From 60c41543469ca59dae5ee9be03d8b05264e84f86 Mon Sep 17 00:00:00 2001
From: Awni Hannun <awni@apple.com>
Date: Mon, 7 Apr 2025 10:47:51 -0700
Subject: [PATCH] Only request residency once (#2051)

---
 mlx/backend/metal/resident.cpp | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/mlx/backend/metal/resident.cpp b/mlx/backend/metal/resident.cpp
index 545f67e49..0a9e1b861 100644
--- a/mlx/backend/metal/resident.cpp
+++ b/mlx/backend/metal/resident.cpp
@@ -22,6 +22,7 @@ ResidencySet::ResidencySet(MTL::Device* d) {
       }
       throw std::runtime_error(msg.str());
     }
+    wired_set_->requestResidency();
   }
 }
 
@@ -32,7 +33,6 @@ void ResidencySet::insert(MTL::Allocation* buf) {
   if (wired_set_->allocatedSize() + buf->allocatedSize() <= capacity_) {
     wired_set_->addAllocation(buf);
     wired_set_->commit();
-    wired_set_->requestResidency();
   } else {
     unwired_set_.insert(buf);
   }
@@ -76,7 +76,6 @@ void ResidencySet::resize(size_t size) {
       }
     }
     wired_set_->commit();
-    wired_set_->requestResidency();
   } else if (current_size > size) {
     auto pool = new_scoped_memory_pool();
     // Remove wired allocations until under capacity