[SYCL][ABI-break] Remove fpga features from SYCL RT #20882

againull · 2025-12-12T00:43:55Z

See #16929 for details.
Removed the following extensions:
sycl_ext_intel_data_flow_pipes_properties
sycl_ext_intel_dataflow_pipes
sycl_ext_intel_fpga_datapath
sycl_ext_intel_fpga_device_selector
sycl_ext_intel_fpga_kernel_arg_properties
sycl_ext_intel_fpga_kernel_interface_properties
sycl_ext_intel_fpga_lsu
sycl_ext_intel_fpga_mem
sycl_ext_intel_fpga_reg
sycl_ext_intel_fpga_task_sequence
sycl_ext_intel_mem_channel_property
sycl_ext_oneapi_annotated_arg
Moved all documentation to removed folder because all listed extensions were implemented in the code, even though some of the extensions documentation remained in proposed folder.

Removed init_mode and implement_in_csr from sycl_ext_oneapi_device_global.

llvm/include/llvm/SYCLLowerIR/DeviceConfigFile.td

againull · 2025-12-15T21:01:47Z

sycl/doc/EnvironmentVariables.md

+| `ONEAPI_DEVICE_SELECTOR` | [See below.](#oneapi_device_selector)  | This device selection environment variable can be used to limit the choice of devices available when the SYCL-using application is run.  Useful for limiting devices to a certain type (like GPUs or CPUs) or backends (like Level Zero or OpenCL).  This device selection mechanism is replacing `SYCL_DEVICE_FILTER` .  The `ONEAPI_DEVICE_SELECTOR` syntax is shared with OpenMP and also allows sub-devices to be chosen. [See below.](#oneapi_device_selector) for a full description.   |
 | `ONEAPI_PVC_SEND_WAR_WA` | '1' or '0' | Controls the workaround for Erratum "FP64 register ordering violation" on Intel Ponte Vecchio GPUs. Setting `ONEAPI_PVC_SEND_WAR_WA=0` disables the workaround and is only safe if the secondary FP64 pipeline is disabled. Default is enabled ('1') and applied throughout the oneAPI software stack - including OneDNN, OneMKL, OpenCL and Level Zero Runtimes, and Intel Graphics Compiler. |
-| `SYCL_DEVICE_ALLOWLIST` | See [below](#sycl_device_allowlist) | Filter out devices that do not match the pattern specified. `BackendName` accepts `host`, `opencl`, `level_zero`, `native_cpu` or `cuda`. `DeviceType` accepts `host`, `cpu`, `gpu`, `fpga`, or `acc`. `fpga` and `acc` are handled in the same manner. `DeviceVendorId` accepts uint32_t in hex form (`0xXYZW`). `DriverVersion`, `PlatformVersion`, `DeviceName` and `PlatformName` accept regular expression. Special characters, such as parenthesis, must be escaped. DPC++ runtime will select only those devices which satisfy provided values above and regex. More than one device can be specified using the piping symbol "\|".|
+| `SYCL_DEVICE_ALLOWLIST` | See [below](#sycl_device_allowlist) | Filter out devices that do not match the pattern specified. `BackendName` accepts `host`, `opencl`, `level_zero`, `native_cpu` or `cuda`. `DeviceType` accepts `host`, `cpu`, `gpu`. `DeviceVendorId` accepts uint32_t in hex form (`0xXYZW`). `DriverVersion`, `PlatformVersion`, `DeviceName` and `PlatformName` accept regular expression. Special characters, such as parenthesis, must be escaped. DPC++ runtime will select only those devices which satisfy provided values above and regex. More than one device can be specified using the piping symbol "\|".|


Note for reviewers: SYCL_DEVICE_ALLOWLIST allowed to use fpga and acc interchangeably.
I removed both. Not sure if we need to save acc.

againull · 2025-12-15T21:02:21Z

sycl/include/sycl/info/info_desc.hpp

  cpu = UR_DEVICE_TYPE_CPU,
  gpu = UR_DEVICE_TYPE_GPU,
-  accelerator = UR_DEVICE_TYPE_FPGA,
+  accelerator = 0x10000,


Note for reviewers: I untied accelerator from UR_DEVICE_TYPE_FPGA and had to assign explicit value because otherwise it's value clashes with UR_DEVICE_TYPE_CPU.

Why untie accelerator from UR_DEVICE_TYPE_FPGA? Are we removing the latter?

againull · 2025-12-15T21:05:18Z

sycl/include/sycl/info/aspects.def

 __SYCL_ASPECT(ext_intel_gpu_subslices_per_slice, 22)
 __SYCL_ASPECT(ext_intel_gpu_eu_count_per_subslice, 23)
 __SYCL_ASPECT(ext_intel_max_mem_bandwidth, 24)
-__SYCL_ASPECT(ext_intel_mem_channel, 25)


Note for reviewers: I wonder if we need to update the IDs of the aspects that follow the removed aspect.
Some of the past PRs didn't update:
#13351
So I haven't updated either.

I think we should rework the Aspect IDs as well, but I won't mind if that's done in a separate PR.

againull · 2025-12-15T21:07:04Z

sycl/source/detail/platform_impl.cpp

  // handles to do a cleanup later
  std::vector<ur_device_handle_t> UrDevicesToCleanUp = UrDevices;

+  // Filter out FPGA devices since they are no longer supported.


Note for reviewers: Currently I am filtering out all UR_DEVICE_TYPE_FPGA devices returned by UR API.
I.e. we just ignore them and don't return from platform::get_devices and everywhere else.

I think this is not needed as FPGA devices are filtered out at UR level: #20014

Thanks, removed this code.

sarnex

esimd lgtm

adamfidel

Graph change to node LGTM.

sycl/source/detail/device_impl.hpp

sycl/source/detail/platform_impl.cpp

sycl/test-e2e/FilterSelector/select_device.cpp

bader

Nice clean-up. 10K lines less code to maintain. 👍

YuriPlyakhin

dpcpp-tools changes lgtm

uditagarwal97

LGTM overall

uditagarwal97 · 2025-12-16T04:12:57Z

sycl/include/sycl/ext/oneapi/experimental/annotated_usm/alloc_util.hpp

 // Transform a compile-time property list to a USM property_list (working at
-// runtime). Right now only the `buffer_location<N>` has its corresponding USM
+// runtime). Right there is no property that has its corresponding USM
 // runtime property and is transformable
 template <typename PropertyListT> inline property_list get_usm_property_list() {
-  if constexpr (PropertyListT::template has_property<buffer_location_key>()) {
-    return property_list{
-        sycl::ext::intel::experimental::property::usm::buffer_location(
-            PropertyListT::template get_property<buffer_location_key>().value)};
-  }
  return {};
 }


I wonder if we can completely get rid of this function and all its uses? As this function just returns an empty list.

uditagarwal97 · 2025-12-16T04:16:04Z

sycl/include/sycl/info/aspects.def

 __SYCL_ASPECT(ext_intel_gpu_subslices_per_slice, 22)
 __SYCL_ASPECT(ext_intel_gpu_eu_count_per_subslice, 23)
 __SYCL_ASPECT(ext_intel_max_mem_bandwidth, 24)
-__SYCL_ASPECT(ext_intel_mem_channel, 25)


I think we should rework the Aspect IDs as well, but I won't mind if that's done in a separate PR.

uditagarwal97 · 2025-12-16T04:19:44Z

sycl/include/sycl/info/info_desc.hpp

  cpu = UR_DEVICE_TYPE_CPU,
  gpu = UR_DEVICE_TYPE_GPU,
-  accelerator = UR_DEVICE_TYPE_FPGA,
+  accelerator = 0x10000,


Why untie accelerator from UR_DEVICE_TYPE_FPGA? Are we removing the latter?

againull had a problem deploying to WindowsCILock December 12, 2025 00:44 — with GitHub Actions Error

againull force-pushed the remove_fpga_features branch from 600c199 to 502506a Compare December 12, 2025 00:53

againull had a problem deploying to WindowsCILock December 12, 2025 00:54 — with GitHub Actions Failure

bader reviewed Dec 12, 2025

View reviewed changes

llvm/include/llvm/SYCLLowerIR/DeviceConfigFile.td Show resolved Hide resolved

bader reviewed Dec 12, 2025

View reviewed changes

llvm/include/llvm/SYCLLowerIR/DeviceConfigFile.td Outdated Show resolved Hide resolved

againull had a problem deploying to WindowsCILock December 12, 2025 01:34 — with GitHub Actions Failure

againull force-pushed the remove_fpga_features branch 2 times, most recently from 6f785e1 to 4ee0563 Compare December 15, 2025 20:44

againull had a problem deploying to WindowsCILock December 15, 2025 20:44 — with GitHub Actions Error

[SYCL][ABI-break] Remove fpga features from SYCL RT

876871d

againull force-pushed the remove_fpga_features branch from 4ee0563 to 876871d Compare December 15, 2025 20:46

againull temporarily deployed to WindowsCILock December 15, 2025 20:47 — with GitHub Actions Inactive

againull marked this pull request as ready for review December 15, 2025 21:20

againull requested review from a team as code owners December 15, 2025 21:20

againull requested review from adamfidel and uditagarwal97 December 15, 2025 21:20

againull commented Dec 15, 2025

View reviewed changes

againull temporarily deployed to WindowsCILock December 15, 2025 21:22 — with GitHub Actions Inactive

againull had a problem deploying to WindowsCILock December 15, 2025 21:22 — with GitHub Actions Failure

againull temporarily deployed to WindowsCILock December 15, 2025 21:22 — with GitHub Actions Inactive

sarnex approved these changes Dec 15, 2025

View reviewed changes

sarnex requested review from a team December 15, 2025 21:29

adamfidel approved these changes Dec 15, 2025

View reviewed changes

bader reviewed Dec 15, 2025

View reviewed changes

sycl/source/detail/device_impl.hpp Outdated Show resolved Hide resolved

bader reviewed Dec 15, 2025

View reviewed changes

sycl/source/detail/platform_impl.cpp Show resolved Hide resolved

bader reviewed Dec 15, 2025

View reviewed changes

sycl/source/detail/platform_impl.cpp Outdated Show resolved Hide resolved

bader requested changes Dec 15, 2025

View reviewed changes

sycl/test-e2e/FilterSelector/select_device.cpp Outdated Show resolved Hide resolved

Address review

722e609

againull had a problem deploying to WindowsCILock December 15, 2025 22:57 — with GitHub Actions Failure

bader approved these changes Dec 15, 2025

View reviewed changes

againull temporarily deployed to WindowsCILock December 15, 2025 23:32 — with GitHub Actions Inactive

YuriPlyakhin approved these changes Dec 15, 2025

View reviewed changes

Fix xfail marker

db2f833

againull temporarily deployed to WindowsCILock December 15, 2025 23:47 — with GitHub Actions Inactive

againull temporarily deployed to WindowsCILock December 16, 2025 00:25 — with GitHub Actions Inactive

uditagarwal97 reviewed Dec 16, 2025

View reviewed changes

[SYCL][ABI-break] Remove fpga features from SYCL RT #20882

Are you sure you want to change the base?

[SYCL][ABI-break] Remove fpga features from SYCL RT #20882

Conversation

againull commented Dec 12, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

againull Dec 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sarnex left a comment

Choose a reason for hiding this comment

Uh oh!

adamfidel left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

bader left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

YuriPlyakhin left a comment

Choose a reason for hiding this comment

Uh oh!

uditagarwal97 left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

againull commented Dec 12, 2025 •

edited

Loading

againull Dec 15, 2025 •

edited

Loading

bader left a comment •

edited

Loading