[PATCH 2/2] lib/hash: avoid implicit conversion to 64 bit number

[Andre Muezerie]

On Thu, Jan 23, 2025 at 08:55:29AM +0100, Morten Brørup wrote:
> > From: Andre Muezerie [mailto:andremue at linux.microsoft.com]
> > Sent: Wednesday, 22 January 2025 22.37
> > 
> > On Wed, Jan 22, 2025 at 04:12:49PM +0000, Bruce Richardson wrote:
> > > On Wed, Nov 27, 2024 at 05:53:57PM -0800, Andre Muezerie wrote:
> > > > MSVC issues the warnings below:
> > > >
> > > > 1) ../lib/hash/rte_thash_gf2_poly_math.c(128): warning C4334: '<<':
> > > >     result of 32-bit shift implicitly converted to 64 bits
> > > >     (was 64-bit shift intended?)
> > > >
> > > > The code would be better off by using 64 bit numbers to begin with.
> > > > That eliminates the need for a conversion to 64 bits later.
> > > >
> > > > 2) ../lib/hash/rte_thash.c(568): warning C4334: '<<':
> > > >     result of 32-bit shift implicitly converted to 64 bits
> > > >     (was 64-bit shift intended?)
> > > >
> > > > 1ULL should be used as the result of the bit shift gets multiplied
> > > > by sizeof(uint32_t).
> > > >
> > > > Signed-off-by: Andre Muezerie <andremue at linux.microsoft.com>
> > > > ---
> > >
> > > Acked-by: Bruce Richardson <bruce.richardson at intel.com>
> > >
> > > >  lib/hash/rte_thash.c               | 2 +-
> > > >  lib/hash/rte_thash_gf2_poly_math.c | 6 +++---
> > > >  2 files changed, 4 insertions(+), 4 deletions(-)
> > > >
> > > > diff --git a/lib/hash/rte_thash.c b/lib/hash/rte_thash.c
> > > > index fa78787143..f076311b57 100644
> > > > --- a/lib/hash/rte_thash.c
> > > > +++ b/lib/hash/rte_thash.c
> > > > @@ -565,7 +565,7 @@ rte_thash_add_helper(struct rte_thash_ctx *ctx,
> > const char *name, uint32_t len,
> > > >  		offset;
> > > >
> > > >  	ent = rte_zmalloc(NULL, sizeof(struct rte_thash_subtuple_helper)
> > +
> > > > -		sizeof(uint32_t) * (1 << ctx->reta_sz_log),
> > > > +		sizeof(uint32_t) * (1ULL << ctx->reta_sz_log),
> > > >  		RTE_CACHE_LINE_SIZE);
> > >
> > > Is there a reason not to use RTE_BIT64 here too?
> > 
> > Here we are calculating the size to be passed to the second argument of
> > rte_zmalloc, which is of type size_t. size_t is implementation
> > dependent, typically 4 bytes on 32-bit systems and 8 bytes on 64-bit
> > systems, so using 1ULL seems more appropriate.
> 
> 1ULL makes it 8 byte on 32-bit systems too. Did you mean 1UL?
> 
> How about reducing the formula to directly shift the sizeof() instead, i.e.:
> sizeof(uint32_t) << ctx->reta_sz_log,

Shifting the sizeof() directly is better indeed. Let me know how we should
proceed. Do you want me to send out a new series incorporating this suggestion?