Tags: rishisinhanj/ao

[xpu][float8] ciflow fix: compile and base to check on CPU only machine

[xpu][mx][test] Enable test_mx_tensor on xpu

- Parametrize mx format kernel tests with device to include xpu.
- Add ids to existing test parameters for clearer test names.

Signed-off-by: Ula Golowicz <urszula.golowicz@intel.com>

[xpu][mx][test] Enable test_kernels on xpu

- Parametrize mx format kernel tests with device to include xpu.
- Add ids to existing test parameters for clearer test names.

Signed-off-by: Ula Golowicz <urszula.golowicz@intel.com>

[mx] RCEIL rounding for non-CUDA backends

The USE_PTX flag in MXFP8 triton kernels was gated by `not IS_ROCM`,
which assumed only CUDA and ROCm backends exist. This would cause
errors on other backends such as XPU.

Signed-off-by: Ula Golowicz <urszula.golowicz@intel.com>

Enable GPTQ on Intel GPU

Remove unnecessary Torch dependency from ExecuTorch ops build

The ExecuTorch ops target defines TORCHAO_SHARED_KERNELS_BUILD_EXECUTORCH=1,
which means shared kernel headers use ExecuTorch includes, not Torch includes.
The find_package(Torch) and TORCH_INCLUDE_DIRS were therefore unnecessary and
caused CMake configure failures in standalone ExecuTorch builds where PyTorch
is not discoverable via find_package.

Remove unnecessary Torch dependency from ExecuTorch ops build

The ExecuTorch ops target defines TORCHAO_SHARED_KERNELS_BUILD_EXECUTORCH=1,
which means shared kernel headers use ExecuTorch includes, not Torch includes.
The find_package(Torch) and TORCH_INCLUDE_DIRS were therefore unnecessary and
caused CMake configure failures in standalone ExecuTorch builds where PyTorch
is not discoverable via find_package.

Remove unnecessary Torch dependency from ExecuTorch ops build

The ExecuTorch ops target defines TORCHAO_SHARED_KERNELS_BUILD_EXECUTORCH=1,
which means shared kernel headers use ExecuTorch includes, not Torch includes.
The find_package(Torch) and TORCH_INCLUDE_DIRS were therefore unnecessary and
caused CMake configure failures in standalone ExecuTorch builds where PyTorch
is not discoverable via find_package.

Remove unnecessary Torch dependency from ExecuTorch ops build

The ExecuTorch ops target defines TORCHAO_SHARED_KERNELS_BUILD_EXECUTORCH=1,
which means shared kernel headers use ExecuTorch includes, not Torch includes.
The find_package(Torch) and TORCH_INCLUDE_DIRS were therefore unnecessary and
caused CMake configure failures in standalone ExecuTorch builds where PyTorch
is not discoverable via find_package.

Remove unnecessary Torch dependency from ExecuTorch ops build

The ExecuTorch ops target defines TORCHAO_SHARED_KERNELS_BUILD_EXECUTORCH=1,
which means shared kernel headers use ExecuTorch includes, not Torch includes.
The find_package(Torch) and TORCH_INCLUDE_DIRS were therefore unnecessary and
caused CMake configure failures in standalone ExecuTorch builds where PyTorch
is not discoverable via find_package.