Re: [PATCH v4 5/8] riscv/runtime-const: Introduce runtime_const_mask_32()

[K Prateek Nayak <kprateek.nayak@amd.com>]

Hello Charlie,

On 6/23/2026 10:54 AM, Charlie Jenkins wrote:
> On Thu, 30 Apr 2026 09:47:27 +0000, K Prateek Nayak <kprateek.nayak@amd.com> wrote:
>> Futex hash computation requires a mask operation with read-only after
>> init data that will be converted to a runtime constant in the subsequent
>> commit.
>>
>> Introduce runtime_const_mask_32 to further optimize the mask operation
>> in the futex hash computation hot path. GCC generates a:
>>
>>   lui   a0, 0x12346       # upper; +0x800 then >>12 for correct rounding
>>   addi  a0, a0, 0x678     # lower 12 bits
>>   and   a1, a1, a0        # a1 = a1 & a0
>>
>> pattern to tackle arbitrary 32-bit masks and the same was also suggested
>> by Claude which is implemented here. The final (__ret & val) operation
>> is intentionally placed outside of asm block to allow compilers to
>> further optimize it if possible.
> 
> If the mask fits in 12 bits, we can nop the lui and the addi and just
> patch an "andi" instruction with the 12 bits of the mask. We already do
> this with the lui+addi block and nop the lui if val fits in 12 bits. I
> would be happy to help draft that optimization.
> 
> But I think the better solution would be to take the power of 2
> assumption since that will also benefit arm. We should still only emit
> an andi if val fits in 12 bits, but if it doesn't we can patch in
> shifts:
> 
> slli    a0,a0,x
> srli    a0,a0,x
> 
> Where x is the constant (arch_size - _futex_shift - 1)

I can do that for the next version and use ubfx for ARM. I can just put
in a BUG_ON() at the arch/ specific __runtime_fixup_mask() and if a
new use case arises which hits that, we can perhaps move on the dynamic
nop patching scheme that you mentioned earlier.

Let me know if that works and I can pivot to that scheme in v5 and send
it out post -rc1 after some testing.

-- 
Thanks and Regards,
Prateek