phiopt: Optimize (x <=> y) cmp z [PR94589]

genericize_spaceship genericizes i <=> j to approximately
({ int c; if (i == j) c = 0; else if (i < j) c = -1; else c = 1; c; })
for strong ordering and
({ int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; c; })
for partial ordering.
The C++ standard supports then == or != comparisons of that against
strong/partial ordering enums, or </<=/==/!=/>/>= comparisons of <=> result
against literal 0.

In some cases we already optimize that but in many cases we keep performing
all the 2 or 3 comparisons, compute the spaceship value and then compare
that.

The following patch recognizes those patterns if the <=> operands are
integral types or floating point (the latter only for -ffast-math) and
optimizes it to the single comparison that is needed (plus adds debug stmts
if needed for the spaceship result).

There is one thing I'd like to address in a follow-up: the pr94589-2.C
testcase should be matching just 12 times each, but runs
into operator>=(partial_ordering, unspecified) being defined as
(_M_value&1)==_M_value
rather than _M_value>=0.  When not honoring NaNs, the 2 case should be
unreachable and so (_M_value&1)==_M_value is then equivalent to _M_value>=0,
but is not a single use but two uses.  I'll need to pattern match that case
specially.

2021-05-06  Jakub Jelinek  <jakub@redhat.com>

	PR tree-optimization/94589
	* tree-ssa-phiopt.c (tree_ssa_phiopt_worker): Call
	spaceship_replacement.
	(cond_only_block_p, spaceship_replacement): New functions.

	* gcc.dg/pr94589-1.c: New test.
	* gcc.dg/pr94589-2.c: New test.
	* gcc.dg/pr94589-3.c: New test.
	* gcc.dg/pr94589-4.c: New test.
	* g++.dg/opt/pr94589-1.C: New test.
	* g++.dg/opt/pr94589-2.C: New test.
	* g++.dg/opt/pr94589-3.C: New test.
	* g++.dg/opt/pr94589-4.C: New test.
This commit is contained in:
Jakub Jelinek 2021-05-06 10:15:40 +02:00
parent b5254d6b75
commit ad96c867e1
9 changed files with 922 additions and 0 deletions

View file

@ -0,0 +1,33 @@
// PR tree-optimization/94589
// { dg-do compile { target c++20 } }
// { dg-options "-O2 -g0 -fdump-tree-optimized" }
// { dg-final { scan-tree-dump-times "\[ij]_\[0-9]+\\(D\\) (?:<|<=|==|!=|>|>=) \[ij]_\[0-9]+\\(D\\)" 12 "optimized" } }
// { dg-final { scan-tree-dump-times "i_\[0-9]+\\(D\\) (?:<|<=|==|!=|>|>=) \[45]" 12 "optimized" } }
#include <compare>
#define A __attribute__((noipa))
A bool f1 (int i, int j) { auto c = i <=> j; return c == 0; }
A bool f2 (int i, int j) { auto c = i <=> j; return c != 0; }
A bool f3 (int i, int j) { auto c = i <=> j; return c > 0; }
A bool f4 (int i, int j) { auto c = i <=> j; return c < 0; }
A bool f5 (int i, int j) { auto c = i <=> j; return c >= 0; }
A bool f6 (int i, int j) { auto c = i <=> j; return c <= 0; }
A bool f7 (int i, int j) { auto c = i <=> j; return c == std::strong_ordering::less; }
A bool f8 (int i, int j) { auto c = i <=> j; return c != std::strong_ordering::less; }
A bool f9 (int i, int j) { auto c = i <=> j; return c == std::strong_ordering::equal; }
A bool f10 (int i, int j) { auto c = i <=> j; return c != std::strong_ordering::equal; }
A bool f11 (int i, int j) { auto c = i <=> j; return c == std::strong_ordering::greater; }
A bool f12 (int i, int j) { auto c = i <=> j; return c != std::strong_ordering::greater; }
A bool f13 (int i) { auto c = i <=> 5; return c == 0; }
A bool f14 (int i) { auto c = i <=> 5; return c != 0; }
A bool f15 (int i) { auto c = i <=> 5; return c > 0; }
A bool f16 (int i) { auto c = i <=> 5; return c < 0; }
A bool f17 (int i) { auto c = i <=> 5; return c >= 0; }
A bool f18 (int i) { auto c = i <=> 5; return c <= 0; }
A bool f19 (int i) { auto c = i <=> 5; return c == std::strong_ordering::less; }
A bool f20 (int i) { auto c = i <=> 5; return c != std::strong_ordering::less; }
A bool f21 (int i) { auto c = i <=> 5; return c == std::strong_ordering::equal; }
A bool f22 (int i) { auto c = i <=> 5; return c != std::strong_ordering::equal; }
A bool f23 (int i) { auto c = i <=> 5; return c == std::strong_ordering::greater; }
A bool f24 (int i) { auto c = i <=> 5; return c != std::strong_ordering::greater; }

View file

@ -0,0 +1,33 @@
// PR tree-optimization/94589
// { dg-do compile { target c++20 } }
// { dg-options "-O2 -g0 -ffast-math -fdump-tree-optimized" }
// { dg-final { scan-tree-dump-times "\[ij]_\[0-9]+\\(D\\) (?:<|<=|==|!=|>|>=) \[ij]_\[0-9]+\\(D\\)" 14 "optimized" } }
// { dg-final { scan-tree-dump-times "i_\[0-9]+\\(D\\) (?:<|<=|==|!=|>|>=) 5\\.0" 14 "optimized" } }
#include <compare>
#define A __attribute__((noipa))
A bool f1 (double i, double j) { auto c = i <=> j; return c == 0; }
A bool f2 (double i, double j) { auto c = i <=> j; return c != 0; }
A bool f3 (double i, double j) { auto c = i <=> j; return c > 0; }
A bool f4 (double i, double j) { auto c = i <=> j; return c < 0; }
A bool f5 (double i, double j) { auto c = i <=> j; return c >= 0; }
A bool f6 (double i, double j) { auto c = i <=> j; return c <= 0; }
A bool f7 (double i, double j) { auto c = i <=> j; return c == std::partial_ordering::less; }
A bool f8 (double i, double j) { auto c = i <=> j; return c != std::partial_ordering::less; }
A bool f9 (double i, double j) { auto c = i <=> j; return c == std::partial_ordering::equivalent; }
A bool f10 (double i, double j) { auto c = i <=> j; return c != std::partial_ordering::equivalent; }
A bool f11 (double i, double j) { auto c = i <=> j; return c == std::partial_ordering::greater; }
A bool f12 (double i, double j) { auto c = i <=> j; return c != std::partial_ordering::greater; }
A bool f13 (double i) { auto c = i <=> 5.0; return c == 0; }
A bool f14 (double i) { auto c = i <=> 5.0; return c != 0; }
A bool f15 (double i) { auto c = i <=> 5.0; return c > 0; }
A bool f16 (double i) { auto c = i <=> 5.0; return c < 0; }
A bool f17 (double i) { auto c = i <=> 5.0; return c >= 0; }
A bool f18 (double i) { auto c = i <=> 5.0; return c <= 0; }
A bool f19 (double i) { auto c = i <=> 5.0; return c == std::partial_ordering::less; }
A bool f20 (double i) { auto c = i <=> 5.0; return c != std::partial_ordering::less; }
A bool f21 (double i) { auto c = i <=> 5.0; return c == std::partial_ordering::equivalent; }
A bool f22 (double i) { auto c = i <=> 5.0; return c != std::partial_ordering::equivalent; }
A bool f23 (double i) { auto c = i <=> 5.0; return c == std::partial_ordering::greater; }
A bool f24 (double i) { auto c = i <=> 5.0; return c != std::partial_ordering::greater; }

View file

@ -0,0 +1,84 @@
// { dg-do run { target c++20 } }
// { dg-options "-O2 -g" }
#include "pr94589-1.C"
#define C(fn, i, j, r) if (fn (i, j) != r) __builtin_abort ()
#define D(fn, i, r) if (fn (i) != r) __builtin_abort ()
int
main ()
{
C (f1, 7, 8, false);
C (f1, 8, 8, true);
C (f1, 9, 8, false);
C (f2, 7, 8, true);
C (f2, 8, 8, false);
C (f2, 9, 8, true);
C (f3, 7, 8, false);
C (f3, 8, 8, false);
C (f3, 9, 8, true);
C (f4, 7, 8, true);
C (f4, 8, 8, false);
C (f4, 9, 8, false);
C (f5, 7, 8, false);
C (f5, 8, 8, true);
C (f5, 9, 8, true);
C (f6, 7, 8, true);
C (f6, 8, 8, true);
C (f6, 9, 8, false);
C (f7, 7, 8, true);
C (f7, 8, 8, false);
C (f7, 9, 8, false);
C (f8, 7, 8, false);
C (f8, 8, 8, true);
C (f8, 9, 8, true);
C (f9, 7, 8, false);
C (f9, 8, 8, true);
C (f9, 9, 8, false);
C (f10, 7, 8, true);
C (f10, 8, 8, false);
C (f10, 9, 8, true);
C (f11, 7, 8, false);
C (f11, 8, 8, false);
C (f11, 9, 8, true);
C (f12, 7, 8, true);
C (f12, 8, 8, true);
C (f12, 9, 8, false);
D (f13, 4, false);
D (f13, 5, true);
D (f13, 6, false);
D (f14, 4, true);
D (f14, 5, false);
D (f14, 6, true);
D (f15, 4, false);
D (f15, 5, false);
D (f15, 6, true);
D (f16, 4, true);
D (f16, 5, false);
D (f16, 6, false);
D (f17, 4, false);
D (f17, 5, true);
D (f17, 6, true);
D (f18, 4, true);
D (f18, 5, true);
D (f18, 6, false);
D (f19, 4, true);
D (f19, 5, false);
D (f19, 6, false);
D (f20, 4, false);
D (f20, 5, true);
D (f20, 6, true);
D (f21, 4, false);
D (f21, 5, true);
D (f21, 6, false);
D (f22, 4, true);
D (f22, 5, false);
D (f22, 6, true);
D (f23, 4, false);
D (f23, 5, false);
D (f23, 6, true);
D (f24, 4, true);
D (f24, 5, true);
D (f24, 6, false);
}

View file

@ -0,0 +1,84 @@
// { dg-do run { target c++20 } }
// { dg-options "-O2 -g -ffast-math" }
#include "pr94589-2.C"
#define C(fn, i, j, r) if (fn (i, j) != r) __builtin_abort ()
#define D(fn, i, r) if (fn (i) != r) __builtin_abort ()
int
main ()
{
C (f1, 7.0, 8.0, false);
C (f1, 8.0, 8.0, true);
C (f1, 9.0, 8.0, false);
C (f2, 7.0, 8.0, true);
C (f2, 8.0, 8.0, false);
C (f2, 9.0, 8.0, true);
C (f3, 7.0, 8.0, false);
C (f3, 8.0, 8.0, false);
C (f3, 9.0, 8.0, true);
C (f4, 7.0, 8.0, true);
C (f4, 8.0, 8.0, false);
C (f4, 9.0, 8.0, false);
C (f5, 7.0, 8.0, false);
C (f5, 8.0, 8.0, true);
C (f5, 9.0, 8.0, true);
C (f6, 7.0, 8.0, true);
C (f6, 8.0, 8.0, true);
C (f6, 9.0, 8.0, false);
C (f7, 7.0, 8.0, true);
C (f7, 8.0, 8.0, false);
C (f7, 9.0, 8.0, false);
C (f8, 7.0, 8.0, false);
C (f8, 8.0, 8.0, true);
C (f8, 9.0, 8.0, true);
C (f9, 7.0, 8.0, false);
C (f9, 8.0, 8.0, true);
C (f9, 9.0, 8.0, false);
C (f10, 7.0, 8.0, true);
C (f10, 8.0, 8.0, false);
C (f10, 9.0, 8.0, true);
C (f11, 7.0, 8.0, false);
C (f11, 8.0, 8.0, false);
C (f11, 9.0, 8.0, true);
C (f12, 7.0, 8.0, true);
C (f12, 8.0, 8.0, true);
C (f12, 9.0, 8.0, false);
D (f13, 4.0, false);
D (f13, 5.0, true);
D (f13, 6.0, false);
D (f14, 4.0, true);
D (f14, 5.0, false);
D (f14, 6.0, true);
D (f15, 4.0, false);
D (f15, 5.0, false);
D (f15, 6.0, true);
D (f16, 4.0, true);
D (f16, 5.0, false);
D (f16, 6.0, false);
D (f17, 4.0, false);
D (f17, 5.0, true);
D (f17, 6.0, true);
D (f18, 4.0, true);
D (f18, 5.0, true);
D (f18, 6.0, false);
D (f19, 4.0, true);
D (f19, 5.0, false);
D (f19, 6.0, false);
D (f20, 4.0, false);
D (f20, 5.0, true);
D (f20, 6.0, true);
D (f21, 4.0, false);
D (f21, 5.0, true);
D (f21, 6.0, false);
D (f22, 4.0, true);
D (f22, 5.0, false);
D (f22, 6.0, true);
D (f23, 4.0, false);
D (f23, 5.0, false);
D (f23, 6.0, true);
D (f24, 4.0, true);
D (f24, 5.0, true);
D (f24, 6.0, false);
}

View file

@ -0,0 +1,35 @@
/* PR tree-optimization/94589 */
/* { dg-do compile } */
/* { dg-options "-O2 -g0 -fdump-tree-optimized" } */
/* { dg-final { scan-tree-dump-times "\[ij]_\[0-9]+\\(D\\) (?:<|<=|==|!=|>|>=) \[ij]_\[0-9]+\\(D\\)" 14 "optimized" } } */
/* { dg-final { scan-tree-dump-times "i_\[0-9]+\\(D\\) (?:<|<=|==|!=|>|>=) \[45]" 14 "optimized" } } */
#define A __attribute__((noipa))
A int f1 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c == 0; }
A int f2 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c != 0; }
A int f3 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c > 0; }
A int f4 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c < 0; }
A int f5 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c >= 0; }
A int f6 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c <= 0; }
A int f7 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c == -1; }
A int f8 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c != -1; }
A int f9 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c > -1; }
A int f10 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c <= -1; }
A int f11 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c == 1; }
A int f12 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c != 1; }
A int f13 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c < 1; }
A int f14 (int i, int j) { int c = i == j ? 0 : i < j ? -1 : 1; return c >= 1; }
A int f15 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c == 0; }
A int f16 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c != 0; }
A int f17 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c > 0; }
A int f18 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c < 0; }
A int f19 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c >= 0; }
A int f20 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c <= 0; }
A int f21 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c == -1; }
A int f22 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c != -1; }
A int f23 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c > -1; }
A int f24 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c <= -1; }
A int f25 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c == 1; }
A int f26 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c != 1; }
A int f27 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c < 1; }
A int f28 (int i) { int c = i == 5 ? 0 : i < 5 ? -1 : 1; return c >= 1; }

View file

@ -0,0 +1,35 @@
/* PR tree-optimization/94589 */
/* { dg-do compile } */
/* { dg-options "-O2 -g0 -ffast-math -fdump-tree-optimized" } */
/* { dg-final { scan-tree-dump-times "\[ij]_\[0-9]+\\(D\\) (?:<|<=|==|!=|>|>=) \[ij]_\[0-9]+\\(D\\)" 14 "optimized" } } */
/* { dg-final { scan-tree-dump-times "i_\[0-9]+\\(D\\) (?:<|<=|==|!=|>|>=) 5\\.0" 14 "optimized" } } */
#define A __attribute__((noipa))
A int f1 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c == 0; }
A int f2 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c != 0; }
A int f3 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c > 0; }
A int f4 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c < 0; }
A int f5 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c >= 0; }
A int f6 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c <= 0; }
A int f7 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c == -1; }
A int f8 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c != -1; }
A int f9 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c > -1; }
A int f10 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c <= -1; }
A int f11 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c == 1; }
A int f12 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c != 1; }
A int f13 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c < 1; }
A int f14 (double i, double j) { int c; if (i == j) c = 0; else if (i < j) c = -1; else if (i > j) c = 1; else c = 2; return c >= 1; }
A int f15 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c == 0; }
A int f16 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c != 0; }
A int f17 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c > 0; }
A int f18 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c < 0; }
A int f19 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c >= 0; }
A int f20 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c <= 0; }
A int f21 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c == -1; }
A int f22 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c != -1; }
A int f23 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c > -1; }
A int f24 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c <= -1; }
A int f25 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c == 1; }
A int f26 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c != 1; }
A int f27 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c < 1; }
A int f28 (double i) { int c; if (i == 5.0) c = 0; else if (i < 5.0) c = -1; else if (i > 5.0) c = 1; else c = 2; return c >= 1; }

View file

@ -0,0 +1,97 @@
/* { dg-do run } */
/* { dg-options "-O2 -g" } */
#include "pr94589-1.c"
#define C(fn, i, j, r) if (fn (i, j) != r) __builtin_abort ()
#define D(fn, i, r) if (fn (i) != r) __builtin_abort ()
int
main ()
{
C (f1, 7, 8, 0);
C (f1, 8, 8, 1);
C (f1, 9, 8, 0);
C (f2, 7, 8, 1);
C (f2, 8, 8, 0);
C (f2, 9, 8, 1);
C (f3, 7, 8, 0);
C (f3, 8, 8, 0);
C (f3, 9, 8, 1);
C (f4, 7, 8, 1);
C (f4, 8, 8, 0);
C (f4, 9, 8, 0);
C (f5, 7, 8, 0);
C (f5, 8, 8, 1);
C (f5, 9, 8, 1);
C (f6, 7, 8, 1);
C (f6, 8, 8, 1);
C (f6, 9, 8, 0);
C (f7, 7, 8, 1);
C (f7, 8, 8, 0);
C (f7, 9, 8, 0);
C (f8, 7, 8, 0);
C (f8, 8, 8, 1);
C (f8, 9, 8, 1);
C (f9, 7, 8, 0);
C (f9, 8, 8, 1);
C (f9, 9, 8, 1);
C (f10, 7, 8, 1);
C (f10, 8, 8, 0);
C (f10, 9, 8, 0);
C (f11, 7, 8, 0);
C (f11, 8, 8, 0);
C (f11, 9, 8, 1);
C (f12, 7, 8, 1);
C (f12, 8, 8, 1);
C (f12, 9, 8, 0);
C (f13, 7, 8, 1);
C (f13, 8, 8, 1);
C (f13, 9, 8, 0);
C (f14, 7, 8, 0);
C (f14, 8, 8, 0);
C (f14, 9, 8, 1);
D (f15, 4, 0);
D (f15, 5, 1);
D (f15, 6, 0);
D (f16, 4, 1);
D (f16, 5, 0);
D (f16, 6, 1);
D (f17, 4, 0);
D (f17, 5, 0);
D (f17, 6, 1);
D (f18, 4, 1);
D (f18, 5, 0);
D (f18, 6, 0);
D (f19, 4, 0);
D (f19, 5, 1);
D (f19, 6, 1);
D (f20, 4, 1);
D (f20, 5, 1);
D (f20, 6, 0);
D (f21, 4, 1);
D (f21, 5, 0);
D (f21, 6, 0);
D (f22, 4, 0);
D (f22, 5, 1);
D (f22, 6, 1);
D (f23, 4, 0);
D (f23, 5, 1);
D (f23, 6, 1);
D (f24, 4, 1);
D (f24, 5, 0);
D (f24, 6, 0);
D (f25, 4, 0);
D (f25, 5, 0);
D (f25, 6, 1);
D (f26, 4, 1);
D (f26, 5, 1);
D (f26, 6, 0);
D (f27, 4, 1);
D (f27, 5, 1);
D (f27, 6, 0);
D (f28, 4, 0);
D (f28, 5, 0);
D (f28, 6, 1);
return 0;
}

View file

@ -0,0 +1,97 @@
/* { dg-do run } */
/* { dg-options "-O2 -g -ffast-math" } */
#include "pr94589-2.c"
#define C(fn, i, j, r) if (fn (i, j) != r) __builtin_abort ()
#define D(fn, i, r) if (fn (i) != r) __builtin_abort ()
int
main ()
{
C (f1, 7.0, 8.0, 0);
C (f1, 8.0, 8.0, 1);
C (f1, 9.0, 8.0, 0);
C (f2, 7.0, 8.0, 1);
C (f2, 8.0, 8.0, 0);
C (f2, 9.0, 8.0, 1);
C (f3, 7.0, 8.0, 0);
C (f3, 8.0, 8.0, 0);
C (f3, 9.0, 8.0, 1);
C (f4, 7.0, 8.0, 1);
C (f4, 8.0, 8.0, 0);
C (f4, 9.0, 8.0, 0);
C (f5, 7.0, 8.0, 0);
C (f5, 8.0, 8.0, 1);
C (f5, 9.0, 8.0, 1);
C (f6, 7.0, 8.0, 1);
C (f6, 8.0, 8.0, 1);
C (f6, 9.0, 8.0, 0);
C (f7, 7.0, 8.0, 1);
C (f7, 8.0, 8.0, 0);
C (f7, 9.0, 8.0, 0);
C (f8, 7.0, 8.0, 0);
C (f8, 8.0, 8.0, 1);
C (f8, 9.0, 8.0, 1);
C (f9, 7.0, 8.0, 0);
C (f9, 8.0, 8.0, 1);
C (f9, 9.0, 8.0, 1);
C (f10, 7.0, 8.0, 1);
C (f10, 8.0, 8.0, 0);
C (f10, 9.0, 8.0, 0);
C (f11, 7.0, 8.0, 0);
C (f11, 8.0, 8.0, 0);
C (f11, 9.0, 8.0, 1);
C (f12, 7.0, 8.0, 1);
C (f12, 8.0, 8.0, 1);
C (f12, 9.0, 8.0, 0);
C (f13, 7.0, 8.0, 1);
C (f13, 8.0, 8.0, 1);
C (f13, 9.0, 8.0, 0);
C (f14, 7.0, 8.0, 0);
C (f14, 8.0, 8.0, 0);
C (f14, 9.0, 8.0, 1);
D (f15, 4.0, 0);
D (f15, 5.0, 1);
D (f15, 6.0, 0);
D (f16, 4.0, 1);
D (f16, 5.0, 0);
D (f16, 6.0, 1);
D (f17, 4.0, 0);
D (f17, 5.0, 0);
D (f17, 6.0, 1);
D (f18, 4.0, 1);
D (f18, 5.0, 0);
D (f18, 6.0, 0);
D (f19, 4.0, 0);
D (f19, 5.0, 1);
D (f19, 6.0, 1);
D (f20, 4.0, 1);
D (f20, 5.0, 1);
D (f20, 6.0, 0);
D (f21, 4.0, 1);
D (f21, 5.0, 0);
D (f21, 6.0, 0);
D (f22, 4.0, 0);
D (f22, 5.0, 1);
D (f22, 6.0, 1);
D (f23, 4.0, 0);
D (f23, 5.0, 1);
D (f23, 6.0, 1);
D (f24, 4.0, 1);
D (f24, 5.0, 0);
D (f24, 6.0, 0);
D (f25, 4.0, 0);
D (f25, 5.0, 0);
D (f25, 6.0, 1);
D (f26, 4.0, 1);
D (f26, 5.0, 1);
D (f26, 6.0, 0);
D (f27, 4.0, 1);
D (f27, 5.0, 1);
D (f27, 6.0, 0);
D (f28, 4.0, 0);
D (f28, 5.0, 0);
D (f28, 6.0, 1);
return 0;
}

View file

@ -64,6 +64,8 @@ static bool abs_replacement (basic_block, basic_block,
edge, edge, gimple *, tree, tree);
static bool xor_replacement (basic_block, basic_block,
edge, edge, gimple *, tree, tree);
static bool spaceship_replacement (basic_block, basic_block,
edge, edge, gphi *, tree, tree);
static bool cond_removal_in_popcount_clz_ctz_pattern (basic_block, basic_block,
edge, edge, gimple *,
tree, tree);
@ -359,6 +361,8 @@ tree_ssa_phiopt_worker (bool do_store_elim, bool do_hoist_loads, bool early_p)
cfgchanged = true;
else if (minmax_replacement (bb, bb1, e1, e2, phi, arg0, arg1))
cfgchanged = true;
else if (spaceship_replacement (bb, bb1, e1, e2, phi, arg0, arg1))
cfgchanged = true;
}
}
@ -1809,6 +1813,426 @@ minmax_replacement (basic_block cond_bb, basic_block middle_bb,
return true;
}
/* Return true if the only executable statement in BB is a GIMPLE_COND. */
static bool
cond_only_block_p (basic_block bb)
{
/* BB must have no executable statements. */
gimple_stmt_iterator gsi = gsi_after_labels (bb);
if (phi_nodes (bb))
return false;
while (!gsi_end_p (gsi))
{
gimple *stmt = gsi_stmt (gsi);
if (is_gimple_debug (stmt))
;
else if (gimple_code (stmt) == GIMPLE_NOP
|| gimple_code (stmt) == GIMPLE_PREDICT
|| gimple_code (stmt) == GIMPLE_COND)
;
else
return false;
gsi_next (&gsi);
}
return true;
}
/* Attempt to optimize (x <=> y) cmp 0 and similar comparisons.
For strong ordering <=> try to match something like:
<bb 2> : // cond3_bb (== cond2_bb)
if (x_4(D) != y_5(D))
goto <bb 3>; [INV]
else
goto <bb 6>; [INV]
<bb 3> : // cond_bb
if (x_4(D) < y_5(D))
goto <bb 6>; [INV]
else
goto <bb 4>; [INV]
<bb 4> : // middle_bb
<bb 6> : // phi_bb
# iftmp.0_2 = PHI <1(4), 0(2), -1(3)>
_1 = iftmp.0_2 == 0;
and for partial ordering <=> something like:
<bb 2> : // cond3_bb
if (a_3(D) == b_5(D))
goto <bb 6>; [50.00%]
else
goto <bb 3>; [50.00%]
<bb 3> [local count: 536870913]: // cond2_bb
if (a_3(D) < b_5(D))
goto <bb 6>; [50.00%]
else
goto <bb 4>; [50.00%]
<bb 4> [local count: 268435456]: // cond_bb
if (a_3(D) > b_5(D))
goto <bb 6>; [50.00%]
else
goto <bb 5>; [50.00%]
<bb 5> [local count: 134217728]: // middle_bb
<bb 6> [local count: 1073741824]: // phi_bb
# SR.27_4 = PHI <0(2), -1(3), 1(4), 2(5)>
_2 = SR.27_4 > 0; */
static bool
spaceship_replacement (basic_block cond_bb, basic_block middle_bb,
edge e0, edge e1, gphi *phi,
tree arg0, tree arg1)
{
if (!INTEGRAL_TYPE_P (TREE_TYPE (PHI_RESULT (phi)))
|| TYPE_UNSIGNED (TREE_TYPE (PHI_RESULT (phi)))
|| !tree_fits_shwi_p (arg0)
|| !tree_fits_shwi_p (arg1)
|| !IN_RANGE (tree_to_shwi (arg0), -1, 2)
|| !IN_RANGE (tree_to_shwi (arg1), -1, 2))
return false;
basic_block phi_bb = gimple_bb (phi);
gcc_assert (phi_bb == e0->dest && phi_bb == e1->dest);
if (!IN_RANGE (EDGE_COUNT (phi_bb->preds), 3, 4))
return false;
use_operand_p use_p;
gimple *use_stmt;
if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (PHI_RESULT (phi)))
return false;
if (!single_imm_use (PHI_RESULT (phi), &use_p, &use_stmt))
return false;
enum tree_code cmp;
tree lhs, rhs;
if (gimple_code (use_stmt) == GIMPLE_COND)
{
cmp = gimple_cond_code (use_stmt);
lhs = gimple_cond_lhs (use_stmt);
rhs = gimple_cond_rhs (use_stmt);
}
else if (is_gimple_assign (use_stmt))
{
if (gimple_assign_rhs_class (use_stmt) == GIMPLE_BINARY_RHS)
{
cmp = gimple_assign_rhs_code (use_stmt);
lhs = gimple_assign_rhs1 (use_stmt);
rhs = gimple_assign_rhs2 (use_stmt);
}
else if (gimple_assign_rhs_code (use_stmt) == COND_EXPR)
{
tree cond = gimple_assign_rhs1 (use_stmt);
if (!COMPARISON_CLASS_P (cond))
return false;
cmp = TREE_CODE (cond);
lhs = TREE_OPERAND (cond, 0);
rhs = TREE_OPERAND (cond, 1);
}
else
return false;
}
else
return false;
switch (cmp)
{
case EQ_EXPR:
case NE_EXPR:
case LT_EXPR:
case GT_EXPR:
case LE_EXPR:
case GE_EXPR:
break;
default:
return false;
}
if (lhs != PHI_RESULT (phi)
|| !tree_fits_shwi_p (rhs)
|| !IN_RANGE (tree_to_shwi (rhs), -1, 1))
return false;
if (!empty_block_p (middle_bb))
return false;
gcond *cond1 = as_a <gcond *> (last_stmt (cond_bb));
enum tree_code cmp1 = gimple_cond_code (cond1);
if (cmp1 != LT_EXPR && cmp1 != GT_EXPR)
return false;
tree lhs1 = gimple_cond_lhs (cond1);
tree rhs1 = gimple_cond_rhs (cond1);
/* The optimization may be unsafe due to NaNs. */
if (HONOR_NANS (TREE_TYPE (lhs1)))
return false;
if (TREE_CODE (lhs1) == SSA_NAME && SSA_NAME_OCCURS_IN_ABNORMAL_PHI (lhs1))
return false;
if (TREE_CODE (rhs1) == SSA_NAME && SSA_NAME_OCCURS_IN_ABNORMAL_PHI (rhs1))
return false;
if (!single_pred_p (cond_bb) || !cond_only_block_p (cond_bb))
return false;
basic_block cond2_bb = single_pred (cond_bb);
if (EDGE_COUNT (cond2_bb->succs) != 2)
return false;
edge cond2_phi_edge;
if (EDGE_SUCC (cond2_bb, 0)->dest == cond_bb)
{
if (EDGE_SUCC (cond2_bb, 1)->dest != phi_bb)
return false;
cond2_phi_edge = EDGE_SUCC (cond2_bb, 1);
}
else if (EDGE_SUCC (cond2_bb, 0)->dest != phi_bb)
return false;
else
cond2_phi_edge = EDGE_SUCC (cond2_bb, 0);
tree arg2 = gimple_phi_arg_def (phi, cond2_phi_edge->dest_idx);
if (!tree_fits_shwi_p (arg2))
return false;
gimple *cond2 = last_stmt (cond2_bb);
if (cond2 == NULL || gimple_code (cond2) != GIMPLE_COND)
return false;
enum tree_code cmp2 = gimple_cond_code (cond2);
tree lhs2 = gimple_cond_lhs (cond2);
tree rhs2 = gimple_cond_rhs (cond2);
if (lhs2 == lhs1)
{
if (!operand_equal_p (rhs2, rhs1, 0))
return false;
}
else if (lhs2 == rhs1)
{
if (rhs2 != lhs1)
return false;
}
else
return false;
tree arg3 = arg2;
basic_block cond3_bb = cond2_bb;
edge cond3_phi_edge = cond2_phi_edge;
gimple *cond3 = cond2;
enum tree_code cmp3 = cmp2;
tree lhs3 = lhs2;
tree rhs3 = rhs2;
if (EDGE_COUNT (phi_bb->preds) == 4)
{
if (absu_hwi (tree_to_shwi (arg2)) != 1)
return false;
if (e1->flags & EDGE_TRUE_VALUE)
{
if (tree_to_shwi (arg0) != 2
|| absu_hwi (tree_to_shwi (arg1)) != 1
|| wi::to_widest (arg1) == wi::to_widest (arg2))
return false;
}
else if (tree_to_shwi (arg1) != 2
|| absu_hwi (tree_to_shwi (arg0)) != 1
|| wi::to_widest (arg0) == wi::to_widest (arg1))
return false;
if (cmp2 != LT_EXPR && cmp2 != GT_EXPR)
return false;
/* if (x < y) goto phi_bb; else fallthru;
if (x > y) goto phi_bb; else fallthru;
bbx:;
phi_bb:;
is ok, but if x and y are swapped in one of the comparisons,
or the comparisons are the same and operands not swapped,
or second goto phi_bb is not the true edge, it is not. */
if ((lhs2 == lhs1)
^ (cmp2 == cmp1)
^ ((e1->flags & EDGE_TRUE_VALUE) != 0))
return false;
if ((cond2_phi_edge->flags & EDGE_TRUE_VALUE) == 0)
return false;
if (!single_pred_p (cond2_bb) || !cond_only_block_p (cond2_bb))
return false;
cond3_bb = single_pred (cond2_bb);
if (EDGE_COUNT (cond2_bb->succs) != 2)
return false;
if (EDGE_SUCC (cond3_bb, 0)->dest == cond2_bb)
{
if (EDGE_SUCC (cond3_bb, 1)->dest != phi_bb)
return false;
cond3_phi_edge = EDGE_SUCC (cond3_bb, 1);
}
else if (EDGE_SUCC (cond3_bb, 0)->dest != phi_bb)
return false;
else
cond3_phi_edge = EDGE_SUCC (cond3_bb, 0);
arg3 = gimple_phi_arg_def (phi, cond3_phi_edge->dest_idx);
cond3 = last_stmt (cond3_bb);
if (cond3 == NULL || gimple_code (cond3) != GIMPLE_COND)
return false;
cmp3 = gimple_cond_code (cond3);
lhs3 = gimple_cond_lhs (cond3);
rhs3 = gimple_cond_rhs (cond3);
if (lhs3 == lhs1)
{
if (!operand_equal_p (rhs3, rhs1, 0))
return false;
}
else if (lhs3 == rhs1)
{
if (rhs3 != lhs1)
return false;
}
else
return false;
}
else if (absu_hwi (tree_to_shwi (arg0)) != 1
|| absu_hwi (tree_to_shwi (arg1)) != 1
|| wi::to_widest (arg0) == wi::to_widest (arg1))
return false;
if (!integer_zerop (arg3) || (cmp3 != EQ_EXPR && cmp3 != NE_EXPR))
return false;
if ((cond3_phi_edge->flags & (cmp3 == EQ_EXPR
? EDGE_TRUE_VALUE : EDGE_FALSE_VALUE)) == 0)
return false;
/* lhs1 one_cmp rhs1 results in PHI_RESULT (phi) of 1. */
enum tree_code one_cmp;
if ((cmp1 == LT_EXPR)
^ (!integer_onep ((e1->flags & EDGE_TRUE_VALUE) ? arg1 : arg0)))
one_cmp = LT_EXPR;
else
one_cmp = GT_EXPR;
enum tree_code res_cmp;
switch (cmp)
{
case EQ_EXPR:
if (integer_zerop (rhs))
res_cmp = EQ_EXPR;
else if (integer_minus_onep (rhs))
res_cmp = one_cmp == LT_EXPR ? GT_EXPR : LT_EXPR;
else if (integer_onep (rhs))
res_cmp = one_cmp;
else
return false;
break;
case NE_EXPR:
if (integer_zerop (rhs))
res_cmp = NE_EXPR;
else if (integer_minus_onep (rhs))
res_cmp = one_cmp == LT_EXPR ? LE_EXPR : GE_EXPR;
else if (integer_onep (rhs))
res_cmp = one_cmp == LT_EXPR ? GE_EXPR : LE_EXPR;
else
return false;
break;
case LT_EXPR:
if (integer_onep (rhs))
res_cmp = one_cmp == LT_EXPR ? GE_EXPR : LE_EXPR;
else if (integer_zerop (rhs))
res_cmp = one_cmp == LT_EXPR ? GT_EXPR : LT_EXPR;
else
return false;
break;
case LE_EXPR:
if (integer_zerop (rhs))
res_cmp = one_cmp == LT_EXPR ? GE_EXPR : LE_EXPR;
else if (integer_minus_onep (rhs))
res_cmp = one_cmp == LT_EXPR ? GT_EXPR : LT_EXPR;
else
return false;
break;
case GT_EXPR:
if (integer_minus_onep (rhs))
res_cmp = one_cmp == LT_EXPR ? LE_EXPR : GE_EXPR;
else if (integer_zerop (rhs))
res_cmp = one_cmp;
else
return false;
break;
case GE_EXPR:
if (integer_zerop (rhs))
res_cmp = one_cmp == LT_EXPR ? LE_EXPR : GE_EXPR;
else if (integer_onep (rhs))
res_cmp = one_cmp;
else
return false;
break;
default:
gcc_unreachable ();
}
if (gimple_code (use_stmt) == GIMPLE_COND)
{
gcond *use_cond = as_a <gcond *> (use_stmt);
gimple_cond_set_code (use_cond, res_cmp);
gimple_cond_set_lhs (use_cond, lhs1);
gimple_cond_set_rhs (use_cond, rhs1);
}
else if (gimple_assign_rhs_class (use_stmt) == GIMPLE_BINARY_RHS)
{
gimple_assign_set_rhs_code (use_stmt, res_cmp);
gimple_assign_set_rhs1 (use_stmt, lhs1);
gimple_assign_set_rhs2 (use_stmt, rhs1);
}
else
{
tree cond = build2 (res_cmp, TREE_TYPE (gimple_assign_rhs1 (use_stmt)),
lhs1, rhs1);
gimple_assign_set_rhs1 (use_stmt, cond);
}
update_stmt (use_stmt);
if (MAY_HAVE_DEBUG_BIND_STMTS)
{
use_operand_p use_p;
imm_use_iterator iter;
bool has_debug_uses = false;
FOR_EACH_IMM_USE_FAST (use_p, iter, PHI_RESULT (phi))
{
gimple *use_stmt = USE_STMT (use_p);
gcc_assert (is_gimple_debug (use_stmt));
has_debug_uses = true;
break;
}
if (has_debug_uses)
{
/* If there are debug uses, emit something like:
# DEBUG D#1 => i_2(D) > j_3(D) ? 1 : -1
# DEBUG D#2 => i_2(D) == j_3(D) ? 0 : D#1
where > stands for the comparison that yielded 1
and replace debug uses of phi result with that D#2.
Ignore the value of 2, because if NaNs aren't expected,
all floating point numbers should be comparable. */
gimple_stmt_iterator gsi = gsi_after_labels (gimple_bb (phi));
tree type = TREE_TYPE (PHI_RESULT (phi));
tree temp1 = make_node (DEBUG_EXPR_DECL);
DECL_ARTIFICIAL (temp1) = 1;
TREE_TYPE (temp1) = type;
SET_DECL_MODE (temp1, TYPE_MODE (type));
tree t = build2 (one_cmp, boolean_type_node, lhs1, rhs2);
t = build3 (COND_EXPR, type, t, build_one_cst (type),
build_int_cst (type, -1));
gimple *g = gimple_build_debug_bind (temp1, t, phi);
gsi_insert_before (&gsi, g, GSI_SAME_STMT);
tree temp2 = make_node (DEBUG_EXPR_DECL);
DECL_ARTIFICIAL (temp2) = 1;
TREE_TYPE (temp2) = type;
SET_DECL_MODE (temp2, TYPE_MODE (type));
t = build2 (EQ_EXPR, boolean_type_node, lhs1, rhs2);
t = build3 (COND_EXPR, type, t, build_zero_cst (type), temp1);
g = gimple_build_debug_bind (temp2, t, phi);
gsi_insert_before (&gsi, g, GSI_SAME_STMT);
replace_uses_by (PHI_RESULT (phi), temp2);
}
}
gimple_stmt_iterator psi = gsi_for_stmt (phi);
remove_phi_node (&psi, true);
return true;
}
/* Convert
<bb 2>