* [PATCH] Extend optimization for integer bit test on __atomic_fetch_[or|and]_*
@ 2022-11-02 19:08 H.J. Lu
2022-11-05 11:21 ` Richard Biener
0 siblings, 1 reply; 2+ messages in thread
From: H.J. Lu @ 2022-11-02 19:08 UTC (permalink / raw)
To: gcc-patches; +Cc: liuhongt
Extend optimization for
_1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
_5 = (signed int) _1;
_4 = _5 >= 0;
to
_1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
_5 = (signed int) _1;
if (_5 >= 0)
gcc/
PR middle-end/102566
* tree-ssa-ccp.cc (optimize_atomic_bit_test_and): Also handle
if (_5 < 0) and if (_5 >= 0).
gcc/testsuite/
PR middle-end/102566
* g++.target/i386/pr102566-7.C
---
gcc/testsuite/g++.target/i386/pr102566-7.C | 22 ++++++
gcc/tree-ssa-ccp.cc | 84 ++++++++++++++++++----
2 files changed, 91 insertions(+), 15 deletions(-)
create mode 100644 gcc/testsuite/g++.target/i386/pr102566-7.C
diff --git a/gcc/testsuite/g++.target/i386/pr102566-7.C b/gcc/testsuite/g++.target/i386/pr102566-7.C
new file mode 100644
index 00000000000..ce90214f33d
--- /dev/null
+++ b/gcc/testsuite/g++.target/i386/pr102566-7.C
@@ -0,0 +1,22 @@
+/* { dg-do compile { target c++11 } } */
+/* { dg-options "-O2" } */
+
+#include <atomic>
+
+template<uint32_t b>
+void lock_bts(std::atomic<uint32_t> &a) { while (!(a.fetch_or(b) & b)); }
+template<uint32_t b>
+void lock_btr(std::atomic<uint32_t> &a) { while (a.fetch_and(~b) & b); }
+template<uint32_t b>
+void lock_btc(std::atomic<uint32_t> &a) { while (a.fetch_xor(b) & b); }
+template void lock_bts<1U<<30>(std::atomic<uint32_t> &a);
+template void lock_btr<1U<<30>(std::atomic<uint32_t> &a);
+template void lock_btc<1U<<30>(std::atomic<uint32_t> &a);
+template void lock_bts<1U<<31>(std::atomic<uint32_t> &a);
+template void lock_btr<1U<<31>(std::atomic<uint32_t> &a);
+template void lock_btc<1U<<31>(std::atomic<uint32_t> &a);
+
+/* { dg-final { scan-assembler-times "lock;?\[ \t\]*btsl" 2 } } */
+/* { dg-final { scan-assembler-times "lock;?\[ \t\]*btrl" 2 } } */
+/* { dg-final { scan-assembler-times "lock;?\[ \t\]*btcl" 2 } } */
+/* { dg-final { scan-assembler-not "cmpxchg" } } */
diff --git a/gcc/tree-ssa-ccp.cc b/gcc/tree-ssa-ccp.cc
index 9778e776cf2..3a4b6bc1118 100644
--- a/gcc/tree-ssa-ccp.cc
+++ b/gcc/tree-ssa-ccp.cc
@@ -3471,17 +3471,35 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
{
gimple *use_nop_stmt;
if (!single_imm_use (use_lhs, &use_p, &use_nop_stmt)
- || !is_gimple_assign (use_nop_stmt))
+ || (!is_gimple_assign (use_nop_stmt)
+ && gimple_code (use_nop_stmt) != GIMPLE_COND))
return false;
- tree use_nop_lhs = gimple_assign_lhs (use_nop_stmt);
- rhs_code = gimple_assign_rhs_code (use_nop_stmt);
- if (rhs_code != BIT_AND_EXPR)
+ /* Handle both
+ _4 = _5 < 0;
+ and
+ if (_5 < 0)
+ */
+ tree use_nop_lhs = nullptr;
+ rhs_code = ERROR_MARK;
+ if (is_gimple_assign (use_nop_stmt))
{
- if (TREE_CODE (use_nop_lhs) == SSA_NAME
+ use_nop_lhs = gimple_assign_lhs (use_nop_stmt);
+ rhs_code = gimple_assign_rhs_code (use_nop_stmt);
+ }
+ if (!use_nop_lhs || rhs_code != BIT_AND_EXPR)
+ {
+ /* Also handle
+ if (_5 < 0)
+ */
+ if (use_nop_lhs
+ && TREE_CODE (use_nop_lhs) == SSA_NAME
&& SSA_NAME_OCCURS_IN_ABNORMAL_PHI (use_nop_lhs))
return false;
- if (rhs_code == BIT_NOT_EXPR)
+ if (use_nop_lhs && rhs_code == BIT_NOT_EXPR)
{
+ /* Handle
+ _7 = ~_2;
+ */
g = convert_atomic_bit_not (fn, use_nop_stmt, lhs,
mask);
if (!g)
@@ -3512,14 +3530,31 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
}
else
{
- if (TREE_CODE (TREE_TYPE (use_nop_lhs)) != BOOLEAN_TYPE)
- return false;
+ tree cmp_rhs1, cmp_rhs2;
+ if (use_nop_lhs)
+ {
+ /* Handle
+ _4 = _5 < 0;
+ */
+ if (TREE_CODE (TREE_TYPE (use_nop_lhs))
+ != BOOLEAN_TYPE)
+ return false;
+ cmp_rhs1 = gimple_assign_rhs1 (use_nop_stmt);
+ cmp_rhs2 = gimple_assign_rhs2 (use_nop_stmt);
+ }
+ else
+ {
+ /* Handle
+ if (_5 < 0)
+ */
+ rhs_code = gimple_cond_code (use_nop_stmt);
+ cmp_rhs1 = gimple_cond_lhs (use_nop_stmt);
+ cmp_rhs2 = gimple_cond_rhs (use_nop_stmt);
+ }
if (rhs_code != GE_EXPR && rhs_code != LT_EXPR)
return false;
- tree cmp_rhs1 = gimple_assign_rhs1 (use_nop_stmt);
if (use_lhs != cmp_rhs1)
return false;
- tree cmp_rhs2 = gimple_assign_rhs2 (use_nop_stmt);
if (!integer_zerop (cmp_rhs2))
return false;
@@ -3547,6 +3582,14 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
_1 = __atomic_fetch_and_4 (ptr_6, 0x7fffffff, _3);
_6 = _1 & 0x80000000;
_4 = _6 != 0 or _6 == 0;
+ and convert
+ _1 = __atomic_fetch_and_4 (ptr_6, 0x7fffffff, _3);
+ _5 = (signed int) _1;
+ if (_5 < 0 or _5 >= 0)
+ to
+ _1 = __atomic_fetch_and_4 (ptr_6, 0x7fffffff, _3);
+ _6 = _1 & 0x80000000;
+ if (_6 != 0 or _6 == 0)
*/
and_mask = build_int_cst (TREE_TYPE (use_rhs),
highest);
@@ -3567,6 +3610,14 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
_1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
_6 = _1 & 0x80000000;
_4 = _6 != 0 or _6 == 0;
+ and convert
+ _1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
+ _5 = (signed int) _1;
+ if (_5 < 0 or _5 >= 0)
+ to
+ _1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
+ _6 = _1 & 0x80000000;
+ if (_6 != 0 or _6 == 0)
*/
}
var = make_ssa_name (TREE_TYPE (use_rhs));
@@ -3577,11 +3628,14 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
gsi = gsi_for_stmt (use_nop_stmt);
gsi_insert_before (&gsi, g, GSI_NEW_STMT);
use_stmt = g;
- g = gimple_build_assign (use_nop_lhs,
- (rhs_code == GE_EXPR
- ? EQ_EXPR : NE_EXPR),
- var,
- build_zero_cst (TREE_TYPE (use_rhs)));
+ rhs_code = rhs_code == GE_EXPR ? EQ_EXPR : NE_EXPR;
+ tree const_zero = build_zero_cst (TREE_TYPE (use_rhs));
+ if (use_nop_lhs)
+ g = gimple_build_assign (use_nop_lhs, rhs_code,
+ var, const_zero);
+ else
+ g = gimple_build_cond (rhs_code, var, const_zero,
+ nullptr, nullptr);
gsi_insert_after (&gsi, g, GSI_NEW_STMT);
gsi = gsi_for_stmt (use_nop_stmt);
gsi_remove (&gsi, true);
--
2.37.3
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH] Extend optimization for integer bit test on __atomic_fetch_[or|and]_*
2022-11-02 19:08 [PATCH] Extend optimization for integer bit test on __atomic_fetch_[or|and]_* H.J. Lu
@ 2022-11-05 11:21 ` Richard Biener
0 siblings, 0 replies; 2+ messages in thread
From: Richard Biener @ 2022-11-05 11:21 UTC (permalink / raw)
To: H.J. Lu; +Cc: gcc-patches, liuhongt
On Wed, Nov 2, 2022 at 8:09 PM H.J. Lu via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> Extend optimization for
>
> _1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
> _5 = (signed int) _1;
> _4 = _5 >= 0;
>
> to
>
> _1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
> _5 = (signed int) _1;
> if (_5 >= 0)
OK.
> gcc/
>
> PR middle-end/102566
> * tree-ssa-ccp.cc (optimize_atomic_bit_test_and): Also handle
> if (_5 < 0) and if (_5 >= 0).
>
> gcc/testsuite/
>
> PR middle-end/102566
> * g++.target/i386/pr102566-7.C
> ---
> gcc/testsuite/g++.target/i386/pr102566-7.C | 22 ++++++
> gcc/tree-ssa-ccp.cc | 84 ++++++++++++++++++----
> 2 files changed, 91 insertions(+), 15 deletions(-)
> create mode 100644 gcc/testsuite/g++.target/i386/pr102566-7.C
>
> diff --git a/gcc/testsuite/g++.target/i386/pr102566-7.C b/gcc/testsuite/g++.target/i386/pr102566-7.C
> new file mode 100644
> index 00000000000..ce90214f33d
> --- /dev/null
> +++ b/gcc/testsuite/g++.target/i386/pr102566-7.C
> @@ -0,0 +1,22 @@
> +/* { dg-do compile { target c++11 } } */
> +/* { dg-options "-O2" } */
> +
> +#include <atomic>
> +
> +template<uint32_t b>
> +void lock_bts(std::atomic<uint32_t> &a) { while (!(a.fetch_or(b) & b)); }
> +template<uint32_t b>
> +void lock_btr(std::atomic<uint32_t> &a) { while (a.fetch_and(~b) & b); }
> +template<uint32_t b>
> +void lock_btc(std::atomic<uint32_t> &a) { while (a.fetch_xor(b) & b); }
> +template void lock_bts<1U<<30>(std::atomic<uint32_t> &a);
> +template void lock_btr<1U<<30>(std::atomic<uint32_t> &a);
> +template void lock_btc<1U<<30>(std::atomic<uint32_t> &a);
> +template void lock_bts<1U<<31>(std::atomic<uint32_t> &a);
> +template void lock_btr<1U<<31>(std::atomic<uint32_t> &a);
> +template void lock_btc<1U<<31>(std::atomic<uint32_t> &a);
> +
> +/* { dg-final { scan-assembler-times "lock;?\[ \t\]*btsl" 2 } } */
> +/* { dg-final { scan-assembler-times "lock;?\[ \t\]*btrl" 2 } } */
> +/* { dg-final { scan-assembler-times "lock;?\[ \t\]*btcl" 2 } } */
> +/* { dg-final { scan-assembler-not "cmpxchg" } } */
> diff --git a/gcc/tree-ssa-ccp.cc b/gcc/tree-ssa-ccp.cc
> index 9778e776cf2..3a4b6bc1118 100644
> --- a/gcc/tree-ssa-ccp.cc
> +++ b/gcc/tree-ssa-ccp.cc
> @@ -3471,17 +3471,35 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
> {
> gimple *use_nop_stmt;
> if (!single_imm_use (use_lhs, &use_p, &use_nop_stmt)
> - || !is_gimple_assign (use_nop_stmt))
> + || (!is_gimple_assign (use_nop_stmt)
> + && gimple_code (use_nop_stmt) != GIMPLE_COND))
> return false;
> - tree use_nop_lhs = gimple_assign_lhs (use_nop_stmt);
> - rhs_code = gimple_assign_rhs_code (use_nop_stmt);
> - if (rhs_code != BIT_AND_EXPR)
> + /* Handle both
> + _4 = _5 < 0;
> + and
> + if (_5 < 0)
> + */
> + tree use_nop_lhs = nullptr;
> + rhs_code = ERROR_MARK;
> + if (is_gimple_assign (use_nop_stmt))
> {
> - if (TREE_CODE (use_nop_lhs) == SSA_NAME
> + use_nop_lhs = gimple_assign_lhs (use_nop_stmt);
> + rhs_code = gimple_assign_rhs_code (use_nop_stmt);
> + }
> + if (!use_nop_lhs || rhs_code != BIT_AND_EXPR)
> + {
> + /* Also handle
> + if (_5 < 0)
> + */
> + if (use_nop_lhs
> + && TREE_CODE (use_nop_lhs) == SSA_NAME
> && SSA_NAME_OCCURS_IN_ABNORMAL_PHI (use_nop_lhs))
> return false;
> - if (rhs_code == BIT_NOT_EXPR)
> + if (use_nop_lhs && rhs_code == BIT_NOT_EXPR)
> {
> + /* Handle
> + _7 = ~_2;
> + */
> g = convert_atomic_bit_not (fn, use_nop_stmt, lhs,
> mask);
> if (!g)
> @@ -3512,14 +3530,31 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
> }
> else
> {
> - if (TREE_CODE (TREE_TYPE (use_nop_lhs)) != BOOLEAN_TYPE)
> - return false;
> + tree cmp_rhs1, cmp_rhs2;
> + if (use_nop_lhs)
> + {
> + /* Handle
> + _4 = _5 < 0;
> + */
> + if (TREE_CODE (TREE_TYPE (use_nop_lhs))
> + != BOOLEAN_TYPE)
> + return false;
> + cmp_rhs1 = gimple_assign_rhs1 (use_nop_stmt);
> + cmp_rhs2 = gimple_assign_rhs2 (use_nop_stmt);
> + }
> + else
> + {
> + /* Handle
> + if (_5 < 0)
> + */
> + rhs_code = gimple_cond_code (use_nop_stmt);
> + cmp_rhs1 = gimple_cond_lhs (use_nop_stmt);
> + cmp_rhs2 = gimple_cond_rhs (use_nop_stmt);
> + }
> if (rhs_code != GE_EXPR && rhs_code != LT_EXPR)
> return false;
> - tree cmp_rhs1 = gimple_assign_rhs1 (use_nop_stmt);
> if (use_lhs != cmp_rhs1)
> return false;
> - tree cmp_rhs2 = gimple_assign_rhs2 (use_nop_stmt);
> if (!integer_zerop (cmp_rhs2))
> return false;
>
> @@ -3547,6 +3582,14 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
> _1 = __atomic_fetch_and_4 (ptr_6, 0x7fffffff, _3);
> _6 = _1 & 0x80000000;
> _4 = _6 != 0 or _6 == 0;
> + and convert
> + _1 = __atomic_fetch_and_4 (ptr_6, 0x7fffffff, _3);
> + _5 = (signed int) _1;
> + if (_5 < 0 or _5 >= 0)
> + to
> + _1 = __atomic_fetch_and_4 (ptr_6, 0x7fffffff, _3);
> + _6 = _1 & 0x80000000;
> + if (_6 != 0 or _6 == 0)
> */
> and_mask = build_int_cst (TREE_TYPE (use_rhs),
> highest);
> @@ -3567,6 +3610,14 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
> _1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
> _6 = _1 & 0x80000000;
> _4 = _6 != 0 or _6 == 0;
> + and convert
> + _1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
> + _5 = (signed int) _1;
> + if (_5 < 0 or _5 >= 0)
> + to
> + _1 = __atomic_fetch_or_4 (ptr_6, 0x80000000, _3);
> + _6 = _1 & 0x80000000;
> + if (_6 != 0 or _6 == 0)
> */
> }
> var = make_ssa_name (TREE_TYPE (use_rhs));
> @@ -3577,11 +3628,14 @@ optimize_atomic_bit_test_and (gimple_stmt_iterator *gsip,
> gsi = gsi_for_stmt (use_nop_stmt);
> gsi_insert_before (&gsi, g, GSI_NEW_STMT);
> use_stmt = g;
> - g = gimple_build_assign (use_nop_lhs,
> - (rhs_code == GE_EXPR
> - ? EQ_EXPR : NE_EXPR),
> - var,
> - build_zero_cst (TREE_TYPE (use_rhs)));
> + rhs_code = rhs_code == GE_EXPR ? EQ_EXPR : NE_EXPR;
> + tree const_zero = build_zero_cst (TREE_TYPE (use_rhs));
> + if (use_nop_lhs)
> + g = gimple_build_assign (use_nop_lhs, rhs_code,
> + var, const_zero);
> + else
> + g = gimple_build_cond (rhs_code, var, const_zero,
> + nullptr, nullptr);
> gsi_insert_after (&gsi, g, GSI_NEW_STMT);
> gsi = gsi_for_stmt (use_nop_stmt);
> gsi_remove (&gsi, true);
> --
> 2.37.3
>
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2022-11-05 11:21 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2022-11-02 19:08 [PATCH] Extend optimization for integer bit test on __atomic_fetch_[or|and]_* H.J. Lu
2022-11-05 11:21 ` Richard Biener
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).