246af00412 ("Don't signal errors for (log qnan)") introduced quiet
float comparison VOPs.  On 32-bit ppc the QUIET variant argument is
never passed to the FROB macro, so it defaults to NIL and both
quiet</single-float and quiet=/single-float (and their double-float
counterparts) emit fcmpo instead of fcmpu.

fcmpo signals VXVC when either operand is a NaN, including a quiet NaN,
which is precisely what the quiet comparisons exist to avoid; the VOPs
are therefore indistinguishable from the plain < and = ones.

ppc64/float.lisp passes T here, as do the arm64 and arm backends.  Only
the 32-bit ppc backend was missed.
---
 src/compiler/ppc/float.lisp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/src/compiler/ppc/float.lisp b/src/compiler/ppc/float.lisp
index a7c9312..35fae25 100644
--- src/compiler/ppc/float.lisp
+++ src/compiler/ppc/float.lisp
@@ -438,10 +438,10 @@
                   (:translate ,translate)
                   (:variant :double ,yep ,nope ,quiet)))))
   (frob < :lt :ge </single-float </double-float)
-  (frob quiet< :lt :ge quiet</single-float quiet</double-float)
+  (frob quiet< :lt :ge quiet</single-float quiet</double-float t)
   (frob > :gt :le >/single-float >/double-float)
   (frob = :eq :ne =/single-float =/double-float)
-  (frob quiet= :eq :ne quiet=/single-float quiet=/double-float))
+  (frob quiet= :eq :ne quiet=/single-float quiet=/double-float t))
 
 
 ;;;; Conversion: