Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add the half ulps values for embedded profile #2265

Merged
merged 2 commits into from
Feb 11, 2025
Merged

Add the half ulps values for embedded profile #2265

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e353abb
Add the half ulps values for embedded profile
AhmedAmraniAkdi Feb 7, 2025
ff1cf61
Formatting, and correct half ulps for embedded tanpi
AhmedAmraniAkdi Feb 10, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
225 changes: 117 additions & 108 deletions test_conformance/math_brute_force/function_list.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,31 +29,32 @@
// Only use ulps information in spir test
#ifdef FUNCTION_LIST_ULPS_ONLY

#define ENTRY(_name, _ulp, _embedded_ulp, _half_ulp, _rmode, _type) \
#define ENTRY(_name, _ulp, _embedded_ulp, _half_ulp, _half_embedded_ulp, \
_rmode, _type) \
{ \
STRINGIFY(_name), STRINGIFY(_name), { NULL }, { NULL }, { NULL }, \
_ulp, _ulp, _half_ulp, _embedded_ulp, INFINITY, INFINITY, _rmode, \
RELAXED_OFF, _type \
_ulp, _ulp, _half_ulp, _half_embedded_ulp, _embedded_ulp, \
INFINITY, INFINITY, _rmode, RELAXED_OFF, _type \
}
#define ENTRY_EXT(_name, _ulp, _embedded_ulp, _half_ulp, _relaxed_ulp, _rmode, \
_type, _relaxed_embedded_ulp) \
#define ENTRY_EXT(_name, _ulp, _embedded_ulp, _half_ulp, _half_embedded_ulp, \
_relaxed_ulp, _rmode, _type, _relaxed_embedded_ulp) \
{ \
STRINGIFY(_name), STRINGIFY(_name), { NULL }, { NULL }, { NULL }, \
_ulp, _ulp, _half_ulp, _embedded_ulp, _relaxed_ulp, \
_relaxed_embedded_ulp, _rmode, RELAXED_ON, _type \
_ulp, _ulp, _half_ulp, _half_embedded_ulp, _embedded_ulp, \
_relaxed_ulp, _relaxed_embedded_ulp, _rmode, RELAXED_ON, _type \
}
#define HALF_ENTRY(_name, _ulp, _embedded_ulp, _rmode, _type) \
{ \
"half_" STRINGIFY(_name), "half_" STRINGIFY(_name), { NULL }, \
{ NULL }, { NULL }, _ulp, _ulp, _ulp, _embedded_ulp, INFINITY, \
INFINITY, _rmode, RELAXED_OFF, _type \
{ NULL }, { NULL }, _ulp, _ulp, _ulp, _ulp, _embedded_ulp, \
INFINITY, INFINITY, _rmode, RELAXED_OFF, _type \
}
#define OPERATOR_ENTRY(_name, _operator, _ulp, _embedded_ulp, _half_ulp, \
_rmode, _type) \
_half_embedded_ulp, _rmode, _type) \
{ \
STRINGIFY(_name), _operator, { NULL }, { NULL }, { NULL }, _ulp, _ulp, \
_half_ulp, _embedded_ulp, INFINITY, INFINITY, _rmode, RELAXED_OFF, \
_type \
_half_ulp, _half_embedded_ulp, _embedded_ulp, INFINITY, INFINITY, \
_rmode, RELAXED_OFF, _type \
}

#define unaryF NULL
Expand Down Expand Up @@ -88,35 +89,37 @@

#else // FUNCTION_LIST_ULPS_ONLY

#define ENTRY(_name, _ulp, _embedded_ulp, _half_ulp, _rmode, _type) \
#define ENTRY(_name, _ulp, _embedded_ulp, _half_ulp, _half_embedded_ulp, \
_rmode, _type) \
{ \
STRINGIFY(_name), STRINGIFY(_name), { (void*)reference_##_name }, \
{ (void*)reference_##_name##l }, { (void*)reference_##_name }, \
_ulp, _ulp, _half_ulp, _embedded_ulp, INFINITY, INFINITY, _rmode, \
RELAXED_OFF, _type \
_ulp, _ulp, _half_ulp, _half_embedded_ulp, _embedded_ulp, \
INFINITY, INFINITY, _rmode, RELAXED_OFF, _type \
}
#define ENTRY_EXT(_name, _ulp, _embedded_ulp, _half_ulp, _relaxed_ulp, _rmode, \
_type, _relaxed_embedded_ulp) \
#define ENTRY_EXT(_name, _ulp, _embedded_ulp, _half_ulp, _half_embedded_ulp, \
_relaxed_ulp, _rmode, _type, _relaxed_embedded_ulp) \
{ \
STRINGIFY(_name), STRINGIFY(_name), { (void*)reference_##_name }, \
{ (void*)reference_##_name##l }, \
{ (void*)reference_##relaxed_##_name }, _ulp, _ulp, _half_ulp, \
_embedded_ulp, _relaxed_ulp, _relaxed_embedded_ulp, _rmode, \
RELAXED_ON, _type \
_half_embedded_ulp, _embedded_ulp, _relaxed_ulp, \
_relaxed_embedded_ulp, _rmode, RELAXED_ON, _type \
}
#define HALF_ENTRY(_name, _ulp, _embedded_ulp, _rmode, _type) \
{ \
"half_" STRINGIFY(_name), "half_" STRINGIFY(_name), \
{ (void*)reference_##_name }, { NULL }, { NULL }, _ulp, _ulp, \
_ulp, _embedded_ulp, INFINITY, INFINITY, _rmode, RELAXED_OFF, \
_type \
_ulp, _ulp, _embedded_ulp, INFINITY, INFINITY, _rmode, \
RELAXED_OFF, _type \
}
#define OPERATOR_ENTRY(_name, _operator, _ulp, _embedded_ulp, _half_ulp, \
_rmode, _type) \
_half_embedded_ulp, _rmode, _type) \
{ \
STRINGIFY(_name), _operator, { (void*)reference_##_name }, \
{ (void*)reference_##_name##l }, { NULL }, _ulp, _ulp, _half_ulp, \
_embedded_ulp, INFINITY, INFINITY, _rmode, RELAXED_OFF, _type \
_half_embedded_ulp, _embedded_ulp, INFINITY, INFINITY, _rmode, \
RELAXED_OFF, _type \
}

static constexpr vtbl _unary = {
Expand Down Expand Up @@ -252,19 +255,19 @@ static constexpr vtbl _mad_tbl = {

// clang-format off
const Func functionList[] = {
ENTRY_EXT(acos, 4.0f, 4.0f, 2.0f, 4096.0f, FTZ_OFF, unaryF, 4096.0f),
ENTRY(acosh, 4.0f, 4.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY(acospi, 5.0f, 5.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY_EXT(asin, 4.0f, 4.0f, 2.0f, 4096.0f, FTZ_OFF, unaryF, 4096.0f),
ENTRY(asinh, 4.0f, 4.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY(asinpi, 5.0f, 5.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY_EXT(atan, 5.0f, 5.0f, 2.0f, 4096.0f, FTZ_OFF, unaryF, 4096.0f),
ENTRY(atanh, 5.0f, 5.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY(atanpi, 5.0f, 5.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY(atan2, 6.0f, 6.0f, 2.0f, FTZ_OFF, binaryF),
ENTRY(atan2pi, 6.0f, 6.0f, 2.0f, FTZ_OFF, binaryF),
ENTRY(cbrt, 2.0f, 4.0f, 2.f, FTZ_OFF, unaryF),
ENTRY(ceil, 0.0f, 0.0f, 0.f, FTZ_OFF, unaryF),
ENTRY_EXT(acos, 4.0f, 4.0f, 2.0f, 3.0f, 4096.0f, FTZ_OFF, unaryF, 4096.0f),
ENTRY(acosh, 4.0f, 4.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY(acospi, 5.0f, 5.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY_EXT(asin, 4.0f, 4.0f, 2.0f, 3.0f, 4096.0f, FTZ_OFF, unaryF, 4096.0f),
ENTRY(asinh, 4.0f, 4.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY(asinpi, 5.0f, 5.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY_EXT(atan, 5.0f, 5.0f, 2.0f, 3.0f, 4096.0f, FTZ_OFF, unaryF, 4096.0f),
ENTRY(atanh, 5.0f, 5.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY(atanpi, 5.0f, 5.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY(atan2, 6.0f, 6.0f, 2.0f, 3.0f, FTZ_OFF, binaryF),
ENTRY(atan2pi, 6.0f, 6.0f, 2.0f, 3.0f, FTZ_OFF, binaryF),
ENTRY(cbrt, 2.0f, 4.0f, 2.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY(ceil, 0.0f, 0.0f, 0.f, 0.f, FTZ_OFF, unaryF),
{ "copysign",
"copysign",
{ (void*)reference_copysignf },
Expand All @@ -274,96 +277,97 @@ const Func functionList[] = {
0.0f,
0.0f,
0.0f,
0.0f,
svenvh marked this conversation as resolved.
Show resolved Hide resolved
INFINITY,
INFINITY,
FTZ_OFF,
RELAXED_OFF,
binaryF },
ENTRY_EXT(cos, 4.0f, 4.0f, 2.f, 0.00048828125f, FTZ_OFF, unaryF,
ENTRY_EXT(cos, 4.0f, 4.0f, 2.0f, 2.0f, 0.00048828125f, FTZ_OFF, unaryF,
0.00048828125f), // relaxed ulp 2^-11
ENTRY(cosh, 4.0f, 4.0f, 2.f, FTZ_OFF, unaryF),
ENTRY_EXT(cospi, 4.0f, 4.0f, 2.f, 0.00048828125f, FTZ_OFF, unaryF,
ENTRY(cosh, 4.0f, 4.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY_EXT(cospi, 4.0f, 4.0f, 2.0f, 2.0f, 0.00048828125f, FTZ_OFF, unaryF,
0.00048828125f), // relaxed ulp 2^-11
ENTRY(erfc, 16.0f, 16.0f, 4.0f, FTZ_OFF, unaryF),
ENTRY(erf, 16.0f, 16.0f, 4.0f, FTZ_OFF, unaryF),
ENTRY(erfc, 16.0f, 16.0f, 4.0f, 4.0f, FTZ_OFF, unaryF),
ENTRY(erf, 16.0f, 16.0f, 4.0f, 4.0f, FTZ_OFF, unaryF),

// relaxed error is overwritten in unary.c as it is 3+floor(fabs(2*x))
ENTRY_EXT(exp, 3.0f, 4.0f, 2.f, 3.0f, FTZ_OFF, unaryF, 4.0f),
ENTRY_EXT(exp, 3.0f, 4.0f, 2.0f, 3.0f, 3.0f, FTZ_OFF, unaryF, 4.0f),

// relaxed error is overwritten in unary.c as it is 3+floor(fabs(2*x))
ENTRY_EXT(exp2, 3.0f, 4.0f, 2.f, 3.0f, FTZ_OFF, unaryF, 4.0f),
ENTRY_EXT(exp2, 3.0f, 4.0f, 2.0f, 3.0f, 3.0f, FTZ_OFF, unaryF, 4.0f),

// relaxed error is overwritten in unary.c as it is 3+floor(fabs(2*x)) in derived mode;
// in non-derived mode it uses the ulp error for half_exp10.
ENTRY_EXT(exp10, 3.0f, 4.0f, 2.f, 8192.0f, FTZ_OFF, unaryF, 8192.0f),

ENTRY(expm1, 3.0f, 4.0f, 2.f, FTZ_OFF, unaryF),
ENTRY(fabs, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(fdim, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(floor, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(fma, 0.0f, 0.0f, 0.0f, FTZ_OFF, ternaryF),
ENTRY(fmax, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(fmin, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(fmod, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(fract, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF_two_results),
ENTRY(frexp, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF_two_results_i),
ENTRY(hypot, 4.0f, 4.0f, 2.0f, FTZ_OFF, binaryF),
ENTRY(ilogb, 0.0f, 0.0f, 0.0f, FTZ_OFF, i_unaryF),
ENTRY(isequal, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isfinite, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY(isgreater, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isgreaterequal, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isinf, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY(isless, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(islessequal, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(islessgreater, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isnan, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY(isnormal, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY(isnotequal, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isordered, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isunordered, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(ldexp, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF_i),
ENTRY(lgamma, INFINITY, INFINITY, INFINITY, FTZ_OFF, unaryF),
ENTRY(lgamma_r, INFINITY, INFINITY, INFINITY, FTZ_OFF,
ENTRY_EXT(exp10, 3.0f, 4.0f, 2.0f, 3.0f, 8192.0f, FTZ_OFF, unaryF, 8192.0f),

ENTRY(expm1, 3.0f, 4.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY(fabs, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(fdim, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(floor, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(fma, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, ternaryF),
ENTRY(fmax, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(fmin, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(fmod, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(fract, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF_two_results),
ENTRY(frexp, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF_two_results_i),
ENTRY(hypot, 4.0f, 4.0f, 2.0f, 3.0f, FTZ_OFF, binaryF),
ENTRY(ilogb, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, i_unaryF),
ENTRY(isequal, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isfinite, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY(isgreater, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isgreaterequal, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isinf, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY(isless, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(islessequal, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(islessgreater, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isnan, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY(isnormal, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY(isnotequal, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isordered, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(isunordered, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_binaryF),
ENTRY(ldexp, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF_i),
ENTRY(lgamma, INFINITY, INFINITY, INFINITY, INFINITY, FTZ_OFF, unaryF),
ENTRY(lgamma_r, INFINITY, INFINITY, INFINITY, INFINITY, FTZ_OFF,
unaryF_two_results_i),
ENTRY_EXT(log, 3.0f, 4.0f, 2.0f, 4.76837158203125e-7f, FTZ_OFF, unaryF,
ENTRY_EXT(log, 3.0f, 4.0f, 2.0f, 3.0f, 4.76837158203125e-7f, FTZ_OFF, unaryF,
4.76837158203125e-7f), // relaxed ulp 2^-21
ENTRY_EXT(log2, 3.0f, 4.0f, 2.0f, 4.76837158203125e-7f, FTZ_OFF, unaryF,
ENTRY_EXT(log2, 3.0f, 4.0f, 2.0f, 3.0f, 4.76837158203125e-7f, FTZ_OFF, unaryF,
4.76837158203125e-7f), // relaxed ulp 2^-21
ENTRY_EXT(log10, 3.0f, 4.0f, 2.0f, 4.76837158203125e-7f, FTZ_OFF, unaryF,
ENTRY_EXT(log10, 3.0f, 4.0f, 2.0f, 3.0f, 4.76837158203125e-7f, FTZ_OFF, unaryF,
4.76837158203125e-7f), // relaxed ulp 2^-21
ENTRY(log1p, 2.0f, 4.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY(logb, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(log1p, 2.0f, 4.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY(logb, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),

// In fast-relaxed-math mode it has to be either exactly rounded fma or exactly rounded a*b+c
ENTRY_EXT(mad, INFINITY, INFINITY, INFINITY, INFINITY, FTZ_OFF, mad_function, INFINITY),
ENTRY_EXT(mad, INFINITY, INFINITY, INFINITY, INFINITY, INFINITY, FTZ_OFF, mad_function, INFINITY),

ENTRY(maxmag, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(minmag, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(modf, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF_two_results),
ENTRY(nan, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF_u),
ENTRY(nextafter, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF_nextafter),
ENTRY(maxmag, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(minmag, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(modf, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF_two_results),
ENTRY(nan, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF_u),
ENTRY(nextafter, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF_nextafter),

// In derived mode the ulp error is calculated as exp2(y*log2(x)).
// In non-derived it is the same as half_pow.
ENTRY_EXT(pow, 16.0f, 16.0f, 4.0f, 8192.0f, FTZ_OFF, binaryF, 8192.0f),

ENTRY(pown, 16.0f, 16.0f, 4.0f, FTZ_OFF, binaryF_i),
ENTRY(powr, 16.0f, 16.0f, 4.0f, FTZ_OFF, binaryF),
ENTRY(remainder, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(remquo, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF_two_results_i),
ENTRY(rint, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(rootn, 16.0f, 16.0f, 4.0f, FTZ_OFF, binaryF_i),
ENTRY(round, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(rsqrt, 2.0f, 4.0f, 1.0f, FTZ_OFF, unaryF),
ENTRY(signbit, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY_EXT(sin, 4.0f, 4.0f, 2.0f, 0.00048828125f, FTZ_OFF, unaryF,
ENTRY_EXT(pow, 16.0f, 16.0f, 4.0f, 5.0f, 8192.0f, FTZ_OFF, binaryF, 8192.0f),

ENTRY(pown, 16.0f, 16.0f, 4.0f, 5.0f, FTZ_OFF, binaryF_i),
ENTRY(powr, 16.0f, 16.0f, 4.0f, 5.0f, FTZ_OFF, binaryF),
ENTRY(remainder, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF),
ENTRY(remquo, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryF_two_results_i),
ENTRY(rint, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(rootn, 16.0f, 16.0f, 4.0f, 5.0f, FTZ_OFF, binaryF_i),
ENTRY(round, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(rsqrt, 2.0f, 4.0f, 1.0f, 1.0f, FTZ_OFF, unaryF),
ENTRY(signbit, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
ENTRY_EXT(sin, 4.0f, 4.0f, 2.0f, 2.0f, 0.00048828125f, FTZ_OFF, unaryF,
0.00048828125f), // relaxed ulp 2^-11
ENTRY_EXT(sincos, 4.0f, 4.0f, 2.0f, 0.00048828125f, FTZ_OFF,
ENTRY_EXT(sincos, 4.0f, 4.0f, 2.0f, 2.0f, 0.00048828125f, FTZ_OFF,
unaryF_two_results,
0.00048828125f), // relaxed ulp 2^-11
ENTRY(sinh, 4.0f, 4.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY_EXT(sinpi, 4.0f, 4.0f, 2.0f, 0.00048828125f, FTZ_OFF, unaryF,
ENTRY(sinh, 4.0f, 4.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY_EXT(sinpi, 4.0f, 4.0f, 2.0f, 2.0f, 0.00048828125f, FTZ_OFF, unaryF,
0.00048828125f), // relaxed ulp 2^-11
{ "sqrt",
"sqrt",
Expand All @@ -373,6 +377,7 @@ const Func functionList[] = {
3.0f,
0.0f,
0.0f,
1.0f,
4.0f,
INFINITY,
INFINITY,
Expand All @@ -387,6 +392,7 @@ const Func functionList[] = {
0.0f,
INFINITY,
INFINITY,
INFINITY,
INFINITY,
INFINITY,
INFINITY,
Expand All @@ -396,12 +402,12 @@ const Func functionList[] = {

// In derived mode it the ulp error is calculated as sin/cos.
// In non-derived mode it is the same as half_tan.
ENTRY_EXT(tan, 5.0f, 5.0f, 2.0f, 8192.0f, FTZ_OFF, unaryF, 8192.0f),
ENTRY_EXT(tan, 5.0f, 5.0f, 2.0f, 3.0f, 8192.0f, FTZ_OFF, unaryF, 8192.0f),

ENTRY(tanh, 5.0f, 5.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY(tanpi, 6.0f, 6.0f, 2.0f, FTZ_OFF, unaryF),
ENTRY(tanh, 5.0f, 5.0f, 2.0f, 3.0f, FTZ_OFF, unaryF),
ENTRY(tanpi, 6.0f, 6.0f, 2.0f, 2.0f, FTZ_OFF, unaryF),
svenvh marked this conversation as resolved.
Show resolved Hide resolved
//ENTRY(tgamma, 16.0f, 16.0f, FTZ_OFF, unaryF), Commented this out until we can be sure this requirement is realistic
ENTRY(trunc, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),
ENTRY(trunc, 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, unaryF),

HALF_ENTRY(cos, 8192.0f, 8192.0f, FTZ_ON, unaryOF),
HALF_ENTRY(divide, 8192.0f, 8192.0f, FTZ_ON, binaryOF),
Expand All @@ -419,8 +425,8 @@ const Func functionList[] = {
HALF_ENTRY(tan, 8192.0f, 8192.0f, FTZ_ON, unaryOF),

// basic operations
OPERATOR_ENTRY(add, "+", 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryOperatorF),
OPERATOR_ENTRY(subtract, "-", 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryOperatorF),
OPERATOR_ENTRY(add, "+", 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryOperatorF),
OPERATOR_ENTRY(subtract, "-", 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryOperatorF),
//ENTRY(reciprocal, 1.0f, 1.0f, FTZ_OFF, unaryF),
{ "reciprocal",
"/",
Expand All @@ -430,6 +436,7 @@ const Func functionList[] = {
2.5f,
0.0f,
0.0f,
1.0f,
3.0f,
2.5f,
INFINITY,
Expand All @@ -444,6 +451,7 @@ const Func functionList[] = {
2.5f,
0.0f,
1.0f,
1.0f,
3.0f,
2.5f,
INFINITY,
Expand All @@ -457,17 +465,18 @@ const Func functionList[] = {
{ nullptr },
0.0f,
INFINITY,
INFINITY,
INFINITY,
INFINITY,
INFINITY,
INFINITY,
FTZ_OFF,
RELAXED_OFF,
binaryOperatorOF /* only for single precision */ },
OPERATOR_ENTRY(multiply, "*", 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryOperatorF),
OPERATOR_ENTRY(assignment, "", 0.0f, 0.0f, 0.0f, FTZ_OFF,
OPERATOR_ENTRY(multiply, "*", 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, binaryOperatorF),
OPERATOR_ENTRY(assignment, "", 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF,
unaryF), // A simple copy operation
OPERATOR_ENTRY(not, "!", 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
OPERATOR_ENTRY(not, "!", 0.0f, 0.0f, 0.0f, 0.0f, FTZ_OFF, macro_unaryF),
};
// clang-format on

Expand Down
Loading
Loading