egraphs: Delete select+fcmp to f{min,max}_pseudo transform (#6859)

This transform is wrong for the inputs in the attached testcase.

It was introduced in #5546 and has been sort of unnoticed until #6843 caused
some wasmtime testcode to fire this optimization rule and produce this
counterexample.

Author: afonso360

cranelift/codegen/src/opts/selects.isle

@@ -44,20 +44,3 @@
 (rule (simplify (bitselect ty @ (multi_lane _ _) (ugt _ x y) y x)) (umin ty x y))
 (rule (simplify (bitselect ty @ (multi_lane _ _) (uge _ x y) y x)) (umin ty x y))
 
-;; For floats convert fcmp lt into pseudo_min and gt into pseudo_max
-;;
-;; fmax_pseudo docs state:
-;; The behaviour for this operations is defined as  fmax_pseudo(a, b) = (a < b) ? b : a, and the behaviour for zero
-;; or NaN inputs follows from the behaviour of < with such inputs.
-;;
-;; That is exactly the operation that we match here!
-(rule (simplify
-       (select ty (fcmp _ (FloatCC.LessThan) x y) x y))
-      (fmin_pseudo ty x y))
-(rule (simplify
-       (select ty (fcmp _ (FloatCC.GreaterThan) x y) x y))
-      (fmax_pseudo ty x y))
-
-;; TODO: perform this same optimization to `f{min,max}_pseudo` for vectors
-;; with the `bitselect` instruction, but the pattern is a bit more complicated
-;; due to most bitselects-over-floats having bitcasts.

cranelift/filetests/filetests/egraph/select.clif

@@ -141,28 +141,3 @@ block0(v0: i32, v1: i32, v2: i32, v3: i32):
 ; check:    v4 = icmp ule v0, v1
 ; check:    v5 = select v4, v2, v3
 ; check:    return v5
-
-
-
-
-function %select_fcmp_gt_to_fmax_pseudo(f32, f32) -> f32 {
-block0(v0: f32, v1: f32):
-    v2 = fcmp gt v0, v1
-    v3 = select v2, v0, v1
-    return v3
-}
-
-; check: block0(v0: f32, v1: f32):
-; check:    v4 = fmax_pseudo v0, v1
-; check:    return v4
-
-function %select_fcmp_lt_to_fmin_pseudo(f32, f32) -> f32 {
-block0(v0: f32, v1: f32):
-    v2 = fcmp lt v0, v1
-    v3 = select v2, v0, v1
-    return v3
-}
-
-; check: block0(v0: f32, v1: f32):
-; check:    v4 = fmin_pseudo v0, v1
-; check:    return v4

cranelift/filetests/filetests/runtests/issue-6859.clif

@@ -0,0 +1,16 @@
+test interpret
+test run
+set opt_level=speed
+target x86_64
+target x86_64 has_avx
+target aarch64
+target s390x
+target riscv64
+
+function %a(f32, f32) -> f32 fast {
+block0(v0: f32, v1: f32):
+    v2 = fcmp lt v0, v1
+    v3 = select v2, v0, v1
+    return v3
+}
+; run: %a(0.0, NaN) == NaN