Auto merge of #58304 - gnzlbg:simd_saturated, r=nagisa

Add generic simd saturated add/sub intrinsics

r? @eddyb

26 files changed, 895 insertions, 9 deletions

diff --git a/src/librustc_codegen_llvm/context.rs b/src/librustc_codegen_llvm/context.rs
index d9c4d22b411..edeef23bdca 100644
--- a/src/librustc_codegen_llvm/context.rs
+++ b/src/librustc_codegen_llvm/context.rs
@@ -515,14 +515,24 @@ impl CodegenCx<'b, 'tcx> {
         let t_f32 = self.type_f32();
         let t_f64 = self.type_f64();
 
-        let t_v2f32 = self.type_vector(t_f32, 2);
-        let t_v4f32 = self.type_vector(t_f32, 4);
-        let t_v8f32 = self.type_vector(t_f32, 8);
-        let t_v16f32 = self.type_vector(t_f32, 16);
-
-        let t_v2f64 = self.type_vector(t_f64, 2);
-        let t_v4f64 = self.type_vector(t_f64, 4);
-        let t_v8f64 = self.type_vector(t_f64, 8);
+        macro_rules! vector_types {
+            ($id_out:ident: $elem_ty:ident, $len:expr) => {
+                let $id_out = self.type_vector($elem_ty, $len);
+            };
+            ($($id_out:ident: $elem_ty:ident, $len:expr;)*) => {
+                $(vector_types!($id_out: $elem_ty, $len);)*
+            }
+        }
+        vector_types! {
+            t_v2f32: t_f32, 2;
+            t_v4f32: t_f32, 4;
+            t_v8f32: t_f32, 8;
+            t_v16f32: t_f32, 16;
+
+            t_v2f64: t_f64, 2;
+            t_v4f64: t_f64, 4;
+            t_v8f64: t_f64, 8;
+        }
 
         ifn!("llvm.memset.p0i8.i16", fn(i8p, t_i8, t_i16, t_i32, i1) -> void);
         ifn!("llvm.memset.p0i8.i32", fn(i8p, t_i8, t_i32, t_i32, i1) -> void);
diff --git a/src/librustc_codegen_llvm/intrinsic.rs b/src/librustc_codegen_llvm/intrinsic.rs
index 3785e19af97..d12ee7649bd 100644
--- a/src/librustc_codegen_llvm/intrinsic.rs
+++ b/src/librustc_codegen_llvm/intrinsic.rs
@@ -1848,7 +1848,52 @@ unsupported {} from `{}` with element `{}` of size `{}` to `{}`"#,
         simd_xor: Uint, Int => xor;
         simd_fmax: Float => maxnum;
         simd_fmin: Float => minnum;
+
+    }
+
+    if name == "simd_saturating_add" || name == "simd_saturating_sub" {
+        let lhs = args[0].immediate();
+        let rhs = args[1].immediate();
+        let is_add = name == "simd_saturating_add";
+        let ptr_bits = bx.tcx().data_layout.pointer_size.bits() as _;
+        let (signed, elem_width, elem_ty) = match in_elem.sty {
+            ty::Int(i) =>
+                (
+                    true,
+                    i.bit_width().unwrap_or(ptr_bits),
+                    bx.cx.type_int_from_ty(i)
+                ),
+            ty::Uint(i) =>
+                (
+                    false,
+                    i.bit_width().unwrap_or(ptr_bits),
+                    bx.cx.type_uint_from_ty(i)
+                ),
+            _ => {
+                return_error!(
+                    "expected element type `{}` of vector type `{}` \
+                     to be a signed or unsigned integer type",
+                    arg_tys[0].simd_type(tcx).sty, arg_tys[0]
+                );
+            }
+        };
+        let llvm_intrinsic = &format!(
+            "llvm.{}{}.sat.v{}i{}",
+            if signed { 's' } else { 'u' },
+            if is_add { "add" } else { "sub" },
+            in_len, elem_width
+        );
+        let vec_ty = bx.cx.type_vector(elem_ty, in_len as u64);
+
+        let f = bx.declare_cfn(
+            &llvm_intrinsic,
+            bx.type_func(&[vec_ty, vec_ty], vec_ty)
+        );
+        llvm::SetUnnamedAddr(f, false);
+        let v = bx.call(f, &[lhs, rhs], None);
+        return Ok(v);
     }
+
     span_bug!(span, "unknown SIMD intrinsic");
 }
 
diff --git a/src/librustc_typeck/check/intrinsic.rs b/src/librustc_typeck/check/intrinsic.rs
index fca9fa0829f..3b174b55f2b 100644
--- a/src/librustc_typeck/check/intrinsic.rs
+++ b/src/librustc_typeck/check/intrinsic.rs
@@ -410,7 +410,8 @@ 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_fpow" => {
+        "simd_fmin" | "simd_fmax" | "simd_fpow" |
+        "simd_saturating_add" | "simd_saturating_sub" => {
             (1, vec![param(0), param(0)], param(0))
         }
         "simd_fsqrt" | "simd_fsin" | "simd_fcos" | "simd_fexp" | "simd_fexp2" |
diff --git a/src/test/codegen/simd-intrinsic-float-abs.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-abs.rs
index acb993d51fb..acb993d51fb 100644
--- a/src/test/codegen/simd-intrinsic-float-abs.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-abs.rs
diff --git a/src/test/codegen/simd-intrinsic-float-ceil.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-ceil.rs
index 58667af7e50..58667af7e50 100644
--- a/src/test/codegen/simd-intrinsic-float-ceil.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-ceil.rs
diff --git a/src/test/codegen/simd-intrinsic-float-cos.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-cos.rs
index affbe17d334..affbe17d334 100644
--- a/src/test/codegen/simd-intrinsic-float-cos.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-cos.rs
diff --git a/src/test/codegen/simd-intrinsic-float-exp.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-exp.rs
index 43472d9dece..43472d9dece 100644
--- a/src/test/codegen/simd-intrinsic-float-exp.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-exp.rs
diff --git a/src/test/codegen/simd-intrinsic-float-exp2.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-exp2.rs
index 471d49b7841..471d49b7841 100644
--- a/src/test/codegen/simd-intrinsic-float-exp2.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-exp2.rs
diff --git a/src/test/codegen/simd-intrinsic-float-floor.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-floor.rs
index bd8ba8b6cd7..bd8ba8b6cd7 100644
--- a/src/test/codegen/simd-intrinsic-float-floor.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-floor.rs
diff --git a/src/test/codegen/simd-intrinsic-float-fma.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-fma.rs
index 8ad9e9004c2..8ad9e9004c2 100644
--- a/src/test/codegen/simd-intrinsic-float-fma.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-fma.rs
diff --git a/src/test/codegen/simd-intrinsic-float-fsqrt.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-fsqrt.rs
index ecffca960da..ecffca960da 100644
--- a/src/test/codegen/simd-intrinsic-float-fsqrt.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-fsqrt.rs
diff --git a/src/test/codegen/simd-intrinsic-float-log.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-log.rs
index 79e6ed54690..79e6ed54690 100644
--- a/src/test/codegen/simd-intrinsic-float-log.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-log.rs
diff --git a/src/test/codegen/simd-intrinsic-float-log10.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-log10.rs
index db92a94fca8..db92a94fca8 100644
--- a/src/test/codegen/simd-intrinsic-float-log10.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-log10.rs
diff --git a/src/test/codegen/simd-intrinsic-float-log2.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-log2.rs
index 90d9ec3cedd..90d9ec3cedd 100644
--- a/src/test/codegen/simd-intrinsic-float-log2.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-log2.rs
diff --git a/src/test/codegen/simd-intrinsic-float-minmax.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-minmax.rs
index 2761392e6a9..2761392e6a9 100644
--- a/src/test/codegen/simd-intrinsic-float-minmax.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-minmax.rs
diff --git a/src/test/codegen/simd-intrinsic-float-pow.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-pow.rs
index 1dd2c2ccb83..1dd2c2ccb83 100644
--- a/src/test/codegen/simd-intrinsic-float-pow.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-pow.rs
diff --git a/src/test/codegen/simd-intrinsic-float-powi.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-powi.rs
index 09f31bdd6bb..09f31bdd6bb 100644
--- a/src/test/codegen/simd-intrinsic-float-powi.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-powi.rs
diff --git a/src/test/codegen/simd-intrinsic-float-sin.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-sin.rs
index dc87651c2aa..dc87651c2aa 100644
--- a/src/test/codegen/simd-intrinsic-float-sin.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-float-sin.rs
diff --git a/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-arithmetic-saturating.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-arithmetic-saturating.rs
new file mode 100644
index 00000000000..adee796d247
--- /dev/null
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-arithmetic-saturating.rs
@@ -0,0 +1,685 @@
+// compile-flags: -C no-prepopulate-passes
+// ignore-tidy-linelength
+
+#![crate_type = "lib"]
+
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+#![deny(unused)]
+
+// signed integer types
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i8x2(i8, i8);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i8x4(i8, i8, i8, i8);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i8x8(
+    i8, i8, i8, i8, i8, i8, i8, i8,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i8x16(
+    i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i8x32(
+    i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8,
+    i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i8x64(
+    i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8,
+    i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8,
+    i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8,
+    i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8, i8,
+);
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i16x2(i16, i16);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i16x4(i16, i16, i16, i16);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i16x8(
+    i16, i16, i16, i16, i16, i16, i16, i16,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i16x16(
+    i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i16x32(
+    i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16,
+    i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16, i16,
+);
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i32x2(i32, i32);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i32x4(i32, i32, i32, i32);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i32x8(
+    i32, i32, i32, i32, i32, i32, i32, i32,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i32x16(
+    i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32, i32,
+);
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i64x2(i64, i64);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i64x4(i64, i64, i64, i64);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i64x8(
+    i64, i64, i64, i64, i64, i64, i64, i64,
+);
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i128x2(i128, i128);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct i128x4(i128, i128, i128, i128);
+
+// unsigned integer types
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u8x2(u8, u8);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u8x4(u8, u8, u8, u8);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u8x8(
+    u8, u8, u8, u8, u8, u8, u8, u8,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u8x16(
+    u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u8x32(
+    u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8,
+    u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u8x64(
+    u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8,
+    u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8,
+    u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8,
+    u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8, u8,
+);
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u16x2(u16, u16);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u16x4(u16, u16, u16, u16);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u16x8(
+    u16, u16, u16, u16, u16, u16, u16, u16,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u16x16(
+    u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u16x32(
+    u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16,
+    u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16, u16,
+);
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u32x2(u32, u32);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u32x4(u32, u32, u32, u32);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u32x8(
+    u32, u32, u32, u32, u32, u32, u32, u32,
+);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u32x16(
+    u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32, u32,
+);
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u64x2(u64, u64);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u64x4(u64, u64, u64, u64);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u64x8(
+    u64, u64, u64, u64, u64, u64, u64, u64,
+);
+
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u128x2(u128, u128);
+#[repr(simd)] #[derive(Copy, Clone)] pub struct u128x4(u128, u128, u128, u128);
+
+extern "platform-intrinsic" {
+    fn simd_saturating_add<T>(x: T, y: T) -> T;
+    fn simd_saturating_sub<T>(x: T, y: T) -> T;
+}
+
+// CHECK-LABEL: @sadd_i8x2
+#[no_mangle]
+pub unsafe fn sadd_i8x2(x: i8x2, y: i8x2) -> i8x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i8> @llvm.sadd.sat.v2i8(<2 x i8> %{{[0-9]+}}, <2 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i8x4
+#[no_mangle]
+pub unsafe fn sadd_i8x4(x: i8x4, y: i8x4) -> i8x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i8> @llvm.sadd.sat.v4i8(<4 x i8> %{{[0-9]+}}, <4 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i8x8
+#[no_mangle]
+pub unsafe fn sadd_i8x8(x: i8x8, y: i8x8) -> i8x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i8> @llvm.sadd.sat.v8i8(<8 x i8> %{{[0-9]+}}, <8 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i8x16
+#[no_mangle]
+pub unsafe fn sadd_i8x16(x: i8x16, y: i8x16) -> i8x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i8> @llvm.sadd.sat.v16i8(<16 x i8> %{{[0-9]+}}, <16 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i8x32
+#[no_mangle]
+pub unsafe fn sadd_i8x32(x: i8x32, y: i8x32) -> i8x32 {
+    // CHECK: %{{[0-9]+}} = call <32 x i8> @llvm.sadd.sat.v32i8(<32 x i8> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i8x64
+#[no_mangle]
+pub unsafe fn sadd_i8x64(x: i8x64, y: i8x64) -> i8x64 {
+    // CHECK: %{{[0-9]+}} = call <64 x i8> @llvm.sadd.sat.v64i8(<64 x i8> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i16x2
+#[no_mangle]
+pub unsafe fn sadd_i16x2(x: i16x2, y: i16x2) -> i16x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i16> @llvm.sadd.sat.v2i16(<2 x i16> %{{[0-9]+}}, <2 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i16x4
+#[no_mangle]
+pub unsafe fn sadd_i16x4(x: i16x4, y: i16x4) -> i16x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i16> @llvm.sadd.sat.v4i16(<4 x i16> %{{[0-9]+}}, <4 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i16x8
+#[no_mangle]
+pub unsafe fn sadd_i16x8(x: i16x8, y: i16x8) -> i16x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i16> @llvm.sadd.sat.v8i16(<8 x i16> %{{[0-9]+}}, <8 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i16x16
+#[no_mangle]
+pub unsafe fn sadd_i16x16(x: i16x16, y: i16x16) -> i16x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i16> @llvm.sadd.sat.v16i16(<16 x i16> %{{[0-9]+}}, <16 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i16x32
+#[no_mangle]
+pub unsafe fn sadd_i16x32(x: i16x32, y: i16x32) -> i16x32 {
+    // CHECK: %{{[0-9]+}} = call <32 x i16> @llvm.sadd.sat.v32i16(<32 x i16> %{{[0-9]+}}, <32 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i32x2
+#[no_mangle]
+pub unsafe fn sadd_i32x2(x: i32x2, y: i32x2) -> i32x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i32> @llvm.sadd.sat.v2i32(<2 x i32> %{{[0-9]+}}, <2 x i32> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i32x4
+#[no_mangle]
+pub unsafe fn sadd_i32x4(x: i32x4, y: i32x4) -> i32x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i32> @llvm.sadd.sat.v4i32(<4 x i32> %{{[0-9]+}}, <4 x i32> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i32x8
+#[no_mangle]
+pub unsafe fn sadd_i32x8(x: i32x8, y: i32x8) -> i32x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i32> @llvm.sadd.sat.v8i32(<8 x i32> %{{[0-9]+}}, <8 x i32> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i32x16
+#[no_mangle]
+pub unsafe fn sadd_i32x16(x: i32x16, y: i32x16) -> i32x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i32> @llvm.sadd.sat.v16i32(<16 x i32> %{{[0-9]+}}, <16 x i32> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i64x2
+#[no_mangle]
+pub unsafe fn sadd_i64x2(x: i64x2, y: i64x2) -> i64x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i64> @llvm.sadd.sat.v2i64(<2 x i64> %{{[0-9]+}}, <2 x i64> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i64x4
+#[no_mangle]
+pub unsafe fn sadd_i64x4(x: i64x4, y: i64x4) -> i64x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i64> @llvm.sadd.sat.v4i64(<4 x i64> %{{[0-9]+}}, <4 x i64> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i64x8
+#[no_mangle]
+pub unsafe fn sadd_i64x8(x: i64x8, y: i64x8) -> i64x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i64> @llvm.sadd.sat.v8i64(<8 x i64> %{{[0-9]+}}, <8 x i64> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i128x2
+#[no_mangle]
+pub unsafe fn sadd_i128x2(x: i128x2, y: i128x2) -> i128x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i128> @llvm.sadd.sat.v2i128(<2 x i128> %{{[0-9]+}}, <2 x i128> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @sadd_i128x4
+#[no_mangle]
+pub unsafe fn sadd_i128x4(x: i128x4, y: i128x4) -> i128x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i128> @llvm.sadd.sat.v4i128(<4 x i128> %{{[0-9]+}}, <4 x i128> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+
+
+// CHECK-LABEL: @uadd_u8x2
+#[no_mangle]
+pub unsafe fn uadd_u8x2(x: u8x2, y: u8x2) -> u8x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i8> @llvm.uadd.sat.v2i8(<2 x i8> %{{[0-9]+}}, <2 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u8x4
+#[no_mangle]
+pub unsafe fn uadd_u8x4(x: u8x4, y: u8x4) -> u8x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i8> @llvm.uadd.sat.v4i8(<4 x i8> %{{[0-9]+}}, <4 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u8x8
+#[no_mangle]
+pub unsafe fn uadd_u8x8(x: u8x8, y: u8x8) -> u8x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i8> @llvm.uadd.sat.v8i8(<8 x i8> %{{[0-9]+}}, <8 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u8x16
+#[no_mangle]
+pub unsafe fn uadd_u8x16(x: u8x16, y: u8x16) -> u8x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i8> @llvm.uadd.sat.v16i8(<16 x i8> %{{[0-9]+}}, <16 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u8x32
+#[no_mangle]
+pub unsafe fn uadd_u8x32(x: u8x32, y: u8x32) -> u8x32 {
+    // CHECK: %{{[0-9]+}} = call <32 x i8> @llvm.uadd.sat.v32i8(<32 x i8> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u8x64
+#[no_mangle]
+pub unsafe fn uadd_u8x64(x: u8x64, y: u8x64) -> u8x64 {
+    // CHECK: %{{[0-9]+}} = call <64 x i8> @llvm.uadd.sat.v64i8(<64 x i8> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u16x2
+#[no_mangle]
+pub unsafe fn uadd_u16x2(x: u16x2, y: u16x2) -> u16x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i16> @llvm.uadd.sat.v2i16(<2 x i16> %{{[0-9]+}}, <2 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u16x4
+#[no_mangle]
+pub unsafe fn uadd_u16x4(x: u16x4, y: u16x4) -> u16x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i16> @llvm.uadd.sat.v4i16(<4 x i16> %{{[0-9]+}}, <4 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u16x8
+#[no_mangle]
+pub unsafe fn uadd_u16x8(x: u16x8, y: u16x8) -> u16x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i16> @llvm.uadd.sat.v8i16(<8 x i16> %{{[0-9]+}}, <8 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u16x16
+#[no_mangle]
+pub unsafe fn uadd_u16x16(x: u16x16, y: u16x16) -> u16x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i16> @llvm.uadd.sat.v16i16(<16 x i16> %{{[0-9]+}}, <16 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u16x32
+#[no_mangle]
+pub unsafe fn uadd_u16x32(x: u16x32, y: u16x32) -> u16x32 {
+    // CHECK: %{{[0-9]+}} = call <32 x i16> @llvm.uadd.sat.v32i16(<32 x i16> %{{[0-9]+}}, <32 x i16> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u32x2
+#[no_mangle]
+pub unsafe fn uadd_u32x2(x: u32x2, y: u32x2) -> u32x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i32> @llvm.uadd.sat.v2i32(<2 x i32> %{{[0-9]+}}, <2 x i32> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u32x4
+#[no_mangle]
+pub unsafe fn uadd_u32x4(x: u32x4, y: u32x4) -> u32x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i32> @llvm.uadd.sat.v4i32(<4 x i32> %{{[0-9]+}}, <4 x i32> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u32x8
+#[no_mangle]
+pub unsafe fn uadd_u32x8(x: u32x8, y: u32x8) -> u32x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i32> @llvm.uadd.sat.v8i32(<8 x i32> %{{[0-9]+}}, <8 x i32> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u32x16
+#[no_mangle]
+pub unsafe fn uadd_u32x16(x: u32x16, y: u32x16) -> u32x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i32> @llvm.uadd.sat.v16i32(<16 x i32> %{{[0-9]+}}, <16 x i32> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u64x2
+#[no_mangle]
+pub unsafe fn uadd_u64x2(x: u64x2, y: u64x2) -> u64x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i64> @llvm.uadd.sat.v2i64(<2 x i64> %{{[0-9]+}}, <2 x i64> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u64x4
+#[no_mangle]
+pub unsafe fn uadd_u64x4(x: u64x4, y: u64x4) -> u64x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i64> @llvm.uadd.sat.v4i64(<4 x i64> %{{[0-9]+}}, <4 x i64> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u64x8
+#[no_mangle]
+pub unsafe fn uadd_u64x8(x: u64x8, y: u64x8) -> u64x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i64> @llvm.uadd.sat.v8i64(<8 x i64> %{{[0-9]+}}, <8 x i64> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u128x2
+#[no_mangle]
+pub unsafe fn uadd_u128x2(x: u128x2, y: u128x2) -> u128x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i128> @llvm.uadd.sat.v2i128(<2 x i128> %{{[0-9]+}}, <2 x i128> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+// CHECK-LABEL: @uadd_u128x4
+#[no_mangle]
+pub unsafe fn uadd_u128x4(x: u128x4, y: u128x4) -> u128x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i128> @llvm.uadd.sat.v4i128(<4 x i128> %{{[0-9]+}}, <4 x i128> %{{[0-9]+}})
+    simd_saturating_add(x, y)
+}
+
+
+
+
+
+// CHECK-LABEL: @ssub_i8x2
+#[no_mangle]
+pub unsafe fn ssub_i8x2(x: i8x2, y: i8x2) -> i8x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i8> @llvm.ssub.sat.v2i8(<2 x i8> %{{[0-9]+}}, <2 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i8x4
+#[no_mangle]
+pub unsafe fn ssub_i8x4(x: i8x4, y: i8x4) -> i8x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i8> @llvm.ssub.sat.v4i8(<4 x i8> %{{[0-9]+}}, <4 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i8x8
+#[no_mangle]
+pub unsafe fn ssub_i8x8(x: i8x8, y: i8x8) -> i8x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i8> @llvm.ssub.sat.v8i8(<8 x i8> %{{[0-9]+}}, <8 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i8x16
+#[no_mangle]
+pub unsafe fn ssub_i8x16(x: i8x16, y: i8x16) -> i8x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i8> @llvm.ssub.sat.v16i8(<16 x i8> %{{[0-9]+}}, <16 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i8x32
+#[no_mangle]
+pub unsafe fn ssub_i8x32(x: i8x32, y: i8x32) -> i8x32 {
+    // CHECK: %{{[0-9]+}} = call <32 x i8> @llvm.ssub.sat.v32i8(<32 x i8> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i8x64
+#[no_mangle]
+pub unsafe fn ssub_i8x64(x: i8x64, y: i8x64) -> i8x64 {
+    // CHECK: %{{[0-9]+}} = call <64 x i8> @llvm.ssub.sat.v64i8(<64 x i8> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i16x2
+#[no_mangle]
+pub unsafe fn ssub_i16x2(x: i16x2, y: i16x2) -> i16x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i16> @llvm.ssub.sat.v2i16(<2 x i16> %{{[0-9]+}}, <2 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i16x4
+#[no_mangle]
+pub unsafe fn ssub_i16x4(x: i16x4, y: i16x4) -> i16x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i16> @llvm.ssub.sat.v4i16(<4 x i16> %{{[0-9]+}}, <4 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i16x8
+#[no_mangle]
+pub unsafe fn ssub_i16x8(x: i16x8, y: i16x8) -> i16x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %{{[0-9]+}}, <8 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i16x16
+#[no_mangle]
+pub unsafe fn ssub_i16x16(x: i16x16, y: i16x16) -> i16x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i16> @llvm.ssub.sat.v16i16(<16 x i16> %{{[0-9]+}}, <16 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i16x32
+#[no_mangle]
+pub unsafe fn ssub_i16x32(x: i16x32, y: i16x32) -> i16x32 {
+    // CHECK: %{{[0-9]+}} = call <32 x i16> @llvm.ssub.sat.v32i16(<32 x i16> %{{[0-9]+}}, <32 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i32x2
+#[no_mangle]
+pub unsafe fn ssub_i32x2(x: i32x2, y: i32x2) -> i32x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i32> @llvm.ssub.sat.v2i32(<2 x i32> %{{[0-9]+}}, <2 x i32> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i32x4
+#[no_mangle]
+pub unsafe fn ssub_i32x4(x: i32x4, y: i32x4) -> i32x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i32> @llvm.ssub.sat.v4i32(<4 x i32> %{{[0-9]+}}, <4 x i32> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i32x8
+#[no_mangle]
+pub unsafe fn ssub_i32x8(x: i32x8, y: i32x8) -> i32x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i32> @llvm.ssub.sat.v8i32(<8 x i32> %{{[0-9]+}}, <8 x i32> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i32x16
+#[no_mangle]
+pub unsafe fn ssub_i32x16(x: i32x16, y: i32x16) -> i32x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i32> @llvm.ssub.sat.v16i32(<16 x i32> %{{[0-9]+}}, <16 x i32> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i64x2
+#[no_mangle]
+pub unsafe fn ssub_i64x2(x: i64x2, y: i64x2) -> i64x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i64> @llvm.ssub.sat.v2i64(<2 x i64> %{{[0-9]+}}, <2 x i64> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i64x4
+#[no_mangle]
+pub unsafe fn ssub_i64x4(x: i64x4, y: i64x4) -> i64x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i64> @llvm.ssub.sat.v4i64(<4 x i64> %{{[0-9]+}}, <4 x i64> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i64x8
+#[no_mangle]
+pub unsafe fn ssub_i64x8(x: i64x8, y: i64x8) -> i64x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i64> @llvm.ssub.sat.v8i64(<8 x i64> %{{[0-9]+}}, <8 x i64> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i128x2
+#[no_mangle]
+pub unsafe fn ssub_i128x2(x: i128x2, y: i128x2) -> i128x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i128> @llvm.ssub.sat.v2i128(<2 x i128> %{{[0-9]+}}, <2 x i128> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @ssub_i128x4
+#[no_mangle]
+pub unsafe fn ssub_i128x4(x: i128x4, y: i128x4) -> i128x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i128> @llvm.ssub.sat.v4i128(<4 x i128> %{{[0-9]+}}, <4 x i128> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+
+
+// CHECK-LABEL: @usub_u8x2
+#[no_mangle]
+pub unsafe fn usub_u8x2(x: u8x2, y: u8x2) -> u8x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i8> @llvm.usub.sat.v2i8(<2 x i8> %{{[0-9]+}}, <2 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u8x4
+#[no_mangle]
+pub unsafe fn usub_u8x4(x: u8x4, y: u8x4) -> u8x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i8> @llvm.usub.sat.v4i8(<4 x i8> %{{[0-9]+}}, <4 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u8x8
+#[no_mangle]
+pub unsafe fn usub_u8x8(x: u8x8, y: u8x8) -> u8x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i8> @llvm.usub.sat.v8i8(<8 x i8> %{{[0-9]+}}, <8 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u8x16
+#[no_mangle]
+pub unsafe fn usub_u8x16(x: u8x16, y: u8x16) -> u8x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i8> @llvm.usub.sat.v16i8(<16 x i8> %{{[0-9]+}}, <16 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u8x32
+#[no_mangle]
+pub unsafe fn usub_u8x32(x: u8x32, y: u8x32) -> u8x32 {
+    // CHECK: %{{[0-9]+}} = call <32 x i8> @llvm.usub.sat.v32i8(<32 x i8> %{{[0-9]+}}, <32 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u8x64
+#[no_mangle]
+pub unsafe fn usub_u8x64(x: u8x64, y: u8x64) -> u8x64 {
+    // CHECK: %{{[0-9]+}} = call <64 x i8> @llvm.usub.sat.v64i8(<64 x i8> %{{[0-9]+}}, <64 x i8> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u16x2
+#[no_mangle]
+pub unsafe fn usub_u16x2(x: u16x2, y: u16x2) -> u16x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i16> @llvm.usub.sat.v2i16(<2 x i16> %{{[0-9]+}}, <2 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u16x4
+#[no_mangle]
+pub unsafe fn usub_u16x4(x: u16x4, y: u16x4) -> u16x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i16> @llvm.usub.sat.v4i16(<4 x i16> %{{[0-9]+}}, <4 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u16x8
+#[no_mangle]
+pub unsafe fn usub_u16x8(x: u16x8, y: u16x8) -> u16x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i16> @llvm.usub.sat.v8i16(<8 x i16> %{{[0-9]+}}, <8 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u16x16
+#[no_mangle]
+pub unsafe fn usub_u16x16(x: u16x16, y: u16x16) -> u16x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i16> @llvm.usub.sat.v16i16(<16 x i16> %{{[0-9]+}}, <16 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u16x32
+#[no_mangle]
+pub unsafe fn usub_u16x32(x: u16x32, y: u16x32) -> u16x32 {
+    // CHECK: %{{[0-9]+}} = call <32 x i16> @llvm.usub.sat.v32i16(<32 x i16> %{{[0-9]+}}, <32 x i16> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u32x2
+#[no_mangle]
+pub unsafe fn usub_u32x2(x: u32x2, y: u32x2) -> u32x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i32> @llvm.usub.sat.v2i32(<2 x i32> %{{[0-9]+}}, <2 x i32> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u32x4
+#[no_mangle]
+pub unsafe fn usub_u32x4(x: u32x4, y: u32x4) -> u32x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i32> @llvm.usub.sat.v4i32(<4 x i32> %{{[0-9]+}}, <4 x i32> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u32x8
+#[no_mangle]
+pub unsafe fn usub_u32x8(x: u32x8, y: u32x8) -> u32x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i32> @llvm.usub.sat.v8i32(<8 x i32> %{{[0-9]+}}, <8 x i32> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u32x16
+#[no_mangle]
+pub unsafe fn usub_u32x16(x: u32x16, y: u32x16) -> u32x16 {
+    // CHECK: %{{[0-9]+}} = call <16 x i32> @llvm.usub.sat.v16i32(<16 x i32> %{{[0-9]+}}, <16 x i32> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u64x2
+#[no_mangle]
+pub unsafe fn usub_u64x2(x: u64x2, y: u64x2) -> u64x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i64> @llvm.usub.sat.v2i64(<2 x i64> %{{[0-9]+}}, <2 x i64> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u64x4
+#[no_mangle]
+pub unsafe fn usub_u64x4(x: u64x4, y: u64x4) -> u64x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i64> @llvm.usub.sat.v4i64(<4 x i64> %{{[0-9]+}}, <4 x i64> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u64x8
+#[no_mangle]
+pub unsafe fn usub_u64x8(x: u64x8, y: u64x8) -> u64x8 {
+    // CHECK: %{{[0-9]+}} = call <8 x i64> @llvm.usub.sat.v8i64(<8 x i64> %{{[0-9]+}}, <8 x i64> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u128x2
+#[no_mangle]
+pub unsafe fn usub_u128x2(x: u128x2, y: u128x2) -> u128x2 {
+    // CHECK: %{{[0-9]+}} = call <2 x i128> @llvm.usub.sat.v2i128(<2 x i128> %{{[0-9]+}}, <2 x i128> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
+
+// CHECK-LABEL: @usub_u128x4
+#[no_mangle]
+pub unsafe fn usub_u128x4(x: u128x4, y: u128x4) -> u128x4 {
+    // CHECK: %{{[0-9]+}} = call <4 x i128> @llvm.usub.sat.v4i128(<4 x i128> %{{[0-9]+}}, <4 x i128> %{{[0-9]+}})
+    simd_saturating_sub(x, y)
+}
diff --git a/src/test/codegen/simd-intrinsic-generic-bitmask.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-bitmask.rs
index cd8130f9231..cd8130f9231 100644
--- a/src/test/codegen/simd-intrinsic-generic-bitmask.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-bitmask.rs
diff --git a/src/test/codegen/simd-intrinsic-generic-gather.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-gather.rs
index 3389104219d..3389104219d 100644
--- a/src/test/codegen/simd-intrinsic-generic-gather.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-gather.rs
diff --git a/src/test/codegen/simd-intrinsic-generic-scatter.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-scatter.rs
index dd0a9801bc5..dd0a9801bc5 100644
--- a/src/test/codegen/simd-intrinsic-generic-scatter.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-scatter.rs
diff --git a/src/test/codegen/simd-intrinsic-generic-select.rs b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-select.rs
index 03bb2265526..03bb2265526 100644
--- a/src/test/codegen/simd-intrinsic-generic-select.rs
+++ b/src/test/codegen/simd-intrinsic/simd-intrinsic-generic-select.rs
diff --git a/src/test/run-pass/simd/simd-intrinsic-generic-arithmetic-saturating.rs b/src/test/run-pass/simd/simd-intrinsic-generic-arithmetic-saturating.rs
new file mode 100644
index 00000000000..b2ddcf023eb
--- /dev/null
+++ b/src/test/run-pass/simd/simd-intrinsic-generic-arithmetic-saturating.rs
@@ -0,0 +1,92 @@
+// run-pass
+// ignore-emscripten
+// min-llvm-version 8.0
+
+#![allow(non_camel_case_types)]
+#![feature(repr_simd, platform_intrinsics)]
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+struct u32x4(pub u32, pub u32, pub u32, pub u32);
+
+#[repr(simd)]
+#[derive(Copy, Clone, PartialEq, Debug)]
+struct i32x4(pub i32, pub i32, pub i32, pub i32);
+
+extern "platform-intrinsic" {
+    fn simd_saturating_add<T>(x: T, y: T) -> T;
+    fn simd_saturating_sub<T>(x: T, y: T) -> T;
+}
+
+fn main() {
+    // unsigned
+    {
+        const M: u32 = u32::max_value();
+
+        let a = u32x4(1, 2, 3, 4);
+        let b = u32x4(2, 4, 6, 8);
+        let m = u32x4(M, M, M, M);
+        let m1 = u32x4(M - 1, M - 1, M - 1, M - 1);
+        let z = u32x4(0, 0, 0, 0);
+
+        unsafe {
+            assert_eq!(simd_saturating_add(z, z), z);
+            assert_eq!(simd_saturating_add(z, a), a);
+            assert_eq!(simd_saturating_add(b, z), b);
+            assert_eq!(simd_saturating_add(a, a), b);
+            assert_eq!(simd_saturating_add(a, m), m);
+            assert_eq!(simd_saturating_add(m, b), m);
+            assert_eq!(simd_saturating_add(m1, a), m);
+
+            assert_eq!(simd_saturating_sub(b, z), b);
+            assert_eq!(simd_saturating_sub(b, a), a);
+            assert_eq!(simd_saturating_sub(a, a), z);
+            assert_eq!(simd_saturating_sub(a, b), z);
+            assert_eq!(simd_saturating_sub(a, m1), z);
+            assert_eq!(simd_saturating_sub(b, m1), z);
+        }
+    }
+
+    // signed
+    {
+        const MIN: i32 = i32::min_value();
+        const MAX: i32 = i32::max_value();
+
+        let a = i32x4(1, 2, 3, 4);
+        let b = i32x4(2, 4, 6, 8);
+        let c = i32x4(-1, -2, -3, -4);
+        let d = i32x4(-2, -4, -6, -8);
+
+        let max = i32x4(MAX, MAX, MAX, MAX);
+        let max1 = i32x4(MAX - 1, MAX - 1, MAX - 1, MAX - 1);
+        let min = i32x4(MIN, MIN, MIN, MIN);
+        let min1 = i32x4(MIN + 1, MIN + 1, MIN + 1, MIN + 1);
+
+        let z = i32x4(0, 0, 0, 0);
+
+        unsafe {
+            assert_eq!(simd_saturating_add(z, z), z);
+            assert_eq!(simd_saturating_add(z, a), a);
+            assert_eq!(simd_saturating_add(b, z), b);
+            assert_eq!(simd_saturating_add(a, a), b);
+            assert_eq!(simd_saturating_add(a, max), max);
+            assert_eq!(simd_saturating_add(max, b), max);
+            assert_eq!(simd_saturating_add(max1, a), max);
+            assert_eq!(simd_saturating_add(min1, z), min1);
+            assert_eq!(simd_saturating_add(min, z), min);
+            assert_eq!(simd_saturating_add(min1, c), min);
+            assert_eq!(simd_saturating_add(min, c), min);
+            assert_eq!(simd_saturating_add(min1, d), min);
+            assert_eq!(simd_saturating_add(min, d), min);
+
+            assert_eq!(simd_saturating_sub(b, z), b);
+            assert_eq!(simd_saturating_sub(b, a), a);
+            assert_eq!(simd_saturating_sub(a, a), z);
+            assert_eq!(simd_saturating_sub(a, b), c);
+            assert_eq!(simd_saturating_sub(z, max), min1);
+            assert_eq!(simd_saturating_sub(min1, z), min1);
+            assert_eq!(simd_saturating_sub(min1, a), min);
+            assert_eq!(simd_saturating_sub(min1, b), min);
+        }
+    }
+}
diff --git a/src/test/ui/simd-intrinsic/simd-intrinsic-generic-arithmetic-saturating.rs b/src/test/ui/simd-intrinsic/simd-intrinsic-generic-arithmetic-saturating.rs
new file mode 100644
index 00000000000..65cf31f2f1f
--- /dev/null
+++ b/src/test/ui/simd-intrinsic/simd-intrinsic-generic-arithmetic-saturating.rs
@@ -0,0 +1,38 @@
+// ignore-emscripten
+// ignore-tidy-linelength
+#![feature(repr_simd, platform_intrinsics)]
+#![allow(non_camel_case_types)]
+#[repr(simd)]
+#[derive(Copy, Clone)]
+pub struct i32x4(pub i32, pub i32, pub i32, pub i32);
+
+#[repr(simd)]
+#[derive(Copy, Clone)]
+pub struct x4<T>(pub T, pub T, pub T, pub T);
+
+#[repr(simd)]
+#[derive(Copy, Clone)]
+pub struct f32x4(pub f32, pub f32, pub f32, pub f32);
+
+extern "platform-intrinsic" {
+    fn simd_saturating_add<T>(x: T, y: T) -> T;
+    fn simd_saturating_sub<T>(x: T, y: T) -> T;
+}
+
+fn main() {
+    let x = i32x4(0, 0, 0, 0);
+    let y = x4(0_usize, 0, 0, 0);
+    let z = f32x4(0.0, 0.0, 0.0, 0.0);
+
+    unsafe {
+        simd_saturating_add(x, x);
+        simd_saturating_add(y, y);
+        simd_saturating_sub(x, x);
+        simd_saturating_sub(y, y);
+
+        simd_saturating_add(z, z);
+        //~^ ERROR expected element type `f32` of vector type `f32x4` to be a signed or unsigned integer type
+        simd_saturating_sub(z, z);
+        //~^ ERROR expected element type `f32` of vector type `f32x4` to be a signed or unsigned integer type
+    }
+}
diff --git a/src/test/ui/simd-intrinsic/simd-intrinsic-generic-arithmetic-saturating.stderr b/src/test/ui/simd-intrinsic/simd-intrinsic-generic-arithmetic-saturating.stderr
new file mode 100644
index 00000000000..0e88540bcc8
--- /dev/null
+++ b/src/test/ui/simd-intrinsic/simd-intrinsic-generic-arithmetic-saturating.stderr
@@ -0,0 +1,15 @@
+error[E0511]: invalid monomorphization of `simd_saturating_add` intrinsic: expected element type `f32` of vector type `f32x4` to be a signed or unsigned integer type
+  --> $DIR/simd-intrinsic-generic-arithmetic-saturating.rs:33:9
+   |
+LL |         simd_saturating_add(z, z);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error[E0511]: invalid monomorphization of `simd_saturating_sub` intrinsic: expected element type `f32` of vector type `f32x4` to be a signed or unsigned integer type
+  --> $DIR/simd-intrinsic-generic-arithmetic-saturating.rs:35:9
+   |
+LL |         simd_saturating_sub(z, z);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error: aborting due to 2 previous errors
+
+For more information about this error, try `rustc --explain E0511`.