[PATCH v7] malloc: Optimize small memory clearing for calloc

[Guo, Wangyang]

On 12/4/2024 3:02 PM, H.J. Lu wrote:
> On Wed, Dec 4, 2024 at 9:45 AM Guo, Wangyang <wangyang.guo@intel.com> wrote:
>>
>> On 12/4/2024 8:16 AM, H.J. Lu wrote:
>>
>> On Wed, Dec 4, 2024 at 7:32 AM Paul Eggert <eggert@cs.ucla.edu> wrote:
>>
>> On 2024-12-02 20:42, H.J. Lu wrote:
>>
>> +  /* Use repeated stores with 1 branch, instead of up to 3.  */
>> +  *(d + 0) = 0;
>> +  *(d + 1) = 0;
>> +  *(d + 2) = 0;
>> +  *(d + nclears - 2) = 0;
>> +  *(d + nclears - 2 + 1) = 0;
>> +  if (nclears > 5)
>> +    {
>> +      *(d + 3) = 0;
>> +      *(d + 3 + 1) = 0;
>> +      *(d + nclears - 4) = 0;
>> +      *(d + nclears - 4 + 1) = 0;
>> +    }
>>
>> The following alternative code has no branches, when I compile with GCC
>> 14 -O2 on x86-64. Would that be a win?
>>
>>     /* Use repeated stores with no branches.  */
>>     int big = 3 * ((nclears + 1) >> 3);
>>     d[0] = 0;
>>     d[1] = 0;
>>     d[2] = 0;
>>     d[big] = 0;
>>     d[big + 1] = 0;
>>     d[nclears - big - 1] = 0;
>>     d[nclears - big] = 0;
>>     d[nclears - 2] = 0;
>>     d[nclears - 2 + 1] = 0;
>>
>> Wangyang,  can you try this against the current master branch?
>>
>> Thanks.
>>
>> I try to apply the alternative code with attached patch, but meet coredump error when running bench-calloc-thread:
>>
>> Fatal glibc error: malloc.c:2601 (sysmalloc): assertion failed: (old_top == initial_top (av) && old_size == 0) || ((unsigned long) (old_size) >= MINSIZE && prev_inuse (old_top) && ((unsigned long) old_end & (pagesize - 1)) == 0)
> 
> Try this.
> 
> --
> H.J.

That works. But looks like the alternative code has no performance 
benefit in Xeon.

Result of bench-calloc-thread benchmark

Test Platform: Xeon-8380
Ratio: New / Original time_per_iteration (Lower is Better)

Threads#   | Ratio
-----------|------
1 thread   | 1.018
4 threads  | 1.005