RE: [PATCH v1 1/2] Match: Support __builtin_add_overflow for branchless unsigned SAT_ADD
Thanks Richard for comments, will merge the rest form of .SAT_ADD in one middle
end patch for fully picture, as well as comments addressing.
Pan
-Original Message-
From: Richard Biener
Sent: Wednesday, May 22, 2024 9:16 PM
To: Li, Pan2
Cc: gcc-patches@gcc.gnu.org; juzhe.zh...@rivai.ai; kito.ch...@gmail.com;
tamar.christ...@arm.com
Subject: Re: [PATCH v1 1/2] Match: Support __builtin_add_overflow for
branchless unsigned SAT_ADD
On Sun, May 19, 2024 at 8:37 AM wrote:
>
> From: Pan Li
>
> This patch would like to support the branchless form for unsigned
> SAT_ADD when leverage __builtin_add_overflow. For example as below:
>
> uint64_t sat_add_u(uint64_t x, uint64_t y)
> {
> uint64_t ret;
> uint64_t overflow = __builtin_add_overflow (x, y, );
>
> return (T)(-overflow) | ret;
> }
>
> Before this patch:
>
> uint64_t sat_add_u (uint64_t x, uint64_t y)
> {
> long unsigned int _1;
> long unsigned int _2;
> long unsigned int _3;
> __complex__ long unsigned int _6;
> uint64_t _8;
>
> ;; basic block 2, loop depth 0
> ;;pred: ENTRY
> _6 = .ADD_OVERFLOW (x_4(D), y_5(D));
> _1 = REALPART_EXPR <_6>;
> _2 = IMAGPART_EXPR <_6>;
> _3 = -_2;
> _8 = _1 | _3;
> return _8;
> ;;succ: EXIT
>
> }
>
> After this patch:
>
> uint64_t sat_add_u (uint64_t x, uint64_t y)
> {
> uint64_t _8;
>
> ;; basic block 2, loop depth 0
> ;;pred: ENTRY
> _8 = .SAT_ADD (x_4(D), y_5(D)); [tail call]
> return _8;
> ;;succ: EXIT
>
> }
>
> The below tests suite are passed for this patch.
> * The rv64gcv fully regression test.
> * The x86 bootstrap test.
> * The x86 fully regression test.
>
> gcc/ChangeLog:
>
> * match.pd: Add SAT_ADD right part 2 for __builtin_add_overflow.
>
> Signed-off-by: Pan Li
> ---
> gcc/match.pd | 4
> 1 file changed, 4 insertions(+)
>
> diff --git a/gcc/match.pd b/gcc/match.pd
> index b291e34bbe4..5328e846aff 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -3064,6 +3064,10 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> (negate (convert (ne (imagpart (IFN_ADD_OVERFLOW:c @0 @1)) integer_zerop)))
> (if (TYPE_UNSIGNED (type) && integer_types_ternary_match (type, @0, @1
>
> +(match (usadd_right_part_2 @0 @1)
> + (negate (imagpart (IFN_ADD_OVERFLOW:c @0 @1)))
> + (if (TYPE_UNSIGNED (type) && integer_types_ternary_match (type, @0, @1
> +
Can you merge this with the patch that makes use of the
usadd_right_part_2 match?
It's difficult to review on its own.
> /* We cannot merge or overload usadd_left_part_1 and usadd_left_part_2
> because the sub part of left_part_2 cannot work with right_part_1.
> For example, left_part_2 pattern focus one .ADD_OVERFLOW but the
> --
> 2.34.1
>
Re: [PATCH v1 1/2] Match: Support __builtin_add_overflow for branchless unsigned SAT_ADD
On Sun, May 19, 2024 at 8:37 AM wrote:
>
> From: Pan Li
>
> This patch would like to support the branchless form for unsigned
> SAT_ADD when leverage __builtin_add_overflow. For example as below:
>
> uint64_t sat_add_u(uint64_t x, uint64_t y)
> {
> uint64_t ret;
> uint64_t overflow = __builtin_add_overflow (x, y, );
>
> return (T)(-overflow) | ret;
> }
>
> Before this patch:
>
> uint64_t sat_add_u (uint64_t x, uint64_t y)
> {
> long unsigned int _1;
> long unsigned int _2;
> long unsigned int _3;
> __complex__ long unsigned int _6;
> uint64_t _8;
>
> ;; basic block 2, loop depth 0
> ;;pred: ENTRY
> _6 = .ADD_OVERFLOW (x_4(D), y_5(D));
> _1 = REALPART_EXPR <_6>;
> _2 = IMAGPART_EXPR <_6>;
> _3 = -_2;
> _8 = _1 | _3;
> return _8;
> ;;succ: EXIT
>
> }
>
> After this patch:
>
> uint64_t sat_add_u (uint64_t x, uint64_t y)
> {
> uint64_t _8;
>
> ;; basic block 2, loop depth 0
> ;;pred: ENTRY
> _8 = .SAT_ADD (x_4(D), y_5(D)); [tail call]
> return _8;
> ;;succ: EXIT
>
> }
>
> The below tests suite are passed for this patch.
> * The rv64gcv fully regression test.
> * The x86 bootstrap test.
> * The x86 fully regression test.
>
> gcc/ChangeLog:
>
> * match.pd: Add SAT_ADD right part 2 for __builtin_add_overflow.
>
> Signed-off-by: Pan Li
> ---
> gcc/match.pd | 4
> 1 file changed, 4 insertions(+)
>
> diff --git a/gcc/match.pd b/gcc/match.pd
> index b291e34bbe4..5328e846aff 100644
> --- a/gcc/match.pd
> +++ b/gcc/match.pd
> @@ -3064,6 +3064,10 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> (negate (convert (ne (imagpart (IFN_ADD_OVERFLOW:c @0 @1)) integer_zerop)))
> (if (TYPE_UNSIGNED (type) && integer_types_ternary_match (type, @0, @1
>
> +(match (usadd_right_part_2 @0 @1)
> + (negate (imagpart (IFN_ADD_OVERFLOW:c @0 @1)))
> + (if (TYPE_UNSIGNED (type) && integer_types_ternary_match (type, @0, @1
> +
Can you merge this with the patch that makes use of the
usadd_right_part_2 match?
It's difficult to review on its own.
> /* We cannot merge or overload usadd_left_part_1 and usadd_left_part_2
> because the sub part of left_part_2 cannot work with right_part_1.
> For example, left_part_2 pattern focus one .ADD_OVERFLOW but the
> --
> 2.34.1
>
[PATCH v1 1/2] Match: Support __builtin_add_overflow for branchless unsigned SAT_ADD
From: Pan Li
This patch would like to support the branchless form for unsigned
SAT_ADD when leverage __builtin_add_overflow. For example as below:
uint64_t sat_add_u(uint64_t x, uint64_t y)
{
uint64_t ret;
uint64_t overflow = __builtin_add_overflow (x, y, );
return (T)(-overflow) | ret;
}
Before this patch:
uint64_t sat_add_u (uint64_t x, uint64_t y)
{
long unsigned int _1;
long unsigned int _2;
long unsigned int _3;
__complex__ long unsigned int _6;
uint64_t _8;
;; basic block 2, loop depth 0
;;pred: ENTRY
_6 = .ADD_OVERFLOW (x_4(D), y_5(D));
_1 = REALPART_EXPR <_6>;
_2 = IMAGPART_EXPR <_6>;
_3 = -_2;
_8 = _1 | _3;
return _8;
;;succ: EXIT
}
After this patch:
uint64_t sat_add_u (uint64_t x, uint64_t y)
{
uint64_t _8;
;; basic block 2, loop depth 0
;;pred: ENTRY
_8 = .SAT_ADD (x_4(D), y_5(D)); [tail call]
return _8;
;;succ: EXIT
}
The below tests suite are passed for this patch.
* The rv64gcv fully regression test.
* The x86 bootstrap test.
* The x86 fully regression test.
gcc/ChangeLog:
* match.pd: Add SAT_ADD right part 2 for __builtin_add_overflow.
Signed-off-by: Pan Li
---
gcc/match.pd | 4
1 file changed, 4 insertions(+)
diff --git a/gcc/match.pd b/gcc/match.pd
index b291e34bbe4..5328e846aff 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -3064,6 +3064,10 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(negate (convert (ne (imagpart (IFN_ADD_OVERFLOW:c @0 @1)) integer_zerop)))
(if (TYPE_UNSIGNED (type) && integer_types_ternary_match (type, @0, @1
+(match (usadd_right_part_2 @0 @1)
+ (negate (imagpart (IFN_ADD_OVERFLOW:c @0 @1)))
+ (if (TYPE_UNSIGNED (type) && integer_types_ternary_match (type, @0, @1
+
/* We cannot merge or overload usadd_left_part_1 and usadd_left_part_2
because the sub part of left_part_2 cannot work with right_part_1.
For example, left_part_2 pattern focus one .ADD_OVERFLOW but the
--
2.34.1
