diff mbox

cgraph offloading error?

Message ID 5633EC2B.3090405@acm.org
State New
Headers show

Commit Message

Nathan Sidwell Oct. 30, 2015, 10:16 p.m. UTC 
On 10/30/15 14:28, Jeff Law wrote:

> So when we don't use src_reg or dst_mode, we'll get a warning about the unused
> variable. I guess this is the first port where HARD_REGNO_NREGS is a constant.

Yeah,  I noticed that when first lookingat the port, but as it wasn't 
(apparently) broken ...

> Second, MOVE_MAX is 4.  That's causing out-of-bounds array access warnings in
> various places.

> There were a variety of other problems associated with MOVE_MAX being smaller
> than a word.   If I change MOVE_MAX to 8, then everything is good.

Makes sense.


Testing attached ...

nathan

Comments

Nathan Sidwell Oct. 30, 2015, 10:17 p.m. UTC | #1 
On 10/30/15 15:16, Nathan Sidwell wrote:

>
> Testing attached ...

... with parens on void cast args ...

nathan
Jeff Law Oct. 30, 2015, 10:30 p.m. UTC | #2 
On 10/30/2015 04:16 PM, Nathan Sidwell wrote:
> On 10/30/15 14:28, Jeff Law wrote:
>
>> So when we don't use src_reg or dst_mode, we'll get a warning about
>> the unused
>> variable. I guess this is the first port where HARD_REGNO_NREGS is a
>> constant.
>
> Yeah,  I noticed that when first lookingat the port, but as it wasn't
> (apparently) broken ...
It's only fatal with -Wjumble-mumble that gets set by config-list.mk, so 
it could have been easily sneaking by.  Certainly wouldn't affect the 
correctness of the compiler though.

>
>> Second, MOVE_MAX is 4.  That's causing out-of-bounds array access
>> warnings in
>> various places.
>
>> There were a variety of other problems associated with MOVE_MAX being
>> smaller
>> than a word.   If I change MOVE_MAX to 8, then everything is good.
>
> Makes sense.
I find myself wondering if that ought to be explicitly sanity-checked 
somewhere.  I'm sure there's a significant number of invariants for 
ports that we could check at build time and hopefully prevent problems 
of this nature in the future.  Though it may not be worth it..

> Testing attached ...
Cool.  I'll go ahead and add nvptx-elf to the list shortly on the 
assumption this stuff will get fixed on way or another.

Jeff
diff mbox

Patch

2015-10-30  Jeff Law <jeff@redhat.com>
	    Nathan Sidwell  <nathan@acm.org>

	* config/nvptx/nvptx.h (HARD_REGNO_NREGS): Avoid warning on unused
	args.
	(MOVE_MAX): Set to 8.

Index: config/nvptx/nvptx.h
===================================================================
--- config/nvptx/nvptx.h	(revision 229595)
+++ config/nvptx/nvptx.h	(working copy)
@@ -88,7 +88,7 @@ 
 #define CALL_USED_REGISTERS				\
   { 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1 }
 
-#define HARD_REGNO_NREGS(regno, mode)	1
+#define HARD_REGNO_NREGS(regno, mode)	((void)regno, (void)mode, 1)
 #define CANNOT_CHANGE_MODE_CLASS(M1, M2, CLS) ((CLS) == RETURN_REG)
 #define HARD_REGNO_MODE_OK(REG, MODE) nvptx_hard_regno_mode_ok (REG, MODE)
 
@@ -356,7 +356,7 @@  struct GTY(()) machine_function
 #define FLOAT_STORE_FLAG_VALUE(MODE) REAL_VALUE_ATOF("1.0", (MODE))
 
 #define CASE_VECTOR_MODE SImode
-#define MOVE_MAX 4
+#define MOVE_MAX 8
 #define MOVE_RATIO(SPEED) 4
 #define TRULY_NOOP_TRUNCATION(outprec, inprec) 1
 #define FUNCTION_MODE QImode