[llvm] [AArch64] Disconnect FeatureUseScalarIncVL from FeatureSVE2. (PR #114568)

David Green via llvm-commits llvm-commits at lists.llvm.org
Fri Nov 1 09:39:50 PDT 2024 

https://github.com/davemgreen created https://github.com/llvm/llvm-project/pull/114568

FeatureUseScalarIncVL is a tuning feature, used to control whether addvl or add+cnt is used. It was previously added as a dependency for FeatureSVE2, an architecture feature but this can be seen as a layering violation. The main disadvantage is that -use-scalar-inc-vl cannot be used without disabling sve2 and all dependant features.

This patch attempt to break that, enabling the feature if SVE2 or SME is present and the use-scalar-inc-vl is not otherwise specified, but not requiring a hard dependency between the two. (Unfortunately I had to check if the value was present in the features string to check that).

It also inverts the meaning of the feature to DontUseScalarIncVL, so that the default state is to use addvl, hopefully the more sensible default going forward.

>From 3f55cc6f0d428dce351e22da8be6e1fbfafdecc9 Mon Sep 17 00:00:00 2001
From: David Green <david.green at arm.com>
Date: Fri, 1 Nov 2024 15:53:52 +0000
Subject: [PATCH] [AArch64] Disconnect FeatureUseScalarIncVL from FeatureSVE2.

FeatureUseScalarIncVL is a tuning feature, used to control whether addvl or
add+cnt is used. It was previously added as a dependency for FeatureSVE2, an
architecture feature but this can be seen as a layering violation. The main
disadvantage is that -use-scalar-inc-vl cannot be used without disabling sve2
and all dependant features.

This patch attempt to break that, enabling the feature if SVE2 or SME is
present and the use-scalar-inc-vl is not otherwise specified, but not requiring
a hard dependency between the two. (Unfortunately I had to check if the value
was present in the features string to check that).

It also inverts the meaning of the feature to DontUseScalarIncVL, so that the
default state is to use addvl, hopefully the more sensible default going
forward.
---
 llvm/lib/Target/AArch64/AArch64Features.td             | 10 +++++-----
 llvm/lib/Target/AArch64/AArch64InstrInfo.td            |  4 ++--
 llvm/lib/Target/AArch64/AArch64Subtarget.cpp           |  7 +++++++
 .../AArch64/sve-intrinsics-counting-elems-i32.ll       |  2 +-
 .../CodeGen/AArch64/sve-intrinsics-counting-elems.ll   |  4 +++-
 llvm/test/CodeGen/AArch64/sve-vl-arith.ll              |  3 ++-
 6 files changed, 20 insertions(+), 10 deletions(-)

diff --git a/llvm/lib/Target/AArch64/AArch64Features.td b/llvm/lib/Target/AArch64/AArch64Features.td
index 642976cd3ea076..29da5e700854b5 100644
--- a/llvm/lib/Target/AArch64/AArch64Features.td
+++ b/llvm/lib/Target/AArch64/AArch64Features.td
@@ -358,12 +358,9 @@ def FeatureTHE : ExtensionWithMArch<"the", "THE", "FEAT_THE",
 //  Armv9.0 Architecture Extensions
 //===----------------------------------------------------------------------===//
 
-def FeatureUseScalarIncVL : SubtargetFeature<"use-scalar-inc-vl",
-  "UseScalarIncVL", "true", "Prefer inc/dec over add+cnt">;
-
 def FeatureSVE2 : ExtensionWithMArch<"sve2", "SVE2", "FEAT_SVE2",
   "Enable Scalable Vector Extension 2 (SVE2) instructions",
-  [FeatureSVE, FeatureUseScalarIncVL]>;
+  [FeatureSVE]>;
 
 def FeatureSVE2AES : ExtensionWithMArch<"sve2-aes", "SVE2AES",
   "FEAT_SVE_AES, FEAT_SVE_PMULL128",
@@ -403,7 +400,7 @@ def FeatureRME : Extension<"rme", "RME", "FEAT_RME",
   "Enable Realm Management Extension">;
 
 def FeatureSME : ExtensionWithMArch<"sme", "SME", "FEAT_SME",
-  "Enable Scalable Matrix Extension (SME)", [FeatureBF16, FeatureUseScalarIncVL]>;
+  "Enable Scalable Matrix Extension (SME)", [FeatureBF16]>;
 
 def FeatureSMEF64F64 : ExtensionWithMArch<"sme-f64f64", "SMEF64F64", "FEAT_SME_F64F64",
   "Enable Scalable Matrix Extension (SME) F64F64 instructions", [FeatureSME]>;
@@ -797,6 +794,9 @@ def FeatureUseFixedOverScalableIfEqualCost: SubtargetFeature<"use-fixed-over-sca
   "UseFixedOverScalableIfEqualCost", "true",
   "Prefer fixed width loop vectorization over scalable if the cost-model assigns equal costs">;
 
+def FeatureDontUseScalarIncVL : SubtargetFeature<"dont-use-scalar-inc-vl",
+  "DontUseScalarIncVL", "true", "Prefer add+cnt over inc/dec">;
+
 //===----------------------------------------------------------------------===//
 // Architectures.
 //
diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.td b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
index 424848252f6aa6..ff11224842e7bf 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.td
@@ -375,9 +375,9 @@ def UseNegativeImmediates
     : Predicate<"false">, AssemblerPredicate<(all_of (not FeatureNoNegativeImmediates)),
                                              "NegativeImmediates">;
 
-def UseScalarIncVL : Predicate<"Subtarget->useScalarIncVL()">;
+def UseScalarIncVL : Predicate<"!Subtarget->dontUseScalarIncVL()">;
 
-def NoUseScalarIncVL : Predicate<"!Subtarget->useScalarIncVL()">;
+def NoUseScalarIncVL : Predicate<"Subtarget->dontUseScalarIncVL()">;
 
 def UseSVEFPLD1R : Predicate<"!Subtarget->noSVEFPLD1R()">;
 
diff --git a/llvm/lib/Target/AArch64/AArch64Subtarget.cpp b/llvm/lib/Target/AArch64/AArch64Subtarget.cpp
index 736d57e6ae2fd9..09dff6265cbacb 100644
--- a/llvm/lib/Target/AArch64/AArch64Subtarget.cpp
+++ b/llvm/lib/Target/AArch64/AArch64Subtarget.cpp
@@ -335,6 +335,13 @@ void AArch64Subtarget::initializeProperties(bool HasMinSize) {
 
   if (AArch64MinimumJumpTableEntries.getNumOccurrences() > 0 || !HasMinSize)
     MinimumJumpTableEntries = AArch64MinimumJumpTableEntries;
+
+  // If only SVE is present (not SVE2 or SME) and DontUseScalarIncVL is not
+  // otherwise set, enable it by default.
+  if (!hasSVE2() && !hasSME()) {
+    if (!getFeatureString().contains("dont-use-scalar-inc-vl"))
+      DontUseScalarIncVL = true;
+  }
 }
 
 AArch64Subtarget::AArch64Subtarget(const Triple &TT, StringRef CPU,
diff --git a/llvm/test/CodeGen/AArch64/sve-intrinsics-counting-elems-i32.ll b/llvm/test/CodeGen/AArch64/sve-intrinsics-counting-elems-i32.ll
index 5062a43da931f8..1855d3673f92c7 100644
--- a/llvm/test/CodeGen/AArch64/sve-intrinsics-counting-elems-i32.ll
+++ b/llvm/test/CodeGen/AArch64/sve-intrinsics-counting-elems-i32.ll
@@ -1,6 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s -check-prefix=NO_SCALAR_INC
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+use-scalar-inc-vl < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=-dont-use-scalar-inc-vl < %s | FileCheck %s
 ; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2 < %s | FileCheck %s
 
 ; INCB
diff --git a/llvm/test/CodeGen/AArch64/sve-intrinsics-counting-elems.ll b/llvm/test/CodeGen/AArch64/sve-intrinsics-counting-elems.ll
index 8f0a9eac87c27c..bf812823e86a99 100644
--- a/llvm/test/CodeGen/AArch64/sve-intrinsics-counting-elems.ll
+++ b/llvm/test/CodeGen/AArch64/sve-intrinsics-counting-elems.ll
@@ -1,8 +1,10 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve < %s | FileCheck %s
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+use-scalar-inc-vl < %s | FileCheck %s -check-prefix=USE_SCALAR_INC
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=-dont-use-scalar-inc-vl < %s | FileCheck %s -check-prefix=USE_SCALAR_INC
 ; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2 < %s | FileCheck %s -check-prefix=USE_SCALAR_INC
 ; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme -force-streaming < %s | FileCheck %s -check-prefix=USE_SCALAR_INC
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2 -mattr=+dont-use-scalar-inc-vl < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sme -mattr=+dont-use-scalar-inc-vl -force-streaming < %s | FileCheck %s
 
 ;
 ; CNTB
diff --git a/llvm/test/CodeGen/AArch64/sve-vl-arith.ll b/llvm/test/CodeGen/AArch64/sve-vl-arith.ll
index de2af590acd1e2..4ddb2fdf56787b 100644
--- a/llvm/test/CodeGen/AArch64/sve-vl-arith.ll
+++ b/llvm/test/CodeGen/AArch64/sve-vl-arith.ll
@@ -1,7 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 2
 ; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -verify-machineinstrs < %s | FileCheck %s -check-prefix=NO_SCALAR_INC
-; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=+use-scalar-inc-vl -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve -mattr=-dont-use-scalar-inc-vl -verify-machineinstrs < %s | FileCheck %s
 ; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2 -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2 -mattr=+dont-use-scalar-inc-vl -verify-machineinstrs < %s | FileCheck %s -check-prefix=NO_SCALAR_INC
 
 define <vscale x 8 x i16> @inch_vec(<vscale x 8 x i16> %a) {
 ; NO_SCALAR_INC-LABEL: inch_vec:

More information about the llvm-commits mailing list