[EXP][Command-buffer] OpenCL kernel command update

Implement the API for updating the kernel commands in a command-buffer
defined by oneapi-src#1089 for
the OpenCL adapter.

However, the following changes to the UR kernel update API have been
made based on implementation experience:

1. Forbid updating the work-dim of the kernel, see KhronosGroup/OpenCL-Docs#1057
2. Remove struct fields to update exec info, after [DPC++ implementation
   prototype](intel/llvm#12840) shows this isn't
   needed.

This adapter implementation depends on support for the
[cl_khr_command_buffer_mutable_dispatch](https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_Ext.html#cl_khr_command_buffer_mutable_dispatch)
extension.

Tested on Intel GPU/CPUs OpenCL implementations with the
[command-buffer emulation
layer](https://github.com/bashbaug/SimpleOpenCLSamples/tree/main/layers/10_cmdbufemu).

```bash
$ OPENCL_LAYERS=<path/to/SimpleOpenCLSamples/build/layers/10_cmdbufemu/libCmdBufEmu.so> ./bin/test-exp_command_buffer --platform="Intel(R) OpenCL Graphics"
```

DPC++ PR intel/llvm#12724

include/ur_api.h

@@ -270,7 +270,6 @@ typedef enum ur_structure_type_t {
  UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_MEMOBJ_ARG_DESC = 0x1002, ///< ::ur_exp_command_buffer_update_memobj_arg_desc_t
  UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_POINTER_ARG_DESC = 0x1003, ///< ::ur_exp_command_buffer_update_pointer_arg_desc_t
  UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_VALUE_ARG_DESC = 0x1004, ///< ::ur_exp_command_buffer_update_value_arg_desc_t
- UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_EXEC_INFO_DESC = 0x1005, ///< ::ur_exp_command_buffer_update_exec_info_desc_t
  UR_STRUCTURE_TYPE_EXP_SAMPLER_MIP_PROPERTIES = 0x2000, ///< ::ur_exp_sampler_mip_properties_t
  UR_STRUCTURE_TYPE_EXP_INTEROP_MEM_DESC = 0x2001, ///< ::ur_exp_interop_mem_desc_t
  UR_STRUCTURE_TYPE_EXP_INTEROP_SEMAPHORE_DESC = 0x2002, ///< ::ur_exp_interop_semaphore_desc_t
@@ -7896,19 +7895,6 @@ typedef struct ur_exp_command_buffer_update_value_arg_desc_t {
 
 } ur_exp_command_buffer_update_value_arg_desc_t;
 
-///////////////////////////////////////////////////////////////////////////////
-/// @brief Descriptor type for updating kernel command execution info.
-typedef struct ur_exp_command_buffer_update_exec_info_desc_t {
- ur_structure_type_t stype; ///< [in] type of this structure, must be
- ///< ::UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_EXEC_INFO_DESC
- const void *pNext; ///< [in][optional] pointer to extension-specific structure
- ur_kernel_exec_info_t propName; ///< [in] Name of execution attribute.
- size_t propSize; ///< [in] Size of execution attribute.
- const ur_kernel_exec_info_properties_t *pProperties; ///< [in][optional] Pointer to execution info properties.
- const void *pNewExecInfo; ///< [in] Pointer to memory location holding the execution info value.
-
-} ur_exp_command_buffer_update_exec_info_desc_t;
-
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Descriptor type for updating a kernel launch command.
 typedef struct ur_exp_command_buffer_update_kernel_launch_desc_t {
@@ -7918,16 +7904,13 @@ typedef struct ur_exp_command_buffer_update_kernel_launch_desc_t {
  uint32_t numNewMemObjArgs; ///< [in] Length of pNewMemObjArgList.
  uint32_t numNewPointerArgs; ///< [in] Length of pNewPointerArgList.
  uint32_t numNewValueArgs; ///< [in] Length of pNewValueArgList.
- uint32_t numNewExecInfos; ///< [in] Length of pNewExecInfoList.
  uint32_t newWorkDim; ///< [in] Number of work dimensions in the kernel ND-range, from 1-3.
  const ur_exp_command_buffer_update_memobj_arg_desc_t *pNewMemObjArgList; ///< [in][optional][range(0, numNewMemObjArgs)] An array describing the new
  ///< kernel mem obj arguments for the command.
  const ur_exp_command_buffer_update_pointer_arg_desc_t *pNewPointerArgList; ///< [in][optional][range(0, numNewPointerArgs)] An array describing the
  ///< new kernel pointer arguments for the command.
  const ur_exp_command_buffer_update_value_arg_desc_t *pNewValueArgList; ///< [in][optional][range(0, numNewValueArgs)] An array describing the new
  ///< kernel value arguments for the command.
- const ur_exp_command_buffer_update_exec_info_desc_t *pNewExecInfoList; ///< [in][optional][range(0, numNewExecInfos)] An array describing the
- ///< execution info objects for the command.
  size_t *pNewGlobalWorkOffset; ///< [in][optional][range(0, newWorkDim)] Array of newWorkDim unsigned
  ///< values that describe the offset used to calculate the global ID.
  size_t *pNewGlobalWorkSize; ///< [in][optional][range(0, newWorkDim)] Array of newWorkDim unsigned
@@ -7970,6 +7953,8 @@ typedef struct ur_exp_command_buffer_command_handle_t_ *ur_exp_command_buffer_co
 /// + `NULL == phCommandBuffer`
 /// - ::UR_RESULT_ERROR_INVALID_CONTEXT
 /// - ::UR_RESULT_ERROR_INVALID_DEVICE
+/// - ::UR_RESULT_ERROR_INVALID_OPERATION
+/// + If `pCommandBufferDesc->isUpdatable` is true and `hDevice` does not support UR_DEVICE_INFO_COMMAND_BUFFER_UPDATE_SUPPORT_EXP.
 /// - ::UR_RESULT_ERROR_OUT_OF_HOST_MEMORY
 /// - ::UR_RESULT_ERROR_OUT_OF_RESOURCES
 UR_APIEXPORT ur_result_t UR_APICALL
@@ -8571,6 +8556,7 @@ urCommandBufferReleaseCommandExp(
 /// - ::UR_RESULT_ERROR_INVALID_OPERATION
 /// + If ::ur_exp_command_buffer_desc_t::isUpdatable was not set to true on creation of the command buffer `hCommand` belongs to.
 /// + If the command-buffer `hCommand` belongs to has not been finalized.
+/// + If `pUpdateKernelLaunch->newWorkDim` is different from the work-dim used on creation of `hCommand`.
 /// - ::UR_RESULT_ERROR_INVALID_COMMAND_BUFFER_COMMAND_HANDLE_EXP
 /// - ::UR_RESULT_ERROR_INVALID_MEM_OBJECT
 /// - ::UR_RESULT_ERROR_INVALID_KERNEL_ARGUMENT_INDEX

include/ur_print.h

@@ -978,14 +978,6 @@ UR_APIEXPORT ur_result_t UR_APICALL urPrintExpCommandBufferUpdatePointerArgDesc(
 /// - `buff_size < out_size`
 UR_APIEXPORT ur_result_t UR_APICALL urPrintExpCommandBufferUpdateValueArgDesc(const struct ur_exp_command_buffer_update_value_arg_desc_t params, char *buffer, const size_t buff_size, size_t *out_size);
 
-///////////////////////////////////////////////////////////////////////////////
-/// @brief Print ur_exp_command_buffer_update_exec_info_desc_t struct
-/// @returns
-/// - ::UR_RESULT_SUCCESS
-/// - ::UR_RESULT_ERROR_INVALID_SIZE
-/// - `buff_size < out_size`
-UR_APIEXPORT ur_result_t UR_APICALL urPrintExpCommandBufferUpdateExecInfoDesc(const struct ur_exp_command_buffer_update_exec_info_desc_t params, char *buffer, const size_t buff_size, size_t *out_size);
-
 ///////////////////////////////////////////////////////////////////////////////
 /// @brief Print ur_exp_command_buffer_update_kernel_launch_desc_t struct
 /// @returns

include/ur_print.hpp

@@ -332,7 +332,6 @@ inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const struct
 inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const struct ur_exp_command_buffer_update_memobj_arg_desc_t params);
 inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const struct ur_exp_command_buffer_update_pointer_arg_desc_t params);
 inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const struct ur_exp_command_buffer_update_value_arg_desc_t params);
-inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const struct ur_exp_command_buffer_update_exec_info_desc_t params);
 inline std::ostream &operator<<(std::ostream &os, [[maybe_unused]] const struct ur_exp_command_buffer_update_kernel_launch_desc_t params);
 inline std::ostream &operator<<(std::ostream &os, enum ur_exp_peer_info_t value);
 
@@ -1044,9 +1043,6 @@ inline std::ostream &operator<<(std::ostream &os, enum ur_structure_type_t value
  case UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_VALUE_ARG_DESC:
  os << "UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_VALUE_ARG_DESC";
  break;
- case UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_EXEC_INFO_DESC:
- os << "UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_EXEC_INFO_DESC";
- break;
  case UR_STRUCTURE_TYPE_EXP_SAMPLER_MIP_PROPERTIES:
  os << "UR_STRUCTURE_TYPE_EXP_SAMPLER_MIP_PROPERTIES";
  break;
@@ -1282,11 +1278,6 @@ inline ur_result_t printStruct(std::ostream &os, const void *ptr) {
  printPtr(os, pstruct);
  } break;
 
- case UR_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_EXEC_INFO_DESC: {
- const ur_exp_command_buffer_update_exec_info_desc_t *pstruct = (const ur_exp_command_buffer_update_exec_info_desc_t *)ptr;
- printPtr(os, pstruct);
- } break;
-
  case UR_STRUCTURE_TYPE_EXP_SAMPLER_MIP_PROPERTIES: {
  const ur_exp_sampler_mip_properties_t *pstruct = (const ur_exp_sampler_mip_properties_t *)ptr;
  printPtr(os, pstruct);
@@ -9513,46 +9504,6 @@ inline std::ostream &operator<<(std::ostream &os, const struct ur_exp_command_bu
  return os;
 }
 ///////////////////////////////////////////////////////////////////////////////
-/// @brief Print operator for the ur_exp_command_buffer_update_exec_info_desc_t type
-/// @returns
-/// std::ostream &
-inline std::ostream &operator<<(std::ostream &os, const struct ur_exp_command_buffer_update_exec_info_desc_t params) {
- os << "(struct ur_exp_command_buffer_update_exec_info_desc_t){";
-
- os << ".stype = ";
-
- os << (params.stype);
-
- os << ", ";
- os << ".pNext = ";
-
- ur::details::printStruct(os,
- (params.pNext));
-
- os << ", ";
- os << ".propName = ";
-
- os << (params.propName);
-
- os << ", ";
- os << ".propSize = ";
-
- os << (params.propSize);
-
- os << ", ";
- os << ".pProperties = ";
-
- os << (params.pProperties);
-
- os << ", ";
- os << ".pNewExecInfo = ";
-
- os << (params.pNewExecInfo);
-
- os << "}";
- return os;
-}
-///////////////////////////////////////////////////////////////////////////////
 /// @brief Print operator for the ur_exp_command_buffer_update_kernel_launch_desc_t type
 /// @returns
 /// std::ostream &
@@ -9584,11 +9535,6 @@ inline std::ostream &operator<<(std::ostream &os, const struct ur_exp_command_bu
 
  os << (params.numNewValueArgs);
 
- os << ", ";
- os << ".numNewExecInfos = ";
-
- os << (params.numNewExecInfos);
-
  os << ", ";
  os << ".newWorkDim = ";
 
@@ -9627,17 +9573,6 @@ inline std::ostream &operator<<(std::ostream &os, const struct ur_exp_command_bu
  }
  os << "}";
 
- os << ", ";
- os << ".pNewExecInfoList = {";
- for (size_t i = 0; (params.pNewExecInfoList) != NULL && i < params.numNewExecInfos; ++i) {
- if (i != 0) {
- os << ", ";
- }
-
- os << ((params.pNewExecInfoList))[i];
- }
- os << "}";
-
  os << ", ";
  os << ".pNewGlobalWorkOffset = {";
  for (size_t i = 0; (params.pNewGlobalWorkOffset) != NULL && i < params.newWorkDim; ++i) {

scripts/core/EXP-COMMAND-BUFFER.rst

@@ -256,7 +256,6 @@ Enums
  * ${X}_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_MEMOBJ_ARG_DESC
  * ${X}_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_POINTER_ARG_DESC
  * ${X}_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_VALUE_ARG_DESC
- * ${X}_STRUCTURE_TYPE_EXP_COMMAND_BUFFER_UPDATE_EXEC_INFO_DESC
 * ${x}_command_t
  * ${X}_COMMAND_COMMAND_BUFFER_ENQUEUE_EXP
 * ${x}_function_t
@@ -290,7 +289,6 @@ Types
 * ${x}_exp_command_buffer_update_memobj_arg_desc_t
 * ${x}_exp_command_buffer_update_pointer_arg_desc_t
 * ${x}_exp_command_buffer_update_value_arg_desc_t
-* ${x}_exp_command_buffer_update_exec_info_desc_t
 * ${x}_exp_command_buffer_sync_point_t
 * ${x}_exp_command_buffer_handle_t
 * ${x}_exp_command_buffer_command_handle_t

scripts/core/exp-command-buffer.yml

@@ -63,9 +63,6 @@ etors:
  - name: EXP_COMMAND_BUFFER_UPDATE_VALUE_ARG_DESC
  desc: $x_exp_command_buffer_update_value_arg_desc_t
  value: "0x1004"
- - name: EXP_COMMAND_BUFFER_UPDATE_EXEC_INFO_DESC
- desc: $x_exp_command_buffer_update_exec_info_desc_t
- value: "0x1005"
 --- #--------------------------------------------------------------------------
 type: enum
 extend: true
@@ -163,24 +160,6 @@ members:
  desc: "[in][optional] Argument value representing matching kernel arg type to set at argument index."
 --- #--------------------------------------------------------------------------
 type: struct
-desc: "Descriptor type for updating kernel command execution info."
-base: $x_base_desc_t
-name: $x_exp_command_buffer_update_exec_info_desc_t
-members:
- - type: ur_kernel_exec_info_t
- name: propName
- desc: "[in] Name of execution attribute."
- - type: size_t
- name: propSize
- desc: "[in] Size of execution attribute."
- - type: "const ur_kernel_exec_info_properties_t *"
- name: pProperties
- desc: "[in][optional] Pointer to execution info properties."
- - type: "const void *"
- name: pNewExecInfo
- desc: "[in] Pointer to memory location holding the execution info value."
---- #--------------------------------------------------------------------------
-type: struct
 desc: "Descriptor type for updating a kernel launch command."
 base: $x_base_desc_t
 name: $x_exp_command_buffer_update_kernel_launch_desc_t
@@ -194,9 +173,6 @@ members:
  - type: uint32_t
  name: numNewValueArgs
  desc: "[in] Length of pNewValueArgList."
- - type: uint32_t
- name: numNewExecInfos
- desc: "[in] Length of pNewExecInfoList."
  - type: uint32_t
  name: newWorkDim
  desc: "[in] Number of work dimensions in the kernel ND-range, from 1-3."
@@ -209,9 +185,6 @@ members:
  - type: "const $x_exp_command_buffer_update_value_arg_desc_t*"
  name: pNewValueArgList
  desc: "[in][optional][range(0, numNewValueArgs)] An array describing the new kernel value arguments for the command."
- - type: "const $x_exp_command_buffer_update_exec_info_desc_t*"
- name: pNewExecInfoList
- desc: "[in][optional][range(0, numNewExecInfos)] An array describing the execution info objects for the command."
  - type: "size_t*"
  name: pNewGlobalWorkOffset
  desc: "[in][optional][range(0, newWorkDim)] Array of newWorkDim unsigned values that describe the offset used to calculate the global ID."
@@ -261,6 +234,8 @@ params:
 returns:
  - $X_RESULT_ERROR_INVALID_CONTEXT
  - $X_RESULT_ERROR_INVALID_DEVICE
+ - $X_RESULT_ERROR_INVALID_OPERATION:
+ - "If `pCommandBufferDesc->isUpdatable` is true and `hDevice` does not support UR_DEVICE_INFO_COMMAND_BUFFER_UPDATE_SUPPORT_EXP."
  - $X_RESULT_ERROR_OUT_OF_HOST_MEMORY
  - $X_RESULT_ERROR_OUT_OF_RESOURCES
 --- #--------------------------------------------------------------------------
@@ -924,6 +899,7 @@ returns:
  - $X_RESULT_ERROR_INVALID_OPERATION:
  - "If $x_exp_command_buffer_desc_t::isUpdatable was not set to true on creation of the command buffer `hCommand` belongs to."
  - "If the command-buffer `hCommand` belongs to has not been finalized."
+ - "If `pUpdateKernelLaunch->newWorkDim` is different from the work-dim used on creation of `hCommand`."
  - $X_RESULT_ERROR_INVALID_COMMAND_BUFFER_COMMAND_HANDLE_EXP
  - $X_RESULT_ERROR_INVALID_MEM_OBJECT
  - $X_RESULT_ERROR_INVALID_KERNEL_ARGUMENT_INDEX

source/adapters/cuda/command_buffer.cpp

@@ -867,6 +867,13 @@ UR_APIEXPORT ur_result_t UR_APICALL urCommandBufferUpdateKernelLaunchExp(
  return UR_RESULT_ERROR_INVALID_OPERATION;
  }
 
+ // Error if work dim changes
+ if (auto NewWorkDim = pUpdateKernelLaunch->newWorkDim) {
+ if (NewWorkDim != hCommand->WorkDim) {
+ return UR_RESULT_ERROR_INVALID_OPERATION;
+ }
+ }
+
  // Kernel corresponding to the command to update
  ur_kernel_handle_t Kernel = hCommand->Kernel;