Half-precision functions are generally tested against the
single-precision reference. This causes double rounding: first to single
precision, then from there to half precision. For the most part, it is
good enough, but specifically in the case of fma, a correctly rounded
result is required and is not obtained, for instance for arguments
0x1.eacp+7, 0x1.3f4p+4, 0x1.c04p+14, which produce an exact result of
0x1.065fffp+15 which should be rounded to half-prefcision 0x1.064p+15,
but was previously first rounded to single-precision 0x1.066p+15, and
from there to half-precision 0x1.068p+15. Testing against reference_fmal
gives us sufficient precision that double rounding does not cause
issues.
The f_fma(..., FLUSHED) calls for FTZ testing cannot be updated the same
way but do not need to be: these calls all have at least one constant
operand of zero. If one operand is zero, double rounding cannot be an
issue.
* Enable fp16 in math bruteforce
* Added modernization of remaining half tests for consistency (issue #142, bruteforce)
* Added kernel types related corrections
* Added more fixes and general cleanup
* Corrected ULP values for half tests (issue #142, bruteforce)
* Corrected presubmit check for clang format
* Added support for ternary, unary_two_result and unary_two_result_i tests for cl_half (issue #142, bruteforce)
* Added missing condition due to vendor's review
* code format correction
* Added check for lack of support for denormals in binary_half scenario
* Corrected procedure to compute nextafter cl_half for flush-to-zero mode
* Added correction for external check of reference value for nextafter test
* Added correction due to code review request
* Changed quantity of tests performed for half in unary and macro_unary procedures from basic
* Added corrections related to code review:
-added binary_operator_half.cpp and binary_two_results_i_half.cpp
-address sanitizer errors fixed
-extending list of special half values
-removed unnecessary relaxed math references in half tests
-corrected conditions to verify ulp narrowing of computation results
-several refactoring and cosmetics corrections
* Print format correction due to failed CI check
* Corrected bug found in code review (fp16 bruteforce)
* Corrections related to code review (cl_khr_fp16 support according to #142)
-gHostFill missing support added
-special half values array extended
-cosmetics and unifying
* clang format applied
* consistency correction
* more consistency corrections for cl_fp16_khr supported tests
* Corrections related to code review (bureforce #142)
* Correction for i_unary_half test capacity
* Corrections related to capacity of cl_khr_fp16 tests in bruteforce (#142)
---------
Co-authored-by: Wawiorko, Grzegorz <grzegorz.wawiorko@intel.com>
