[00/13] Saturating arithmetic matching improvements

Message ID	20260902145254.77832-1-ktkachov@nvidia.com
Headers	DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 1454C4BA2E27 Received-SPF: Pass (protection.outlook.com: domain of nvidia.com designates 216.228.117.161 as permitted sender) receiver=protection.outlook.com; client-ip=216.228.117.161; helo=mail.nvidia.com; pr=C From: <ktkachov@nvidia.com> To: <gcc-patches@gcc.gnu.org> CC: Kyrylo Tkachov <ktkachov@nvidia.com> Subject: [PATCH 00/13] Saturating arithmetic matching improvements Date: Wed, 2 Sep 2026 16:52:40 +0200 Message-ID: <20260902145254.77832-1-ktkachov@nvidia.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Content-Type: text/plain Precedence: list Errors-To: gcc-patches-bounces~patchwork=sourceware.org@gcc.gnu.org
Series	Saturating arithmetic matching improvements \| [00/13] Saturating arithmetic matching improvements [01/13] match-sat-alu.pd: Allow wider ranges for signed saturating truncation [02/13] match-sat-alu.pd: Recognize X - MIN (X, Y) as saturating subtraction [03/13] match-sat-alu.pd: Recognize signed saturation clamps [04/13] match-sat-alu.pd: Recognize borrow-test saturating subtraction [05/13] match-sat-alu.pd: Recognize more unsigned saturating subtraction forms [06/13] match-sat-alu.pd: Recognize complement tests for unsigned saturating addition [07/13] match-sat-alu.pd: Recognize X + MIN (~X, Y) as unsigned saturating addition [08/13] match-sat-alu.pd: Recognize strict widened unsigned saturation clamps [09/13] match-sat-alu.pd: Recognize signed saturating truncation after a shift [10/13] match-sat-alu.pd: Recognize unsigned constant-clamp truncation [11/13] match-sat-alu.pd: Recognize signed constant-clamp truncation [12/13] match-sat-alu.pd: Recognize complement masks for unsigned saturating truncation [13/13] match.pd: Narrow promoted saturating addition and subtraction

Message ID

20260902145254.77832-1-ktkachov@nvidia.com

Headers

DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 1454C4BA2E27
Received-SPF: Pass (protection.outlook.com: domain of nvidia.com designates
 216.228.117.161 as permitted sender) receiver=protection.outlook.com;
 client-ip=216.228.117.161; helo=mail.nvidia.com; pr=C
From: <ktkachov@nvidia.com>
To: <gcc-patches@gcc.gnu.org>
CC: Kyrylo Tkachov <ktkachov@nvidia.com>
Subject: [PATCH 00/13] Saturating arithmetic matching improvements
Date: Wed, 2 Sep 2026 16:52:40 +0200
Message-ID: <20260902145254.77832-1-ktkachov@nvidia.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Content-Type: text/plain
X-MS-Exchange-AntiSpam-MessageData-ChunkCount: 1
X-MS-Exchange-AntiSpam-MessageData-0: 
 NgS7X1ptmYV/jRrhoilhQhsnkAnEe+hgwDxPnAa3Cri6M/3Dd8xwXEH1m+Ii/EqUXmxlguCIZma5kFhf6nO01gfp1y1eiDsLmD6kzaA/2bQYz7BWCVcTASt3cnBNjVNGEUyV692uPH5Kx06ZSXdfhy76dxWGDdnrWcBaA0eYwmH+LSEUwLkDG0OgRph6JI42nyDP/678Ojh1co7hEmOFfplHlUVKpfW3TtPada0fFAqqtTE59/O8LH/2OJfBqZJspfhZCMgUeL1JN/um7GcinrR2caM5izGzpsAHtqdbgOW0VdNvLr7M0EcoMPqueijH3mWAWGYDUOnPKZcw+et86ExTmw2fE7wm3NBZMRUrTMhzibrsXjI/NZVNY8d1DcnFrGPAlyTpWuqSRXTOV0YhClVbcsg8sXsnHsk2xy/yW6HNJ+tvGuWRwGCpUu2j2C22
X-OriginatorOrg: Nvidia.com
X-MS-Exchange-CrossTenant-OriginalArrivalTime: 02 Sep 2026 14:53:32.0789 (UTC)
X-MS-Exchange-CrossTenant-Network-Message-Id: 
 a842beb4-d78b-4bbb-7cca-08df0901f522
X-MS-Exchange-CrossTenant-Id: 43083d15-7273-40c1-b7db-39efd9ccc17a
X-MS-Exchange-CrossTenant-OriginalAttributedTenantConnectingIp: 
 TenantId=43083d15-7273-40c1-b7db-39efd9ccc17a; Ip=[216.228.117.161];
 Helo=[mail.nvidia.com]
X-MS-Exchange-CrossTenant-AuthSource: 
 SA2PEPF00001504.namprd04.prod.outlook.com
X-MS-Exchange-CrossTenant-AuthAs: Anonymous
X-MS-Exchange-CrossTenant-FromEntityHeader: HybridOnPrem
X-MS-Exchange-Transport-CrossTenantHeadersStamped: SN7PR12MB7809
X-Spam-Status: No, score=-1.2 required=5.0 tests=BAYES_00, DKIMWL_WL_HIGH,
 DKIM_SIGNED, DKIM_VALID, DKIM_VALID_AU, DKIM_VALID_EF, KAM_SHORT,
 LOCAL_AUTHENTICATION_FAIL_SPF, RCVD_IN_DNSWL_NONE, RCVD_IN_MSPIKE_H2,
 SPF_HELO_NONE, SPF_NONE,
 TXREP shortcircuit=no autolearn=no autolearn_force=no version=3.4.6
X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on sourceware.org
X-BeenThere: gcc-patches@gcc.gnu.org
X-Mailman-Version: 2.1.30
Precedence: list
List-Id: Gcc-patches mailing list <gcc-patches.gcc.gnu.org>
List-Unsubscribe: <https://gcc.gnu.org/mailman/options/gcc-patches>,
 <mailto:gcc-patches-request@gcc.gnu.org?subject=unsubscribe>
List-Archive: <https://gcc.gnu.org/pipermail/gcc-patches/>
List-Post: <mailto:gcc-patches@gcc.gnu.org>
List-Help: <mailto:gcc-patches-request@gcc.gnu.org?subject=help>
List-Subscribe: <https://gcc.gnu.org/mailman/listinfo/gcc-patches>,
 <mailto:gcc-patches-request@gcc.gnu.org?subject=subscribe>
Errors-To: gcc-patches-bounces~patchwork=sourceware.org@gcc.gnu.org

Series

Saturating arithmetic matching improvements |

Message

Kyrylo Tkachov Sept. 2, 2026, 2:52 p.m. UTC

  From: Kyrylo Tkachov <ktkachov@nvidia.com>

Hi all,

This series teaches GCC to recognise some more saturating arithmetic patterns,
primarily through match-sat-alu.pd and match.pd matchers.
These patterns were identified in a number of OSS workloads like video encoding
with the help of AI agents, though the patches themselves are manually written.

Bootstrapped and tested on aarch64-none-linux-gnu.
Ok for trunk?
Thanks,
Kyrill

Kyrylo Tkachov (13):
  match-sat-alu.pd: Allow wider ranges for signed saturating truncation
  match-sat-alu.pd: Recognize X - MIN (X, Y) as saturating subtraction
  match-sat-alu.pd: Recognize signed saturation clamps
  match-sat-alu.pd: Recognize borrow-test saturating subtraction
  match-sat-alu.pd: Recognize more unsigned saturating subtraction forms
  match-sat-alu.pd: Recognize complement tests for unsigned saturating
    addition
  match-sat-alu.pd: Recognize X + MIN (~X, Y) as unsigned saturating
    addition
  match-sat-alu.pd: Recognize strict widened unsigned saturation clamps
  match-sat-alu.pd: Recognize signed saturating truncation after a shift
  match-sat-alu.pd: Recognize unsigned constant-clamp truncation
  match-sat-alu.pd: Recognize signed constant-clamp truncation
  match-sat-alu.pd: Recognize complement masks for unsigned saturating
    truncation
  match.pd: Narrow promoted saturating addition and subtraction

 gcc/match-sat-alu.pd                          | 341 ++++++++++++++++--
 gcc/match.pd                                  |  37 ++
 gcc/testsuite/gcc.dg/sat-trunc-signed-cmp-2.c |  42 +++
 .../gcc.dg/vect/vect-sat-trunc-clamp-1.c      | 121 +++++++
 .../gcc.dg/vect/vect-sat-trunc-clamp-2.c      | 126 +++++++
 .../gcc.dg/vect/vect-sat-trunc-complement-1.c |  70 ++++
 .../vect/vect-sat-trunc-range-precision-1.c   | 122 +++++++
 .../vect/vect-sat-trunc-shift-sign-source-1.c | 221 ++++++++++++
 .../gcc.target/aarch64/sat_s_clamp-1.c        |  78 ++++
 .../gcc.target/aarch64/sat_s_clamp-2.c        |  43 +++
 .../gcc.target/aarch64/sat_s_val_ashr-1.c     |  30 ++
 .../gcc.target/aarch64/sat_u_add_min_not-1.c  |  18 +
 .../gcc.target/aarch64/sat_u_add_not_cmp-1.c  |  30 ++
 .../gcc.target/aarch64/sat_u_add_not_cmp-2.c  |  17 +
 .../gcc.target/aarch64/sat_u_add_widen_lt-1.c |  32 ++
 .../gcc.target/aarch64/sat_u_promoted-1.c     |  29 ++
 .../gcc.target/aarch64/sat_u_promoted-2.c     |  32 ++
 .../gcc.target/aarch64/sat_u_sub_borrow-1.c   |  22 ++
 .../gcc.target/aarch64/sat_u_sub_minmax-1.c   |  18 +
 .../gcc.target/aarch64/sat_u_sub_minmax-2.c   |  32 ++
 .../gcc.target/aarch64/sat_u_sub_minmax-3.c   |  27 ++
 .../gcc.target/aarch64/sat_u_sub_swapped-1.c  |  27 ++
 .../aarch64/saturating_arithmetic_1.c         |   2 +-
 .../aarch64/sve/saturating_arithmetic_1.c     |   2 +-
 .../aarch64/sve/saturating_arithmetic_3.c     |   2 +-
 .../aarch64/sve/saturating_arithmetic_4.c     |   2 +-
 .../vect-sat-trunc-clamp-inner-use-1.c        |  22 ++
 .../vect-sat-trunc-clamp-multi-use-1.c        |  21 ++
 .../aarch64/vect-sat-trunc-clamp-s-1.c        |   7 +
 .../vect-sat-trunc-clamp-s-negative-1.c       |  30 ++
 .../aarch64/vect-sat-trunc-clamp-u-1.c        |   7 +
 .../vect-sat-trunc-clamp-u-negative-1.c       |  30 ++
 .../aarch64/vect-sat-trunc-complement-1.c     |  30 ++
 .../vect-sat-trunc-range-precision-1.c        |   7 +
 ...ect-sat-trunc-range-precision-negative-1.c |  24 ++
 .../vect-sat-trunc-shift-sign-source-1.c      |   7 +
 ...t-sat-trunc-shift-sign-source-negative-1.c | 122 +++++++
 gcc/tree-ssa-math-opts.cc                     |  17 +-
 38 files changed, 1807 insertions(+), 40 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/sat-trunc-signed-cmp-2.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-clamp-1.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-clamp-2.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-complement-1.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-range-precision-1.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-shift-sign-source-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_s_clamp-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_s_clamp-2.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_s_val_ashr-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_add_min_not-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_add_not_cmp-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_add_not_cmp-2.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_add_widen_lt-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_promoted-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_promoted-2.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_borrow-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_minmax-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_minmax-2.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_minmax-3.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_swapped-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-inner-use-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-multi-use-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-s-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-s-negative-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-u-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-u-negative-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-complement-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-range-precision-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-range-precision-negative-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-shift-sign-source-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-shift-sign-source-negative-1.c

Comments

Kyrylo Tkachov Sept. 2, 2026, 2:52 p.m. UTC | #1

From: Kyrylo Tkachov <ktkachov@nvidia.com>

Hi all,

This series teaches GCC to recognise some more saturating arithmetic patterns,
primarily through match-sat-alu.pd and match.pd matchers.
These patterns were identified in a number of OSS workloads like video encoding
with the help of AI agents, though the patches themselves are manually written.

Bootstrapped and tested on aarch64-none-linux-gnu.
Ok for trunk?
Thanks,
Kyrill

Kyrylo Tkachov (13):
  match-sat-alu.pd: Allow wider ranges for signed saturating truncation
  match-sat-alu.pd: Recognize X - MIN (X, Y) as saturating subtraction
  match-sat-alu.pd: Recognize signed saturation clamps
  match-sat-alu.pd: Recognize borrow-test saturating subtraction
  match-sat-alu.pd: Recognize more unsigned saturating subtraction forms
  match-sat-alu.pd: Recognize complement tests for unsigned saturating
    addition
  match-sat-alu.pd: Recognize X + MIN (~X, Y) as unsigned saturating
    addition
  match-sat-alu.pd: Recognize strict widened unsigned saturation clamps
  match-sat-alu.pd: Recognize signed saturating truncation after a shift
  match-sat-alu.pd: Recognize unsigned constant-clamp truncation
  match-sat-alu.pd: Recognize signed constant-clamp truncation
  match-sat-alu.pd: Recognize complement masks for unsigned saturating
    truncation
  match.pd: Narrow promoted saturating addition and subtraction

 gcc/match-sat-alu.pd                          | 341 ++++++++++++++++--
 gcc/match.pd                                  |  37 ++
 gcc/testsuite/gcc.dg/sat-trunc-signed-cmp-2.c |  42 +++
 .../gcc.dg/vect/vect-sat-trunc-clamp-1.c      | 121 +++++++
 .../gcc.dg/vect/vect-sat-trunc-clamp-2.c      | 126 +++++++
 .../gcc.dg/vect/vect-sat-trunc-complement-1.c |  70 ++++
 .../vect/vect-sat-trunc-range-precision-1.c   | 122 +++++++
 .../vect/vect-sat-trunc-shift-sign-source-1.c | 221 ++++++++++++
 .../gcc.target/aarch64/sat_s_clamp-1.c        |  78 ++++
 .../gcc.target/aarch64/sat_s_clamp-2.c        |  43 +++
 .../gcc.target/aarch64/sat_s_val_ashr-1.c     |  30 ++
 .../gcc.target/aarch64/sat_u_add_min_not-1.c  |  18 +
 .../gcc.target/aarch64/sat_u_add_not_cmp-1.c  |  30 ++
 .../gcc.target/aarch64/sat_u_add_not_cmp-2.c  |  17 +
 .../gcc.target/aarch64/sat_u_add_widen_lt-1.c |  32 ++
 .../gcc.target/aarch64/sat_u_promoted-1.c     |  29 ++
 .../gcc.target/aarch64/sat_u_promoted-2.c     |  32 ++
 .../gcc.target/aarch64/sat_u_sub_borrow-1.c   |  22 ++
 .../gcc.target/aarch64/sat_u_sub_minmax-1.c   |  18 +
 .../gcc.target/aarch64/sat_u_sub_minmax-2.c   |  32 ++
 .../gcc.target/aarch64/sat_u_sub_minmax-3.c   |  27 ++
 .../gcc.target/aarch64/sat_u_sub_swapped-1.c  |  27 ++
 .../aarch64/saturating_arithmetic_1.c         |   2 +-
 .../aarch64/sve/saturating_arithmetic_1.c     |   2 +-
 .../aarch64/sve/saturating_arithmetic_3.c     |   2 +-
 .../aarch64/sve/saturating_arithmetic_4.c     |   2 +-
 .../vect-sat-trunc-clamp-inner-use-1.c        |  22 ++
 .../vect-sat-trunc-clamp-multi-use-1.c        |  21 ++
 .../aarch64/vect-sat-trunc-clamp-s-1.c        |   7 +
 .../vect-sat-trunc-clamp-s-negative-1.c       |  30 ++
 .../aarch64/vect-sat-trunc-clamp-u-1.c        |   7 +
 .../vect-sat-trunc-clamp-u-negative-1.c       |  30 ++
 .../aarch64/vect-sat-trunc-complement-1.c     |  30 ++
 .../vect-sat-trunc-range-precision-1.c        |   7 +
 ...ect-sat-trunc-range-precision-negative-1.c |  24 ++
 .../vect-sat-trunc-shift-sign-source-1.c      |   7 +
 ...t-sat-trunc-shift-sign-source-negative-1.c | 122 +++++++
 gcc/tree-ssa-math-opts.cc                     |  17 +-
 38 files changed, 1807 insertions(+), 40 deletions(-)
 create mode 100644 gcc/testsuite/gcc.dg/sat-trunc-signed-cmp-2.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-clamp-1.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-clamp-2.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-complement-1.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-range-precision-1.c
 create mode 100644 gcc/testsuite/gcc.dg/vect/vect-sat-trunc-shift-sign-source-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_s_clamp-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_s_clamp-2.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_s_val_ashr-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_add_min_not-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_add_not_cmp-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_add_not_cmp-2.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_add_widen_lt-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_promoted-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_promoted-2.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_borrow-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_minmax-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_minmax-2.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_minmax-3.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sat_u_sub_swapped-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-inner-use-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-multi-use-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-s-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-s-negative-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-u-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-clamp-u-negative-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-complement-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-range-precision-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-range-precision-negative-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-shift-sign-source-1.c
 create mode 100644 gcc/testsuite/gcc.target/aarch64/vect-sat-trunc-shift-sign-source-negative-1.c