SDNQ disable fp8 mm with pre-quants on unsupported gpus

modules/sdnq/common.py

@@ -6,7 +6,7 @@ import torch
 
 from modules import shared, devices
 
-sdnq_version = "0.1.7"
+sdnq_version = "0.1.8"
 
 dtype_dict = {
     ### Integers
@@ -335,6 +335,17 @@ def check_torch_compile(): # dynamo can be disabled after startup
     return use_torch_compile and not torch._dynamo.config.disable # pylint: disable=protected-access
 
 
+if os.environ.get("SDNQ_ALLOW_FP8_MM", None) is None:
+    if devices.backend == "cuda":
+        is_fp8_mm_supported = bool(torch.cuda.get_device_capability(devices.device) >= (8,9))
+    elif devices.backend == "rocm":
+        gfx_version = devices.get_hip_agent().gfx_version
+        is_fp8_mm_supported = bool(gfx_version >= 0x1200 or (gfx_version >= 0x940 and gfx_version < 0x1000))
+    else:
+        is_fp8_mm_supported = False
+else:
+    is_fp8_mm_supported = os.environ.get("SDNQ_ALLOW_FP8_MM", "0").lower() not in {"0", "false", "no"}
+
 if os.environ.get("SDNQ_USE_TENSORWISE_FP8_MM", None) is None:
     # row-wise FP8 only exist on H100 hardware, sdnq will use software row-wise with tensorwise hardware with this setting
     use_tensorwise_fp8_matmul = bool(devices.backend != "cuda" or (devices.backend == "cuda" and torch.cuda.get_device_capability(devices.device) < (9,0)))

modules/sdnq/loader.py

@@ -3,7 +3,7 @@ import json
 import torch
 from diffusers.models.modeling_utils import ModelMixin
 
-from .common import dtype_dict, use_tensorwise_fp8_matmul, check_torch_compile, conv_types, linear_types
+from .common import dtype_dict, use_tensorwise_fp8_matmul, is_fp8_mm_supported, check_torch_compile, conv_types, linear_types
 from .quantizer import SDNQConfig, sdnq_post_load_quant, prepare_weight_for_matmul, prepare_svd_for_matmul, get_quant_args_from_config
 from .forward import get_forward_func
 from .file_loader import load_files
@@ -172,6 +172,10 @@ def apply_sdnq_options_to_module(model, dtype: torch.dtype | None = None, dequan
         if hasattr(module, "sdnq_dequantizer"):
             layer_class_name = module.original_class.__name__
             current_use_quantized_matmul = use_quantized_matmul
+
+            if not is_fp8_mm_supported and module.sdnq_dequantizer.quantized_matmul_dtype in {"fp8", "float8_e4m3fn"}:
+                current_use_quantized_matmul = False
+
             if current_use_quantized_matmul:
                 if layer_class_name in conv_types:
                     output_channel_size, channel_size = module.sdnq_dequantizer.original_shape[:2]