add codegen test for simd_insert_dyn and simd_extract_dyn

Author: folkertdev

library/core/src/intrinsics/simd.rs

@@ -4,7 +4,7 @@
 
 /// Inserts an element into a vector, returning the updated vector.
 ///
-/// `T` must be a vector with element type `U`, and `idx` must be `const`.
+/// `T` must be a vector with element type `U`, and `idx` must be const.
 ///
 /// # Safety
 ///
@@ -20,7 +20,7 @@ pub unsafe fn simd_insert<T, U>(_x: T, _idx: u32, _val: U) -> T {
 ///
 /// `T` must be a vector with element type `U`.
 ///
-/// If the index is `const`, [`simd_insert`] may emit better assembly.
+/// If the index is `const`, using [`simd_insert`] instead may emit better assembly.
 ///
 /// # Safety
 ///
@@ -35,7 +35,7 @@ pub unsafe fn simd_insert_dyn<T, U>(_x: T, _idx: u32, _val: U) -> T {
 
 /// Extracts an element from a vector.
 ///
-/// `T` must be a vector with element type `U`, and `idx` must be `const`.
+/// `T` must be a vector with element type `U`, and `idx` must be const.
 ///
 /// # Safety
 ///
@@ -51,7 +51,7 @@ pub unsafe fn simd_extract<T, U>(_x: T, _idx: u32) -> U {
 ///
 /// `T` must be a vector with element type `U`.
 ///
-/// If the index is `const`, [`simd_extract`] may emit better assembly.
+/// If the index is `const`, using [`simd_extract`] instead may emit better assembly.
 ///
 /// # Safety
 ///

tests/codegen/simd/extract-insert-dyn.rs

@@ -0,0 +1,88 @@
+//@compile-flags: -Copt-level=3 -Z merge-functions=disabled
+
+#![feature(core_intrinsics, s390x_target_feature, repr_simd)]
+#![no_std]
+#![crate_type = "lib"]
+#![allow(non_camel_case_types)]
+
+// Test that
+//
+// - `core::intrinsics::simd::simd_extract_dyn` is lowered to an `extractelement`
+// - `core::intrinsics::simd::simd_insert_dyn` is lowered to an `insertelement`
+//
+// On many platforms (e.g. x86_64) this will ultimately lower to a pointer load, but some platforms
+// (e.g. s390x, powerpc) have dedicated instructions for these operations.
+
+use core::intrinsics::simd::{simd_extract, simd_extract_dyn, simd_insert, simd_insert_dyn};
+
+#[repr(simd)]
+#[derive(Clone, Copy)]
+pub struct u32x16([u32; 16]);
+
+#[repr(simd)]
+#[derive(Clone, Copy)]
+pub struct i8x16([i8; 16]);
+
+// CHECK-LABEL: dyn_simd_extract
+// CHECK: extractelement <16 x i8> %x, i32 %idx
+// CHECK-NEXT: ret
+#[no_mangle]
+unsafe extern "C" fn dyn_simd_extract(x: i8x16, idx: u32) -> i8 {
+    simd_extract_dyn(x, idx)
+}
+
+// CHECK-LABEL: literal_dyn_simd_extract
+// CHECK: extractelement <16 x i8> %x, i64 7
+// CHECK-NEXT: ret
+#[no_mangle]
+unsafe extern "C" fn literal_dyn_simd_extract(x: i8x16) -> i8 {
+    simd_extract_dyn(x, 7)
+}
+
+// CHECK-LABEL: const_dyn_simd_extract
+// CHECK: extractelement <16 x i8> %x, i64 7
+// CHECK-NEXT: ret
+#[no_mangle]
+unsafe extern "C" fn const_dyn_simd_extract(x: i8x16) -> i8 {
+    simd_extract_dyn(x, const { 3 + 4 })
+}
+
+// CHECK-LABEL: const_simd_extract
+// CHECK: extractelement <16 x i8> %x, i64 7
+// CHECK-NEXT: ret
+#[no_mangle]
+unsafe extern "C" fn const_simd_extract(x: i8x16) -> i8 {
+    simd_extract(x, const { 3 + 4 })
+}
+
+// CHECK-LABEL: dyn_simd_insert
+// CHECK: insertelement <16 x i8> %x, i8 %e, i32 %idx
+// CHECK-NEXT: ret
+#[no_mangle]
+unsafe extern "C" fn dyn_simd_insert(x: i8x16, e: i8, idx: u32) -> i8x16 {
+    simd_insert_dyn(x, idx, e)
+}
+
+// CHECK-LABEL: literal_dyn_simd_insert
+// CHECK: insertelement <16 x i8> %x, i8 %e, i64 7
+// CHECK-NEXT: ret
+#[no_mangle]
+unsafe extern "C" fn literal_dyn_simd_insert(x: i8x16, e: i8) -> i8x16 {
+    simd_insert_dyn(x, 7, e)
+}
+
+// CHECK-LABEL: const_dyn_simd_insert
+// CHECK: insertelement <16 x i8> %x, i8 %e, i64 7
+// CHECK-NEXT: ret
+#[no_mangle]
+unsafe extern "C" fn const_dyn_simd_insert(x: i8x16, e: i8) -> i8x16 {
+    simd_insert_dyn(x, const { 3 + 4 }, e)
+}
+
+// CHECK-LABEL: const_simd_insert
+// CHECK: insertelement <16 x i8> %x, i8 %e, i64 7
+// CHECK-NEXT: ret
+#[no_mangle]
+unsafe extern "C" fn const_simd_insert(x: i8x16, e: i8) -> i8x16 {
+    simd_insert(x, const { 3 + 4 }, e)
+}