diff --git a/tests/unit/test_d1_hd128_debug.py b/tests/unit/test_d1_hd128_debug.py
new file mode 100644
index 00000000..37e91870
--- /dev/null
+++ b/tests/unit/test_d1_hd128_debug.py
@@ -0,0 +1,73 @@
+"""D1: Debug hd=128 — check if the QK output is correct."""
+import torch, math
+import cutlass.cute as cute
+import cutlass.torch as ct
+import cuda.bindings.driver as cuda
+from dsv4.kernels.attention.fmha import FmhaKernel
+
+
+def test_hd128_debug():
+    hd = 128
+    n_kv = 128
+    m = 128
+    torch.manual_seed(42)
+    q = torch.randn(m, hd, 1, dtype=torch.bfloat16, device='cuda')
+    k = torch.randn(n_kv, hd, 1, dtype=torch.bfloat16, device='cuda')
+    v = torch.randn(n_kv, hd, dtype=torch.bfloat16, device='cuda')
+
+    # Reference: just the QK @ V attention (un-normalized)
+    qf = q[:, :, 0].float()
+    kf = k[:, :, 0].float()
+    scale = 1.0 / math.sqrt(hd)
+    attn = qf @ kf.T * scale
+    attn_max = attn.max(dim=-1, keepdim=True)[0]
+    attn_exp = torch.exp(attn - attn_max)
+    attn_sum = attn_exp.sum(dim=-1, keepdim=True)
+    ref_unnorm = attn_exp @ v.float()
+    ref_lse = (torch.log(attn_sum.squeeze(-1)) + attn_max.squeeze(-1))[0].item()
+
+    # Run kernel with TMEM-P (force)
+    lse_tensor = torch.zeros(m, 1, 1, dtype=torch.float32, device='cuda')
+    kernel = FmhaKernel(head_dim=hd, s_k=n_kv, use_smem_p=False)
+    pv_n_tile = kernel.pv_n_tile
+    print(f'pv_n_tile={pv_n_tile}, n_pv_tiles={kernel.n_pv_tiles}')
+    print(f'tmem_o0_offset={kernel.tmem_o0_offset}, tmem_p0_offset={kernel.tmem_p0_offset}')
+    print(f'tOrP0_offset={kernel.tOrP0_offset}')
+    print(f'num_tmem_alloc_cols={kernel.num_tmem_alloc_cols}')
+    print(f'scale_softmax={kernel.scale_softmax}, scale_softmax_log2={kernel.scale_softmax_log2}')
+
+    stream = cuda.CUstream(torch.cuda.current_stream().cuda_stream)
+    v_tile = v[:, 0:pv_n_tile].contiguous().unsqueeze(-1)
+    c_tile = torch.zeros(m, pv_n_tile, 1, dtype=torch.bfloat16, device='cuda')
+    mQ = ct.from_dlpack(q).mark_layout_dynamic(leading_dim=ct.get_leading_dim(q))
+    mK = ct.from_dlpack(k).mark_layout_dynamic(leading_dim=ct.get_leading_dim(k))
+    mV = ct.from_dlpack(v_tile).mark_layout_dynamic(leading_dim=ct.get_leading_dim(v_tile))
+    mC = ct.from_dlpack(c_tile).mark_layout_dynamic(leading_dim=ct.get_leading_dim(c_tile))
+    mLSE = ct.from_dlpack(lse_tensor).mark_layout_dynamic(leading_dim=ct.get_leading_dim(lse_tensor))
+
+    compiled = cute.compile(kernel, mQ, mK, mV, mC, stream, mLSE)
+    compiled(mQ, mK, mV, mC, stream, mLSE)
+    torch.cuda.synchronize()
+
+    out = c_tile[:, :, 0].float()
+    kernel_lse = lse_tensor[0, 0, 0].item()
+    cos = torch.nn.functional.cosine_similarity(out.flatten().unsqueeze(0), ref_unnorm.flatten().unsqueeze(0)).item()
+
+    print(f'\nResults:')
+    print(f'  cos_unnorm={cos:.6f}')
+    print(f'  kernel_lse={kernel_lse:.6f}  ref_lse={ref_lse:.6f}  err={abs(kernel_lse - ref_lse):.6f}')
+    print(f'  out[0,:4]={out[0,:4].tolist()}')
+    print(f'  ref[0,:4]={ref_unnorm[0,:4].tolist()}')
+
+    # Check: is the output roughly the right magnitude?
+    print(f'  out.abs().max()={out.abs().max().item():.4f}')
+    print(f'  ref.abs().max()={ref_unnorm.abs().max().item():.4f}')
+
+    # Check row-by-row: is O[0] proportional to ref[0]?
+    if cos < 0.99:
+        row_cos = torch.nn.functional.cosine_similarity(out[0].unsqueeze(0), ref_unnorm[0].unsqueeze(0)).item()
+        print(f'  row0_cos={row_cos:.6f}')
+
+
+if __name__ == '__main__':
+    test_hd128_debug()