Remove op_registration.h (#241)

Summary:
Pull Request resolved: #241

As titled.

Reviewed By: henrylhtsang

Differential Revision: D97536913

Author: cthi

csrc/conv/conv_ops.cpp

@@ -7,7 +7,6 @@
  */
 
 #include <mslk/conv/conv.h> // @manual
-#include <mslk/utils/torch/op_registration.h> // @manual
 #include <torch/library.h>
 
 namespace mslk::conv {

csrc/gemm/gemm_ops.cpp

@@ -8,7 +8,6 @@
 
 #include <mslk/gemm/gemm.h> // @manual
 #include <mslk/gemm/gemm_torch.h> // @manual
-#include <mslk/utils/torch/op_registration.h> // @manual
 #include <torch/library.h>
 
 namespace mslk::gemm {

csrc/moe/moe_ops.cpp

@@ -8,7 +8,6 @@
 
 #include <ATen/ATen.h>
 #include <mslk/moe/moe.h> // @manual
-#include <mslk/utils/torch/op_registration.h> // @manual
 #include <torch/library.h>
 #include <cstdint>
 #include <optional>
@@ -31,9 +30,9 @@ TORCH_LIBRARY_FRAGMENT(mslk, m) {
 }
 
 TORCH_LIBRARY_IMPL(mslk, CUDA, m) {
-  DISPATCH_TO_CUDA("index_shuffling", index_shuffling_torch);
+  m.impl("index_shuffling", index_shuffling_torch);
 #ifndef USE_ROCM
-  DISPATCH_TO_CUDA("scatter_add_along_first_dim", scatter_add_along_first_dim);
+  m.impl("scatter_add_along_first_dim", scatter_add_along_first_dim);
 #endif
 }
 
@@ -62,10 +61,9 @@ void scatter_add_along_first_dim_meta(
 }
 
 TORCH_LIBRARY_IMPL(mslk, Meta, m) {
-  DISPATCH_TO_META("index_shuffling", index_shuffling_torch_meta);
+  m.impl("index_shuffling", index_shuffling_torch_meta);
 #ifndef USE_ROCM
-  DISPATCH_TO_META(
-      "scatter_add_along_first_dim", scatter_add_along_first_dim_meta);
+  m.impl("scatter_add_along_first_dim", scatter_add_along_first_dim_meta);
 #endif
 }
 

csrc/quantize/quantize_ops.cpp

@@ -7,7 +7,6 @@
  */
 
 #include <mslk/quantize/quantize.h> // @manual
-#include <mslk/utils/torch/op_registration.h> // @manual
 #include <torch/library.h>
 
 namespace mslk::quantize {
@@ -31,11 +30,10 @@ TORCH_LIBRARY_FRAGMENT(mslk, m) {
 }
 
 TORCH_LIBRARY_IMPL(mslk, CUDA, m) {
-  DISPATCH_TO_CUDA("per_tensor_quantize_i8", per_tensor_quantize_i8);
-  DISPATCH_TO_CUDA(
-      "per_tensor_dynamic_quantize_i8", per_tensor_dynamic_quantize_i8);
-  DISPATCH_TO_CUDA("get_fp8_per_tensor_scale", get_fp8_per_tensor_scale);
-  DISPATCH_TO_CUDA(
+  m.impl("per_tensor_quantize_i8", per_tensor_quantize_i8);
+  m.impl("per_tensor_dynamic_quantize_i8", per_tensor_dynamic_quantize_i8);
+  m.impl("get_fp8_per_tensor_scale", get_fp8_per_tensor_scale);
+  m.impl(
       "quantize_fp8_per_tensor_fixed_scale",
       quantize_fp8_per_tensor_fixed_scale);
 }