[Beignet] [PATCH V3] GBE: add param to switch the behavior of math func

[Guo Yejun]

Add OCL_STRICT_CONFORMANCE to switch the behavior of math func,
The funcs will be high precision with perf drops if it is 1, Fast
path with good enough precision will be selected if it is 0.

This change is to add the code basis, with 'sin' and 'cos' implemented
as examples, other math functions support will be added later.

Signed-off-by: Guo Yejun <yejun.guo at intel.com>
---
 backend/src/backend/program.cpp      |   12 ++++++++++++
 backend/src/builtin_vector_proto.def |    4 ++++
 backend/src/ocl_stdlib.tmpl.h        |    8 ++++++++
 3 files changed, 24 insertions(+)

diff --git a/backend/src/backend/program.cpp b/backend/src/backend/program.cpp
index 2492a8b..98fcded 100644
--- a/backend/src/backend/program.cpp
+++ b/backend/src/backend/program.cpp
@@ -459,12 +459,20 @@ namespace gbe {
 
   /*********************** End of Program class member function *************************/
 
+#define REDEF_MATH_FUNC(x) "#ifdef "#x"\n#undef "#x"\n#endif\n#define "#x" __gen_ocl_internal_fastpath_"#x"\n"
+  std::string ocl_mathfunc_fastpath_str =
+    REDEF_MATH_FUNC(sin)
+    REDEF_MATH_FUNC(cos)
+    "\n"
+  ;
+
   static void programDelete(gbe_program gbeProgram) {
     gbe::Program *program = (gbe::Program*)(gbeProgram);
     GBE_SAFE_DELETE(program);
   }
 
   BVAR(OCL_OUTPUT_BUILD_LOG, false);
+  BVAR(OCL_STRICT_CONFORMANCE, true);
   SVAR(OCL_PCH_PATH, PCH_OBJECT_DIR);
   SVAR(OCL_PCM_PATH, PCM_OBJECT_DIR);
 
@@ -760,6 +768,10 @@ namespace gbe {
     } else
       fwrite(ocl_stdlib_str.c_str(), strlen(ocl_stdlib_str.c_str()), 1, clFile);
 
+    if (!OCL_STRICT_CONFORMANCE) {
+        fwrite(ocl_mathfunc_fastpath_str.c_str(), strlen(ocl_mathfunc_fastpath_str.c_str()), 1, clFile);
+    }
+
     // Write the source to the cl file
     fwrite(source, strlen(source), 1, clFile);
     fclose(clFile);
diff --git a/backend/src/builtin_vector_proto.def b/backend/src/builtin_vector_proto.def
index 4393ad5..7bc7c48 100644
--- a/backend/src/builtin_vector_proto.def
+++ b/backend/src/builtin_vector_proto.def
@@ -129,6 +129,10 @@ gentype tanpi (gentype x)
 gentype tgamma (gentype)
 gentype trunc (gentype)
 
+##math function fast path
+gentype __gen_ocl_internal_fastpath_sin (gentype)
+gentype __gen_ocl_internal_fastpath_cos (gentype)
+
 ##half_native_math
 #gentype half_cos (gentype x)
 #gentype half_divide (gentype x, gentype y)
diff --git a/backend/src/ocl_stdlib.tmpl.h b/backend/src/ocl_stdlib.tmpl.h
index d191b8e..6f60577 100755
--- a/backend/src/ocl_stdlib.tmpl.h
+++ b/backend/src/ocl_stdlib.tmpl.h
@@ -4462,6 +4462,14 @@ INLINE_OVERLOADABLE  size_t get_image_array_size(image1d_array_t image)
   { return __gen_ocl_get_image_array_size(image); }
 #endif
 
+INLINE_OVERLOADABLE float __gen_ocl_internal_fastpath_sin(float x) {
+    return native_sin(x);
+}
+
+INLINE_OVERLOADABLE float __gen_ocl_internal_fastpath_cos(float x) {
+    return native_cos(x);
+}
+
 #pragma OPENCL EXTENSION cl_khr_fp64 : disable
 
 #undef DECL_IMAGE
-- 
1.7.9.5