This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Arch-independent double-word clz expansion and a few other tweaks for bitscanning

Rask Ingemann Lambertsen wrote:
On Fri, Aug 10, 2007 at 11:45:33PM -0700, Zack Weinberg wrote: 
==================================================================
--- config/i386/i386.md	(revision 127388)
+++ config/i386/i386.md	(local)
@@ -14583,98 +14583,16 @@ (define_insn "leave_rex64"
... 
+(define_insn "ctzsi2"
+  [(set (match_operand:SI 0 "register_operand" "=r")
+	(ctz:SI (match_operand:SI 1 "nonimmediate_operand" "rm")))
+   (set (reg:CCZ FLAGS_REG)
+        (compare:CCZ (match_dup 1) (const_int 0)))]


   Usually this sort of insn has the (set (reg:CCx FLAGS_REG) ...) part
first because that's how combine wants it. Likewise with the other ones you
added.

I'm happy to change it around, but empirically, it doesn't seem to make any difference: initial RTL generation produces

(insn 7 6 8 test.c:2 (parallel [
            (set (reg:CCZ 17 flags)
                (compare:CCZ (reg/v:DI 59 [ x ])
                    (const_int 0 [0x0])))
            (set (reg:DI 61)
                (ctz:DI (reg/v:DI 59 [ x ])))
        ]) -1 (nil))

despite that I wrote it the other way around in the insn pattern.

zw

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]