Cortex-M: remove extractScalarToInt helpers, use static_cast directly (#17543)

### Summary
Since #16322 changed all Scalar parameters to int64_t, the
extractScalarToInt32 and extractScalarToInt helpers in
cortex_m_ops_common.h are unnecessary indirection — they called
Scalar::to<int64_t>() then cast, but the values are already int64_t.

Replace all call sites with direct static_cast and remove the helpers,
the unused Scalar type alias, and the scalar_utils.h include. Also
update the quantized_mul Python schema in operators.py to use int
instead of Scalar, matching the C++ kernel and operators.yaml.

Additionally remove a duplicate kernel_includes.h include from
cortex_m_ops_common.h.

### Test plan
```
pytest backends/cortex_m/test/
```

cc @digantdesai @SS-JIA @freddan80 @per @zingo @oscarandersson8218
@mansnils @Sebastian-Larsson @robell @psiddh @AdrianLundell

Co-authored-by: Claude <noreply@anthropic.com>

Author: rascani

backends/cortex_m/ops/cortex_m_ops_common.h

@@ -12,10 +12,8 @@
 #include <executorch/kernels/portable/cpu/util/broadcast_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 
-#include <executorch/kernels/portable/cpu/scalar_utils.h>
 #include <executorch/kernels/portable/cpu/util/elementwise_util.h>
 #include <executorch/kernels/portable/cpu/util/kernel_ops_util.h>
-#include <executorch/runtime/kernel/kernel_includes.h>
 #include <executorch/runtime/platform/assert.h>
 
 #include <limits>
@@ -27,7 +25,6 @@ extern "C" {
 
 using Tensor = torch::executor::Tensor;
 using ScalarType = executorch::aten::ScalarType;
-using Scalar = torch::executor::Scalar;
 using Error = executorch::runtime::Error;
 using Int64ArrayRef = executorch::aten::ArrayRef<int64_t>;
 using KernelRuntimeContext = torch::executor::KernelRuntimeContext;
@@ -391,19 +388,3 @@ inline Error resize_to_broadcast_target_size(
   return executorch::runtime::resize_tensor(
       output, {expected_output_size, expected_output_dim});
 }
-
-/**
- * Convert Scalar to CMSIS-NN int32 format
- * For multipliers, zero_points, etc. from quantize_multiplier_aot
- */
-inline int32_t extractScalarToInt32(const Scalar& scalar_value) {
-  return static_cast<int32_t>(scalar_value.to<int64_t>());
-}
-
-/**
- * Convert Scalar to CMSIS-NN int format
- * For shift values from quantize_multiplier_aot
- */
-inline int extractScalarToInt(const Scalar& scalar_value) {
-  return static_cast<int>(scalar_value.to<int64_t>());
-}

backends/cortex_m/ops/op_quantized_add.cpp

@@ -60,15 +60,15 @@ Tensor& quantized_add_out(
       "quantized_add_out: input1_int8.sizes() = %zu",
       input1_int8.sizes().size());
 
-  int32_t zp1 = extractScalarToInt32(input1_zero_point);
-  int32_t input1_mult = extractScalarToInt32(input1_multiplier);
-  int input1_shift_val = extractScalarToInt(input1_shift);
-  int32_t zp2 = extractScalarToInt32(input2_zero_point);
-  int32_t input2_mult = extractScalarToInt32(input2_multiplier);
-  int input2_shift_val = extractScalarToInt(input2_shift);
-  int32_t out_zp = extractScalarToInt32(output_zero_point);
-  int32_t output_mult = extractScalarToInt32(output_multiplier);
-  int output_shift_val = extractScalarToInt(output_shift);
+  int32_t zp1 = static_cast<int32_t>(input1_zero_point);
+  int32_t input1_mult = static_cast<int32_t>(input1_multiplier);
+  int input1_shift_val = static_cast<int>(input1_shift);
+  int32_t zp2 = static_cast<int32_t>(input2_zero_point);
+  int32_t input2_mult = static_cast<int32_t>(input2_multiplier);
+  int input2_shift_val = static_cast<int>(input2_shift);
+  int32_t out_zp = static_cast<int32_t>(output_zero_point);
+  int32_t output_mult = static_cast<int32_t>(output_multiplier);
+  int output_shift_val = static_cast<int>(output_shift);
   int8_t* input1_ptr = input1_int8.data_ptr<int8_t>();
   int8_t* input2_ptr = input2_int8.data_ptr<int8_t>();
 

backends/cortex_m/ops/op_quantized_mul.cpp

@@ -61,11 +61,11 @@ Tensor& quantized_mul_out(
   // Extract quantization parameters
   int8_t* input1_ptr = input1_int8.data_ptr<int8_t>();
   int8_t* input2_ptr = input2_int8.data_ptr<int8_t>();
-  int32_t zp1 = extractScalarToInt32(input1_zero_point);
-  int32_t zp2 = extractScalarToInt32(input2_zero_point);
-  const int32_t out_zp = extractScalarToInt32(output_zero_point);
-  const int32_t output_mult = extractScalarToInt32(output_multiplier);
-  const int32_t output_shift_val = extractScalarToInt32(output_shift);
+  int32_t zp1 = static_cast<int32_t>(input1_zero_point);
+  int32_t zp2 = static_cast<int32_t>(input2_zero_point);
+  const int32_t out_zp = static_cast<int32_t>(output_zero_point);
+  const int32_t output_mult = static_cast<int32_t>(output_multiplier);
+  const int32_t output_shift_val = static_cast<int32_t>(output_shift);
 
   int32_t muls_per_loop = 0;
 

backends/cortex_m/ops/operators.py

@@ -195,15 +195,15 @@ def quantized_add_impl(
 # ===================================================================
 lib.define(
     "quantized_mul("
-    "Tensor self, Scalar self_zero_point, "
-    "Tensor other, Scalar other_zero_point, "
-    "Scalar output_zero_point, Scalar output_multiplier, Scalar output_shift) -> Tensor"
+    "Tensor self, int self_zero_point, "
+    "Tensor other, int other_zero_point, "
+    "int output_zero_point, int output_multiplier, int output_shift) -> Tensor"
 )
 lib.define(
     "quantized_mul.out("
-    "Tensor self, Scalar self_zero_point, "
-    "Tensor other, Scalar other_zero_point, "
-    "Scalar output_zero_point, Scalar output_multiplier, Scalar output_shift, "
+    "Tensor self, int self_zero_point, "
+    "Tensor other, int other_zero_point, "
+    "int output_zero_point, int output_multiplier, int output_shift, "
     "*, Tensor(a!) out) -> Tensor(a!)"
 )