From 13ba9328d2e38af8f45c2626bf6d94d65117a339 Mon Sep 17 00:00:00 2001
From: YOUNGJOON KIM <god102104@gmail.com>
Date: Sat, 21 Feb 2026 19:17:26 +0900
Subject: [PATCH 1/3] ROI pool CUDA: compare in acc_type to fix half/MSVC build

---
 torchvision/csrc/ops/cuda/roi_pool_kernel.cu | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)

diff --git a/torchvision/csrc/ops/cuda/roi_pool_kernel.cu b/torchvision/csrc/ops/cuda/roi_pool_kernel.cu
index 3a9374bb438..ccb8888cacb 100644
--- a/torchvision/csrc/ops/cuda/roi_pool_kernel.cu
+++ b/torchvision/csrc/ops/cuda/roi_pool_kernel.cu
@@ -63,10 +63,14 @@ __global__ void roi_pool_forward_kernel_impl(
     int maxidx = -1;
     const T* offset_input =
         input + (roi_batch_ind * channels + c) * height * width;
+    using acc_t = at::acc_type<T, /*is_cuda=*/true>;
     for (int h = hstart; h < hend; ++h) {
       for (int w = wstart; w < wend; ++w) {
         int input_index = h * width + w;
-        if (offset_input[input_index] > maxval) {
+        acc_t v  = static_cast<acc_t>(offset_input[input_index]);
+        acc_t mv = static_cast<acc_t>(maxval);
+
+        if (v > mv) {
           maxval = offset_input[input_index];
           maxidx = input_index;
         }

From c03784b5f66ca6584d71a7eba779dcc110da740c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Patrick=20Kim=20=28=EA=B9=80=EC=98=81=EC=A4=80=29?=
 <43011129+kimchioverfit@users.noreply.github.com>
Date: Mon, 23 Feb 2026 18:11:28 +0900
Subject: [PATCH 2/3] Add AccumulateType header to roi_pool_kernel.cu

Add <ATen/AccumulateType.h>
---
 torchvision/csrc/ops/cuda/roi_pool_kernel.cu | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/torchvision/csrc/ops/cuda/roi_pool_kernel.cu b/torchvision/csrc/ops/cuda/roi_pool_kernel.cu
index ccb8888cacb..4b78dc2e705 100644
--- a/torchvision/csrc/ops/cuda/roi_pool_kernel.cu
+++ b/torchvision/csrc/ops/cuda/roi_pool_kernel.cu
@@ -4,7 +4,7 @@
 #include <float.h>
 #include <torch/library.h>
 #include <ATen/native/cuda/KernelUtils.cuh>
-
+#include <ATen/AccumulateType.h>
 #include "cuda_helpers.h"
 
 namespace vision {

From 788bf6a94235d599ccd01d0ad8846e72df3a2a03 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Patrick=20Kim=20=28=EA=B9=80=EC=98=81=EC=A4=80=29?=
 <43011129+kimchioverfit@users.noreply.github.com>
Date: Tue, 24 Feb 2026 15:14:38 +0900
Subject: [PATCH 3/3] Apply clang-format to roi_pool_kernel.cu

No functional changes.
Only formatting adjustments required by clang-format (include order and spacing).
---
 torchvision/csrc/ops/cuda/roi_pool_kernel.cu | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/torchvision/csrc/ops/cuda/roi_pool_kernel.cu b/torchvision/csrc/ops/cuda/roi_pool_kernel.cu
index 4b78dc2e705..894531901ea 100644
--- a/torchvision/csrc/ops/cuda/roi_pool_kernel.cu
+++ b/torchvision/csrc/ops/cuda/roi_pool_kernel.cu
@@ -1,10 +1,11 @@
 #include <ATen/ATen.h>
+#include <ATen/AccumulateType.h>
 #include <ATen/cuda/CUDAContext.h>
 #include <c10/cuda/CUDAGuard.h>
 #include <float.h>
 #include <torch/library.h>
 #include <ATen/native/cuda/KernelUtils.cuh>
-#include <ATen/AccumulateType.h>
+
 #include "cuda_helpers.h"
 
 namespace vision {
@@ -67,7 +68,7 @@ __global__ void roi_pool_forward_kernel_impl(
     for (int h = hstart; h < hend; ++h) {
       for (int w = wstart; w < wend; ++w) {
         int input_index = h * width + w;
-        acc_t v  = static_cast<acc_t>(offset_input[input_index]);
+        acc_t v = static_cast<acc_t>(offset_input[input_index]);
         acc_t mv = static_cast<acc_t>(maxval);
 
         if (v > mv) {