From b9700d7ea4757e35406dada4b017ed8221f845bb Mon Sep 17 00:00:00 2001
From: Luo Xionghu <xionghu.luo@intel.com>
Date: Tue, 14 Oct 2014 08:08:46 +0800
Subject: [PATCH] add utest popcount for all types.

v2: add all types to test.
v3: fix signed type count bits error.

Signed-off-by: Luo Xionghu <xionghu.luo@intel.com>
Reviewed-by: Zhigang Gong <zhigang.gong@linux.intel.com>
---
 kernels/compiler_popcount.cl | 16 ++++++++++
 utests/CMakeLists.txt        |  1 +
 utests/compiler_popcount.cpp | 75 ++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 92 insertions(+)
 create mode 100644 kernels/compiler_popcount.cl
 create mode 100644 utests/compiler_popcount.cpp

diff --git a/kernels/compiler_popcount.cl b/kernels/compiler_popcount.cl
new file mode 100644
index 0000000..1636118
--- /dev/null
+++ b/kernels/compiler_popcount.cl
@@ -0,0 +1,16 @@
+#define TEST_TYPE(TYPE)                                       \
+kernel void test_##TYPE(global TYPE *src, global TYPE *dst) { \
+  int i = get_global_id(0);                                   \
+  dst[i] = popcount(src[i]);                                  \
+}
+
+TEST_TYPE(char)
+TEST_TYPE(uchar)
+TEST_TYPE(short)
+TEST_TYPE(ushort)
+TEST_TYPE(int)
+TEST_TYPE(uint)
+TEST_TYPE(long)
+TEST_TYPE(ulong)
+
+#undef TEST_TYPE
diff --git a/utests/CMakeLists.txt b/utests/CMakeLists.txt
index b45ecf9..1b8caca 100644
--- a/utests/CMakeLists.txt
+++ b/utests/CMakeLists.txt
@@ -41,6 +41,7 @@ set (utests_sources
   compiler_ceil.cpp
   compiler_clz_short.cpp
   compiler_clz_int.cpp
+  compiler_popcount.cpp
   compiler_convert_uchar_sat.cpp
   compiler_copy_buffer.cpp
   compiler_copy_image.cpp
diff --git a/utests/compiler_popcount.cpp b/utests/compiler_popcount.cpp
new file mode 100644
index 0000000..c960ae6
--- /dev/null
+++ b/utests/compiler_popcount.cpp
@@ -0,0 +1,75 @@
+#include "utest_helper.hpp"
+
+namespace {
+
+template<typename T>
+T get_max();
+
+#define DEF_TEMPLATE(TYPE, NAME)                                    \
+template <>                                                         \
+TYPE get_max<TYPE>()                                                \
+{                                                                   \
+  static TYPE max = CL_##NAME##_MAX;                                \
+  return max;                                                       \
+}                                                                   \
+                                                                    \
+template <>                                                         \
+u##TYPE get_max<u##TYPE>()                                          \
+{                                                                   \
+  static u##TYPE max = CL_U##NAME##_MAX;                            \
+  return max;                                                       \
+}
+
+DEF_TEMPLATE(int8_t, CHAR)
+DEF_TEMPLATE(int16_t, SHRT)
+DEF_TEMPLATE(int32_t, INT)
+DEF_TEMPLATE(int64_t, LONG)
+
+template<typename T>
+void test(const char *kernel_name, int s_type)
+{
+  const int n = sizeof(T) * 8;
+
+  // Setup kernel and buffers
+  OCL_CREATE_KERNEL_FROM_FILE("compiler_popcount", kernel_name);
+  OCL_CREATE_BUFFER(buf[0], 0, n * sizeof(T), NULL);
+  OCL_CREATE_BUFFER(buf[1], 0, n * sizeof(T), NULL);
+  OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]);
+  OCL_SET_ARG(1, sizeof(cl_mem), &buf[1]);
+  globals[0] = n;
+  locals[0] = n;
+
+  OCL_MAP_BUFFER(0);
+  ((T*)buf_data[0])[0] = 0;
+  for (int32_t i = 1; i < (int32_t) n; ++i){
+    ((T*)buf_data[0])[i] = get_max<T>() >> i;
+  }
+  OCL_UNMAP_BUFFER(0);
+
+  OCL_NDRANGE(1);
+
+  OCL_MAP_BUFFER(1);
+  OCL_ASSERT(((T*)buf_data[1])[0] == 0);
+  for (int i = 1; i < n; ++i){
+    OCL_ASSERT(((T*)buf_data[1])[i] == n-i-s_type);
+  }
+  OCL_UNMAP_BUFFER(1);
+}
+
+}
+
+#define compiler_popcount(type, kernel, s_type) \
+static void compiler_popcount_ ##type(void)\
+{\
+  test<type>(# kernel, s_type);\
+}\
+MAKE_UTEST_FROM_FUNCTION(compiler_popcount_ ## type);
+
+compiler_popcount(int8_t, test_char, 1)
+compiler_popcount(uint8_t, test_uchar, 0)
+compiler_popcount(int16_t, test_short, 1)
+compiler_popcount(uint16_t, test_ushort, 0)
+compiler_popcount(int32_t, test_int, 1)
+compiler_popcount(uint32_t, test_uint, 0)
+compiler_popcount(int64_t, test_long, 1)
+compiler_popcount(uint64_t, test_ulong, 0)
-- 
2.7.4