On Thu, 15 Mar 2007, Linus Torvalds wrote:
quoted text
> 
> Same obvious performance problems go for
>
>         case COPY:

As an example, I *think* this patch to zlib-1.2.3 not only generates 
better code, but is (a) shorter and (b) more logical anyway.

Together with Davide's suggestion on using C macro expansion to make most 
of the mode switches simple branches, it might get rid of most of the 
indirect branches (to get rid of them all, you'd have to also find the 
places where we *don't* set a new state, because it stays the same like 
this one, and the ones where we have conditionals on what the mode is 
going to be..

Of course, the zlib sources are pretty horrid for other reasons (K&R 
source code meant to be compiled on 16-bit architectures too). But that's 
a separate issue, and at least shouldn't affect the resulting code 
quality..

		Linus

---
 inflate.c |    3 +--
 1 files changed, 1 insertions(+), 2 deletions(-)

diff --git a/inflate.c b/inflate.c
index 792fdee..fb26f39 100644
--- a/inflate.c
+++ b/inflate.c
@@ -819,7 +819,7 @@ int flush;
             state->mode = COPY;
         case COPY:
             copy = state->length;
-            if (copy) {
+            while (copy) {
                 if (copy > have) copy = have;
                 if (copy > left) copy = left;
                 if (copy == 0) goto inf_leave;
@@ -829,7 +829,6 @@ int flush;
                 left -= copy;
                 put += copy;
                 state->length -= copy;
-                break;
             }
             Tracev((stderr, "inflate:       stored end\n"));
             state->mode = TYPE;
-
To unsubscribe from this list: send the line "unsubscribe git" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html