Added implementation for fmin/fmax for aarch64.

The two built-in function map to the fmax and fmin instruction.
Bug 30035

Author: bschommer

aarch64/Asm.v

@@ -283,6 +283,8 @@ Inductive instruction: Type :=
   | Pfmsub (sz: fsize) (rd r1 r2 r3: freg)                            (**r [rd = r3 - r1 * r2] *)
   | Pfnmadd (sz: fsize) (rd r1 r2 r3: freg)                           (**r [rd = - r3 - r1 * r2] *)
   | Pfnmsub (sz: fsize) (rd r1 r2 r3: freg)                           (**r [rd = - r3 + r1 * r2] *)
+  | Pfmax (sz: fsize) (rd r1 r2: freg)                                (**r maximum *)
+  | Pfmin (sz: fsize) (rd r1 r2: freg)                                (**r minimum *)
   (** Floating-point comparison *)
   | Pfcmp (sz: fsize) (r1 r2: freg)                                   (**r compare [r1] and [r2] *)
   | Pfcmp0 (sz: fsize) (r1: freg)                                     (**r compare [r1] and [+0.0] *)
@@ -1114,6 +1116,8 @@ Definition exec_instr (f: function) (i: instruction) (rs: regset) (m: mem) : out
   | Pfmsub _ _ _ _ _
   | Pfnmadd _ _ _ _ _
   | Pfnmsub _ _ _ _ _
+  | Pfmax _ _ _ _
+  | Pfmin _ _ _ _
   | Pnop
   | Pcfi_adjust _ 
   | Pcfi_rel_offset _ =>

aarch64/Asmexpand.ml

@@ -363,6 +363,10 @@ let expand_builtin_inline name args res =
       emit (Pfnmadd(D, res, a1, a2, a3))
   | "__builtin_fnmsub", [BA(FR a1); BA(FR a2); BA(FR a3)], BR(FR res) ->
       emit (Pfnmsub(D, res, a1, a2, a3))
+  | "__builtin_fmax", [BA (FR a1); BA (FR a2)], BR (FR res) ->
+      emit (Pfmax (D, res, a1, a2))
+  | "__builtin_fmin", [BA (FR a1); BA (FR a2)], BR (FR res) ->
+      emit (Pfmin (D, res, a1, a2))
   (* Vararg *)
   | "__builtin_va_start", [BA(IR a)], _ ->
       expand_builtin_va_start a

aarch64/TargetPrinter.ml

@@ -467,6 +467,10 @@ module Target : TARGET =
         fprintf oc "	fnmadd	%a, %a, %a, %a\n" freg (sz, rd) freg (sz, r1) freg (sz, r2) freg (sz, r3)
     | Pfnmsub(sz, rd, r1, r2, r3) ->
         fprintf oc "	fnmsub	%a, %a, %a, %a\n" freg (sz, rd) freg (sz, r1) freg (sz, r2) freg (sz, r3)
+    | Pfmax (sz, rd, r1, r2) ->
+        fprintf oc "	fmax	%a, %a, %a\n" freg (sz, rd) freg (sz, r1) freg (sz, r2)
+    | Pfmin (sz, rd, r1, r2) ->
+        fprintf oc "	fmin	%a, %a, %a\n" freg (sz, rd) freg (sz, r1) freg (sz, r2)
     (* Floating-point comparison *)
     | Pfcmp(sz, r1, r2) ->
         fprintf oc "	fcmp	%a, %a\n" freg (sz, r1) freg (sz, r2)