Re-use common_includes to propagate shared function

Author: Mousius

python/tvm/topi/arm_cpu/mprofile/dsp/micro_kernel/multi_channel_convolve.py

@@ -23,7 +23,7 @@
 import textwrap
 
 from tvm import te, tir
-from .common import num_simd_lanes_per_word
+from .common import num_simd_lanes_per_word, common_includes
 
 
 def _get_func_name(in_dtype, tensor_w, channels, kernel_h, kernel_w, suffix):
@@ -107,10 +107,8 @@ def multi_channel_convolve_impl(in_dtype, *args) -> str:
 def _quad_int8_channel_convolve_impl(_tensor_h, tensor_w, channels, kernel_h, kernel_w, suffix):
     return textwrap.dedent(
         (
-            f"""
-        #include <stdint.h>
-        #include <arm_acle.h>
-
+            common_includes
+            + f"""
         // __SXTB16(_ROR(X, Y)) is combined into one assembly instruction
 
         #define TVMGEN_QUAD_INT8_CHANNEL_REARRANGE_SUM_DSP( \
@@ -172,7 +170,8 @@ def _quad_int8_channel_convolve_impl(_tensor_h, tensor_w, channels, kernel_h, ke
 def _dual_int16_channel_convolve_impl(_tensor_h, tensor_w, channels, kernel_h, kernel_w, suffix):
     return textwrap.dedent(
         (
-            f"""
+            common_includes
+            + f"""
         #include <stdint.h>
 
         /* We do four channels at once to get this speed boost. */

python/tvm/topi/arm_cpu/mprofile/dsp/micro_kernel/tensordot.py

@@ -23,7 +23,7 @@
 
 from tvm import te, tir
 
-from .common import num_simd_lanes_per_word
+from .common import num_simd_lanes_per_word, common_includes
 
 
 def _get_func_name(in_dtype, tensor_h, jump, tensor_w, suffix):
@@ -123,23 +123,8 @@ def tensordot_impl(in_dtype: str, tensor_h: int, jump: int, tensor_w: int, suffi
     function_name = _get_func_name(in_dtype, tensor_h, jump, tensor_w, suffix)
     return textwrap.dedent(
         (
-            f"""
-        #include <stdint.h>
-        #include <arm_acle.h>
-
-        #ifndef ARM_CPU_ROR_EXISTS
-        #define ARM_CPU_ROR_EXISTS
-        __attribute__((always_inline)) uint32_t __ror(uint32_t op1, uint32_t op2)
-        {{
-            op2 %= 32U;
-            if (op2 == 0U)
-            {{
-                return op1;
-            }}
-            return (op1 >> op2) | (op1 << (32U - op2));
-        }}
-        #endif
-
+            common_includes
+            + f"""
         #ifdef __cplusplus
         extern "C"
         #endif