Re: [patch V4 09/31] bitops: Add x86-specific parity functions
From: H. Peter Anvin
Date: Mon May 16 2016 - 15:20:41 EST
On May 11, 2016 2:31:39 AM PDT, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>On Wed, May 11, 2016 at 05:16:38PM +0800, zengzhaoxiu@xxxxxxx wrote:
>
>> +static inline unsigned int __arch_parity4(unsigned int w)
>> +{
>> + unsigned int res = 0;
>> +
>> + asm("test $0xf, %1; setpo %b0"
>> + : "+q" (res)
>> + : "r" (w)
>> + : "cc");
>> +
>> + return res;
>> +}
>> +
>> +static inline unsigned int __arch_parity8(unsigned int w)
>> +{
>> + unsigned int res = 0;
>> +
>> + asm("test %1, %1; setpo %b0"
>> + : "+q" (res)
>> + : "r" (w)
>> + : "cc");
>> +
>> + return res;
>> +}
>> +
>> +static inline unsigned int __arch_parity16(unsigned int w)
>> +{
>> + unsigned int res = 0;
>> +
>> + asm("xor %h1, %b1; setpo %b0"
>> + : "+q" (res), "+q" (w)
>> + : : "cc");
>> +
>> + return res;
>> +}
>
>Please use the GEN_*_RMWcc() stuff to avoid the setpo where possible.
>
>> +
>> +#ifdef CONFIG_64BIT
>> +/* popcnt %eax, %eax -- redundant REX prefix for alignment */
>> +#define POPCNT32 ".byte 0xf3,0x40,0x0f,0xb8,0xc0"
>> +/* popcnt %rax, %rax */
>> +#define POPCNT64 ".byte 0xf3,0x48,0x0f,0xb8,0xc0"
>> +#else
>> +/* popcnt %eax, %eax */
>> +#define POPCNT32 ".byte 0xf3,0x0f,0xb8,0xc0"
>> +#endif
>
>Yuck, please don't duplicate stuff like this.
>
>> +
>> +static __always_inline unsigned int __arch_parity32(unsigned int w)
>> +{
>> + unsigned int res;
>> + unsigned int tmp;
>> +
>> + asm(ALTERNATIVE(
>> + " mov %%eax, %1 \n"
>> + " shr $16, %%eax \n"
>> + " xor %1, %%eax \n"
>> + " xor %%ah, %%al \n"
>> + " mov $0, %%eax \n"
>> + " setpo %%al \n",
>> + POPCNT32 " \n"
>> + " and $1, %%eax \n",
>> + X86_FEATURE_POPCNT)
>> + : "=a" (res), "=&r" (tmp)
>> + : "a" (w)
>> + : "cc");
>> +
>> + return res;
>> +}
>
>How many bytes does that end up being? Should we make it a call?
Also, setxx is used with a "bool" or "u8", not " unsigned int"!!!!
--
Sent from my Android device with K-9 Mail. Please excuse brevity and formatting.