restructure results

1 files changed, 23 insertions, 0 deletions

diff --git a/results/classifier/deepseek-r1:14b/reasoning/performance/693 b/results/classifier/deepseek-r1:14b/reasoning/performance/693
new file mode 100644
index 000000000..6f268a289
--- /dev/null
+++ b/results/classifier/deepseek-r1:14b/reasoning/performance/693
@@ -0,0 +1,23 @@
+<think>
+Okay, so I need to classify this bug report about QEMU's increased memory usage with TCG. Let me read through the problem description again.
+
+The user reports that when using QEMU with accel=tcg, instances allocated 256MB of memory end up using around 512MB of RSS (Resident Set Size). This doesn't happen when using accel=kvm or with older versions like qemu-4.2. It's observed across multiple distributions: Debian 11, Centos 8 Stream, Ubuntu 21.10, and a pre-release of Ubuntu 22.04.
+
+So the main issue is related to memory usage when running under TCG (Translation-Cache Guest), which is QEMU's user-mode emulation without KVM. The bug doesn't occur with KVM acceleration, suggesting it's specific to how TCG handles memory in newer QEMU versions.
+
+Looking at the possible categories: mistranslation, assembly lowering, other, device, graphic, socket, network, KVM, boot, vnc, debug, files, permissions, performance, kernel, peripherals, hypervisor.
+
+The problem is about increased memory usage. "Performance" seems relevant because it's related to how resources are used. But the issue is specifically with TCG, which is part of QEMU's user-mode translation. So maybe "mistranslation" or another category? Alternatively, "performance" could be a broad enough category that covers this.
+
+But wait, "mistranslation" refers more to incorrect mapping from source to target architecture. The issue here isn't about incorrect code generation but about higher memory consumption. So perhaps "performance" is the right fit since it's affecting resource usage.
+
+Another thought: Could it be related to assembly lowering? If TCG generates more efficient or inefficient code, that could affect memory. But in this case, it's a significant increase without any functional issues, just resource usage. 
+
+Alternatively, maybe "hypervisor" isn't the right category since KVM is being used as an alternative, but when accel=tcg is on, QEMU uses its own translation which doesn't involve the hypervisor as much.
+
+Wait, the possible categories include "performance", so that seems appropriate here because the problem is about memory usage (a performance metric) under TCG. 
+
+I think "performance" would be the best fit for this issue.
+</think>
+
+performance
\ No newline at end of file