Auto merge of #121223 - RalfJung:simd-intrinsics, r=Amanieu

intrinsics::simd: add missing functions, avoid UB-triggering fast-math

Turns out stdarch declares a bunch more SIMD intrinsics that are still missing from libcore.
I hope I got the docs and in particular the safety requirements right for these "unordered" and "nanless" intrinsics.

Many of these are unused even in stdarch, but they are implemented in the codegen backend, so we may as well list them here.

r? `@Amanieu`
Cc `@calebzulawski` `@workingjubilee`

2 files changed, 4 insertions, 7 deletions

diff --git a/compiler/rustc_codegen_gcc/src/builder.rs b/compiler/rustc_codegen_gcc/src/builder.rs
index 5f1e4538376..7e2139866f4 100644
--- a/compiler/rustc_codegen_gcc/src/builder.rs
+++ b/compiler/rustc_codegen_gcc/src/builder.rs
@@ -1752,7 +1752,7 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
         self.vector_reduce(src, |a, b, context| context.new_binary_op(None, op, a.get_type(), a, b))
     }
 
-    pub fn vector_reduce_fadd_fast(&mut self, _acc: RValue<'gcc>, _src: RValue<'gcc>) -> RValue<'gcc> {
+    pub fn vector_reduce_fadd_reassoc(&mut self, _acc: RValue<'gcc>, _src: RValue<'gcc>) -> RValue<'gcc> {
         unimplemented!();
     }
 
@@ -1772,7 +1772,7 @@ impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
         unimplemented!();
     }
 
-    pub fn vector_reduce_fmul_fast(&mut self, _acc: RValue<'gcc>, _src: RValue<'gcc>) -> RValue<'gcc> {
+    pub fn vector_reduce_fmul_reassoc(&mut self, _acc: RValue<'gcc>, _src: RValue<'gcc>) -> RValue<'gcc> {
         unimplemented!();
     }
 
diff --git a/compiler/rustc_codegen_gcc/src/intrinsic/simd.rs b/compiler/rustc_codegen_gcc/src/intrinsic/simd.rs
index 9fa978cd2ef..d8091724d86 100644
--- a/compiler/rustc_codegen_gcc/src/intrinsic/simd.rs
+++ b/compiler/rustc_codegen_gcc/src/intrinsic/simd.rs
@@ -989,14 +989,14 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
 
     arith_red!(
         simd_reduce_add_unordered: BinaryOp::Plus,
-        vector_reduce_fadd_fast,
+        vector_reduce_fadd_reassoc,
         false,
         add,
         0.0 // TODO: Use this argument.
     );
     arith_red!(
         simd_reduce_mul_unordered: BinaryOp::Mult,
-        vector_reduce_fmul_fast,
+        vector_reduce_fmul_reassoc,
         false,
         mul,
         1.0
@@ -1041,9 +1041,6 @@ pub fn generic_simd_intrinsic<'a, 'gcc, 'tcx>(
 
     minmax_red!(simd_reduce_min: vector_reduce_min, vector_reduce_fmin);
     minmax_red!(simd_reduce_max: vector_reduce_max, vector_reduce_fmax);
-    // TODO(sadlerap): revisit these intrinsics to generate more optimal reductions
-    minmax_red!(simd_reduce_min_nanless: vector_reduce_min, vector_reduce_fmin);
-    minmax_red!(simd_reduce_max_nanless: vector_reduce_max, vector_reduce_fmax);
 
     macro_rules! bitwise_red {
         ($name:ident : $op:expr, $boolean:expr) => {