Fix unused-variable issues in caffe2 (pytorch#143639)

Summary: LLVM-15 has a warning `-Wunused-variable` which we treat as an error because it's so often diagnostic of a code issue. Unused variables can compromise readability or, worse, performance. This diff either (a) removes an unused variable and, possibly, it's associated code or (b) qualifies the variable with `[[maybe_unused]]`. - If you approve of this diff, please use the "Accept & Ship" button :-) Test Plan: Sandcastle Pull Request resolved: pytorch#143639 Approved by: https://github.com/kit1980, https://github.com/malfet, https://github.com/cyyever
abhishek-iitmadras · Dec 21, 2024 · 518b505 · 518b505
1 parent f443100
commit 518b505
Show file tree

Hide file tree

Showing 4 changed files with 6 additions and 6 deletions.
diff --git a/aten/src/ATen/cuda/CUDABlas.cpp b/aten/src/ATen/cuda/CUDABlas.cpp
@@ -1442,7 +1442,6 @@ void scaled_gemm(
 #if CUDA_VERSION >= 11080 || defined(USE_ROCM)
   const auto computeType = CUBLAS_COMPUTE_32F;
   const auto scaleType = CUDA_R_32F;
-  const int8_t fastAccuMode = use_fast_accum ? 1 : 0;
   const float alpha_val = 1.0;
   const float beta_val = 0.0;
   CuBlasLtMatmulDescriptor computeDesc(computeType, scaleType);
@@ -1454,7 +1453,7 @@ void scaled_gemm(
     computeDesc.setAttribute(CUBLASLT_MATMUL_DESC_D_SCALE_POINTER, result_scale_ptr);
   }
 #ifndef USE_ROCM
-  computeDesc.setAttribute(CUBLASLT_MATMUL_DESC_FAST_ACCUM, fastAccuMode);
+  computeDesc.setAttribute(CUBLASLT_MATMUL_DESC_FAST_ACCUM, use_fast_accum ? 1 : 0);
 #endif
   CuBlasLtMatrixLayout Adesc(ScalarTypeToCudaDataType(mat1_dtype), m, k, mat1_ld, transa == 't');
   CuBlasLtMatrixLayout Bdesc(ScalarTypeToCudaDataType(mat2_dtype), k, n, mat2_ld, transb == 't');

diff --git a/aten/src/ATen/native/ao_sparse/quantized/cpu/fbgemm_utils.cpp b/aten/src/ATen/native/ao_sparse/quantized/cpu/fbgemm_utils.cpp
@@ -10,7 +10,7 @@
 
 namespace ao::sparse {
 int register_linear_params() {
-  static auto register_linear_params =
+  [[maybe_unused]] static auto register_linear_params =
       torch::selective_class_<LinearPackedParamsBase>(
           "sparse", TORCH_SELECTIVE_CLASS("LinearPackedParamsBase"))
           .def_pickle(

diff --git a/aten/src/ATen/native/sparse/cuda/SparseCUDABlas.cpp b/aten/src/ATen/native/sparse/cuda/SparseCUDABlas.cpp
@@ -154,9 +154,9 @@ void _csrmm2(
 
 
   auto handle = at::cuda::getCurrentCUDASparseHandle();
-  cudaDeviceProp* prop = at::cuda::getCurrentDeviceProperties();
   // ALG1 is broken on SM89 as of CUDA 11.8+
 #if !defined(USE_ROCM)
+  cudaDeviceProp* prop = at::cuda::getCurrentDeviceProperties();
   auto default_alg = prop->major == 8 && prop->minor == 9 ? CUSPARSE_SPMM_CSR_ALG2 : CUSPARSE_SPMM_CSR_ALG1;
 #else
   auto default_alg = CUSPARSE_SPMM_CSR_ALG1;

diff --git a/aten/src/ATen/native/transformers/cuda/sdp_utils.cpp b/aten/src/ATen/native/transformers/cuda/sdp_utils.cpp
@@ -650,13 +650,14 @@ bool can_use_mem_efficient_attention(sdp_params const& params, bool debug) {
   return false;
 #endif
   // Constraints specific to mem efficient attention
-  constexpr auto greater_than_or_equal_sm80_mem_efficient_dtypes =
-      array_of<at::ScalarType>(at::kHalf, at::kFloat, at::kBFloat16);
   constexpr auto less_than_sm80_mem_efficient_dtypes =
       array_of<at::ScalarType>(at::kHalf, at::kFloat);
 #ifdef USE_ROCM
   constexpr auto aotriton_mem_efficient_dtypes =
       array_of<at::ScalarType>(at::kHalf, at::kFloat, at::kBFloat16);
+#else
+  constexpr auto greater_than_or_equal_sm80_mem_efficient_dtypes =
+      array_of<at::ScalarType>(at::kHalf, at::kFloat, at::kBFloat16);
 #endif
 
   //  Define gate functions that determine if a mem efficient kernel can be ran