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.cc
> b/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.
> +
> + /* 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
