[llvm] [LV] Add target-independent tests for type-mismatch (PR #130295)

David Sherwood via llvm-commits llvm-commits at lists.llvm.org
Thu Mar 27 08:27:26 PDT 2025


================
@@ -0,0 +1,421 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt -passes=loop-vectorize -force-vector-width=4 -S %s | FileCheck %s
+
+;; Tests where load type is different from store type.
+
+define void @load_store_type_mismatch(ptr noalias %src.1, ptr noalias %src.2, ptr noalias %dst.1, ptr noalias %dst.2, i64 %n) {
+; CHECK-LABEL: define void @load_store_type_mismatch(
+; CHECK-SAME: ptr noalias [[SRC_1:%.*]], ptr noalias [[SRC_2:%.*]], ptr noalias [[DST_1:%.*]], ptr noalias [[DST_2:%.*]], i64 [[N:%.*]]) {
+; CHECK-NEXT:  [[ENTRY:.*]]:
+; CHECK-NEXT:    [[UMAX:%.*]] = call i64 @llvm.umax.i64(i64 [[N]], i64 1)
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[UMAX]], 4
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
+; CHECK:       [[VECTOR_PH]]:
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[UMAX]], 4
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[UMAX]], [[N_MOD_VF]]
+; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
+; CHECK:       [[VECTOR_BODY]]:
+; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 1
+; CHECK-NEXT:    [[TMP2:%.*]] = add i64 [[INDEX]], 2
+; CHECK-NEXT:    [[TMP3:%.*]] = add i64 [[INDEX]], 3
+; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr i64, ptr [[SRC_1]], i64 [[TMP0]]
+; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i64, ptr [[TMP4]], i32 0
+; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x i64>, ptr [[TMP5]], align 4
+; CHECK-NEXT:    [[TMP6:%.*]] = trunc <4 x i64> [[WIDE_LOAD]] to <4 x i32>
+; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr i64, ptr [[SRC_2]], i64 [[TMP0]]
+; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr i64, ptr [[TMP7]], i32 0
+; CHECK-NEXT:    [[WIDE_LOAD1:%.*]] = load <4 x i64>, ptr [[TMP8]], align 4
+; CHECK-NEXT:    [[TMP9:%.*]] = add <4 x i64> [[WIDE_LOAD]], [[WIDE_LOAD1]]
+; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr nusw i64, ptr [[DST_1]], i64 [[TMP0]]
+; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr nusw i64, ptr [[DST_1]], i64 [[TMP1]]
+; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr nusw i64, ptr [[DST_1]], i64 [[TMP2]]
+; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr nusw i64, ptr [[DST_1]], i64 [[TMP3]]
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <4 x i32> [[TMP6]], i32 0
+; CHECK-NEXT:    store i32 [[TMP14]], ptr [[TMP10]], align 4
+; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <4 x i32> [[TMP6]], i32 1
+; CHECK-NEXT:    store i32 [[TMP15]], ptr [[TMP11]], align 4
+; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <4 x i32> [[TMP6]], i32 2
+; CHECK-NEXT:    store i32 [[TMP16]], ptr [[TMP12]], align 4
+; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <4 x i32> [[TMP6]], i32 3
+; CHECK-NEXT:    store i32 [[TMP17]], ptr [[TMP13]], align 4
+; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr nusw i64, ptr [[DST_2]], i64 [[TMP0]]
+; CHECK-NEXT:    [[TMP19:%.*]] = getelementptr nusw i64, ptr [[TMP18]], i32 0
+; CHECK-NEXT:    store <4 x i64> [[TMP9]], ptr [[TMP19]], align 4
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-NEXT:    [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP20]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       [[MIDDLE_BLOCK]]:
+; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[UMAX]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[SCALAR_PH]]
+; CHECK:       [[SCALAR_PH]]:
+; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[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_SRC_1:%.*]] = getelementptr i64, ptr [[SRC_1]], i64 [[IV]]
+; CHECK-NEXT:    [[LD_SRC_1:%.*]] = load i64, ptr [[GEP_SRC_1]], align 4
+; CHECK-NEXT:    [[LD_SRC_1_I32:%.*]] = trunc i64 [[LD_SRC_1]] to i32
+; CHECK-NEXT:    [[GEP_SRC_2:%.*]] = getelementptr i64, ptr [[SRC_2]], i64 [[IV]]
+; CHECK-NEXT:    [[LD_SRC_2:%.*]] = load i64, ptr [[GEP_SRC_2]], align 4
+; CHECK-NEXT:    [[ADD:%.*]] = add i64 [[LD_SRC_1]], [[LD_SRC_2]]
+; CHECK-NEXT:    [[GEP_DST_1:%.*]] = getelementptr nusw i64, ptr [[DST_1]], i64 [[IV]]
+; CHECK-NEXT:    store i32 [[LD_SRC_1_I32]], ptr [[GEP_DST_1]], align 4
+; CHECK-NEXT:    [[GEP_DST_2:%.*]] = getelementptr nusw i64, ptr [[DST_2]], i64 [[IV]]
+; CHECK-NEXT:    store i64 [[ADD]], ptr [[GEP_DST_2]], align 4
+; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT:    [[COND:%.*]] = icmp ult i64 [[IV_NEXT]], [[N]]
+; CHECK-NEXT:    br i1 [[COND]], label %[[LOOP]], label %[[EXIT]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK:       [[EXIT]]:
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop
+
+loop:
+  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
+
+  %gep.src.1 = getelementptr i64, ptr %src.1, i64 %iv
+  %ld.src.1 = load i64, ptr %gep.src.1
+  %ld.src.1.i32 = trunc i64 %ld.src.1 to i32
+
+  %gep.src.2 = getelementptr i64, ptr %src.2, i64 %iv
+  %ld.src.2 = load i64, ptr %gep.src.2
+  %add = add i64 %ld.src.1, %ld.src.2
+
+  %gep.dst.1 = getelementptr nusw i64, ptr %dst.1, i64 %iv
+  store i32 %ld.src.1.i32, ptr %gep.dst.1
+
+  %gep.dst.2 = getelementptr nusw i64, ptr %dst.2, i64 %iv
+  store i64 %add, ptr %gep.dst.2
+
+  %iv.next = add nuw nsw i64 %iv, 1
+  %cond = icmp ult i64 %iv.next, %n
+  br i1 %cond, label %loop, label %exit
+
+exit:
+  ret void
+}
+
+define void @load_store_type_mismatch_rt_memcheck(ptr %src.1, ptr %src.2, ptr %dst.1, ptr %dst.2, i64 %n) {
----------------
david-arm wrote:

I wonder if the runtime check tests should live in test/Transforms/LoopVectorize/runtime-check.ll?

Also, is it worth having a test with just a single i64 load, a trunc to i32, and a store of i32? I thought that if there were only two accesses in the loop we used diff checks instead, which would exercise a different code path?

https://github.com/llvm/llvm-project/pull/130295


More information about the llvm-commits mailing list