[Beignet] [PATCH] BDW: set the BDW GT3's max_compute_unit to 47 per spec.

[Yang Rong]

Need double check BDW GT3's max thread count when SLM enable.

Signed-off-by: Yang Rong <rong.r.yang at intel.com>
---
 src/cl_device_id.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/src/cl_device_id.c b/src/cl_device_id.c
index e3b0c64..545b944 100644
--- a/src/cl_device_id.c
+++ b/src/cl_device_id.c
@@ -129,7 +129,7 @@ static struct _cl_device_id intel_brw_gt2_device = {
 
 static struct _cl_device_id intel_brw_gt3_device = {
   INIT_ICD(dispatch)
-  .max_compute_unit = 48,
+  .max_compute_unit = 47,
   .max_thread_per_unit = 7,
   .sub_slice_count = 6,
   .max_work_item_sizes = {1024, 1024, 1024},
@@ -660,8 +660,9 @@ cl_get_kernel_max_wg_sz(cl_kernel kernel)
       work_group_size = kernel->program->ctx->device->max_compute_unit *
                         kernel->program->ctx->device->max_thread_per_unit * simd_width;
   } else {
-    thread_cnt = kernel->program->ctx->device->max_compute_unit *
-                 kernel->program->ctx->device->max_thread_per_unit / kernel->program->ctx->device->sub_slice_count;
+    cl_device_id device = kernel->program->ctx->device;
+    thread_cnt = (device->max_compute_unit + device->sub_slice_count - 1) / device->sub_slice_count;
+    thread_cnt = thread_cnt * device->max_thread_per_unit;
     if(thread_cnt > 64)
       thread_cnt = 64;
     work_group_size = thread_cnt * simd_width;
-- 
1.9.1