[AArch64] Enable UseFixedOverScalableIfEqualCost for more Cortex-x cpus. (#122807)

For similar reasons for fixed-width being prefered to scalable for
Neoverse V2, this patch enables the UseFixedOverScalableIfEqualCost
feature when using -mcpu=cortex-x2, x3, x4 and x925 that are similar to
Neoverse V2.

Author: davemgreen

llvm/lib/Target/AArch64/AArch64Processors.td

@@ -229,6 +229,7 @@ def TuneX2 : SubtargetFeature<"cortex-x2", "ARMProcFamily", "CortexX2",
                                   FeatureALULSLFast,
                                   FeaturePostRAScheduler,
                                   FeatureEnableSelectOptimize,
+                                  FeatureUseFixedOverScalableIfEqualCost,
                                   FeaturePredictableSelectIsExpensive]>;
 
 def TuneX3 : SubtargetFeature<"cortex-x3", "ARMProcFamily", "CortexX3",
@@ -238,6 +239,7 @@ def TuneX3 : SubtargetFeature<"cortex-x3", "ARMProcFamily", "CortexX3",
                                FeatureFuseAES,
                                FeaturePostRAScheduler,
                                FeatureEnableSelectOptimize,
+                               FeatureUseFixedOverScalableIfEqualCost,
                                FeaturePredictableSelectIsExpensive]>;
 
 def TuneX4 : SubtargetFeature<"cortex-x4", "ARMProcFamily", "CortexX4",
@@ -247,6 +249,7 @@ def TuneX4 : SubtargetFeature<"cortex-x4", "ARMProcFamily", "CortexX4",
                                FeatureFuseAES,
                                FeaturePostRAScheduler,
                                FeatureEnableSelectOptimize,
+                               FeatureUseFixedOverScalableIfEqualCost,
                                FeaturePredictableSelectIsExpensive]>;
 
 def TuneX925 : SubtargetFeature<"cortex-x925", "ARMProcFamily",
@@ -256,6 +259,7 @@ def TuneX925 : SubtargetFeature<"cortex-x925", "ARMProcFamily",
                                 FeatureFuseAES,
                                 FeaturePostRAScheduler,
                                 FeatureEnableSelectOptimize,
+                                FeatureUseFixedOverScalableIfEqualCost,
                                 FeaturePredictableSelectIsExpensive]>;
 
 def TuneA64FX : SubtargetFeature<"a64fx", "ARMProcFamily", "A64FX",

llvm/test/Transforms/LoopVectorize/AArch64/prefer-fixed-if-equal-to-scalable.ll

@@ -1,14 +1,21 @@
-; RUN: opt -S < %s -passes=loop-vectorize -force-target-instruction-cost=1 | FileCheck %s
+; RUN: opt -S < %s -passes=loop-vectorize -mcpu=generic -mattr=+sve2 | FileCheck %s --check-prefix=CHECK-GENERIC
+; RUN: opt -S < %s -passes=loop-vectorize -mcpu=neoverse-v2 | FileCheck %s --check-prefix=CHECK-PREFERFIXED
+; RUN: opt -S < %s -passes=loop-vectorize -mcpu=cortex-x2 | FileCheck %s --check-prefix=CHECK-PREFERFIXED
+; RUN: opt -S < %s -passes=loop-vectorize -mcpu=cortex-x3 | FileCheck %s --check-prefix=CHECK-PREFERFIXED
+; RUN: opt -S < %s -passes=loop-vectorize -mcpu=cortex-x4 | FileCheck %s --check-prefix=CHECK-PREFERFIXED
+; RUN: opt -S < %s -passes=loop-vectorize -mcpu=cortex-x925 | FileCheck %s --check-prefix=CHECK-PREFERFIXED
 
 target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32"
 target triple = "aarch64-unknown-linux-gnu"
 
 @a = dso_local local_unnamed_addr global [32000 x float] zeroinitializer, align 64
 @b = dso_local local_unnamed_addr global [32000 x float] zeroinitializer, align 64
 
-define void @NeoverseV2() #0 {
-; CHECK-LABEL: define void @NeoverseV2(
-; CHECK:       store <4 x float>
+define void @test() #0 {
+; CHECK-GENERIC-LABEL: define void @test(
+; CHECK-GENERIC:       store <vscale x 4 x float>
+; CHECK-PREFERFIXED-LABEL: define void @test(
+; CHECK-PREFERFIXED:       store <4 x float>
 ;
 entry:
   br label %for.body
@@ -31,30 +38,4 @@ for.body:
   br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
 }
 
-define void @GenericCPU() #1 {
-; CHECK-LABEL: define void @GenericCPU(
-; CHECK:       store <vscale x 4 x float>
-;
-entry:
-  br label %for.body
-
-for.cond.cleanup:
-  ret void
-
-for.body:
-  %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
-  %arrayidx = getelementptr inbounds [32000 x float], ptr @a, i64 0, i64 %indvars.iv
-  %0 = load float, ptr %arrayidx, align 4
-  %arrayidx2 = getelementptr inbounds [32000 x float], ptr @b, i64 0, i64 %indvars.iv
-  %1 = load float, ptr %arrayidx2, align 4
-  %add = fadd fast float %1, %0
-  %2 = add nuw nsw i64 %indvars.iv, 16000
-  %arrayidx5 = getelementptr inbounds [32000 x float], ptr @a, i64 0, i64 %2
-  store float %add, ptr %arrayidx5, align 4
-  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
-  %exitcond.not = icmp eq i64 %indvars.iv.next, 16000
-  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
-}
-
-attributes #0 = { vscale_range(1,16) "target-cpu"="neoverse-v2" "target-features"="+sve,+sve2,+v9a" }
-attributes #1 = { vscale_range(1,16) "target-cpu"="generic" "target-features"="+sve,+v9a" }
+attributes #0 = { vscale_range(1,16) }