Auto merge of #115933 - oli-obk:simd_shuffle_const, r=workingjubilee

Prototype using const generic for simd_shuffle IDX array

cc https://github.com/rust-lang/rust/issues/85229

r? `@workingjubilee` on the design

TLDR: there is now a `fn simd_shuffle_generic<T, U, const IDX: &'static [u32]>(x: T, y: T) -> U;` intrinsic that allows replacing

```rust
simd_shuffle(a, b, const { stuff })
```

with

```rust
simd_shuffle_generic::<_, _, {&stuff}>(a, b)
```

which makes the compiler implementations much simpler, if we manage to at some point eliminate `simd_shuffle`.

There are some issues with this today though (can't do math without bubbling it up in the generic arguments). With this change, we can start porting the simple cases and get better data on the others.

11 files changed, 313 insertions, 46 deletions

diff --git a/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs b/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs
index c64a4008996..6efbe149863 100644
--- a/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs
+++ b/compiler/rustc_codegen_cranelift/src/intrinsics/simd.rs
@@ -21,7 +21,7 @@ fn report_simd_type_validation_error(
 pub(super) fn codegen_simd_intrinsic_call<'tcx>(
     fx: &mut FunctionCx<'_, '_, 'tcx>,
     intrinsic: Symbol,
-    _args: GenericArgsRef<'tcx>,
+    generic_args: GenericArgsRef<'tcx>,
     args: &[mir::Operand<'tcx>],
     ret: CPlace<'tcx>,
     target: BasicBlock,
@@ -117,6 +117,54 @@ pub(super) fn codegen_simd_intrinsic_call<'tcx>(
             });
         }
 
+        // simd_shuffle_generic<T, U, const I: &[u32]>(x: T, y: T) -> U
+        sym::simd_shuffle_generic => {
+            let [x, y] = args else {
+                bug!("wrong number of args for intrinsic {intrinsic}");
+            };
+            let x = codegen_operand(fx, x);
+            let y = codegen_operand(fx, y);
+
+            if !x.layout().ty.is_simd() {
+                report_simd_type_validation_error(fx, intrinsic, span, x.layout().ty);
+                return;
+            }
+
+            let idx = generic_args[2]
+                .expect_const()
+                .eval(fx.tcx, ty::ParamEnv::reveal_all(), Some(span))
+                .unwrap()
+                .unwrap_branch();
+
+            assert_eq!(x.layout(), y.layout());
+            let layout = x.layout();
+
+            let (lane_count, lane_ty) = layout.ty.simd_size_and_type(fx.tcx);
+            let (ret_lane_count, ret_lane_ty) = ret.layout().ty.simd_size_and_type(fx.tcx);
+
+            assert_eq!(lane_ty, ret_lane_ty);
+            assert_eq!(idx.len() as u64, ret_lane_count);
+
+            let total_len = lane_count * 2;
+
+            let indexes =
+                idx.iter().map(|idx| idx.unwrap_leaf().try_to_u16().unwrap()).collect::<Vec<u16>>();
+
+            for &idx in &indexes {
+                assert!(u64::from(idx) < total_len, "idx {} out of range 0..{}", idx, total_len);
+            }
+
+            for (out_idx, in_idx) in indexes.into_iter().enumerate() {
+                let in_lane = if u64::from(in_idx) < lane_count {
+                    x.value_lane(fx, in_idx.into())
+                } else {
+                    y.value_lane(fx, u64::from(in_idx) - lane_count)
+                };
+                let out_lane = ret.place_lane(fx, u64::try_from(out_idx).unwrap());
+                out_lane.write_cvalue(fx, in_lane);
+            }
+        }
+
         // simd_shuffle<T, I, U>(x: T, y: T, idx: I) -> U
         sym::simd_shuffle => {
             let (x, y, idx) = match args {
diff --git a/compiler/rustc_codegen_llvm/src/intrinsic.rs b/compiler/rustc_codegen_llvm/src/intrinsic.rs
index 9289c37d763..a97b803fc64 100644
--- a/compiler/rustc_codegen_llvm/src/intrinsic.rs
+++ b/compiler/rustc_codegen_llvm/src/intrinsic.rs
@@ -15,7 +15,7 @@ use rustc_codegen_ssa::mir::place::PlaceRef;
 use rustc_codegen_ssa::traits::*;
 use rustc_hir as hir;
 use rustc_middle::ty::layout::{FnAbiOf, HasTyCtxt, LayoutOf};
-use rustc_middle::ty::{self, Ty};
+use rustc_middle::ty::{self, GenericArgsRef, Ty};
 use rustc_middle::{bug, span_bug};
 use rustc_span::{sym, symbol::kw, Span, Symbol};
 use rustc_target::abi::{self, Align, HasDataLayout, Primitive};
@@ -376,7 +376,9 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> {
             }
 
             _ if name.as_str().starts_with("simd_") => {
-                match generic_simd_intrinsic(self, name, callee_ty, args, ret_ty, llret_ty, span) {
+                match generic_simd_intrinsic(
+                    self, name, callee_ty, fn_args, args, ret_ty, llret_ty, span,
+                ) {
                     Ok(llval) => llval,
                     Err(()) => return,
                 }
@@ -911,6 +913,7 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
     bx: &mut Builder<'_, 'll, 'tcx>,
     name: Symbol,
     callee_ty: Ty<'tcx>,
+    fn_args: GenericArgsRef<'tcx>,
     args: &[OperandRef<'tcx, &'ll Value>],
     ret_ty: Ty<'tcx>,
     llret_ty: &'ll Type,
@@ -1030,6 +1033,56 @@ fn generic_simd_intrinsic<'ll, 'tcx>(
         ));
     }
 
+    if name == sym::simd_shuffle_generic {
+        let idx = fn_args[2]
+            .expect_const()
+            .eval(tcx, ty::ParamEnv::reveal_all(), Some(span))
+            .unwrap()
+            .unwrap_branch();
+        let n = idx.len() as u64;
+
+        require_simd!(ret_ty, InvalidMonomorphization::SimdReturn { span, name, ty: ret_ty });
+        let (out_len, out_ty) = ret_ty.simd_size_and_type(bx.tcx());
+        require!(
+            out_len == n,
+            InvalidMonomorphization::ReturnLength { span, name, in_len: n, ret_ty, out_len }
+        );
+        require!(
+            in_elem == out_ty,
+            InvalidMonomorphization::ReturnElement { span, name, in_elem, in_ty, ret_ty, out_ty }
+        );
+
+        let total_len = in_len * 2;
+
+        let indices: Option<Vec<_>> = idx
+            .iter()
+            .enumerate()
+            .map(|(arg_idx, val)| {
+                let idx = val.unwrap_leaf().try_to_i32().unwrap();
+                if idx >= i32::try_from(total_len).unwrap() {
+                    bx.sess().emit_err(InvalidMonomorphization::ShuffleIndexOutOfBounds {
+                        span,
+                        name,
+                        arg_idx: arg_idx as u64,
+                        total_len: total_len.into(),
+                    });
+                    None
+                } else {
+                    Some(bx.const_i32(idx))
+                }
+            })
+            .collect();
+        let Some(indices) = indices else {
+            return Ok(bx.const_null(llret_ty));
+        };
+
+        return Ok(bx.shuffle_vector(
+            args[0].immediate(),
+            args[1].immediate(),
+            bx.const_vector(&indices),
+        ));
+    }
+
     if name == sym::simd_shuffle {
         // Make sure this is actually an array, since typeck only checks the length-suffixed
         // version of this intrinsic.
diff --git a/compiler/rustc_hir_analysis/src/check/intrinsic.rs b/compiler/rustc_hir_analysis/src/check/intrinsic.rs
index 3c0d977917f..c61719c1fd2 100644
--- a/compiler/rustc_hir_analysis/src/check/intrinsic.rs
+++ b/compiler/rustc_hir_analysis/src/check/intrinsic.rs
@@ -20,6 +20,7 @@ fn equate_intrinsic_type<'tcx>(
     it: &hir::ForeignItem<'_>,
     n_tps: usize,
     n_lts: usize,
+    n_cts: usize,
     sig: ty::PolyFnSig<'tcx>,
 ) {
     let (own_counts, span) = match &it.kind {
@@ -51,7 +52,7 @@ fn equate_intrinsic_type<'tcx>(
 
     if gen_count_ok(own_counts.lifetimes, n_lts, "lifetime")
         && gen_count_ok(own_counts.types, n_tps, "type")
-        && gen_count_ok(own_counts.consts, 0, "const")
+        && gen_count_ok(own_counts.consts, n_cts, "const")
     {
         let it_def_id = it.owner_id.def_id;
         check_function_signature(
@@ -489,7 +490,7 @@ pub fn check_intrinsic_type(tcx: TyCtxt<'_>, it: &hir::ForeignItem<'_>) {
     };
     let sig = tcx.mk_fn_sig(inputs, output, false, unsafety, Abi::RustIntrinsic);
     let sig = ty::Binder::bind_with_vars(sig, bound_vars);
-    equate_intrinsic_type(tcx, it, n_tps, n_lts, sig)
+    equate_intrinsic_type(tcx, it, n_tps, n_lts, 0, sig)
 }
 
 /// Type-check `extern "platform-intrinsic" { ... }` functions.
@@ -501,9 +502,9 @@ pub fn check_platform_intrinsic_type(tcx: TyCtxt<'_>, it: &hir::ForeignItem<'_>)
 
     let name = it.ident.name;
 
-    let (n_tps, inputs, output) = match name {
+    let (n_tps, n_cts, inputs, output) = match name {
         sym::simd_eq | sym::simd_ne | sym::simd_lt | sym::simd_le | sym::simd_gt | sym::simd_ge => {
-            (2, vec![param(0), param(0)], param(1))
+            (2, 0, vec![param(0), param(0)], param(1))
         }
         sym::simd_add
         | sym::simd_sub
@@ -519,8 +520,8 @@ pub fn check_platform_intrinsic_type(tcx: TyCtxt<'_>, it: &hir::ForeignItem<'_>)
         | sym::simd_fmax
         | sym::simd_fpow
         | sym::simd_saturating_add
-        | sym::simd_saturating_sub => (1, vec![param(0), param(0)], param(0)),
-        sym::simd_arith_offset => (2, vec![param(0), param(1)], param(0)),
+        | sym::simd_saturating_sub => (1, 0, vec![param(0), param(0)], param(0)),
+        sym::simd_arith_offset => (2, 0, vec![param(0), param(1)], param(0)),
         sym::simd_neg
         | sym::simd_bswap
         | sym::simd_bitreverse
@@ -538,25 +539,25 @@ pub fn check_platform_intrinsic_type(tcx: TyCtxt<'_>, it: &hir::ForeignItem<'_>)
         | sym::simd_ceil
         | sym::simd_floor
         | sym::simd_round
-        | sym::simd_trunc => (1, vec![param(0)], param(0)),
-        sym::simd_fpowi => (1, vec![param(0), tcx.types.i32], param(0)),
-        sym::simd_fma => (1, vec![param(0), param(0), param(0)], param(0)),
-        sym::simd_gather => (3, vec![param(0), param(1), param(2)], param(0)),
-        sym::simd_scatter => (3, vec![param(0), param(1), param(2)], Ty::new_unit(tcx)),
-        sym::simd_insert => (2, vec![param(0), tcx.types.u32, param(1)], param(0)),
-        sym::simd_extract => (2, vec![param(0), tcx.types.u32], param(1)),
+        | sym::simd_trunc => (1, 0, vec![param(0)], param(0)),
+        sym::simd_fpowi => (1, 0, vec![param(0), tcx.types.i32], param(0)),
+        sym::simd_fma => (1, 0, vec![param(0), param(0), param(0)], param(0)),
+        sym::simd_gather => (3, 0, vec![param(0), param(1), param(2)], param(0)),
+        sym::simd_scatter => (3, 0, vec![param(0), param(1), param(2)], Ty::new_unit(tcx)),
+        sym::simd_insert => (2, 0, vec![param(0), tcx.types.u32, param(1)], param(0)),
+        sym::simd_extract => (2, 0, vec![param(0), tcx.types.u32], param(1)),
         sym::simd_cast
         | sym::simd_as
         | sym::simd_cast_ptr
         | sym::simd_expose_addr
-        | sym::simd_from_exposed_addr => (2, vec![param(0)], param(1)),
-        sym::simd_bitmask => (2, vec![param(0)], param(1)),
+        | sym::simd_from_exposed_addr => (2, 0, vec![param(0)], param(1)),
+        sym::simd_bitmask => (2, 0, vec![param(0)], param(1)),
         sym::simd_select | sym::simd_select_bitmask => {
-            (2, vec![param(0), param(1), param(1)], param(1))
+            (2, 0, vec![param(0), param(1), param(1)], param(1))
         }
-        sym::simd_reduce_all | sym::simd_reduce_any => (1, vec![param(0)], tcx.types.bool),
+        sym::simd_reduce_all | sym::simd_reduce_any => (1, 0, vec![param(0)], tcx.types.bool),
         sym::simd_reduce_add_ordered | sym::simd_reduce_mul_ordered => {
-            (2, vec![param(0), param(1)], param(1))
+            (2, 0, vec![param(0), param(1)], param(1))
         }
         sym::simd_reduce_add_unordered
         | sym::simd_reduce_mul_unordered
@@ -566,8 +567,9 @@ pub fn check_platform_intrinsic_type(tcx: TyCtxt<'_>, it: &hir::ForeignItem<'_>)
         | sym::simd_reduce_min
         | sym::simd_reduce_max
         | sym::simd_reduce_min_nanless
-        | sym::simd_reduce_max_nanless => (2, vec![param(0)], param(1)),
-        sym::simd_shuffle => (3, vec![param(0), param(0), param(1)], param(2)),
+        | sym::simd_reduce_max_nanless => (2, 0, vec![param(0)], param(1)),
+        sym::simd_shuffle => (3, 0, vec![param(0), param(0), param(1)], param(2)),
+        sym::simd_shuffle_generic => (2, 1, vec![param(0), param(0)], param(1)),
         _ => {
             let msg = format!("unrecognized platform-specific intrinsic function: `{name}`");
             tcx.sess.struct_span_err(it.span, msg).emit();
@@ -577,5 +579,5 @@ pub fn check_platform_intrinsic_type(tcx: TyCtxt<'_>, it: &hir::ForeignItem<'_>)
 
     let sig = tcx.mk_fn_sig(inputs, output, false, hir::Unsafety::Unsafe, Abi::PlatformIntrinsic);
     let sig = ty::Binder::dummy(sig);
-    equate_intrinsic_type(tcx, it, n_tps, 0, sig)
+    equate_intrinsic_type(tcx, it, n_tps, 0, n_cts, sig)
 }
diff --git a/compiler/rustc_span/src/symbol.rs b/compiler/rustc_span/src/symbol.rs
index 382754be2ca..4f46256626f 100644
--- a/compiler/rustc_span/src/symbol.rs
+++ b/compiler/rustc_span/src/symbol.rs
@@ -1465,6 +1465,7 @@ symbols! {
         simd_shl,
         simd_shr,
         simd_shuffle,
+        simd_shuffle_generic,
         simd_sub,
         simd_trunc,
         simd_xor,
diff --git a/src/tools/miri/src/shims/intrinsics/mod.rs b/src/tools/miri/src/shims/intrinsics/mod.rs
index d54145dbdc7..063f74f989d 100644
--- a/src/tools/miri/src/shims/intrinsics/mod.rs
+++ b/src/tools/miri/src/shims/intrinsics/mod.rs
@@ -60,7 +60,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
         }
 
         // The rest jumps to `ret` immediately.
-        this.emulate_intrinsic_by_name(intrinsic_name, args, dest)?;
+        this.emulate_intrinsic_by_name(intrinsic_name, instance.args, args, dest)?;
 
         trace!("{:?}", this.dump_place(dest));
         this.go_to_block(ret);
@@ -71,6 +71,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
     fn emulate_intrinsic_by_name(
         &mut self,
         intrinsic_name: &str,
+        generic_args: ty::GenericArgsRef<'tcx>,
         args: &[OpTy<'tcx, Provenance>],
         dest: &PlaceTy<'tcx, Provenance>,
     ) -> InterpResult<'tcx> {
@@ -80,7 +81,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
             return this.emulate_atomic_intrinsic(name, args, dest);
         }
         if let Some(name) = intrinsic_name.strip_prefix("simd_") {
-            return this.emulate_simd_intrinsic(name, args, dest);
+            return this.emulate_simd_intrinsic(name, generic_args, args, dest);
         }
 
         match intrinsic_name {
diff --git a/src/tools/miri/src/shims/intrinsics/simd.rs b/src/tools/miri/src/shims/intrinsics/simd.rs
index de0c58de44f..5e488679b81 100644
--- a/src/tools/miri/src/shims/intrinsics/simd.rs
+++ b/src/tools/miri/src/shims/intrinsics/simd.rs
@@ -12,6 +12,7 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
     fn emulate_simd_intrinsic(
         &mut self,
         intrinsic_name: &str,
+        generic_args: ty::GenericArgsRef<'tcx>,
         args: &[OpTy<'tcx, Provenance>],
         dest: &PlaceTy<'tcx, Provenance>,
     ) -> InterpResult<'tcx> {
@@ -488,6 +489,38 @@ pub trait EvalContextExt<'mir, 'tcx: 'mir>: crate::MiriInterpCxExt<'mir, 'tcx> {
                     this.write_immediate(*val, &dest)?;
                 }
             }
+            "shuffle_generic" => {
+                let [left, right] = check_arg_count(args)?;
+                let (left, left_len) = this.operand_to_simd(left)?;
+                let (right, right_len) = this.operand_to_simd(right)?;
+                let (dest, dest_len) = this.place_to_simd(dest)?;
+
+                let index = generic_args[2].expect_const().eval(*this.tcx, this.param_env(), Some(this.tcx.span)).unwrap().unwrap_branch();
+                let index_len = index.len();
+
+                assert_eq!(left_len, right_len);
+                assert_eq!(index_len as u64, dest_len);
+
+                for i in 0..dest_len {
+                    let src_index: u64 = index[i as usize].unwrap_leaf()
+                        .try_to_u32().unwrap()
+                        .into();
+                    let dest = this.project_index(&dest, i)?;
+
+                    let val = if src_index < left_len {
+                        this.read_immediate(&this.project_index(&left, src_index)?)?
+                    } else if src_index < left_len.checked_add(right_len).unwrap() {
+                        let right_idx = src_index.checked_sub(left_len).unwrap();
+                        this.read_immediate(&this.project_index(&right, right_idx)?)?
+                    } else {
+                        span_bug!(
+                            this.cur_span(),
+                            "simd_shuffle index {src_index} is out of bounds for 2 vectors of size {left_len}",
+                        );
+                    };
+                    this.write_immediate(*val, &dest)?;
+                }
+            }
             "shuffle" => {
                 let [left, right, index] = check_arg_count(args)?;
                 let (left, left_len) = this.operand_to_simd(left)?;
diff --git a/src/tools/miri/tests/pass/portable-simd.rs b/src/tools/miri/tests/pass/portable-simd.rs
index ee67a65a4f9..c1f5618c163 100644
--- a/src/tools/miri/tests/pass/portable-simd.rs
+++ b/src/tools/miri/tests/pass/portable-simd.rs
@@ -1,5 +1,6 @@
 //@compile-flags: -Zmiri-strict-provenance
-#![feature(portable_simd, platform_intrinsics)]
+#![feature(portable_simd, platform_intrinsics, adt_const_params, inline_const)]
+#![allow(incomplete_features)]
 use std::simd::*;
 
 extern "platform-intrinsic" {
@@ -390,6 +391,8 @@ fn simd_intrinsics() {
         fn simd_reduce_any<T>(x: T) -> bool;
         fn simd_reduce_all<T>(x: T) -> bool;
         fn simd_select<M, T>(m: M, yes: T, no: T) -> T;
+        fn simd_shuffle_generic<T, U, const IDX: &'static [u32]>(x: T, y: T) -> U;
+        fn simd_shuffle<T, IDX, U>(x: T, y: T, idx: IDX) -> U;
     }
     unsafe {
         // Make sure simd_eq returns all-1 for `true`
@@ -413,6 +416,22 @@ fn simd_intrinsics() {
             simd_select(i8x4::from_array([0, -1, -1, 0]), b, a),
             i32x4::from_array([10, 2, 10, 10])
         );
+        assert_eq!(
+            simd_shuffle_generic::<_, i32x4, {&[3, 1, 0, 2]}>(a, b),
+            a,
+        );
+        assert_eq!(
+            simd_shuffle::<_, _, i32x4>(a, b, const {[3, 1, 0, 2]}),
+            a,
+        );
+        assert_eq!(
+            simd_shuffle_generic::<_, i32x4, {&[7, 5, 4, 6]}>(a, b),
+            i32x4::from_array([4, 2, 1, 10]),
+        );
+        assert_eq!(
+            simd_shuffle::<_, _, i32x4>(a, b, const {[7, 5, 4, 6]}),
+            i32x4::from_array([4, 2, 1, 10]),
+        );
     }
 }
 
diff --git a/tests/ui/simd/intrinsic/generic-elements.rs b/tests/ui/simd/intrinsic/generic-elements.rs
index 0ff2203ec72..6ba93e46f75 100644
--- a/tests/ui/simd/intrinsic/generic-elements.rs
+++ b/tests/ui/simd/intrinsic/generic-elements.rs
@@ -1,6 +1,7 @@
 // build-fail
 
-#![feature(repr_simd, platform_intrinsics, rustc_attrs)]
+#![feature(repr_simd, platform_intrinsics, rustc_attrs, adt_const_params)]
+#![allow(incomplete_features)]
 
 #[repr(simd)]
 #[derive(Copy, Clone)]
@@ -35,6 +36,7 @@ extern "platform-intrinsic" {
     fn simd_extract<T, E>(x: T, idx: u32) -> E;
 
     fn simd_shuffle<T, I, U>(x: T, y: T, idx: I) -> U;
+    fn simd_shuffle_generic<T, U, const IDX: &'static [u32]>(x: T, y: T) -> U;
 }
 
 fn main() {
@@ -71,5 +73,29 @@ fn main() {
         //~^ ERROR expected return type of length 4, found `i32x8` with length 8
         simd_shuffle::<_, _, i32x2>(x, x, IDX8);
         //~^ ERROR expected return type of length 8, found `i32x2` with length 2
+
+        const I2: &[u32] = &[0; 2];
+        simd_shuffle_generic::<i32, i32, I2>(0, 0);
+        //~^ ERROR expected SIMD input type, found non-SIMD `i32`
+        const I4: &[u32] = &[0; 4];
+        simd_shuffle_generic::<i32, i32, I4>(0, 0);
+        //~^ ERROR expected SIMD input type, found non-SIMD `i32`
+        const I8: &[u32] = &[0; 8];
+        simd_shuffle_generic::<i32, i32, I8>(0, 0);
+        //~^ ERROR expected SIMD input type, found non-SIMD `i32`
+
+        simd_shuffle_generic::<_, f32x2, I2>(x, x);
+//~^ ERROR element type `i32` (element of input `i32x4`), found `f32x2` with element type `f32`
+        simd_shuffle_generic::<_, f32x4, I4>(x, x);
+//~^ ERROR element type `i32` (element of input `i32x4`), found `f32x4` with element type `f32`
+        simd_shuffle_generic::<_, f32x8, I8>(x, x);
+//~^ ERROR element type `i32` (element of input `i32x4`), found `f32x8` with element type `f32`
+
+        simd_shuffle_generic::<_, i32x8, I2>(x, x);
+        //~^ ERROR expected return type of length 2, found `i32x8` with length 8
+        simd_shuffle_generic::<_, i32x8, I4>(x, x);
+        //~^ ERROR expected return type of length 4, found `i32x8` with length 8
+        simd_shuffle_generic::<_, i32x2, I8>(x, x);
+        //~^ ERROR expected return type of length 8, found `i32x2` with length 2
     }
 }
diff --git a/tests/ui/simd/intrinsic/generic-elements.stderr b/tests/ui/simd/intrinsic/generic-elements.stderr
index 115d9d4b3f3..26e01344939 100644
--- a/tests/ui/simd/intrinsic/generic-elements.stderr
+++ b/tests/ui/simd/intrinsic/generic-elements.stderr
@@ -1,75 +1,129 @@
 error[E0511]: invalid monomorphization of `simd_insert` intrinsic: expected SIMD input type, found non-SIMD `i32`
-  --> $DIR/generic-elements.rs:44:9
+  --> $DIR/generic-elements.rs:46:9
    |
 LL |         simd_insert(0, 0, 0);
    |         ^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_insert` intrinsic: expected inserted type `i32` (element of input `i32x4`), found `f64`
-  --> $DIR/generic-elements.rs:46:9
+  --> $DIR/generic-elements.rs:48:9
    |
 LL |         simd_insert(x, 0, 1.0);
    |         ^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_extract` intrinsic: expected return type `i32` (element of input `i32x4`), found `f32`
-  --> $DIR/generic-elements.rs:48:9
+  --> $DIR/generic-elements.rs:50:9
    |
 LL |         simd_extract::<_, f32>(x, 0);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_shuffle` intrinsic: expected SIMD input type, found non-SIMD `i32`
-  --> $DIR/generic-elements.rs:52:9
+  --> $DIR/generic-elements.rs:54:9
    |
 LL |         simd_shuffle::<i32, _, i32>(0, 0, IDX2);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_shuffle` intrinsic: expected SIMD input type, found non-SIMD `i32`
-  --> $DIR/generic-elements.rs:55:9
+  --> $DIR/generic-elements.rs:57:9
    |
 LL |         simd_shuffle::<i32, _, i32>(0, 0, IDX4);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_shuffle` intrinsic: expected SIMD input type, found non-SIMD `i32`
-  --> $DIR/generic-elements.rs:58:9
+  --> $DIR/generic-elements.rs:60:9
    |
 LL |         simd_shuffle::<i32, _, i32>(0, 0, IDX8);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_shuffle` intrinsic: expected return element type `i32` (element of input `i32x4`), found `f32x2` with element type `f32`
-  --> $DIR/generic-elements.rs:61:9
+  --> $DIR/generic-elements.rs:63:9
    |
 LL |         simd_shuffle::<_, _, f32x2>(x, x, IDX2);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_shuffle` intrinsic: expected return element type `i32` (element of input `i32x4`), found `f32x4` with element type `f32`
-  --> $DIR/generic-elements.rs:63:9
+  --> $DIR/generic-elements.rs:65:9
    |
 LL |         simd_shuffle::<_, _, f32x4>(x, x, IDX4);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_shuffle` intrinsic: expected return element type `i32` (element of input `i32x4`), found `f32x8` with element type `f32`
-  --> $DIR/generic-elements.rs:65:9
+  --> $DIR/generic-elements.rs:67:9
    |
 LL |         simd_shuffle::<_, _, f32x8>(x, x, IDX8);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_shuffle` intrinsic: expected return type of length 2, found `i32x8` with length 8
-  --> $DIR/generic-elements.rs:68:9
+  --> $DIR/generic-elements.rs:70:9
    |
 LL |         simd_shuffle::<_, _, i32x8>(x, x, IDX2);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_shuffle` intrinsic: expected return type of length 4, found `i32x8` with length 8
-  --> $DIR/generic-elements.rs:70:9
+  --> $DIR/generic-elements.rs:72:9
    |
 LL |         simd_shuffle::<_, _, i32x8>(x, x, IDX4);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 error[E0511]: invalid monomorphization of `simd_shuffle` intrinsic: expected return type of length 8, found `i32x2` with length 2
-  --> $DIR/generic-elements.rs:72:9
+  --> $DIR/generic-elements.rs:74:9
    |
 LL |         simd_shuffle::<_, _, i32x2>(x, x, IDX8);
    |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-error: aborting due to 12 previous errors
+error[E0511]: invalid monomorphization of `simd_shuffle_generic` intrinsic: expected SIMD input type, found non-SIMD `i32`
+  --> $DIR/generic-elements.rs:78:9
+   |
+LL |         simd_shuffle_generic::<i32, i32, I2>(0, 0);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error[E0511]: invalid monomorphization of `simd_shuffle_generic` intrinsic: expected SIMD input type, found non-SIMD `i32`
+  --> $DIR/generic-elements.rs:81:9
+   |
+LL |         simd_shuffle_generic::<i32, i32, I4>(0, 0);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error[E0511]: invalid monomorphization of `simd_shuffle_generic` intrinsic: expected SIMD input type, found non-SIMD `i32`
+  --> $DIR/generic-elements.rs:84:9
+   |
+LL |         simd_shuffle_generic::<i32, i32, I8>(0, 0);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error[E0511]: invalid monomorphization of `simd_shuffle_generic` intrinsic: expected return element type `i32` (element of input `i32x4`), found `f32x2` with element type `f32`
+  --> $DIR/generic-elements.rs:87:9
+   |
+LL |         simd_shuffle_generic::<_, f32x2, I2>(x, x);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error[E0511]: invalid monomorphization of `simd_shuffle_generic` intrinsic: expected return element type `i32` (element of input `i32x4`), found `f32x4` with element type `f32`
+  --> $DIR/generic-elements.rs:89:9
+   |
+LL |         simd_shuffle_generic::<_, f32x4, I4>(x, x);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error[E0511]: invalid monomorphization of `simd_shuffle_generic` intrinsic: expected return element type `i32` (element of input `i32x4`), found `f32x8` with element type `f32`
+  --> $DIR/generic-elements.rs:91:9
+   |
+LL |         simd_shuffle_generic::<_, f32x8, I8>(x, x);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error[E0511]: invalid monomorphization of `simd_shuffle_generic` intrinsic: expected return type of length 2, found `i32x8` with length 8
+  --> $DIR/generic-elements.rs:94:9
+   |
+LL |         simd_shuffle_generic::<_, i32x8, I2>(x, x);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error[E0511]: invalid monomorphization of `simd_shuffle_generic` intrinsic: expected return type of length 4, found `i32x8` with length 8
+  --> $DIR/generic-elements.rs:96:9
+   |
+LL |         simd_shuffle_generic::<_, i32x8, I4>(x, x);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error[E0511]: invalid monomorphization of `simd_shuffle_generic` intrinsic: expected return type of length 8, found `i32x2` with length 2
+  --> $DIR/generic-elements.rs:98:9
+   |
+LL |         simd_shuffle_generic::<_, i32x2, I8>(x, x);
+   |         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+error: aborting due to 21 previous errors
 
 For more information about this error, try `rustc --explain E0511`.
diff --git a/tests/ui/simd/monomorphize-shuffle-index.generic.stderr b/tests/ui/simd/monomorphize-shuffle-index.generic.stderr
new file mode 100644
index 00000000000..fc66b195674
--- /dev/null
+++ b/tests/ui/simd/monomorphize-shuffle-index.generic.stderr
@@ -0,0 +1,12 @@
+error: overly complex generic constant
+  --> $DIR/monomorphize-shuffle-index.rs:29:45
+   |
+LL |         return simd_shuffle_generic::<_, _, { &Self::I }>(a, b);
+   |                                             ^^--------^^
+   |                                               |
+   |                                               pointer casts are not allowed in generic constants
+   |
+   = help: consider moving this anonymous constant into a `const` function
+
+error: aborting due to previous error
+
diff --git a/tests/ui/simd/monomorphize-shuffle-index.rs b/tests/ui/simd/monomorphize-shuffle-index.rs
index 2467baa08b0..db7953f06dd 100644
--- a/tests/ui/simd/monomorphize-shuffle-index.rs
+++ b/tests/ui/simd/monomorphize-shuffle-index.rs
@@ -1,8 +1,14 @@
-//run-pass
-#![feature(repr_simd, platform_intrinsics)]
+//[old]run-pass
+//[generic_with_fn]run-pass
+// revisions: old generic generic_with_fn
+#![feature(repr_simd, platform_intrinsics, adt_const_params, generic_const_exprs)]
+#![allow(incomplete_features)]
 
 extern "platform-intrinsic" {
+    #[cfg(old)]
     fn simd_shuffle<T, I, U>(a: T, b: T, i: I) -> U;
+    #[cfg(any(generic, generic_with_fn))]
+    fn simd_shuffle_generic<T, U, const I: &'static [u32]>(a: T, b: T) -> U;
 }
 
 #[derive(Copy, Clone)]
@@ -11,12 +17,24 @@ struct Simd<T, const N: usize>([T; N]);
 
 trait Shuffle<const N: usize> {
     const I: [u32; N];
-
-    unsafe fn shuffle<T, const M: usize>(&self, a: Simd<T, M>, b: Simd<T, M>) -> Simd<T, N> {
-        simd_shuffle(a, b, Self::I)
+    const J: &'static [u32] = &Self::I;
+
+    unsafe fn shuffle<T, const M: usize>(&self, a: Simd<T, M>, b: Simd<T, M>) -> Simd<T, N>
+    where
+        Thing<{ Self::J }>:,
+    {
+        #[cfg(old)]
+        return simd_shuffle(a, b, Self::I);
+        #[cfg(generic)]
+        return simd_shuffle_generic::<_, _, { &Self::I }>(a, b);
+        //[generic]~^ overly complex generic constant
+        #[cfg(generic_with_fn)]
+        return simd_shuffle_generic::<_, _, { Self::J }>(a, b);
     }
 }
 
+struct Thing<const X: &'static [u32]>;
+
 fn main() {
     struct I1;
     impl Shuffle<4> for I1 {