Skip to content

Commit 4476020

Browse files
author
morelos
committed
Update base for Update on "[ET-VK] double, short, and uint16 dtype runtime support"
Creating support for double, short, and uint16 for quantization ops. Registering the short keyword since theres already support. Also changing the cpu implementation to support half Differential Revision: [D75959063](https://our.internmc.facebook.com/intern/diff/D75959063/) [ghstack-poisoned]
2 parents 9923be9 + 4a14fdd commit 4476020

File tree

174 files changed

+1307
-828
lines changed

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.

174 files changed

+1307
-828
lines changed

‎.ci/scripts/test_llama.sh‎

Lines changed: 1 addition & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -156,8 +156,7 @@ cmake_install_executorch_libraries() {
156156
-DCMAKE_INSTALL_PREFIX=cmake-out \
157157
-DCMAKE_BUILD_TYPE="$CMAKE_BUILD_TYPE" \
158158
-DEXECUTORCH_BUILD_QNN="$QNN" \
159-
-DQNN_SDK_ROOT="$QNN_SDK_ROOT" \
160-
-Bcmake-out .
159+
-DQNN_SDK_ROOT="$QNN_SDK_ROOT"
161160
cmake --build cmake-out -j9 --target install --config "$CMAKE_BUILD_TYPE"
162161
}
163162

‎.ci/scripts/unittest-buck2.sh‎

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -25,7 +25,9 @@ BUILDABLE_KERNELS_PRIM_OPS_TARGETS=$(buck2 query //kernels/prim_ops/... | grep -
2525
# //runtime/kernel/... is failing because //third-party:torchgen_files's shell script can't find python on PATH.
2626
# //runtime/test/... requires Python torch, which we don't have in our OSS buck setup.
2727
for op in "build" "test"; do
28-
buck2 $op $BUILDABLE_OPTIMIZED_OPS //kernels/portable/... \
28+
buck2 $op $BUILDABLE_OPTIMIZED_OPS \
29+
//examples/selective_build:select_all_dtype_selective_lib_portable_lib \
30+
//kernels/portable/... \
2931
$BUILDABLE_KERNELS_PRIM_OPS_TARGETS //runtime/backend/... //runtime/core/... \
3032
//runtime/executor: //runtime/kernel/... //runtime/platform/...
3133
done

‎CMakeLists.txt‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -554,6 +554,10 @@ if(EXECUTORCH_BUILD_PTHREADPOOL AND EXECUTORCH_BUILD_CPUINFO)
554554
endif()
555555

556556
if(EXECUTORCH_BUILD_PYBIND)
557+
558+
# Add codegen tools subdirectory for selective_build pybind module
559+
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/codegen/tools)
560+
557561
if(NOT EXECUTORCH_BUILD_EXTENSION_DATA_LOADER)
558562
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/extension/data_loader)
559563
endif()

‎backends/cadence/fusion_g3/operators/op_clamp.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -21,13 +21,13 @@
2121
#include <executorch/kernels/portable/cpu/util/math_util.h>
2222
#include <executorch/runtime/kernel/kernel_includes.h>
2323

24-
using ::executorch::aten::optional;
2524
using ::executorch::aten::Scalar;
2625
using ::executorch::aten::ScalarType;
2726
using ::executorch::aten::Tensor;
2827
using ::executorch::runtime::canCast;
2928
using ::executorch::runtime::Error;
3029
using ::executorch::runtime::KernelRuntimeContext;
30+
using std::optional;
3131

3232
namespace cadence {
3333
namespace impl {

‎backends/cadence/fusion_g3/operators/op_dequantize.cpp‎

Lines changed: 10 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -24,7 +24,7 @@ using ::executorch::runtime::Error;
2424
using ::executorch::runtime::KernelRuntimeContext;
2525

2626
template <typename T>
27-
using optional = ::executorch::aten::optional<T>;
27+
using optional = std::optional<T>;
2828
/* ScalarType in Executorch do not have support for below data types.
2929
* So, creating a placeholder for these data types. Once, ScalarTypes is
3030
* updated to have support for below data types, these can be removed and
@@ -51,7 +51,7 @@ void check_dequantize_per_tensor_args(
5151
int64_t quant_min,
5252
int64_t quant_max,
5353
ScalarType dtype,
54-
::executorch::aten::optional<ScalarType>& out_dtype,
54+
std::optional<ScalarType>& out_dtype,
5555
Tensor& out) {
5656
ET_CHECK_MSG(
5757
input.scalar_type() == ScalarType::Byte ||
@@ -93,7 +93,7 @@ Tensor& dequantize_impl(
9393
float* scale_data,
9494
int* zero_point_data,
9595
int* axis,
96-
::executorch::aten::optional<ScalarType> out_dtype) {
96+
std::optional<ScalarType> out_dtype) {
9797
const ::executorch::aten::ArrayRef<Tensor::SizesType> input_size =
9898
input.sizes();
9999

@@ -260,8 +260,8 @@ Tensor& dequantize_impl(
260260
}
261261
}
262262

263-
::executorch::aten::optional<::executorch::aten::ArrayRef<int64_t>>
264-
optional_dim_list{::executorch::aten::ArrayRef<int64_t>{
263+
std::optional<::executorch::aten::ArrayRef<int64_t>> optional_dim_list{
264+
::executorch::aten::ArrayRef<int64_t>{
265265
dims, size_t(input.dim() - 1)}};
266266

267267
// Actual dequantization logic
@@ -466,8 +466,8 @@ Tensor& dequantize_impl(
466466
}
467467
}
468468

469-
::executorch::aten::optional<::executorch::aten::ArrayRef<int64_t>>
470-
optional_dim_list{::executorch::aten::ArrayRef<int64_t>{
469+
std::optional<::executorch::aten::ArrayRef<int64_t>> optional_dim_list{
470+
::executorch::aten::ArrayRef<int64_t>{
471471
dims, size_t(input.dim() - 1)}};
472472

473473
// Actual dequantization logic
@@ -600,7 +600,7 @@ Tensor& dequantize_per_tensor_tensor_args_out(
600600
int64_t quant_min,
601601
int64_t quant_max,
602602
ScalarType dtype,
603-
::executorch::aten::optional<ScalarType> out_dtype,
603+
std::optional<ScalarType> out_dtype,
604604
Tensor& out) {
605605
#ifdef OP_ARG_CHECK
606606
ET_CHECK_MSG(
@@ -639,12 +639,12 @@ Tensor& dequantize_per_channel_out(
639639
KernelRuntimeContext& context,
640640
const Tensor& input,
641641
const Tensor& scale,
642-
const ::executorch::aten::optional<Tensor>& opt_zero_points,
642+
const std::optional<Tensor>& opt_zero_points,
643643
int64_t axis,
644644
int64_t quant_min,
645645
int64_t quant_max,
646646
ScalarType dtype,
647-
::executorch::aten::optional<ScalarType> out_dtype,
647+
std::optional<ScalarType> out_dtype,
648648
Tensor& out) {
649649
if (axis < 0) {
650650
axis += executorch::runtime::nonzero_dim(input);

‎backends/cadence/fusion_g3/operators/op_div.cpp‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -19,14 +19,14 @@
1919
#include <executorch/runtime/kernel/kernel_includes.h>
2020
#include <executorch/runtime/platform/assert.h>
2121

22-
using ::executorch::aten::optional;
2322
using ::executorch::aten::Scalar;
2423
using ::executorch::aten::ScalarType;
25-
using ::executorch::aten::string_view;
2624
using ::executorch::aten::Tensor;
2725
using ::executorch::runtime::canCast;
2826
using ::executorch::runtime::Error;
2927
using ::executorch::runtime::KernelRuntimeContext;
28+
using std::optional;
29+
using std::string_view;
3030

3131
namespace cadence {
3232
namespace impl {
@@ -686,4 +686,4 @@ Tensor& div_scalar_mode_out(
686686
} // namespace native
687687
} // namespace G3
688688
} // namespace impl
689-
} // namespace cadence
689+
} // namespace cadence

‎backends/cadence/fusion_g3/operators/op_mean.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -17,11 +17,11 @@
1717
#include <executorch/runtime/platform/assert.h>
1818

1919
using ::executorch::aten::ArrayRef;
20-
using ::executorch::aten::optional;
2120
using ::executorch::aten::ScalarType;
2221
using ::executorch::aten::Tensor;
2322
using ::executorch::runtime::Error;
2423
using ::executorch::runtime::KernelRuntimeContext;
24+
using std::optional;
2525

2626
namespace cadence {
2727
namespace impl {

‎backends/cadence/fusion_g3/operators/op_native_layer_norm.cpp‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -19,11 +19,11 @@
1919
#include <executorch/runtime/kernel/kernel_includes.h>
2020

2121
using ::executorch::aten::IntArrayRef;
22-
using ::executorch::aten::optional;
2322
using ::executorch::aten::ScalarType;
2423
using ::executorch::aten::Tensor;
2524
using ::executorch::runtime::Error;
2625
using ::executorch::runtime::KernelRuntimeContext;
26+
using std::optional;
2727

2828
namespace cadence {
2929
namespace impl {

‎backends/cadence/fusion_g3/operators/op_quantize.cpp‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -329,8 +329,8 @@ Tensor& quantize_impl(
329329
}
330330
}
331331

332-
::executorch::aten::optional<::executorch::aten::ArrayRef<int64_t>>
333-
optional_dim_list{::executorch::aten::ArrayRef<int64_t>{
332+
std::optional<::executorch::aten::ArrayRef<int64_t>> optional_dim_list{
333+
::executorch::aten::ArrayRef<int64_t>{
334334
dims, size_t(input.dim() - 1)}};
335335

336336
// Actual quantization logic
@@ -534,8 +534,8 @@ Tensor& quantize_impl(
534534
}
535535
}
536536

537-
::executorch::aten::optional<::executorch::aten::ArrayRef<int64_t>>
538-
optional_dim_list{::executorch::aten::ArrayRef<int64_t>{
537+
std::optional<::executorch::aten::ArrayRef<int64_t>> optional_dim_list{
538+
::executorch::aten::ArrayRef<int64_t>{
539539
dims, size_t(input.dim() - 1)}};
540540

541541
// Actual quantization logic

‎backends/cadence/fusion_g3/operators/op_slice_copy.cpp‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -37,8 +37,8 @@ Tensor& slice_copy_Tensor_out(
3737
KernelRuntimeContext& ctx,
3838
const Tensor& in,
3939
int64_t dim,
40-
::executorch::aten::optional<int64_t> start_val,
41-
::executorch::aten::optional<int64_t> end_val,
40+
std::optional<int64_t> start_val,
41+
std::optional<int64_t> end_val,
4242
int64_t step,
4343
Tensor& out) {
4444
(void)ctx;

0 commit comments

Comments
 (0)