Auto merge of rust-lang#120718 - saethlin:reasonable-fast-math, r=nnethercote

Add "algebraic" fast-math intrinsics, based on fast-math ops that cannot return poison

Setting all of LLVM's fast-math flags makes our fast-math intrinsics very dangerous, because some inputs are UB. This set of flags permits common algebraic transformations, but according to the [LangRef](https://llvm.org/docs/LangRef.html#fastmath), only the flags `nnan` (no nans) and `ninf` (no infs) can produce poison.

And this uses the algebraic float ops to fix rust-lang#120720

cc `@orlp`

Author: bors

src/intrinsics/mod.rs

@@ -1152,17 +1152,26 @@ fn codegen_regular_intrinsic_call<'tcx>(
             ret.write_cvalue(fx, ret_val);
         }
 
-        sym::fadd_fast | sym::fsub_fast | sym::fmul_fast | sym::fdiv_fast | sym::frem_fast => {
+        sym::fadd_fast
+        | sym::fsub_fast
+        | sym::fmul_fast
+        | sym::fdiv_fast
+        | sym::frem_fast
+        | sym::fadd_algebraic
+        | sym::fsub_algebraic
+        | sym::fmul_algebraic
+        | sym::fdiv_algebraic
+        | sym::frem_algebraic => {
             intrinsic_args!(fx, args => (x, y); intrinsic);
 
             let res = crate::num::codegen_float_binop(
                 fx,
                 match intrinsic {
-                    sym::fadd_fast => BinOp::Add,
-                    sym::fsub_fast => BinOp::Sub,
-                    sym::fmul_fast => BinOp::Mul,
-                    sym::fdiv_fast => BinOp::Div,
-                    sym::frem_fast => BinOp::Rem,
+                    sym::fadd_fast | sym::fadd_algebraic => BinOp::Add,
+                    sym::fsub_fast | sym::fsub_algebraic => BinOp::Sub,
+                    sym::fmul_fast | sym::fmul_algebraic => BinOp::Mul,
+                    sym::fdiv_fast | sym::fdiv_algebraic => BinOp::Div,
+                    sym::frem_fast | sym::frem_algebraic => BinOp::Rem,
                     _ => unreachable!(),
                 },
                 x,