On 09/14/2016 01:56 AM, Paolo Bonzini wrote:
>> I was hoping to avoid an extra RELATIVE relocation in the (normal) PIE
>> case.
>
> There would be no relocation for 0 and NULL, right? GCC would actually
> put them in bss, IIRC.
Ah, I now see your complaint.
You'd prefer "type var = INIT" over "
On 14/09/2016 03:11, Richard Henderson wrote:
> On 09/13/2016 04:21 PM, Paolo Bonzini wrote:
>>
>>
>> On 13/09/2016 22:57, Richard Henderson wrote:
>>> -#if defined(CONFIG_AVX2_OPT) || (defined(CONFIG_CPUID_H) &&
>>> defined(__SSE2__))
>>> -#include
>>> -
>>> +#if (defined(CONFIG_AVX2_OPT) && de
On 09/13/2016 04:21 PM, Paolo Bonzini wrote:
On 13/09/2016 22:57, Richard Henderson wrote:
-#if defined(CONFIG_AVX2_OPT) || (defined(CONFIG_CPUID_H) && defined(__SSE2__))
-#include
-
+#if (defined(CONFIG_AVX2_OPT) && defined(CONFIG_CPUID_H)) || defined(__SSE2__)
Your __SSE2__ version is bet
On 13/09/2016 22:57, Richard Henderson wrote:
> -#if defined(CONFIG_AVX2_OPT) || (defined(CONFIG_CPUID_H) &&
> defined(__SSE2__))
> -#include
> -
> +#if (defined(CONFIG_AVX2_OPT) && defined(CONFIG_CPUID_H)) ||
> defined(__SSE2__)
Your __SSE2__ version is better than mine which required cpuid.
Handle alignment of buffers, so that the vector paths
can be used more often.
Signed-off-by: Richard Henderson
Signed-off-by: Paolo Bonzini
---
util/bufferiszero.c | 236
1 file changed, 162 insertions(+), 74 deletions(-)
Tested with Fedora
