[llvm] AMDGPU/NewPM Port GCNDPPCombine to NPM (PR #105816)

[Pravin Jagtap via llvm-commits]

================
@@ -0,0 +1,55 @@
+//=======- GCNDPPCombine.h - optimization for DPP instructions ---==========//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+// The pass combines V_MOV_B32_dpp instruction with its VALU uses as a DPP src0
+// operand. If any of the use instruction cannot be combined with the mov the
+// whole sequence is reverted.
+//
+// $old = ...
+// $dpp_value = V_MOV_B32_dpp $old, $vgpr_to_be_read_from_other_lane,
+//                            dpp_controls..., $row_mask, $bank_mask,
+//                            $bound_ctrl
+// $res = VALU $dpp_value [, src1]
+//
+// to
+//
+// $res = VALU_DPP $combined_old, $vgpr_to_be_read_from_other_lane, [src1,]
+//                 dpp_controls..., $row_mask, $bank_mask, $combined_bound_ctrl
+//
+// Combining rules :
+//
+// if $row_mask and $bank_mask are fully enabled (0xF) and
+//    $bound_ctrl==DPP_BOUND_ZERO or $old==0
+// -> $combined_old = undef,
+//    $combined_bound_ctrl = DPP_BOUND_ZERO
+//
+// if the VALU op is binary and
+//    $bound_ctrl==DPP_BOUND_OFF and
+//    $old==identity value (immediate) for the VALU op
+// -> $combined_old = src1,
+//    $combined_bound_ctrl = DPP_BOUND_OFF
+//
+// Otherwise cancel.
+//
+// The mov_dpp instruction should reside in the same BB as all its uses
+//===----------------------------------------------------------------------===//
+
----------------
pravinjagtap wrote:

This is same as llvm/lib/Target/AMDGPU/GCNDPPCombine.cpp description. Avoid the duplication.

https://github.com/llvm/llvm-project/pull/105816