On Mon, Jul 1, 2024 at 11:48 PM Andrew Pinski <quic_apin...@quicinc.com> wrote:
>
> Expanding cabs in powcab might be too late as forwprop might
> recombine the load from a memory with the complex expr. Moving
> instead to complex lowering allows us to use directly the real/imag
> component from the loads instead. This allows for vectorization too.
>
> Bootstrapped and tested on x86_64-linux-gnu with no regressions.
OK.
Thanks,
Richard.
> PR tree-optimization/115710
>
> gcc/ChangeLog:
>
> * tree-complex.cc (init_dont_simulate_again): Handle CABS.
> (gimple_expand_builtin_cabs): New function, moved mostly
> from tree-ssa-math-opts.cc.
> (expand_complex_operations_1): Call gimple_expand_builtin_cabs.
> * tree-ssa-math-opts.cc (gimple_expand_builtin_cabs): Remove.
> (build_and_insert_binop): Remove.
> (pass_data_expand_powcabs): Update comment.
> (pass_expand_powcabs::execute): Don't handle CABS.
>
> gcc/testsuite/ChangeLog:
>
> * gcc.dg/tree-ssa/cabs-1.c: New test.
> * gcc.dg/tree-ssa/cabs-2.c: New test.
> * gfortran.dg/vect/pr115710.f90: New test.
>
> Signed-off-by: Andrew Pinski <quic_apin...@quicinc.com>
> ---
> gcc/testsuite/gcc.dg/tree-ssa/cabs-1.c | 14 +++++
> gcc/testsuite/gcc.dg/tree-ssa/cabs-2.c | 13 ++++
> gcc/testsuite/gfortran.dg/vect/pr115710.f90 | 18 ++++++
> gcc/tree-complex.cc | 68 +++++++++++++++++++-
> gcc/tree-ssa-math-opts.cc | 70 +--------------------
> 5 files changed, 113 insertions(+), 70 deletions(-)
> create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/cabs-1.c
> create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/cabs-2.c
> create mode 100644 gcc/testsuite/gfortran.dg/vect/pr115710.f90
>
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/cabs-1.c
> b/gcc/testsuite/gcc.dg/tree-ssa/cabs-1.c
> new file mode 100644
> index 00000000000..12ff6049e63
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/cabs-1.c
> @@ -0,0 +1,14 @@
> +/* { dg-do compile { target sqrt_insn } } */
> +/* { dg-options "-Ofast -fdump-tree-cplxlower1" } */
> +/* { dg-add-options sqrt_insn } */
> +
> +
> +double f(_Complex double a)
> +{
> + a+= 1.0f;
> + return __builtin_cabs(a);
> +}
> +
> +/* Check that cabs is expanded during complex lowering. */
> +/* { dg-final { scan-tree-dump-not "__builtin_cabs " "cplxlower1" } } */
> +/* { dg-final { scan-tree-dump "__builtin_sqrt " "cplxlower1" } } */
> diff --git a/gcc/testsuite/gcc.dg/tree-ssa/cabs-2.c
> b/gcc/testsuite/gcc.dg/tree-ssa/cabs-2.c
> new file mode 100644
> index 00000000000..efe3de90cba
> --- /dev/null
> +++ b/gcc/testsuite/gcc.dg/tree-ssa/cabs-2.c
> @@ -0,0 +1,13 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -fdump-tree-cplxlower1" } */
> +
> +
> +double f(_Complex double a)
> +{
> + a+= 1.0f;
> + return __builtin_cabs(a);
> +}
> +
> +/* Check that cabs is not expanded during complex lowering. */
> +/* { dg-final { scan-tree-dump "__builtin_cabs " "cplxlower1" } } */
> +/* { dg-final { scan-tree-dump-not "__builtin_sqrt " "cplxlower1" } } */
> diff --git a/gcc/testsuite/gfortran.dg/vect/pr115710.f90
> b/gcc/testsuite/gfortran.dg/vect/pr115710.f90
> new file mode 100644
> index 00000000000..3749210ac80
> --- /dev/null
> +++ b/gcc/testsuite/gfortran.dg/vect/pr115710.f90
> @@ -0,0 +1,18 @@
> +! { dg-do compile }
> +! { dg-additional-options "-Ofast" }
> +! { dg-require-effective-target vect_float }
> +! { dg-require-effective-target vect_call_sqrtf }
> +
> +! { dg-final { scan-tree-dump "LOOP VECTORIZED" "vect" } }
> +! CABS expansion should allow for the vectorization to happen.
> +
> +subroutine foo(a,b,n)
> + complex(kind(1.0))::a(*)
> + real(kind(1.0))::b(*)
> + integer::i,n
> +
> + do i=1,n
> + b(i)=abs(a(i))**2
> + end do
> +
> +end subroutine foo
> diff --git a/gcc/tree-complex.cc b/gcc/tree-complex.cc
> index dfebec18ec3..d1276dc1c2f 100644
> --- a/gcc/tree-complex.cc
> +++ b/gcc/tree-complex.cc
> @@ -21,6 +21,7 @@ along with GCC; see the file COPYING3. If not see
> #include "system.h"
> #include "coretypes.h"
> #include "backend.h"
> +#include "target.h"
> #include "rtl.h"
> #include "tree.h"
> #include "gimple.h"
> @@ -42,7 +43,9 @@ along with GCC; see the file COPYING3. If not see
> #include "cfganal.h"
> #include "gimple-fold.h"
> #include "diagnostic-core.h"
> -
> +#include "case-cfn-macros.h"
> +#include "builtins.h"
> +#include "optabs-tree.h"
>
> /* For each complex ssa name, a lattice value. We're interested in finding
> out whether a complex number is degenerate in some way, having only real
> @@ -238,7 +241,18 @@ init_dont_simulate_again (void)
> {
> case GIMPLE_CALL:
> if (gimple_call_lhs (stmt))
> - sim_again_p = is_complex_reg (gimple_call_lhs (stmt));
> + {
> + sim_again_p = is_complex_reg (gimple_call_lhs (stmt));
> + switch (gimple_call_combined_fn (stmt))
> + {
> + CASE_CFN_CABS:
> + /* Expand cabs only if unsafe math and optimizing. */
> + if (optimize && flag_unsafe_math_optimizations)
> + saw_a_complex_op = true;
> + break;
> + default:;
> + }
> + }
> break;
>
> case GIMPLE_ASSIGN:
> @@ -1686,6 +1700,46 @@ expand_complex_asm (gimple_stmt_iterator *gsi)
> }
> }
>
> +
> +/* ARG is the argument to a cabs builtin call in GSI with location info
> + LOC. Create a sequence of statements prior to GSI that calculates
> + sqrt(R*R + I*I), where R and I are the real and imaginary components
> + of ARG, respectively. */
> +
> +static void
> +gimple_expand_builtin_cabs (gimple_stmt_iterator *gsi, gimple *old_stmt)
> +{
> + tree real_part, imag_part, addend1, addend2, sum;
> + tree arg = gimple_call_arg (old_stmt, 0);
> + tree type = TREE_TYPE (TREE_TYPE (arg));
> + tree sqrtfn = mathfn_built_in (type, BUILT_IN_SQRT);
> + machine_mode mode = TYPE_MODE (type);
> + gimple *new_stmt;
> +
> + if (!flag_unsafe_math_optimizations
> + || !optimize_bb_for_speed_p (gimple_bb (old_stmt))
> + || !sqrtfn
> + || optab_handler (sqrt_optab, mode) == CODE_FOR_nothing)
> + return;
> +
> + real_part = extract_component (gsi, arg, false, true);
> + imag_part = extract_component (gsi, arg, true, true);
> + location_t loc = gimple_location (old_stmt);
> +
> + gimple_seq stmts = NULL;
> + addend1 = gimple_build (&stmts, loc, MULT_EXPR, type, real_part,
> real_part);
> + addend2 = gimple_build (&stmts, loc, MULT_EXPR, type, imag_part,
> imag_part);
> + sum = gimple_build (&stmts, loc, PLUS_EXPR, type, addend1, addend2);
> + gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
> +
> + /* Build the sqrt call. */
> + new_stmt = gimple_build_call (sqrtfn, 1, sum);
> + gimple_set_location (new_stmt, loc);
> + tree lhs = gimple_call_lhs (old_stmt);
> + gimple_call_set_lhs (new_stmt, lhs);
> + gsi_replace (gsi, new_stmt, true);
> +}
> +
> /* Process one statement. If we identify a complex operation, expand it. */
>
> static void
> @@ -1696,6 +1750,16 @@ expand_complex_operations_1 (gimple_stmt_iterator *gsi)
> tree ac, ar, ai, bc, br, bi;
> complex_lattice_t al, bl;
> enum tree_code code;
> + if (gimple_code (stmt) == GIMPLE_CALL)
> + {
> + switch (gimple_call_combined_fn (stmt))
> + {
> + CASE_CFN_CABS:
> + gimple_expand_builtin_cabs (gsi, stmt);
> + return;
> + default:;
> + }
> + }
>
> if (gimple_code (stmt) == GIMPLE_ASM)
> {
> diff --git a/gcc/tree-ssa-math-opts.cc b/gcc/tree-ssa-math-opts.cc
> index 3b5433ec000..71f896a9790 100644
> --- a/gcc/tree-ssa-math-opts.cc
> +++ b/gcc/tree-ssa-math-opts.cc
> @@ -1622,22 +1622,6 @@ build_and_insert_binop (gimple_stmt_iterator *gsi,
> location_t loc,
> return result;
> }
>
> -/* Build a gimple reference operation with the given CODE and argument
> - ARG, assigning the result to a new SSA name of TYPE with NAME.
> - Insert the statement prior to GSI's current position, and return
> - the fresh SSA name. */
> -
> -static inline tree
> -build_and_insert_ref (gimple_stmt_iterator *gsi, location_t loc, tree type,
> - const char *name, enum tree_code code, tree arg0)
> -{
> - tree result = make_temp_ssa_name (type, NULL, name);
> - gimple *stmt = gimple_build_assign (result, build1 (code, type, arg0));
> - gimple_set_location (stmt, loc);
> - gsi_insert_before (gsi, stmt, GSI_SAME_STMT);
> - return result;
> -}
> -
> /* Build a gimple assignment to cast VAL to TYPE. Insert the statement
> prior to GSI's current position, and return the fresh SSA name. */
>
> @@ -2193,39 +2177,6 @@ gimple_expand_builtin_pow (gimple_stmt_iterator *gsi,
> location_t loc,
> return NULL_TREE;
> }
>
> -/* ARG is the argument to a cabs builtin call in GSI with location info
> - LOC. Create a sequence of statements prior to GSI that calculates
> - sqrt(R*R + I*I), where R and I are the real and imaginary components
> - of ARG, respectively. Return an expression holding the result. */
> -
> -static tree
> -gimple_expand_builtin_cabs (gimple_stmt_iterator *gsi, location_t loc, tree
> arg)
> -{
> - tree real_part, imag_part, addend1, addend2, sum, result;
> - tree type = TREE_TYPE (TREE_TYPE (arg));
> - tree sqrtfn = mathfn_built_in (type, BUILT_IN_SQRT);
> - machine_mode mode = TYPE_MODE (type);
> -
> - if (!flag_unsafe_math_optimizations
> - || !optimize_bb_for_speed_p (gimple_bb (gsi_stmt (*gsi)))
> - || !sqrtfn
> - || optab_handler (sqrt_optab, mode) == CODE_FOR_nothing)
> - return NULL_TREE;
> -
> - real_part = build_and_insert_ref (gsi, loc, type, "cabs",
> - REALPART_EXPR, arg);
> - addend1 = build_and_insert_binop (gsi, loc, "cabs", MULT_EXPR,
> - real_part, real_part);
> - imag_part = build_and_insert_ref (gsi, loc, type, "cabs",
> - IMAGPART_EXPR, arg);
> - addend2 = build_and_insert_binop (gsi, loc, "cabs", MULT_EXPR,
> - imag_part, imag_part);
> - sum = build_and_insert_binop (gsi, loc, "cabs", PLUS_EXPR, addend1,
> addend2);
> - result = build_and_insert_call (gsi, loc, sqrtfn, sum);
> -
> - return result;
> -}
> -
> /* Go through all calls to sin, cos and cexpi and call execute_cse_sincos_1
> on the SSA_NAME argument of each of them. */
>
> @@ -2322,7 +2273,8 @@ make_pass_cse_sincos (gcc::context *ctxt)
> }
>
> /* Expand powi(x,n) into an optimal number of multiplies, when n is a
> constant.
> - Also expand CABS. */
> + Note the name is powcabs but cabs expansion was moved to the lower complex
> + pass. */
> namespace {
>
> const pass_data pass_data_expand_powcabs =
> @@ -2455,24 +2407,6 @@ pass_expand_powcabs::execute (function *fun)
> }
> break;
>
> - CASE_CFN_CABS:
> - arg0 = gimple_call_arg (stmt, 0);
> - loc = gimple_location (stmt);
> - result = gimple_expand_builtin_cabs (&gsi, loc, arg0);
> -
> - if (result)
> - {
> - tree lhs = gimple_get_lhs (stmt);
> - gassign *new_stmt = gimple_build_assign (lhs, result);
> - gimple_set_location (new_stmt, loc);
> - unlink_stmt_vdef (stmt);
> - gsi_replace (&gsi, new_stmt, true);
> - cleanup_eh = true;
> - if (gimple_vdef (stmt))
> - release_ssa_name (gimple_vdef (stmt));
> - }
> - break;
> -
> default:;
> }
> }
> --
> 2.43.0
>
