ahmed/OpenCL-CTS
1
0
Fork 0
mirror of https://github.com/KhronosGroup/OpenCL-CTS.git synced 2026-03-24 15:59:03 +00:00
Code Packages Projects Releases Wiki Activity

Fix 'fpclassify: ambiguous call' compile fail in MSVC 2022 (#2426)

Browse Source 
Similar to #2219, we see "'fpclassify': ambiguous call" error in
test_conformance\basic\test_fpmath.cpp
due to missing constexpr at
https://github.com/KhronosGroup/OpenCL-CTS/blob/9265cbb2c274/test_conformance/basic/test_fpmath.cpp#L104
This PR fixes the issue by moving utility function isnan_fp in
testHarness.h and use it.
Note this PR doesn't modify use of isnan in many tests where only
float/double values are checked.
This commit is contained in:
Wenju He
2025-08-06 00:08:04 +08:00
committed by GitHub
parent 9ca0126c54
commit e15c6eb760
12 changed files with 94 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
test_conformance/math_brute_force/utility.h
View File

@@ -19,6 +19,7 @@
#include "harness/compat.h"
#include "harness/rounding_mode.h"
#include "harness/fpcontrol.h"
#include "harness/mathHelpers.h"
#include "harness/testHarness.h"
#include "harness/ThreadPool.h"
#include "harness/conversions.h"
@@ -172,16 +173,6 @@ inline int IsFloatNaN(double x)
return ((u.u & 0x7fffffffU) > 0x7F800000U);
}
inline bool IsHalfNaN(const cl_half v)
{
// Extract FP16 exponent and mantissa
uint16_t h_exp = (((cl_half)v) >> (CL_HALF_MANT_DIG - 1)) & 0x1F;
uint16_t h_mant = ((cl_half)v) & 0x3FF;
// NaN test
return (h_exp == 0x1F && h_mant != 0);
}
inline bool IsHalfInfinity(const cl_half v)
{
// Extract FP16 exponent and mantissa