[ROCm] Fix MIOpen CTC loss crash on Windows (pytorch#179264)

<h2>Fix MIOpen CTC loss access violation on Windows discrete GPUs</h2>

<h3>Problem</h3>

<p>A failing unit test on Windows started showing a couple weeks ago and a missing <code>#include</code> was added in [](pytorch#178284), but CI on TheRock kept failing. The fix was tested on gfx1151 (APU), where the test passed, but CI showed failures on gfx1100. </p>

<p><code>test_CTCLoss_no_batch_dim</code> (and any code path hitting <code>miopen_ctc_loss</code>) crashes with a fatal access violation on Windows systems with discrete AMD GPUs:</p>

<pre><code>Windows fatal exception: access violation
Exception Code: 0xC0000005
#0 miopen::CTCLossDescriptor::GetCTCLossWorkspaceSize (MIOpen.dll+0x14fde4)
#1 miopenGetCTCLossWorkspaceSize (MIOpen.dll+0x150912)
#2 at::native::miopen_ctc_loss (torch_hip.dll)
</code></pre>

<h3>Root Cause</h3>

<p><code>miopenGetCTCLossWorkspaceSize</code> and <code>miopenCTCLoss</code> read the <code>labels</code>, <code>label_lengths</code>, and <code>input_lengths</code> arrays <strong>on the host side</strong> to plan the computation and calculate workspace requirements. The existing code copies these arrays to GPU memory and passes device pointers:</p>

<pre><code>Tensor labels_gpu = targets_t.to(Device(at::kCUDA), at::kInt);
// ... hipMemcpy to GPU ...
MIOPEN_CHECK(miopenGetCTCLossWorkspaceSize(...,
    labels_gpu.data_ptr&lt;int&gt;(),          // device pointer
    label_lengths_gpu.data_ptr&lt;int&gt;(),   // device pointer
    input_lengths_gpu.data_ptr&lt;int&gt;()    // device pointer
));
</code></pre>

<p>This works on:</p>
<ul>
<li><strong>Linux</strong> — HSA (Heterogeneous System Architecture) maps GPU allocations into the process virtual address space, making device pointers host-readable</li>
<li><strong>Windows APUs</strong> — CPU and iGPU share system RAM, so device pointers point to host-accessible memory</li>
</ul>

<p>This crashes on:</p>
<ul>
<li><strong>Windows dGPUs</strong> — GPU has dedicated VRAM across PCIe; device pointers are opaque handles that cannot be dereferenced from host code</li>
</ul>

<h3>Verification</h3>

<p>Tested on gfx1201:</p>

<table border="1" cellpadding="6" cellspacing="0">
<tr><th>Check</th><th>Result</th></tr>
<tr><td><code>hipDeviceAttributeIntegrated</code></td><td><code>0</code> (discrete GPU)</td></tr>
<tr><td><code>hipDeviceAttributeCanUseHostPointerForRegisteredMem</code></td><td><code>0</code></td></tr>
<tr><td><code>hipDeviceAttributeManagedMemory</code></td><td><code>0x7FFFFFFF</code> (unsupported)</td></tr>
<tr><td><code>hipDeviceAttributeUnifiedAddressing</code></td><td><code>0x7FFFFFFF</code> (unsupported)</td></tr>
<tr><td>Host read of <code>hipMalloc</code> pointer via <code>ctypes</code></td><td>Access violation</td></tr>
<tr><td>CTC loss with CPU pointers</td><td>Pass (forward + backward)</td></tr>
</table>

<h3>Fix</h3>

<p>Use host pointers since this is what MIOpen expects should be used.</p>

<h3>Testing</h3>

<p>Run all existing CTCLoss unit tests.</p>

Pull Request resolved: pytorch#179264
Approved by: https://github.com/jeffdaily

Co-authored-by: Jeff Daily <jeff.daily@amd.com>

Author: mstankov-amd

aten/src/ATen/native/miopen/LossCTC_miopen.cpp

@@ -207,35 +207,16 @@ std::tuple<Tensor, Tensor> miopen_ctc_loss(
   Tensor costs = at::empty({batch_size}, log_probs->options());
   Tensor grad = at::empty_like(log_probs_t, LEGACY_CONTIGUOUS_MEMORY_FORMAT);
 
-  // MIOpen requires labels and lengths on GPU
-  Tensor labels_gpu = targets_t.to(Device(at::kCUDA), at::kInt);
-  Tensor label_lengths_gpu = at::empty(
-      {static_cast<int64_t>(target_lengths.size())},
-      at::TensorOptions().dtype(at::kInt).device(at::kCUDA));
-  Tensor input_lengths_gpu = at::empty(
-      {static_cast<int64_t>(input_lengths.size())},
-      at::TensorOptions().dtype(at::kInt).device(at::kCUDA));
-
-  C10_CUDA_CHECK(hipMemcpy(
-      label_lengths_gpu.data_ptr<int>(),
-      target_lengths.data(),
-      target_lengths.size() * sizeof(int),
-      hipMemcpyHostToDevice));
-  C10_CUDA_CHECK(hipMemcpy(
-      input_lengths_gpu.data_ptr<int>(),
-      input_lengths.data(),
-      input_lengths.size() * sizeof(int),
-      hipMemcpyHostToDevice));
-
+  // MIOpen reads labels/lengths on the host.
   size_t workspace_size;
   (void)deterministic; // MIOpen only supports deterministic algorithm
   MIOPEN_CHECK(miopenGetCTCLossWorkspaceSize(
       handle,
       probs_desc,
       grads_desc,
-      labels_gpu.data_ptr<int>(),
-      label_lengths_gpu.data_ptr<int>(),
-      input_lengths_gpu.data_ptr<int>(),
+      targets_t.data_ptr<int>(),
+      target_lengths.data(),
+      input_lengths.data(),
       MIOPEN_CTC_LOSS_ALGO_DETERMINISTIC,
       ctc_desc,
       &workspace_size));
@@ -246,9 +227,9 @@ std::tuple<Tensor, Tensor> miopen_ctc_loss(
       handle,
       probs_desc,
       log_probs_t.data_ptr(),
-      labels_gpu.data_ptr<int>(),
-      label_lengths_gpu.data_ptr<int>(),
-      input_lengths_gpu.data_ptr<int>(),
+      targets_t.data_ptr<int>(),
+      target_lengths.data(),
+      input_lengths.data(),
       costs.data_ptr(),
       grads_desc,
       grad.data_ptr(),