[llvm] [LV][EVL] Generate negative strided load/store for reversed load/store (PR #123608)

[Luke Lau via llvm-commits]

================
@@ -39,16 +39,14 @@ define void @reverse_load_store(i64 %startval, ptr noalias %ptr, ptr noalias %pt
 ; IF-EVL-NEXT:    [[TMP10:%.*]] = sub i64 1, [[TMP18]]
 ; IF-EVL-NEXT:    [[TMP16:%.*]] = getelementptr i32, ptr [[TMP8]], i64 [[TMP9]]
 ; IF-EVL-NEXT:    [[TMP12:%.*]] = getelementptr i32, ptr [[TMP16]], i64 [[TMP10]]
-; IF-EVL-NEXT:    [[VP_OP_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.vp.load.nxv4i32.p0(ptr align 4 [[TMP12]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
-; IF-EVL-NEXT:    [[VP_REVERSE:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.reverse.nxv4i32(<vscale x 4 x i32> [[VP_OP_LOAD]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
+; IF-EVL-NEXT:    [[VP_REVERSE_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.experimental.vp.strided.load.nxv4i32.p0.i32(ptr align 4 [[TMP12]], i32 -4, <vscale x 4 x i1> splat (i1 true), i32 [[TMP5]])
----------------
lukel97 wrote:

I presume that LoopVectorizationLegality would have checked beforehand that either ordering is correct. Because assuming that the scalar loop was written something like:

```c
for (int i = n - 1; i >= 0; i--)
  ... = x[i];
```

Then it would have originally loaded the values like `<%ptr, %ptr - 1, %ptr -2 ...>`.

https://github.com/llvm/llvm-project/pull/123608