Merge pull request #4 from dijopaul/rushikesh_code

integrated operators div and floor_div nad sqrt

Author: dijopaul

backends/cadence/aot/functions_hifi.yaml

@@ -101,6 +101,11 @@
   kernels:
     - arg_meta: null
       kernel_name: torch::executor::where_out
+      
+- op: rsqrt.out
+  kernels:
+    - arg_meta: null
+      kernel_name: torch::executor::rsqrt_out        
 
 # custom ops
 - func: cadence::quantize_per_tensor.out(Tensor input, float scale, int zero_point, int quant_min, int quant_max, ScalarType dtype, *, Tensor(a!) out) -> Tensor(a!)

backends/cadence/hifi/kernels/CMakeLists.txt

@@ -9,6 +9,8 @@ add_library(
   cadence_kernels
   kernels.cpp
   ${EXECUTORCH_ROOT}/backends/cadence/hifi/third-party/nnlib/matmul_asym8uxasym8u_asym8u.cpp
+  ${EXECUTORCH_ROOT}/backends/cadence/hifi/third-party/nnlib/xa_nn_elm_div_broadcast_f32.c
+  ${EXECUTORCH_ROOT}/backends/cadence/hifi/third-party/nnlib/xa_nn_elm_floor_div_broadcast_f32.c
 )
 
 target_include_directories(

backends/cadence/hifi/kernels/kernels.h

@@ -14,6 +14,28 @@
 
 /* For NNLIB APIs */
 #include "xa_nnlib_kernels_api.h"
+
+extern "C" WORD32 xa_nn_elm_floor_div_f32xf32_f32(
+                                    FLOAT32 * __restrict__ p_out,
+                                    const FLOAT32 * __restrict__ p_inp1,
+                                    const FLOAT32 * __restrict__ p_inp2,
+                                    WORD32 num_elm);
+                                    
+extern "C" WORD32 xa_nn_elm_floor_div_broadcast_4D_f32xf32_f32(
+                                    FLOAT32 * __restrict__ p_out,
+                                    const WORD32 *const p_out_shape,
+                                    const FLOAT32 * __restrict__ p_inp1,
+                                    const WORD32 *const p_inp1_shape,
+                                    const FLOAT32 * __restrict__ p_inp2,
+                                    const WORD32 *const p_inp2_shape);
+                                    
+extern "C" WORD32 xa_nn_elm_div_broadcast_4D_f32xf32_f32(
+                                    FLOAT32 * __restrict__ p_out,
+                                    const WORD32 *const p_out_shape,
+                                    const FLOAT32 * __restrict__ p_inp1,
+                                    const WORD32 *const p_inp1_shape,
+                                    const FLOAT32 * __restrict__ p_inp2,
+                                    const WORD32 *const p_inp2_shape);
 
 
 namespace impl {

backends/cadence/hifi/operators/CMakeLists.txt

@@ -35,7 +35,8 @@ set(_aten_ops__srcs
     "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_bmm.cpp"
     "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_cat.cpp"
     "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_clone.cpp"
-    "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_div.cpp"
+    "${EXECUTORCH_ROOT}/backends/cadence/hifi/operators/op_div.cpp"
+    "${EXECUTORCH_ROOT}/backends/cadence/hifi/operators/op_rsqrt.cpp"
     "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_mul.cpp"
     "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_permute_copy.cpp"
     "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_sigmoid.cpp"
@@ -44,7 +45,8 @@ set(_aten_ops__srcs
     "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_split_with_sizes_copy.cpp"
     "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_sub.cpp"
     "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_to_copy.cpp"
-    "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_where.cpp")
+    "${EXECUTORCH_ROOT}/kernels/portable/cpu/op_where.cpp"
+    "${EXECUTORCH_ROOT}/kernels/portable/cpu/pattern/unary_ufunc_realhb_to_floath.cpp")
 add_library(aten_ops_cadence ${_aten_ops__srcs})
 target_link_libraries(aten_ops_cadence PUBLIC executorch)
 target_link_libraries(aten_ops_cadence PRIVATE cadence_kernels)