[Beignet] [PATCH 2/4] Runtime: fixed one missing case for clGetKernelWorkGroupInfo.

[Zhigang Gong]

CL_KERNEL_PRIVATE_MEM_SIZE is not implemented, this patch fix
this issue and can pass the piglit test case.

Signed-off-by: Zhigang Gong <zhigang.gong at intel.com>
---
 src/cl_command_queue_gen7.c |    2 +-
 src/cl_device_id.c          |   54 +++++++++++++++++++++----------------------
 src/cl_kernel.c             |    2 ++
 src/cl_kernel.h             |    1 +
 4 files changed, 30 insertions(+), 29 deletions(-)

diff --git a/src/cl_command_queue_gen7.c b/src/cl_command_queue_gen7.c
index 9faaafa..923a881 100644
--- a/src/cl_command_queue_gen7.c
+++ b/src/cl_command_queue_gen7.c
@@ -236,7 +236,7 @@ cl_bind_stack(cl_gpgpu gpgpu, cl_kernel ker)
 {
   cl_context ctx = ker->program->ctx;
   cl_device_id device = ctx->device;
-  const int32_t per_lane_stack_sz = gbe_kernel_get_stack_size(ker->opaque);
+  const int32_t per_lane_stack_sz = ker->stack_size;
   const int32_t value = GBE_CURBE_EXTRA_ARGUMENT;
   const int32_t sub_value = GBE_STACK_BUFFER;
   const int32_t offset = gbe_kernel_get_curbe_offset(ker->opaque, value, sub_value);
diff --git a/src/cl_device_id.c b/src/cl_device_id.c
index 1124d30..871cda3 100644
--- a/src/cl_device_id.c
+++ b/src/cl_device_id.c
@@ -259,17 +259,19 @@ cl_device_get_version(cl_device_id device, cl_int *ver)
 }
 #undef DECL_FIELD
 
-#define DECL_FIELD(CASE,FIELD)                                      \
-  case JOIN(CL_KERNEL_,CASE):                                       \
-      if (param_value_size < sizeof(((cl_device_id)NULL)->FIELD))   \
-        return CL_INVALID_VALUE;                                    \
-      if (param_value_size_ret != NULL)                             \
-        *param_value_size_ret = sizeof(((cl_device_id)NULL)->FIELD);\
-      memcpy(param_value,                                           \
-             &device->FIELD,                                        \
-             sizeof(((cl_device_id)NULL)->FIELD));                  \
+#define _DECL_FIELD(FIELD)                                 \
+      if (param_value && param_value_size < sizeof(FIELD)) \
+        return CL_INVALID_VALUE;                           \
+      if (param_value_size_ret != NULL)                    \
+        *param_value_size_ret = sizeof(FIELD);             \
+      if (param_value)                                     \
+        memcpy(param_value, &FIELD, sizeof(FIELD));        \
         return CL_SUCCESS;
 
+#define DECL_FIELD(CASE,FIELD)                             \
+  case JOIN(CL_KERNEL_,CASE):                              \
+  _DECL_FIELD(FIELD)
+
 #include "cl_kernel.h"
 LOCAL cl_int
 cl_get_kernel_workgroup_info(cl_kernel kernel,
@@ -279,31 +281,27 @@ cl_get_kernel_workgroup_info(cl_kernel kernel,
                              void* param_value,
                              size_t* param_value_size_ret)
 {
+  int err = CL_SUCCESS;
   if (UNLIKELY(device != &intel_ivb_gt1_device &&
                device != &intel_ivb_gt2_device))
     return CL_INVALID_DEVICE;
-  if (UNLIKELY(param_value == NULL))
-    return CL_INVALID_VALUE;
 
+  CHECK_KERNEL(kernel);
   switch (param_name) {
-    DECL_FIELD(WORK_GROUP_SIZE, wg_sz)
-    DECL_FIELD(PREFERRED_WORK_GROUP_SIZE_MULTIPLE, preferred_wg_sz_mul)
+    DECL_FIELD(WORK_GROUP_SIZE, device->wg_sz)
+    DECL_FIELD(PREFERRED_WORK_GROUP_SIZE_MULTIPLE, device->preferred_wg_sz_mul)
     case CL_KERNEL_LOCAL_MEM_SIZE:
-      if (param_value_size < sizeof(cl_ulong))
-        return CL_INVALID_VALUE;
-      if (param_value_size_ret != NULL)
-        *param_value_size_ret = sizeof(cl_ulong);
-      *(cl_ulong*)param_value = gbe_kernel_get_slm_size(kernel->opaque) + kernel->local_mem_sz;
-      return CL_SUCCESS;
-    case CL_KERNEL_COMPILE_WORK_GROUP_SIZE:
-      if (param_value_size < sizeof(kernel->compile_wg_sz))
-        return CL_INVALID_VALUE;
-      if (param_value_size_ret != NULL)
-        *param_value_size_ret = sizeof(kernel->compile_wg_sz);
-      memcpy(param_value, kernel->compile_wg_sz, sizeof(kernel->compile_wg_sz));
-      return CL_SUCCESS;
-
-    default: return CL_INVALID_VALUE;
+      {
+        size_t local_mem_sz =  gbe_kernel_get_slm_size(kernel->opaque) + kernel->local_mem_sz;
+        _DECL_FIELD(local_mem_sz)
+      }
+    DECL_FIELD(COMPILE_WORK_GROUP_SIZE, kernel->compile_wg_sz)
+    DECL_FIELD(PRIVATE_MEM_SIZE, kernel->stack_size)
+    default:
+      return CL_INVALID_VALUE;
   };
+
+error:
+  return err;
 }
 
diff --git a/src/cl_kernel.c b/src/cl_kernel.c
index 9a2a737..1869eba 100644
--- a/src/cl_kernel.c
+++ b/src/cl_kernel.c
@@ -231,6 +231,7 @@ cl_kernel_setup(cl_kernel k, gbe_kernel opaque)
   if (k->sampler_sz > 0)
     gbe_kernel_get_sampler_data(k->opaque, k->samplers);
   gbe_kernel_get_compile_wg_size(k->opaque, k->compile_wg_sz);
+  k->stack_size = gbe_kernel_get_stack_size(k->opaque);
   /* Get image data & size */
   k->image_sz = gbe_kernel_get_image_size(k->opaque);
   assert(k->sampler_sz <= GEN_MAX_SURFACES);
@@ -265,6 +266,7 @@ cl_kernel_dup(cl_kernel from)
   to->sampler_sz = from->sampler_sz;
   to->image_sz = from->image_sz;
   memcpy(to->compile_wg_sz, from->compile_wg_sz, sizeof(from->compile_wg_sz));
+  to->stack_size = from->stack_size;
   if (to->sampler_sz)
     memcpy(to->samplers, from->samplers, to->sampler_sz * sizeof(uint32_t));
   if (to->image_sz) {
diff --git a/src/cl_kernel.h b/src/cl_kernel.h
index 608ed8e..6524a69 100644
--- a/src/cl_kernel.h
+++ b/src/cl_kernel.h
@@ -59,6 +59,7 @@ struct _cl_kernel {
   cl_ulong local_mem_sz;      /* local memory size specified in kernel args. */
   size_t compile_wg_sz[3];    /* Required workgroup size by __attribute__((reqd_work_gro
                                  up_size(X, Y, Z))) qualifier.*/
+  size_t stack_size;          /* stack size per work item. */
   cl_argument *args;          /* To track argument setting */
   uint32_t arg_n:31;          /* Number of arguments */
   uint32_t ref_its_program:1; /* True only for the user kernel (created by clCreateKernel) */
-- 
1.7.9.5