> On 30 Aug 2024, at 14:17, Richard Sandiford <richard.sandif...@arm.com> wrote:
>
> External email: Use caution opening links or attachments
>
>
> Jennifer Schmitz <jschm...@nvidia.com> writes:
>> This patch implements constant folding for svdiv. If the predicate is
>> ptrue or predication is _x, it uses vector_const_binop with
>> aarch64_const_binop as callback and tree_code TRUNC_DIV_EXPR to fold constant
>> integer operands.
>> In aarch64_const_binop, a case was added for TRUNC_DIV_EXPR to return 0
>> for division by 0, as defined in the semantics for svdiv.
>> Tests were added to check the produced assembly for different
>> predicates, signed and unsigned integers, and the svdiv_n_* case.
>>
>> The patch was bootstrapped and regtested on aarch64-linux-gnu, no regression.
>> OK for mainline?
>>
>> Signed-off-by: Jennifer Schmitz <jschm...@nvidia.com>
>>
>> gcc/
>> * config/aarch64/aarch64-sve-builtins-base.cc (svdiv_impl::fold):
>> Try constant folding.
>> * config/aarch64/aarch64-sve-builtins.cc (aarch64_const_binop):
>> Add special case for division by 0.
>>
>> gcc/testsuite/
>> * gcc.target/aarch64/sve/const_fold_div_1.c: New test.
>>
>> From 92583732da28f6eb4a8db484fa3b24d55a7265e6 Mon Sep 17 00:00:00 2001
>> From: Jennifer Schmitz <jschm...@nvidia.com>
>> Date: Thu, 29 Aug 2024 05:04:51 -0700
>> Subject: [PATCH 2/3] SVE intrinsics: Fold constant operands for svdiv.
>>
>> This patch implements constant folding for svdiv. If the predicate is
>> ptrue or predication is _x, it uses vector_const_binop with
>> aarch64_const_binop as callback and tree_code TRUNC_DIV_EXPR to fold constant
>> integer operands.
>> In aarch64_const_binop, a case was added for TRUNC_DIV_EXPR to return 0
>> for division by 0, as defined in the semantics for svdiv.
>> Tests were added to check the produced assembly for different
>> predicates, signed and unsigned integers, and the svdiv_n_* case.
>>
>> The patch was bootstrapped and regtested on aarch64-linux-gnu, no regression.
>> OK for mainline?
>>
>> Signed-off-by: Jennifer Schmitz <jschm...@nvidia.com>
>>
>> gcc/
>> * config/aarch64/aarch64-sve-builtins-base.cc (svdiv_impl::fold):
>> Try constant folding.
>> * config/aarch64/aarch64-sve-builtins.cc (aarch64_const_binop):
>> Add special case for division by 0.
>>
>> gcc/testsuite/
>> * gcc.target/aarch64/sve/const_fold_div_1.c: New test.
>> ---
>> .../aarch64/aarch64-sve-builtins-base.cc | 19 +-
>> gcc/config/aarch64/aarch64-sve-builtins.cc | 4 +
>> .../gcc.target/aarch64/sve/const_fold_div_1.c | 336 ++++++++++++++++++
>> 3 files changed, 356 insertions(+), 3 deletions(-)
>> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/const_fold_div_1.c
>>
>> diff --git
>> a/gcc/config/aarch64/aarch64-sve-builtins-base.ccb/gcc/config/aarch64/aarch64-sve-builtins-base.cc
>> index d55bee0b72f..617c7fc87e5 100644
>> --- a/gcc/config/aarch64/aarch64-sve-builtins-base.cc
>> +++ b/gcc/config/aarch64/aarch64-sve-builtins-base.cc
>> @@ -755,8 +755,21 @@ public:
>> gimple *
>> fold (gimple_folder &f) const override
>> {
>> - tree divisor = gimple_call_arg (f.call, 2);
>> - tree divisor_cst = uniform_integer_cst_p (divisor);
>> + tree pg = gimple_call_arg (f.call, 0);
>> + tree op1 = gimple_call_arg (f.call, 1);
>> + tree op2 = gimple_call_arg (f.call, 2);
>> +
>> + /* Try to fold constant integer operands. */
>> + if (f.type_suffix (0).integer_p
>> + && (f.pred == PRED_x
>> + || is_ptrue (pg, f.type_suffix (0).element_bytes)))
>> + if (tree res = vector_const_binop (TRUNC_DIV_EXPR, op1, op2,
>> + aarch64_const_binop))
>> + return gimple_build_assign (f.lhs, res);
>
> To reduce cut-&-paste, it'd be good to put this in a helper:
>
> gimple *gimple_folder::fold_const_binary (tree_code code);
>
> that does the outermost "if" above for "code" rather than TRUNC_DIV_EXPR.
> It could return null on failure. Then the caller can just be:
>
> if (auto *res = f.fold_const_binary (TRUNC_DIV_EXPR))
> return res;
>
> This could go right at the top of the function, since it doesn't rely
> on any of the local variables above.
Done.
>
>> +
>> + /* If the divisor is a uniform power of 2, fold to a shift
>> + instruction. */
>> + tree divisor_cst = uniform_integer_cst_p (op2);
>>
>> if (!divisor_cst || !integer_pow2p (divisor_cst))
>> return NULL;
>> @@ -770,7 +783,7 @@ public:
>> shapes::binary_uint_opt_n, MODE_n,
>> f.type_suffix_ids, GROUP_none, f.pred);
>> call = f.redirect_call (instance);
>> - tree d = INTEGRAL_TYPE_P (TREE_TYPE (divisor)) ? divisor : divisor_cst;
>> + tree d = INTEGRAL_TYPE_P (TREE_TYPE (op2)) ? op2 : divisor_cst;
>> new_divisor = wide_int_to_tree (TREE_TYPE (d), tree_log2 (d));
>> }
>> else
>> diff --git a/gcc/config/aarch64/aarch64-sve-builtins.cc
>> b/gcc/config/aarch64/aarch64-sve-builtins.cc
>> index 315d5ac4177..c1b28ebfe4e 100644
>> --- a/gcc/config/aarch64/aarch64-sve-builtins.cc
>> +++ b/gcc/config/aarch64/aarch64-sve-builtins.cc
>> @@ -3444,6 +3444,10 @@ aarch64_const_binop (enum tree_code code, tree arg1,
>> tree arg2)
>> signop sign = TYPE_SIGN (type);
>> wi::overflow_type overflow = wi::OVF_NONE;
>>
>> + /* Return 0 for division by 0. */
>> + if (code == TRUNC_DIV_EXPR && integer_zerop (arg2))
>> + return arg2;
>> +
>> if (!poly_int_binop (poly_res, code, arg1, arg2, sign, &overflow))
>> return NULL_TREE;
>> return force_fit_type (type, poly_res, false,
>> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/const_fold_div_1.c
>> b/gcc/testsuite/gcc.target/aarch64/sve/const_fold_div_1.c
>> new file mode 100644
>> index 00000000000..062fb6e560e
>> --- /dev/null
>> +++ b/gcc/testsuite/gcc.target/aarch64/sve/const_fold_div_1.c
>> @@ -0,0 +1,336 @@
>> +/* { dg-final { check-function-bodies "**" "" } } */
>> +/* { dg-options "-O2" } */
>> +
>> +#include "arm_sve.h"
>> +
>> +/*
>> +** s64_x_pg:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svint64_t s64_x_pg (svbool_t pg)
>> +{
>> + return svdiv_x (pg, svdup_s64 (5), svdup_s64 (3));
>> +}
>> +
>> +/*
>> +** s64_x_pg_0:
>> +** mov z[0-9]+\.b, #0
>> +** ret
>> +*/
>> +svint64_t s64_x_pg_0 (svbool_t pg)
>> +{
>> + return svdiv_x (pg, svdup_s64 (0), svdup_s64 (3));
>> +}
>> +
>> +/*
>> +** s64_x_pg_by0:
>> +** mov z[0-9]+\.b, #0
>> +** ret
>> +*/
>> +svint64_t s64_x_pg_by0 (svbool_t pg)
>> +{
>> + return svdiv_x (pg, svdup_s64 (5), svdup_s64 (0));
>> +}
>> +
>> +/*
>> +** s64_z_pg:
>> +** mov z[0-9]+\.d, p[0-7]/z, #1
>> +** ret
>> +*/
>> +svint64_t s64_z_pg (svbool_t pg)
>> +{
>> + return svdiv_z (pg, svdup_s64 (5), svdup_s64 (3));
>> +}
>> +
>> +/*
>> +** s64_z_pg_0:
>> +** mov z[0-9]+\.d, p[0-7]/z, #0
>> +** ret
>> +*/
>> +svint64_t s64_z_pg_0 (svbool_t pg)
>> +{
>> + return svdiv_z (pg, svdup_s64 (0), svdup_s64 (3));
>> +}
>> +
>> +/*
>> +** s64_z_pg_by0:
>> +** mov (z[0-9]+\.d), #5
>> +** mov (z[0-9]+)\.b, #0
>> +** sdivr \2\.d, p[0-7]/m, \2\.d, \1
>> +** ret
>> +*/
>> +svint64_t s64_z_pg_by0 (svbool_t pg)
>> +{
>> + return svdiv_z (pg, svdup_s64 (5), svdup_s64 (0));
>> +}
>> +
>> +/*
>> +** s64_m_pg:
>> +** mov (z[0-9]+\.d), #3
>> +** mov (z[0-9]+\.d), #5
>> +** sdiv \2, p[0-7]/m, \2, \1
>> +** ret
>> +*/
>> +svint64_t s64_m_pg (svbool_t pg)
>> +{
>> + return svdiv_m (pg, svdup_s64 (5), svdup_s64 (3));
>> +}
>> +
>> +/*
>> +** s64_x_ptrue:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svint64_t s64_x_ptrue ()
>> +{
>> + return svdiv_x (svptrue_b64 (), svdup_s64 (5), svdup_s64 (3));
>> +}
>> +
>> +/*
>> +** s64_z_ptrue:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svint64_t s64_z_ptrue ()
>> +{
>> + return svdiv_z (svptrue_b64 (), svdup_s64 (5), svdup_s64 (3));
>> +}
>> +
>> +/*
>> +** s64_m_ptrue:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svint64_t s64_m_ptrue ()
>> +{
>> + return svdiv_m (svptrue_b64 (), svdup_s64 (5), svdup_s64 (3));
>> +}
>> +
>> +/*
>> +** s64_x_pg_n:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svint64_t s64_x_pg_n (svbool_t pg)
>> +{
>> + return svdiv_n_s64_x (pg, svdup_s64 (5), 3);
>> +}
>> +
>> +/*
>> +** s64_x_pg_n_s64_0:
>> +** mov z[0-9]+\.b, #0
>> +** ret
>> +*/
>> +svint64_t s64_x_pg_n_s64_0 (svbool_t pg)
>> +{
>> + return svdiv_n_s64_x (pg, svdup_s64 (0), 3);
>> +}
>> +
>> +/*
>> +** s64_x_pg_n_s64_by0:
>> +** mov z[0-9]+\.b, #0
>> +** ret
>> +*/
>> +svint64_t s64_x_pg_n_s64_by0 (svbool_t pg)
>> +{
>> + return svdiv_n_s64_x (pg, svdup_s64 (5), 0);
>> +}
>> +
>> +/*
>> +** s64_z_pg_n:
>> +** mov z[0-9]+\.d, p[0-7]/z, #1
>> +** ret
>> +*/
>> +svint64_t s64_z_pg_n (svbool_t pg)
>> +{
>> + return svdiv_n_s64_z (pg, svdup_s64 (5), 3);
>> +}
>> +
>> +/*
>> +** s64_z_pg_n_s64_0:
>> +** mov z[0-9]+\.d, p[0-7]/z, #0
>> +** ret
>> +*/
>> +svint64_t s64_z_pg_n_s64_0 (svbool_t pg)
>> +{
>> + return svdiv_n_s64_z (pg, svdup_s64 (0), 3);
>> +}
>> +
>> +/*
>> +** s64_z_pg_n_s64_by0:
>> +** mov (z[0-9]+\.d), #5
>> +** mov (z[0-9]+)\.b, #0
>> +** sdivr \2\.d, p[0-7]/m, \2\.d, \1
>> +** ret
>> +*/
>> +svint64_t s64_z_pg_n_s64_by0 (svbool_t pg)
>> +{
>> + return svdiv_n_s64_z (pg, svdup_s64 (5), 0);
>> +}
>> +
>> +/*
>> +** s64_m_pg_n:
>> +** mov (z[0-9]+\.d), #3
>> +** mov (z[0-9]+\.d), #5
>> +** sdiv \2, p[0-7]/m, \2, \1
>> +** ret
>> +*/
>> +svint64_t s64_m_pg_n (svbool_t pg)
>> +{
>> + return svdiv_n_s64_m (pg, svdup_s64 (5), 3);
>> +}
>> +
>> +/*
>> +** s64_x_ptrue_n:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svint64_t s64_x_ptrue_n ()
>> +{
>> + return svdiv_n_s64_x (svptrue_b64 (), svdup_s64 (5), 3);
>> +}
>> +
>> +/*
>> +** s64_z_ptrue_n:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svint64_t s64_z_ptrue_n ()
>> +{
>> + return svdiv_n_s64_z (svptrue_b64 (), svdup_s64 (5), 3);
>> +}
>> +
>> +/*
>> +** s64_m_ptrue_n:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svint64_t s64_m_ptrue_n ()
>> +{
>> + return svdiv_n_s64_m (svptrue_b64 (), svdup_s64 (5), 3);
>> +}
>> +
>> +/*
>> +** u64_x_pg:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svuint64_t u64_x_pg (svbool_t pg)
>> +{
>> + return svdiv_x (pg, svdup_u64 (5), svdup_u64 (3));
>> +}
>> +
>> +/*
>> +** u64_z_pg:
>> +** mov z[0-9]+\.d, p[0-7]/z, #1
>> +** ret
>> +*/
>> +svuint64_t u64_z_pg (svbool_t pg)
>> +{
>> + return svdiv_z (pg, svdup_u64 (5), svdup_u64 (3));
>> +}
>> +
>> +/*
>> +** u64_m_pg:
>> +** mov (z[0-9]+\.d), #3
>> +** mov (z[0-9]+\.d), #5
>> +** udiv \2, p[0-7]/m, \2, \1
>> +** ret
>> +*/
>> +svuint64_t u64_m_pg (svbool_t pg)
>> +{
>> + return svdiv_m (pg, svdup_u64 (5), svdup_u64 (3));
>> +}
>> +
>> +/*
>> +** u64_x_ptrue:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svuint64_t u64_x_ptrue ()
>> +{
>> + return svdiv_x (svptrue_b64 (), svdup_u64 (5), svdup_u64 (3));
>> +}
>> +
>> +/*
>> +** u64_z_ptrue:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svuint64_t u64_z_ptrue ()
>> +{
>> + return svdiv_z (svptrue_b64 (), svdup_u64 (5), svdup_u64 (3));
>> +}
>> +
>> +/*
>> +** u64_m_ptrue:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svuint64_t u64_m_ptrue ()
>> +{
>> + return svdiv_m (svptrue_b64 (), svdup_u64 (5), svdup_u64 (3));
>> +}
>> +
>> +/*
>> +** u64_x_pg_n:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svuint64_t u64_x_pg_n (svbool_t pg)
>> +{
>> + return svdiv_n_u64_x (pg, svdup_u64 (5), 3);
>> +}
>> +
>> +/*
>> +** u64_z_pg_n:
>> +** mov z[0-9]+\.d, p[0-7]/z, #1
>> +** ret
>> +*/
>> +svuint64_t u64_z_pg_n (svbool_t pg)
>> +{
>> + return svdiv_n_u64_z (pg, svdup_u64 (5), 3);
>> +}
>> +
>> +/*
>> +** u64_m_pg_n:
>> +** mov (z[0-9]+\.d), #3
>> +** mov (z[0-9]+\.d), #5
>> +** udiv \2, p[0-7]/m, \2, \1
>> +** ret
>> +*/
>> +svuint64_t u64_m_pg_n (svbool_t pg)
>> +{
>> + return svdiv_n_u64_m (pg, svdup_u64 (5), 3);
>> +}
>> +
>> +/*
>> +** u64_x_ptrue_n:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svuint64_t u64_x_ptrue_n ()
>> +{
>> + return svdiv_n_u64_x (svptrue_b64 (), svdup_u64 (5), 3);
>> +}
>> +
>> +/*
>> +** u64_z_ptrue_n:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svuint64_t u64_z_ptrue_n ()
>> +{
>> + return svdiv_n_u64_z (svptrue_b64 (), svdup_u64 (5), 3);
>> +}
>> +
>> +/*
>> +** u64_m_ptrue_n:
>> +** mov z[0-9]+\.d, #1
>> +** ret
>> +*/
>> +svuint64_t u64_m_ptrue_n ()
>> +{
>> + return svdiv_n_u64_m (svptrue_b64 (), svdup_u64 (5), 3);
>> +}
>
> These are good tests, but maybe we could throw in a small number
> of svdupq tests as well, to test for non-uniform cases. E.g.:
>
> svdiv_s32_m (svptrue_b32 (), svdupq_s32 (3, 0, -5, 11), svdupq_s32 (4, 1,
> -6, 0));
>
> which hopefully should get optimised to zero.
>
> Similarly:
>
> svdiv_s32_z (svptrue_b32 (), svdupq_s32 (6, -30, 100, -4), svdupq_s32 (-3,
> 15, -50, 2));
>
> should get optimised to -2.
>
> Looks good to me otherwise.
>
> Thanks,
> Richard
Thanks for suggesting these tests, I added them to the test file.
Best, Jennifer
0002-SVE-intrinsics-Fold-constant-operands-for-svdiv.patch
Description: Binary data
smime.p7s
Description: S/MIME cryptographic signature
