Eliminate c10/cuda/CUDAException.h

Author: pearu

src/libtorchaudio/cuda_utils.h

@@ -5,10 +5,6 @@
 
 #include <cuda_runtime_api.h>
 
-// TODO: replace TA_CUDA_CHECK with STD_CUDA_CHECK after
-// https://github.com/pytorch/pytorch/pull/169385 has landed.
-#define TA_CUDA_CHECK(...) __VA_ARGS__
-
 namespace libtorchaudio::cuda {
 
 inline cudaStream_t getCurrentCUDAStream(

src/libtorchaudio/forced_align/gpu/compute.cu

@@ -1,9 +1,9 @@
 #include <libtorchaudio/cuda_utils.h>
 #include <libtorchaudio/utils.h>
 #include <torch/csrc/stable/library.h>
+#include <torch/csrc/stable/macros.h>
 #include <torch/headeronly/core/Dispatch_v2.h>
 #include <torch/headeronly/core/ScalarType.h>
-#include <c10/cuda/CUDAException.h>
 
 #include <cub/cub.cuh>
 #include <limits.h>
@@ -207,7 +207,7 @@ void forced_align_impl(
             backPtrBufferLen,
             torchaudio::packed_accessor32<scalar_t, 2>(alphas),
             torchaudio::packed_accessor32<int8_t, 2>(backPtrBuffer));
-    C10_CUDA_KERNEL_LAUNCH_CHECK();
+    STD_CUDA_KERNEL_LAUNCH_CHECK();
     ++backPtrBufferLen;
     if (backPtrBufferLen == kBackPtrBufferSize || t == T - 1) {
       libtorchaudio::cuda::synchronize(cpuDataTranferStream, device_index);
@@ -219,7 +219,7 @@ void forced_align_impl(
       // Copy ASYNC from GPU to CPU
       int64_t offset =
           static_cast<int64_t>(t + 1 - backPtrBufferLen) * S * sizeof(int8_t);
-      C10_CUDA_CHECK(cudaMemcpyAsync(
+      STD_CUDA_CHECK(cudaMemcpyAsync(
           static_cast<int8_t*>(backPtrCpu.data_ptr()) + offset,
           bufferCopy.data_ptr(),
           backPtrBufferLen * S * sizeof(int8_t),

src/libtorchaudio/iir_cuda.cu

@@ -1,8 +1,8 @@
 #include <libtorchaudio/utils.h>
 #include <torch/csrc/stable/accelerator.h>
+#include <torch/csrc/stable/macros.h>
 #include <torch/headeronly/core/Dispatch_v2.h>
 #include <torch/headeronly/core/ScalarType.h>
-#include <c10/cuda/CUDAException.h>
 
 using torch::headeronly::ScalarType;
 using torch::stable::Tensor;
@@ -74,7 +74,7 @@ Tensor cuda_lfilter_core_loop(
             torchaudio::packed_accessor_size_t<scalar_t, 3>(in),
             torchaudio::packed_accessor_size_t<scalar_t, 2>(a_flipped),
             torchaudio::packed_accessor_size_t<scalar_t, 3>(padded_out)));
-        C10_CUDA_KERNEL_LAUNCH_CHECK();
+        STD_CUDA_KERNEL_LAUNCH_CHECK();
         }), AT_FLOATING_TYPES);
   return padded_out;
 }

src/libtorchaudio/stable/ops.h

@@ -106,7 +106,7 @@ inline Tensor cuda(const Tensor& self, int32_t cuda_index) {
 inline Tensor new_zeros(
     const Tensor& self,
     std::vector<int64_t> size,
-    std::optional<c10::ScalarType> dtype = std::nullopt,
+    std::optional<torch::headeronly::ScalarType> dtype = std::nullopt,
     std::optional<Layout> layout = std::nullopt,
     std::optional<Device> device = std::nullopt,
     std::optional<bool> pin_memory = std::nullopt) {