fix include

Signed-off-by: jiqing-feng <[email protected]>

Author: jiqing-feng

csrc/cpu_ops.cpp

@@ -1,5 +1,4 @@
 #include <BinSearch.h>
-#include <common.h>
 #include <cpu_ops.h>
 #include <thread>
 

csrc/cpu_ops.h

@@ -1,6 +1,7 @@
 #ifndef BITSANDBYTES_CPU_OPS_H
 #define BITSANDBYTES_CPU_OPS_H
 
+#include <common.h>
 #include <cstdint>
 #include <cstring>
 

csrc/ops.cu

@@ -5,7 +5,6 @@
 
 #include <BinSearch.h>
 #include <cassert>
-#include <common.h>
 #include <cub/device/device_scan.cuh>
 #include <kernels.cuh>
 #include <limits>

csrc/ops.cuh

@@ -11,6 +11,7 @@
 #include <iostream>
 #include <stdio.h>
 
+#include <common.h>
 #include <cublasLt.h>
 #include <cublas_v2.h>
 #include <cuda_fp16.h>
@@ -69,14 +70,6 @@ typedef enum Optimizer_t {
     ADEMAMIX = 6
 } Optimizer_t;
 
-typedef enum Transform_t {
-    ROW = 0,
-    COL = 1,
-    COL32 = 2,
-    COL_TURING = 3,
-    COL_AMPERE = 4,
-} Transform_t;
-
 typedef enum Funcs_t {
     FILL = 0,
     ARANGE = 1,

csrc/ops_hip.cuh

@@ -13,6 +13,7 @@
 #include <stdio.h>
 #include <unistd.h>
 
+#include <common.h>
 #include <functional>
 #include <hip/hip_fp16.h>
 #include <hip/hip_runtime_api.h>
@@ -71,14 +72,6 @@ typedef enum Optimizer_t {
     ADEMAMIX = 6,
 } Optimizer_t;
 
-typedef enum Transform_t {
-    ROW = 0,
-    COL = 1,
-    COL32 = 2,
-    COL_TURING = 3,
-    COL_AMPERE = 4,
-} Transform_t;
-
 typedef enum Funcs_t {
     FILL = 0,
     ARANGE = 1,

csrc/pythonInterface.cpp

@@ -17,7 +17,6 @@
 #include <xpu_ops.h>
 #endif
 #include <cpu_ops.h>
-#include <common.h>
 
 // Compatibility between HIP/CUDA APIs
 #if BUILD_HIP

csrc/xpu_ops.cpp

@@ -1,4 +1,3 @@
-#include <common.h>
 #include <xpu_kernels.h>
 #include <xpu_ops.h>
 

csrc/xpu_ops.h

@@ -2,6 +2,7 @@
 #define xpu_ops_H
 
 #include <assert.h>
+#include <common.h>
 #include <cstdint>
 #include <iostream>
 #include <stdio.h>