Skip to content

MIGraphX EP Add FP4 support #176

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

Merged
merged 3 commits into from
Oct 17, 2025
Merged

MIGraphX EP Add FP4 support #176

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1c3bacd
Support fp4 type in ORT (#25767)
hariharans29 Sep 3, 2025
077dd2a
[Core] Fix debug node input output compilation after Fp4 support was …
hariharans29 Sep 4, 2025
1a87ee2
Link FP4 types between OnnxRT and MIGraphX APIs
TedThemistokleous Oct 1, 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
14 changes: 7 additions & 7 deletions .github/workflows/linux_minimal_build.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -325,7 +325,7 @@ jobs:
--build_wheel
--use_binskim_compliant_compile_flags
--disable_ml_ops
--disable_types sparsetensor float8 optional
--disable_types sparsetensor float4 float8 optional
--include_ops_by_config /onnxruntime_src/build/.test_data/include_no_operators.config
--cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=OFF

Expand All @@ -341,7 +341,7 @@ jobs:
--build_wheel
--use_binskim_compliant_compile_flags
--disable_ml_ops
--disable_types sparsetensor float8 optional
--disable_types sparsetensor float4 float8 optional
--include_ops_by_config /onnxruntime_src/build/.test_data/include_no_operators.config
--cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=OFF

Expand All @@ -358,7 +358,7 @@ jobs:
--build_wheel
--use_binskim_compliant_compile_flags
--disable_ml_ops
--disable_types sparsetensor float8 optional
--disable_types sparsetensor float4 float8 optional
--include_ops_by_config /onnxruntime_src/build/.test_data/include_no_operators.config
--cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=OFF

Expand Down Expand Up @@ -408,7 +408,7 @@ jobs:
--disable_ml_ops
--skip_tests
--enable_reduced_operator_type_support
--disable_types sparsetensor optional float8
--disable_types sparsetensor optional float4 float8
--include_ops_by_config /onnxruntime_src/build/.test_data/include_no_operators.config
--cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=OFF

Expand All @@ -427,7 +427,7 @@ jobs:
--disable_ml_ops
--skip_tests
--enable_reduced_operator_type_support
--disable_types sparsetensor optional float8
--disable_types sparsetensor optional float4 float8
--include_ops_by_config /onnxruntime_src/build/.test_data/include_no_operators.config
--cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=OFF

Expand Down Expand Up @@ -483,7 +483,7 @@ jobs:
--disable_ml_ops
--skip_tests
--enable_reduced_operator_type_support
--disable_types sparsetensor optional float8
--disable_types sparsetensor optional float4 float8
--include_ops_by_config /onnxruntime_src/build/.test_data/include_no_operators.config
--cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=OFF

Expand All @@ -502,7 +502,7 @@ jobs:
--disable_ml_ops
--skip_tests
--enable_reduced_operator_type_support
--disable_types sparsetensor optional float8
--disable_types sparsetensor optional float4 float8
--include_ops_by_config /onnxruntime_src/build/.test_data/include_no_operators.config
--cmake_extra_defines onnxruntime_BUILD_UNIT_TESTS=OFF

Expand Down
5 changes: 5 additions & 0 deletions cmake/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -155,6 +155,7 @@ option(onnxruntime_DISABLE_ML_OPS "Disable traditional ML ops" OFF)
option(onnxruntime_DISABLE_SPARSE_TENSORS "Disable sparse tensors data types" OFF)
option(onnxruntime_DISABLE_OPTIONAL_TYPE "Disable optional type" OFF)
option(onnxruntime_DISABLE_FLOAT8_TYPES "Disable float 8 types" OFF)
option(onnxruntime_DISABLE_FLOAT4_TYPES "Disable float 4 types" OFF)
option(onnxruntime_MINIMAL_BUILD "Exclude as much as possible from the build. Support ORT format models. No support for ONNX format models." OFF)
option(onnxruntime_CLIENT_PACKAGE_BUILD "Enables default settings that are more appropriate for client/on-device workloads." OFF)
cmake_dependent_option(onnxruntime_DISABLE_RTTI "Disable RTTI" ON "NOT onnxruntime_ENABLE_PYTHON;NOT onnxruntime_USE_CUDA" OFF)
Expand Down Expand Up @@ -1029,6 +1030,10 @@ function(onnxruntime_set_compile_flags target_name)
target_compile_definitions(${target_name} PRIVATE DISABLE_FLOAT8_TYPES)
endif()

if (onnxruntime_DISABLE_FLOAT4_TYPES)
target_compile_definitions(${target_name} PRIVATE DISABLE_FLOAT4_TYPES)
endif()

if (onnxruntime_ENABLE_ATEN)
target_compile_definitions(${target_name} PRIVATE ENABLE_ATEN)
endif()
Expand Down
10 changes: 6 additions & 4 deletions docs/OperatorKernels.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -622,10 +622,12 @@ Do not modify directly.*
|||14|**T** = tensor(double), tensor(float), tensor(float16)<br/> **U** = tensor(double), tensor(float), tensor(float16)|
|||[9, 13]|**T** = tensor(double), tensor(float), tensor(float16)|
|||[7, 8]|**T** = tensor(double), tensor(float), tensor(float16)|
|Cast|*in* input:**T1**<br> *out* output:**T2**|19+|**T1** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[13, 18]|**T1** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[9, 12]|**T1** = tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[6, 8]|**T1** = tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|Cast|*in* input:**T1**<br> *out* output:**T2**|23+|**T1** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float4e2m1), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float4e2m1), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[21, 22]|**T1** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float4e2m1), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[19, 20]|**T1** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float4e2m1), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[13, 18]|**T1** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float4e2m1), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[9, 12]|**T1** = tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float4e2m1), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|||[6, 8]|**T1** = tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)<br/> **T2** = tensor(bfloat16), tensor(bool), tensor(double), tensor(float), tensor(float16), tensor(float4e2m1), tensor(float8e4m3fn), tensor(float8e5m2), tensor(int16), tensor(int32), tensor(int64), tensor(int8), tensor(uint16), tensor(uint32), tensor(uint64), tensor(uint8)|
|Ceil|*in* X:**T**<br> *out* Y:**T**|13+|**T** = tensor(double), tensor(float), tensor(float16)|
|||[6, 12]|**T** = tensor(double), tensor(float), tensor(float16)|
|Clip|*in* input:**T**<br> *in* min:**T**<br> *in* max:**T**<br> *out* output:**T**<br><br>or<br><br>*in* input:**T**<br> *out* output:**T**|13+|**T** = tensor(double), tensor(float), tensor(float16), tensor(int64), tensor(int8), tensor(uint64), tensor(uint8)|
Expand Down
13 changes: 12 additions & 1 deletion include/onnxruntime/core/framework/data_types.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,7 @@
#include "core/framework/float8.h"
#include "core/framework/float16.h"
#include "core/framework/int4.h"
#include "core/framework/float4.h"
#include "core/graph/onnx_protobuf.h"
#include "core/framework/to_tensor_proto_element_type.h"

Expand Down Expand Up @@ -209,6 +210,7 @@ class DataTypeImpl {
static const std::vector<MLDataType>& AllTensorTypesIRv4();
static const std::vector<MLDataType>& AllTensorTypesIRv9();
static const std::vector<MLDataType>& AllTensorTypesIRv10();
static const std::vector<MLDataType>& AllTensorTypesIRv11();

static const std::vector<MLDataType>& AllFixedSizeTensorTypes(); // up to IR4 (no float 8), deprecated
static const std::vector<MLDataType>& AllFixedSizeTensorTypesIRv4();
Expand Down Expand Up @@ -287,6 +289,10 @@ struct IsTensorContainedType : public IsAnyOf<T, float, uint8_t, int8_t, uint16_
#if !defined(DISABLE_FLOAT8_TYPES)
,
Float8E4M3FN, Float8E4M3FNUZ, Float8E5M2, Float8E5M2FNUZ
#endif
#if !defined(DISABLE_FLOAT4_TYPES)
,
Float4E2M1x2
#endif
> {
};
Expand All @@ -302,6 +308,10 @@ struct IsSparseTensorContainedType : public IsAnyOf<T, float, uint8_t, int8_t, u
#if !defined(DISABLE_FLOAT8_TYPES)
,
Float8E4M3FN, Float8E4M3FNUZ, Float8E5M2, Float8E5M2FNUZ
#endif
#if !defined(DISABLE_FLOAT4_TYPES)
,
Float4E2M1x2
#endif
> {
};
Expand Down Expand Up @@ -921,7 +931,7 @@ class OpaqueType : public NonTensorType<T> {
*
* \details This class contains an integer constant that can be
* used for input data type dispatching. This class also stores the number of subelements per size units.
* Example: For int4, the size unit is 1 byte and the number of subelements is 2.
* Example: For float4/int4, the size unit is 1 byte and the number of subelements is 2.
*
*/
class PrimitiveDataTypeBase : public DataTypeImpl {
Expand Down Expand Up @@ -1101,6 +1111,7 @@ inline const PrimitiveDataTypeBase* DataTypeImpl::AsPrimitiveDataType() const {
// Registers a subbyte primitive.
// Examples:
// - Int4x2 stores 2 packed 4-bit elements in 1 byte: ORT_*_SUBBYTE_TYPE(Int4x2, 2)
// - Float4E2M1x2 stores 2 packed 4-bit elements in 1 byte: ORT_*_SUBBYTE_TYPE(Float4E2M1x2, 2)
// - [not supported] Int3x8 could store 8 packed 3-bit elements in 3 bytes: ORT_*_SUBBYTE_TYPE(Int3x8, 8)
#define ORT_REGISTER_PRIM_SUBBYTE_TYPE(TYPE, NUM_SUB_ELEMS) \
template <> \
Expand Down
Loading
Loading