[cuDNN][Quantization] Don't print when plan finalization fails in cuDNN quantization backend (#128177)
Similar in spirit to #125790, hopefully addresses failures seen for cuDNN 9.1 upgrade: #https://github.com/pytorch/pytorch/pull/128166
CC @nWEIdia @atalman
Pull Request resolved: https://github.com/pytorch/pytorch/pull/128177
Approved by: https://github.com/nWEIdia, https://github.com/Skylion007
diff --git a/aten/src/ATen/native/quantized/cudnn/BinaryOps.cpp b/aten/src/ATen/native/quantized/cudnn/BinaryOps.cpp
index 07ccc19..9e9e675 100644
--- a/aten/src/ATen/native/quantized/cudnn/BinaryOps.cpp
+++ b/aten/src/ATen/native/quantized/cudnn/BinaryOps.cpp
@@ -242,7 +242,7 @@
run(plan_desc);
execution_plan_cache[key] = plan_desc;
return quantized_output.view(orig_sizes);
- } catch (cudnn_frontend::cudnnException &e) {std::cout << "cudnn error:" << e.what() << std::endl;} catch(c10::CuDNNError &e) { std::cout << "other error" << e.what() << std::endl;}
+ } catch (cudnn_frontend::cudnnException &e) {} catch(c10::CuDNNError &e) {}
}
TORCH_CHECK(false, "Unable to find an engine to execute this computation in Quantized Add Cudnn");
diff --git a/aten/src/ATen/native/quantized/cudnn/Conv.cpp b/aten/src/ATen/native/quantized/cudnn/Conv.cpp
index 606d769..8823038 100644
--- a/aten/src/ATen/native/quantized/cudnn/Conv.cpp
+++ b/aten/src/ATen/native/quantized/cudnn/Conv.cpp
@@ -252,7 +252,7 @@
run(plan);
execution_plan_cache.emplace(key, plan);
return;
- } catch (cudnn_frontend::cudnnException &e) {std::cout << "cudnn error:" << e.what() << std::endl;} catch(c10::CuDNNError &e) { std::cout << "other error" << e.what() << std::endl;}
+ } catch (cudnn_frontend::cudnnException &e) {} catch(c10::CuDNNError &e) {}
}
TORCH_CHECK(false, "Unable to find an engine to execute this computation in Quantized Conv2D Cudnn");
diff --git a/aten/src/ATen/native/quantized/cudnn/Linear.cpp b/aten/src/ATen/native/quantized/cudnn/Linear.cpp
index d321959..54eb084 100644
--- a/aten/src/ATen/native/quantized/cudnn/Linear.cpp
+++ b/aten/src/ATen/native/quantized/cudnn/Linear.cpp
@@ -286,7 +286,7 @@
run(plan);
execution_plan_cache.emplace(key, plan);
return;
- } catch (cudnn_frontend::cudnnException &e) {std::cout << "cudnn error:" << e.what() << std::endl;} catch(c10::CuDNNError &e) { std::cout << "other error" << e.what() << std::endl;}
+ } catch (cudnn_frontend::cudnnException &e) {} catch(c10::CuDNNError &e) {}
}
TORCH_CHECK(false, "Unable to find an engine to execute this computation Quantized Linear Cudnn");
diff --git a/test/quantization/core/test_quantized_op.py b/test/quantization/core/test_quantized_op.py
index 5b86693..6671b66 100644
--- a/test/quantization/core/test_quantized_op.py
+++ b/test/quantization/core/test_quantized_op.py
@@ -4052,7 +4052,6 @@
use_channelwise=st.sampled_from([False])) # channelwise currently not supported for qlinear cudnn
@skipIfNoFBGEMM
@unittest.skipIf(not TEST_CUDNN, "cudnn is not enabled.")
- @unittest.skipIf(TEST_CUDNN and torch.backends.cudnn.version() == 90100, "expected failure on cuDNN 9.1.0")
@unittest.skipIf(not SM80OrLater, "requires sm80 or later.")
@unittest.skipIf(TEST_ROCM, "not supported on rocm.")
# TODO: check with yang regarding CUDNN flags
