[llvm] Add an all-in-one histogram intrinsic, along with lowering for AArch64 (PR #88106)

[Jay Foad via llvm-commits]

================
@@ -19030,6 +19030,65 @@ will be on any later loop iteration.
 This intrinsic will only return 0 if the input count is also 0. A non-zero input
 count will produce a non-zero result.
 
+'``llvm.experimental.vector.histogram.*``' Intrinsic
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+This is an overloaded intrinsic.
+
+These intrinsics represent histogram-like operations; that is, updating values
+in memory that may not be contiguous, and where multiple elements within a
+single vector may be updating the same value in memory.
+
+The update operation must be specified as part of the intrinsic name. For a
+simple histogram like the following the ``add`` operation would be used.
+
+.. code-block:: c
+
+    void simple_histogram(int *restrict buckets, unsigned *indices, int N, int inc) {
+      for (int i = 0; i < N; ++i)
+        buckets[indices[i]] += inc;
+    }
+
+More update operation types may be added in the future.
+
+::
+
+    declare <8 x i32> @llvm.experimental.vector.histogram.add.v8p0.i32(<8 x ptr> %ptrs, i32 %inc, <8 x i1> %mask)
+    declare <vscale x 2 x i64> @llvm.experimental.vector.histogram.add.nxv2p0.i64(<vscale x 2 x ptr> %ptrs, i64 %inc, <vscale x 2 x i1> %mask)
+
+Arguments:
+""""""""""
+
+The first argument is a vector of pointers to the memory locations to be
+updated. The second argument is a scalar used to update the value from
+memory; it must match the type of value to be updated. The final argument
+is a mask value which will exclude that vector element from being updated,
+and will exclude it from any cross-lane calculations to determine the
+final values for each memory location.
+
+Semantics:
+""""""""""
+
+The '``llvm.experimental.vector.histogram``' intrinsics are used to perform
+updates on potentially overlapping values in memory. The intrinsics represent
+the follow sequence of operations:
+
+1. Gather load from the ``ptrs`` operand, with element type matching that of
+   the ``inc`` operand.
+2. Update of the values loaded from memory. In the case of the ``add``
+   update operation, this means:
+
+   1. Perform a cross-vector histogram operation on the ``ptrs`` operand,
+      or a set of index values if it can be decomposed into a base pointer
+      with smaller indices matching the type of ``inc``.
+   2. Multiply the result by the ``inc`` operand.
+   3. Add the result to the values loaded from memory
+3. Scatter the result of the update operation to the memory locations from
+   the ``ptrs`` operand.
+
+The ``mask`` operand will apply to at least the gather and scatter operations,
+and potentially the update if supported.
----------------
jayfoad wrote:

"potentially the update if supported" sounds like an implementation optimization. I have a slight preference to keep the semantics section as simple as possible by omitting it. Same for the "or a set of index value..." part above.

https://github.com/llvm/llvm-project/pull/88106