Auto merge of #50521 - gnzlbg:simd_float, r=alexcrichton

Add simd math intrinsics and gather/scatter

This PR adds simd math intrinsics for floating-point vectors (sqrt, sin, cos, pow, exp, log, fma, abs, etc.) and the generic simd gather/scatter intrinsics.

src/librustc_codegen_llvm/context.rs

@@ -522,6 +522,15 @@ fn declare_intrinsic(cx: &CodegenCx, key: &str) -> Option<ValueRef> {
     let t_f32 = Type::f32(cx);
     let t_f64 = Type::f64(cx);
 
+    let t_v2f32 = Type::vector(&t_f32, 2);
+    let t_v4f32 = Type::vector(&t_f32, 4);
+    let t_v8f32 = Type::vector(&t_f32, 8);
+    let t_v16f32 = Type::vector(&t_f32, 16);
+
+    let t_v2f64 = Type::vector(&t_f64, 2);
+    let t_v4f64 = Type::vector(&t_f64, 4);
+    let t_v8f64 = Type::vector(&t_f64, 8);
+
     ifn!("llvm.memcpy.p0i8.p0i8.i16", fn(i8p, i8p, t_i16, t_i32, i1) -> void);
     ifn!("llvm.memcpy.p0i8.p0i8.i32", fn(i8p, i8p, t_i32, t_i32, i1) -> void);
     ifn!("llvm.memcpy.p0i8.p0i8.i64", fn(i8p, i8p, t_i64, t_i32, i1) -> void);
@@ -537,37 +546,145 @@ fn declare_intrinsic(cx: &CodegenCx, key: &str) -> Option<ValueRef> {
     ifn!("llvm.frameaddress", fn(t_i32) -> i8p);
 
     ifn!("llvm.powi.f32", fn(t_f32, t_i32) -> t_f32);
+    ifn!("llvm.powi.v2f32", fn(t_v2f32, t_i32) -> t_v2f32);
+    ifn!("llvm.powi.v4f32", fn(t_v4f32, t_i32) -> t_v4f32);
+    ifn!("llvm.powi.v8f32", fn(t_v8f32, t_i32) -> t_v8f32);
+    ifn!("llvm.powi.v16f32", fn(t_v16f32, t_i32) -> t_v16f32);
     ifn!("llvm.powi.f64", fn(t_f64, t_i32) -> t_f64);
+    ifn!("llvm.powi.v2f64", fn(t_v2f64, t_i32) -> t_v2f64);
+    ifn!("llvm.powi.v4f64", fn(t_v4f64, t_i32) -> t_v4f64);
+    ifn!("llvm.powi.v8f64", fn(t_v8f64, t_i32) -> t_v8f64);
+
     ifn!("llvm.pow.f32", fn(t_f32, t_f32) -> t_f32);
+    ifn!("llvm.pow.v2f32", fn(t_v2f32, t_v2f32) -> t_v2f32);
+    ifn!("llvm.pow.v4f32", fn(t_v4f32, t_v4f32) -> t_v4f32);
+    ifn!("llvm.pow.v8f32", fn(t_v8f32, t_v8f32) -> t_v8f32);
+    ifn!("llvm.pow.v16f32", fn(t_v16f32, t_v16f32) -> t_v16f32);
     ifn!("llvm.pow.f64", fn(t_f64, t_f64) -> t_f64);
+    ifn!("llvm.pow.v2f64", fn(t_v2f64, t_v2f64) -> t_v2f64);
+    ifn!("llvm.pow.v4f64", fn(t_v4f64, t_v4f64) -> t_v4f64);
+    ifn!("llvm.pow.v8f64", fn(t_v8f64, t_v8f64) -> t_v8f64);
 
     ifn!("llvm.sqrt.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.sqrt.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.sqrt.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.sqrt.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.sqrt.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.sqrt.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.sqrt.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.sqrt.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.sqrt.v8f64", fn(t_v8f64) -> t_v8f64);
+
     ifn!("llvm.sin.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.sin.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.sin.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.sin.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.sin.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.sin.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.sin.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.sin.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.sin.v8f64", fn(t_v8f64) -> t_v8f64);
+
     ifn!("llvm.cos.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.cos.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.cos.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.cos.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.cos.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.cos.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.cos.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.cos.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.cos.v8f64", fn(t_v8f64) -> t_v8f64);
+
     ifn!("llvm.exp.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.exp.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.exp.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.exp.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.exp.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.exp.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.exp.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.exp.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.exp.v8f64", fn(t_v8f64) -> t_v8f64);
+
     ifn!("llvm.exp2.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.exp2.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.exp2.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.exp2.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.exp2.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.exp2.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.exp2.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.exp2.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.exp2.v8f64", fn(t_v8f64) -> t_v8f64);
+
     ifn!("llvm.log.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.log.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.log.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.log.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.log.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.log.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.log.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.log.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.log.v8f64", fn(t_v8f64) -> t_v8f64);
+
     ifn!("llvm.log10.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.log10.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.log10.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.log10.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.log10.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.log10.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.log10.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.log10.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.log10.v8f64", fn(t_v8f64) -> t_v8f64);
+
     ifn!("llvm.log2.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.log2.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.log2.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.log2.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.log2.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.log2.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.log2.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.log2.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.log2.v8f64", fn(t_v8f64) -> t_v8f64);
 
     ifn!("llvm.fma.f32", fn(t_f32, t_f32, t_f32) -> t_f32);
+    ifn!("llvm.fma.v2f32", fn(t_v2f32, t_v2f32, t_v2f32) -> t_v2f32);
+    ifn!("llvm.fma.v4f32", fn(t_v4f32, t_v4f32, t_v4f32) -> t_v4f32);
+    ifn!("llvm.fma.v8f32", fn(t_v8f32, t_v8f32, t_v8f32) -> t_v8f32);
+    ifn!("llvm.fma.v16f32", fn(t_v16f32, t_v16f32, t_v16f32) -> t_v16f32);
     ifn!("llvm.fma.f64", fn(t_f64, t_f64, t_f64) -> t_f64);
+    ifn!("llvm.fma.v2f64", fn(t_v2f64, t_v2f64, t_v2f64) -> t_v2f64);
+    ifn!("llvm.fma.v4f64", fn(t_v4f64, t_v4f64, t_v4f64) -> t_v4f64);
+    ifn!("llvm.fma.v8f64", fn(t_v8f64, t_v8f64, t_v8f64) -> t_v8f64);
 
     ifn!("llvm.fabs.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.fabs.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.fabs.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.fabs.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.fabs.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.fabs.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.fabs.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.fabs.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.fabs.v8f64", fn(t_v8f64) -> t_v8f64);
 
     ifn!("llvm.floor.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.floor.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.floor.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.floor.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.floor.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.floor.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.floor.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.floor.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.floor.v8f64", fn(t_v8f64) -> t_v8f64);
+
     ifn!("llvm.ceil.f32", fn(t_f32) -> t_f32);
+    ifn!("llvm.ceil.v2f32", fn(t_v2f32) -> t_v2f32);
+    ifn!("llvm.ceil.v4f32", fn(t_v4f32) -> t_v4f32);
+    ifn!("llvm.ceil.v8f32", fn(t_v8f32) -> t_v8f32);
+    ifn!("llvm.ceil.v16f32", fn(t_v16f32) -> t_v16f32);
     ifn!("llvm.ceil.f64", fn(t_f64) -> t_f64);
+    ifn!("llvm.ceil.v2f64", fn(t_v2f64) -> t_v2f64);
+    ifn!("llvm.ceil.v4f64", fn(t_v4f64) -> t_v4f64);
+    ifn!("llvm.ceil.v8f64", fn(t_v8f64) -> t_v8f64);
+
     ifn!("llvm.trunc.f32", fn(t_f32) -> t_f32);
     ifn!("llvm.trunc.f64", fn(t_f64) -> t_f64);
 

src/librustc_codegen_llvm/intrinsic.rs

@@ -1140,6 +1140,359 @@ fn generic_simd_intrinsic<'a, 'tcx>(
         return Ok(bx.select(m_i1s, args[1].immediate(), args[2].immediate()));
     }
 
+    fn simd_simple_float_intrinsic<'a, 'tcx>(name: &str,
+                                             in_elem: &::rustc::ty::TyS,
+                                             in_ty: &::rustc::ty::TyS,
+                                             in_len: usize,
+                                             bx: &Builder<'a, 'tcx>,
+                                             span: Span,
+                                             args: &[OperandRef<'tcx>])
+                                             -> Result<ValueRef, ()> {
+        macro_rules! emit_error {
+            ($msg: tt) => {
+                emit_error!($msg, )
+            };
+            ($msg: tt, $($fmt: tt)*) => {
+                span_invalid_monomorphization_error(
+                    bx.sess(), span,
+                    &format!(concat!("invalid monomorphization of `{}` intrinsic: ",
+                                     $msg),
+                             name, $($fmt)*));
+            }
+        }
+        macro_rules! return_error {
+            ($($fmt: tt)*) => {
+                {
+                    emit_error!($($fmt)*);
+                    return Err(());
+                }
+            }
+        }
+        let ety = match in_elem.sty {
+            ty::TyFloat(f) if f.bit_width() == 32 => {
+                if in_len < 2 || in_len > 16 {
+                    return_error!(
+                        "unsupported floating-point vector `{}` with length `{}` \
+                         out-of-range [2, 16]",
+                        in_ty, in_len);
+                }
+                "f32"
+            },
+            ty::TyFloat(f) if f.bit_width() == 64 => {
+                if in_len < 2 || in_len > 8 {
+                    return_error!("unsupported floating-point vector `{}` with length `{}` \
+                                   out-of-range [2, 8]",
+                                  in_ty, in_len);
+                }
+                "f64"
+            },
+            ty::TyFloat(f) => {
+                return_error!("unsupported element type `{}` of floating-point vector `{}`",
+                              f, in_ty);
+            },
+            _ => {
+                return_error!("`{}` is not a floating-point type", in_ty);
+            }
+        };
+
+        let llvm_name = &format!("llvm.{0}.v{1}{2}", name, in_len, ety);
+        let intrinsic = bx.cx.get_intrinsic(&llvm_name);
+        let c = bx.call(intrinsic,
+                        &args.iter().map(|arg| arg.immediate()).collect::<Vec<_>>(),
+                        None);
+        unsafe { llvm::LLVMRustSetHasUnsafeAlgebra(c) };
+        return Ok(c);
+    }
+
+    if name == "simd_fsqrt" {
+        return simd_simple_float_intrinsic("sqrt", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_fsin" {
+        return simd_simple_float_intrinsic("sin", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_fcos" {
+        return simd_simple_float_intrinsic("cos", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_fabs" {
+        return simd_simple_float_intrinsic("fabs", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_floor" {
+        return simd_simple_float_intrinsic("floor", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_ceil" {
+        return simd_simple_float_intrinsic("ceil", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_fexp" {
+        return simd_simple_float_intrinsic("exp", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_fexp2" {
+        return simd_simple_float_intrinsic("exp2", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_flog10" {
+        return simd_simple_float_intrinsic("log10", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_flog2" {
+        return simd_simple_float_intrinsic("log2", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_flog" {
+        return simd_simple_float_intrinsic("log", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_fpowi" {
+        return simd_simple_float_intrinsic("powi", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_fpow"  {
+        return simd_simple_float_intrinsic("pow", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    if name == "simd_fma" {
+        return simd_simple_float_intrinsic("fma", in_elem, in_ty, in_len, bx, span, args);
+    }
+
+    // FIXME: use:
+    //  https://github.com/llvm-mirror/llvm/blob/master/include/llvm/IR/Function.h#L182
+    //  https://github.com/llvm-mirror/llvm/blob/master/include/llvm/IR/Intrinsics.h#L81
+    fn llvm_vector_str(elem_ty: ty::Ty, vec_len: usize, no_pointers: usize) -> String {
+        let p0s: String = "p0".repeat(no_pointers);
+        match elem_ty.sty {
+            ty::TyInt(v) => format!("v{}{}i{}", vec_len, p0s, v.bit_width().unwrap()),
+            ty::TyUint(v) => format!("v{}{}i{}", vec_len, p0s, v.bit_width().unwrap()),
+            ty::TyFloat(v) => format!("v{}{}f{}", vec_len, p0s, v.bit_width()),
+            _ => unreachable!(),
+        }
+    }
+
+    fn llvm_vector_ty(cx: &CodegenCx, elem_ty: ty::Ty, vec_len: usize,
+                      mut no_pointers: usize) -> Type {
+        // FIXME: use cx.layout_of(ty).llvm_type() ?
+        let mut elem_ty = match elem_ty.sty {
+            ty::TyInt(v) => Type::int_from_ty(cx, v),
+            ty::TyUint(v) => Type::uint_from_ty(cx, v),
+            ty::TyFloat(v) => Type::float_from_ty(cx, v),
+            _ => unreachable!(),
+        };
+        while no_pointers > 0 {
+            elem_ty = elem_ty.ptr_to();
+            no_pointers -= 1;
+        }
+        Type::vector(&elem_ty, vec_len as u64)
+    }
+
+
+    if name == "simd_gather"  {
+        // simd_gather(values: <N x T>, pointers: <N x *_ T>,
+        //             mask: <N x i{M}>) -> <N x T>
+        // * N: number of elements in the input vectors
+        // * T: type of the element to load
+        // * M: any integer width is supported, will be truncated to i1
+
+        // All types must be simd vector types
+        require_simd!(in_ty, "first");
+        require_simd!(arg_tys[1], "second");
+        require_simd!(arg_tys[2], "third");
+        require_simd!(ret_ty, "return");
+
+        // Of the same length:
+        require!(in_len == arg_tys[1].simd_size(tcx),
+                 "expected {} argument with length {} (same as input type `{}`), \
+                  found `{}` with length {}", "second", in_len, in_ty, arg_tys[1],
+                 arg_tys[1].simd_size(tcx));
+        require!(in_len == arg_tys[2].simd_size(tcx),
+                 "expected {} argument with length {} (same as input type `{}`), \
+                  found `{}` with length {}", "third", in_len, in_ty, arg_tys[2],
+                 arg_tys[2].simd_size(tcx));
+
+        // The return type must match the first argument type
+        require!(ret_ty == in_ty,
+                 "expected return type `{}`, found `{}`",
+                 in_ty, ret_ty);
+
+        // This counts how many pointers
+        fn ptr_count(t: ty::Ty) -> usize {
+            match t.sty {
+                ty::TyRawPtr(p) => 1 + ptr_count(p.ty),
+                _ => 0,
+            }
+        }
+
+        // Non-ptr type
+        fn non_ptr(t: ty::Ty) -> ty::Ty {
+            match t.sty {
+                ty::TyRawPtr(p) => non_ptr(p.ty),
+                _ => t,
+            }
+        }
+
+        // The second argument must be a simd vector with an element type that's a pointer
+        // to the element type of the first argument
+        let (pointer_count, underlying_ty) = match arg_tys[1].simd_type(tcx).sty {
+            ty::TyRawPtr(p) if p.ty == in_elem => (ptr_count(arg_tys[1].simd_type(tcx)),
+                                                   non_ptr(arg_tys[1].simd_type(tcx))),
+            _ => {
+                require!(false, "expected element type `{}` of second argument `{}` \
+                                 to be a pointer to the element type `{}` of the first \
+                                 argument `{}`, found `{}` != `*_ {}`",
+                         arg_tys[1].simd_type(tcx).sty, arg_tys[1], in_elem, in_ty,
+                         arg_tys[1].simd_type(tcx).sty, in_elem);
+                unreachable!();
+            }
+        };
+        assert!(pointer_count > 0);
+        assert!(pointer_count - 1 == ptr_count(arg_tys[0].simd_type(tcx)));
+        assert_eq!(underlying_ty, non_ptr(arg_tys[0].simd_type(tcx)));
+
+        // The element type of the third argument must be a signed integer type of any width:
+        match arg_tys[2].simd_type(tcx).sty {
+            ty::TyInt(_) => (),
+            _ => {
+                require!(false, "expected element type `{}` of third argument `{}` \
+                                 to be a signed integer type",
+                         arg_tys[2].simd_type(tcx).sty, arg_tys[2]);
+            }
+        }
+
+        // Alignment of T, must be a constant integer value:
+        let alignment_ty = Type::i32(bx.cx);
+        let alignment = C_i32(bx.cx, bx.cx.align_of(in_elem).abi() as i32);
+
+        // Truncate the mask vector to a vector of i1s:
+        let (mask, mask_ty) = {
+            let i1 = Type::i1(bx.cx);
+            let i1xn = Type::vector(&i1, in_len as u64);
+            (bx.trunc(args[2].immediate(), i1xn), i1xn)
+        };
+
+        // Type of the vector of pointers:
+        let llvm_pointer_vec_ty = llvm_vector_ty(bx.cx, underlying_ty, in_len, pointer_count);
+        let llvm_pointer_vec_str = llvm_vector_str(underlying_ty, in_len, pointer_count);
+
+        // Type of the vector of elements:
+        let llvm_elem_vec_ty = llvm_vector_ty(bx.cx, underlying_ty, in_len, pointer_count - 1);
+        let llvm_elem_vec_str = llvm_vector_str(underlying_ty, in_len, pointer_count - 1);
+
+        let llvm_intrinsic = format!("llvm.masked.gather.{}.{}",
+                                     llvm_elem_vec_str, llvm_pointer_vec_str);
+        let f = declare::declare_cfn(bx.cx, &llvm_intrinsic,
+                                     Type::func(&[llvm_pointer_vec_ty, alignment_ty, mask_ty,
+                                                  llvm_elem_vec_ty], &llvm_elem_vec_ty));
+        llvm::SetUnnamedAddr(f, false);
+        let v = bx.call(f, &[args[1].immediate(), alignment, mask, args[0].immediate()],
+                        None);
+        return Ok(v);
+    }
+
+    if name == "simd_scatter"  {
+        // simd_scatter(values: <N x T>, pointers: <N x *mut T>,
+        //             mask: <N x i{M}>) -> ()
+        // * N: number of elements in the input vectors
+        // * T: type of the element to load
+        // * M: any integer width is supported, will be truncated to i1
+
+        // All types must be simd vector types
+        require_simd!(in_ty, "first");
+        require_simd!(arg_tys[1], "second");
+        require_simd!(arg_tys[2], "third");
+
+        // Of the same length:
+        require!(in_len == arg_tys[1].simd_size(tcx),
+                 "expected {} argument with length {} (same as input type `{}`), \
+                  found `{}` with length {}", "second", in_len, in_ty, arg_tys[1],
+                 arg_tys[1].simd_size(tcx));
+        require!(in_len == arg_tys[2].simd_size(tcx),
+                 "expected {} argument with length {} (same as input type `{}`), \
+                  found `{}` with length {}", "third", in_len, in_ty, arg_tys[2],
+                 arg_tys[2].simd_size(tcx));
+
+        // This counts how many pointers
+        fn ptr_count(t: ty::Ty) -> usize {
+            match t.sty {
+                ty::TyRawPtr(p) => 1 + ptr_count(p.ty),
+                _ => 0,
+            }
+        }
+
+        // Non-ptr type
+        fn non_ptr(t: ty::Ty) -> ty::Ty {
+            match t.sty {
+                ty::TyRawPtr(p) => non_ptr(p.ty),
+                _ => t,
+            }
+        }
+
+        // The second argument must be a simd vector with an element type that's a pointer
+        // to the element type of the first argument
+        let (pointer_count, underlying_ty) = match arg_tys[1].simd_type(tcx).sty {
+            ty::TyRawPtr(p) if p.ty == in_elem && p.mutbl == hir::MutMutable
+                => (ptr_count(arg_tys[1].simd_type(tcx)),
+                    non_ptr(arg_tys[1].simd_type(tcx))),
+            _ => {
+                require!(false, "expected element type `{}` of second argument `{}` \
+                                 to be a pointer to the element type `{}` of the first \
+                                 argument `{}`, found `{}` != `*mut {}`",
+                         arg_tys[1].simd_type(tcx).sty, arg_tys[1], in_elem, in_ty,
+                         arg_tys[1].simd_type(tcx).sty, in_elem);
+                unreachable!();
+            }
+        };
+        assert!(pointer_count > 0);
+        assert!(pointer_count - 1 == ptr_count(arg_tys[0].simd_type(tcx)));
+        assert_eq!(underlying_ty, non_ptr(arg_tys[0].simd_type(tcx)));
+
+        // The element type of the third argument must be a signed integer type of any width:
+        match arg_tys[2].simd_type(tcx).sty {
+            ty::TyInt(_) => (),
+            _ => {
+                require!(false, "expected element type `{}` of third argument `{}` \
+                                 to be a signed integer type",
+                         arg_tys[2].simd_type(tcx).sty, arg_tys[2]);
+            }
+        }
+
+        // Alignment of T, must be a constant integer value:
+        let alignment_ty = Type::i32(bx.cx);
+        let alignment = C_i32(bx.cx, bx.cx.align_of(in_elem).abi() as i32);
+
+        // Truncate the mask vector to a vector of i1s:
+        let (mask, mask_ty) = {
+            let i1 = Type::i1(bx.cx);
+            let i1xn = Type::vector(&i1, in_len as u64);
+            (bx.trunc(args[2].immediate(), i1xn), i1xn)
+        };
+
+        let ret_t = Type::void(bx.cx);
+
+        // Type of the vector of pointers:
+        let llvm_pointer_vec_ty = llvm_vector_ty(bx.cx, underlying_ty, in_len, pointer_count);
+        let llvm_pointer_vec_str = llvm_vector_str(underlying_ty, in_len, pointer_count);
+
+        // Type of the vector of elements:
+        let llvm_elem_vec_ty = llvm_vector_ty(bx.cx, underlying_ty, in_len, pointer_count - 1);
+        let llvm_elem_vec_str = llvm_vector_str(underlying_ty, in_len, pointer_count - 1);
+
+        let llvm_intrinsic = format!("llvm.masked.scatter.{}.{}",
+                                     llvm_elem_vec_str, llvm_pointer_vec_str);
+        let f = declare::declare_cfn(bx.cx, &llvm_intrinsic,
+                                     Type::func(&[llvm_elem_vec_ty,
+                                                  llvm_pointer_vec_ty,
+                                                  alignment_ty,
+                                                  mask_ty], &ret_t));
+        llvm::SetUnnamedAddr(f, false);
+        let v = bx.call(f, &[args[0].immediate(), args[1].immediate(), alignment, mask],
+                        None);
+        return Ok(v);
+    }
+
     macro_rules! arith_red {
         ($name:tt : $integer_reduce:ident, $float_reduce:ident, $ordered:expr) => {
             if name == $name {

src/librustc_typeck/check/intrinsic.rs

@@ -351,9 +351,26 @@ pub fn check_platform_intrinsic_type<'a, 'tcx>(tcx: TyCtxt<'a, 'tcx, 'tcx>,
         "simd_add" | "simd_sub" | "simd_mul" | "simd_rem" |
         "simd_div" | "simd_shl" | "simd_shr" |
         "simd_and" | "simd_or" | "simd_xor" |
-        "simd_fmin" | "simd_fmax" => {
+        "simd_fmin" | "simd_fmax" | "simd_fpow" => {
             (1, vec![param(0), param(0)], param(0))
         }
+        "simd_fsqrt" | "simd_fsin" | "simd_fcos" | "simd_fexp" | "simd_fexp2" |
+        "simd_flog2" | "simd_flog10" | "simd_flog" |
+        "simd_fabs" | "simd_floor" | "simd_ceil" => {
+            (1, vec![param(0)], param(0))
+        }
+        "simd_fpowi" => {
+            (1, vec![param(0), tcx.types.i32], param(0))
+        }
+        "simd_fma" => {
+            (1, vec![param(0), param(0), param(0)], param(0))
+        }
+        "simd_gather" => {
+            (3, vec![param(0), param(1), param(2)], param(0))
+        }
+        "simd_scatter" => {
+            (3, vec![param(0), param(1), param(2)], tcx.mk_nil())
+        }
         "simd_insert" => (2, vec![param(0), tcx.types.u32, param(1)], param(0)),
         "simd_extract" => (2, vec![param(0), tcx.types.u32], param(1)),
         "simd_cast" => (2, vec![param(0)], param(1)),

src/test/codegen/simd-intrinsic-float-abs.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fabs<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @fabs_32x2
+#[no_mangle]
+pub unsafe fn fabs_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.fabs.v2f32
+    simd_fabs(a)
+}
+
+// CHECK-LABEL: @fabs_32x4
+#[no_mangle]
+pub unsafe fn fabs_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.fabs.v4f32
+    simd_fabs(a)
+}
+
+// CHECK-LABEL: @fabs_32x8
+#[no_mangle]
+pub unsafe fn fabs_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.fabs.v8f32
+    simd_fabs(a)
+}
+
+// CHECK-LABEL: @fabs_32x16
+#[no_mangle]
+pub unsafe fn fabs_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.fabs.v16f32
+    simd_fabs(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @fabs_64x4
+#[no_mangle]
+pub unsafe fn fabs_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.fabs.v4f64
+    simd_fabs(a)
+}
+
+// CHECK-LABEL: @fabs_64x2
+#[no_mangle]
+pub unsafe fn fabs_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.fabs.v2f64
+    simd_fabs(a)
+}
+
+// CHECK-LABEL: @fabs_64x8
+#[no_mangle]
+pub unsafe fn fabs_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.fabs.v8f64
+    simd_fabs(a)
+}

src/test/codegen/simd-intrinsic-float-ceil.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_ceil<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @ceil_32x2
+#[no_mangle]
+pub unsafe fn ceil_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.ceil.v2f32
+    simd_ceil(a)
+}
+
+// CHECK-LABEL: @ceil_32x4
+#[no_mangle]
+pub unsafe fn ceil_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.ceil.v4f32
+    simd_ceil(a)
+}
+
+// CHECK-LABEL: @ceil_32x8
+#[no_mangle]
+pub unsafe fn ceil_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.ceil.v8f32
+    simd_ceil(a)
+}
+
+// CHECK-LABEL: @ceil_32x16
+#[no_mangle]
+pub unsafe fn ceil_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.ceil.v16f32
+    simd_ceil(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @ceil_64x4
+#[no_mangle]
+pub unsafe fn ceil_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.ceil.v4f64
+    simd_ceil(a)
+}
+
+// CHECK-LABEL: @ceil_64x2
+#[no_mangle]
+pub unsafe fn ceil_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.ceil.v2f64
+    simd_ceil(a)
+}
+
+// CHECK-LABEL: @ceil_64x8
+#[no_mangle]
+pub unsafe fn ceil_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.ceil.v8f64
+    simd_ceil(a)
+}

src/test/codegen/simd-intrinsic-float-cos.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fcos<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @fcos_32x2
+#[no_mangle]
+pub unsafe fn fcos_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.cos.v2f32
+    simd_fcos(a)
+}
+
+// CHECK-LABEL: @fcos_32x4
+#[no_mangle]
+pub unsafe fn fcos_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.cos.v4f32
+    simd_fcos(a)
+}
+
+// CHECK-LABEL: @fcos_32x8
+#[no_mangle]
+pub unsafe fn fcos_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.cos.v8f32
+    simd_fcos(a)
+}
+
+// CHECK-LABEL: @fcos_32x16
+#[no_mangle]
+pub unsafe fn fcos_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.cos.v16f32
+    simd_fcos(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @fcos_64x4
+#[no_mangle]
+pub unsafe fn fcos_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.cos.v4f64
+    simd_fcos(a)
+}
+
+// CHECK-LABEL: @fcos_64x2
+#[no_mangle]
+pub unsafe fn fcos_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.cos.v2f64
+    simd_fcos(a)
+}
+
+// CHECK-LABEL: @fcos_64x8
+#[no_mangle]
+pub unsafe fn fcos_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.cos.v8f64
+    simd_fcos(a)
+}

src/test/codegen/simd-intrinsic-float-exp.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fexp<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @exp_32x2
+#[no_mangle]
+pub unsafe fn exp_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.exp.v2f32
+    simd_fexp(a)
+}
+
+// CHECK-LABEL: @exp_32x4
+#[no_mangle]
+pub unsafe fn exp_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.exp.v4f32
+    simd_fexp(a)
+}
+
+// CHECK-LABEL: @exp_32x8
+#[no_mangle]
+pub unsafe fn exp_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.exp.v8f32
+    simd_fexp(a)
+}
+
+// CHECK-LABEL: @exp_32x16
+#[no_mangle]
+pub unsafe fn exp_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.exp.v16f32
+    simd_fexp(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @exp_64x4
+#[no_mangle]
+pub unsafe fn exp_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.exp.v4f64
+    simd_fexp(a)
+}
+
+// CHECK-LABEL: @exp_64x2
+#[no_mangle]
+pub unsafe fn exp_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.exp.v2f64
+    simd_fexp(a)
+}
+
+// CHECK-LABEL: @exp_64x8
+#[no_mangle]
+pub unsafe fn exp_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.exp.v8f64
+    simd_fexp(a)
+}

src/test/codegen/simd-intrinsic-float-exp2.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fexp2<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @exp2_32x2
+#[no_mangle]
+pub unsafe fn exp2_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.exp2.v2f32
+    simd_fexp2(a)
+}
+
+// CHECK-LABEL: @exp2_32x4
+#[no_mangle]
+pub unsafe fn exp2_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.exp2.v4f32
+    simd_fexp2(a)
+}
+
+// CHECK-LABEL: @exp2_32x8
+#[no_mangle]
+pub unsafe fn exp2_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.exp2.v8f32
+    simd_fexp2(a)
+}
+
+// CHECK-LABEL: @exp2_32x16
+#[no_mangle]
+pub unsafe fn exp2_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.exp2.v16f32
+    simd_fexp2(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @exp2_64x4
+#[no_mangle]
+pub unsafe fn exp2_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.exp2.v4f64
+    simd_fexp2(a)
+}
+
+// CHECK-LABEL: @exp2_64x2
+#[no_mangle]
+pub unsafe fn exp2_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.exp2.v2f64
+    simd_fexp2(a)
+}
+
+// CHECK-LABEL: @exp2_64x8
+#[no_mangle]
+pub unsafe fn exp2_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.exp2.v8f64
+    simd_fexp2(a)
+}

src/test/codegen/simd-intrinsic-float-floor.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_floor<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @floor_32x2
+#[no_mangle]
+pub unsafe fn floor_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.floor.v2f32
+    simd_floor(a)
+}
+
+// CHECK-LABEL: @floor_32x4
+#[no_mangle]
+pub unsafe fn floor_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.floor.v4f32
+    simd_floor(a)
+}
+
+// CHECK-LABEL: @floor_32x8
+#[no_mangle]
+pub unsafe fn floor_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.floor.v8f32
+    simd_floor(a)
+}
+
+// CHECK-LABEL: @floor_32x16
+#[no_mangle]
+pub unsafe fn floor_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.floor.v16f32
+    simd_floor(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @floor_64x4
+#[no_mangle]
+pub unsafe fn floor_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.floor.v4f64
+    simd_floor(a)
+}
+
+// CHECK-LABEL: @floor_64x2
+#[no_mangle]
+pub unsafe fn floor_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.floor.v2f64
+    simd_floor(a)
+}
+
+// CHECK-LABEL: @floor_64x8
+#[no_mangle]
+pub unsafe fn floor_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.floor.v8f64
+    simd_floor(a)
+}

src/test/codegen/simd-intrinsic-float-fma.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fma<T>(x: T, b: T, c: T) -> T;
+}
+
+// CHECK-LABEL: @fma_32x2
+#[no_mangle]
+pub unsafe fn fma_32x2(a: f32x2, b: f32x2, c: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.fma.v2f32
+    simd_fma(a, b, c)
+}
+
+// CHECK-LABEL: @fma_32x4
+#[no_mangle]
+pub unsafe fn fma_32x4(a: f32x4, b: f32x4, c: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.fma.v4f32
+    simd_fma(a, b, c)
+}
+
+// CHECK-LABEL: @fma_32x8
+#[no_mangle]
+pub unsafe fn fma_32x8(a: f32x8, b: f32x8, c: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.fma.v8f32
+    simd_fma(a, b, c)
+}
+
+// CHECK-LABEL: @fma_32x16
+#[no_mangle]
+pub unsafe fn fma_32x16(a: f32x16, b: f32x16, c: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.fma.v16f32
+    simd_fma(a, b, c)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @fma_64x4
+#[no_mangle]
+pub unsafe fn fma_64x4(a: f64x4, b: f64x4, c: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.fma.v4f64
+    simd_fma(a, b, c)
+}
+
+// CHECK-LABEL: @fma_64x2
+#[no_mangle]
+pub unsafe fn fma_64x2(a: f64x2, b: f64x2, c: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.fma.v2f64
+    simd_fma(a, b, c)
+}
+
+// CHECK-LABEL: @fma_64x8
+#[no_mangle]
+pub unsafe fn fma_64x8(a: f64x8, b: f64x8, c: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.fma.v8f64
+    simd_fma(a, b, c)
+}

src/test/codegen/simd-intrinsic-float-fsqrt.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fsqrt<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @fsqrt_32x2
+#[no_mangle]
+pub unsafe fn fsqrt_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.sqrt.v2f32
+    simd_fsqrt(a)
+}
+
+// CHECK-LABEL: @fsqrt_32x4
+#[no_mangle]
+pub unsafe fn fsqrt_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.sqrt.v4f32
+    simd_fsqrt(a)
+}
+
+// CHECK-LABEL: @fsqrt_32x8
+#[no_mangle]
+pub unsafe fn fsqrt_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.sqrt.v8f32
+    simd_fsqrt(a)
+}
+
+// CHECK-LABEL: @fsqrt_32x16
+#[no_mangle]
+pub unsafe fn fsqrt_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.sqrt.v16f32
+    simd_fsqrt(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @fsqrt_64x4
+#[no_mangle]
+pub unsafe fn fsqrt_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.sqrt.v4f64
+    simd_fsqrt(a)
+}
+
+// CHECK-LABEL: @fsqrt_64x2
+#[no_mangle]
+pub unsafe fn fsqrt_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.sqrt.v2f64
+    simd_fsqrt(a)
+}
+
+// CHECK-LABEL: @fsqrt_64x8
+#[no_mangle]
+pub unsafe fn fsqrt_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.sqrt.v8f64
+    simd_fsqrt(a)
+}

src/test/codegen/simd-intrinsic-float-log.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_flog<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @log_32x2
+#[no_mangle]
+pub unsafe fn log_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.log.v2f32
+    simd_flog(a)
+}
+
+// CHECK-LABEL: @log_32x4
+#[no_mangle]
+pub unsafe fn log_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.log.v4f32
+    simd_flog(a)
+}
+
+// CHECK-LABEL: @log_32x8
+#[no_mangle]
+pub unsafe fn log_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.log.v8f32
+    simd_flog(a)
+}
+
+// CHECK-LABEL: @log_32x16
+#[no_mangle]
+pub unsafe fn log_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.log.v16f32
+    simd_flog(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @log_64x4
+#[no_mangle]
+pub unsafe fn log_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.log.v4f64
+    simd_flog(a)
+}
+
+// CHECK-LABEL: @log_64x2
+#[no_mangle]
+pub unsafe fn log_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.log.v2f64
+    simd_flog(a)
+}
+
+// CHECK-LABEL: @log_64x8
+#[no_mangle]
+pub unsafe fn log_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.log.v8f64
+    simd_flog(a)
+}

src/test/codegen/simd-intrinsic-float-log10.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_flog10<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @log10_32x2
+#[no_mangle]
+pub unsafe fn log10_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.log10.v2f32
+    simd_flog10(a)
+}
+
+// CHECK-LABEL: @log10_32x4
+#[no_mangle]
+pub unsafe fn log10_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.log10.v4f32
+    simd_flog10(a)
+}
+
+// CHECK-LABEL: @log10_32x8
+#[no_mangle]
+pub unsafe fn log10_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.log10.v8f32
+    simd_flog10(a)
+}
+
+// CHECK-LABEL: @log10_32x16
+#[no_mangle]
+pub unsafe fn log10_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.log10.v16f32
+    simd_flog10(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @log10_64x4
+#[no_mangle]
+pub unsafe fn log10_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.log10.v4f64
+    simd_flog10(a)
+}
+
+// CHECK-LABEL: @log10_64x2
+#[no_mangle]
+pub unsafe fn log10_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.log10.v2f64
+    simd_flog10(a)
+}
+
+// CHECK-LABEL: @log10_64x8
+#[no_mangle]
+pub unsafe fn log10_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.log10.v8f64
+    simd_flog10(a)
+}

src/test/codegen/simd-intrinsic-float-log2.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_flog2<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @log2_32x2
+#[no_mangle]
+pub unsafe fn log2_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.log2.v2f32
+    simd_flog2(a)
+}
+
+// CHECK-LABEL: @log2_32x4
+#[no_mangle]
+pub unsafe fn log2_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.log2.v4f32
+    simd_flog2(a)
+}
+
+// CHECK-LABEL: @log2_32x8
+#[no_mangle]
+pub unsafe fn log2_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.log2.v8f32
+    simd_flog2(a)
+}
+
+// CHECK-LABEL: @log2_32x16
+#[no_mangle]
+pub unsafe fn log2_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.log2.v16f32
+    simd_flog2(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @log2_64x4
+#[no_mangle]
+pub unsafe fn log2_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.log2.v4f64
+    simd_flog2(a)
+}
+
+// CHECK-LABEL: @log2_64x2
+#[no_mangle]
+pub unsafe fn log2_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.log2.v2f64
+    simd_flog2(a)
+}
+
+// CHECK-LABEL: @log2_64x8
+#[no_mangle]
+pub unsafe fn log2_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.log2.v8f64
+    simd_flog2(a)
+}

src/test/codegen/simd-intrinsic-float-minmax.rs

@@ -16,7 +16,7 @@
 #![crate_type = "lib"]
 
 #![feature(repr_simd, platform_intrinsics)]
-#[allow(non_camel_case_types)]
+#![allow(non_camel_case_types)]
 
 #[repr(simd)]
 #[derive(Copy, Clone, PartialEq, Debug)]

src/test/codegen/simd-intrinsic-float-pow.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fpow<T>(x: T, b: T) -> T;
+}
+
+// CHECK-LABEL: @fpow_32x2
+#[no_mangle]
+pub unsafe fn fpow_32x2(a: f32x2, b: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.pow.v2f32
+    simd_fpow(a, b)
+}
+
+// CHECK-LABEL: @fpow_32x4
+#[no_mangle]
+pub unsafe fn fpow_32x4(a: f32x4, b: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.pow.v4f32
+    simd_fpow(a, b)
+}
+
+// CHECK-LABEL: @fpow_32x8
+#[no_mangle]
+pub unsafe fn fpow_32x8(a: f32x8, b: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.pow.v8f32
+    simd_fpow(a, b)
+}
+
+// CHECK-LABEL: @fpow_32x16
+#[no_mangle]
+pub unsafe fn fpow_32x16(a: f32x16, b: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.pow.v16f32
+    simd_fpow(a, b)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @fpow_64x4
+#[no_mangle]
+pub unsafe fn fpow_64x4(a: f64x4, b: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.pow.v4f64
+    simd_fpow(a, b)
+}
+
+// CHECK-LABEL: @fpow_64x2
+#[no_mangle]
+pub unsafe fn fpow_64x2(a: f64x2, b: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.pow.v2f64
+    simd_fpow(a, b)
+}
+
+// CHECK-LABEL: @fpow_64x8
+#[no_mangle]
+pub unsafe fn fpow_64x8(a: f64x8, b: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.pow.v8f64
+    simd_fpow(a, b)
+}

src/test/codegen/simd-intrinsic-float-powi.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fpowi<T>(x: T, b: i32) -> T;
+}
+
+// CHECK-LABEL: @fpowi_32x2
+#[no_mangle]
+pub unsafe fn fpowi_32x2(a: f32x2, b: i32) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.powi.v2f32
+    simd_fpowi(a, b)
+}
+
+// CHECK-LABEL: @fpowi_32x4
+#[no_mangle]
+pub unsafe fn fpowi_32x4(a: f32x4, b: i32) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.powi.v4f32
+    simd_fpowi(a, b)
+}
+
+// CHECK-LABEL: @fpowi_32x8
+#[no_mangle]
+pub unsafe fn fpowi_32x8(a: f32x8, b: i32) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.powi.v8f32
+    simd_fpowi(a, b)
+}
+
+// CHECK-LABEL: @fpowi_32x16
+#[no_mangle]
+pub unsafe fn fpowi_32x16(a: f32x16, b: i32) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.powi.v16f32
+    simd_fpowi(a, b)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @fpowi_64x4
+#[no_mangle]
+pub unsafe fn fpowi_64x4(a: f64x4, b: i32) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.powi.v4f64
+    simd_fpowi(a, b)
+}
+
+// CHECK-LABEL: @fpowi_64x2
+#[no_mangle]
+pub unsafe fn fpowi_64x2(a: f64x2, b: i32) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.powi.v2f64
+    simd_fpowi(a, b)
+}
+
+// CHECK-LABEL: @fpowi_64x8
+#[no_mangle]
+pub unsafe fn fpowi_64x8(a: f64x8, b: i32) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.powi.v8f64
+    simd_fpowi(a, b)
+}

src/test/codegen/simd-intrinsic-float-sin.rs

@@ -0,0 +1,104 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x2(pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x8(pub f32, pub f32, pub f32, pub f32,
+                 pub f32, pub f32, pub f32, pub f32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f32x16(pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32,
+                  pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fsin<T>(x: T) -> T;
+}
+
+// CHECK-LABEL: @fsin_32x2
+#[no_mangle]
+pub unsafe fn fsin_32x2(a: f32x2) -> f32x2 {
+    // CHECK: call fast <2 x float> @llvm.sin.v2f32
+    simd_fsin(a)
+}
+
+// CHECK-LABEL: @fsin_32x4
+#[no_mangle]
+pub unsafe fn fsin_32x4(a: f32x4) -> f32x4 {
+    // CHECK: call fast <4 x float> @llvm.sin.v4f32
+    simd_fsin(a)
+}
+
+// CHECK-LABEL: @fsin_32x8
+#[no_mangle]
+pub unsafe fn fsin_32x8(a: f32x8) -> f32x8 {
+    // CHECK: call fast <8 x float> @llvm.sin.v8f32
+    simd_fsin(a)
+}
+
+// CHECK-LABEL: @fsin_32x16
+#[no_mangle]
+pub unsafe fn fsin_32x16(a: f32x16) -> f32x16 {
+    // CHECK: call fast <16 x float> @llvm.sin.v16f32
+    simd_fsin(a)
+}
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x2(pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x4(pub f64, pub f64, pub f64, pub f64);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct f64x8(pub f64, pub f64, pub f64, pub f64,
+                 pub f64, pub f64, pub f64, pub f64);
+
+// CHECK-LABEL: @fsin_64x4
+#[no_mangle]
+pub unsafe fn fsin_64x4(a: f64x4) -> f64x4 {
+    // CHECK: call fast <4 x double> @llvm.sin.v4f64
+    simd_fsin(a)
+}
+
+// CHECK-LABEL: @fsin_64x2
+#[no_mangle]
+pub unsafe fn fsin_64x2(a: f64x2) -> f64x2 {
+    // CHECK: call fast <2 x double> @llvm.sin.v2f64
+    simd_fsin(a)
+}
+
+// CHECK-LABEL: @fsin_64x8
+#[no_mangle]
+pub unsafe fn fsin_64x8(a: f64x8) -> f64x8 {
+    // CHECK: call fast <8 x double> @llvm.sin.v8f64
+    simd_fsin(a)
+}

src/test/codegen/simd-intrinsic-generic-gather.rs

@@ -0,0 +1,48 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+// ignore-tidy-linelength
+// min-llvm-version 6.0
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct Vec2<T>(pub T, pub T);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct Vec4<T>(pub T, pub T, pub T, pub T);
+
+extern "platform-intrinsic" {
+    fn simd_gather<T, P, M>(value: T, pointers: P, mask: M) -> T;
+}
+
+// CHECK-LABEL: @gather_f32x2
+#[no_mangle]
+pub unsafe fn gather_f32x2(pointers: Vec2<*const f32>, mask: Vec2<i32>,
+                           values: Vec2<f32>) -> Vec2<f32> {
+    // CHECK: call <2 x float> @llvm.masked.gather.v2f32.v2p0f32(<2 x float*> {{.*}}, i32 {{.*}}, <2 x i1> {{.*}}, <2 x float> {{.*}})
+    simd_gather(values, pointers, mask)
+}
+
+// CHECK-LABEL: @gather_pf32x2
+#[no_mangle]
+pub unsafe fn gather_pf32x2(pointers: Vec2<*const *const f32>, mask: Vec2<i32>,
+                           values: Vec2<*const f32>) -> Vec2<*const f32> {
+    // CHECK: call <2 x float*> @llvm.masked.gather.v2p0f32.v2p0p0f32(<2 x float**> {{.*}}, i32 {{.*}}, <2 x i1> {{.*}}, <2 x float*> {{.*}})
+    simd_gather(values, pointers, mask)
+}

src/test/codegen/simd-intrinsic-generic-scatter.rs

@@ -0,0 +1,49 @@
+// Copyright 2016 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+// ignore-tidy-linelength
+// min-llvm-version 6.0
+
+// compile-flags: -C no-prepopulate-passes
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct Vec2<T>(pub T, pub T);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+pub struct Vec4<T>(pub T, pub T, pub T, pub T);
+
+extern "platform-intrinsic" {
+    fn simd_scatter<T, P, M>(value: T, pointers: P, mask: M);
+}
+
+// CHECK-LABEL: @scatter_f32x2
+#[no_mangle]
+pub unsafe fn scatter_f32x2(pointers: Vec2<*mut f32>, mask: Vec2<i32>,
+                            values: Vec2<f32>) {
+    // CHECK: call void @llvm.masked.scatter.v2f32.v2p0f32(<2 x float> {{.*}}, <2 x float*> {{.*}}, i32 {{.*}}, <2 x i1> {{.*}})
+    simd_scatter(values, pointers, mask)
+}
+
+
+// CHECK-LABEL: @scatter_pf32x2
+#[no_mangle]
+pub unsafe fn scatter_pf32x2(pointers: Vec2<*mut *const f32>, mask: Vec2<i32>,
+                             values: Vec2<*const f32>) {
+    // CHECK: call void @llvm.masked.scatter.v2p0f32.v2p0p0f32(<2 x float*> {{.*}}, <2 x float**> {{.*}}, i32 {{.*}}, <2 x i1> {{.*}})
+    simd_scatter(values, pointers, mask)
+}

src/test/run-pass/simd-intrinsic-float-math.rs

@@ -0,0 +1,94 @@
+// Copyright 2015 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+// ignore-android
+
+// FIXME: this test fails on arm-android because the NDK version 14 is too old.
+// It needs at least version 18. We disable it on all android build bots because
+// there is no way in compile-test to disable it for an (arch,os) pair.
+
+// Test that the simd floating-point math intrinsics produce correct results.
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_fsqrt<T>(x: T) -> T;
+    fn simd_fabs<T>(x: T) -> T;
+    fn simd_fsin<T>(x: T) -> T;
+    fn simd_fcos<T>(x: T) -> T;
+    fn simd_ceil<T>(x: T) -> T;
+    fn simd_fexp<T>(x: T) -> T;
+    fn simd_fexp2<T>(x: T) -> T;
+    fn simd_floor<T>(x: T) -> T;
+    fn simd_fma<T>(x: T, y: T, z: T) -> T;
+    fn simd_flog<T>(x: T) -> T;
+    fn simd_flog10<T>(x: T) -> T;
+    fn simd_flog2<T>(x: T) -> T;
+    fn simd_fpow<T>(x: T, y: T) -> T;
+    fn simd_fpowi<T>(x: T, y: i32) -> T;
+}
+
+fn main() {
+    let x = f32x4(1.0, 1.0, 1.0, 1.0);
+    let y = f32x4(-1.0, -1.0, -1.0, -1.0);
+    let z = f32x4(0.0, 0.0, 0.0, 0.0);
+
+    let h = f32x4(0.5, 0.5, 0.5, 0.5);
+
+    unsafe {
+        let r = simd_fabs(y);
+        assert_eq!(x, r);
+
+        let r = simd_fcos(z);
+        assert_eq!(x, r);
+
+        let r = simd_ceil(h);
+        assert_eq!(x, r);
+
+        let r = simd_fexp(z);
+        assert_eq!(x, r);
+
+        let r = simd_fexp2(z);
+        assert_eq!(x, r);
+
+        let r = simd_floor(h);
+        assert_eq!(z, r);
+
+        let r = simd_fma(x, h, h);
+        assert_eq!(x, r);
+
+        let r = simd_fsqrt(x);
+        assert_eq!(x, r);
+
+        let r = simd_flog(x);
+        assert_eq!(z, r);
+
+        let r = simd_flog2(x);
+        assert_eq!(z, r);
+
+        let r = simd_flog10(x);
+        assert_eq!(z, r);
+
+        let r = simd_fpow(h, x);
+        assert_eq!(h, r);
+
+        let r = simd_fpowi(h, 1);
+        assert_eq!(h, r);
+
+        let r = simd_fsin(z);
+        assert_eq!(z, r);
+    }
+}

src/test/run-pass/simd-intrinsic-float-minmax.rs

@@ -9,13 +9,13 @@
 // except according to those terms.
 
 // ignore-emscripten
-// min-llvm-version 6.0
+// min-llvm-version 7.0
 // error-pattern: panicked
 
 // Test that the simd_f{min,max} intrinsics produce the correct results.
 
 #![feature(repr_simd, platform_intrinsics)]
-#[allow(non_camel_case_types)]
+#![allow(non_camel_case_types)]
 
 #[repr(simd)]
 #[derive(Copy, Clone, PartialEq, Debug)]
@@ -43,7 +43,6 @@ fn main() {
         let minn = simd_fmin(y, n);
         assert_eq!(minn, y);
 
-        // FIXME(49261)
         let max0 = simd_fmax(x, y);
         let max1 = simd_fmax(y, x);
         assert_eq!(max0, max1);

src/test/run-pass/simd-intrinsic-generic-gather.rs

@@ -0,0 +1,151 @@
+// Copyright 2015 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+// ignore-emscripten
+// min-llvm-version 6.0
+
+// Test that the simd_{gather,scatter} intrinsics produce the correct results.
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+struct x4<T>(pub T, pub T, pub T, pub T);
+
+extern "platform-intrinsic" {
+    fn simd_gather<T, U, V>(x: T, y: U, z: V) -> T;
+    fn simd_scatter<T, U, V>(x: T, y: U, z: V) -> ();
+}
+
+fn main() {
+    let mut x = [0_f32, 1., 2., 3., 4., 5., 6., 7.];
+
+    let default = x4(-3_f32, -3., -3., -3.);
+    let s_strided = x4(0_f32, 2., -3., 6.);
+    let mask = x4(-1_i32, -1, 0, -1);
+
+    // reading from *const
+    unsafe {
+        let pointer = &x[0] as *const f32;
+        let pointers =  x4(
+            pointer.offset(0) as *const f32,
+            pointer.offset(2),
+            pointer.offset(4),
+            pointer.offset(6)
+        );
+
+        let r_strided = simd_gather(default, pointers, mask);
+
+        assert_eq!(r_strided, s_strided);
+    }
+
+    // reading from *mut
+    unsafe {
+        let pointer = &mut x[0] as *mut f32;
+        let pointers = x4(
+            pointer.offset(0) as *mut f32,
+            pointer.offset(2),
+            pointer.offset(4),
+            pointer.offset(6)
+        );
+
+        let r_strided = simd_gather(default, pointers, mask);
+
+        assert_eq!(r_strided, s_strided);
+    }
+
+    // writing to *mut
+    unsafe {
+        let pointer = &mut x[0] as *mut f32;
+        let pointers = x4(
+            pointer.offset(0) as *mut f32,
+            pointer.offset(2),
+            pointer.offset(4),
+            pointer.offset(6)
+        );
+
+        let values = x4(42_f32, 43_f32, 44_f32, 45_f32);
+        simd_scatter(values, pointers, mask);
+
+        assert_eq!(x, [42., 1., 43., 3., 4., 5., 45., 7.]);
+    }
+
+    // test modifying array of *const f32
+    let mut y = [
+        &x[0] as *const f32,
+        &x[1] as *const f32,
+        &x[2] as *const f32,
+        &x[3] as *const f32,
+        &x[4] as *const f32,
+        &x[5] as *const f32,
+        &x[6] as *const f32,
+        &x[7] as *const f32
+    ];
+
+    let default = x4(y[0], y[0], y[0], y[0]);
+    let s_strided = x4(y[0], y[2], y[0], y[6]);
+
+    // reading from *const
+    unsafe {
+        let pointer = &y[0] as *const *const f32;
+        let pointers = x4(
+            pointer.offset(0) as *const *const f32,
+            pointer.offset(2),
+            pointer.offset(4),
+            pointer.offset(6)
+        );
+
+        let r_strided = simd_gather(default, pointers, mask);
+
+        assert_eq!(r_strided, s_strided);
+    }
+
+    // reading from *mut
+    unsafe {
+        let pointer = &mut y[0] as *mut *const f32;
+        let pointers = x4(
+            pointer.offset(0) as *mut *const f32,
+            pointer.offset(2),
+            pointer.offset(4),
+            pointer.offset(6)
+        );
+
+        let r_strided = simd_gather(default, pointers, mask);
+
+        assert_eq!(r_strided, s_strided);
+    }
+
+    // writing to *mut
+    unsafe {
+        let pointer = &mut y[0] as *mut *const f32;
+        let pointers = x4(
+            pointer.offset(0) as *mut *const f32,
+            pointer.offset(2),
+            pointer.offset(4),
+            pointer.offset(6)
+        );
+
+        let values = x4(y[7], y[6], y[5], y[1]);
+        simd_scatter(values, pointers, mask);
+
+        let s = [
+            &x[7] as *const f32,
+            &x[1] as *const f32,
+            &x[6] as *const f32,
+            &x[3] as *const f32,
+            &x[4] as *const f32,
+            &x[5] as *const f32,
+            &x[1] as *const f32,
+            &x[7] as *const f32
+        ];
+        assert_eq!(y, s);
+    }
+}