From 01735f33e8c584b20f9d605007389f13f0b0687d Mon Sep 17 00:00:00 2001
From: EgorBo <egorbo@gmail.com>
Date: Sun, 13 Oct 2024 19:42:29 +0200
Subject: [PATCH] Remove redundant sign/zero extension for SIMD broadcasts

---
 src/coreclr/jit/lowerxarch.cpp | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)

diff --git a/src/coreclr/jit/lowerxarch.cpp b/src/coreclr/jit/lowerxarch.cpp
index 1da40d1dc453b..1a20e57a7c66e 100644
--- a/src/coreclr/jit/lowerxarch.cpp
+++ b/src/coreclr/jit/lowerxarch.cpp
@@ -10356,6 +10356,22 @@ void Lowering::ContainCheckHWIntrinsic(GenTreeHWIntrinsic* node)
 
                                 op1 = node->Op(1);
                                 op1->ClearContained();
+
+                                // Drop GT_CAST if it doesn't change the op1's bits we're about to broadcast
+                                if (op1->OperIs(GT_CAST) && !op1->gtOverflow())
+                                {
+                                    GenTreeCast* cast = op1->AsCast();
+                                    if (!varTypeIsFloating(cast->CastToType()) &&
+                                        !varTypeIsFloating(cast->CastFromType()) &&
+                                        (genTypeSize(cast->CastToType()) >= genTypeSize(simdBaseType)) &&
+                                        (genTypeSize(cast->CastFromType()) >= genTypeSize(simdBaseType)))
+                                    {
+                                        BlockRange().Remove(op1);
+                                        op1 = cast->CastOp();
+                                        op1->ClearContained();
+                                        node->Op(1) = op1;
+                                    }
+                                }
                             }
                         }
                         break;