[SLP]Allow interleaving check only if it is less than number of elements

Need to check if the interleaving factor is less than total number of
elements in loads slice to handle it correctly and avoid compiler crash.

Fixes report #112361 (comment)

Author: alexey-bataev

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

@@ -7046,7 +7046,8 @@ void BoUpSLP::tryToVectorizeGatheredLoads(
                 OrdersType Order;
                 SmallVector<Value *> PointerOps;
                 // Segmented load detected - vectorize at maximum vector factor.
-                if (TTI.isLegalInterleavedAccessType(
+                if (InterleaveFactor <= Slice.size() &&
+                    TTI.isLegalInterleavedAccessType(
                         getWidenedType(Slice.front()->getType(), VF),
                         InterleaveFactor,
                         cast<LoadInst>(Slice.front())->getAlign(),

llvm/test/Transforms/SLPVectorizer/RISCV/interleave-greater-than-slice.ll

@@ -0,0 +1,74 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -S --passes=slp-vectorizer -mtriple=riscv64-unknown-linux -mattr=+v,+zvl128b < %s | FileCheck %s
+
+define void @test(ptr %a, float %0) {
+; CHECK-LABEL: define void @test(
+; CHECK-SAME: ptr [[A:%.*]], float [[TMP0:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT:  [[ENTRY:.*:]]
+; CHECK-NEXT:    [[TMP1:%.*]] = load ptr, ptr [[A]], align 8
+; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr i8, ptr [[TMP1]], i64 84
+; CHECK-NEXT:    [[TMP2:%.*]] = load float, ptr [[ARRAYIDX]], align 4
+; CHECK-NEXT:    [[TMP3:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP2]], float 0.000000e+00, float 0.000000e+00)
+; CHECK-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr i8, ptr [[TMP1]], i64 28
+; CHECK-NEXT:    [[TMP4:%.*]] = load float, ptr [[ARRAYIDX1]], align 4
+; CHECK-NEXT:    [[TMP5:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP4]], float 0.000000e+00, float [[TMP3]])
+; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr i8, ptr [[TMP1]], i64 8
+; CHECK-NEXT:    [[TMP6:%.*]] = load float, ptr [[ARRAYIDX2]], align 4
+; CHECK-NEXT:    [[TMP7:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP6]], float 0.000000e+00, float 0.000000e+00)
+; CHECK-NEXT:    [[ARRAYIDX3:%.*]] = getelementptr i8, ptr [[TMP1]], i64 68
+; CHECK-NEXT:    [[TMP8:%.*]] = load float, ptr [[ARRAYIDX3]], align 4
+; CHECK-NEXT:    [[TMP9:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP8]], float 0.000000e+00, float [[TMP5]])
+; CHECK-NEXT:    [[ARRAYIDX4:%.*]] = getelementptr i8, ptr [[TMP1]], i64 88
+; CHECK-NEXT:    [[TMP10:%.*]] = load float, ptr [[ARRAYIDX4]], align 4
+; CHECK-NEXT:    [[TMP11:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP10]], float 0.000000e+00, float [[TMP7]])
+; CHECK-NEXT:    [[ARRAYIDX5:%.*]] = getelementptr i8, ptr [[TMP1]], i64 92
+; CHECK-NEXT:    [[TMP12:%.*]] = load float, ptr [[ARRAYIDX5]], align 4
+; CHECK-NEXT:    [[TMP13:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP12]], float 0.000000e+00, float [[TMP11]])
+; CHECK-NEXT:    [[TMP14:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP0]], float 0.000000e+00, float [[TMP9]])
+; CHECK-NEXT:    [[ARRAYIDX6:%.*]] = getelementptr i8, ptr [[TMP1]], i64 96
+; CHECK-NEXT:    [[TMP15:%.*]] = load float, ptr [[ARRAYIDX6]], align 4
+; CHECK-NEXT:    [[TMP16:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP15]], float 0.000000e+00, float [[TMP13]])
+; CHECK-NEXT:    [[ARRAYIDX7:%.*]] = getelementptr i8, ptr [[TMP1]], i64 80
+; CHECK-NEXT:    [[TMP17:%.*]] = load float, ptr [[ARRAYIDX7]], align 4
+; CHECK-NEXT:    [[TMP18:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP0]], float [[TMP17]], float [[TMP16]])
+; CHECK-NEXT:    [[ARRAYIDX8:%.*]] = getelementptr i8, ptr [[TMP1]], i64 100
+; CHECK-NEXT:    [[TMP19:%.*]] = load float, ptr [[ARRAYIDX8]], align 4
+; CHECK-NEXT:    [[TMP20:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP19]], float 0.000000e+00, float [[TMP14]])
+; CHECK-NEXT:    [[ADD:%.*]] = fadd float [[TMP18]], [[TMP20]]
+; CHECK-NEXT:    store float [[ADD]], ptr [[A]], align 4
+; CHECK-NEXT:    ret void
+;
+entry:
+  %1 = load ptr, ptr %a, align 8
+  %arrayidx = getelementptr i8, ptr %1, i64 84
+  %2 = load float, ptr %arrayidx, align 4
+  %3 = tail call float @llvm.fmuladd.f32(float %2, float 0.000000e+00, float 0.000000e+00)
+  %arrayidx1 = getelementptr i8, ptr %1, i64 28
+  %4 = load float, ptr %arrayidx1, align 4
+  %5 = tail call float @llvm.fmuladd.f32(float %4, float 0.000000e+00, float %3)
+  %arrayidx2 = getelementptr i8, ptr %1, i64 8
+  %6 = load float, ptr %arrayidx2, align 4
+  %7 = tail call float @llvm.fmuladd.f32(float %6, float 0.000000e+00, float 0.000000e+00)
+  %arrayidx3 = getelementptr i8, ptr %1, i64 68
+  %8 = load float, ptr %arrayidx3, align 4
+  %9 = tail call float @llvm.fmuladd.f32(float %8, float 0.000000e+00, float %5)
+  %arrayidx4 = getelementptr i8, ptr %1, i64 88
+  %10 = load float, ptr %arrayidx4, align 4
+  %11 = tail call float @llvm.fmuladd.f32(float %10, float 0.000000e+00, float %7)
+  %arrayidx5 = getelementptr i8, ptr %1, i64 92
+  %12 = load float, ptr %arrayidx5, align 4
+  %13 = tail call float @llvm.fmuladd.f32(float %12, float 0.000000e+00, float %11)
+  %14 = tail call float @llvm.fmuladd.f32(float %0, float 0.000000e+00, float %9)
+  %arrayidx6 = getelementptr i8, ptr %1, i64 96
+  %15 = load float, ptr %arrayidx6, align 4
+  %16 = tail call float @llvm.fmuladd.f32(float %15, float 0.000000e+00, float %13)
+  %arrayidx7 = getelementptr i8, ptr %1, i64 80
+  %17 = load float, ptr %arrayidx7, align 4
+  %18 = tail call float @llvm.fmuladd.f32(float %0, float %17, float %16)
+  %arrayidx8 = getelementptr i8, ptr %1, i64 100
+  %19 = load float, ptr %arrayidx8, align 4
+  %20 = tail call float @llvm.fmuladd.f32(float %19, float 0.000000e+00, float %14)
+  %add = fadd float %18, %20
+  store float %add, ptr %a, align 4
+  ret void
+}