Skip to content

Cortex-M backend: Update Scalar types to be int #16322

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
SaoirseARM wants to merge 2 commits into
base: main
Choose a base branch
from
+53 −57
Open

Cortex-M backend: Update Scalar types to be int #16322

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
df21577
Cortex-M backend: Update Scalar types to be int
SaoirseARM Dec 4, 2025
5dc1aa4
Merge branch 'main' into toupstream/scalars
SaoirseARM Dec 19, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
42 changes: 19 additions & 23 deletions backends/cortex_m/ops/cortex_m_ops_common.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,26 +74,22 @@ inline void validate_cmsis_nn_tensor_requirements(
}

inline void validate_single_quant_params(
const Scalar& zero_point,
const Scalar& multiplier,
const Scalar& shift,
const int64_t zero_point,
const int64_t multiplier,
const int64_t shift,
const char* param_name) {
int64_t zp_val = zero_point.to<int64_t>();
int64_t mult_val = multiplier.to<int64_t>();
int64_t shift_val = shift.to<int64_t>();

ET_CHECK_MSG(
mult_val >= std::numeric_limits<int32_t>::min() &&
mult_val <= std::numeric_limits<int32_t>::max(),
multiplier >= std::numeric_limits<int32_t>::min() &&
multiplier <= std::numeric_limits<int32_t>::max(),
"%s multiplier must be in int32 range [Value: %d]",
param_name,
mult_val);
multiplier);

ET_CHECK_MSG(
shift_val >= -31 && shift_val <= 31,
shift >= -31 && shift <= 31,
"%s shift must be in range [-31, 31] [Value: %d]",
param_name,
shift_val);
shift);
}

/**
Expand All @@ -108,15 +104,15 @@ inline void validate_single_quant_params(
* Raises errors via ET_KERNEL_CHECK if any check fails.
*/
inline void validate_quantization_params(
const Scalar& zero_point1,
const Scalar& multiplier1,
const Scalar& shift1,
const Scalar& zero_point2,
const Scalar& multiplier2,
const Scalar& shift2,
const Scalar& output_zero_point,
const Scalar& output_multiplier,
const Scalar& output_shift,
const int64_t zero_point1,
const int64_t multiplier1,
const int64_t shift1,
const int64_t zero_point2,
const int64_t multiplier2,
const int64_t shift2,
const int64_t output_zero_point,
const int64_t output_multiplier,
const int64_t output_shift,
Tensor& output) {
validate_single_quant_params(
zero_point1, multiplier1, shift1, "Single quant Input1");
Expand Down Expand Up @@ -171,8 +167,8 @@ inline bool is_channel_broadcast(const Tensor& tensor1, const Tensor& tensor2) {
// multiplier: Range {ARM_NN_Q31_MIN + 1, Q32_MAX}
// shift : Range {-31, 30}
inline bool validate_per_channel_quant_params(
const int32_t* multipliers,
const int32_t* shifts,
const IntArrayRef multipliers,
const IntArrayRef shifts,
int num_channels) {
for (int i = 0; i < num_channels; ++i) {
// Multiplier: {ARM_NN_Q31_MIN + 1, ARM_NN_Q31_MAX}
Expand Down
18 changes: 9 additions & 9 deletions backends/cortex_m/ops/op_quantized_add.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,16 +21,16 @@ using KernelRuntimeContext = torch::executor::KernelRuntimeContext;
Tensor& quantized_add_out(
KernelRuntimeContext& context,
const Tensor& input1_int8,
const Scalar& input1_zero_point,
const Scalar& input1_multiplier,
const Scalar& input1_shift,
const int64_t input1_zero_point,
const int64_t input1_multiplier,
const int64_t input1_shift,
const Tensor& input2_int8,
const Scalar& input2_zero_point,
const Scalar& input2_multiplier,
const Scalar& input2_shift,
const Scalar& output_zero_point,
const Scalar& output_multiplier,
const Scalar& output_shift,
const int64_t input2_zero_point,
const int64_t input2_multiplier,
const int64_t input2_shift,
const int64_t output_zero_point,
const int64_t output_multiplier,
const int64_t output_shift,
Tensor& out) {
// Validate tensor types and dim order
bool channel_broadcast = is_channel_broadcast(input1_int8, input2_int8);
Expand Down
6 changes: 3 additions & 3 deletions backends/cortex_m/ops/op_quantized_avg_pool2d.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,9 +22,9 @@ Tensor& quantized_avg_pool2d_out(
const IntArrayRef kernel_size,
const IntArrayRef stride,
const IntArrayRef padding,
const Scalar& zero_point,
const Scalar& multiplier,
const Scalar& shift,
const int64_t zero_point,
const int64_t multiplier,
const int64_t shift,
Tensor& out) {
if (input.dim() != 4 || out.dim() != 4) {
ET_LOG(Error, "quantized_avg_pool2d_out: tensors must be 4-D");
Expand Down
14 changes: 7 additions & 7 deletions backends/cortex_m/ops/op_quantized_mul.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,12 +26,12 @@ using KernelRuntimeContext = torch::executor::KernelRuntimeContext;
Tensor& quantized_mul_out(
KernelRuntimeContext& context,
const Tensor& input1_int8,
const Scalar& input1_zero_point,
const int64_t input1_zero_point,
const Tensor& input2_int8,
const Scalar& input2_zero_point,
const Scalar& output_zero_point,
const Scalar& output_multiplier,
const Scalar& output_shift,
const int64_t input2_zero_point,
const int64_t output_zero_point,
const int64_t output_multiplier,
const int64_t output_shift,
Tensor& out) {
// Validate tensor types and quantization parameters

Expand All @@ -44,8 +44,8 @@ Tensor& quantized_mul_out(
/*require_channels_last=*/channel_broadcast,
/*require_same_sizes=*/!channel_broadcast);

const Scalar kIdentityMultiplier(/*value=*/1);
const Scalar kZeroShift(/*value=*/0);
const int32_t kIdentityMultiplier(/*value=*/1);
const int32_t kZeroShift(/*value=*/0);
validate_quantization_params(
input1_zero_point,
kIdentityMultiplier,
Expand Down
24 changes: 12 additions & 12 deletions backends/cortex_m/ops/operators.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -114,17 +114,17 @@ def dequantize_per_tensor_impl(
# Define the operator schema with multipliers and shifts (11 args)
lib.define(
"quantized_add("
"Tensor self, Scalar self_zero_point, Scalar self_multiplier, Scalar self_shift, "
"Tensor other, Scalar other_zero_point, Scalar other_multiplier, Scalar other_shift, "
"Scalar output_zero_point, Scalar output_multiplier, Scalar output_shift) -> Tensor"
"Tensor self, int self_zero_point, int self_multiplier, int self_shift, "
"Tensor other, int other_zero_point, int other_multiplier, int other_shift, "
"int output_zero_point, int output_multiplier, int output_shift) -> Tensor"
)

# Define the operator schema with multipliers and shifts (11 args + out tensor)
lib.define(
"quantized_add.out("
"Tensor self, Scalar self_zero_point, Scalar self_multiplier, Scalar self_shift, "
"Tensor other, Scalar other_zero_point, Scalar other_multiplier, Scalar other_shift, "
"Scalar output_zero_point, Scalar output_multiplier, Scalar output_shift, "
"Tensor self, int self_zero_point, int self_multiplier, int self_shift, "
"Tensor other, int other_zero_point, int other_multiplier, int other_shift, "
"int output_zero_point, int output_multiplier, int output_shift, "
"*, Tensor(a!) out) -> Tensor(a!)"
)

Expand Down Expand Up @@ -591,9 +591,9 @@ def quantized_conv2d_impl(
"int[] kernel_size, "
"int[] stride, "
"int[] padding, "
"Scalar zero_point, "
"Scalar multiplier, "
"Scalar shift"
"int zero_point, "
"int multiplier, "
"int shift"
") -> Tensor"
)
lib.define(
Expand All @@ -602,9 +602,9 @@ def quantized_conv2d_impl(
"int[] kernel_size, "
"int[] stride, "
"int[] padding, "
"Scalar zero_point, "
"Scalar multiplier, "
"Scalar shift, "
"int zero_point, "
"int multiplier, "
"int shift, "
"*, Tensor(a!) out) -> Tensor(a!)"
)

Expand Down
6 changes: 3 additions & 3 deletions backends/cortex_m/ops/operators.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,13 +17,13 @@
- arg_meta: null
kernel_name: cortex_m::dequantize_per_tensor_out

- func: cortex_m::quantized_add.out(Tensor self, Scalar self_zero_point, Scalar self_multiplier, Scalar self_shift, Tensor other, Scalar other_zero_point, Scalar other_multiplier, Scalar other_shift, Scalar output_zero_point, Scalar output_multiplier, Scalar output_shift, *, Tensor(a!) out) -> Tensor(a!)
- func: cortex_m::quantized_add.out(Tensor self, int self_zero_point, int self_multiplier, int self_shift, Tensor other, int other_zero_point, int other_multiplier, int other_shift, int output_zero_point, int output_multiplier, int output_shift, *, Tensor(a!) out) -> Tensor(a!)
variants: function
kernels:
- arg_meta: null
kernel_name: cortex_m::quantized_add_out

- func: cortex_m::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(a!) out) -> Tensor(a!)
- func: cortex_m::quantized_mul.out(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!)
variants: function
kernels:
- arg_meta: null
Expand Down Expand Up @@ -59,7 +59,7 @@
- arg_meta: null
kernel_name: cortex_m::quantized_conv2d_out

- func: cortex_m::quantized_avg_pool2d.out(Tensor input, int[] kernel_size, int[] stride, int[] padding, Scalar zero_point, Scalar multiplier, Scalar shift, *, Tensor(a!) out) -> Tensor(a!)
- func: cortex_m::quantized_avg_pool2d.out(Tensor input, int[] kernel_size, int[] stride, int[] padding, int zero_point, int multiplier, int shift, *, Tensor(a!) out) -> Tensor(a!)
variants: function
kernels:
- arg_meta: null
Expand Down
Loading