refactor(triton): share INT8 autotune kwargs

Keep both INT8 matmul kernels on one configs, cache-key, and pruning contract so future device-scoping changes cannot diverge between kernel paths. Add a focused assertion covering the shared mapping.

Signed-off-by: liminfei-amd <91481003+liminfei-amd@users.noreply.github.com>

Author: liminfei-amd

comfy_kitchen/backends/triton/quantization.py

@@ -906,11 +906,13 @@ def _prune_int8_autotune_configs(configs, named_args, **kwargs):
 
 _INT8_PRUNE_CONFIGS_BY = {'early_config_prune': _prune_int8_autotune_configs}
 
-@triton.autotune(
-    configs=_INT8_MATMUL_CONFIGS,
-    key=['m', 'n', 'k', 'device_index'],
-    prune_configs_by=_INT8_PRUNE_CONFIGS_BY,
-)
+_INT8_AUTOTUNE_KWARGS = {
+    'configs': _INT8_MATMUL_CONFIGS,
+    'key': ['m', 'n', 'k', 'device_index'],
+    'prune_configs_by': _INT8_PRUNE_CONFIGS_BY,
+}
+
+@triton.autotune(**_INT8_AUTOTUNE_KWARGS)
 @triton.jit
 def _int8_matmul_dequant_kernel(
     # Pointers
@@ -982,11 +984,7 @@ def _int8_matmul_dequant_kernel(
     c_mask = (offs_am[:, None] < m) & (offs_bn[None, :] < n)
     tl.store(c_ptrs, c, mask=c_mask)
 
-@triton.autotune(
-    configs=_INT8_MATMUL_CONFIGS,
-    key=['m', 'n', 'k', 'device_index'],
-    prune_configs_by=_INT8_PRUNE_CONFIGS_BY,
-)
+@triton.autotune(**_INT8_AUTOTUNE_KWARGS)
 @triton.jit
 def _int8_matmul_dequant_per_row_kernel(
     # Pointers

tests/test_int8.py

@@ -204,6 +204,12 @@ def test_int8_autotune_pool_creation_does_not_probe_devices(monkeypatch):
 def test_int8_autotune_cache_key_includes_device():
     from comfy_kitchen.backends.triton import quantization
 
+    assert quantization._INT8_AUTOTUNE_KWARGS == {
+        "configs": quantization._INT8_MATMUL_CONFIGS,
+        "key": ["m", "n", "k", "device_index"],
+        "prune_configs_by": quantization._INT8_PRUNE_CONFIGS_BY,
+    }
+
     assert "device_index" in quantization._int8_matmul_dequant_kernel.keys
     assert "device_index" in quantization._int8_matmul_dequant_per_row_kernel.keys