Kyrylo Tkachov <[email protected]> writes:
>> On 29 Jul 2025, at 18:41, Richard Sandiford <[email protected]>
>> wrote:
>>
>> This patch continues the work of making ACLE intrinsics use VNx16BI
>> for svbool_t results. It deals with the svpnext* intrinsics.
>>
>
> I wonder if the new patterns need pred_clobber alternatives in this and the
> other patches?
> If they do we can add them in follow-up patches I suppose.
I think the series preserves the status quo in terms of where
pred_clobber is applied and where it isn't. But yeah, it seems
that the status quo misses cases, here and in the *_cc patterns
for MATCH/NMATCH.
Like you say, I think we can deal with that separately, although
the series does unfortunately add to the number of patterns that
would need changing.
Thanks for the reviews!
Richard
> Ok.
> Thanks,
> Kyrill
>
>> gcc/
>> * config/aarch64/iterators.md (PNEXT_ONLY): New int iterator.
>> * config/aarch64/aarch64-sve.md
>> (@aarch64_sve_<sve_pred_op><mode>): Restrict SVE_PITER pattern
>> to VNx16BI_ONLY.
>> (@aarch64_sve_<sve_pred_op><mode>): New PNEXT_ONLY pattern for
>> PRED_HSD.
>> (*aarch64_sve_<sve_pred_op><mode>): Likewise.
>> (*aarch64_sve_<sve_pred_op><mode>_cc): Likewise.
>>
>> gcc/testsuite/
>> * gcc.target/aarch64/sve/acle/general/pnext_3.c: New test.
>> ---
>> gcc/config/aarch64/aarch64-sve.md | 77 ++++++++++-
>> gcc/config/aarch64/iterators.md | 2 +
>> .../aarch64/sve/acle/general/pnext_3.c | 130 ++++++++++++++++++
>> 3 files changed, 204 insertions(+), 5 deletions(-)
>> create mode 100644
>> gcc/testsuite/gcc.target/aarch64/sve/acle/general/pnext_3.c
>>
>> diff --git a/gcc/config/aarch64/aarch64-sve.md
>> b/gcc/config/aarch64/aarch64-sve.md
>> index b881ba851b6..8011227e2d9 100644
>> --- a/gcc/config/aarch64/aarch64-sve.md
>> +++ b/gcc/config/aarch64/aarch64-sve.md
>> @@ -11171,14 +11171,49 @@ (define_insn "*aarch64_brk<brk_op>_ptest"
>> ;; -------------------------------------------------------------------------
>>
>> (define_insn "@aarch64_sve_<sve_pred_op><mode>"
>> - [(set (match_operand:PRED_ALL 0 "register_operand" "=Upa")
>> - (unspec:PRED_ALL
>> - [(match_operand:PRED_ALL 1 "register_operand" "Upa")
>> + [(set (match_operand:VNx16BI_ONLY 0 "register_operand" "=Upa")
>> + (unspec:VNx16BI_ONLY
>> + [(match_operand:VNx16BI_ONLY 1 "register_operand" "Upa")
>> (match_operand:SI 2 "aarch64_sve_ptrue_flag")
>> - (match_operand:PRED_ALL 3 "register_operand" "0")]
>> + (match_operand:VNx16BI_ONLY 3 "register_operand" "0")]
>> SVE_PITER))
>> (clobber (reg:CC_NZC CC_REGNUM))]
>> - "TARGET_SVE && <max_elem_bits> >= <elem_bits>"
>> + "TARGET_SVE"
>> + "<sve_pred_op>\t%0.<Vetype>, %1, %0.<Vetype>"
>> +)
>> +
>> +(define_expand "@aarch64_sve_<sve_pred_op><mode>"
>> + [(parallel
>> + [(set (match_operand:VNx16BI 0 "register_operand" "=Upa")
>> + (and:VNx16BI
>> + (subreg:VNx16BI
>> + (unspec:PRED_HSD
>> + [(match_operand:PRED_HSD 1 "register_operand" "Upa")
>> + (match_operand:SI 2 "aarch64_sve_ptrue_flag")
>> + (match_operand:PRED_HSD 3 "register_operand" "0")]
>> + PNEXT_ONLY)
>> + 0)
>> + (match_dup 4)))
>> + (clobber (reg:CC_NZC CC_REGNUM))])]
>> + "TARGET_SVE"
>> + {
>> + operands[4] = aarch64_ptrue_all (<data_bytes>);
>> + }
>> +)
>> +
>> +(define_insn "*aarch64_sve_<sve_pred_op><mode>"
>> + [(set (match_operand:VNx16BI 0 "register_operand" "=Upa")
>> + (and:VNx16BI
>> + (subreg:VNx16BI
>> + (unspec:PRED_HSD
>> + [(match_operand:PRED_HSD 1 "register_operand" "Upa")
>> + (match_operand:SI 2 "aarch64_sve_ptrue_flag")
>> + (match_operand:PRED_HSD 3 "register_operand" "0")]
>> + PNEXT_ONLY)
>> + 0)
>> + (match_operand:PRED_HSD 4 "aarch64_ptrue_all_operand")))
>> + (clobber (reg:CC_NZC CC_REGNUM))]
>> + "TARGET_SVE"
>> "<sve_pred_op>\t%0.<Vetype>, %1, %0.<Vetype>"
>> )
>>
>> @@ -11212,6 +11247,38 @@ (define_insn_and_rewrite
>> "*aarch64_sve_<sve_pred_op><mode>_cc"
>> }
>> )
>>
>> +(define_insn_and_rewrite "*aarch64_sve_<sve_pred_op><mode>_cc"
>> + [(set (reg:CC_NZC CC_REGNUM)
>> + (unspec:CC_NZC
>> + [(match_operand:VNx16BI 1 "register_operand" "Upa")
>> + (match_operand 2)
>> + (match_operand:SI 3 "aarch64_sve_ptrue_flag")
>> + (unspec:PRED_HSD
>> + [(match_operand 4)
>> + (match_operand:SI 5 "aarch64_sve_ptrue_flag")
>> + (match_operand:PRED_HSD 6 "register_operand" "0")]
>> + PNEXT_ONLY)]
>> + UNSPEC_PTEST))
>> + (set (match_operand:VNx16BI 0 "register_operand" "=Upa")
>> + (and:VNx16BI
>> + (subreg:VNx16BI
>> + (unspec:PRED_HSD
>> + [(match_dup 4)
>> + (match_dup 5)
>> + (match_dup 6)]
>> + PNEXT_ONLY)
>> + 0)
>> + (match_operand:PRED_HSD 7 "aarch64_ptrue_all_operand")))]
>> + "TARGET_SVE
>> + && aarch64_sve_same_pred_for_ptest_p (&operands[2], &operands[4])"
>> + "<sve_pred_op>\t%0.<Vetype>, %1, %0.<Vetype>"
>> + "&& !rtx_equal_p (operands[2], operands[4])"
>> + {
>> + operands[4] = operands[2];
>> + operands[5] = operands[3];
>> + }
>> +)
>> +
>> ;; Same, but with only the flags result being interesting.
>> (define_insn_and_rewrite "*aarch64_sve_<sve_pred_op><mode>_ptest"
>> [(set (reg:CC_NZC CC_REGNUM)
>> diff --git a/gcc/config/aarch64/iterators.md
>> b/gcc/config/aarch64/iterators.md
>> index e619af14928..8f8237edf6c 100644
>> --- a/gcc/config/aarch64/iterators.md
>> +++ b/gcc/config/aarch64/iterators.md
>> @@ -3880,6 +3880,8 @@ (define_int_iterator SVE_BRK_BINARY [UNSPEC_BRKN
>> UNSPEC_BRKPA UNSPEC_BRKPB])
>>
>> (define_int_iterator SVE_PITER [UNSPEC_PFIRST UNSPEC_PNEXT])
>>
>> +(define_int_iterator PNEXT_ONLY [UNSPEC_PNEXT])
>> +
>> (define_int_iterator MATMUL [UNSPEC_SMATMUL UNSPEC_UMATMUL
>> UNSPEC_USMATMUL])
>>
>> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/acle/general/pnext_3.c
>> b/gcc/testsuite/gcc.target/aarch64/sve/acle/general/pnext_3.c
>> new file mode 100644
>> index 00000000000..d9c009062d7
>> --- /dev/null
>> +++ b/gcc/testsuite/gcc.target/aarch64/sve/acle/general/pnext_3.c
>> @@ -0,0 +1,130 @@
>> +/* { dg-do compile } */
>> +/* { dg-options "-O2" } */
>> +/* { dg-final { check-function-bodies "**" "" } } */
>> +
>> +#include <arm_sve.h>
>> +
>> +#ifdef __cplusplus
>> +extern "C" {
>> +#endif
>> +
>> +/*
>> +** test1:
>> +** pnext p0\.h, p1, p0\.h
>> +** ret
>> +*/
>> +svbool_t
>> +test1 (svbool_t pg, svbool_t prev)
>> +{
>> + return svand_z (svptrue_b8 (),
>> + svpnext_b16 (prev, pg),
>> + svptrue_b16 ());
>> +}
>> +
>> +/*
>> +** test2:
>> +** pnext p0\.h, p1, p0\.h
>> +** ret
>> +*/
>> +svbool_t
>> +test2 (svbool_t pg, svbool_t prev)
>> +{
>> + return svand_z (svptrue_b16 (),
>> + svpnext_b16 (prev, pg),
>> + svptrue_b8 ());
>> +}
>> +
>> +/*
>> +** test3:
>> +** pnext p0\.h, p1, p0\.h
>> +** ret
>> +*/
>> +svbool_t
>> +test3 (svbool_t pg, svbool_t prev)
>> +{
>> + return svand_z (svptrue_b16 (),
>> + svpnext_b16 (prev, pg),
>> + svptrue_b16 ());
>> +}
>> +
>> +/*
>> +** test4:
>> +** pnext p0\.s, p1, p0\.s
>> +** ret
>> +*/
>> +svbool_t
>> +test4 (svbool_t pg, svbool_t prev)
>> +{
>> + return svand_z (svptrue_b32 (),
>> + svpnext_b32 (prev, pg),
>> + svptrue_b8 ());
>> +}
>> +
>> +/*
>> +** test5:
>> +** pnext p0\.s, p1, p0\.s
>> +** ret
>> +*/
>> +svbool_t
>> +test5 (svbool_t pg, svbool_t prev)
>> +{
>> + return svand_z (svptrue_b16 (),
>> + svpnext_b32 (prev, pg),
>> + svptrue_b8 ());
>> +}
>> +
>> +/*
>> +** test6:
>> +** pnext p0\.s, p1, p0\.s
>> +** ret
>> +*/
>> +svbool_t
>> +test6 (svbool_t pg, svbool_t prev)
>> +{
>> + return svand_z (svptrue_b8 (),
>> + svpnext_b32 (prev, pg),
>> + svptrue_b32 ());
>> +}
>> +
>> +/*
>> +** test7:
>> +** pnext p0\.d, p1, p0\.d
>> +** ret
>> +*/
>> +svbool_t
>> +test7 (svbool_t pg, svbool_t prev)
>> +{
>> + return svand_z (svptrue_b16 (),
>> + svpnext_b64 (prev, pg),
>> + svptrue_b8 ());
>> +}
>> +
>> +/*
>> +** test8:
>> +** pnext p0\.d, p1, p0\.d
>> +** ret
>> +*/
>> +svbool_t
>> +test8 (svbool_t pg, svbool_t prev)
>> +{
>> + return svand_z (svptrue_b32 (),
>> + svpnext_b64 (prev, pg),
>> + svptrue_b8 ());
>> +}
>> +
>> +/*
>> +** test9:
>> +** pnext p0\.d, p1, p0\.d
>> +** ret
>> +*/
>> +svbool_t
>> +test9 (svbool_t pg, svbool_t prev)
>> +{
>> + return svand_z (svptrue_b8 (),
>> + svpnext_b64 (prev, pg),
>> + svptrue_b64 ());
>> +}
>> +
>> +#ifdef __cplusplus
>> +}
>> +#endif
>> --
>> 2.43.0
>>
