[Beignet] [PATCH 6/8] OCL20: add a cl_kernel pointer to gpgpu.

[Yang Rong]

Because in flush the command queue, must check the currunt flushed
command queue has device enqueue or not, it need the cl_kernel. So store
the cl_kernel pointer to gpgpu. And add two function intel_gpgpu_set_kernel
and intel_gpgpu_get_kernel for it.

Signed-off-by: Yang Rong <rong.r.yang at intel.com>
---
 src/cl_driver.h         | 18 +++++++++++++-----
 src/cl_driver_defs.c    |  3 +++
 src/intel/intel_gpgpu.c | 14 ++++++++++++++
 src/intel/intel_gpgpu.h |  1 +
 4 files changed, 31 insertions(+), 5 deletions(-)

diff --git a/src/cl_driver.h b/src/cl_driver.h
index 5a3053e..bf685c0 100644
--- a/src/cl_driver.h
+++ b/src/cl_driver.h
@@ -117,7 +117,7 @@ typedef enum gpu_command_status {
 typedef struct cl_gpgpu_kernel {
   const char *name;        /* kernel name and bo name */
   uint32_t grf_blocks;     /* register blocks kernel wants (in 8 reg blocks) */
-  uint32_t curbe_sz;         /* total size of all curbes */
+  uint32_t curbe_sz;       /* total size of all curbes */
   cl_buffer bo;            /* kernel code in the proper addr space */
   int32_t barrierID;       /* barrierID for _this_ kernel */
   uint32_t use_slm:1;      /* For gen7 (automatic barrier management) */
@@ -141,6 +141,12 @@ extern cl_gpgpu_sync_cb *cl_gpgpu_sync;
 typedef void (cl_gpgpu_bind_buf_cb)(cl_gpgpu, cl_buffer, uint32_t offset, uint32_t internal_offset, size_t size, uint8_t bti);
 extern cl_gpgpu_bind_buf_cb *cl_gpgpu_bind_buf;
 
+typedef void (cl_gpgpu_set_kernel_cb)(cl_gpgpu, void *);
+extern cl_gpgpu_set_kernel_cb *cl_gpgpu_set_kernel;
+
+typedef void* (cl_gpgpu_get_kernel_cb)(cl_gpgpu);
+extern cl_gpgpu_get_kernel_cb *cl_gpgpu_get_kernel;
+
 /* bind samplers defined in both kernel and kernel args. */
 typedef void (cl_gpgpu_bind_sampler_cb)(cl_gpgpu, uint32_t *samplers, size_t sampler_sz);
 extern cl_gpgpu_bind_sampler_cb *cl_gpgpu_bind_sampler;
@@ -289,7 +295,6 @@ typedef void (cl_gpgpu_walker_cb)(cl_gpgpu,
                                   const size_t global_wk_sz[3],
                                   const size_t local_wk_sz[3]);
 extern cl_gpgpu_walker_cb *cl_gpgpu_walker;
-
 /**************************************************************************
  * Buffer
  **************************************************************************/
@@ -300,14 +305,17 @@ extern cl_buffer_alloc_cb *cl_buffer_alloc;
 typedef cl_buffer (cl_buffer_alloc_userptr_cb)(cl_buffer_mgr, const char*, void *, size_t, unsigned long);
 extern cl_buffer_alloc_userptr_cb *cl_buffer_alloc_userptr;
 
-typedef cl_buffer (cl_buffer_set_softpin_offset_cb)(cl_buffer, uint64_t);
+typedef int (cl_buffer_set_softpin_offset_cb)(cl_buffer, uint64_t);
 extern cl_buffer_set_softpin_offset_cb *cl_buffer_set_softpin_offset;
 
-typedef cl_buffer (cl_buffer_set_bo_use_full_range_cb)(cl_buffer, uint32_t);
+typedef int (cl_buffer_set_bo_use_full_range_cb)(cl_buffer, uint32_t);
 extern cl_buffer_set_bo_use_full_range_cb *cl_buffer_set_bo_use_full_range;
 
+typedef int (cl_buffer_disable_reuse_cb)(cl_buffer);
+extern cl_buffer_disable_reuse_cb *cl_buffer_disable_reuse;
+
 /* Set a buffer's tiling mode */
-typedef cl_buffer (cl_buffer_set_tiling_cb)(cl_buffer, int tiling, size_t stride);
+typedef int (cl_buffer_set_tiling_cb)(cl_buffer, int tiling, size_t stride);
 extern cl_buffer_set_tiling_cb *cl_buffer_set_tiling;
 
 #include "cl_context.h"
diff --git a/src/cl_driver_defs.c b/src/cl_driver_defs.c
index ae4262d..c1ae868 100644
--- a/src/cl_driver_defs.c
+++ b/src/cl_driver_defs.c
@@ -34,6 +34,7 @@ LOCAL cl_buffer_alloc_cb *cl_buffer_alloc = NULL;
 LOCAL cl_buffer_alloc_userptr_cb *cl_buffer_alloc_userptr = NULL;
 LOCAL cl_buffer_set_softpin_offset_cb *cl_buffer_set_softpin_offset = NULL;
 LOCAL cl_buffer_set_bo_use_full_range_cb *cl_buffer_set_bo_use_full_range = NULL;
+LOCAL cl_buffer_disable_reuse_cb *cl_buffer_disable_reuse = NULL;
 LOCAL cl_buffer_set_tiling_cb *cl_buffer_set_tiling = NULL;
 LOCAL cl_buffer_alloc_from_texture_cb *cl_buffer_alloc_from_texture = NULL;
 LOCAL cl_buffer_release_from_texture_cb *cl_buffer_release_from_texture = NULL;
@@ -99,4 +100,6 @@ LOCAL cl_gpgpu_unmap_printf_buffer_cb *cl_gpgpu_unmap_printf_buffer = NULL;
 LOCAL cl_gpgpu_set_printf_info_cb *cl_gpgpu_set_printf_info = NULL;
 LOCAL cl_gpgpu_get_printf_info_cb *cl_gpgpu_get_printf_info = NULL;
 LOCAL cl_gpgpu_release_printf_buffer_cb *cl_gpgpu_release_printf_buffer = NULL;
+LOCAL cl_gpgpu_set_kernel_cb *cl_gpgpu_set_kernel = NULL;
+LOCAL cl_gpgpu_get_kernel_cb *cl_gpgpu_get_kernel = NULL;
 
diff --git a/src/intel/intel_gpgpu.c b/src/intel/intel_gpgpu.c
index 914e9c3..1f3cc99 100644
--- a/src/intel/intel_gpgpu.c
+++ b/src/intel/intel_gpgpu.c
@@ -2127,6 +2127,18 @@ intel_gpgpu_get_printf_info(intel_gpgpu_t *gpgpu)
   return gpgpu->printf_info;
 }
 
+static void
+intel_gpgpu_set_kernel(intel_gpgpu_t *gpgpu, void * kernel)
+{
+  gpgpu->kernel = kernel;
+}
+
+static void*
+intel_gpgpu_get_kernel(intel_gpgpu_t *gpgpu)
+{
+  return gpgpu->kernel;
+}
+
 LOCAL void
 intel_set_gpgpu_callbacks(int device_id)
 {
@@ -2161,6 +2173,8 @@ intel_set_gpgpu_callbacks(int device_id)
   cl_gpgpu_release_printf_buffer = (cl_gpgpu_release_printf_buffer_cb *)intel_gpgpu_release_printf_buf;
   cl_gpgpu_set_printf_info = (cl_gpgpu_set_printf_info_cb *)intel_gpgpu_set_printf_info;
   cl_gpgpu_get_printf_info = (cl_gpgpu_get_printf_info_cb *)intel_gpgpu_get_printf_info;
+  cl_gpgpu_set_kernel = (cl_gpgpu_set_kernel_cb *)intel_gpgpu_set_kernel;
+  cl_gpgpu_get_kernel = (cl_gpgpu_get_kernel_cb *)intel_gpgpu_get_kernel;
 
   if (IS_BROADWELL(device_id) || IS_CHERRYVIEW(device_id)) {
     cl_gpgpu_bind_image = (cl_gpgpu_bind_image_cb *) intel_gpgpu_bind_image_gen8;
diff --git a/src/intel/intel_gpgpu.h b/src/intel/intel_gpgpu.h
index de7ddf6..15bfb2f 100644
--- a/src/intel/intel_gpgpu.h
+++ b/src/intel/intel_gpgpu.h
@@ -52,6 +52,7 @@ struct intel_gpgpu
   uint32_t target_buf_offset[max_buf_n];/* internal offset for buffers binded for the call */
   uint32_t binded_offset[max_buf_n];    /* their offsets in the curbe buffer */
   uint32_t binded_n;                    /* number of buffers binded */
+  void *kernel;                         /* cl_kernel with this gpgpu */
 
   unsigned long img_bitmap;              /* image usage bitmap. */
   unsigned int img_index_base;          /* base index for image surface.*/
-- 
1.9.1