[llvm] [DirectX][Docs] Update DXILArchitecture documentation (PR #86198)

[Farzon Lotfi via llvm-commits]

https://github.com/farzonl updated https://github.com/llvm/llvm-project/pull/86198

>From be275a02593ce59fe917acabff7dca057904e6b4 Mon Sep 17 00:00:00 2001
From: Farzon Lotfi <farzonlotfi at microsoft.com>
Date: Thu, 21 Mar 2024 17:07:30 -0400
Subject: [PATCH] [DirectX][Docs] Update DXILArchitecture documentation for
 instruction expansion

---
 llvm/docs/DirectX/DXILArchitecture.rst | 13 +++++++++++++
 1 file changed, 13 insertions(+)

diff --git a/llvm/docs/DirectX/DXILArchitecture.rst b/llvm/docs/DirectX/DXILArchitecture.rst
index d6712bea4f7712..dd39429d53d87f 100644
--- a/llvm/docs/DirectX/DXILArchitecture.rst
+++ b/llvm/docs/DirectX/DXILArchitecture.rst
@@ -61,6 +61,19 @@ on the utilities described in "Common Code" above in order to share
 logic with both the DirectX backend and with Clang's codegen of HLSL
 support as much as possible.
 
+The DirectX Intrinsic Expansion Pass
+====================================
+There are intrinsics that don't map directly to DXIL Ops. In some cases
+an intrinsic needs to be expanded to a set of LLVM IR instructions. In
+other cases an intrinsic needs modifications to the arguments or return
+values of a DXIL Op. This pass (`DXILIntrinsicExpansion.cpp`) handles all 
+the cases where our intrinsics don't have a one to one mapping. You should 
+also consider using this pass when the expansion is specific to DXIL so 
+as to keep implementation details out of CodeGen. Finally, there is an 
+expectation that we maintain vectors through this pass. Therefore, best 
+practice would be to avoid scalarization in this pass.
+
+
 The DirectX Backend
 ===================