Re: [PATCH 1/3] arm: Fix sign of MVE predicate mve_pred16_t [PR 107674]

["Andre Vieira (lists)" <andre.simoesdiasvieira@arm.com>]

On 26/01/2023 15:02, Kyrylo Tkachov wrote:
> Hi Andre,
> 
>> -----Original Message-----
>> From: Andre Vieira (lists) <andre.simoesdiasvieira@arm.com>
>> Sent: Tuesday, January 24, 2023 1:41 PM
>> To: gcc-patches@gcc.gnu.org
>> Cc: Kyrylo Tkachov <Kyrylo.Tkachov@arm.com>; Richard Earnshaw
>> <Richard.Earnshaw@arm.com>
>> Subject: [PATCH 1/3] arm: Fix sign of MVE predicate mve_pred16_t [PR
>> 107674]
>>
>> Hi,
>>
>> The ACLE defines mve_pred16_t as an unsigned short.  This patch makes
>> sure GCC treats the predicate as an unsigned type, rather than signed.
>>
>> Bootstrapped on aarch64-none-eabi and regression tested on arm-none-eabi
>> and armeb-none-eabi for armv8.1-m.main+mve.fp.
>>
>> OK for trunk?
>>
>> gcc/ChangeLog:
>>
>> 	PR target/107674
>> 	* config/arm/arm-builtins.cc (arm_simd_builtin_type): Rewrite to
>> use
>> 	new qualifiers parameter and use unsigned short type for MVE
>> predicate.
>> 	(arm_init_builtin): Call arm_simd_builtin_type with qualifiers
>> 	parameter.
>> 	(arm_init_crypto_builtins): Likewise.
>>
>> gcc/testsuite/ChangeLog:
>>
>> 	PR target/107674
>> 	* gcc.target/arm/mve/mve_vpt.c: New test.
> 
> diff --git a/gcc/config/arm/arm-builtins.cc b/gcc/config/arm/arm-builtins.cc
> index 11d7478d9df69139802a9d42c09dd0de7480b60e..6c67cec93ff76a4b42f3a0b305f697142e88fcd9 100644
> --- a/gcc/config/arm/arm-builtins.cc
> +++ b/gcc/config/arm/arm-builtins.cc
> @@ -1489,12 +1489,14 @@ arm_lookup_simd_builtin_type (machine_mode mode,
>   }
>   
>   static tree
> -arm_simd_builtin_type (machine_mode mode, bool unsigned_p, bool poly_p)
> +arm_simd_builtin_type (machine_mode mode, enum arm_type_qualifiers qualifiers)
>   {
> 
> I think in C++ now we can leave out the "enum" here.
> 
> diff --git a/gcc/testsuite/gcc.target/arm/mve/mve_vpt.c b/gcc/testsuite/gcc.target/arm/mve/mve_vpt.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..26a565b79dd1348e361b3aa23a1d6e6d13bffce8
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/arm/mve/mve_vpt.c
> @@ -0,0 +1,27 @@
> +/* { dg-options "-O2" } */
> +/* { dg-require-effective-target arm_v8_1m_mve_ok } */
> +/* { dg-add-options arm_v8_1m_mve } */
> +/* { dg-final { check-function-bodies "**" "" } } */
> +#include <arm_mve.h>
> +void test0 (uint8_t *a, uint8_t *b, uint8_t *c)
> +{
> +    uint8x16_t va = vldrbq_u8 (a);
> +    uint8x16_t vb = vldrbq_u8 (b);
> +    mve_pred16_t p = vcmpeqq_u8 (va, vb);
> +    uint8x16_t vc = vaddq_x_u8 (va, vb, p);
> +    vstrbq_p_u8 (c, vc, p);
> +}
> +/*
> +** test0:
> +**	vldrb.8	q2, \[r0\]
> +**	vldrb.8	q1, \[r1\]
> +**	vcmp.i8	eq, q2, q1
> +**	vmrs	r3, p0	@ movhi
> +**	uxth	r3, r3
> +**	vmsr	p0, r3	@ movhi
> +**	vpst
> +**	vaddt.i8	q3, q2, q1
> +**	vpst
> +**	vstrbt.8	q3, \[r2\]
> +**	bx	lr
> +*/
> 
> This explicit assembly matching looks quite fragile and sensitive to future scheduling and RA changes.
> Is there something more targeted we could scan for to check that the predicate is unsigned now?
No not really, as it's not unsigned everywhere, only in its intermediate 
representation between intrinsics. GCC is aware that mve_pred16_t is an 
unsigned short, so as soon as you try to use the value on its own or 
pass it as an argument or return, there is an implicit cast.

I could make this particular test-case more robust by not checking 
specific registers. Though the sequence of loads-cmp-add-store will 
always be the same as it's data-dependent.