This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH v6] Enable tracing of pseudo-registers on ARM


Antoine Tremblay <antoine.tremblay@ericsson.com> writes:

> @@ -20,7 +20,11 @@
>   * registers on x86_64.
>   */
>  

The comments above should be updated as well.

> +#if (defined __x86_64__)
>  #include <immintrin.h>
> +#elif (defined __arm__ || defined __thumb2__ || defined __thumb__)

__arm__ is defined even in thumb mode, so only "defined __arm__" is enough.

> +#include <arm_neon.h>

Why do you include arm_neon.h?  I don't see anything NEON specific is used.

> +#endif
>  
>  void
>  dummy (void)
> @@ -37,6 +41,7 @@ main (void)
>  {
>    /* Strictly speaking, it should be ymm15 (xmm15 is 128-bit), but gcc older
>       than 4.9 doesn't recognize "ymm15" as a valid register name.  */
> +#if (defined __x86_64__)
>    register __v8si a asm("xmm15") = {
>      0x12340001,
>      0x12340002,
> @@ -48,6 +53,13 @@ main (void)
>      0x12340008,
>    };
>    asm volatile ("traceme: call dummy" : : "x" (a));
> +#elif (defined __arm__ || defined __thumb2__ || defined __thumb__)

Only "defined __arm__" is needed.

> +  register uint32_t a asm("s5") = {
> +    0x2
> +  };

I'd like to write an inline asm to set s5 a value and the value can be shown as
an integer so that the test is more reliable (current test tests float
"2.80259693e-45").

> +  asm volatile ("traceme: bl dummy" : : "x" (a));
> +#endif
> +
>    end ();
>    return 0;
>  }
> diff --git a/gdb/testsuite/gdb.trace/tracefile-pseudo-reg.exp b/gdb/testsuite/gdb.trace/tracefile-pseudo-reg.exp
> index 4c52c64..12a2740 100644
> --- a/gdb/testsuite/gdb.trace/tracefile-pseudo-reg.exp
> +++ b/gdb/testsuite/gdb.trace/tracefile-pseudo-reg.exp
> @@ -12,8 +12,8 @@
>  # You should have received a copy of the GNU General Public License
>  # along with this program.  If not, see <http://www.gnu.org/licenses/>.
>  
> -if { ! [is_amd64_regs_target] } {
> -    verbose "Skipping tfile AVX test (target is not x86_64)."
> +if { ! [is_amd64_regs_target] && ! [istarget "arm*-*-*"] } {
> +    verbose "Skipping tracefile pseudo register tests, target is not supported."
>      return
>  }
>  
> @@ -21,8 +21,14 @@ load_lib "trace-support.exp"
>  
>  standard_testfile
>  
> +if { [is_amd64_regs_target] } {
> + set add_flags "-mavx"
> +} elseif { [istarget "arm*-*-*"] } {
> + set add_flags "-mfpu=neon"

Don't have to pass -mfpu=neon, because the case is also valid for vfp.

> +}
> +
>  if {[prepare_for_testing $testfile.exp $testfile $srcfile \
> -     [list debug additional_flags=-mavx]]} {
> +     [list debug additional_flags=$add_flags]]} {
>      return -1
>  }
>  
> @@ -36,20 +42,31 @@ if ![gdb_target_supports_trace] {
>      return -1
>  }
>  
> -gdb_test_multiple "print \$ymm15" "check for AVX support" {
> +if { [is_amd64_regs_target] } {
> +    set reg "\$ymm15"
> +    set reg_message "check for AVX support"
> +} elseif { [istarget "arm*-*-*"] } {
> +    set reg "\$s5"
> +    set reg_message "check for Neon support"

$s5 exists on the processors which support NEON or VFP, so the
$reg_message isn't accurate.  We can change reg_message to "check
register $reg".

-- 
Yao (éå)


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]