ahmed/OpenCL-CTS
1
0
Fork 0
mirror of https://github.com/KhronosGroup/OpenCL-CTS.git synced 2026-03-25 00:09:02 +00:00
Code Packages Projects Releases Wiki Activity

Add fp16 testing to conversions and bruteforce (#1975)

Browse Source 
Merge the `fp16-staging` branch into `main`, adding fp16 (`half`)
testing to the conversions and math bruteforce tests.

---------

Signed-off-by: Ahmed Hesham <ahmed.hesham@arm.com>
Signed-off-by: Sven van Haastregt <sven.vanhaastregt@arm.com>
Signed-off-by: Guo, Yilong <yilong.guo@intel.com>
Signed-off-by: John Kesapides <john.kesapides@arm.com>
Co-authored-by: Marcin Hajder <marcin.hajder@gmail.com>
Co-authored-by: Ewan Crawford <ewan@codeplay.com>
Co-authored-by: Wawiorko, Grzegorz <grzegorz.wawiorko@intel.com>
Co-authored-by: Sreelakshmi Haridas Maruthur <sharidas@quicinc.com>
Co-authored-by: Harald van Dijk <harald@gigawatt.nl>
Co-authored-by: Ben Ashbaugh <ben.ashbaugh@intel.com>
Co-authored-by: Haonan Yang <haonan.yang@intel.com>
Co-authored-by: Ahmed Hesham <117350656+ahesham-arm@users.noreply.github.com>
Co-authored-by: niranjanjoshi121 <43807392+niranjanjoshi121@users.noreply.github.com>
Co-authored-by: Wenwan Xing <wenwan.xing@intel.com>
Co-authored-by: Yilong Guo <yilong.guo@intel.com>
Co-authored-by: Romaric Jodin <89833130+rjodinchr@users.noreply.github.com>
Co-authored-by: joshqti <127994991+joshqti@users.noreply.github.com>
Co-authored-by: Pekka Jääskeläinen <pekka.jaaskelainen@tuni.fi>
Co-authored-by: imilenkovic00 <155085410+imilenkovic00@users.noreply.github.com>
Co-authored-by: John Kesapides <46718829+JohnKesapidesARM@users.noreply.github.com>
Co-authored-by: Aharon Abramson <aharon.abramson@mobileye.com>
This commit is contained in:
Sven van Haastregt
2024-06-18 18:43:11 +02:00
committed by GitHub
parent b3c89ebde0
commit b6941b6c61
30 changed files with 7149 additions and 350 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
test_conformance/math_brute_force/utility.h
View File

@@ -1,5 +1,5 @@
//
// Copyright (c) 2017 The Khronos Group Inc.
// Copyright (c) 2017-2024 The Khronos Group Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
@@ -22,6 +22,7 @@
#include "harness/testHarness.h"
#include "harness/ThreadPool.h"
#include "harness/conversions.h"
#include "CL/cl_half.h"
#define BUFFER_SIZE (1024 * 1024 * 2)
#define EMBEDDED_REDUCTION_FACTOR (64)
@@ -61,10 +62,21 @@ extern int gFastRelaxedDerived;
extern int gWimpyMode;
extern int gHostFill;
extern int gIsInRTZMode;
extern int gHasHalf;
extern int gInfNanSupport;
extern int gIsEmbedded;
extern int gVerboseBruteForce;
extern uint32_t gMaxVectorSizeIndex;
extern uint32_t gMinVectorSizeIndex;
extern cl_device_fp_config gFloatCapabilities;
extern cl_device_fp_config gHalfCapabilities;
extern RoundingMode gFloatToHalfRoundingMode;
extern cl_half_rounding_mode gHalfRoundingMode;
#define HFF(num) cl_half_from_float(num, gHalfRoundingMode)
#define HFD(num) cl_half_from_double(num, gHalfRoundingMode)
#define HTF(num) cl_half_to_float(num)
#define LOWER_IS_BETTER 0
#define HIGHER_IS_BETTER 1
@@ -115,6 +127,12 @@ inline int IsFloatResultSubnormal(double x, float ulps)
return x < MAKE_HEX_DOUBLE(0x1.0p-126, 0x1, -126);
}
inline int IsHalfResultSubnormal(float x, float ulps)
{
x = fabs(x) - MAKE_HEX_FLOAT(0x1.0p-24, 0x1, -24) * ulps;
return x < MAKE_HEX_FLOAT(0x1.0p-14, 0x1, -14);
}
inline int IsFloatResultSubnormalAbsError(double x, float abs_err)
{
x = x - abs_err;
@@ -157,6 +175,26 @@ 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
uint16_t h_exp = (((cl_half)v) >> (CL_HALF_MANT_DIG - 1)) & 0x1F;
uint16_t h_mant = ((cl_half)v) & 0x3FF;
// Inf test
return (h_exp == 0x1F && h_mant == 0);
}
cl_uint RoundUpToNextPowerOfTwo(cl_uint x);
// Windows (since long double got deprecated) sets the x87 to 53-bit precision