[LAA] Make sure MaxVF for Store-Load forward safe dep distances is pow2.

MaxVF computed in couldPreventStoreLoadFowrard may not be a power of 2,
as CommonStride may not be a power-of-2.

This can cause crashes after 78777a2. Use bit_floor to make sure it is
a suitable power-of-2.

Fixes #134696.

Author: fhahn

llvm/lib/Analysis/LoopAccessAnalysis.cpp

@@ -1781,7 +1781,8 @@ bool MemoryDepChecker::couldPreventStoreLoadForward(uint64_t Distance,
           MaxStoreLoadForwardSafeDistanceInBits &&
       MaxVFWithoutSLForwardIssuesPowerOf2 !=
           VectorizerParams::MaxVectorWidth * TypeByteSize) {
-    uint64_t MaxVF = MaxVFWithoutSLForwardIssuesPowerOf2 / CommonStride;
+    uint64_t MaxVF =
+        bit_floor(MaxVFWithoutSLForwardIssuesPowerOf2 / CommonStride);
     uint64_t MaxVFInBits = MaxVF * TypeByteSize * 8;
     MaxStoreLoadForwardSafeDistanceInBits =
         std::min(MaxStoreLoadForwardSafeDistanceInBits, MaxVFInBits);

llvm/test/Transforms/LoopVectorize/RISCV/safe-dep-distance.ll

@@ -256,3 +256,66 @@ loop:
 exit:
   ret void
 }
+
+@a = external global [10 x [12 x i16]]
+
+; Test case for https://github.com/llvm/llvm-project/issues/134696.
+define void @safe_load_store_distance_not_pow_of_2(i64 %N) {
+; CHECK-LABEL: @safe_load_store_distance_not_pow_of_2(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[N:%.*]], i64 1)
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i64 [[N]], [[UMIN]]
+; CHECK-NEXT:    [[TMP1:%.*]] = udiv i64 [[TMP0]], 3
+; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[UMIN]], [[TMP1]]
+; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[TMP2]], 1
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP3]], 8
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK:       vector.ph:
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 8
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
+; CHECK-NEXT:    [[TMP5:%.*]] = select i1 [[TMP4]], i64 8, i64 [[N_MOD_VF]]
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP3]], [[TMP5]]
+; CHECK-NEXT:    [[TMP6:%.*]] = mul i64 [[N_VEC]], 3
+; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
+; CHECK:       vector.body:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <8 x i64> [ <i64 0, i64 3, i64 6, i64 9, i64 12, i64 15, i64 18, i64 21>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr [10 x [12 x i16]], ptr @a, i64 0, i64 8, <8 x i64> [[VEC_IND]]
+; CHECK-NEXT:    call void @llvm.masked.scatter.v8i16.v8p0(<8 x i16> zeroinitializer, <8 x ptr> [[TMP7]], i32 2, <8 x i1> splat (i1 true))
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
+; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <8 x i64> [[VEC_IND]], splat (i64 24)
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
+; CHECK:       middle.block:
+; CHECK-NEXT:    br label [[SCALAR_PH]]
+; CHECK:       scalar.ph:
+; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[TMP6]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr [10 x [12 x i16]], ptr @a, i64 0, i64 0, i64 [[IV]]
+; CHECK-NEXT:    [[TMP9:%.*]] = load i16, ptr [[GEP]], align 2
+; CHECK-NEXT:    [[GEP_OFF:%.*]] = getelementptr [10 x [12 x i16]], ptr @a, i64 0, i64 8, i64 [[IV]]
+; CHECK-NEXT:    store i16 0, ptr [[GEP_OFF]], align 2
+; CHECK-NEXT:    [[IV_NEXT]] = add nsw i64 [[IV]], 3
+; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i64 [[IV]], [[N]]
+; CHECK-NEXT:    br i1 [[CMP]], label [[LOOP]], label [[EXIT:%.*]], !llvm.loop [[LOOP11:![0-9]+]]
+; CHECK:       exit:
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+  %gep = getelementptr [10 x [12 x i16]], ptr @a, i64 0, i64 0, i64 %iv
+  %1 = load i16, ptr %gep, align 2
+  %gep.off = getelementptr [10 x [12 x i16]], ptr @a, i64 0, i64 8, i64 %iv
+  store i16 0, ptr %gep.off, align 2
+  %iv.next = add nsw i64 %iv, 3
+  %cmp = icmp ult i64 %iv, %N
+  br i1 %cmp, label %loop, label %exit
+
+exit:
+  ret void
+}