[llvm-branch-commits] [llvm] [AMDGPU] Add amdgcn.av.global.(load|store).b128 intrinsics (PR #191390)

[Shilei Tian via llvm-branch-commits]

================
@@ -1775,6 +1775,111 @@ The AMDGPU backend implements the following LLVM IR intrinsics.
                                                    * :ref:`Synchronization Scope<amdgpu-intrinsics-syncscope-metadata-operand>`.
                                                      Note that the scope used must ensure that the L2 cache will be hit.
 
+  llvm.amdgcn.av.global.load.b128                     This intrinsic is supported on gfx9, gfx10, gfx11, and gfx12 targets.
+
+                                                   Signature:
+
+                                                   .. code-block:: llvm
+
+                                                      <4 x i32> @llvm.amdgcn.av.global.load.b128(
+                                                          ptr addrspace(1), ; source
+                                                          metadata)         ; scope    - e.g. '!0' where '!0 = !{!"wavegroup"}'
+
+                                                   Reads the value from the source address with cache behavior specified by the scope.
+
+                                                   The following table shows the mapping between valid scope values and target
+                                                   instruction flags or field values.
+
+                                                   ============== ========================== ========================== ========================== ========================== ==========================
+                                                   targets        instruction                           ``"wavefront"``            ``"workgroup"``                ``"agent"``      ``""`` (empty string)
+                                                   ============== ========================== ========================== ========================== ========================== ==========================
+                                                   gfx90*         ``global_load_dwordx4``                                                                             ``glc``                    ``glc``
+
+                                                   gfx942, gfx950 ``global_load_dwordx4``                        (wave)            ``sc0`` (group)           ``sc1`` (device)       ``sc0 sc1`` (system)
+
+                                                   gfx10*         ``global_load_dwordx4``                                                  ``glc``                ``glc dlc``                ``glc dlc``
+
+                                                   gfx11*         ``global_load_dwordx4``                                                  ``glc``                    ``glc``                    ``glc``
+
+                                                   gfx120*        ``av_global_load_b128``                             (CU)    ``scope:SCOPE_SE`` (SE)  ``scope:SCOPE_DEV`` (DEV)  ``scope:SCOPE_SYS`` (SYS)
----------------
shiltian wrote:

The alignment here and below for `(CU)` might be off.

https://github.com/llvm/llvm-project/pull/191390