Added comments

microsoft · May 16, 2024 · dfc8c6c · dfc8c6c
1 parent d288d36
commit dfc8c6c
Show file tree

Hide file tree

Showing 8 changed files with 8 additions and 0 deletions.
diff --git a/csrc/includes/cublas_wrappers.h b/csrc/includes/cublas_wrappers.h
@@ -30,6 +30,7 @@ int cublas_gemm_ex(cublasHandle_t handle,
                    const float* A,
                    const float* B,
                    float* C,
+//TODO HIP: Remove backward compatibility for torch<=2.0 in future
 #if defined(__HIP_PLATFORM_AMD__) && TORCH_VERSION_MAJOR <= 2 && TORCH_VERSION_MINOR <=0
                    rocblas_gemm_algo algo = rocblas_gemm_algo_standard);
 #else

diff --git a/csrc/includes/feed_forward.h b/csrc/includes/feed_forward.h
@@ -48,6 +48,7 @@ class FeedForward {
                        weights,
                        input_ptr,
                        out,
+//TODO HIP: Remove backward compatibility for torch<=2.0 in future
 #if defined(__HIP_PLATFORM_AMD__) && TORCH_VERSION_MAJOR <= 2 && TORCH_VERSION_MINOR <=0
                        rocblas_gemm_algo(config_.gemm_algos[0]));
 #else

diff --git a/csrc/includes/gemm_test.h b/csrc/includes/gemm_test.h
@@ -67,6 +67,7 @@ class GemmTest {
                            B,
                            A,
                            C,
+//TODO HIP: Remove backward compatibility for torch<=2.0 in future
 #if defined(__HIP_PLATFORM_AMD__) && TORCH_VERSION_MAJOR <= 2 && TORCH_VERSION_MINOR <=0			   
                            static_cast<rocblas_gemm_algo>(algo));
 #else

diff --git a/csrc/includes/strided_batch_gemm.h b/csrc/includes/strided_batch_gemm.h
@@ -77,6 +77,7 @@ class StridedBatchGemm {
                                     stride_b,
                                     stride_c,
                                     bsz,
+//TODO HIP: Remove backward compatibility for torch<=2.0 in future
 #if defined(__HIP_PLATFORM_AMD__) && TORCH_VERSION_MAJOR <= 2 && TORCH_VERSION_MINOR <=0
                                     rocblas_gemm_algo(_config.gemm_algos[0]));
 #else

diff --git a/csrc/transformer/cublas_wrappers.cu b/csrc/transformer/cublas_wrappers.cu
@@ -5,6 +5,7 @@
 
 #include "cublas_wrappers.h"
 
+//TODO HIP: Remove backward compatibility for torch<=2.0 in future
 #if defined(__HIP_PLATFORM_AMD__) && TORCH_VERSION_MAJOR <= 2 && TORCH_VERSION_MINOR <=0
 int cublas_gemm_ex(rocblas_handle handle,
                    rocblas_operation transa,

diff --git a/csrc/transformer/inference/csrc/pt_binding.cpp b/csrc/transformer/inference/csrc/pt_binding.cpp
@@ -163,6 +163,7 @@ at::Tensor einsum_sec_sm_ecm(at::Tensor& Q, at::Tensor& W)
                    (T*)W.data_ptr(),
                    (T*)Q.data_ptr(),
                    (T*)O.data_ptr(),
+//TODO HIP: Remove backward compatibility for torch<=2.0 in future
 #if defined(__HIP_PLATFORM_AMD__) && TORCH_VERSION_MAJOR <= 2 && TORCH_VERSION_MINOR <=0
                    rocblas_gemm_algo_standard);
 #else

diff --git a/csrc/transformer/inference/includes/inference_cublas_wrappers.h b/csrc/transformer/inference/includes/inference_cublas_wrappers.h
@@ -18,6 +18,7 @@
 #endif
 #include <stdio.h>
 
+//TODO HIP: Remove backward compatibility for torch<=2.0 in future
 #if defined(__HIP_PLATFORM_AMD__) && TORCH_VERSION_MAJOR <= 2 && TORCH_VERSION_MINOR <=0
 int cublas_gemm_ex(rocblas_handle handle,
                    rocblas_operation transa,

diff --git a/deepspeed/inference/v2/kernels/core_ops/blas_kernels/blas_utils.h b/deepspeed/inference/v2/kernels/core_ops/blas_kernels/blas_utils.h
@@ -55,6 +55,7 @@ class BlasContext {
 
 enum class BlasType { FP32, FP16, BF16 };
 
+//TODO HIP: Remove backward compatibility for torch<=2.0 in future
 #if defined(__HIP_PLATFORM_AMD__) && TORCH_VERSION_MAJOR <= 2 && TORCH_VERSION_MINOR <=0
 rocblas_operation get_trans_op(bool do_trans)
 {
-Original file line number
+Diff line change
@@ Expand Up / @@ -67,6 +67,7 @@ class GemmTest { @@
                                B,
                                A,
                                C,
+    //TODO HIP: Remove backward compatibility for torch<=2.0 in future
     #if defined(__HIP_PLATFORM_AMD__) && TORCH_VERSION_MAJOR <= 2 && TORCH_VERSION_MINOR <=0
                                static_cast<rocblas_gemm_algo>(algo));
     #else
@@ Expand Down @@