Recommit [AArch64][SVE]Use FeatureUseFixedOverScalableIfEqualCost for A510/A520 (#134606)

Recommit. This work was done by #132246 but failed buildbots due to the
test introduced needing updates

Inefficient SVE codegen occurs on at least two in-order cores, those
being Cortex-A510 and Cortex-A520. For example a simple vector add

```
void foo(float a, float b, float dst, unsigned n) {
    for (unsigned i = 0; i < n; ++i)
        dst[i] = a[i] + b[i];
}
```

Vectorizes the inner loop into the following interleaved sequence of
instructions.

```
        add     x12, x1, x10
        ld1b    { z0.b }, p0/z, [x1, x10]
        add     x13, x2, x10
        ld1b    { z1.b }, p0/z, [x2, x10]
        ldr     z2, [x12, #1, mul vl]
        ldr     z3, [x13, #1, mul vl]
        dech    x11
        add     x12, x0, x10
        fadd    z0.s, z1.s, z0.s
        fadd    z1.s, z3.s, z2.s
        st1b    { z0.b }, p0, [x0, x10]
        addvl   x10, x10, #2
        str     z1, [x12, #1, mul vl]
```

By adjusting the target features to prefer fixed over scalable if the
cost is equal we get the following vectorized loop.

```
         ldp q0, q3, [x11, #-16]
         subs    x13, x13, #8
         ldp q1, q2, [x10, #-16]
         add x10, x10, #32
         add x11, x11, #32
         fadd    v0.4s, v1.4s, v0.4s
         fadd    v1.4s, v2.4s, v3.4s
         stp q0, q1, [x12, #-16]
         add x12, x12, #32
```

Which is more efficient.

Author: nasherm

llvm/lib/Target/AArch64/AArch64Processors.td

@@ -723,6 +723,7 @@ def ProcessorFeatures {
                                  FeatureSB, FeaturePAuth, FeatureSSBS, FeatureSVE, FeatureSVE2,
                                  FeatureComplxNum, FeatureCRC, FeatureDotProd,
                                  FeatureFPARMv8,FeatureFullFP16, FeatureJS, FeatureLSE,
+                                 FeatureUseFixedOverScalableIfEqualCost,
                                  FeatureRAS, FeatureRCPC, FeatureRDM];
   list<SubtargetFeature> A520 = [HasV9_2aOps, FeaturePerfMon, FeatureAM,
                                  FeatureMTE, FeatureETE, FeatureSVEBitPerm,
@@ -732,6 +733,7 @@ def ProcessorFeatures {
                                  FeatureSVE, FeatureSVE2, FeatureBF16, FeatureComplxNum, FeatureCRC,
                                  FeatureFPARMv8, FeatureFullFP16, FeatureMatMulInt8, FeatureJS,
                                  FeatureNEON, FeatureLSE, FeatureRAS, FeatureRCPC, FeatureRDM,
+                                 FeatureUseFixedOverScalableIfEqualCost,
                                  FeatureDotProd];
   list<SubtargetFeature> A520AE = [HasV9_2aOps, FeaturePerfMon, FeatureAM,
                                  FeatureMTE, FeatureETE, FeatureSVEBitPerm,

llvm/test/Transforms/LoopVectorize/AArch64/sve-fixed-width-inorder-core.ll

@@ -0,0 +1,168 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -mtriple=aarch64-none-elf -mcpu=cortex-a510 -mattr=+sve -passes=loop-vectorize -S | FileCheck %s --check-prefix=CHECK-CA510
+; RUN: opt < %s -mtriple=aarch64-none-elf -mcpu=cortex-a520 -mattr=+sve -passes=loop-vectorize -S | FileCheck %s --check-prefix=CHECK-CA520
+
+define void @sve_add(ptr  %dst, ptr  %a, ptr  %b, i64 %n) {
+; CHECK-CA510-LABEL: define void @sve_add(
+; CHECK-CA510-SAME: ptr [[DST:%.*]], ptr [[A:%.*]], ptr [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-CA510-NEXT:  [[ENTRY:.*:]]
+; CHECK-CA510-NEXT:    [[B3:%.*]] = ptrtoint ptr [[B]] to i64
+; CHECK-CA510-NEXT:    [[A2:%.*]] = ptrtoint ptr [[A]] to i64
+; CHECK-CA510-NEXT:    [[DST1:%.*]] = ptrtoint ptr [[DST]] to i64
+; CHECK-CA510-NEXT:    [[CMP9_NOT:%.*]] = icmp eq i64 [[N]], 0
+; CHECK-CA510-NEXT:    br i1 [[CMP9_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK-CA510:       [[FOR_BODY_PREHEADER]]:
+; CHECK-CA510-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 8
+; CHECK-CA510-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]]
+; CHECK-CA510:       [[VECTOR_MEMCHECK]]:
+; CHECK-CA510-NEXT:    [[TMP0:%.*]] = sub i64 [[DST1]], [[A2]]
+; CHECK-CA510-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 32
+; CHECK-CA510-NEXT:    [[TMP1:%.*]] = sub i64 [[DST1]], [[B3]]
+; CHECK-CA510-NEXT:    [[DIFF_CHECK4:%.*]] = icmp ult i64 [[TMP1]], 32
+; CHECK-CA510-NEXT:    [[CONFLICT_RDX:%.*]] = or i1 [[DIFF_CHECK]], [[DIFF_CHECK4]]
+; CHECK-CA510-NEXT:    br i1 [[CONFLICT_RDX]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
+; CHECK-CA510:       [[VECTOR_PH]]:
+; CHECK-CA510-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 8
+; CHECK-CA510-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
+; CHECK-CA510-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK-CA510:       [[VECTOR_BODY]]:
+; CHECK-CA510-NEXT:    [[TMP2:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-CA510-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[TMP2]]
+; CHECK-CA510-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw float, ptr [[TMP3]], i32 0
+; CHECK-CA510-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw float, ptr [[TMP3]], i32 4
+; CHECK-CA510-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP4]], align 4
+; CHECK-CA510-NEXT:    [[WIDE_LOAD5:%.*]] = load <4 x float>, ptr [[TMP5]], align 4
+; CHECK-CA510-NEXT:    [[TMP6:%.*]] = getelementptr inbounds nuw float, ptr [[B]], i64 [[TMP2]]
+; CHECK-CA510-NEXT:    [[TMP7:%.*]] = getelementptr inbounds nuw float, ptr [[TMP6]], i32 0
+; CHECK-CA510-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw float, ptr [[TMP6]], i32 4
+; CHECK-CA510-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x float>, ptr [[TMP7]], align 4
+; CHECK-CA510-NEXT:    [[WIDE_LOAD7:%.*]] = load <4 x float>, ptr [[TMP8]], align 4
+; CHECK-CA510-NEXT:    [[TMP9:%.*]] = fadd fast <4 x float> [[WIDE_LOAD6]], [[WIDE_LOAD]]
+; CHECK-CA510-NEXT:    [[TMP10:%.*]] = fadd fast <4 x float> [[WIDE_LOAD7]], [[WIDE_LOAD5]]
+; CHECK-CA510-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw float, ptr [[DST]], i64 [[TMP2]]
+; CHECK-CA510-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw float, ptr [[TMP11]], i32 0
+; CHECK-CA510-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw float, ptr [[TMP11]], i32 4
+; CHECK-CA510-NEXT:    store <4 x float> [[TMP9]], ptr [[TMP12]], align 4
+; CHECK-CA510-NEXT:    store <4 x float> [[TMP10]], ptr [[TMP13]], align 4
+; CHECK-CA510-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[TMP2]], 8
+; CHECK-CA510-NEXT:    [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-CA510-NEXT:    br i1 [[TMP14]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-CA510:       [[MIDDLE_BLOCK]]:
+; CHECK-CA510-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
+; CHECK-CA510-NEXT:    br i1 [[CMP_N]], label %[[FOR_COND_CLEANUP_LOOPEXIT:.*]], label %[[SCALAR_PH]]
+; CHECK-CA510:       [[SCALAR_PH]]:
+; CHECK-CA510-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[FOR_BODY_PREHEADER]] ], [ 0, %[[VECTOR_MEMCHECK]] ]
+; CHECK-CA510-NEXT:    br label %[[FOR_BODY:.*]]
+; CHECK-CA510:       [[FOR_BODY]]:
+; CHECK-CA510-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
+; CHECK-CA510-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV]]
+; CHECK-CA510-NEXT:    [[TMP15:%.*]] = load float, ptr [[ARRAYIDX]], align 4
+; CHECK-CA510-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds nuw float, ptr [[B]], i64 [[INDVARS_IV]]
+; CHECK-CA510-NEXT:    [[TMP16:%.*]] = load float, ptr [[ARRAYIDX2]], align 4
+; CHECK-CA510-NEXT:    [[ADD:%.*]] = fadd fast float [[TMP16]], [[TMP15]]
+; CHECK-CA510-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds nuw float, ptr [[DST]], i64 [[INDVARS_IV]]
+; CHECK-CA510-NEXT:    store float [[ADD]], ptr [[ARRAYIDX4]], align 4
+; CHECK-CA510-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-CA510-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[N]]
+; CHECK-CA510-NEXT:    br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP_LOOPEXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK-CA510:       [[FOR_COND_CLEANUP_LOOPEXIT]]:
+; CHECK-CA510-NEXT:    br label %[[FOR_COND_CLEANUP]]
+; CHECK-CA510:       [[FOR_COND_CLEANUP]]:
+; CHECK-CA510-NEXT:    ret void
+;
+; CHECK-CA520-LABEL: define void @sve_add(
+; CHECK-CA520-SAME: ptr [[DST:%.*]], ptr [[A:%.*]], ptr [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-CA520-NEXT:  [[ENTRY:.*:]]
+; CHECK-CA520-NEXT:    [[B3:%.*]] = ptrtoint ptr [[B]] to i64
+; CHECK-CA520-NEXT:    [[A2:%.*]] = ptrtoint ptr [[A]] to i64
+; CHECK-CA520-NEXT:    [[DST1:%.*]] = ptrtoint ptr [[DST]] to i64
+; CHECK-CA520-NEXT:    [[CMP9_NOT:%.*]] = icmp eq i64 [[N]], 0
+; CHECK-CA520-NEXT:    br i1 [[CMP9_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK-CA520:       [[FOR_BODY_PREHEADER]]:
+; CHECK-CA520-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 8
+; CHECK-CA520-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]]
+; CHECK-CA520:       [[VECTOR_MEMCHECK]]:
+; CHECK-CA520-NEXT:    [[TMP0:%.*]] = sub i64 [[DST1]], [[A2]]
+; CHECK-CA520-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP0]], 32
+; CHECK-CA520-NEXT:    [[TMP1:%.*]] = sub i64 [[DST1]], [[B3]]
+; CHECK-CA520-NEXT:    [[DIFF_CHECK4:%.*]] = icmp ult i64 [[TMP1]], 32
+; CHECK-CA520-NEXT:    [[CONFLICT_RDX:%.*]] = or i1 [[DIFF_CHECK]], [[DIFF_CHECK4]]
+; CHECK-CA520-NEXT:    br i1 [[CONFLICT_RDX]], label %[[SCALAR_PH]], label %[[VECTOR_PH:.*]]
+; CHECK-CA520:       [[VECTOR_PH]]:
+; CHECK-CA520-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], 8
+; CHECK-CA520-NEXT:    [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
+; CHECK-CA520-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK-CA520:       [[VECTOR_BODY]]:
+; CHECK-CA520-NEXT:    [[TMP2:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-CA520-NEXT:    [[TMP3:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[TMP2]]
+; CHECK-CA520-NEXT:    [[TMP4:%.*]] = getelementptr inbounds nuw float, ptr [[TMP3]], i32 0
+; CHECK-CA520-NEXT:    [[TMP5:%.*]] = getelementptr inbounds nuw float, ptr [[TMP3]], i32 4
+; CHECK-CA520-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP4]], align 4
+; CHECK-CA520-NEXT:    [[WIDE_LOAD5:%.*]] = load <4 x float>, ptr [[TMP5]], align 4
+; CHECK-CA520-NEXT:    [[TMP6:%.*]] = getelementptr inbounds nuw float, ptr [[B]], i64 [[TMP2]]
+; CHECK-CA520-NEXT:    [[TMP7:%.*]] = getelementptr inbounds nuw float, ptr [[TMP6]], i32 0
+; CHECK-CA520-NEXT:    [[TMP8:%.*]] = getelementptr inbounds nuw float, ptr [[TMP6]], i32 4
+; CHECK-CA520-NEXT:    [[WIDE_LOAD6:%.*]] = load <4 x float>, ptr [[TMP7]], align 4
+; CHECK-CA520-NEXT:    [[WIDE_LOAD7:%.*]] = load <4 x float>, ptr [[TMP8]], align 4
+; CHECK-CA520-NEXT:    [[TMP9:%.*]] = fadd fast <4 x float> [[WIDE_LOAD6]], [[WIDE_LOAD]]
+; CHECK-CA520-NEXT:    [[TMP10:%.*]] = fadd fast <4 x float> [[WIDE_LOAD7]], [[WIDE_LOAD5]]
+; CHECK-CA520-NEXT:    [[TMP11:%.*]] = getelementptr inbounds nuw float, ptr [[DST]], i64 [[TMP2]]
+; CHECK-CA520-NEXT:    [[TMP12:%.*]] = getelementptr inbounds nuw float, ptr [[TMP11]], i32 0
+; CHECK-CA520-NEXT:    [[TMP13:%.*]] = getelementptr inbounds nuw float, ptr [[TMP11]], i32 4
+; CHECK-CA520-NEXT:    store <4 x float> [[TMP9]], ptr [[TMP12]], align 4
+; CHECK-CA520-NEXT:    store <4 x float> [[TMP10]], ptr [[TMP13]], align 4
+; CHECK-CA520-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[TMP2]], 8
+; CHECK-CA520-NEXT:    [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-CA520-NEXT:    br i1 [[TMP14]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK-CA520:       [[MIDDLE_BLOCK]]:
+; CHECK-CA520-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
+; CHECK-CA520-NEXT:    br i1 [[CMP_N]], label %[[FOR_COND_CLEANUP_LOOPEXIT:.*]], label %[[SCALAR_PH]]
+; CHECK-CA520:       [[SCALAR_PH]]:
+; CHECK-CA520-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[MIDDLE_BLOCK]] ], [ 0, %[[FOR_BODY_PREHEADER]] ], [ 0, %[[VECTOR_MEMCHECK]] ]
+; CHECK-CA520-NEXT:    br label %[[FOR_BODY:.*]]
+; CHECK-CA520:       [[FOR_BODY]]:
+; CHECK-CA520-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[FOR_BODY]] ], [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ]
+; CHECK-CA520-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds nuw float, ptr [[A]], i64 [[INDVARS_IV]]
+; CHECK-CA520-NEXT:    [[TMP15:%.*]] = load float, ptr [[ARRAYIDX]], align 4
+; CHECK-CA520-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds nuw float, ptr [[B]], i64 [[INDVARS_IV]]
+; CHECK-CA520-NEXT:    [[TMP16:%.*]] = load float, ptr [[ARRAYIDX2]], align 4
+; CHECK-CA520-NEXT:    [[ADD:%.*]] = fadd fast float [[TMP16]], [[TMP15]]
+; CHECK-CA520-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr inbounds nuw float, ptr [[DST]], i64 [[INDVARS_IV]]
+; CHECK-CA520-NEXT:    store float [[ADD]], ptr [[ARRAYIDX4]], align 4
+; CHECK-CA520-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-CA520-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[N]]
+; CHECK-CA520-NEXT:    br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP_LOOPEXIT]], label %[[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK-CA520:       [[FOR_COND_CLEANUP_LOOPEXIT]]:
+; CHECK-CA520-NEXT:    br label %[[FOR_COND_CLEANUP]]
+; CHECK-CA520:       [[FOR_COND_CLEANUP]]:
+; CHECK-CA520-NEXT:    ret void
+;
+entry:
+  %cmp9.not = icmp eq i64 %n, 0
+  br i1 %cmp9.not, label %for.cond.cleanup, label %for.body
+for.body:                                         ; preds = %for.body.preheader, %for.body
+  %indvars.iv = phi i64 [ 0, %entry], [ %indvars.iv.next, %for.body ]
+  %arrayidx = getelementptr inbounds nuw float, ptr %a, i64 %indvars.iv
+  %0 = load float, ptr %arrayidx, align 4
+  %arrayidx2 = getelementptr inbounds nuw float, ptr %b, i64 %indvars.iv
+  %1 = load float, ptr %arrayidx2, align 4
+  %add = fadd fast float %1, %0
+  %arrayidx4 = getelementptr inbounds nuw float, ptr %dst, i64 %indvars.iv
+  store float %add, ptr %arrayidx4, align 4
+  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
+  %exitcond.not = icmp eq i64 %indvars.iv.next, %n
+  br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+for.cond.cleanup:                                 ; preds = %for.cond.cleanup.loopexit, %entry
+  ret void
+}
+;.
+; CHECK-CA510: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
+; CHECK-CA510: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
+; CHECK-CA510: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
+; CHECK-CA510: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]}
+;.
+; CHECK-CA520: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
+; CHECK-CA520: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
+; CHECK-CA520: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
+; CHECK-CA520: [[LOOP3]] = distinct !{[[LOOP3]], [[META1]]}
+;.