From 41d30395daabe253227bf96df68d8ba879f4f2b3 Mon Sep 17 00:00:00 2001
From: Duncan Tourolle <duncan@tourolle.paris>
Date: Fri, 31 Jul 2026 22:25:00 +0200
Subject: [PATCH] fix(ort): CUDA detection was gated behind the TensorRT-EP
 build flag

detect_ort_provider() only tested for CUDAExecutionProvider inside
#ifdef SAE_ORT_WITH_TRT_EP, so any build that did not also opt into the
TensorRT execution provider could never select CUDA and fell straight
through to the CPU. The two are independent: the TRT EP needs the
headers and profile plumbing and is rightly an opt-in, CUDA is a plain
ORT provider and is not.

The failure is silent rather than loud, which is why it survived --
inference runs on the CPU and every answer is still correct, just far
slower. Measured on the VR-012 study: 0.32 s/crop against 0.0021 s/crop
once a GPU backend is actually used, with the card sitting at 212 MiB
and 0% utilisation throughout.

Only the TensorrtExecutionProvider line stays inside the guard.
---
 src/backends/ort_provider.hpp | 13 ++++++++++++-
 1 file changed, 12 insertions(+), 1 deletion(-)

diff --git a/src/backends/ort_provider.hpp b/src/backends/ort_provider.hpp
index 7f27c26..a420435 100644
--- a/src/backends/ort_provider.hpp
+++ b/src/backends/ort_provider.hpp
@@ -23,12 +23,23 @@
 enum class OrtProvider { CPU, CUDA, ROCm, TensorRT };
 
 inline OrtProvider detect_ort_provider() {
+    // ORT returns these in its own preference order (TensorRT, CUDA, ..., CPU
+    // last), so the first recognised entry is the best available and the loop
+    // returns on it.
     auto available = Ort::GetAvailableProviders();
     for (const auto& p : available) {
+        // Only the TensorRT *EP* is a build-time opt-in — it needs the headers
+        // and the profile plumbing below. CUDA is not: it is a plain ORT
+        // provider, and gating its detection on the TRT flag (as this did) made
+        // the CUDA branch unreachable in every build that did not also ask for
+        // TensorRT. The symptom is silent rather than loud — inference simply
+        // runs on the CPU and everything still returns correct answers — which
+        // is why it survived: a 300-actor VR-012 grid cell took 76 s on the CPU
+        // with the GPU idle at 212 MiB.
 #ifdef SAE_ORT_WITH_TRT_EP
         if (p == "TensorrtExecutionProvider") return OrtProvider::TensorRT;
-        if (p == "CUDAExecutionProvider")     return OrtProvider::CUDA;
 #endif
+        if (p == "CUDAExecutionProvider")     return OrtProvider::CUDA;
         if (p == "ROCMExecutionProvider")     return OrtProvider::ROCm;
     }
     return OrtProvider::CPU;