Utest: Add subgroup work item test cases

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

1 files changed, 61 insertions, 0 deletions

diff --git a/utests/builtin_sub_group_size.cpp b/utests/builtin_sub_group_size.cpp
new file mode 100644
index 00000000..1dc24edb
--- /dev/null
+++ b/utests/builtin_sub_group_size.cpp
@@ -0,0 +1,61 @@
+/*
+According to the OpenCL cl_intel_subgroups.
+Now define local and global size as following:
+  globals[0] = 4;
+  globals[1] = 9;
+  globals[2] = 16;
+  locals[0] = 2;
+  locals[1] = 3;
+  locals[2] = 4;
+*/
+
+#define udebug 0
+#include "utest_helper.hpp"
+static void builtin_sub_group_size(void)
+{
+
+  // Setup kernel and buffers
+  size_t dim, i,local_sz = 1,buf_len = 1;
+  OCL_CREATE_KERNEL("builtin_sub_group_size");
+  size_t max_sub_sz;
+
+
+  OCL_CREATE_BUFFER(buf[0], CL_MEM_READ_WRITE, sizeof(int)*576, NULL);
+  OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]);
+
+  for( dim=1; dim <= 3; dim++ )
+  {
+    buf_len = 1;
+    local_sz = 1;
+    for(i=1; i <= dim; i++)
+    {
+      locals[i - 1] = i + 1;
+      globals[i - 1] = (i + 1) * (i + 1);
+      buf_len *= ((i + 1) * (i + 1));
+      local_sz *= i + 1;
+    }
+    for(i = dim+1; i <= 3; i++)
+    {
+      globals[i - 1] = 0;
+      locals[i - 1] = 0;
+    }
+
+    OCL_CALL(clGetKernelSubGroupInfoKHR,kernel,device,CL_KERNEL_MAX_SUB_GROUP_SIZE_FOR_NDRANGE_KHR,sizeof(size_t)*dim,locals,sizeof(size_t),&max_sub_sz,NULL);
+    // Run the kernel
+    OCL_NDRANGE( dim );
+    clFinish(queue);
+
+    OCL_MAP_BUFFER(0);
+
+    for( i = 0; i < buf_len; i++) {
+      size_t expect_sz = (i % local_sz) < (local_sz / max_sub_sz * max_sub_sz) ? max_sub_sz : (local_sz % max_sub_sz);
+#if udebug
+      printf("%zu get %d, expect %zu\n",i, ((uint32_t*)buf_data[0])[i], expect_sz);
+#endif
+      OCL_ASSERT( ((uint32_t*)buf_data[0])[i] == expect_sz);
+    }
+    OCL_UNMAP_BUFFER(0);
+  }
+}
+
+MAKE_UTEST_FROM_FUNCTION(builtin_sub_group_size);