Runtime: Add new API enums for cl_intel_required_subgroup_size extension

Add CL_DEVICE_SUB_GROUP_SIZES_INTEL for clGetDeviceInfo, add
CL_KERNEL_SPILL_MEM_SIZE_INTEL for clGetKernelWorkGroupInfo and add
CL_KERNEL_COMPILE_SUB_GROUP_SIZE_INTEL for clGetKernelSubGroupInfo.
We only have this extension for LLVM 40+ for frontend support.
V2: Add opencl-c define

Signed-off-by: Pan Xiuli <xiuli.pan@intel.com>
Reviewed-by: Yang Rong <rong.r.yang@intel.com>

7 files changed, 50 insertions, 0 deletions

diff --git a/backend/src/libocl/include/ocl.h b/backend/src/libocl/include/ocl.h
index dded5748..5819f8cb 100644
--- a/backend/src/libocl/include/ocl.h
+++ b/backend/src/libocl/include/ocl.h
@@ -126,6 +126,10 @@
 #define cl_intel_planar_yuv
 #define cl_intel_media_block_io
 
+#if __clang_major__*10 + __clang_minor__ > 40
+#define cl_intel_required_subgroup_size
+#endif
+
 #pragma OPENCL EXTENSION cl_khr_fp64 : disable
 #pragma OPENCL EXTENSION cl_khr_fp16 : disable
 #endif
diff --git a/include/CL/cl_intel.h b/include/CL/cl_intel.h
index 47bae46f..3cb85151 100644
--- a/include/CL/cl_intel.h
+++ b/include/CL/cl_intel.h
@@ -197,6 +197,12 @@ typedef CL_API_ENTRY cl_int
 						      void* /*param_value*/,
 						      size_t* /*param_value_size_ret*/ );
 #endif
+
+/* cl_intel_required_subgroup_size extension*/
+#define CL_DEVICE_SUB_GROUP_SIZES_INTEL                 0x4108
+#define CL_KERNEL_SPILL_MEM_SIZE_INTEL                  0x4109
+#define CL_KERNEL_COMPILE_SUB_GROUP_SIZE_INTEL          0x410A
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/cl_device_id.c b/src/cl_device_id.c
index 6cba2b57..76549a43 100644
--- a/src/cl_device_id.c
+++ b/src/cl_device_id.c
@@ -1377,6 +1377,10 @@ cl_get_device_info(cl_device_id     device,
       src_ptr = device->driver_version;
       src_size = device->driver_version_sz;
       break;
+    case CL_DEVICE_SUB_GROUP_SIZES_INTEL:
+      src_ptr = device->sub_group_sizes;
+      src_size = device->sub_group_sizes_sz;
+      break;
 
     default:
       return CL_INVALID_VALUE;
@@ -1520,6 +1524,7 @@ cl_get_kernel_workgroup_info(cl_kernel kernel,
     DECL_FIELD(COMPILE_WORK_GROUP_SIZE, kernel->compile_wg_sz)
     DECL_FIELD(PRIVATE_MEM_SIZE, kernel->stack_size)
     case CL_KERNEL_GLOBAL_WORK_SIZE:
+    {
       dimension = cl_check_builtin_kernel_dimension(kernel, device);
       if ( !dimension ) return CL_INVALID_VALUE;
       if (param_value_size_ret != NULL)
@@ -1537,6 +1542,18 @@ cl_get_kernel_workgroup_info(cl_kernel kernel,
         return CL_SUCCESS;
       }
       return CL_SUCCESS;
+    }
+    case CL_KERNEL_SPILL_MEM_SIZE_INTEL:
+    {
+      if (param_value && param_value_size < sizeof(cl_ulong))
+        return CL_INVALID_VALUE;
+      if (param_value_size_ret != NULL)
+        *param_value_size_ret = sizeof(cl_ulong);
+      if (param_value)
+        *(cl_ulong*)param_value = (cl_ulong)interp_kernel_get_scratch_size(kernel->opaque);
+      return CL_SUCCESS;
+    }
+
     default:
       return CL_INVALID_VALUE;
   };
@@ -1620,6 +1637,16 @@ cl_get_kernel_subgroup_info(cl_kernel kernel,
       }
       break;
     }
+    case CL_KERNEL_COMPILE_SUB_GROUP_SIZE_INTEL:
+    {
+      if (param_value && param_value_size < sizeof(size_t))
+        return CL_INVALID_VALUE;
+      if (param_value_size_ret != NULL)
+        *param_value_size_ret = sizeof(size_t);
+      if (param_value)
+        *(size_t*)param_value = interp_kernel_get_simd_width(kernel->opaque);
+      return CL_SUCCESS;
+    }
     default:
       return CL_INVALID_VALUE;
   };
diff --git a/src/cl_device_id.h b/src/cl_device_id.h
index 6b8f2ebe..93bd2f1f 100644
--- a/src/cl_device_id.h
+++ b/src/cl_device_id.h
@@ -136,6 +136,8 @@ struct _cl_device_id {
   uint32_t atomic_test_result;
   cl_uint image_pitch_alignment;
   cl_uint image_base_address_alignment;
+  size_t sub_group_sizes[2];
+  size_t sub_group_sizes_sz;
 
   //inited as NULL, created only when cmrt kernel is used
   void* cmrt_device;  //realtype: CmDevice*
diff --git a/src/cl_extensions.c b/src/cl_extensions.c
index d49d202f..56099ad0 100644
--- a/src/cl_extensions.c
+++ b/src/cl_extensions.c
@@ -69,8 +69,16 @@ check_intel_extension(cl_extensions_t *extensions)
 {
   int id;
   for(id = INTEL_EXT_START_ID; id <= INTEL_EXT_END_ID; id++)
+  {
     if(id != EXT_ID(intel_motion_estimation))
       extensions->extensions[id].base.ext_enabled = 1;
+    if(id == EXT_ID(intel_required_subgroup_size))
+#if LLVM_VERSION_MAJOR * 10 + LLVM_VERSION_MINOR > 40
+      extensions->extensions[id].base.ext_enabled = 1;
+#else
+      extensions->extensions[id].base.ext_enabled = 0;
+#endif
+  }
 }
 
 void
diff --git a/src/cl_extensions.h b/src/cl_extensions.h
index af0855e4..bb61c0bc 100644
--- a/src/cl_extensions.h
+++ b/src/cl_extensions.h
@@ -30,6 +30,7 @@
   DECL_EXT(intel_motion_estimation) \
   DECL_EXT(intel_subgroups) \
   DECL_EXT(intel_subgroups_short) \
+  DECL_EXT(intel_required_subgroup_size) \
   DECL_EXT(intel_media_block_io) \
   DECL_EXT(intel_planar_yuv)
 
diff --git a/src/cl_gt_device.h b/src/cl_gt_device.h
index ca4f3c57..f6cb5f82 100644
--- a/src/cl_gt_device.h
+++ b/src/cl_gt_device.h
@@ -153,4 +153,6 @@ DECL_INFO_STRING(spir_versions, "1.2")
 .partition_type = {0},
 .image_pitch_alignment = 1,
 .image_base_address_alignment = 4096,
+.sub_group_sizes = {8, 16},
+.sub_group_sizes_sz = sizeof(size_t) * 2,
 .cmrt_device = NULL