Implement fmax using __builtin_fmax

This ensures correct handling of NaNi.

This has been tested with piglit, OpenCV, and the ocl conformance tests.

llvm-svn: 233713

3 files changed, 28 insertions, 8 deletions

diff --git a/libclc/generic/include/clc/math/fmax.h b/libclc/generic/include/clc/math/fmax.h
index d6956af85a5..71ee859be4f 100644
--- a/libclc/generic/include/clc/math/fmax.h
+++ b/libclc/generic/include/clc/math/fmax.h
@@ -1,8 +1,5 @@
-#undef fmax
-#define fmax __clc_fmax
-
 #define __CLC_BODY <clc/math/binary_decl.inc>
-#define __CLC_FUNCTION __clc_fmax
+#define __CLC_FUNCTION fmax
 
 #include <clc/math/gentype.inc>
 
diff --git a/libclc/generic/lib/math/fmax.cl b/libclc/generic/lib/math/fmax.cl
index 58583d6767a..239da3d3a61 100644
--- a/libclc/generic/lib/math/fmax.cl
+++ b/libclc/generic/lib/math/fmax.cl
@@ -1,11 +1,16 @@
 #include <clc/clc.h>
 
+#include "../clcmacro.h"
+
+_CLC_DEFINE_BINARY_BUILTIN(float, fmax, __builtin_fmaxf, float, float);
+
 #ifdef cl_khr_fp64
+
 #pragma OPENCL EXTENSION cl_khr_fp64 : enable
-#endif
 
-#define FUNCTION __clc_fmax
-#define FUNCTION_IMPL(x, y) ((x) < (y) ? (y) : (x))
+_CLC_DEFINE_BINARY_BUILTIN(double, fmax, __builtin_fmax, double, double);
+
+#endif
 
-#define __CLC_BODY <binary_impl.inc>
+#define __CLC_BODY <fmax.inc>
 #include <clc/math/gentype.inc>
diff --git a/libclc/generic/lib/math/fmax.inc b/libclc/generic/lib/math/fmax.inc
new file mode 100644
index 00000000000..8315c5fb6e0
--- /dev/null
+++ b/libclc/generic/lib/math/fmax.inc
@@ -0,0 +1,18 @@
+
+#if !defined(__CLC_SCALAR)
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE fmax(__CLC_GENTYPE x, float y) {
+  return fmax(x, (__CLC_GENTYPE)((__CLC_SCALAR_GENTYPE)y));
+}
+
+#ifdef cl_khr_fp64
+
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE fmax(__CLC_GENTYPE x, double y) {
+  return fmax(x, (__CLC_GENTYPE)((__CLC_SCALAR_GENTYPE)y));
+}
+
+#endif // ifdef cl_khr_fp64
+
+#endif // !defined(__CLC_SCALAR)