sse3;
+mod ssse3;
 
 impl<'mir, 'tcx: 'mir> EvalContextExt<'mir, 'tcx> for crate::MiriInterpCx<'mir, 'tcx> {}
 pub(super) trait EvalContextExt<'mir, 'tcx: 'mir>:
@@ -88,6 +90,16 @@ pub(super) trait EvalContextExt<'mir, 'tcx: 'mir>:
                     this, link_name, abi, args, dest,
                 );
             }
+            name if name.starts_with("sse3.") => {
+                return sse3::EvalContextExt::emulate_x86_sse3_intrinsic(
+                    this, link_name, abi, args, dest,
+                );
+            }
+            name if name.starts_with("ssse3.") => {
+                return ssse3::EvalContextExt::emulate_x86_ssse3_intrinsic(
+                    this, link_name, abi, args, dest,
+                );
+            }
             _ => return Ok(EmulateByNameResult::NotSupported),
         }
         Ok(EmulateByNameResult::NeedsJumping)
@@ -286,3 +298,44 @@ fn bin_op_simd_float_all<'tcx, F: rustc_apfloat::Float>(
 
     Ok(())
 }
+
+/// Horizontaly performs `which` operation on adjacent values of
+/// `left` and `right` SIMD vectors and stores the result in `dest`.
+fn horizontal_bin_op<'tcx>(
+    this: &mut crate::MiriInterpCx<'_, 'tcx>,
+    which: mir::BinOp,
+    saturating: bool,
+    left: &OpTy<'tcx, Provenance>,
+    right: &OpTy<'tcx, Provenance>,
+    dest: &PlaceTy<'tcx, Provenance>,
+) -> InterpResult<'tcx, ()> {
+    let (left, left_len) = this.operand_to_simd(left)?;
+    let (right, right_len) = this.operand_to_simd(right)?;
+    let (dest, dest_len) = this.place_to_simd(dest)?;
+
+    assert_eq!(dest_len, left_len);
+    assert_eq!(dest_len, right_len);
+    assert_eq!(dest_len % 2, 0);
+
+    let middle = dest_len / 2;
+    for i in 0..dest_len {
+        // `i` is the index in `dest`
+        // `j` is the index of the 2-item chunk in `src`
+        let (j, src) =
+            if i < middle { (i, &left) } else { (i.checked_sub(middle).unwrap(), &right) };
+        // `base_i` is the index of the first item of the 2-item chunk in `src`
+        let base_i = j.checked_mul(2).unwrap();
+        let lhs = this.read_immediate(&this.project_index(src, base_i)?)?;
+        let rhs = this.read_immediate(&this.project_index(src, base_i.checked_add(1).unwrap())?)?;
+
+        let res = if saturating {
+            Immediate::from(this.saturating_arith(which, &lhs, &rhs)?)
+        } else {
+            *this.wrapping_binary_op(which, &lhs, &rhs)?
+        };
+
+        this.write_immediate(res, &this.project_index(&dest, i)?)?;
+    }
+
+    Ok(())
+}

src/tools/miri/src/shims/x86/sse2.rs

@@ -82,6 +82,42 @@ pub(super) trait EvalContextExt<'mir, 'tcx: 'mir>:
                     this.write_immediate(*res, &dest)?;
                 }
             }
+            // Used to implement the _mm_madd_epi16 function.
+            // Multiplies packed signed 16-bit integers in `left` and `right`, producing
+            // intermediate signed 32-bit integers. Horizontally add adjacent pairs of
+            // intermediate 32-bit integers, and pack the results in `dest`.
+            "pmadd.wd" => {
+                let [left, right] =
+                    this.check_shim(abi, Abi::C { unwind: false }, link_name, args)?;
+
+                let (left, left_len) = this.operand_to_simd(left)?;
+                let (right, right_len) = this.operand_to_simd(right)?;
+                let (dest, dest_len) = this.place_to_simd(dest)?;
+
+                assert_eq!(left_len, right_len);
+                assert_eq!(dest_len.checked_mul(2).unwrap(), left_len);
+
+                for i in 0..dest_len {
+                    let j1 = i.checked_mul(2).unwrap();
+                    let left1 = this.read_scalar(&this.project_index(&left, j1)?)?.to_i16()?;
+                    let right1 = this.read_scalar(&this.project_index(&right, j1)?)?.to_i16()?;
+
+                    let j2 = j1.checked_add(1).unwrap();
+                    let left2 = this.read_scalar(&this.project_index(&left, j2)?)?.to_i16()?;
+                    let right2 = this.read_scalar(&this.project_index(&right, j2)?)?.to_i16()?;
+
+                    let dest = this.project_index(&dest, i)?;
+
+                    // Multiplications are i16*i16->i32, which will not overflow.
+                    let mul1 = i32::from(left1).checked_mul(right1.into()).unwrap();
+                    let mul2 = i32::from(left2).checked_mul(right2.into()).unwrap();
+                    // However, this addition can overflow in the most extreme case
+                    // (-0x8000)*(-0x8000)+(-0x8000)*(-0x8000) = 0x80000000
+                    let res = mul1.wrapping_add(mul2);
+
+                    this.write_scalar(Scalar::from_i32(res), &dest)?;
+                }
+            }
             // Used to implement the _mm_mulhi_epi16 and _mm_mulhi_epu16 functions.
             "pmulh.w" | "pmulhu.w" => {
                 let [left, right] =

src/tools/miri/src/shims/x86/sse3.rs

@@ -0,0 +1,90 @@
+use rustc_middle::mir;
+use rustc_span::Symbol;
+use rustc_target::abi::Align;
+use rustc_target::spec::abi::Abi;
+
+use super::horizontal_bin_op;
+use crate::*;
+use shims::foreign_items::EmulateByNameResult;
+
+impl<'mir, 'tcx: 'mir> EvalContextExt<'mir, 'tcx> for crate::MiriInterpCx<'mir, 'tcx> {}
+pub(super) trait EvalContextExt<'mir, 'tcx: 'mir>:
+    crate::MiriInterpCxExt<'mir, 'tcx>
+{
+    fn emulate_x86_sse3_intrinsic(
+        &mut self,
+        link_name: Symbol,
+        abi: Abi,
+        args: &[OpTy<'tcx, Provenance>],
+        dest: &PlaceTy<'tcx, Provenance>,
+    ) -> InterpResult<'tcx, EmulateByNameResult<'mir, 'tcx>> {
+        let this = self.eval_context_mut();
+        // Prefix should have already been checked.
+        let unprefixed_name = link_name.as_str().strip_prefix("llvm.x86.sse3.").unwrap();
+
+        match unprefixed_name {
+            // Used to implement the _mm_addsub_ps and _mm_addsub_pd functions.
+            // Alternatingly add and subtract floating point (f32 or f64) from
+            // `left` and `right`
+            "addsub.ps" | "addsub.pd" => {
+                let [left, right] =
+                    this.check_shim(abi, Abi::C { unwind: false }, link_name, args)?;
+
+                let (left, left_len) = this.operand_to_simd(left)?;
+                let (right, right_len) = this.operand_to_simd(right)?;
+                let (dest, dest_len) = this.place_to_simd(dest)?;
+
+                assert_eq!(dest_len, left_len);
+                assert_eq!(dest_len, right_len);
+
+                for i in 0..dest_len {
+                    let left = this.read_immediate(&this.project_index(&left, i)?)?;
+                    let right = this.read_immediate(&this.project_index(&right, i)?)?;
+                    let dest = this.project_index(&dest, i)?;
+
+                    // Even elements are subtracted and odd elements are added.
+                    let op = if i % 2 == 0 { mir::BinOp::Sub } else { mir::BinOp::Add };
+                    let res = this.wrapping_binary_op(op, &left, &right)?;
+
+                    this.write_immediate(*res, &dest)?;
+                }
+            }
+            // Used to implement the _mm_h{add,sub}_p{s,d} functions.
+            // Horizontally add/subtract adjacent floating point values
+            // in `left` and `right`.
+            "hadd.ps" | "hadd.pd" | "hsub.ps" | "hsub.pd" => {
+                let [left, right] =
+                    this.check_shim(abi, Abi::C { unwind: false }, link_name, args)?;
+
+                let which = match unprefixed_name {
+                    "hadd.ps" | "hadd.pd" => mir::BinOp::Add,
+                    "hsub.ps" | "hsub.pd" => mir::BinOp::Sub,
+                    _ => unreachable!(),
+                };
+
+                horizontal_bin_op(this, which, /*saturating*/ false, left, right, dest)?;
+            }
+            // Used to implement the _mm_lddqu_si128 function.
+            // Reads a 128-bit vector from an unaligned pointer. This intrinsic
+            // is expected to perform better than a regular unaligned read when
+            // the data crosses a cache line, but for Miri this is just a regular
+            // unaligned read.
+            "ldu.dq" => {
+                let [src_ptr] = this.check_shim(abi, Abi::C { unwind: false }, link_name, args)?;
+                let src_ptr = this.read_pointer(src_ptr)?;
+                let dest = dest.force_mplace(this)?;
+
+                this.mem_copy(
+                    src_ptr,
+                    Align::ONE,
+                    dest.ptr(),
+                    Align::ONE,
+                    dest.layout.size,
+                    /*nonoverlapping*/ true,
+                )?;
+            }
+            _ => return Ok(EmulateByNameResult::NotSupported),
+        }
+        Ok(EmulateByNameResult::NeedsJumping)
+    }
+}