Return bytes for tdesc_register_size()
Commit Message
> On 10 Jul 2018, at 19:44, Simon Marchi <simon.marchi@ericsson.com> wrote:
>
> On 2018-07-10 11:05 AM, Alan Hayward wrote:
>> tdesc_register_size returns number of bits.
>> Change this to use bytes, the same as regcache::register_size, memcpy and sizeof.
>>
>> This fixes a bug in aarch64_get_tdesc_vq which assumed bytes.
>>
>> Update all other calls to tdesc_register_size.
>>
>> I originally planned to fix aarch64_get_tdesc_vq and push as OBV.
>> However, this seemed the better fix.
>> Required for 8.2
>> Tested with make check on a target all build.
>> I don't have a rs6000 machine, however change is simple enough.
>
> Hi Alan,
>
> Since I work with processors that have 16-bit bytes, I always prefer
> expressing sizes in bits, otherwise it's ambiguous. Are we talking
> about 8-bit bytes or target-sized bytes? So I would have perhaps
> chosen to rename tdesc_register_size to tdesc_register_size_bits
> (for clarity) and fix up the caller.
>
> Would you be ok with that? Other than that, the patch looks fine.
>
Right, that makes sense. This means regcache::register_size is at odds
with that
Happy with just renaming the function as it clears up the ambiguities.
Pushed the following as it's obvious:
@@ -2922,8 +2922,8 @@ aarch64_get_tdesc_vq (const struct target_desc *tdesc)
if (feature_sve == nullptr)
return 0;
- uint64_t vl = tdesc_register_size (feature_sve,
- aarch64_sve_register_names[0]);
+ uint64_t vl = tdesc_register_bitsize (feature_sve,
+ aarch64_sve_register_names[0]) / 8;
return sve_vq_from_vl (vl);
}
@@ -5953,7 +5953,7 @@ rs6000_gdbarch_init (struct gdbarch_info info, struct gdbarch_list *arches)
have_mq = tdesc_numbered_register (feature, tdesc_data, PPC_MQ_REGNUM,
"mq");
- tdesc_wordsize = tdesc_register_size (feature, "pc") / 8;
+ tdesc_wordsize = tdesc_register_bitsize (feature, "pc") / 8;
if (wordsize == -1)
wordsize = tdesc_wordsize;
@@ -5984,7 +5984,7 @@ rs6000_gdbarch_init (struct gdbarch_info info, struct gdbarch_list *arches)
/* The fpscr register was expanded in isa 2.05 to 64 bits
along with the addition of the decimal floating point
facility. */
- if (tdesc_register_size (feature, "fpscr") > 32)
+ if (tdesc_register_bitsize (feature, "fpscr") > 32)
have_dfp = 1;
}
else
@@ -125,8 +125,8 @@ int tdesc_unnumbered_register (const struct tdesc_feature *feature,
/* Search FEATURE for a register named NAME, and return its size in
bits. The register must exist. */
-int tdesc_register_size (const struct tdesc_feature *feature,
- const char *name);
+int tdesc_register_bitsize (const struct tdesc_feature *feature,
+ const char *name);
/* Search FEATURE for a register with any of the names from NAMES
(NULL-terminated). Record REGNO and the register in DATA; when
@@ -816,8 +816,7 @@ tdesc_numbered_register_choices (const struct tdesc_feature *feature,
bits. The register must exist. */
int
-tdesc_register_size (const struct tdesc_feature *feature,
- const char *name)
+tdesc_register_bitsize (const struct tdesc_feature *feature, const char *name)
{
struct tdesc_reg *reg = tdesc_find_register_early (feature, name);