Skip to contentSkip to conversation

tests/gem_pwrite_snooped: Fix #pragma GCC weirdness

PATCH In review [email protected] 2 participants 2 messages Oct 16, 2015

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tests/gem_pwrite_snooped: Fix #pragma GCC weirdness

From: Ville Syrjälä <[email protected]>

gem_pwrite_snooped was broken when the #pragma GCC stuff was added in b04691b tests/gem_pwrite_snooped: disable const cast warning

Apparently gcc treats '#pragma GCC' as a C statement. With the current code memchr_inv() pretty much disappears entirely because gcc thinks there's an unconditional return in the loop body. Or at least that's my assumption. Put braces around the if body to fix it.

This is the original asm:
00000000000002f0 <memchr_inv.isra.0>:

 2f0:   48 85 f6                test   %rsi,%rsi
 2f3:   b8 00 00 00 00          mov    $0x0,%eax
 2f8:   48 0f 45 c7             cmovne %rdi,%rax
 2fc:   c3                      retq

This is with the fix:
00000000000002f0 <memchr_inv.constprop.1>:

 2f0:   48 85 f6                test   %rsi,%rsi
 2f3:   74 25                   je     31a <memchr_inv.constprop.1+0x2a>
 2f5:   48 01 fe                add    %rdi,%rsi
 2f8:   80 3f ff                cmpb   $0xff,(%rdi)
 2fb:   48 8d 57 01             lea    0x1(%rdi),%rdx
 2ff:   74 11                   je     312 <memchr_inv.constprop.1+0x22>
 301:   eb 1a                   jmp    31d <memchr_inv.constprop.1+0x2d>
 303:   0f 1f 44 00 00          nopl   0x0(%rax,%rax,1)
 308:   48 83 c2 01             add    $0x1,%rdx
 30c:   80 7a ff ff             cmpb   $0xff,-0x1(%rdx)
 310:   75 0b                   jne    31d <memchr_inv.constprop.1+0x2d>
 312:   48 39 f2                cmp    %rsi,%rdx
 315:   48 89 d7                mov    %rdx,%rdi
 318:   75 ee                   jne    308 <memchr_inv.constprop.1+0x18>
 31a:   31 c0                   xor    %eax,%eax
 31c:   c3                      retq
 31d:   48 89 f8                mov    %rdi,%rax
 320:   c3                      retq

Cc: Thomas Wood <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92227

---
 tests/gem_pwrite_snooped.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
tests/gem_pwrite_snooped.c +2 -1
index 5783e3b..61b5404 100644
@@ -82,11 +82,12 @@ static void *memchr_inv(const void *s, int c, size_t n)
 	unsigned char uc = c;
 
 	while (n--) {
-		if (*us != uc)
+		if (*us != uc) {
 #pragma GCC diagnostic push
 #pragma GCC diagnostic ignored "-Wcast-qual"
 			return (void *) us;
 #pragma GCC diagnostic pop
+		}
 		us++;
 	}
 
Signature · 6 lines
2.4.9

_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw
Quoted 38 lines · On Fri, Oct 16, 2015 at 12:44:05PM +0300, [email protected] wrote:

From: Ville Syrjälä <[email protected]>

gem_pwrite_snooped was broken when the #pragma GCC stuff was added in b04691b tests/gem_pwrite_snooped: disable const cast warning

Apparently gcc treats '#pragma GCC' as a C statement. With the current code memchr_inv() pretty much disappears entirely because gcc thinks there's an unconditional return in the loop body. Or at least that's my assumption. Put braces around the if body to fix it.

This is the original asm:
00000000000002f0 <memchr_inv.isra.0>:

 2f0:   48 85 f6                test   %rsi,%rsi
 2f3:   b8 00 00 00 00          mov    $0x0,%eax
 2f8:   48 0f 45 c7             cmovne %rdi,%rax
 2fc:   c3                      retq

This is with the fix:
00000000000002f0 <memchr_inv.constprop.1>:

 2f0:   48 85 f6                test   %rsi,%rsi
 2f3:   74 25                   je     31a <memchr_inv.constprop.1+0x2a>
 2f5:   48 01 fe                add    %rdi,%rsi
 2f8:   80 3f ff                cmpb   $0xff,(%rdi)
 2fb:   48 8d 57 01             lea    0x1(%rdi),%rdx
 2ff:   74 11                   je     312 <memchr_inv.constprop.1+0x22>
 301:   eb 1a                   jmp    31d <memchr_inv.constprop.1+0x2d>
 303:   0f 1f 44 00 00          nopl   0x0(%rax,%rax,1)
 308:   48 83 c2 01             add    $0x1,%rdx
 30c:   80 7a ff ff             cmpb   $0xff,-0x1(%rdx)
 310:   75 0b                   jne    31d <memchr_inv.constprop.1+0x2d>
 312:   48 39 f2                cmp    %rsi,%rdx
 315:   48 89 d7                mov    %rdx,%rdi
 318:   75 ee                   jne    308 <memchr_inv.constprop.1+0x18>
 31a:   31 c0                   xor    %eax,%eax
 31c:   c3                      retq
 31d:   48 89 f8                mov    %rdi,%rax
 320:   c3                      retq
Quoted 3 lines

Cc: Thomas Wood <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92227

Bah, hangs head in shame. You'll see why.
-Chris

Signature · 5 lines
Chris Wilson, Intel Open Source Technology Centre
_______________________________________________
Intel-gfx mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/intel-gfx