refactor (csrc): Restructure C++ code organization to facilitate adding new kernels (#169)

This PR does not change any logic or functionality in the main branch.
It restructures the existing C++ code organization to facilitate the
addition of new kernels in subsequent PRs.

In general, each customized CUDA kernel involves modifying the following
three files at the backend:

1. Registering Python bindings and adding declarations in `ops.cc`.
1. Placing the host's kernel launch function in `x.cc`, where `x` is the
operator's name.
1. Implementing CUDA kernels in `x.cuh`.

Author: lcy-seso

.clang-format

@@ -3,24 +3,39 @@ UseTab: Never
 IndentWidth: 2
 ColumnLimit: 80
 
+AccessModifierOffset: -2
+
 # Force pointers to the type for C++.
 DerivePointerAlignment: false
 PointerAlignment: Left
 
-# Reordering #include statements can (and currently will) introduce errors
-SortIncludes: false
-
 # Style choices
 AlignConsecutiveAssignments: false
 AlignConsecutiveDeclarations: false
 IndentPPDirectives: BeforeHash
 
+SortIncludes: true
+IncludeBlocks: Regroup
 IncludeCategories:
-  - Regex:           '^<'
-    Priority:        4
-  - Regex:           '^"(llvm|llvm-c|clang|clang-c|mlir|mlir-c)/'
-    Priority:        3
-  - Regex:           '^"(qoda|\.\.)/'
-    Priority:        2
-  - Regex:           '.*'
-    Priority:        1
+  - Regex: '<([A-Za-z0-9\Q/-_\E])+>'
+    Priority: 4
+  - Regex: '<(catch2|boost)\/'
+    Priority: 3
+  - Regex: '<([A-Za-z0-9.\Q/-_\E])+>'
+    Priority: 2
+  - Regex: '"([A-Za-z0-9.\Q/-_\E])+"'
+    Priority: 1
+
+# If true, empty lines at the start of blocks are kept.
+KeepEmptyLinesAtTheStartOfBlocks: false
+AllowShortLoopsOnASingleLine: true
+AllowShortIfStatementsOnASingleLine: true
+Cpp11BracedListStyle: true
+# If true, always break after the template<...> of a template declaration.
+AlwaysBreakTemplateDeclarations: true
+# If false, a function declaration's or function definition's parameters will 
+# either all be on the same line or will have one line each.
+BinPackArguments: true
+BreakConstructorInitializersBeforeComma: true
+# The maximum number of consecutive empty lines to keep.
+MaxEmptyLinesToKeep: 1

.vscode/settings.json

@@ -1,3 +1,8 @@
 {
-    "yapf.args":["--style={based_on_s'tyle: google, column_limit: 80, indent_width: 4}"]
+    "yapf.args": [
+        "--style={based_on_s'tyle: google, column_limit: 80, indent_width: 4}"
+    ],
+    "files.associations": {
+        "optional": "cpp"
+    }
 }

csrc/common.h

@@ -3,14 +3,26 @@
 #pragma once
 
 #include <ATen/cuda/CUDAContext.h>
+#include <c10/cuda/CUDAGuard.h>
 #include <torch/extension.h>
 
 namespace vptq {
+
+#define CHECK_CUDA(x) \
+  TORCH_CHECK(x.device().is_cuda(), #x " must be a CUDA tensor")
+#define CHECK_CONTIGUOUS(x) \
+  TORCH_CHECK(x.is_contiguous(), #x " must be contiguous")
+#define CHECK_INPUT(x) \
+  CHECK_CUDA(x);       \
+  CHECK_CONTIGUOUS(x)
+
+#define gpuErrchk(ret) gpuAssert((ret), __FILE__, __LINE__);
+
 class OptionalCUDAGuard {
   int set_device_ = -1;
   int current_device_ = -1;
 
- public:
+public:
   OptionalCUDAGuard(int device) : set_device_(device) {
     cudaError_t err = cudaGetDevice(&current_device_);
     std::stringstream ss;
@@ -32,13 +44,12 @@ class OptionalCUDAGuard {
   }
 };
 
-#define gpuErrchk(ret) gpuAssert((ret), __FILE__, __LINE__);
-
 inline void gpuAssert(cudaError_t code, const char* file, int line) {
   if (code != cudaSuccess) {
     fprintf(stderr, "GPUassert: %s %s %d\n", cudaGetErrorString(code), file,
             line);
     TORCH_CHECK(false, cudaGetErrorString(code));
   }
 }
+
 }  // namespace vptq

csrc/utils.cuh csrc/cuda_utils.cuh

@@ -2,27 +2,47 @@
 // Licensed under the MIT License.
 #pragma once
 
+#include <ATen/cuda/CUDAContext.h>
+
 #if defined(USE_ROCM)
-  #include <hip/hip_fp16.h>
   #include <hip/hip_bf16.h>
+  #include <hip/hip_fp16.h>
 
   #define VPTQ_LDG(arg) __ldg(arg)
   #define SHFL_DOWN(val, offset) __shfl_down(val, offset)
   #define WARP_SIZE warpSize
+
 typedef __hip_bfloat162 __bfloat162;
 typedef __hip_bfloat16 __bfloat16;
 #else
-  #include <cuda_fp16.h>
   #include <cuda_bf16.h>
+  #include <cuda_fp16.h>
 
   #define WARP_SIZE 32
   #define VPTQ_LDG(arg) *(arg)
   #define SHFL_DOWN(val, offset) __shfl_down_sync(0xffffffff, val, offset)
+
 typedef __nv_bfloat162 __bfloat162;
 typedef __nv_bfloat16 __bfloat16;
 #endif
 
 namespace vptq {
+
+template <typename T>
+struct C10ToNvType {
+  typedef __bfloat16 type;
+};
+
+template <>
+struct C10ToNvType<c10::Half> {
+  typedef __half type;
+};
+
+template <>
+struct C10ToNvType<float> {
+  typedef float type;
+};
+
 namespace cuda {
 
 constexpr int kBlockSize = 256;
@@ -243,4 +263,5 @@ __device__ __half operator*(const __half& a, const __half& b) {
   return __hmul(a, b);
 }
 #endif
+
 }  // namespace vptq