diff --git a/backends/cadence/fusion_g3/operators/op_add.cpp b/backends/cadence/fusion_g3/operators/op_add.cpp
index d51fee5338..409c4cc510 100644
--- a/backends/cadence/fusion_g3/operators/op_add.cpp
+++ b/backends/cadence/fusion_g3/operators/op_add.cpp
@@ -35,21 +35,7 @@ Tensor& add_out(
const Tensor& b,
const Scalar& alpha,
Tensor& out) {
- // Common Dtype
- ScalarType common_type =
- executorch::runtime::promoteTypes(a.scalar_type(), b.scalar_type());
-
#ifdef OP_ARG_CHECK
- // Check Common Dtype
- ET_KERNEL_CHECK(
- ctx,
- (canCast(common_type, out.scalar_type()) &&
- torch::executor::check_alpha_type(
- torch::executor::native::utils::get_scalar_dtype(alpha),
- common_type)),
- InvalidArgument,
- out);
-
// Check Dim Order
ET_KERNEL_CHECK(
ctx,
@@ -65,10 +51,6 @@ Tensor& add_out(
out);
#endif
- // Compute Dtype
- ScalarType compute_type =
- torch::executor::native::utils::get_compute_type(common_type);
-
static constexpr const char op_name[] = "add.out";
int kTensorDimensionLimit = 5;
@@ -77,12 +59,12 @@ Tensor& add_out(
int inp2_shape[kTensorDimensionLimit];
int out_shape[kTensorDimensionLimit];
- bool broadcast = 0;
+ bool broadcast = false;
int max_dim = a.dim() > b.dim() ? a.dim() : b.dim();
max_dim = out.dim() > max_dim ? out.dim() : max_dim;
- bool optimized = 1;
+ bool optimized = true;
/* Added change to work with input dimensions more than 5 */
for (int i = 0; i < max_dim; i++) {
@@ -109,15 +91,19 @@ Tensor& add_out(
for (int i = 0; i < out.dim(); i++) {
if (((inp1_shape[i]) != (out_shape[i])) ||
((inp2_shape[i]) != (out_shape[i]))) {
- broadcast = 1;
+ broadcast = true;
}
}
- if ((broadcast == 1) && (max_dim > kTensorDimensionLimit)) {
- optimized = 0;
+ if (((broadcast) && (max_dim > kTensorDimensionLimit)) ||
+ (!(((a.scalar_type() == ScalarType::Int) ||
+ (a.scalar_type() == ScalarType::Float)) &&
+ (a.scalar_type() == b.scalar_type()) &&
+ (a.scalar_type() == out.scalar_type())))) {
+ optimized = false;
}
- if ((compute_type == ScalarType::Int) && (optimized)) {
+ if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
const int* const inp1_data = a.const_data_ptr<int>();
const int* const inp2_data = b.const_data_ptr<int>();
int* const out_data = out.mutable_data_ptr<int>();
@@ -169,7 +155,7 @@ Tensor& add_out(
alpha_val,
out.numel());
}
- } else if ((compute_type == ScalarType::Float) && (optimized)) {
+ } else if ((a.scalar_type() == ScalarType::Float) && (optimized)) {
const float* const inp1_data = a.const_data_ptr<float>();
const float* const inp2_data = b.const_data_ptr<float>();
float* const out_data = out.mutable_data_ptr<float>();
@@ -222,6 +208,23 @@ Tensor& add_out(
out.numel());
}
} else {
+ // Common Dtype
+ ScalarType common_type =
+ executorch::runtime::promoteTypes(a.scalar_type(), b.scalar_type());
+ // Compute Dtype
+ ScalarType compute_type =
+ torch::executor::native::utils::get_compute_type(common_type);
+
+ // Check Common Dtype
+ ET_KERNEL_CHECK(
+ ctx,
+ (canCast(common_type, out.scalar_type()) &&
+ torch::executor::check_alpha_type(
+ torch::executor::native::utils::get_scalar_dtype(alpha),
+ common_type)),
+ InvalidArgument,
+ out);
+
ET_SWITCH_REALB_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
const CTYPE_COMPUTE val_alpha =
torch::executor::native::utils::scalar_to<CTYPE_COMPUTE>(alpha);
@@ -249,22 +252,7 @@ Tensor& add_scalar_out(
const Scalar& b,
const Scalar& alpha,
Tensor& out) {
- // Common Dtype
- ScalarType common_type =
- torch::executor::native::utils::promote_type_with_scalar(
- a.scalar_type(), b);
-
#ifdef OP_ARG_CHECK
- // Check Common Dtype
- ET_KERNEL_CHECK(
- ctx,
- (common_type == out.scalar_type() &&
- torch::executor::check_alpha_type(
- torch::executor::native::utils::get_scalar_dtype(alpha),
- common_type)),
- InvalidArgument,
- out);
-
// Check Dim Order
ET_KERNEL_CHECK(
ctx,
@@ -279,14 +267,23 @@ Tensor& add_scalar_out(
InvalidArgument,
out);
#endif
- // Compute Dtype
- ScalarType compute_type =
- torch::executor::native::utils::get_compute_type(common_type);
// @lint-ignore CLANGTIDY facebook-hte-CArray
static constexpr const char op_name[] = "add.Scalar_out";
- if (compute_type == ScalarType::Int) {
+ bool optimized = true;
+
+ if (!(((a.scalar_type() == ScalarType::Int) ||
+ (a.scalar_type() == ScalarType::Float)) &&
+ (a.scalar_type() == out.scalar_type()))) {
+ optimized = false;
+ }
+
+ if ((b.isFloatingPoint()) && (a.scalar_type() == ScalarType::Int)) {
+ optimized = false;
+ }
+
+ if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
const int* const inp1_data = a.const_data_ptr<int>();
int inp2_val;
torch::executor::native::utils::extract_scalar(b, &inp2_val);
@@ -306,7 +303,7 @@ Tensor& add_scalar_out(
alpha_val,
out.numel());
- } else if (compute_type == ScalarType::Float) {
+ } else if ((a.scalar_type() == ScalarType::Float) && (optimized)) {
const float* const inp1_data = a.const_data_ptr<float>();
float inp2_val;
torch::executor::native::utils::extract_scalar(b, &inp2_val);
@@ -327,6 +324,24 @@ Tensor& add_scalar_out(
out.numel());
} else {
+ // Common Dtype
+ ScalarType common_type =
+ torch::executor::native::utils::promote_type_with_scalar(
+ a.scalar_type(), b);
+ // Compute Dtype
+ ScalarType compute_type =
+ torch::executor::native::utils::get_compute_type(common_type);
+
+ // Check Common Dtype
+ ET_KERNEL_CHECK(
+ ctx,
+ (common_type == out.scalar_type() &&
+ torch::executor::check_alpha_type(
+ torch::executor::native::utils::get_scalar_dtype(alpha),
+ common_type)),
+ InvalidArgument,
+ out);
+
ET_SWITCH_REALB_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
torch::executor::native::utils::
apply_unitensor_elementwise_fn<CTYPE_COMPUTE, op_name>(
diff --git a/backends/cadence/fusion_g3/operators/op_cat.cpp b/backends/cadence/fusion_g3/operators/op_cat.cpp
index 74fd96a212..84224b37b0 100644
--- a/backends/cadence/fusion_g3/operators/op_cat.cpp
+++ b/backends/cadence/fusion_g3/operators/op_cat.cpp
@@ -46,11 +46,6 @@ Tensor& cat_out(
int kTensorDimensionLimit = executorch::runtime::kTensorDimensionLimit;
#ifdef OP_ARG_CHECK
- ET_KERNEL_CHECK(
- ctx,
- torch::executor::check_cat_args(tensors, dim, out),
- InvalidArgument,
- out);
Tensor::SizesType expected_out_size[kTensorDimensionLimit];
size_t expected_out_dim = 0;
@@ -106,7 +101,16 @@ Tensor& cat_out(
out_shapes[i] = out_size[i];
}
- if ((out.scalar_type() == ScalarType::Int) ||
+ bool optimized = true;
+
+ for (int i = 0; i < tensors.size(); i++) {
+ if (out.scalar_type() != tensors[i].scalar_type()) {
+ optimized = false;
+ break;
+ }
+ }
+
+ if ((optimized) && (out.scalar_type() == ScalarType::Int) ||
(out.scalar_type() == ScalarType::Short) ||
(out.scalar_type() == ScalarType::Char) ||
(out.scalar_type() == ScalarType::UInt32) ||
@@ -125,6 +129,12 @@ Tensor& cat_out(
(int)dim,
get_element_size(out.scalar_type()));
} else {
+ ET_KERNEL_CHECK(
+ ctx,
+ torch::executor::check_cat_args(tensors, dim, out),
+ InvalidArgument,
+ out);
+
const size_t outer = executorch::runtime::getLeadingDims(out, dim);
const size_t dim_stride = executorch::runtime::getTrailingDims(out, dim);
const size_t ninputs = tensors.size();
diff --git a/backends/cadence/fusion_g3/operators/op_dequantize.cpp b/backends/cadence/fusion_g3/operators/op_dequantize.cpp
index 3e0235170b..dd9d4f2a51 100644
--- a/backends/cadence/fusion_g3/operators/op_dequantize.cpp
+++ b/backends/cadence/fusion_g3/operators/op_dequantize.cpp
@@ -117,7 +117,7 @@ Tensor& dequantize_impl(
}
}
} else {
- if (*zero_point_data != 0) // tesor
+ if (*zero_point_data != 0) // tensor
{
is_asym_dequant |= 1;
}
@@ -125,8 +125,14 @@ Tensor& dequantize_impl(
}
float* out_data = out.mutable_data_ptr<float>();
+ bool optimized = true;
+
+ if (out.scalar_type() != ScalarType::Float) {
+ optimized = false;
+ }
+
if (is_asym_dequant) {
- if (input.scalar_type() == ScalarType::Byte) {
+ if ((input.scalar_type() == ScalarType::Byte) && (optimized)) {
const uint8_t* input_data = input.const_data_ptr<uint8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -139,7 +145,7 @@ Tensor& dequantize_impl(
axis,
zero_point_data,
scale_data);
- } else if (input.scalar_type() == ScalarType::Char) {
+ } else if ((input.scalar_type() == ScalarType::Char) && (optimized)) {
const int8_t* input_data = input.const_data_ptr<int8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -152,7 +158,7 @@ Tensor& dequantize_impl(
axis,
zero_point_data,
scale_data);
- } else if (input.scalar_type() == ScalarType::UInt16) {
+ } else if ((input.scalar_type() == ScalarType::UInt16) && (optimized)) {
const uint16_t* input_data = input.const_data_ptr<uint16_t>();
XT_KERNEL_CHECK(
ctx,
@@ -165,7 +171,7 @@ Tensor& dequantize_impl(
axis,
zero_point_data,
scale_data);
- } else if (input.scalar_type() == ScalarType::Short) {
+ } else if ((input.scalar_type() == ScalarType::Short) && (optimized)) {
const int16_t* input_data = input.const_data_ptr<int16_t>();
XT_KERNEL_CHECK(
ctx,
@@ -178,7 +184,7 @@ Tensor& dequantize_impl(
axis,
zero_point_data,
scale_data);
- } else if (input.scalar_type() == (ScalarType)Bits4u) {
+ } else if ((input.scalar_type() == (ScalarType)Bits4u) && (optimized)) {
const uint8_t* input_data = input.const_data_ptr<uint8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -191,7 +197,7 @@ Tensor& dequantize_impl(
axis,
zero_point_data,
scale_data);
- } else if (input.scalar_type() == (ScalarType)Bits4) {
+ } else if ((input.scalar_type() == (ScalarType)Bits4) && (optimized)) {
const int8_t* input_data = input.const_data_ptr<int8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -338,7 +344,7 @@ Tensor& dequantize_impl(
}
}
} else {
- if (input.scalar_type() == ScalarType::Byte) {
+ if ((input.scalar_type() == ScalarType::Byte) && (optimized)) {
const uint8_t* input_data = input.const_data_ptr<uint8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -350,7 +356,7 @@ Tensor& dequantize_impl(
input.dim(),
axis,
scale_data);
- } else if (input.scalar_type() == ScalarType::Char) {
+ } else if ((input.scalar_type() == ScalarType::Char) && (optimized)) {
const int8_t* input_data = input.const_data_ptr<int8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -362,7 +368,7 @@ Tensor& dequantize_impl(
input.dim(),
axis,
scale_data);
- } else if (input.scalar_type() == ScalarType::UInt16) {
+ } else if ((input.scalar_type() == ScalarType::UInt16) && (optimized)) {
const uint16_t* input_data = input.const_data_ptr<uint16_t>();
XT_KERNEL_CHECK(
ctx,
@@ -374,7 +380,7 @@ Tensor& dequantize_impl(
input.dim(),
axis,
scale_data);
- } else if (input.scalar_type() == ScalarType::Short) {
+ } else if ((input.scalar_type() == ScalarType::Short) && (optimized)) {
const int16_t* input_data = input.const_data_ptr<int16_t>();
XT_KERNEL_CHECK(
ctx,
@@ -386,7 +392,7 @@ Tensor& dequantize_impl(
input.dim(),
axis,
scale_data);
- } else if (input.scalar_type() == (ScalarType)Bits4u) {
+ } else if ((input.scalar_type() == (ScalarType)Bits4u) && (optimized)) {
const uint8_t* input_data = input.const_data_ptr<uint8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -398,7 +404,7 @@ Tensor& dequantize_impl(
input.dim(),
axis,
scale_data);
- } else if (input.scalar_type() == (ScalarType)Bits4) {
+ } else if ((input.scalar_type() == (ScalarType)Bits4) && (optimized)) {
const int8_t* input_data = input.const_data_ptr<int8_t>();
XT_KERNEL_CHECK(
ctx,
diff --git a/backends/cadence/fusion_g3/operators/op_div.cpp b/backends/cadence/fusion_g3/operators/op_div.cpp
index 1461f643a8..85e5da4276 100644
--- a/backends/cadence/fusion_g3/operators/op_div.cpp
+++ b/backends/cadence/fusion_g3/operators/op_div.cpp
@@ -54,10 +54,6 @@ Tensor& div_out(
const Tensor& a,
const Tensor& b,
Tensor& out) {
- // Common Dtype
- ScalarType common_type =
- executorch::runtime::promoteTypes(a.scalar_type(), b.scalar_type());
-
#ifdef OP_ARG_CHECK
// Check Dim Order
ET_KERNEL_CHECK(
@@ -73,11 +69,6 @@ Tensor& div_out(
InvalidArgument,
out);
#endif
-
- // Compute Dtype
- ScalarType compute_type =
- torch::executor::native::utils::get_compute_type(common_type);
-
// @lint-ignore CLANGTIDY facebook-hte-CArray
static constexpr const char op_name[] = "div.out";
@@ -87,12 +78,12 @@ Tensor& div_out(
int inp2_shape[kTensorDimensionLimit];
int out_shape[kTensorDimensionLimit];
- bool broadcast = 0;
+ bool broadcast = false;
int max_dim = a.dim() > b.dim() ? a.dim() : b.dim();
max_dim = out.dim() > max_dim ? out.dim() : max_dim;
- bool optimized = 1;
+ bool optimized = true;
for (int i = 0; i < max_dim; i++) {
out_shape[i] = 1;
@@ -118,15 +109,19 @@ Tensor& div_out(
for (int i = 0; i < out.dim(); i++) {
if (((inp1_shape[i]) != (out_shape[i])) ||
((inp2_shape[i]) != (out_shape[i]))) {
- broadcast = 1;
+ broadcast = true;
}
}
- if ((broadcast == 1) && (max_dim > kTensorDimensionLimit)) {
- optimized = 0;
+ if (((broadcast) && (max_dim > kTensorDimensionLimit)) ||
+ (!(((a.scalar_type() == ScalarType::Int) ||
+ (a.scalar_type() == ScalarType::Float)) &&
+ (a.scalar_type() == b.scalar_type()) &&
+ (out.scalar_type() == ScalarType::Float)))) {
+ optimized = false;
}
- if ((compute_type == ScalarType::Int) && (optimized)) {
+ if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
const int* const inp1_data = a.const_data_ptr<int>();
const int* const inp2_data = b.const_data_ptr<int>();
float* const out_data = out.mutable_data_ptr<float>();
@@ -162,7 +157,7 @@ Tensor& div_out(
inp2_data,
out.numel());
}
- } else if ((compute_type == ScalarType::Float) && (optimized)) {
+ } else if ((a.scalar_type() == ScalarType::Float) && (optimized)) {
const float* const inp1_data = a.const_data_ptr<float>();
const float* const inp2_data = b.const_data_ptr<float>();
float* const out_data = out.mutable_data_ptr<float>();
@@ -244,19 +239,7 @@ Tensor& div_out_mode(
ET_KERNEL_CHECK(
ctx, mode_val == "trunc" || mode_val == "floor", InvalidArgument, out);
- // Common Dtype
- ScalarType common_type =
- executorch::runtime::promoteTypes(a.scalar_type(), b.scalar_type());
-
#ifdef OP_ARG_CHECK
- // Check Common Dtype
- ET_KERNEL_CHECK(
- ctx,
- (canCast(common_type, out.scalar_type()) &&
- common_type != ScalarType::Bool),
- InvalidArgument,
- out);
-
// Check Dim Order
ET_KERNEL_CHECK(
ctx,
@@ -271,9 +254,6 @@ Tensor& div_out_mode(
InvalidArgument,
out);
#endif
- // Compute Dtype
- ScalarType compute_type =
- torch::executor::native::utils::get_compute_type(common_type);
// @lint-ignore CLANGTIDY facebook-hte-CArray
static constexpr const char op_name[] = "div.out_mode";
@@ -287,12 +267,12 @@ Tensor& div_out_mode(
int inp2_shape[kTensorDimensionLimit];
int out_shape[kTensorDimensionLimit];
- bool broadcast = 0;
+ bool broadcast = false;
int max_dim = a.dim() > b.dim() ? a.dim() : b.dim();
max_dim = out.dim() > max_dim ? out.dim() : max_dim;
- bool optimized = 1;
+ bool optimized = true;
for (int i = 0; i < max_dim; i++) {
out_shape[i] = 1;
@@ -318,17 +298,21 @@ Tensor& div_out_mode(
for (int i = 0; i < out.dim(); i++) {
if (((inp1_shape[i]) != (out_shape[i])) ||
((inp2_shape[i]) != (out_shape[i]))) {
- broadcast = 1;
+ broadcast = true;
}
}
- if ((broadcast == 1) && (max_dim > kTensorDimensionLimit)) {
- optimized = 0;
+ if (((broadcast) && (max_dim > kTensorDimensionLimit)) ||
+ (!(((a.scalar_type() == ScalarType::Int) ||
+ (a.scalar_type() == ScalarType::Float)) &&
+ (a.scalar_type() == b.scalar_type()) &&
+ (a.scalar_type() == out.scalar_type())))) {
+ optimized = false;
}
int mode_value = (mode_val == "trunc") ? 1 : 2;
- if ((compute_type == ScalarType::Int) && (optimized)) {
+ if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
const int* const inp1_data = a.const_data_ptr<int>();
const int* const inp2_data = b.const_data_ptr<int>();
int* const out_data = out.mutable_data_ptr<int>();
@@ -367,7 +351,7 @@ Tensor& div_out_mode(
mode_value,
out.numel());
}
- } else if ((compute_type == ScalarType::Float) && (optimized)) {
+ } else if ((a.scalar_type() == ScalarType::Float) && (optimized)) {
const float* const inp1_data = a.const_data_ptr<float>();
const float* const inp2_data = b.const_data_ptr<float>();
float* const out_data = out.mutable_data_ptr<float>();
@@ -407,6 +391,21 @@ Tensor& div_out_mode(
out.numel());
}
} else {
+ // Common Dtype
+ ScalarType common_type =
+ executorch::runtime::promoteTypes(a.scalar_type(), b.scalar_type());
+ // Compute Dtype
+ ScalarType compute_type =
+ torch::executor::native::utils::get_compute_type(common_type);
+
+ // Check Common Dtype
+ ET_KERNEL_CHECK(
+ ctx,
+ (canCast(common_type, out.scalar_type()) &&
+ common_type != ScalarType::Bool),
+ InvalidArgument,
+ out);
+
ET_SWITCH_REAL_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
torch::executor::native::utils::
apply_bitensor_elementwise_fn<CTYPE_COMPUTE, op_name>(
@@ -456,15 +455,7 @@ Tensor& div_scalar_out(
const Tensor& a,
const Scalar& b,
Tensor& out) {
- // Common Dtype
- ScalarType common_type =
- torch::executor::native::utils::promote_type_with_scalar(
- a.scalar_type(), b);
-
#ifdef OP_ARG_CHECK
- // Check Common Dtype
- ET_KERNEL_CHECK(ctx, common_type == out.scalar_type(), InvalidArgument, out);
-
// Check Dim Order
ET_KERNEL_CHECK(
ctx,
@@ -480,14 +471,22 @@ Tensor& div_scalar_out(
out);
#endif
- // Compute Dtype
- ScalarType compute_type =
- torch::executor::native::utils::get_compute_type(common_type);
+ bool optimized = true;
+
+ if (!(((a.scalar_type() == ScalarType::Int) ||
+ (a.scalar_type() == ScalarType::Float)) &&
+ (out.scalar_type() == ScalarType::Float))) {
+ optimized = false;
+ }
+
+ if ((b.isFloatingPoint()) && (a.scalar_type() == ScalarType::Int)) {
+ optimized = false;
+ }
// @lint-ignore CLANGTIDY facebook-hte-CArray
static constexpr const char op_name[] = "div.Scalar_out";
- if (compute_type == ScalarType::Int) {
+ if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
const int* const inp1_data = a.const_data_ptr<int>();
int inp2_val;
torch::executor::native::utils::extract_scalar(b, &inp2_val);
@@ -502,7 +501,7 @@ Tensor& div_scalar_out(
inp1_data,
inp2_val,
out.numel());
- } else if (compute_type == ScalarType::Float) {
+ } else if ((a.scalar_type() == ScalarType::Float) && (optimized)) {
const float* const inp1_data = a.const_data_ptr<float>();
float inp2_val;
torch::executor::native::utils::extract_scalar(b, &inp2_val);
@@ -526,6 +525,11 @@ Tensor& div_scalar_out(
: ScalarType::Float;
ScalarType compute_type =
torch::executor::native::utils::get_compute_type(common_type);
+
+ // Check Common Dtype
+ ET_KERNEL_CHECK(
+ ctx, common_type == out.scalar_type(), InvalidArgument, out);
+
ET_SWITCH_FLOAT_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
const CTYPE_COMPUTE val_b =
torch::executor::native::utils::scalar_to<CTYPE_COMPUTE>(b);
@@ -560,29 +564,7 @@ Tensor& div_scalar_mode_out(
ET_KERNEL_CHECK(
ctx, mode_val == "trunc" || mode_val == "floor", InvalidArgument, out);
- // Common Dtype
- ScalarType common_type =
- torch::executor::native::utils::promote_type_with_scalar(
- a.scalar_type(), b);
-
#ifdef OP_ARG_CHECK
- // Check Common Dtype
- ET_KERNEL_CHECK(
- ctx,
- (canCast(common_type, out.scalar_type()) &&
- common_type != ScalarType::Bool),
- InvalidArgument,
- out);
-
- // Check for intergral division by zero
- ET_KERNEL_CHECK_MSG(
- ctx,
- !(executorch::runtime::isIntegralType(common_type, true) &&
- torch::executor::native::utils::scalar_to<double>(b) == 0),
- InvalidArgument,
- out,
- "Div mode operation encountered integer division by zero");
-
// Check Dim Order
ET_KERNEL_CHECK(
ctx,
@@ -598,18 +580,26 @@ Tensor& div_scalar_mode_out(
out);
#endif
- // Compute Dtype
- ScalarType compute_type =
- torch::executor::native::utils::get_compute_type(common_type);
-
const bool mode_is_trunc = mode_val == "trunc";
+ bool optimized = true;
+
+ if (!(((a.scalar_type() == ScalarType::Int) ||
+ (a.scalar_type() == ScalarType::Float)) &&
+ (a.scalar_type() == out.scalar_type()))) {
+ optimized = false;
+ }
+
+ if ((b.isFloatingPoint()) && (a.scalar_type() == ScalarType::Int)) {
+ optimized = false;
+ }
+
// @lint-ignore CLANGTIDY facebook-hte-CArray
static constexpr const char op_name[] = "div.Scalar_mode_out";
int mode_value = (mode_val == "trunc") ? 1 : 2;
- if (compute_type == ScalarType::Int) {
+ if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
const int* const inp1_data = a.const_data_ptr<int>();
int inp2_val;
torch::executor::native::utils::extract_scalar(b, &inp2_val);
@@ -625,7 +615,7 @@ Tensor& div_scalar_mode_out(
inp2_val,
mode_value,
out.numel());
- } else if (compute_type == ScalarType::Float) {
+ } else if ((a.scalar_type() == ScalarType::Float) && (optimized)) {
const float* const inp1_data = a.const_data_ptr<float>();
float inp2_val;
torch::executor::native::utils::extract_scalar(b, &inp2_val);
@@ -642,6 +632,31 @@ Tensor& div_scalar_mode_out(
mode_value,
out.numel());
} else {
+ // Common Dtype
+ ScalarType common_type =
+ torch::executor::native::utils::promote_type_with_scalar(
+ a.scalar_type(), b);
+ // Compute Dtype
+ ScalarType compute_type =
+ torch::executor::native::utils::get_compute_type(common_type);
+
+ // Check Common Dtype
+ ET_KERNEL_CHECK(
+ ctx,
+ (canCast(common_type, out.scalar_type()) &&
+ common_type != ScalarType::Bool),
+ InvalidArgument,
+ out);
+
+ // Check for intergral division by zero
+ ET_KERNEL_CHECK_MSG(
+ ctx,
+ !(executorch::runtime::isIntegralType(common_type, true) &&
+ torch::executor::native::utils::scalar_to<double>(b) == 0),
+ InvalidArgument,
+ out,
+ "Div mode operation encountered integer division by zero");
+
ET_SWITCH_REAL_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
const CTYPE_COMPUTE val_b =
torch::executor::native::utils::scalar_to<CTYPE_COMPUTE>(b);
diff --git a/backends/cadence/fusion_g3/operators/op_exp.cpp b/backends/cadence/fusion_g3/operators/op_exp.cpp
index 4b6b898b17..41b5d70b22 100644
--- a/backends/cadence/fusion_g3/operators/op_exp.cpp
+++ b/backends/cadence/fusion_g3/operators/op_exp.cpp
@@ -49,9 +49,10 @@ Tensor& exp_out(KernelRuntimeContext& ctx, const Tensor& in, Tensor& out) {
out);
#endif
- if (in.scalar_type() == ScalarType::Float) {
- float* __restrict__ out_data = out.mutable_data_ptr<float>();
- const float* __restrict__ in_data = in.const_data_ptr<float>();
+ if ((in.scalar_type() == ScalarType::Float) &&
+ (out.scalar_type() == ScalarType::Float)) {
+ float* const out_data = out.mutable_data_ptr<float>();
+ const float* const in_data = in.const_data_ptr<float>();
XT_KERNEL_CHECK(
ctx, out, xa_nn_elm_exp_f32_f32, out_data, in_data, out.numel());
diff --git a/backends/cadence/fusion_g3/operators/op_mean.cpp b/backends/cadence/fusion_g3/operators/op_mean.cpp
index 289baceb12..cd02714113 100644
--- a/backends/cadence/fusion_g3/operators/op_mean.cpp
+++ b/backends/cadence/fusion_g3/operators/op_mean.cpp
@@ -44,22 +44,23 @@ int prepare_data(
for (int i = 0; i < num_out_dims; i++) {
out_shape[i] = out.size(i);
}
-
int num_axis_dims = 0;
- for (const auto& d : dim_list.value()) {
- if (d < 0) {
- p_axis[num_axis_dims] = num_inp_dims + d;
- num_axis_dims++;
- } else {
- p_axis[num_axis_dims] = d;
- num_axis_dims++;
+ if (dim_list.has_value()) {
+ for (const auto& d : dim_list.value()) {
+ if (d < 0) {
+ p_axis[num_axis_dims] = num_inp_dims + d;
+ num_axis_dims++;
+ } else {
+ p_axis[num_axis_dims] = d;
+ num_axis_dims++;
+ }
}
}
return num_axis_dims;
}
-Tensor& mean_out(
+Tensor& mean_dim_out(
KernelRuntimeContext& ctx,
const Tensor& in,
optional<ArrayRef<int64_t>> dim_list,
@@ -69,12 +70,6 @@ Tensor& mean_out(
(void)ctx;
#ifdef OP_ARG_CHECK
- ET_KERNEL_CHECK(
- ctx,
- torch::executor::check_mean_dim_args(in, dim_list, keepdim, dtype, out),
- InvalidArgument,
- out);
-
ET_KERNEL_CHECK(
ctx,
executorch::runtime::tensors_have_same_dim_order(in, out),
@@ -97,13 +92,14 @@ Tensor& mean_out(
constexpr int kNnlibMaxDim = 5;
- bool optimized = 1;
+ bool optimized = true;
- if (out.scalar_type() != ScalarType::Float)
- optimized = 0;
+ if (!((out.scalar_type() == ScalarType::Float) &&
+ (in.scalar_type() == ScalarType::Float)))
+ optimized = false;
if (in.dim() > kNnlibMaxDim)
- optimized = 0;
+ optimized = false;
if (optimized) {
float* __restrict__ p_out = out.mutable_data_ptr<float>();
@@ -135,9 +131,8 @@ Tensor& mean_out(
num_inp_dims,
num_out_dims);
- if (num_axis_dims == num_inp_dims) {
+ if ((num_axis_dims == num_inp_dims) || (!dim_list.has_value())) {
num_out_dims = 1;
- out_shape[0] = 1;
}
int inp_shape_max = inp_shape[p_axis[0]];
@@ -168,6 +163,12 @@ Tensor& mean_out(
num_axis_dims,
p_scratch_in);
} else {
+ ET_KERNEL_CHECK(
+ ctx,
+ torch::executor::check_mean_dim_args(in, dim_list, keepdim, dtype, out),
+ InvalidArgument,
+ out);
+
ET_SWITCH_REALHB_TYPES(in.scalar_type(), ctx, "mean.out", CTYPE_IN, [&] {
ET_SWITCH_FLOATH_TYPES(
out.scalar_type(), ctx, "mean.out", CTYPE_OUT, [&] {
diff --git a/backends/cadence/fusion_g3/operators/op_mul.cpp b/backends/cadence/fusion_g3/operators/op_mul.cpp
index 93b4c5a992..bee6ac9cbd 100644
--- a/backends/cadence/fusion_g3/operators/op_mul.cpp
+++ b/backends/cadence/fusion_g3/operators/op_mul.cpp
@@ -33,15 +33,7 @@ Tensor& mul_out(
const Tensor& a,
const Tensor& b,
Tensor& out) {
- // Common Dtype
- ScalarType common_type =
- executorch::runtime::promoteTypes(a.scalar_type(), b.scalar_type());
-
#ifdef OP_ARG_CHECK
- // Check Common Dtype
- ET_KERNEL_CHECK(
- ctx, canCast(common_type, out.scalar_type()), InvalidArgument, out);
-
// Check Dim Order
ET_KERNEL_CHECK(
ctx,
@@ -57,10 +49,6 @@ Tensor& mul_out(
out);
#endif
- // Compute Dtype
- ScalarType compute_type =
- torch::executor::native::utils::get_compute_type(common_type);
-
// @lint-ignore CLANGTIDY facebook-hte-CArray
static constexpr const char op_name[] = "mul.out";
int kTensorDimensionLimit = 5;
@@ -69,12 +57,12 @@ Tensor& mul_out(
int inp2_shape[kTensorDimensionLimit];
int out_shape[kTensorDimensionLimit];
- bool broadcast = 0;
+ bool broadcast = false;
int max_dim = a.dim() > b.dim() ? a.dim() : b.dim();
max_dim = out.dim() > max_dim ? out.dim() : max_dim;
- bool optimized = 1;
+ bool optimized = true;
/* Added change to work with input dimensions more than 5 */
for (int i = 0; i < max_dim; i++) {
@@ -101,15 +89,19 @@ Tensor& mul_out(
for (int i = 0; i < out.dim(); i++) {
if (((inp1_shape[i]) != (out_shape[i])) ||
((inp2_shape[i]) != (out_shape[i]))) {
- broadcast = 1;
+ broadcast = true;
}
}
- if ((broadcast == 1) && (max_dim > kTensorDimensionLimit)) {
- optimized = 0;
+ if (((broadcast) && (max_dim > kTensorDimensionLimit)) ||
+ (!(((a.scalar_type() == ScalarType::Int) ||
+ (a.scalar_type() == ScalarType::Float)) &&
+ (a.scalar_type() == b.scalar_type()) &&
+ (a.scalar_type() == out.scalar_type())))) {
+ optimized = false;
}
- if ((compute_type == ScalarType::Int) && (optimized)) {
+ if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
const int* const inp1_data = a.const_data_ptr<int>();
const int* const inp2_data = b.const_data_ptr<int>();
int* const out_data = out.mutable_data_ptr<int>();
@@ -154,7 +146,7 @@ Tensor& mul_out(
inp2_data,
out.numel());
}
- } else if ((compute_type == ScalarType::Float) && (optimized)) {
+ } else if ((a.scalar_type() == ScalarType::Float) && (optimized)) {
const float* const inp1_data = a.const_data_ptr<float>();
const float* const inp2_data = b.const_data_ptr<float>();
float* const out_data = out.mutable_data_ptr<float>();
@@ -200,6 +192,16 @@ Tensor& mul_out(
out.numel());
}
} else {
+ // Common Dtype
+ ScalarType common_type =
+ executorch::runtime::promoteTypes(a.scalar_type(), b.scalar_type());
+ // Compute Dtype
+ ScalarType compute_type =
+ torch::executor::native::utils::get_compute_type(common_type);
+ // Check Common Dtype
+ ET_KERNEL_CHECK(
+ ctx, canCast(common_type, out.scalar_type()), InvalidArgument, out);
+
ET_SWITCH_REALB_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
torch::executor::native::utils::apply_bitensor_elementwise_fn<
CTYPE_COMPUTE,
@@ -224,15 +226,7 @@ Tensor& mul_scalar_out(
const Tensor& a,
const Scalar& b,
Tensor& out) {
- // Common Dtype
- ScalarType common_type =
- torch::executor::native::utils::promote_type_with_scalar(
- a.scalar_type(), b);
-
#ifdef OP_ARG_CHECK
- // Check Common Dtype
- ET_KERNEL_CHECK(ctx, common_type == out.scalar_type(), InvalidArgument, out);
-
// Check Dim Order
ET_KERNEL_CHECK(
ctx,
@@ -244,13 +238,23 @@ Tensor& mul_scalar_out(
ET_KERNEL_CHECK(
ctx, resize_tensor(out, a.sizes()) == Error::Ok, InvalidArgument, out);
#endif
- // Compute Dtype
- ScalarType compute_type =
- torch::executor::native::utils::get_compute_type(common_type);
// @lint-ignore CLANGTIDY facebook-hte-CArray
static constexpr const char op_name[] = "mul.Scalar_out";
- if (compute_type == ScalarType::Int) {
+
+ bool optimized = true;
+
+ if (!(((a.scalar_type() == ScalarType::Int) ||
+ (a.scalar_type() == ScalarType::Float)) &&
+ (a.scalar_type() == out.scalar_type()))) {
+ optimized = false;
+ }
+
+ if ((b.isFloatingPoint()) && (a.scalar_type() == ScalarType::Int)) {
+ optimized = false;
+ }
+
+ if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
const int* const inp1_data = a.const_data_ptr<int>();
int inp2_val;
torch::executor::native::utils::extract_scalar(b, &inp2_val);
@@ -264,7 +268,7 @@ Tensor& mul_scalar_out(
inp1_data,
inp2_val,
out.numel());
- } else if (compute_type == ScalarType::Float) {
+ } else if ((a.scalar_type() == ScalarType::Float) && (optimized)) {
const float* const inp1_data = a.const_data_ptr<float>();
float inp2_val;
torch::executor::native::utils::extract_scalar(b, &inp2_val);
@@ -279,6 +283,17 @@ Tensor& mul_scalar_out(
inp2_val,
out.numel());
} else {
+ // Common Dtype
+ ScalarType common_type =
+ torch::executor::native::utils::promote_type_with_scalar(
+ a.scalar_type(), b);
+ // Compute Dtype
+ ScalarType compute_type =
+ torch::executor::native::utils::get_compute_type(common_type);
+ // Check Common Dtype
+ ET_KERNEL_CHECK(
+ ctx, common_type == out.scalar_type(), InvalidArgument, out);
+
ET_SWITCH_REALB_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
const CTYPE_COMPUTE val_b =
torch::executor::native::utils::scalar_to<CTYPE_COMPUTE>(b);
diff --git a/backends/cadence/fusion_g3/operators/op_native_layer_norm.cpp b/backends/cadence/fusion_g3/operators/op_native_layer_norm.cpp
index 9857bbce37..b4f076e810 100644
--- a/backends/cadence/fusion_g3/operators/op_native_layer_norm.cpp
+++ b/backends/cadence/fusion_g3/operators/op_native_layer_norm.cpp
@@ -123,14 +123,7 @@ std::tuple<Tensor&, Tensor&, Tensor&> native_layer_norm_out(
std::tuple<Tensor&, Tensor&, Tensor&> ret_val(out, mean_out, rstd_out);
int kTensorDimensionLimit = executorch::runtime::kTensorDimensionLimit;
-
#ifdef OP_ARG_CHECK
- ET_KERNEL_CHECK(
- ctx,
- torch::executor::check_layer_norm_args(
- input, normalized_shape, weight, bias, out, mean_out, rstd_out),
- InvalidArgument,
- ret_val);
// Only support default dim order for now.
// TODO: Support other dim orders.
@@ -189,12 +182,34 @@ std::tuple<Tensor&, Tensor&, Tensor&> native_layer_norm_out(
ret_val);
#endif
+ bool optimized = true;
+
int input_shape[kTensorDimensionLimit];
for (int i = 0; i < input.dim(); i++) {
input_shape[i] = input.size(i);
}
- if (out.scalar_type() == ScalarType::Float) {
+ if (!(((input.scalar_type() == ScalarType::Float) &&
+ (input.scalar_type() == out.scalar_type()) &&
+ (out.scalar_type() == mean_out.scalar_type()) &&
+ (mean_out.scalar_type() == rstd_out.scalar_type())))) {
+ optimized = false;
+ }
+
+ if (optimized) {
+ if (weight.has_value()) {
+ if (!(input.scalar_type() == weight.value().scalar_type())) {
+ optimized = false;
+ }
+ }
+ if (bias.has_value()) {
+ if (!(input.scalar_type() == bias.value().scalar_type())) {
+ optimized = false;
+ }
+ }
+ }
+
+ if ((input.scalar_type() == ScalarType::Float) && (optimized)) {
float* const out_data = out.mutable_data_ptr<float>();
float* const mean_data = mean_out.mutable_data_ptr<float>();
float* const rstd_data = rstd_out.mutable_data_ptr<float>();
@@ -247,6 +262,13 @@ std::tuple<Tensor&, Tensor&, Tensor&> native_layer_norm_out(
free(weight_data);
}
} else {
+ ET_KERNEL_CHECK(
+ ctx,
+ torch::executor::check_layer_norm_args(
+ input, normalized_shape, weight, bias, out, mean_out, rstd_out),
+ InvalidArgument,
+ ret_val);
+
ET_SWITCH_FLOAT_TYPES(
input.scalar_type(), ctx, "native_layer_norm.out", CTYPE, [&]() {
layer_norm<CTYPE>(
diff --git a/backends/cadence/fusion_g3/operators/op_permute_copy.cpp b/backends/cadence/fusion_g3/operators/op_permute_copy.cpp
index 23c2d1e5fb..34def4fd1b 100644
--- a/backends/cadence/fusion_g3/operators/op_permute_copy.cpp
+++ b/backends/cadence/fusion_g3/operators/op_permute_copy.cpp
@@ -65,12 +65,6 @@ Tensor& permute_copy_out(
* the checks only in operator level(As there are no checks in kernel).
*/
#ifdef OP_ARG_CHECK
- ET_KERNEL_CHECK(
- ctx,
- torch::executor::check_permute_copy_args(in, dims, out),
- InvalidArgument,
- out);
-
ET_KERNEL_CHECK(
ctx,
executorch::runtime::tensors_have_same_dim_order(in, out),
@@ -112,7 +106,8 @@ Tensor& permute_copy_out(
signed char* out_data = out.mutable_data_ptr<signed char>();
const signed char* const inp_data = in.const_data_ptr<signed char>();
- if (((out.scalar_type() == ScalarType::Int) ||
+ if (((out.scalar_type() == in.scalar_type()) &&
+ (out.scalar_type() == ScalarType::Int) ||
(out.scalar_type() == ScalarType::Short) ||
(out.scalar_type() == ScalarType::Char) ||
(out.scalar_type() == ScalarType::UInt32) ||
@@ -131,9 +126,15 @@ Tensor& permute_copy_out(
in.dim(),
get_element_size(out.scalar_type()));
} else {
+ ET_KERNEL_CHECK(
+ ctx,
+ torch::executor::check_permute_copy_args(in, dims, out),
+ InvalidArgument,
+ out);
+
const auto in_type = out.scalar_type();
- size_t in_coord[5] = {0};
- size_t trailing_dims_memo[kTensorDimensionLimit];
+ size_t in_coord[executorch::runtime::kTensorDimensionLimit] = {0};
+ size_t trailing_dims_memo[executorch::runtime::kTensorDimensionLimit];
executorch::runtime::memoizeTrailingDims(in, trailing_dims_memo);
// in and out must be the same dtype
ET_SWITCH_ALL_TYPES(in_type, ctx, "permute_copy.out", CTYPE, [&] {
diff --git a/backends/cadence/fusion_g3/operators/op_quantize.cpp b/backends/cadence/fusion_g3/operators/op_quantize.cpp
index 8237c3c266..2af77eca6c 100644
--- a/backends/cadence/fusion_g3/operators/op_quantize.cpp
+++ b/backends/cadence/fusion_g3/operators/op_quantize.cpp
@@ -159,6 +159,12 @@ Tensor& quantize_impl(
bool is_asym_quant = 0;
+ bool optimized = true;
+
+ if (input.scalar_type() != ScalarType::Float) {
+ optimized = false;
+ }
+
if (zero_point_data != NULL) // asymmetric quant
{
if (axis != NULL) // channel
@@ -177,7 +183,7 @@ Tensor& quantize_impl(
}
if (is_asym_quant) {
- if (out.scalar_type() == ScalarType::Byte) {
+ if ((out.scalar_type() == ScalarType::Byte) && (optimized)) {
uint8_t* out_data = out.mutable_data_ptr<uint8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -192,7 +198,7 @@ Tensor& quantize_impl(
zero_point_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == ScalarType::Char) {
+ } else if ((out.scalar_type() == ScalarType::Char) && (optimized)) {
int8_t* out_data = out.mutable_data_ptr<int8_t>();
XT_KERNEL_CHECK(
@@ -208,7 +214,7 @@ Tensor& quantize_impl(
zero_point_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == ScalarType::UInt16) {
+ } else if ((out.scalar_type() == ScalarType::UInt16) && (optimized)) {
uint16_t* out_data = out.mutable_data_ptr<uint16_t>();
XT_KERNEL_CHECK(
ctx,
@@ -223,7 +229,7 @@ Tensor& quantize_impl(
zero_point_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == ScalarType::Short) {
+ } else if ((out.scalar_type() == ScalarType::Short) && (optimized)) {
int16_t* out_data = out.mutable_data_ptr<int16_t>();
XT_KERNEL_CHECK(
ctx,
@@ -238,7 +244,7 @@ Tensor& quantize_impl(
zero_point_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == (ScalarType)Bits4u) {
+ } else if ((out.scalar_type() == (ScalarType)Bits4u) && (optimized)) {
uint8_t* out_data = out.mutable_data_ptr<uint8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -253,7 +259,7 @@ Tensor& quantize_impl(
zero_point_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == (ScalarType)Bits4) {
+ } else if ((out.scalar_type() == (ScalarType)Bits4) && (optimized)) {
int8_t* out_data = out.mutable_data_ptr<int8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -391,7 +397,7 @@ Tensor& quantize_impl(
#undef ASYM_QUANTIZE_IMPL_CHANNEL
}
} else {
- if (out.scalar_type() == ScalarType::Byte) {
+ if ((out.scalar_type() == ScalarType::Byte) && (optimized)) {
uint8_t* out_data = out.mutable_data_ptr<uint8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -405,7 +411,7 @@ Tensor& quantize_impl(
scale_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == ScalarType::Char) {
+ } else if ((out.scalar_type() == ScalarType::Char) && (optimized)) {
int8_t* out_data = out.mutable_data_ptr<int8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -419,7 +425,7 @@ Tensor& quantize_impl(
scale_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == ScalarType::UInt16) {
+ } else if ((out.scalar_type() == ScalarType::UInt16) && (optimized)) {
uint16_t* out_data = out.mutable_data_ptr<uint16_t>();
XT_KERNEL_CHECK(
ctx,
@@ -433,7 +439,7 @@ Tensor& quantize_impl(
scale_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == ScalarType::Short) {
+ } else if ((out.scalar_type() == ScalarType::Short) && (optimized)) {
int16_t* out_data = out.mutable_data_ptr<int16_t>();
XT_KERNEL_CHECK(
ctx,
@@ -447,7 +453,7 @@ Tensor& quantize_impl(
scale_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == (ScalarType)Bits4u) {
+ } else if ((out.scalar_type() == (ScalarType)Bits4u) && (optimized)) {
uint8_t* out_data = out.mutable_data_ptr<uint8_t>();
XT_KERNEL_CHECK(
ctx,
@@ -461,7 +467,7 @@ Tensor& quantize_impl(
scale_data,
quant_min,
quant_max);
- } else if (out.scalar_type() == (ScalarType)Bits4) {
+ } else if ((out.scalar_type() == (ScalarType)Bits4) && (optimized)) {
int8_t* out_data = out.mutable_data_ptr<int8_t>();
XT_KERNEL_CHECK(
ctx,
diff --git a/backends/cadence/fusion_g3/operators/op_slice_copy.cpp b/backends/cadence/fusion_g3/operators/op_slice_copy.cpp
index c481cf726b..9158eecf13 100644
--- a/backends/cadence/fusion_g3/operators/op_slice_copy.cpp
+++ b/backends/cadence/fusion_g3/operators/op_slice_copy.cpp
@@ -58,12 +58,6 @@ Tensor& slice_copy_Tensor_out(
int kTensorDimensionLimit = executorch::runtime::kTensorDimensionLimit;
#ifdef OP_ARG_CHECK
- ET_KERNEL_CHECK(
- ctx,
- torch::executor::check_slice_copy_args(in, dim, step, out),
- InvalidArgument,
- out);
-
ET_KERNEL_CHECK(
ctx,
executorch::runtime::tensors_have_same_dim_order(in, out),
@@ -101,12 +95,13 @@ Tensor& slice_copy_Tensor_out(
signed char* out_data = out.mutable_data_ptr<signed char>();
const signed char* const inp_data = in.const_data_ptr<signed char>();
- if ((out.scalar_type() == ScalarType::Int) ||
- (out.scalar_type() == ScalarType::Short) ||
- (out.scalar_type() == ScalarType::Char) ||
- (out.scalar_type() == ScalarType::UInt32) ||
- (out.scalar_type() == ScalarType::UInt16) ||
- (out.scalar_type() == ScalarType::Byte)) {
+ if ((out.scalar_type() == in.scalar_type()) &&
+ ((out.scalar_type() == ScalarType::Int) ||
+ (out.scalar_type() == ScalarType::Short) ||
+ (out.scalar_type() == ScalarType::Char) ||
+ (out.scalar_type() == ScalarType::UInt32) ||
+ (out.scalar_type() == ScalarType::UInt16) ||
+ (out.scalar_type() == ScalarType::Byte))) {
XT_KERNEL_CHECK(
ctx,
out,
@@ -122,6 +117,12 @@ Tensor& slice_copy_Tensor_out(
(int)dim,
get_element_size(out.scalar_type()));
} else {
+ ET_KERNEL_CHECK(
+ ctx,
+ torch::executor::check_slice_copy_args(in, dim, step, out),
+ InvalidArgument,
+ out);
+
torch::executor::compute_slice(in, dim, start, length, step, out);
}
diff --git a/backends/cadence/fusion_g3/operators/op_softmax.cpp b/backends/cadence/fusion_g3/operators/op_softmax.cpp
index ee87ebaf5a..14b128e928 100644
--- a/backends/cadence/fusion_g3/operators/op_softmax.cpp
+++ b/backends/cadence/fusion_g3/operators/op_softmax.cpp
@@ -39,14 +39,7 @@ Tensor& _softmax_out(
// Adjust for negative dim
dim = dim < 0 ? dim + executorch::runtime::nonzero_dim(in) : dim;
-
#ifdef OP_ARG_CHECK
- ET_KERNEL_CHECK(
- ctx,
- torch::executor::check_softmax_args(in, dim, half_to_float, out),
- InvalidArgument,
- out);
-
ET_KERNEL_CHECK(
ctx, resize_tensor(out, in.sizes()) == Error::Ok, InvalidArgument, out);
@@ -63,7 +56,8 @@ Tensor& _softmax_out(
inp_shapes[i] = in_size[i];
}
- if (out.scalar_type() == ScalarType::Float) {
+ if ((in.scalar_type() == ScalarType::Float) &&
+ (out.scalar_type() == ScalarType::Float)) {
const float* const inp_data = in.const_data_ptr<float>();
float* const out_data = out.mutable_data_ptr<float>();
int axis = dim;
@@ -77,6 +71,12 @@ Tensor& _softmax_out(
in.dim(),
&axis);
} else {
+ ET_KERNEL_CHECK(
+ ctx,
+ torch::executor::check_softmax_args(in, dim, half_to_float, out),
+ InvalidArgument,
+ out);
+
ET_SWITCH_FLOATH_TYPES(in.scalar_type(), ctx, "_softmax.out", CTYPE, [&]() {
const CTYPE* const in_data = in.const_data_ptr<CTYPE>();
CTYPE* const out_data = out.mutable_data_ptr<CTYPE>();
diff --git a/backends/cadence/fusion_g3/operators/op_sub.cpp b/backends/cadence/fusion_g3/operators/op_sub.cpp
index 4bae81c5b2..9bafec5df9 100644
--- a/backends/cadence/fusion_g3/operators/op_sub.cpp
+++ b/backends/cadence/fusion_g3/operators/op_sub.cpp
@@ -35,19 +35,6 @@ Tensor& sub_out(
const Scalar& alpha,
Tensor& out) {
#ifdef OP_ARG_CHECK
- ScalarType alpha_type =
- torch::executor::native::utils::get_scalar_dtype(alpha);
- // Check alpha type
- ET_KERNEL_CHECK(ctx, alpha_type != ScalarType::Bool, InvalidArgument, out);
-
- // Check Common Dtype
- ET_KERNEL_CHECK(
- ctx,
- (canCast(common_type, out.scalar_type()) &&
- canCast(alpha_type, common_type)),
- InvalidArgument,
- out);
-
// Check Dim Order
ET_KERNEL_CHECK(
ctx,
@@ -72,12 +59,12 @@ Tensor& sub_out(
int inp2_shape[kTensorDimensionLimit];
int out_shape[kTensorDimensionLimit];
- bool broadcast = 0;
+ bool broadcast = false;
int max_dim = a.dim() > b.dim() ? a.dim() : b.dim();
max_dim = out.dim() > max_dim ? out.dim() : max_dim;
- bool optimized = 1;
+ bool optimized = true;
for (int i = 0; i < max_dim; i++) {
out_shape[i] = 1;
@@ -103,16 +90,16 @@ Tensor& sub_out(
for (int i = 0; i < out.dim(); i++) {
if (((inp1_shape[i]) != (out_shape[i])) ||
((inp2_shape[i]) != (out_shape[i]))) {
- broadcast = 1;
+ broadcast = true;
}
}
- if (((broadcast == 1) && (max_dim > kTensorDimensionLimit)) ||
+ if (((broadcast) && (max_dim > kTensorDimensionLimit)) ||
(!(((a.scalar_type() == ScalarType::Int) ||
(a.scalar_type() == ScalarType::Float)) &&
(a.scalar_type() == b.scalar_type()) &&
(a.scalar_type() == out.scalar_type())))) {
- optimized = 0;
+ optimized = false;
}
if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
@@ -207,6 +194,19 @@ Tensor& sub_out(
ScalarType compute_type =
torch::executor::native::utils::get_compute_type(common_type);
+ ScalarType alpha_type =
+ torch::executor::native::utils::get_scalar_dtype(alpha);
+ // Check alpha type
+ ET_KERNEL_CHECK(ctx, alpha_type != ScalarType::Bool, InvalidArgument, out);
+
+ // Check Common Dtype
+ ET_KERNEL_CHECK(
+ ctx,
+ (canCast(common_type, out.scalar_type()) &&
+ canCast(alpha_type, common_type)),
+ InvalidArgument,
+ out);
+
ET_SWITCH_REAL_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
const CTYPE_COMPUTE val_alpha =
torch::executor::native::utils::scalar_to<CTYPE_COMPUTE>(alpha);
@@ -236,18 +236,6 @@ Tensor& sub_scalar_out(
const Scalar& alpha,
Tensor& out) {
#ifdef OP_ARG_CHECK
- ScalarType alpha_type =
- torch::executor::native::utils::get_scalar_dtype(alpha);
- // Check alpha type
- ET_KERNEL_CHECK(ctx, alpha_type != ScalarType::Bool, InvalidArgument, out);
-
- // Check Common Dtype
- ET_KERNEL_CHECK(
- ctx,
- (common_type == out.scalar_type() && canCast(alpha_type, common_type)),
- InvalidArgument,
- out);
-
// Check Dim Order
ET_KERNEL_CHECK(
ctx,
@@ -266,14 +254,16 @@ Tensor& sub_scalar_out(
// @lint-ignore CLANGTIDY facebook-hte-CArray
static constexpr const char op_name[] = "sub.Scalar_out";
- bool optimized = 1;
- ScalarType b_type = torch::executor::native::utils::get_scalar_dtype(b);
+ bool optimized = true;
if (!(((a.scalar_type() == ScalarType::Int) ||
(a.scalar_type() == ScalarType::Float)) &&
- (a.scalar_type() == b_type) &&
(a.scalar_type() == out.scalar_type()))) {
- optimized = 0;
+ optimized = false;
+ }
+
+ if ((b.isFloatingPoint()) && (a.scalar_type() == ScalarType::Int)) {
+ optimized = false;
}
if ((a.scalar_type() == ScalarType::Int) && (optimized)) {
@@ -322,6 +312,19 @@ Tensor& sub_scalar_out(
// Compute Dtype
ScalarType compute_type =
torch::executor::native::utils::get_compute_type(common_type);
+
+ ScalarType alpha_type =
+ torch::executor::native::utils::get_scalar_dtype(alpha);
+ // Check alpha type
+ ET_KERNEL_CHECK(ctx, alpha_type != ScalarType::Bool, InvalidArgument, out);
+
+ // Check Common Dtype
+ ET_KERNEL_CHECK(
+ ctx,
+ (common_type == out.scalar_type() && canCast(alpha_type, common_type)),
+ InvalidArgument,
+ out);
+
ET_SWITCH_REAL_TYPES(compute_type, ctx, op_name, CTYPE_COMPUTE, [&]() {
const CTYPE_COMPUTE val_b =
torch::executor::native::utils::scalar_to<CTYPE_COMPUTE>(b);