FPMP Specification — Accuracy, Special Values and Performance

Generated: 2026-10-05 21:11:43

Overview

This document is the per-function reference for the fpmp multi-precision types of the CCCL FP component: measured accuracy, special-value behavior and GPU performance, for every function the two types implement.

The types live in namespace cuda. Arithmetic and the math functions both come in with <cuda/fpmp>. Names appear unqualified in the tables below for width; every one of them is a cuda:: name.

Each function is reported at three accuracy levels, low, def and high. def is the default selector and is equal to mid, not to high; mid is the name to use in code (cuda::fpmp2_accuracy::mid).

Test Platforms

GPU	Clock (MHz)	SMs	FP64:FP32
NVIDIA B300 SXM6 AC	2032	148	1:64
NVIDIA B200	1965	148	1:2

FP64:FP32 is the rate at which the part runs double precision against single, as NVIDIA specifies it: results per clock per SM, from the throughput table in the CUDA C++ Best Practices Guide. A benchmark that times the two side by side can read a narrower gap, since its FP32 baseline runs below peak. It is the single number that decides which of these types is worth using: fp32mp2 is built on FP32 and fp64mp2 on FP64, so each one wins on the hardware where the other’s arithmetic is the scarce resource.

Supported Data Types

Type	Representation	Mantissa	Range	Size
fp32mp2	float-float	46 bits = 2×24 − 2	float’s, ~±10^38	8 bytes
fp64mp2	double-double	104 bits = 2×53 − 2	double’s, ~±10^308	16 bytes

Significand counts include the implicit leading bit, the convention cuda::std::numeric_limits::digits uses: float has 24 and double has 53. A non-overlapping pair guarantees 2p − 2 of them, the two subtracted bits being what keeping the halves disjoint costs. Widening the mantissa does not widen the exponent: each type keeps the range of the format it is built from.

Function Families

The math functions are organized into families that mirror the CUDA C++ mathematical standard library taxonomy (CUDA C++ Programming Guide, “Mathematical Functions”). Each family lives in a dedicated implementation header (fpmp_math_impl_<family>.h) that contains both the fp32mp2 implementation and the fp64mp2 specialization for its functions. Shared kernels and constants live in fpmp_math_impl.h. Include <cuda/fpmp>, which pulls in all family headers and provides the overloaded fpmp2 API wrappers (template declarations, float/double specializations, the freestanding API, and library-mode declarations), along with the basic arithmetic (add, sub, mul, div, fma, mad) and sqrt/rsqrt. Functions that CUDA lists as “non-standard” are folded into their natural standard family.

The implementation headers named below are internal and sit under cuda/__fp/; they are listed to show how the implementation is partitioned, not as headers to include directly.

Family	Implementation header	Functions
Common utilities	fpmp_math_impl.h	error-free transforms, Horner/polynomial evaluation, fp32mp2/fp64mp2 constants, argument reduction (Cody–Waite, Payne–Hanek), exponent split/scale kernels
Exponential	fpmp_math_impl_exp.h	exp, exp2, exp10, expm1, log, log2, log10, log1p
Power	fpmp_math_impl_pow.h	pow, cbrt, rcbrt, hypot, rhypot, norm3d, norm4d, rnorm3d, rnorm4d
Trigonometric	fpmp_math_impl_trig.h	sin, cos, tan, asin, acos, atan, atan2, sincos, sinpi, cospi, sincospi
Hyperbolic	fpmp_math_impl_hyperbolic.h	sinh, cosh, tanh, asinh, acosh, atanh
Error, gamma & special	fpmp_math_impl_special.h	erf, erfc, erfinv, erfcinv, erfcx, tgamma, lgamma, normcdf, normcdfinv, boys_f0, icdf, j0, j1, jn, y0, y1, yn, cyl_bessel_i0, cyl_bessel_i1
Nearest integer & remainder	fpmp_math_impl_nearint.h	ceil, floor, trunc, round, nearbyint, rint, lrint, llrint, lround, llround, fmod, remainder, remquo
Floating-point manipulation	fpmp_math_impl_manip.h	frexp, ldexp, modf, scalbn, scalbln, ilogb, logb, nextafter, copysign, fabs
Classification & comparison	fpmp_math_impl_classify.h	isfinite, isinf, isnan, signbit, fmax, fmin, max, min, fdim

Table of Contents

Function Families

Arithmetic Operations

• Addition (add)

• Subtraction (sub)

• Multiplication (mul)

• Division (div)

• Accumulate (acc)

• Fused Multiply-Add (fma)

• Multiply-Add (mad)

Mathematical Functions

• Square Root (sqrt)

• Reciprocal Square Root (rsqrt)

• Cube Root (cbrt)

• Reciprocal Cube Root (rcbrt)

• Power (pow)

• Exponential (exp)

• Base-2 Exponential (exp2)

• Base-10 Exponential (exp10)

• Exponential Minus One (expm1)

• Natural Logarithm (log)

• Base-2 Logarithm (log2)

• Base-10 Logarithm (log10)

• Logarithm of One Plus x (log1p)

• Sine (sin)

• Cosine (cos)

• Tangent (tan)

• Sine of Pi Times x (sinpi)

• Cosine of Pi Times x (cospi)

• Arc Sine (asin)

• Arc Cosine (acos)

• Arc Tangent (atan)

• Two-Argument Arc Tangent (atan2)

• Hyperbolic Sine (sinh)

• Hyperbolic Cosine (cosh)

• Hyperbolic Tangent (tanh)

• Inverse Hyperbolic Sine (asinh)

• Inverse Hyperbolic Cosine (acosh)

• Inverse Hyperbolic Tangent (atanh)

• Error Function (erf)

• Complementary Error Function (erfc)

• Inverse Normal CDF (normcdfinv)

• Boys Function F0 (boys_f0)

• Floor (floor)

• Ceiling (ceil)

• Round to Nearest (round)

• Truncate (trunc)

• Floating-Point Remainder (fmod)

• IEEE Remainder (remainder)

• Scale by Power of Two (ldexp)

• Scale by Power of Two (scalbn)

• Scale by Power of Two, Long Exponent (scalbln)

• Split into Significand and Exponent (frexp)

Comparison Operations

• Equal (eq)

• Not Equal (ne)

• Less Than (lt)

• Less Than or Equal (le)

• Greater Than (gt)

• Greater Than or Equal (ge)

Type Conversions

• To Int32 (mp2int)

• To UInt32 (mp2uint)

• To Int64 (mp2ll)

• To UInt64 (mp2ull)

• From Int32 (int2mp)

• From UInt32 (uint2mp)

• From Int64 (ll2mp)

• From UInt64 (ull2mp)

• To Native Float (mp2fp)

• From Native Float (fp2mp)

Appendix: Legends

• Measured Accuracy Legend

• Special Values Legend (Floating Point)

• Special Values Legend (Integer Conversions)

• Performance Metrics Legend

• SASS Instructions Legend

• SASS Instructions Summary

Arithmetic Operations

Addition (add)

Type: fp32mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261021648	99.99%	1.00e-13	1.63e-16	43
output special	130965	3e-03%	–	–	–
input special	5	1e-07%	–	–	–
output denormal	15196	4e-04%	0.00e+00	0.00e+00	0
input denormal	18	4e-07%	3.95e-13	2.02e-13	41
input near inf	1180	3e-05%	2.17e-10	8.95e-13	32
cancellation	14	3e-07%	1.27e-08	4.27e-09	26
unclassified	133269	3e-03%	1.21e-09	9.65e-13	29
TOTAL	4261302295	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	3471.0	-0.23x	6.96x	3349.1	-0.23x	-0.44x
ev/clk/SM	11.54	-0.23x	6.96x	11.52	-0.23x	-0.44x
clk/ev	20.7	-0.62x	5.12x	21.1	-0.62x	1.02x

SASS Instructions:

Class	Count
fp32	8
fp64	0
other	1
total	9

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261021648	100.00%	1.00e-13	1.63e-16	43
input denormal	18	4e-07%	3.95e-13	2.02e-13	41
input near inf	1180	3e-05%	2.17e-10	8.95e-13	32
cancellation	14	3e-07%	1.27e-08	4.27e-09	26
unclassified	133269	3e-03%	1.21e-09	9.65e-13	29
TOTAL	4261156129	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	2781.5	-0.19x	5.57x	2688.6	-0.19x	-0.36x
ev/clk/SM	9.25	-0.19x	5.57x	9.24	-0.19x	-0.36x
clk/ev	39.1	-0.33x	2.70x	39.0	-0.34x	-0.55x

SASS Instructions:

Class	Count
fp32	11
fp64	0
other	0
total	11

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261156129	100.00%	7.11e-15	1.06e-16	47
TOTAL	4261156129	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1643.1	-0.11x	3.29x	1588.3	-0.11x	-0.21x
ev/clk/SM	5.46	-0.11x	3.29x	5.46	-0.11x	-0.21x
clk/ev	57.2	-0.23x	1.85x	57.7	-0.23x	-0.38x

SASS Instructions:

Class	Count
fp32	20
fp64	0
other	0
total	20

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	nan	nan
-maxN	-inf	-inf	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	+0	+inf	nan
-1	-inf	-3.4e+38	-2	-1	-1	-1	-1	-1	-1	-1	-1	+0	3.4e+38	+inf	nan
-minN	-inf	-3.4e+38	-1	-2.4e-38	-2.4e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.4e-45	+0	1	3.4e+38	+inf	nan
-maxD	-inf	-3.4e+38	-1	-2.4e-38	-2.4e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	+0	1.4e-45	1	3.4e+38	+inf	nan
-minD	-inf	-3.4e+38	-1	-1.2e-38	-1.2e-38	-2.8e-45	-1.4e-45	-1.4e-45	+0	1.2e-38	1.2e-38	1	3.4e+38	+inf	nan
-0	-inf	-3.4e+38	-1	-1.2e-38	-1.2e-38	-1.4e-45	-0	+0	1.4e-45	1.2e-38	1.2e-38	1	3.4e+38	+inf	nan
+0	-inf	-3.4e+38	-1	-1.2e-38	-1.2e-38	-1.4e-45	+0	+0	1.4e-45	1.2e-38	1.2e-38	1	3.4e+38	+inf	nan
+minD	-inf	-3.4e+38	-1	-1.2e-38	-1.2e-38	+0	1.4e-45	1.4e-45	2.8e-45	1.2e-38	1.2e-38	1	3.4e+38	+inf	nan
+maxD	-inf	-3.4e+38	-1	-1.4e-45	+0	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	2.4e-38	1	3.4e+38	+inf	nan
+minN	-inf	-3.4e+38	-1	+0	1.4e-45	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	2.4e-38	1	3.4e+38	+inf	nan
+1	-inf	-3.4e+38	+0	1	1	1	1	1	1	1	1	2	3.4e+38	+inf	nan
+maxN	-inf	+0	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	+inf	+inf	nan
+INF	nan	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Type: fp64mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760813	100.00%	9.54e-29	7.62e-36	93
output special	2	1e-05%	–	–	–
output denormal	8	5e-05%	0.00e+00	0.00e+00	0
unclassified	1	6e-06%	2.31e-27	2.31e-27	88
TOTAL	16760824	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	62.5	-0.13x	-0.34x	1793.8	-0.24x	10.34x
ev/clk/SM	0.21	-0.13x	-0.34x	6.17	-0.24x	10.34x
clk/ev	563.2	-0.19x	-0.29x	40.4	-0.54x	4.10x

SASS Instructions:

Class	Count
fp32	0
fp64	8
other	4
total	12

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760813	100.00%	9.54e-29	7.62e-36	93
unclassified	1	6e-06%	2.31e-27	2.31e-27	88
TOTAL	16760814	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	45.4	-0.09x	-0.24x	1347.0	-0.18x	7.81x
ev/clk/SM	0.15	-0.09x	-0.24x	4.63	-0.18x	7.81x
clk/ev	788.2	-0.13x	-0.21x	75.8	-0.29x	2.18x

SASS Instructions:

Class	Count
fp32	0
fp64	11
other	0
total	11

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760814	100.00%	2.46e-32	-2.94e-37	105
TOTAL	16760814	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	25.0	-0.05x	-0.13x	816.3	-0.11x	4.66x
ev/clk/SM	0.08	-0.05x	-0.13x	2.81	-0.11x	4.66x
clk/ev	1342.7	-0.08x	-0.12x	107.3	-0.20x	1.54x

SASS Instructions:

Class	Count
fp32	0
fp64	20
other	0
total	20

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	nan	nan
-maxN	-inf	-inf	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	+0	+inf	nan
-1	-inf	-1.8e+308	-2	-1	-1	-1	-1	-1	-1	-1	-1	+0	1.8e+308	+inf	nan
-minN	-inf	-1.8e+308	-1	-4.5e-308	-4.5e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-4.9e-324	+0	1	1.8e+308	+inf	nan
-maxD	-inf	-1.8e+308	-1	-4.5e-308	-4.5e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	+0	4.9e-324	1	1.8e+308	+inf	nan
-minD	-inf	-1.8e+308	-1	-2.2e-308	-2.2e-308	-9.9e-324	-4.9e-324	-4.9e-324	+0	2.2e-308	2.2e-308	1	1.8e+308	+inf	nan
-0	-inf	-1.8e+308	-1	-2.2e-308	-2.2e-308	-4.9e-324	-0	+0	4.9e-324	2.2e-308	2.2e-308	1	1.8e+308	+inf	nan
+0	-inf	-1.8e+308	-1	-2.2e-308	-2.2e-308	-4.9e-324	+0	+0	4.9e-324	2.2e-308	2.2e-308	1	1.8e+308	+inf	nan
+minD	-inf	-1.8e+308	-1	-2.2e-308	-2.2e-308	+0	4.9e-324	4.9e-324	9.9e-324	2.2e-308	2.2e-308	1	1.8e+308	+inf	nan
+maxD	-inf	-1.8e+308	-1	-4.9e-324	+0	2.2e-308	2.2e-308	2.2e-308	2.2e-308	4.5e-308	4.5e-308	1	1.8e+308	+inf	nan
+minN	-inf	-1.8e+308	-1	+0	4.9e-324	2.2e-308	2.2e-308	2.2e-308	2.2e-308	4.5e-308	4.5e-308	1	1.8e+308	+inf	nan
+1	-inf	-1.8e+308	+0	1	1	1	1	1	1	1	1	2	1.8e+308	+inf	nan
+maxN	-inf	+0	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	+inf	+inf	nan
+INF	nan	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

---

Subtraction (sub)

Type: fp32mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261021649	99.99%	1.00e-13	1.64e-16	43
output special	131026	3e-03%	–	–	–
input special	5	1e-07%	–	–	–
output denormal	15041	4e-04%	0.00e+00	0.00e+00	0
input denormal	19	4e-07%	4.12e-13	1.76e-13	41
input near inf	1170	3e-05%	1.53e-10	9.65e-13	32
cancellation	12	3e-07%	4.97e-08	9.45e-09	24
unclassified	133285	3e-03%	1.13e-09	9.52e-13	29
TOTAL	4261302207	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	3468.4	-0.23x	6.95x	3349.0	-0.23x	-0.45x
ev/clk/SM	11.53	-0.23x	6.95x	11.52	-0.23x	-0.45x
clk/ev	20.8	-0.61x	5.12x	21.2	-0.63x	1.02x

SASS Instructions:

Class	Count
fp32	8
fp64	0
other	1
total	9

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261021648	100.00%	1.00e-13	1.64e-16	43
input denormal	19	4e-07%	4.12e-13	1.76e-13	41
input near inf	1170	3e-05%	1.53e-10	9.65e-13	32
cancellation	12	3e-07%	4.97e-08	9.45e-09	24
unclassified	133285	3e-03%	1.13e-09	9.52e-13	29
TOTAL	4261156134	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	2780.3	-0.19x	5.57x	2687.8	-0.19x	-0.36x
ev/clk/SM	9.25	-0.19x	5.57x	9.24	-0.19x	-0.36x
clk/ev	38.8	-0.33x	2.73x	39.3	-0.34x	-0.56x

SASS Instructions:

Class	Count
fp32	11
fp64	0
other	0
total	11

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261156134	100.00%	7.11e-15	1.06e-16	47
TOTAL	4261156134	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1644.3	-0.11x	3.29x	1588.3	-0.11x	-0.21x
ev/clk/SM	5.47	-0.11x	3.29x	5.46	-0.11x	-0.21x
clk/ev	57.1	-0.22x	1.85x	57.5	-0.23x	-0.38x

SASS Instructions:

Class	Count
fp32	20
fp64	0
other	0
total	20

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	nan	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	nan
-maxN	+inf	+0	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-inf	-inf	nan
-1	+inf	3.4e+38	+0	-1	-1	-1	-1	-1	-1	-1	-1	-2	-3.4e+38	-inf	nan
-minN	+inf	3.4e+38	1	+0	-1.4e-45	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-2.4e-38	-2.4e-38	-1	-3.4e+38	-inf	nan
-maxD	+inf	3.4e+38	1	1.4e-45	+0	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-2.4e-38	-2.4e-38	-1	-3.4e+38	-inf	nan
-minD	+inf	3.4e+38	1	1.2e-38	1.2e-38	+0	-1.4e-45	-1.4e-45	-2.8e-45	-1.2e-38	-1.2e-38	-1	-3.4e+38	-inf	nan
-0	+inf	3.4e+38	1	1.2e-38	1.2e-38	1.4e-45	+0	-0	-1.4e-45	-1.2e-38	-1.2e-38	-1	-3.4e+38	-inf	nan
+0	+inf	3.4e+38	1	1.2e-38	1.2e-38	1.4e-45	+0	+0	-1.4e-45	-1.2e-38	-1.2e-38	-1	-3.4e+38	-inf	nan
+minD	+inf	3.4e+38	1	1.2e-38	1.2e-38	2.8e-45	1.4e-45	1.4e-45	+0	-1.2e-38	-1.2e-38	-1	-3.4e+38	-inf	nan
+maxD	+inf	3.4e+38	1	2.4e-38	2.4e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	+0	-1.4e-45	-1	-3.4e+38	-inf	nan
+minN	+inf	3.4e+38	1	2.4e-38	2.4e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.4e-45	+0	-1	-3.4e+38	-inf	nan
+1	+inf	3.4e+38	2	1	1	1	1	1	1	1	1	+0	-3.4e+38	-inf	nan
+maxN	+inf	+inf	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	+0	-inf	nan
+INF	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	nan	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Type: fp64mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760814	100.00%	8.49e-29	-2.46e-36	93
output special	2	1e-05%	–	–	–
output denormal	7	4e-05%	0.00e+00	0.00e+00	0
unclassified	1	6e-06%	1.45e-28	1.45e-28	92
TOTAL	16760824	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	62.5	-0.13x	-0.55x	1799.8	-0.24x	16.13x
ev/clk/SM	0.21	-0.13x	-0.55x	6.19	-0.24x	16.13x
clk/ev	562.7	-0.19x	-0.68x	40.2	-0.54x	9.45x

SASS Instructions:

Class	Count
fp32	0
fp64	8
other	4
total	12

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760814	100.00%	8.49e-29	-2.46e-36	93
unclassified	1	6e-06%	1.45e-28	1.45e-28	92
TOTAL	16760815	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	45.5	-0.09x	-0.40x	1350.6	-0.18x	12.10x
ev/clk/SM	0.15	-0.09x	-0.40x	4.64	-0.18x	12.10x
clk/ev	786.6	-0.13x	-0.48x	75.7	-0.29x	5.01x

SASS Instructions:

Class	Count
fp32	0
fp64	11
other	0
total	11

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760815	100.00%	2.47e-32	-8.39e-38	105
TOTAL	16760815	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	25.0	-0.05x	-0.22x	814.5	-0.11x	7.30x
ev/clk/SM	0.08	-0.05x	-0.22x	2.80	-0.11x	7.30x
clk/ev	1340.9	-0.08x	-0.28x	107.7	-0.20x	3.52x

SASS Instructions:

Class	Count
fp32	0
fp64	20
other	0
total	20

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	nan	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	nan
-maxN	+inf	+0	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-inf	-inf	nan
-1	+inf	1.8e+308	+0	-1	-1	-1	-1	-1	-1	-1	-1	-2	-1.8e+308	-inf	nan
-minN	+inf	1.8e+308	1	+0	-4.9e-324	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-4.5e-308	-4.5e-308	-1	-1.8e+308	-inf	nan
-maxD	+inf	1.8e+308	1	4.9e-324	+0	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-4.5e-308	-4.5e-308	-1	-1.8e+308	-inf	nan
-minD	+inf	1.8e+308	1	2.2e-308	2.2e-308	+0	-4.9e-324	-4.9e-324	-9.9e-324	-2.2e-308	-2.2e-308	-1	-1.8e+308	-inf	nan
-0	+inf	1.8e+308	1	2.2e-308	2.2e-308	4.9e-324	+0	-0	-4.9e-324	-2.2e-308	-2.2e-308	-1	-1.8e+308	-inf	nan
+0	+inf	1.8e+308	1	2.2e-308	2.2e-308	4.9e-324	+0	+0	-4.9e-324	-2.2e-308	-2.2e-308	-1	-1.8e+308	-inf	nan
+minD	+inf	1.8e+308	1	2.2e-308	2.2e-308	9.9e-324	4.9e-324	4.9e-324	+0	-2.2e-308	-2.2e-308	-1	-1.8e+308	-inf	nan
+maxD	+inf	1.8e+308	1	4.5e-308	4.5e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	+0	-4.9e-324	-1	-1.8e+308	-inf	nan
+minN	+inf	1.8e+308	1	4.5e-308	4.5e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	4.9e-324	+0	-1	-1.8e+308	-inf	nan
+1	+inf	1.8e+308	2	1	1	1	1	1	1	1	1	+0	-1.8e+308	-inf	nan
+maxN	+inf	+inf	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	+0	-inf	nan
+INF	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	nan	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

---

Multiplication (mul)

Type: fp32mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3127081961	83.86%	1.00e-13	1.31e-15	43
output special	537824645	14.42%	–	–	–
input special	5	1e-07%	–	–	–
output denormal	3569762	0.10%	3.33e-01	1.68e-06	1
input denormal	20988831	0.56%	5.96e-08	4.70e-09	24
output near denormal	3219189	0.09%	1.19e-07	8.44e-08	23
cancellation	36315300	0.97%	5.96e-08	4.23e-09	24
TOTAL	3728999693	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	4064.1	-0.27x	8.15x	3923.9	-0.27x	-0.52x
ev/clk/SM	13.51	-0.27x	8.15x	13.49	-0.27x	-0.52x
clk/ev	21.3	-0.61x	4.95x	21.8	-0.61x	1.00x

SASS Instructions:

Class	Count
fp32	5
fp64	0
other	1
total	6

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3127081985	97.99%	1.00e-13	1.68e-15	43
output special	1	3e-08%	–	–	–
output denormal	3569762	0.11%	3.33e-01	1.68e-06	1
input denormal	20988807	0.66%	5.96e-08	4.70e-09	24
output near denormal	3219189	0.10%	1.19e-07	8.44e-08	23
cancellation	36315300	1.14%	5.96e-08	4.23e-09	24
TOTAL	3191175044	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	3091.9	-0.21x	6.20x	2989.4	-0.21x	-0.40x
ev/clk/SM	10.28	-0.21x	6.20x	10.28	-0.21x	-0.40x
clk/ev	33.8	-0.37x	3.14x	34.1	-0.39x	-0.64x

SASS Instructions:

Class	Count
fp32	9
fp64	0
other	0
total	9

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3127081985	97.99%	1.00e-13	1.68e-15	43
output special	1	3e-08%	–	–	–
output denormal	3569762	0.11%	3.33e-01	1.68e-06	1
input denormal	20988807	0.66%	5.96e-08	4.70e-09	24
output near denormal	3219189	0.10%	1.19e-07	8.44e-08	23
cancellation	36315300	1.14%	5.96e-08	4.23e-09	24
TOTAL	3191175044	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	3093.4	-0.21x	6.20x	2987.3	-0.21x	-0.40x
ev/clk/SM	10.29	-0.21x	6.20x	10.27	-0.21x	-0.40x
clk/ev	34.0	-0.38x	3.11x	34.1	-0.39x	-0.64x

SASS Instructions:

Class	Count
fp32	9
fp64	0
other	0
total	9

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	+inf	+inf	+inf	+inf	+inf	+inf	nan	nan	-inf	-inf	-inf	-inf	-inf	-inf	nan
-maxN	+inf	+inf	3.4e+38	4	4	4.8e-07	+0	-0	-4.8e-07	-4	-4	-3.4e+38	-inf	-inf	nan
-1	+inf	3.4e+38	1	1.2e-38	1.2e-38	1.4e-45	+0	-0	-1.4e-45	-1.2e-38	-1.2e-38	-1	-3.4e+38	-inf	nan
-minN	+inf	4	1.2e-38	+0	+0	+0	+0	-0	-0	-0	-0	-1.2e-38	-4	-inf	nan
-maxD	+inf	4	1.2e-38	+0	+0	+0	+0	-0	-0	-0	-0	-1.2e-38	-4	-inf	nan
-minD	+inf	4.8e-07	1.4e-45	+0	+0	+0	+0	-0	-0	-0	-0	-1.4e-45	-4.8e-07	-inf	nan
-0	nan	+0	+0	+0	+0	+0	+0	-0	-0	-0	-0	-0	-0	nan	nan
+0	nan	-0	-0	-0	-0	-0	-0	+0	+0	+0	+0	+0	+0	nan	nan
+minD	-inf	-4.8e-07	-1.4e-45	-0	-0	-0	-0	+0	+0	+0	+0	1.4e-45	4.8e-07	+inf	nan
+maxD	-inf	-4	-1.2e-38	-0	-0	-0	-0	+0	+0	+0	+0	1.2e-38	4	+inf	nan
+minN	-inf	-4	-1.2e-38	-0	-0	-0	-0	+0	+0	+0	+0	1.2e-38	4	+inf	nan
+1	-inf	-3.4e+38	-1	-1.2e-38	-1.2e-38	-1.4e-45	-0	+0	1.4e-45	1.2e-38	1.2e-38	1	3.4e+38	+inf	nan
+maxN	-inf	-inf	-3.4e+38	-4	-4	-4.8e-07	-0	+0	4.8e-07	4	4	3.4e+38	+inf	+inf	nan
+INF	-inf	-inf	-inf	-inf	-inf	-inf	nan	nan	+inf	+inf	+inf	+inf	+inf	+inf	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Type: fp64mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	12537828	85.50%	9.99e-29	-2.40e-20	93
output special	2095104	14.29%	–	–	–
output denormal	1757	0.01%	2.05e-13	2.49e-16	42
input denormal	2850	0.02%	1.10e-16	4.54e-18	53
output near denormal	871	6e-03%	2.22e-16	1.97e-16	52
cancellation	24953	0.17%	1.11e-16	4.57e-18	53
TOTAL	14663363	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	106.7	-0.21x	-0.76x	2191.1	-0.29x	16.14x
ev/clk/SM	0.35	-0.21x	-0.76x	7.53	-0.29x	16.14x
clk/ev	389.0	-0.27x	-0.51x	43.4	-0.50x	4.59x

SASS Instructions:

Class	Count
fp32	0
fp64	5
other	2
total	7

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	12537828	99.76%	9.99e-29	-2.40e-20	93
output denormal	1757	0.01%	2.05e-13	2.49e-16	42
input denormal	2850	0.02%	1.10e-16	4.54e-18	53
output near denormal	871	7e-03%	2.22e-16	1.97e-16	52
cancellation	24953	0.20%	1.11e-16	4.57e-18	53
TOTAL	12568259	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	56.4	-0.11x	-0.40x	1553.9	-0.21x	11.46x
ev/clk/SM	0.19	-0.11x	-0.40x	5.34	-0.21x	11.46x
clk/ev	658.6	-0.16x	-0.30x	65.0	-0.34x	3.07x

SASS Instructions:

Class	Count
fp32	0
fp64	9
other	0
total	9

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	12537828	99.76%	9.99e-29	-2.40e-20	93
output denormal	1757	0.01%	2.05e-13	2.49e-16	42
input denormal	2850	0.02%	1.10e-16	4.54e-18	53
output near denormal	871	7e-03%	2.22e-16	1.97e-16	52
cancellation	24953	0.20%	1.11e-16	4.57e-18	53
TOTAL	12568259	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	56.4	-0.11x	-0.40x	1551.6	-0.21x	11.43x
ev/clk/SM	0.19	-0.11x	-0.40x	5.34	-0.21x	11.43x
clk/ev	658.7	-0.16x	-0.30x	65.0	-0.33x	3.07x

SASS Instructions:

Class	Count
fp32	0
fp64	9
other	0
total	9

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	+inf	+inf	+inf	+inf	+inf	+inf	nan	nan	-inf	-inf	-inf	-inf	-inf	-inf	nan
-maxN	+inf	+inf	1.8e+308	4	4	8.9e-16	+0	-0	-8.9e-16	-4	-4	-1.8e+308	-inf	-inf	nan
-1	+inf	1.8e+308	1	2.2e-308	2.2e-308	4.9e-324	+0	-0	-4.9e-324	-2.2e-308	-2.2e-308	-1	-1.8e+308	-inf	nan
-minN	+inf	4	2.2e-308	+0	+0	+0	+0	-0	-0	-0	-0	-2.2e-308	-4	-inf	nan
-maxD	+inf	4	2.2e-308	+0	+0	+0	+0	-0	-0	-0	-0	-2.2e-308	-4	-inf	nan
-minD	+inf	8.9e-16	4.9e-324	+0	+0	+0	+0	-0	-0	-0	-0	-4.9e-324	-8.9e-16	-inf	nan
-0	nan	+0	+0	+0	+0	+0	+0	-0	-0	-0	-0	-0	-0	nan	nan
+0	nan	-0	-0	-0	-0	-0	-0	+0	+0	+0	+0	+0	+0	nan	nan
+minD	-inf	-8.9e-16	-4.9e-324	-0	-0	-0	-0	+0	+0	+0	+0	4.9e-324	8.9e-16	+inf	nan
+maxD	-inf	-4	-2.2e-308	-0	-0	-0	-0	+0	+0	+0	+0	2.2e-308	4	+inf	nan
+minN	-inf	-4	-2.2e-308	-0	-0	-0	-0	+0	+0	+0	+0	2.2e-308	4	+inf	nan
+1	-inf	-1.8e+308	-1	-2.2e-308	-2.2e-308	-4.9e-324	-0	+0	4.9e-324	2.2e-308	2.2e-308	1	1.8e+308	+inf	nan
+maxN	-inf	-inf	-1.8e+308	-4	-4	-8.9e-16	-0	+0	8.9e-16	4	4	1.8e+308	+inf	+inf	nan
+INF	-inf	-inf	-inf	-inf	-inf	-inf	nan	nan	+inf	+inf	+inf	+inf	+inf	+inf	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

---

Division (div)

Type: fp32mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3006532757	80.74%	1.00e-13	2.85e-15	43
output special	536936544	14.42%	–	–	–
input special	5	1e-07%	–	–	–
output denormal	7612929	0.20%	1.00e+00	4.13e-01	0
input denormal	155637297	4.18%	1.79e-07	5.24e-09	22
output near denormal	208307	6e-03%	1.00e+00	6.29e-01	0
input near inf	16515092	0.44%	1.00e+00	1.00e+00	0
cancellation	290372	8e-03%	1.85e-08	1.14e-12	25
TOTAL	3723733303	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	4424.9	1.95x	63.78x	4267.6	2.08x	3.50x
ev/clk/SM	14.71	1.95x	63.78x	14.67	2.08x	3.50x
clk/ev	21.1	2.87x	26.92x	21.5	2.87x	6.12x

SASS Instructions:

Class	Count
fp32	6
fp64	0
other	1
total	7

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3006485741	94.22%	1.00e-13	2.08e-15	43
output special	8355823	0.26%	–	–	–
input special	1	3e-08%	–	–	–
output denormal	3233022	0.10%	1.00e+00	9.73e-01	0
input denormal	155782543	4.88%	1.79e-07	5.19e-09	22
output near denormal	207877	7e-03%	1.00e+00	6.30e-01	0
input near inf	16515092	0.52%	1.00e+00	1.00e+00	0
cancellation	192572	6e-03%	1.26e-08	1.11e-12	26
TOTAL	3190772671	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1987.5	-0.88x	28.65x	1922.5	-0.94x	1.57x
ev/clk/SM	6.61	-0.88x	28.65x	6.61	-0.94x	1.57x
clk/ev	49.9	1.21x	11.40x	50.3	1.23x	2.62x

SASS Instructions:

Class	Count
fp32	13
fp64	0
other	0
total	13

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3187273516	93.42%	1.00e-13	1.70e-15	43
output special	222266516	6.51%	–	–	–
input special	1	3e-08%	–	–	–
output denormal	2097440	0.06%	1.00e+00	2.17e-06	0
input denormal	49841	1e-03%	1.25e-08	1.19e-12	26
input near inf	4317	1e-04%	2.03e-09	1.69e-12	28
cancellation	211974	6e-03%	1.26e-08	1.10e-12	26
TOTAL	3411903605	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1263.9	-0.56x	18.22x	1220.7	-0.60x	1.00x
ev/clk/SM	4.20	-0.56x	18.22x	4.20	-0.60x	1.00x
clk/ev	84.8	-0.72x	6.70x	85.2	-0.73x	1.55x

SASS Instructions:

Class	Count
fp32	23
fp64	0
other	23
total	46

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	nan	+inf	+inf	+inf	+inf	+inf	+inf	-inf	-inf	-inf	-inf	-inf	-inf	nan	nan
-maxN	+0	1	3.4e+38	+inf	+inf	+inf	+inf	-inf	-inf	-inf	-inf	-3.4e+38	-1	-0	nan
-1	+0	2.9e-39	1	8.5e+37	8.5e+37	+inf	+inf	-inf	-inf	-8.5e+37	-8.5e+37	-1	-2.9e-39	-0	nan
-minN	+0	+0	1.2e-38	1	1	+inf	+inf	-inf	-inf	-1	-1	-1.2e-38	-0	-0	nan
-maxD	+0	+0	1.2e-38	1	1	+inf	+inf	-inf	-inf	-1	-1	-1.2e-38	-0	-0	nan
-minD	+0	+0	1.4e-45	1.2e-07	1.2e-07	+inf	+inf	-inf	-inf	-1.2e-07	-1.2e-07	-1.4e-45	-0	-0	nan
-0	+0	+0	+0	+0	+0	nan	nan	nan	nan	-0	-0	-0	-0	-0	nan
+0	-0	-0	-0	-0	-0	nan	nan	nan	nan	+0	+0	+0	+0	+0	nan
+minD	-0	-0	-1.4e-45	-1.2e-07	-1.2e-07	-inf	-inf	+inf	+inf	1.2e-07	1.2e-07	1.4e-45	+0	+0	nan
+maxD	-0	-0	-1.2e-38	-1	-1	-inf	-inf	+inf	+inf	1	1	1.2e-38	+0	+0	nan
+minN	-0	-0	-1.2e-38	-1	-1	-inf	-inf	+inf	+inf	1	1	1.2e-38	+0	+0	nan
+1	-0	-2.9e-39	-1	-8.5e+37	-8.5e+37	-inf	-inf	+inf	+inf	8.5e+37	8.5e+37	1	2.9e-39	+0	nan
+maxN	-0	-1	-3.4e+38	-inf	-inf	-inf	-inf	+inf	+inf	+inf	+inf	3.4e+38	1	+0	nan
+INF	nan	-inf	-inf	-inf	-inf	-inf	-inf	+inf	+inf	+inf	+inf	+inf	+inf	nan	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Type: fp64mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	12480461	85.12%	1.00e-28	-3.87e-20	93
output special	2097129	14.30%	–	–	–
output denormal	2577	0.02%	1.86e-15	1.02e-18	48
input denormal	81990	0.56%	2.32e-16	5.83e-18	51
output near denormal	2	1e-05%	1.21e-16	1.18e-16	52
TOTAL	14662159	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	103.2	1.49x	1.68x	2342.5	1.92x	39.06x
ev/clk/SM	0.34	1.49x	1.68x	8.05	1.92x	39.06x
clk/ev	393.2	1.45x	1.24x	44.7	2.96x	10.90x

SASS Instructions:

Class	Count
fp32	1
fp64	11
other	27
total	39

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	12480681	99.32%	1.00e-28	-3.89e-20	93
output special	4083	0.03%	–	–	–
output denormal	8	6e-05%	1.86e-15	3.27e-16	48
input denormal	81770	0.65%	2.32e-16	5.87e-18	51
output near denormal	2	2e-05%	1.21e-16	1.17e-16	52
TOTAL	12566544	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	44.6	-0.64x	-0.73x	1070.3	-0.88x	17.80x
ev/clk/SM	0.15	-0.64x	-0.73x	3.68	-0.88x	17.80x
clk/ev	852.2	-0.67x	-0.57x	100.3	1.31x	4.85x

SASS Instructions:

Class	Count
fp32	1
fp64	18
other	25
total	44

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	12566536	98.79%	1.19e-29	-6.30e-33	96
output special	153482	1.21%	–	–	–
output denormal	229	2e-03%	2.26e-13	6.53e-16	42
TOTAL	12720247	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	26.2	-0.38x	-0.43x	587.4	-0.48x	9.78x
ev/clk/SM	0.09	-0.38x	-0.43x	2.02	-0.48x	9.78x
clk/ev	1385.9	-0.41x	-0.35x	174.1	-0.76x	2.80x

SASS Instructions:

Class	Count
fp32	1
fp64	28
other	49
total	78

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	nan	+inf	+inf	+inf	+inf	+inf	+inf	-inf	-inf	-inf	-inf	-inf	-inf	nan	nan
-maxN	+0	1	1.8e+308	+inf	+inf	+inf	+inf	-inf	-inf	-inf	-inf	-1.8e+308	-1	-0	nan
-1	+0	5.6e-309	1	4.5e+307	4.5e+307	+inf	+inf	-inf	-inf	-4.5e+307	-4.5e+307	-1	-5.6e-309	-0	nan
-minN	+0	+0	2.2e-308	1	1	+inf	+inf	-inf	-inf	-1	-1	-2.2e-308	-0	-0	nan
-maxD	+0	+0	2.2e-308	1	1	+inf	+inf	-inf	-inf	-1	-1	-2.2e-308	-0	-0	nan
-minD	+0	+0	4.9e-324	2.2e-16	2.2e-16	+inf	+inf	-inf	-inf	-2.2e-16	-2.2e-16	-4.9e-324	-0	-0	nan
-0	+0	+0	+0	+0	+0	nan	nan	nan	nan	-0	-0	-0	-0	-0	nan
+0	-0	-0	-0	-0	-0	nan	nan	nan	nan	+0	+0	+0	+0	+0	nan
+minD	-0	-0	-4.9e-324	-2.2e-16	-2.2e-16	-inf	-inf	+inf	+inf	2.2e-16	2.2e-16	4.9e-324	+0	+0	nan
+maxD	-0	-0	-2.2e-308	-1	-1	-inf	-inf	+inf	+inf	1	1	2.2e-308	+0	+0	nan
+minN	-0	-0	-2.2e-308	-1	-1	-inf	-inf	+inf	+inf	1	1	2.2e-308	+0	+0	nan
+1	-0	-5.6e-309	-1	-4.5e+307	-4.5e+307	-inf	-inf	+inf	+inf	4.5e+307	4.5e+307	1	5.6e-309	+0	nan
+maxN	-0	-1	-1.8e+308	-inf	-inf	-inf	-inf	+inf	+inf	+inf	+inf	1.8e+308	1	+0	nan
+INF	nan	-inf	-inf	-inf	-inf	-inf	-inf	+inf	+inf	+inf	+inf	+inf	+inf	nan	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

---

Accumulate (acc)

Type: fp32mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261156129	100.00%	7.10e-15	5.63e-17	47
output special	130965	3e-03%	–	–	–
input special	5	1e-07%	–	–	–
output denormal	16604	4e-04%	0.00e+00	0.00e+00	0
TOTAL	4261303703	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	3632.3	-0.24x	7.28x	3515.2	-0.24x	-0.47x
ev/clk/SM	12.08	-0.24x	7.28x	12.09	-0.24x	-0.47x
clk/ev	17.2	-0.76x	6.17x	17.5	-0.76x	1.25x

SASS Instructions:

Class	Count
fp32	7
fp64	0
other	1
total	8

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261156129	100.00%	7.10e-15	5.63e-17	47
TOTAL	4261156129	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	2996.9	-0.20x	6.01x	2895.6	-0.20x	-0.38x
ev/clk/SM	9.97	-0.20x	6.01x	9.96	-0.20x	-0.38x
clk/ev	39.1	-0.33x	2.71x	39.3	-0.33x	-0.56x

SASS Instructions:

Class	Count
fp32	10
fp64	0
other	0
total	10

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261156129	100.00%	7.10e-15	5.63e-17	47
TOTAL	4261156129	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	2317.2	-0.15x	4.64x	2238.8	-0.15x	-0.30x
ev/clk/SM	7.71	-0.15x	4.64x	7.70	-0.15x	-0.30x
clk/ev	50.8	-0.26x	2.08x	51.2	-0.26x	-0.42x

SASS Instructions:

Class	Count
fp32	13
fp64	0
other	0
total	13

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	nan	nan
-maxN	-inf	-inf	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	+0	+inf	nan
-1	-inf	-3.4e+38	-2	-1	-1	-1	-1	-1	-1	-1	-1	+0	3.4e+38	+inf	nan
-minN	-inf	-3.4e+38	-1	-2.4e-38	-2.4e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.4e-45	+0	1	3.4e+38	+inf	nan
-maxD	-inf	-3.4e+38	-1	-2.4e-38	-2.4e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	+0	1.4e-45	1	3.4e+38	+inf	nan
-minD	-inf	-3.4e+38	-1	-1.2e-38	-1.2e-38	-2.8e-45	-1.4e-45	-1.4e-45	+0	1.2e-38	1.2e-38	1	3.4e+38	+inf	nan
-0	-inf	-3.4e+38	-1	-1.2e-38	-1.2e-38	-1.4e-45	-0	+0	1.4e-45	1.2e-38	1.2e-38	1	3.4e+38	+inf	nan
+0	-inf	-3.4e+38	-1	-1.2e-38	-1.2e-38	-1.4e-45	+0	+0	1.4e-45	1.2e-38	1.2e-38	1	3.4e+38	+inf	nan
+minD	-inf	-3.4e+38	-1	-1.2e-38	-1.2e-38	+0	1.4e-45	1.4e-45	2.8e-45	1.2e-38	1.2e-38	1	3.4e+38	+inf	nan
+maxD	-inf	-3.4e+38	-1	-1.4e-45	+0	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	2.4e-38	1	3.4e+38	+inf	nan
+minN	-inf	-3.4e+38	-1	+0	1.4e-45	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	2.4e-38	1	3.4e+38	+inf	nan
+1	-inf	-3.4e+38	+0	1	1	1	1	1	1	1	1	2	3.4e+38	+inf	nan
+maxN	-inf	+0	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	+inf	+inf	nan
+INF	nan	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Type: fp64mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760814	100.00%	1.32e-32	-3.35e-38	105
output special	2	1e-05%	–	–	–
output denormal	8	5e-05%	0.00e+00	0.00e+00	0
TOTAL	16760824	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	71.4	-0.14x	-0.38x	1939.5	-0.26x	11.19x
ev/clk/SM	0.24	-0.14x	-0.38x	6.67	-0.26x	11.19x
clk/ev	508.9	-0.21x	-0.32x	34.6	-0.63x	4.78x

SASS Instructions:

Class	Count
fp32	0
fp64	7
other	2
total	9

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760814	100.00%	1.32e-32	-3.35e-38	105
TOTAL	16760814	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	50.0	-0.10x	-0.27x	1459.2	-0.19x	8.43x
ev/clk/SM	0.17	-0.10x	-0.27x	5.02	-0.19x	8.43x
clk/ev	722.8	-0.15x	-0.23x	75.5	-0.29x	2.19x

SASS Instructions:

Class	Count
fp32	0
fp64	10
other	0
total	10

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760814	100.00%	1.32e-32	-3.35e-38	105
TOTAL	16760814	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	38.5	-0.08x	-0.21x	1173.3	-0.16x	6.66x
ev/clk/SM	0.13	-0.08x	-0.21x	4.03	-0.16x	6.66x
clk/ev	914.8	-0.12x	-0.18x	99.4	-0.22x	1.66x

SASS Instructions:

Class	Count
fp32	0
fp64	13
other	0
total	13

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	nan	nan
-maxN	-inf	-inf	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	+0	+inf	nan
-1	-inf	-1.8e+308	-2	-1	-1	-1	-1	-1	-1	-1	-1	+0	1.8e+308	+inf	nan
-minN	-inf	-1.8e+308	-1	-4.5e-308	-4.5e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-4.9e-324	+0	1	1.8e+308	+inf	nan
-maxD	-inf	-1.8e+308	-1	-4.5e-308	-4.5e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	+0	4.9e-324	1	1.8e+308	+inf	nan
-minD	-inf	-1.8e+308	-1	-2.2e-308	-2.2e-308	-9.9e-324	-4.9e-324	-4.9e-324	+0	2.2e-308	2.2e-308	1	1.8e+308	+inf	nan
-0	-inf	-1.8e+308	-1	-2.2e-308	-2.2e-308	-4.9e-324	-0	+0	4.9e-324	2.2e-308	2.2e-308	1	1.8e+308	+inf	nan
+0	-inf	-1.8e+308	-1	-2.2e-308	-2.2e-308	-4.9e-324	+0	+0	4.9e-324	2.2e-308	2.2e-308	1	1.8e+308	+inf	nan
+minD	-inf	-1.8e+308	-1	-2.2e-308	-2.2e-308	+0	4.9e-324	4.9e-324	9.9e-324	2.2e-308	2.2e-308	1	1.8e+308	+inf	nan
+maxD	-inf	-1.8e+308	-1	-4.9e-324	+0	2.2e-308	2.2e-308	2.2e-308	2.2e-308	4.5e-308	4.5e-308	1	1.8e+308	+inf	nan
+minN	-inf	-1.8e+308	-1	+0	4.9e-324	2.2e-308	2.2e-308	2.2e-308	2.2e-308	4.5e-308	4.5e-308	1	1.8e+308	+inf	nan
+1	-inf	-1.8e+308	+0	1	1	1	1	1	1	1	1	2	1.8e+308	+inf	nan
+maxN	-inf	+0	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	+inf	+inf	nan
+INF	nan	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

---

Fused Multiply-Add (fma)

Type: fp32mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3705991498	87.35%	1.00e-13	8.17e-16	43
output special	531120823	12.52%	–	–	–
input special	41	1e-06%	–	–	–
output denormal	64932	2e-03%	1.29e-03	7.37e-07	9
input denormal	5220413	0.12%	5.96e-08	1.37e-09	24
output near denormal	40469	1e-03%	1.19e-07	8.32e-08	23
input near inf	4306	1e-04%	3.82e-09	1.89e-12	27
cancellation	174428	4e-03%	1.09e-08	1.27e-12	26
unclassified	74732	2e-03%	2.58e-09	9.39e-13	28
TOTAL	4242691642	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1887.2	-0.13x	3.78x	1822.0	-0.13x	-0.24x
ev/clk/SM	6.28	-0.13x	3.78x	6.27	-0.13x	-0.24x
clk/ev	32.3	-0.41x	3.42x	32.6	-0.41x	-0.68x

SASS Instructions:

Class	Count
fp32	16
fp64	0
other	1
total	17

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3705991498	99.85%	1.00e-13	8.17e-16	43
output special	18531	5e-04%	–	–	–
output denormal	42617	1e-03%	1.29e-03	1.12e-06	9
input denormal	5220413	0.14%	5.96e-08	1.37e-09	24
output near denormal	40469	1e-03%	1.19e-07	8.32e-08	23
input near inf	4306	1e-04%	3.82e-09	1.89e-12	27
cancellation	174428	5e-03%	1.09e-08	1.27e-12	26
unclassified	74732	2e-03%	2.58e-09	9.39e-13	28
TOTAL	3711566994	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1628.0	-0.11x	3.26x	1573.4	-0.11x	-0.21x
ev/clk/SM	5.41	-0.11x	3.26x	5.41	-0.11x	-0.21x
clk/ev	64.3	-0.20x	1.72x	65.0	-0.21x	-0.34x

SASS Instructions:

Class	Count
fp32	19
fp64	0
other	0
total	19

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3706072849	99.85%	1.00e-13	5.02e-16	43
output special	18531	5e-04%	–	–	–
output denormal	42617	1e-03%	1.29e-03	1.12e-06	9
input denormal	5210818	0.14%	5.96e-08	1.37e-09	24
output near denormal	40469	1e-03%	1.19e-07	8.32e-08	23
input near inf	3100	8e-05%	1.22e-09	1.36e-12	29
cancellation	125702	3e-03%	1.22e-08	1.32e-12	26
unclassified	52908	1e-03%	3.70e-09	1.00e-12	28
TOTAL	3711566994	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	842.5	-0.06x	1.69x	813.9	-0.06x	-0.11x
ev/clk/SM	2.80	-0.06x	1.69x	2.80	-0.06x	-0.11x
clk/ev	90.3	-0.14x	1.22x	91.0	-0.15x	-0.24x

SASS Instructions:

Class	Count
fp32	37
fp64	0
other	0
total	37

Type: fp64mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16646144	100.00%	0.00e+00	0.00e+00	106
TOTAL	16646144	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	31.6	-0.06x	-0.70x	1002.6	-0.13x	22.58x
ev/clk/SM	0.11	-0.06x	-0.70x	3.45	-0.13x	22.58x
clk/ev	1086.9	-0.10x	-0.96x	60.5	-0.36x	17.37x

SASS Instructions:

Class	Count
fp32	0
fp64	16
other	2
total	18

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16646144	100.00%	0.00e+00	0.00e+00	106
TOTAL	16646144	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	26.6	-0.05x	-0.59x	841.5	-0.11x	18.96x
ev/clk/SM	0.09	-0.05x	-0.59x	2.89	-0.11x	18.96x
clk/ev	1302.1	-0.09x	-0.80x	123.7	-0.18x	8.49x

SASS Instructions:

Class	Count
fp32	0
fp64	19
other	0
total	19

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16646144	100.00%	0.00e+00	0.00e+00	106
TOTAL	16646144	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	13.6	-0.03x	-0.30x	460.8	-0.06x	10.38x
ev/clk/SM	0.05	-0.03x	-0.30x	1.58	-0.06x	10.38x
clk/ev	2452.8	-0.04x	-0.42x	171.2	-0.13x	6.13x

SASS Instructions:

Class	Count
fp32	0
fp64	37
other	0
total	37

---

Multiply-Add (mad)

Type: fp32mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3705975209	87.35%	1.00e-13	7.88e-16	43
output special	531120823	12.52%	–	–	–
input special	41	1e-06%	–	–	–
output denormal	68114	2e-03%	1.29e-03	7.03e-07	9
input denormal	5222880	0.12%	5.96e-08	1.36e-09	24
output near denormal	40469	1e-03%	1.19e-07	8.32e-08	23
input near inf	4527	1e-04%	1.38e-09	1.31e-12	29
cancellation	184095	4e-03%	1.22e-08	1.30e-12	26
unclassified	78666	2e-03%	3.70e-09	9.60e-13	28
TOTAL	4242694824	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	2227.1	-0.15x	4.46x	2149.7	-0.15x	-0.29x
ev/clk/SM	7.41	-0.15x	4.46x	7.39	-0.15x	-0.29x
clk/ev	30.9	-0.42x	3.58x	31.0	-0.43x	-0.71x

SASS Instructions:

Class	Count
fp32	13
fp64	0
other	0
total	13

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3705975209	99.85%	1.00e-13	7.88e-16	43
output special	18531	5e-04%	–	–	–
output denormal	42617	1e-03%	1.29e-03	1.12e-06	9
input denormal	5222880	0.14%	5.96e-08	1.36e-09	24
output near denormal	40469	1e-03%	1.19e-07	8.32e-08	23
input near inf	4527	1e-04%	1.38e-09	1.31e-12	29
cancellation	184095	5e-03%	1.22e-08	1.30e-12	26
unclassified	78666	2e-03%	3.70e-09	9.60e-13	28
TOTAL	3711566994	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1862.4	-0.12x	3.73x	1797.8	-0.12x	-0.24x
ev/clk/SM	6.19	-0.12x	3.73x	6.18	-0.12x	-0.24x
clk/ev	45.2	-0.28x	2.45x	45.7	-0.29x	-0.48x

SASS Instructions:

Class	Count
fp32	16
fp64	0
other	0
total	16

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3705962819	99.85%	1.00e-13	9.67e-16	43
output special	18531	5e-04%	–	–	–
output denormal	42617	1e-03%	1.29e-03	1.12e-06	9
input denormal	5223153	0.14%	5.96e-08	1.36e-09	24
output near denormal	40469	1e-03%	1.19e-07	8.32e-08	23
input near inf	4776	1e-04%	1.38e-09	1.31e-12	29
cancellation	192586	5e-03%	7.89e-09	1.28e-12	26
unclassified	82043	2e-03%	2.83e-09	9.66e-13	28
TOTAL	3711566994	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1193.5	-0.08x	2.39x	1152.5	-0.08x	-0.15x
ev/clk/SM	3.97	-0.08x	2.39x	3.96	-0.08x	-0.15x
clk/ev	79.0	-0.17x	1.40x	79.1	-0.17x	-0.28x

SASS Instructions:

Class	Count
fp32	29
fp64	0
other	0
total	29

Type: fp64mp2

Accuracy: low

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16646144	100.00%	0.00e+00	0.00e+00	106
TOTAL	16646144	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	39.0	-0.08x	-0.86x	1207.3	-0.16x	27.20x
ev/clk/SM	0.13	-0.08x	-0.86x	4.15	-0.16x	27.20x
clk/ev	889.9	-0.12x	1.17x	60.5	-0.37x	17.36x

SASS Instructions:

Class	Count
fp32	0
fp64	13
other	0
total	13

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16646144	100.00%	0.00e+00	0.00e+00	106
TOTAL	16646144	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	31.6	-0.06x	-0.70x	985.3	-0.13x	22.20x
ev/clk/SM	0.11	-0.06x	-0.70x	3.39	-0.13x	22.20x
clk/ev	1081.6	-0.10x	-0.96x	86.2	-0.26x	12.18x

SASS Instructions:

Class	Count
fp32	0
fp64	16
other	0
total	16

Accuracy: high

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16646144	100.00%	0.00e+00	0.00e+00	106
TOTAL	16646144	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	17.3	-0.03x	-0.38x	577.7	-0.08x	13.01x
ev/clk/SM	0.06	-0.03x	-0.38x	1.99	-0.08x	13.01x
clk/ev	1926.5	-0.06x	-0.54x	155.6	-0.14x	6.75x

SASS Instructions:

Class	Count
fp32	0
fp64	29
other	0
total	29

Mathematical Functions

Square Root (sqrt)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	1992354374	93.14%	1.00e-13	2.30e-15	43
output special	8388605	0.39%	–	–	–
input denormal	138352058	6.47%	2.98e-08	1.25e-09	25
TOTAL	2139095037	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1558.5	-0.58x	23.45x	1510.0	-0.58x	1.30x
ev/clk/SM	5.18	-0.58x	23.45x	5.19	-0.58x	1.30x
clk/ev	71.4	-0.93x	8.16x	70.9	-0.94x	1.46x

SASS Instructions:

Class	Count
fp32	18
fp64	0
other	1
total	19

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	8305868	99.06%	1.00e-28	-2.09e-20	93
input denormal	78644	0.94%	1.58e-16	2.27e-18	52
TOTAL	8384512	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	23.3	-0.35x	-0.31x	582.8	-0.50x	7.92x
ev/clk/SM	0.08	-0.35x	-0.31x	2.00	-0.50x	7.92x
clk/ev	1519.6	-0.38x	-0.43x	172.7	-0.60x	3.78x

SASS Instructions:

Class	Count
fp32	0
fp64	23
other	25
total	48

Special Values Table:

Input	Value
-INF	-inf
-maxN	-1.8e+308
-1	-1
-minN	-2.2e-308
-maxD	-2.2e-308
-minD	-4.9e-324
-0	-0
+0	+0
+minD	4.9e-324
+maxD	2.2e-308
+minN	2.2e-308
+1	1
+maxN	1.8e+308
+INF	+inf
QNAN	nan

---

Reciprocal Square Root (rsqrt)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2130706432	99.61%	5.06e-14	2.36e-15	44
output special	8388605	0.39%	–	–	–
input special	1	5e-08%	–	–	–
TOTAL	2139095038	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1568.2	-0.41x	15.18x	1517.0	-0.40x	-0.95x
ev/clk/SM	5.21	-0.41x	15.18x	5.22	-0.40x	-0.95x
clk/ev	72.5	-0.61x	4.65x	72.0	-0.60x	1.07x

SASS Instructions:

Class	Count
fp32	17
fp64	0
other	0
total	17

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	8384512	100.00%	2.45e-32	-4.81e-33	105
TOTAL	8384512	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	24.7	-0.24x	-0.72x	611.8	-0.38x	18.55x
ev/clk/SM	0.08	-0.24x	-0.72x	2.10	-0.38x	18.55x
clk/ev	1433.8	-0.24x	1.12x	168.8	-0.46x	9.53x

SASS Instructions:

Class	Count
fp32	0
fp64	22
other	23
total	45

Special Values Table:

Input	Value
-INF	-inf
-maxN	-1.8e+308
-1	-1
-minN	-2.2e-308
-maxD	-2.2e-308
-minD	-4.9e-324
-0	-0
+0	+0
+minD	4.9e-324
+maxD	2.2e-308
+minN	2.2e-308
+1	1
+maxN	1.8e+308
+INF	+inf
QNAN	nan

---

Cube Root (cbrt)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4278190075	100.00%	3.17e-14	1.37e-15	44
input special	3	7e-08%	–	–	–
TOTAL	4278190078	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	329.1	-0.16x	8.15x	317.9	-0.17x	-0.53x
ev/clk/SM	1.09	-0.16x	8.15x	1.09	-0.17x	-0.53x
clk/ev	290.1	-0.29x	3.66x	291.5	-0.29x	-0.88x

SASS Instructions:

Class	Count
fp32	75
fp64	0
other	27
total	102

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Reciprocal Cube Root (rcbrt)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4278190075	100.00%	1.06e-14	1.19e-15	46
output special	5	1e-07%	–	–	–
TOTAL	4278190080	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	266.2	-0.22x	9.03x	257.1	-0.22x	-0.53x
ev/clk/SM	0.89	-0.22x	9.03x	0.88	-0.22x	-0.53x
clk/ev	333.1	-0.49x	4.14x	334.4	-0.50x	1.03x

SASS Instructions:

Class	Count
fp32	95
fp64	0
other	35
total	130

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Power (pow)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	1066806685	54.96%	1.00e-13	9.15e-16	43
output special	867368869	44.68%	–	–	–
input special	2	1e-07%	–	–	–
output denormal	1029046	0.05%	1.00e+00	1.00e+00	0
input denormal	642056	0.03%	3.40e-09	2.14e-13	28
output near inf	57508	3e-03%	2.27e-12	2.59e-13	38
input near inf	46077	2e-03%	4.60e-11	1.96e-13	34
cancellation	2722373	0.14%	5.68e-09	2.19e-13	27
unclassified	2455173	0.13%	2.66e-12	1.98e-13	38
TOTAL	1941127789	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	79.3	-0.17x	13.31x	77.1	-0.18x	-0.56x
ev/clk/SM	0.26	-0.17x	13.31x	0.27	-0.18x	-0.56x
clk/ev	800.4	-0.29x	7.18x	801.5	-0.29x	-0.97x

SASS Instructions:

Class	Count
fp32	305
fp64	0
other	69
total	374

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	+0	+0	-0	nan	nan	nan	1	1	nan	nan	nan	-inf	+inf	+inf	nan
-maxN	+0	nan	-0	nan	nan	nan	1	1	nan	nan	nan	-3.4e+38	nan	+inf	nan
-1	1	1	-1	nan	nan	nan	1	1	nan	nan	nan	-1	1	1	nan
-minN	+inf	nan	-8.5e+37	nan	nan	nan	1	1	nan	nan	nan	-0	nan	+0	nan
-maxD	+inf	nan	-8.5e+37	nan	nan	nan	1	1	nan	nan	nan	-0	nan	+0	nan
-minD	+inf	nan	-inf	nan	nan	nan	1	1	nan	nan	nan	-0	nan	+0	nan
-0	+inf	+inf	+inf	+inf	+inf	+inf	1	1	+0	+0	+0	+0	+0	+0	nan
+0	+inf	+inf	+inf	+inf	+inf	+inf	1	1	+0	+0	+0	+0	+0	+0	nan
+minD	+inf	nan	+inf	1	1	1	1	1	1	1	1	+0	nan	+0	nan
+maxD	+inf	nan	8.5e+37	1	1	1	1	1	1	1	1	+0	nan	+0	nan
+minN	+inf	nan	8.5e+37	1	1	1	1	1	1	1	1	+0	nan	+0	nan
+1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1
+maxN	+0	nan	+0	1	1	1	1	1	1	1	1	3.4e+38	nan	+inf	nan
+INF	+0	+0	+0	+0	+0	+0	1	1	+inf	+inf	+inf	+inf	+inf	+inf	nan
QNAN	nan	nan	nan	nan	nan	nan	1	1	nan	nan	nan	nan	nan	nan	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Exponential (exp)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2233993655	68.53%	1.00e-13	6.19e-16	43
output special	1020169796	31.29%	–	–	–
input special	1	3e-08%	–	–	–
output denormal	2180472	0.07%	1.00e+00	1.00e+00	0
output near inf	66521	2e-03%	3.53e-13	1.73e-13	41
unclassified	3608715	0.11%	2.83e-08	1.94e-13	25
TOTAL	3260019160	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	218.6	-0.06x	6.55x	211.2	-0.06x	-0.24x
ev/clk/SM	0.73	-0.06x	6.55x	0.73	-0.06x	-0.24x
clk/ev	322.8	-0.16x	3.06x	323.4	-0.17x	-0.50x

SASS Instructions:

Class	Count
fp32	136
fp64	0
other	14
total	150

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Base-2 Exponential (exp2)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2245060170	68.79%	1.00e-13	6.76e-16	43
output special	1015021568	31.10%	–	–	–
input special	1	3e-08%	–	–	–
output denormal	573105	0.02%	1.67e-01	6.81e-06	2
output near inf	48145	1e-03%	3.49e-13	1.37e-13	41
unclassified	2776486	0.09%	1.52e-08	2.11e-13	25
TOTAL	3263479475	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	217.8	-0.06x	6.97x	210.4	-0.06x	-0.25x
ev/clk/SM	0.72	-0.06x	6.97x	0.72	-0.06x	-0.25x
clk/ev	316.8	-0.13x	3.31x	315.6	-0.14x	-0.52x

SASS Instructions:

Class	Count
fp32	126
fp64	0
other	21
total	147

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Base-10 Exponential (exp10)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2217857141	68.28%	9.99e-14	8.53e-17	43
output special	1030086470	31.71%	–	–	–
input special	1	3e-08%	–	–	–
output denormal	39806	1e-03%	1.54e-02	2.17e-06	6
unclassified	2079	6e-05%	6.18e-10	1.01e-12	30
TOTAL	3247985497	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	183.9	-0.05x	6.25x	176.7	-0.05x	-0.22x
ev/clk/SM	0.61	-0.05x	6.25x	0.61	-0.05x	-0.22x
clk/ev	490.0	-0.10x	2.28x	489.5	-0.11x	-0.36x

SASS Instructions:

Class	Count
fp32	198
fp64	0
other	21
total	219

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Exponential Minus One (expm1)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3239165628	76.01%	1.00e-13	3.33e-16	43
output special	1020169796	23.94%	–	–	–
input special	1	2e-08%	–	–	–
output near inf	66521	2e-03%	3.53e-13	1.73e-13	41
unclassified	2010918	0.05%	6.53e-13	1.56e-13	40
TOTAL	4261412864	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	160.2	-0.11x	6.00x	154.7	-0.11x	-0.32x
ev/clk/SM	0.53	-0.11x	6.00x	0.53	-0.11x	-0.32x
clk/ev	478.1	-0.15x	2.76x	478.3	-0.15x	-0.46x

SASS Instructions:

Class	Count
fp32	285
fp64	0
other	28
total	313

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Natural Logarithm (log)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2139095037	49.80%	5.03e-14	8.02e-16	44
output special	2139095043	49.80%	–	–	–
input special	16777216	0.39%	–	–	–
TOTAL	4294967296	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	211.2	-0.16x	12.35x	202.3	-0.16x	-0.59x
ev/clk/SM	0.70	-0.16x	12.35x	0.70	-0.16x	-0.59x
clk/ev	330.7	-0.27x	6.00x	332.5	-0.27x	-0.82x

SASS Instructions:

Class	Count
fp32	135
fp64	0
other	20
total	155

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Base-2 Logarithm (log2)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2139095037	49.80%	5.01e-14	1.37e-15	44
output special	2139095043	49.80%	–	–	–
input special	16777216	0.39%	–	–	–
TOTAL	4294967296	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	191.7	-0.15x	11.98x	185.1	-0.15x	-0.55x
ev/clk/SM	0.64	-0.15x	11.98x	0.64	-0.15x	-0.55x
clk/ev	384.1	-0.25x	5.51x	385.3	-0.25x	-0.75x

SASS Instructions:

Class	Count
fp32	145
fp64	0
other	21
total	166

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Base-10 Logarithm (log10)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2139095037	49.80%	5.15e-14	1.47e-15	44
output special	2139095043	49.80%	–	–	–
input special	16777216	0.39%	–	–	–
TOTAL	4294967296	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	190.9	-0.15x	11.94x	184.4	-0.15x	-0.55x
ev/clk/SM	0.63	-0.15x	11.94x	0.63	-0.15x	-0.55x
clk/ev	384.0	-0.24x	5.51x	385.5	-0.24x	-0.75x

SASS Instructions:

Class	Count
fp32	145
fp64	0
other	21
total	166

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Logarithm of One Plus x (log1p)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3187671040	100.00%	8.71e-14	4.48e-16	43
input special	1	3e-08%	–	–	–
TOTAL	3187671041	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	140.0	-0.10x	8.49x	136.0	-0.10x	-0.41x
ev/clk/SM	0.47	-0.10x	8.49x	0.47	-0.10x	-0.41x
clk/ev	538.8	-0.16x	3.91x	539.5	-0.16x	-0.62x

SASS Instructions:

Class	Count
fp32	288
fp64	0
other	52
total	340

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Sine (sin)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4233002838	99.33%	1.00e-13	2.64e-15	43
input near inf	441308	0.01%	1.24e-12	1.44e-13	39
unclassified	27968718	0.66%	1.75e-08	1.45e-13	25
TOTAL	4261412864	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	120.1	-0.10x	4.19x	115.5	-0.09x	-0.21x
ev/clk/SM	0.40	-0.10x	4.19x	0.40	-0.09x	-0.21x
clk/ev	505.0	-0.22x	2.45x	506.6	-0.23x	-0.52x

SASS Instructions:

Class	Count
fp32	226
fp64	0
other	227
total	453

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Cosine (cos)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4249101016	99.32%	1.00e-13	2.61e-15	43
input near inf	441836	0.01%	2.72e-12	1.44e-13	38
unclassified	28647228	0.67%	1.99e-09	1.46e-13	28
TOTAL	4278190080	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	120.2	-0.11x	4.20x	115.4	-0.10x	-0.23x
ev/clk/SM	0.40	-0.11x	4.20x	0.40	-0.10x	-0.23x
clk/ev	505.0	-0.23x	2.49x	506.3	-0.23x	-0.56x

SASS Instructions:

Class	Count
fp32	226
fp64	0
other	230
total	456

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Tangent (tan)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4199082652	98.54%	1.00e-13	4.81e-15	43
input near inf	961493	0.02%	2.72e-12	1.46e-13	38
unclassified	61368719	1.44%	1.75e-08	1.48e-13	25
TOTAL	4261412864	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	110.2	-0.08x	6.68x	106.5	-0.08x	-0.30x
ev/clk/SM	0.37	-0.08x	6.68x	0.37	-0.08x	-0.30x
clk/ev	562.3	-0.23x	3.84x	561.4	-0.24x	-0.69x

SASS Instructions:

Class	Count
fp32	242
fp64	0
other	229
total	471

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Sine of Pi Times x (sinpi)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2801853403	96.12%	1.00e-13	2.98e-15	43
output denormal	1076580	0.04%	2.98e-05	2.38e-07	15
input denormal	100811746	3.46%	5.96e-08	2.79e-09	24
output near denormal	1986404	0.07%	1.19e-07	8.60e-08	23
unclassified	9245451	0.32%	2.63e-13	1.45e-13	41
TOTAL	2914973584	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	97.5	-0.07x	3.38x	94.7	-0.07x	-0.16x
ev/clk/SM	0.32	-0.07x	3.38x	0.33	-0.07x	-0.16x
clk/ev	649.0	-0.11x	2.03x	648.2	-0.11x	-0.32x

SASS Instructions:

Class	Count
fp32	269
fp64	0
other	73
total	342

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

---

Cosine of Pi Times x (cospi)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4235387247	99.39%	1.00e-13	9.13e-16	43
unclassified	26023294	0.61%	2.64e-13	1.73e-13	41
TOTAL	4261410541	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	103.6	-0.07x	3.19x	99.4	-0.07x	-0.16x
ev/clk/SM	0.34	-0.07x	3.19x	0.34	-0.07x	-0.16x
clk/ev	613.6	-0.13x	1.92x	613.2	-0.12x	-0.36x

SASS Instructions:

Class	Count
fp32	265
fp64	0
other	66
total	331

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

---

Arc Sine (asin)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2113929218	100.00%	4.75e-14	9.84e-17	44
TOTAL	2113929218	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	113.3	-0.07x	5.98x	109.5	-0.07x	-0.24x
ev/clk/SM	0.38	-0.07x	5.98x	0.38	-0.07x	-0.24x
clk/ev	425.1	-0.23x	4.56x	426.4	-0.23x	-0.47x

SASS Instructions:

Class	Count
fp32	432
fp64	0
other	9
total	441

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Arc Cosine (acos)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2130706434	100.00%	3.41e-14	5.00e-16	44
TOTAL	2130706434	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	124.5	-0.09x	6.57x	120.5	-0.09x	-0.31x
ev/clk/SM	0.41	-0.09x	6.57x	0.41	-0.09x	-0.31x
clk/ev	375.2	-0.29x	5.25x	375.6	-0.29x	-0.61x

SASS Instructions:

Class	Count
fp32	454
fp64	0
other	10
total	464

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Arc Tangent (atan)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261412864	100.00%	2.82e-14	3.29e-16	45
input special	3	7e-08%	–	–	–
TOTAL	4261412867	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	101.8	-0.06x	5.62x	97.8	-0.06x	-0.20x
ev/clk/SM	0.34	-0.06x	5.62x	0.34	-0.06x	-0.20x
clk/ev	434.7	-0.20x	4.19x	434.4	-0.20x	-0.62x

SASS Instructions:

Class	Count
fp32	264
fp64	0
other	6
total	270

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Two-Argument Arc Tangent (atan2)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3883669366	97.32%	1.00e-13	8.70e-16	43
output special	65536	2e-03%	–	–	–
output denormal	1616789	0.04%	1.00e+00	9.73e-01	0
input denormal	88376055	2.21%	1.79e-07	4.59e-09	22
output near denormal	103777	3e-03%	1.00e+00	6.31e-01	0
input near inf	16487858	0.41%	1.00e+00	5.06e-01	0
cancellation	96863	2e-03%	4.02e-09	1.02e-12	27
TOTAL	3990416244	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	90.5	-0.11x	6.90x	87.0	-0.11x	-0.31x
ev/clk/SM	0.30	-0.11x	6.90x	0.30	-0.11x	-0.31x
clk/ev	486.7	-0.26x	5.10x	486.6	-0.26x	-0.72x

SASS Instructions:

Class	Count
fp32	299
fp64	0
other	46
total	345

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	-2.4	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-0.79	nan
-maxN	-3.1	-2.4	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-0.79	-0	nan
-1	-3.1	-3.1	-2.4	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-1.6	-0.79	-2.9e-39	-0	nan
-minN	-3.1	-3.1	-3.1	-2.4	-2.4	-1.6	-1.6	-1.6	-1.6	-0.79	-0.79	-1.2e-38	-0	-0	nan
-maxD	-3.1	-3.1	-3.1	-2.4	-2.4	-1.6	-1.6	-1.6	-1.6	-0.79	-0.79	-1.2e-38	-0	-0	nan
-minD	-3.1	-3.1	-3.1	-3.1	-3.1	nan	nan	nan	nan	-1.2e-07	-1.2e-07	-1.4e-45	-0	-0	nan
-0	3.1	3.1	3.1	3.1	3.1	nan	+0	+0	nan	+0	+0	+0	+0	+0	nan
+0	3.1	3.1	3.1	3.1	3.1	nan	+0	+0	nan	+0	+0	+0	+0	+0	nan
+minD	3.1	3.1	3.1	3.1	3.1	nan	nan	nan	nan	1.2e-07	1.2e-07	1.4e-45	+0	+0	nan
+maxD	3.1	3.1	3.1	2.4	2.4	1.6	1.6	1.6	1.6	0.79	0.79	1.2e-38	+0	+0	nan
+minN	3.1	3.1	3.1	2.4	2.4	1.6	1.6	1.6	1.6	0.79	0.79	1.2e-38	+0	+0	nan
+1	3.1	3.1	2.4	1.6	1.6	1.6	1.6	1.6	1.6	1.6	1.6	0.79	2.9e-39	+0	nan
+maxN	3.1	2.4	1.6	1.6	1.6	1.6	1.6	1.6	1.6	1.6	1.6	1.6	0.79	+0	nan
+INF	2.4	1.6	1.6	1.6	1.6	1.6	1.6	1.6	1.6	1.6	1.6	1.6	1.6	0.79	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Hyperbolic Sine (sinh)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2216918324	99.80%	1.00e-13	7.17e-16	43
output special	181863	8e-03%	–	–	–
output near inf	22838	1e-03%	2.98e-13	2.27e-13	41
unclassified	4132326	0.19%	6.53e-13	1.57e-13	40
TOTAL	2221255351	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	167.1	-0.12x	9.71x	161.3	-0.13x	-0.52x
ev/clk/SM	0.56	-0.12x	9.71x	0.55	-0.13x	-0.52x
clk/ev	427.2	-0.25x	5.16x	426.1	-0.26x	-0.93x

SASS Instructions:

Class	Count
fp32	288
fp64	0
other	23
total	311

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Hyperbolic Cosine (cosh)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2233695454	99.81%	1.00e-13	7.95e-16	43
output special	181863	8e-03%	–	–	–
output near inf	22838	1e-03%	2.98e-13	2.27e-13	41
unclassified	4132412	0.18%	6.53e-13	1.57e-13	40
TOTAL	2238032567	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	181.2	-0.11x	7.47x	175.6	-0.13x	-0.35x
ev/clk/SM	0.60	-0.11x	7.47x	0.60	-0.13x	-0.35x
clk/ev	400.1	-0.26x	3.49x	399.6	-0.26x	-0.61x

SASS Instructions:

Class	Count
fp32	160
fp64	0
other	18
total	178

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Hyperbolic Tangent (tanh)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261412864	100.00%	2.47e-14	4.61e-17	45
TOTAL	4261412864	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	156.0	-0.07x	7.12x	151.0	-0.08x	-0.33x
ev/clk/SM	0.52	-0.07x	7.12x	0.52	-0.08x	-0.33x
clk/ev	330.7	-0.20x	2.75x	331.5	-0.21x	-0.47x

SASS Instructions:

Class	Count
fp32	310
fp64	0
other	28
total	338

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Inverse Hyperbolic Sine (asinh)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261412864	100.00%	8.97e-14	7.61e-16	43
input special	3	7e-08%	–	–	–
TOTAL	4261412867	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	91.7	-0.11x	7.88x	88.5	-0.11x	-0.38x
ev/clk/SM	0.30	-0.11x	7.88x	0.30	-0.11x	-0.38x
clk/ev	857.4	-0.19x	3.49x	856.6	-0.19x	-0.55x

SASS Instructions:

Class	Count
fp32	521
fp64	0
other	88
total	609

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Inverse Hyperbolic Cosine (acosh)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	1073741822	100.00%	6.57e-14	1.14e-15	43
input special	1	9e-08%	–	–	–
TOTAL	1073741823	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	94.9	-0.11x	7.70x	91.0	-0.11x	-0.37x
ev/clk/SM	0.32	-0.11x	7.70x	0.31	-0.11x	-0.37x
clk/ev	606.3	-0.24x	3.85x	606.4	-0.24x	-0.67x

SASS Instructions:

Class	Count
fp32	517
fp64	0
other	85
total	602

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Inverse Hyperbolic Tangent (atanh)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2113929216	100.00%	3.00e-14	1.00e-16	44
output special	2	9e-08%	–	–	–
TOTAL	2113929218	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	121.7	-0.12x	9.11x	116.8	-0.12x	-0.43x
ev/clk/SM	0.40	-0.12x	9.11x	0.40	-0.12x	-0.43x
clk/ev	358.0	-0.36x	6.49x	358.3	-0.37x	1.22x

SASS Instructions:

Class	Count
fp32	463
fp64	0
other	68
total	531

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Error Function (erf)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4165995376	97.72%	1.00e-13	5.74e-16	43
output denormal	53642	1e-03%	3.45e-05	2.38e-07	14
input denormal	97165519	2.28%	5.96e-08	1.44e-09	24
output near denormal	160571	4e-03%	1.19e-07	7.95e-08	23
TOTAL	4263375108	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	70.2	-0.05x	6.23x	67.8	-0.05x	-0.20x
ev/clk/SM	0.23	-0.05x	6.23x	0.23	-0.05x	-0.20x
clk/ev	660.4	-0.11x	4.53x	658.6	-0.11x	-0.59x

SASS Instructions:

Class	Count
fp32	560
fp64	0
other	22
total	582

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Complementary Error Function (erfc)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	1465663971	45.36%	1.00e-13	1.28e-14	43
output denormal	35872	1e-03%	1.00e+00	2.12e-02	0
input denormal	436207615	13.50%	1.03e-13	1.03e-13	43
output near denormal	13089	4e-04%	1.19e-07	8.60e-08	23
unclassified	1328981987	41.13%	5.96e-08	9.69e-13	24
TOTAL	3230902534	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	46.0	-0.06x	5.58x	44.5	-0.06x	-0.24x
ev/clk/SM	0.15	-0.06x	5.58x	0.15	-0.06x	-0.24x
clk/ev	724.8	-0.19x	5.91x	730.6	-0.19x	-0.56x

SASS Instructions:

Class	Count
fp32	525
fp64	0
other	20
total	545

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Inverse Normal CDF (normcdfinv)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	162681277	97.20%	1.00e-13	7.38e-15	43
unclassified	4683464	2.80%	4.37e-13	1.47e-13	41
TOTAL	167364741	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	61.1	-0.04x	6.75x	59.4	-0.04x	-0.31x
ev/clk/SM	0.20	-0.04x	6.75x	0.20	-0.04x	-0.31x
clk/ev	869.4	-0.13x	4.20x	869.5	-0.13x	-0.58x

SASS Instructions:

Class	Count
fp32	735
fp64	0
other	36
total	771

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

---

Boys Function F0 (boys_f0)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	325851359	100.00%	9.88e-14	1.84e-14	43
unclassified	9	3e-06%	1.09e-13	1.04e-13	43
TOTAL	325851368	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	93.7	13.50x	13.13x	90.5	-0.64x	-0.64x
ev/clk/SM	0.31	13.50x	13.13x	0.31	-0.64x	-0.64x
clk/ev	352.6	14.15x	13.79x	352.3	2.24x	2.18x

SASS Instructions:

Class	Count
fp32	749
fp64	0
other	11
total	760

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Floor (floor)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3212836860	100.00%	0.00e+00	0.00e+00	48
TOTAL	3212836860	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1971.3	-0.43x	3.33x	1908.8	-0.45x	-0.47x
ev/clk/SM	6.55	-0.43x	3.33x	6.56	-0.45x	-0.47x
clk/ev	102.0	-0.26x	1.04x	101.3	-0.26x	-0.28x

SASS Instructions:

Class	Count
fp32	28
fp64	0
other	13
total	41

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Ceiling (ceil)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	3212836861	100.00%	0.00e+00	0.00e+00	48
TOTAL	3212836861	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1972.0	-0.43x	3.33x	1911.4	-0.45x	-0.47x
ev/clk/SM	6.56	-0.43x	3.33x	6.57	-0.45x	-0.47x
clk/ev	102.2	-0.26x	1.04x	101.3	-0.26x	-0.28x

SASS Instructions:

Class	Count
fp32	28
fp64	0
other	13
total	41

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Round to Nearest (round)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2164260863	100.00%	3.55e-15	9.54e-18	48
TOTAL	2164260863	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	952.7	-0.21x	3.66x	923.0	-0.21x	-0.25x
ev/clk/SM	3.17	-0.21x	3.66x	3.17	-0.21x	-0.25x
clk/ev	204.2	-0.18x	-0.82x	203.2	-0.18x	-0.22x

SASS Instructions:

Class	Count
fp32	71
fp64	0
other	29
total	100

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Truncate (trunc)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2147483646	100.00%	0.00e+00	0.00e+00	48
TOTAL	2147483646	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	2038.4	-0.44x	3.44x	1964.8	-0.46x	-0.49x
ev/clk/SM	6.78	-0.44x	3.44x	6.76	-0.46x	-0.49x
clk/ev	102.0	-0.26x	1.03x	101.4	-0.26x	-0.28x

SASS Instructions:

Class	Count
fp32	35
fp64	0
other	23
total	58

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Floating-Point Remainder (fmod)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4205136149	100.00%	3.55e-15	8.72e-17	48
TOTAL	4205136149	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	845.5	-0.20x	3.39x	816.6	-0.20x	-0.49x
ev/clk/SM	2.81	-0.20x	3.39x	2.81	-0.20x	-0.49x
clk/ev	105.9	-0.44x	1.12x	106.5	-0.44x	-0.78x

SASS Instructions:

Class	Count
fp32	41
fp64	0
other	235
total	276

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan
-maxN	-3.4e+38	-0	-0	-0	-1.4e-45	-0	nan	nan	-0	-1.4e-45	-0	-0	-0	-3.4e+38	nan
-1	-1	-1	-0	-0	-2.9e-42	-0	nan	nan	-0	-2.9e-42	-0	-0	-1	-1	nan
-minN	-1.2e-38	-1.2e-38	-1.2e-38	-0	-1.4e-45	-0	nan	nan	-0	-1.4e-45	-0	-1.2e-38	-1.2e-38	-1.2e-38	nan
-maxD	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-0	-0	nan	nan	-0	-0	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	nan
-minD	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-0	nan	nan	-0	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	nan
-0	-0	-0	-0	-0	-0	-0	nan	nan	-0	-0	-0	-0	-0	-0	nan
+0	+0	+0	+0	+0	+0	+0	nan	nan	+0	+0	+0	+0	+0	+0	nan
+minD	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	+0	nan	nan	+0	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	nan
+maxD	1.2e-38	1.2e-38	1.2e-38	1.2e-38	+0	+0	nan	nan	+0	+0	1.2e-38	1.2e-38	1.2e-38	1.2e-38	nan
+minN	1.2e-38	1.2e-38	1.2e-38	+0	1.4e-45	+0	nan	nan	+0	1.4e-45	+0	1.2e-38	1.2e-38	1.2e-38	nan
+1	1	1	+0	+0	2.9e-42	+0	nan	nan	+0	2.9e-42	+0	+0	1	1	nan
+maxN	3.4e+38	+0	+0	+0	1.4e-45	+0	nan	nan	+0	1.4e-45	+0	+0	+0	3.4e+38	nan
+INF	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

IEEE Remainder (remainder)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4188533852	100.00%	2.93e-15	7.00e-25	48
TOTAL	4188533852	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	627.1	-0.26x	7.52x	606.5	-0.27x	-0.66x
ev/clk/SM	2.09	-0.26x	7.52x	2.09	-0.27x	-0.66x
clk/ev	216.8	-0.35x	2.13x	216.4	-0.35x	-0.53x

SASS Instructions:

Class	Count
fp32	65
fp64	0
other	275
total	340

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan
-maxN	-3.4e+38	-0	-0	-0	-1.4e-45	-0	nan	nan	-0	-1.4e-45	-0	-0	-0	-3.4e+38	nan
-1	-1	-1	-0	-0	-2.9e-42	-0	nan	nan	-0	-2.9e-42	-0	-0	-1	-1	nan
-minN	-1.2e-38	-1.2e-38	-1.2e-38	-0	-1.4e-45	-0	nan	nan	-0	-1.4e-45	-0	-1.2e-38	-1.2e-38	-1.2e-38	nan
-maxD	-1.2e-38	-1.2e-38	-1.2e-38	1.4e-45	-0	-0	nan	nan	-0	-0	1.4e-45	-1.2e-38	-1.2e-38	-1.2e-38	nan
-minD	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-0	nan	nan	-0	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	nan
-0	-0	-0	-0	-0	-0	-0	nan	nan	-0	-0	-0	-0	-0	-0	nan
+0	+0	+0	+0	+0	+0	+0	nan	nan	+0	+0	+0	+0	+0	+0	nan
+minD	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	+0	nan	nan	+0	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	nan
+maxD	1.2e-38	1.2e-38	1.2e-38	-1.4e-45	+0	+0	nan	nan	+0	+0	-1.4e-45	1.2e-38	1.2e-38	1.2e-38	nan
+minN	1.2e-38	1.2e-38	1.2e-38	+0	1.4e-45	+0	nan	nan	+0	1.4e-45	+0	1.2e-38	1.2e-38	1.2e-38	nan
+1	1	1	+0	+0	2.9e-42	+0	nan	nan	+0	2.9e-42	+0	+0	1	1	nan
+maxN	3.4e+38	+0	+0	+0	1.4e-45	+0	nan	nan	+0	1.4e-45	+0	+0	+0	3.4e+38	nan
+INF	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Scale by Power of Two (ldexp)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2222529003	68.64%	1.00e-13	2.72e-19	43
output special	1015103359	31.35%	–	–	–
input special	16225	5e-04%	–	–	–
output denormal	119527	4e-03%	1.00e+00	6.30e-04	0
input denormal	122385	4e-03%	5.96e-08	6.69e-09	24
output near denormal	2982	9e-05%	1.19e-07	8.24e-08	23
cancellation	9038	3e-04%	5.74e-08	2.23e-10	24
TOTAL	3237902519	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	259.8	-0.04x	2.08x	1391.7	-0.21x	-0.41x
ev/clk/SM	0.86	-0.04x	2.08x	4.79	-0.21x	-0.41x
clk/ev	149.7	-0.17x	1.97x	58.0	-0.44x	-0.80x

SASS Instructions:

Class	Count
fp32	22
fp64	9
other	52
total	83

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf
-maxN	-0	-0	-1.7e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-inf	-inf	-inf	-3.4e+38
-1	-0	-0	-0.5	-1	-1	-1	-1	-1	-1	-1	-1	-2	-inf	-inf	-1
-minN	-0	-0	-5.9e-39	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-2.4e-38	-inf	-inf	-1.2e-38
-maxD	-0	-0	-5.9e-39	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-2.4e-38	-inf	-inf	-1.2e-38
-minD	-0	-0	-0	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-2.8e-45	-inf	-inf	-1.4e-45
-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0
+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0
+minD	+0	+0	+0	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	2.8e-45	+inf	+inf	1.4e-45
+maxD	+0	+0	5.9e-39	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	+inf	+inf	1.2e-38
+minN	+0	+0	5.9e-39	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	+inf	+inf	1.2e-38
+1	+0	+0	0.5	1	1	1	1	1	1	1	1	2	+inf	+inf	1
+maxN	+0	+0	1.7e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	+inf	+inf	+inf	3.4e+38
+INF	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Scale by Power of Two (scalbn)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2222529003	68.64%	1.00e-13	2.72e-19	43
output special	1015103359	31.35%	–	–	–
input special	16225	5e-04%	–	–	–
output denormal	119527	4e-03%	1.00e+00	6.30e-04	0
input denormal	122385	4e-03%	5.96e-08	6.69e-09	24
output near denormal	2982	9e-05%	1.19e-07	8.24e-08	23
cancellation	9038	3e-04%	5.74e-08	2.23e-10	24
TOTAL	3237902519	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	259.8	-0.04x	2.08x	1391.5	-0.21x	-0.40x
ev/clk/SM	0.86	-0.04x	2.08x	4.78	-0.21x	-0.40x
clk/ev	149.8	-0.17x	1.97x	58.1	-0.44x	-0.80x

SASS Instructions:

Class	Count
fp32	22
fp64	9
other	52
total	83

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf
-maxN	-0	-0	-1.7e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-inf	-inf	-inf	-3.4e+38
-1	-0	-0	-0.5	-1	-1	-1	-1	-1	-1	-1	-1	-2	-inf	-inf	-1
-minN	-0	-0	-5.9e-39	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-2.4e-38	-inf	-inf	-1.2e-38
-maxD	-0	-0	-5.9e-39	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-2.4e-38	-inf	-inf	-1.2e-38
-minD	-0	-0	-0	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-2.8e-45	-inf	-inf	-1.4e-45
-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0
+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0
+minD	+0	+0	+0	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	2.8e-45	+inf	+inf	1.4e-45
+maxD	+0	+0	5.9e-39	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	+inf	+inf	1.2e-38
+minN	+0	+0	5.9e-39	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	+inf	+inf	1.2e-38
+1	+0	+0	0.5	1	1	1	1	1	1	1	1	2	+inf	+inf	1
+maxN	+0	+0	1.7e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	+inf	+inf	+inf	3.4e+38
+INF	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Note: ``fp64mp2`` is a thin wrapper over the system ``fp64`` (or ``fp128`` reference) math for this function and is omitted from the spec.

---

Scale by Power of Two, Long Exponent (scalbln)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2222529003	68.64%	1.00e-13	2.72e-19	43
output special	1015103359	31.35%	–	–	–
input special	16225	5e-04%	–	–	–
output denormal	119527	4e-03%	1.00e+00	6.30e-04	0
input denormal	122385	4e-03%	5.96e-08	6.69e-09	24
output near denormal	2982	9e-05%	1.19e-07	8.24e-08	23
cancellation	9038	3e-04%	5.74e-08	2.23e-10	24
TOTAL	3237902519	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	259.6	-0.04x	2.08x	1391.7	-0.21x	-0.40x
ev/clk/SM	0.86	-0.04x	2.08x	4.79	-0.21x	-0.40x
clk/ev	149.7	-0.17x	1.97x	57.6	-0.45x	-0.80x

SASS Instructions:

Class	Count
fp32	22
fp64	9
other	52
total	83

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf
-maxN	-0	-0	-1.7e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-3.4e+38	-inf	-inf	-inf	-3.4e+38
-1	-0	-0	-0.5	-1	-1	-1	-1	-1	-1	-1	-1	-2	-inf	-inf	-1
-minN	-0	-0	-5.9e-39	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-2.4e-38	-inf	-inf	-1.2e-38
-maxD	-0	-0	-5.9e-39	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-1.2e-38	-2.4e-38	-inf	-inf	-1.2e-38
-minD	-0	-0	-0	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-1.4e-45	-2.8e-45	-inf	-inf	-1.4e-45
-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0
+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0
+minD	+0	+0	+0	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	1.4e-45	2.8e-45	+inf	+inf	1.4e-45
+maxD	+0	+0	5.9e-39	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	+inf	+inf	1.2e-38
+minN	+0	+0	5.9e-39	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	1.2e-38	2.4e-38	+inf	+inf	1.2e-38
+1	+0	+0	0.5	1	1	1	1	1	1	1	1	2	+inf	+inf	1
+maxN	+0	+0	1.7e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	3.4e+38	+inf	+inf	+inf	3.4e+38
+INF	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	15534530	96.21%	0.00e+00	0.00e+00	106
output special	610078	3.78%	–	–	–
output denormal	1035	6e-03%	2.27e-13	-3.64e-15	42
TOTAL	16145643	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	55.6	-0.44x	3.88x	1520.5	-0.44x	10.91x
ev/clk/SM	0.18	-0.44x	3.88x	5.23	-0.44x	10.91x
clk/ev	636.7	-0.46x	3.45x	51.7	-0.90x	6.30x

SASS Instructions:

Class	Count
fp32	0
fp64	14
other	22
total	36

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf	-inf
-maxN	-8.8e+217	-8.8e+217	-9.0e+307	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-1.8e+308	-inf	-inf	-inf	-1.8e+308
-1	-4.9e-91	-4.9e-91	-0.5	-1	-1	-1	-1	-1	-1	-1	-1	-2	-2.0e+90	-2.0e+90	-1
-minN	-0	-0	-1.1e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-4.5e-308	-4.5e-218	-4.5e-218	-2.2e-308
-maxD	-0	-0	-1.1e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-2.2e-308	-4.5e-308	-4.5e-218	-4.5e-218	-2.2e-308
-minD	-0	-0	-0	-4.9e-324	-4.9e-324	-4.9e-324	-4.9e-324	-4.9e-324	-4.9e-324	-4.9e-324	-4.9e-324	-9.9e-324	-1.0e-233	-1.0e-233	-4.9e-324
-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0	-0
+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0	+0
+minD	+0	+0	+0	4.9e-324	4.9e-324	4.9e-324	4.9e-324	4.9e-324	4.9e-324	4.9e-324	4.9e-324	9.9e-324	1.0e-233	1.0e-233	4.9e-324
+maxD	+0	+0	1.1e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	4.5e-308	4.5e-218	4.5e-218	2.2e-308
+minN	+0	+0	1.1e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	2.2e-308	4.5e-308	4.5e-218	4.5e-218	2.2e-308
+1	4.9e-91	4.9e-91	0.5	1	1	1	1	1	1	1	1	2	2.0e+90	2.0e+90	1
+maxN	8.8e+217	8.8e+217	9.0e+307	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	1.8e+308	+inf	+inf	+inf	1.8e+308
+INF	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf	+inf
QNAN	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan	nan

---

Split into Significand and Exponent (frexp)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4278190075	100.00%	0.00e+00	0.00e+00	48
input special	3	7e-08%	–	–	–
TOTAL	4278190078	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	296.5	-0.10x	1.75x	706.7	-0.24x	-0.33x
ev/clk/SM	0.99	-0.10x	1.75x	2.43	-0.24x	-0.33x
clk/ev	189.8	-0.21x	1.64x	147.4	-0.28x	-0.45x

SASS Instructions:

Class	Count
fp32	13
fp64	2
other	25
total	40

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16769024	100.00%	0.00e+00	0.00e+00	106
TOTAL	16769024	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	46.1	-0.27x	-0.16x	695.0	-0.32x	2.55x
ev/clk/SM	0.15	-0.27x	-0.16x	2.39	-0.32x	2.55x
clk/ev	950.9	-0.33x	-0.72x	173.1	-0.38x	2.88x

SASS Instructions:

Class	Count
fp32	0
fp64	11
other	28
total	39

Special Values Table:

Input	Value
-INF	-inf
-maxN	-1.8e+308
-1	-1
-minN	-2.2e-308
-maxD	-2.2e-308
-minD	-4.9e-324
-0	-0
+0	+0
+minD	4.9e-324
+maxD	2.2e-308
+minN	2.2e-308
+1	1
+maxN	1.8e+308
+INF	+inf
QNAN	nan

Comparison Operations

Equal (eq)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261478400	100.00%	0.00e+00	0.00e+00	1
TOTAL	4261478400	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	4094.9	-0.66x	8.21x	3960.3	-0.66x	-0.95x
ev/clk/SM	13.62	-0.66x	8.21x	13.62	-0.66x	-0.95x
clk/ev	18.6	-0.80x	5.84x	18.4	-0.83x	1.41x

SASS Instructions:

Class	Count
fp32	2
fp64	0
other	1
total	3

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0
-maxN	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0
-1	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0
-minN	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0
-maxD	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0
-minD	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0
-0	0	0	0	0	0	0	1	1	0	0	0	0	0	0	0
+0	0	0	0	0	0	0	1	1	0	0	0	0	0	0	0
+minD	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0
+maxD	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0
+minN	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0
+1	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0
+maxN	0	0	0	0	0	0	0	0	0	0	0	0	1	0	0
+INF	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760836	100.00%	0.00e+00	0.00e+00	1
TOTAL	16760836	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	249.6	-0.50x	-0.50x	2372.8	-0.57x	4.84x
ev/clk/SM	0.83	-0.50x	-0.50x	8.16	-0.57x	4.84x
clk/ev	207.4	-0.52x	-0.53x	30.2	-0.85x	3.61x

SASS Instructions:

Class	Count
fp32	0
fp64	2
other	1
total	3

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0
-maxN	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0
-1	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0
-minN	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0
-maxD	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0
-minD	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0
-0	0	0	0	0	0	0	1	1	0	0	0	0	0	0	0
+0	0	0	0	0	0	0	1	1	0	0	0	0	0	0	0
+minD	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0
+maxD	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0
+minN	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0
+1	0	0	0	0	0	0	0	0	0	0	0	1	0	0	0
+maxN	0	0	0	0	0	0	0	0	0	0	0	0	1	0	0
+INF	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

---

Not Equal (ne)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261478400	100.00%	0.00e+00	0.00e+00	1
TOTAL	4261478400	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	4089.9	-0.66x	8.20x	3956.0	-0.66x	-0.95x
ev/clk/SM	13.60	-0.66x	8.20x	13.60	-0.66x	-0.95x
clk/ev	18.2	-0.82x	5.98x	18.8	-0.81x	1.37x

SASS Instructions:

Class	Count
fp32	2
fp64	0
other	1
total	3

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	0	1	1	1	1	1	1	1	1	1	1	1	1	1	1
-maxN	1	0	1	1	1	1	1	1	1	1	1	1	1	1	1
-1	1	1	0	1	1	1	1	1	1	1	1	1	1	1	1
-minN	1	1	1	0	1	1	1	1	1	1	1	1	1	1	1
-maxD	1	1	1	1	0	1	1	1	1	1	1	1	1	1	1
-minD	1	1	1	1	1	0	1	1	1	1	1	1	1	1	1
-0	1	1	1	1	1	1	0	0	1	1	1	1	1	1	1
+0	1	1	1	1	1	1	0	0	1	1	1	1	1	1	1
+minD	1	1	1	1	1	1	1	1	0	1	1	1	1	1	1
+maxD	1	1	1	1	1	1	1	1	1	0	1	1	1	1	1
+minN	1	1	1	1	1	1	1	1	1	1	0	1	1	1	1
+1	1	1	1	1	1	1	1	1	1	1	1	0	1	1	1
+maxN	1	1	1	1	1	1	1	1	1	1	1	1	0	1	1
+INF	1	1	1	1	1	1	1	1	1	1	1	1	1	0	1
QNAN	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760836	100.00%	0.00e+00	0.00e+00	1
TOTAL	16760836	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	249.6	-0.50x	-0.49x	2372.3	-0.57x	4.83x
ev/clk/SM	0.83	-0.50x	-0.49x	8.16	-0.57x	4.83x
clk/ev	206.6	-0.53x	-0.50x	30.4	-0.85x	3.35x

SASS Instructions:

Class	Count
fp32	0
fp64	2
other	1
total	3

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	0	1	1	1	1	1	1	1	1	1	1	1	1	1	1
-maxN	1	0	1	1	1	1	1	1	1	1	1	1	1	1	1
-1	1	1	0	1	1	1	1	1	1	1	1	1	1	1	1
-minN	1	1	1	0	1	1	1	1	1	1	1	1	1	1	1
-maxD	1	1	1	1	0	1	1	1	1	1	1	1	1	1	1
-minD	1	1	1	1	1	0	1	1	1	1	1	1	1	1	1
-0	1	1	1	1	1	1	0	0	1	1	1	1	1	1	1
+0	1	1	1	1	1	1	0	0	1	1	1	1	1	1	1
+minD	1	1	1	1	1	1	1	1	0	1	1	1	1	1	1
+maxD	1	1	1	1	1	1	1	1	1	0	1	1	1	1	1
+minN	1	1	1	1	1	1	1	1	1	1	0	1	1	1	1
+1	1	1	1	1	1	1	1	1	1	1	1	0	1	1	1
+maxN	1	1	1	1	1	1	1	1	1	1	1	1	0	1	1
+INF	1	1	1	1	1	1	1	1	1	1	1	1	1	0	1
QNAN	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1

---

Less Than (lt)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261478400	100.00%	0.00e+00	0.00e+00	1
TOTAL	4261478400	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	3025.9	-0.49x	6.07x	2943.6	-0.49x	-0.71x
ev/clk/SM	10.06	-0.49x	6.07x	10.12	-0.49x	-0.71x
clk/ev	32.1	-0.46x	3.38x	32.5	-0.47x	-0.80x

SASS Instructions:

Class	Count
fp32	3
fp64	0
other	2
total	5

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	0	1	1	1	1	1	1	1	1	1	1	1	1	1	0
-maxN	0	0	1	1	1	1	1	1	1	1	1	1	1	1	0
-1	0	0	0	1	1	1	1	1	1	1	1	1	1	1	0
-minN	0	0	0	0	1	1	1	1	1	1	1	1	1	1	0
-maxD	0	0	0	0	0	1	1	1	1	1	1	1	1	1	0
-minD	0	0	0	0	0	0	1	1	1	1	1	1	1	1	0
-0	0	0	0	0	0	0	0	0	1	1	1	1	1	1	0
+0	0	0	0	0	0	0	0	0	1	1	1	1	1	1	0
+minD	0	0	0	0	0	0	0	0	0	1	1	1	1	1	0
+maxD	0	0	0	0	0	0	0	0	0	0	1	1	1	1	0
+minN	0	0	0	0	0	0	0	0	0	0	0	1	1	1	0
+1	0	0	0	0	0	0	0	0	0	0	0	0	1	1	0
+maxN	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0
+INF	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760836	100.00%	0.00e+00	0.00e+00	1
TOTAL	16760836	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	166.9	-0.33x	-0.35x	2080.1	-0.50x	4.47x
ev/clk/SM	0.55	-0.33x	-0.35x	7.15	-0.50x	4.47x
clk/ev	279.5	-0.39x	-0.38x	38.9	-0.66x	2.71x

SASS Instructions:

Class	Count
fp32	0
fp64	3
other	2
total	5

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	0	1	1	1	1	1	1	1	1	1	1	1	1	1	0
-maxN	0	0	1	1	1	1	1	1	1	1	1	1	1	1	0
-1	0	0	0	1	1	1	1	1	1	1	1	1	1	1	0
-minN	0	0	0	0	1	1	1	1	1	1	1	1	1	1	0
-maxD	0	0	0	0	0	1	1	1	1	1	1	1	1	1	0
-minD	0	0	0	0	0	0	1	1	1	1	1	1	1	1	0
-0	0	0	0	0	0	0	0	0	1	1	1	1	1	1	0
+0	0	0	0	0	0	0	0	0	1	1	1	1	1	1	0
+minD	0	0	0	0	0	0	0	0	0	1	1	1	1	1	0
+maxD	0	0	0	0	0	0	0	0	0	0	1	1	1	1	0
+minN	0	0	0	0	0	0	0	0	0	0	0	1	1	1	0
+1	0	0	0	0	0	0	0	0	0	0	0	0	1	1	0
+maxN	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0
+INF	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

---

Less Than or Equal (le)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261478400	100.00%	0.00e+00	0.00e+00	1
TOTAL	4261478400	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	3042.5	-0.49x	6.10x	2946.0	-0.49x	-0.71x
ev/clk/SM	10.12	-0.49x	6.10x	10.13	-0.49x	-0.71x
clk/ev	31.6	-0.47x	3.44x	32.1	-0.47x	-0.80x

SASS Instructions:

Class	Count
fp32	3
fp64	0
other	2
total	5

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	1	1	1	1	1	1	1	1	1	1	1	1	1	1	0
-maxN	0	1	1	1	1	1	1	1	1	1	1	1	1	1	0
-1	0	0	1	1	1	1	1	1	1	1	1	1	1	1	0
-minN	0	0	0	1	1	1	1	1	1	1	1	1	1	1	0
-maxD	0	0	0	0	1	1	1	1	1	1	1	1	1	1	0
-minD	0	0	0	0	0	1	1	1	1	1	1	1	1	1	0
-0	0	0	0	0	0	0	1	1	1	1	1	1	1	1	0
+0	0	0	0	0	0	0	1	1	1	1	1	1	1	1	0
+minD	0	0	0	0	0	0	0	0	1	1	1	1	1	1	0
+maxD	0	0	0	0	0	0	0	0	0	1	1	1	1	1	0
+minN	0	0	0	0	0	0	0	0	0	0	1	1	1	1	0
+1	0	0	0	0	0	0	0	0	0	0	0	1	1	1	0
+maxN	0	0	0	0	0	0	0	0	0	0	0	0	1	1	0
+INF	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760836	100.00%	0.00e+00	0.00e+00	1
TOTAL	16760836	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	166.9	-0.33x	-0.35x	2081.8	-0.50x	4.47x
ev/clk/SM	0.55	-0.33x	-0.35x	7.16	-0.50x	4.47x
clk/ev	279.7	-0.39x	-0.41x	39.0	-0.66x	2.90x

SASS Instructions:

Class	Count
fp32	0
fp64	3
other	2
total	5

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	1	1	1	1	1	1	1	1	1	1	1	1	1	1	0
-maxN	0	1	1	1	1	1	1	1	1	1	1	1	1	1	0
-1	0	0	1	1	1	1	1	1	1	1	1	1	1	1	0
-minN	0	0	0	1	1	1	1	1	1	1	1	1	1	1	0
-maxD	0	0	0	0	1	1	1	1	1	1	1	1	1	1	0
-minD	0	0	0	0	0	1	1	1	1	1	1	1	1	1	0
-0	0	0	0	0	0	0	1	1	1	1	1	1	1	1	0
+0	0	0	0	0	0	0	1	1	1	1	1	1	1	1	0
+minD	0	0	0	0	0	0	0	0	1	1	1	1	1	1	0
+maxD	0	0	0	0	0	0	0	0	0	1	1	1	1	1	0
+minN	0	0	0	0	0	0	0	0	0	0	1	1	1	1	0
+1	0	0	0	0	0	0	0	0	0	0	0	1	1	1	0
+maxN	0	0	0	0	0	0	0	0	0	0	0	0	1	1	0
+INF	0	0	0	0	0	0	0	0	0	0	0	0	0	1	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

---

Greater Than (gt)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261478400	100.00%	0.00e+00	0.00e+00	1
TOTAL	4261478400	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	3045.1	-0.49x	6.10x	2942.5	-0.49x	-0.71x
ev/clk/SM	10.13	-0.49x	6.10x	10.12	-0.49x	-0.71x
clk/ev	31.8	-0.48x	3.41x	32.5	-0.47x	-0.79x

SASS Instructions:

Class	Count
fp32	3
fp64	0
other	2
total	5

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
-maxN	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0
-1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0
-minN	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0
-maxD	1	1	1	1	0	0	0	0	0	0	0	0	0	0	0
-minD	1	1	1	1	1	0	0	0	0	0	0	0	0	0	0
-0	1	1	1	1	1	1	0	0	0	0	0	0	0	0	0
+0	1	1	1	1	1	1	0	0	0	0	0	0	0	0	0
+minD	1	1	1	1	1	1	1	1	0	0	0	0	0	0	0
+maxD	1	1	1	1	1	1	1	1	1	0	0	0	0	0	0
+minN	1	1	1	1	1	1	1	1	1	1	0	0	0	0	0
+1	1	1	1	1	1	1	1	1	1	1	1	0	0	0	0
+maxN	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0
+INF	1	1	1	1	1	1	1	1	1	1	1	1	1	0	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760836	100.00%	0.00e+00	0.00e+00	1
TOTAL	16760836	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	166.5	-0.33x	-0.35x	2076.1	-0.50x	4.46x
ev/clk/SM	0.55	-0.33x	-0.35x	7.14	-0.50x	4.46x
clk/ev	282.4	-0.39x	-0.38x	38.9	-0.67x	2.76x

SASS Instructions:

Class	Count
fp32	0
fp64	3
other	2
total	5

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
-maxN	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0
-1	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0
-minN	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0
-maxD	1	1	1	1	0	0	0	0	0	0	0	0	0	0	0
-minD	1	1	1	1	1	0	0	0	0	0	0	0	0	0	0
-0	1	1	1	1	1	1	0	0	0	0	0	0	0	0	0
+0	1	1	1	1	1	1	0	0	0	0	0	0	0	0	0
+minD	1	1	1	1	1	1	1	1	0	0	0	0	0	0	0
+maxD	1	1	1	1	1	1	1	1	1	0	0	0	0	0	0
+minN	1	1	1	1	1	1	1	1	1	1	0	0	0	0	0
+1	1	1	1	1	1	1	1	1	1	1	1	0	0	0	0
+maxN	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0
+INF	1	1	1	1	1	1	1	1	1	1	1	1	1	0	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

---

Greater Than or Equal (ge)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4261478400	100.00%	0.00e+00	0.00e+00	1
TOTAL	4261478400	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	3044.0	-0.49x	6.10x	2940.7	-0.49x	-0.71x
ev/clk/SM	10.12	-0.49x	6.10x	10.11	-0.49x	-0.71x
clk/ev	31.8	-0.46x	3.43x	32.2	-0.48x	-0.80x

SASS Instructions:

Class	Count
fp32	3
fp64	0
other	2
total	5

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0
-maxN	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0
-1	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0
-minN	1	1	1	1	0	0	0	0	0	0	0	0	0	0	0
-maxD	1	1	1	1	1	0	0	0	0	0	0	0	0	0	0
-minD	1	1	1	1	1	1	0	0	0	0	0	0	0	0	0
-0	1	1	1	1	1	1	1	1	0	0	0	0	0	0	0
+0	1	1	1	1	1	1	1	1	0	0	0	0	0	0	0
+minD	1	1	1	1	1	1	1	1	1	0	0	0	0	0	0
+maxD	1	1	1	1	1	1	1	1	1	1	0	0	0	0	0
+minN	1	1	1	1	1	1	1	1	1	1	1	0	0	0	0
+1	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0
+maxN	1	1	1	1	1	1	1	1	1	1	1	1	1	0	0
+INF	1	1	1	1	1	1	1	1	1	1	1	1	1	1	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16760836	100.00%	0.00e+00	0.00e+00	1
TOTAL	16760836	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	166.5	-0.33x	-0.35x	2074.6	-0.50x	4.46x
ev/clk/SM	0.55	-0.33x	-0.35x	7.13	-0.50x	4.46x
clk/ev	282.4	-0.39x	-0.36x	39.2	-0.66x	2.58x

SASS Instructions:

Class	Count
fp32	0
fp64	3
other	2
total	5

Special Values Table:

a\b	-INF	-maxN	-1	-minN	-maxD	-minD	-0	+0	+minD	+maxD	+minN	+1	+maxN	+INF	QNAN
-INF	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0
-maxN	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0
-1	1	1	1	0	0	0	0	0	0	0	0	0	0	0	0
-minN	1	1	1	1	0	0	0	0	0	0	0	0	0	0	0
-maxD	1	1	1	1	1	0	0	0	0	0	0	0	0	0	0
-minD	1	1	1	1	1	1	0	0	0	0	0	0	0	0	0
-0	1	1	1	1	1	1	1	1	0	0	0	0	0	0	0
+0	1	1	1	1	1	1	1	1	0	0	0	0	0	0	0
+minD	1	1	1	1	1	1	1	1	1	0	0	0	0	0	0
+maxD	1	1	1	1	1	1	1	1	1	1	0	0	0	0	0
+minN	1	1	1	1	1	1	1	1	1	1	1	0	0	0	0
+1	1	1	1	1	1	1	1	1	1	1	1	1	0	0	0
+maxN	1	1	1	1	1	1	1	1	1	1	1	1	1	0	0
+INF	1	1	1	1	1	1	1	1	1	1	1	1	1	1	0
QNAN	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0

Type Conversions

To Int32 (mp2int)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2533359616	100.00%	0.00e+00	0.00e+00	32
TOTAL	2533359616	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1544.4	-0.33x	-1.00x	1480.3	-0.35x	-1.00x
ev/clk/SM	5.14	-0.33x	-1.00x	5.09	-0.35x	-1.00x
clk/ev	71.0	-0.38x	-0.99x	69.2	-0.38x	1.00x

SASS Instructions:

Class	Count
fp32	2
fp64	0
other	5
total	7

Special Values Table:

Input	Value
-INF	-inf
-LMAX	-9.2e+18
-IMAX	-2.1e+09
-1	-1
-0	-0
+0	+0
+1	1
+IMAX	2.1e+09
+UMAX	4.3e+09
+LMAX	9.2e+18
+INF	+inf
QNAN	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	8577024	100.00%	0.00e+00	0.00e+00	32
TOTAL	8577024	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	104.4	-0.18x	-1.00x	1415.4	-0.36x	-1.00x
ev/clk/SM	0.35	-0.18x	-1.00x	4.87	-0.36x	-1.00x
clk/ev	377.8	-0.28x	1.00x	75.8	-0.43x	-1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	5
other	3
total	8

Special Values Table:

Input	Value
-INF	-inf
-LMAX	-9.2e+18
-IMAX	-2.1e+09
-1	-1
-0	-0
+0	+0
+1	1
+IMAX	2.1e+09
+UMAX	4.3e+09
+LMAX	9.2e+18
+INF	+inf
QNAN	nan

---

To UInt32 (mp2uint)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	1266679810	100.00%	0.00e+00	0.00e+00	32
TOTAL	1266679810	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1544.3	-0.34x	-1.00x	1480.7	-0.35x	-1.00x
ev/clk/SM	5.13	-0.34x	-1.00x	5.09	-0.35x	-1.00x
clk/ev	71.0	-0.38x	1.00x	69.3	-0.38x	1.00x

SASS Instructions:

Class	Count
fp32	2
fp64	0
other	5
total	7

Special Values Table:

Input	Value
-INF	-inf
-LMAX	-9.2e+18
-IMAX	-2.1e+09
-1	-1
-0	-0
+0	+0
+1	1
+IMAX	2.1e+09
+UMAX	4.3e+09
+LMAX	9.2e+18
+INF	+inf
QNAN	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4288512	100.00%	0.00e+00	0.00e+00	32
TOTAL	4288512	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	104.4	-0.18x	-1.00x	1414.6	-0.36x	-1.00x
ev/clk/SM	0.35	-0.18x	-1.00x	4.86	-0.36x	-1.00x
clk/ev	378.0	-0.28x	-1.00x	75.4	-0.43x	1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	5
other	3
total	8

Special Values Table:

Input	Value
-INF	-inf
-LMAX	-9.2e+18
-IMAX	-2.1e+09
-1	-1
-0	-0
+0	+0
+1	1
+IMAX	2.1e+09
+UMAX	4.3e+09
+LMAX	9.2e+18
+INF	+inf
QNAN	nan

---

To Int64 (mp2ll)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	2533359616	100.00%	0.00e+00	0.00e+00	64
TOTAL	2533359616	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	176.2	-0.30x	1.00x	1497.6	-0.35x	-1.00x
ev/clk/SM	0.59	-0.30x	1.00x	5.15	-0.35x	-1.00x
clk/ev	262.5	-0.40x	-1.00x	60.5	-0.43x	-1.00x

SASS Instructions:

Class	Count
fp32	2
fp64	0
other	7
total	9

Special Values Table:

Input	Value
-INF	-inf
-LMAX	-9.2e+18
-IMAX	-2.1e+09
-1	-1
-0	-0
+0	+0
+1	1
+IMAX	2.1e+09
+UMAX	4.3e+09
+LMAX	9.2e+18
+INF	+inf
QNAN	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	8577024	100.00%	0.00e+00	0.00e+00	64
TOTAL	8577024	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	103.5	-0.17x	-1.00x	1417.6	-0.35x	-1.00x
ev/clk/SM	0.34	-0.17x	-1.00x	4.87	-0.35x	-1.00x
clk/ev	391.0	-0.27x	-1.00x	72.4	-0.39x	-1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	5
other	4
total	9

Special Values Table:

Input	Value
-INF	-inf
-LMAX	-9.2e+18
-IMAX	-2.1e+09
-1	-1
-0	-0
+0	+0
+1	1
+IMAX	2.1e+09
+UMAX	4.3e+09
+LMAX	9.2e+18
+INF	+inf
QNAN	nan

---

To UInt64 (mp2ull)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	1266679810	100.00%	0.00e+00	0.00e+00	64
TOTAL	1266679810	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	176.2	-0.30x	1.00x	1496.1	-0.35x	-1.00x
ev/clk/SM	0.59	-0.30x	1.00x	5.14	-0.35x	-1.00x
clk/ev	262.0	-0.41x	1.00x	60.5	-0.43x	-0.99x

SASS Instructions:

Class	Count
fp32	2
fp64	0
other	7
total	9

Special Values Table:

Input	Value
-INF	-inf
-LMAX	-9.2e+18
-IMAX	-2.1e+09
-1	-1
-0	-0
+0	+0
+1	1
+IMAX	2.1e+09
+UMAX	4.3e+09
+LMAX	9.2e+18
+INF	+inf
QNAN	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4288512	100.00%	0.00e+00	0.00e+00	64
TOTAL	4288512	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	103.5	-0.17x	-1.00x	1416.7	-0.35x	-1.00x
ev/clk/SM	0.34	-0.17x	-1.00x	4.87	-0.35x	-1.00x
clk/ev	391.1	-0.27x	-1.00x	72.4	-0.39x	1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	5
other	4
total	9

Special Values Table:

Input	Value
-INF	-inf
-LMAX	-9.2e+18
-IMAX	-2.1e+09
-1	-1
-0	-0
+0	+0
+1	1
+IMAX	2.1e+09
+UMAX	4.3e+09
+LMAX	9.2e+18
+INF	+inf
QNAN	nan

---

From Int32 (int2mp)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4294967295	100.00%	0.00e+00	0.00e+00	48
TOTAL	4294967295	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	9120.5	-0.99x	-0.99x	8816.2	-0.99x	-0.99x
ev/clk/SM	30.33	-0.99x	-0.99x	30.31	-0.99x	-0.99x
clk/ev	11.0	1.05x	-0.99x	11.3	1.00x	1.01x

SASS Instructions:

Class	Count
fp32	2
fp64	0
other	3
total	5

Special Values Table:

Input	Value
0	0
+1	1
-1	-1
+2	2
-2	-2
+MAX	2147483647
-MAX	-2147483648
+Mx-1	2147483646
-Mx+1	-2147483647
+half	1073741823
-half	-1073741824
+100	100
-100	-100
+1M	1000000
-1M	-1000000
~MAX	2147483632

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16777216	100.00%	0.00e+00	0.00e+00	106
TOTAL	16777216	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	593.0	3.31x	-1.00x	4242.8	2.93x	1.00x
ev/clk/SM	1.97	3.31x	-1.00x	14.59	2.93x	1.00x
clk/ev	105.5	2.27x	1.00x	26.2	2.48x	1.01x

SASS Instructions:

Class	Count
fp32	0
fp64	1
other	1
total	2

Special Values Table:

Input	Value
0	0
+1	1
-1	-1
+2	2
-2	-2
+MAX	2147483647
-MAX	-2147483648
+Mx-1	2147483646
-Mx+1	-2147483647
+half	1073741823
-half	-1073741824
+100	100
-100	-100
+1M	1000000
-1M	-1000000
~MAX	2147483632

---

From UInt32 (uint2mp)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4294967295	100.00%	0.00e+00	0.00e+00	48
TOTAL	4294967295	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	9115.0	-0.99x	-0.99x	8844.1	-1.00x	-1.00x
ev/clk/SM	30.31	-0.99x	-0.99x	30.41	-1.00x	-1.00x
clk/ev	10.9	1.02x	1.01x	11.2	1.03x	1.01x

SASS Instructions:

Class	Count
fp32	2
fp64	0
other	3
total	5

Special Values Table:

Input	Value
0	0
1	1
2	2
MAX	4294967295
Mx-1	4294967294
half	2147483647
hlf+1	2147483648
100	100
1K	1000
1M	1000000
1B	1000000000
MSB	2147483648
~MSB	2147483647
hi16	4294901760
lo16	65535
0xAA	2863311530

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16777216	100.00%	0.00e+00	0.00e+00	106
TOTAL	16777216	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	593.1	3.31x	-1.00x	4234.9	2.92x	-1.00x
ev/clk/SM	1.97	3.31x	-1.00x	14.56	2.92x	-1.00x
clk/ev	105.6	2.26x	1.01x	26.5	2.44x	1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	1
other	1
total	2

Special Values Table:

Input	Value
0	0
1	1
2	2
MAX	4294967295
Mx-1	4294967294
half	2147483647
hlf+1	2147483648
100	100
1K	1000
1M	1000000
1B	1000000000
MSB	2147483648
~MSB	2147483647
hi16	4294901760
lo16	65535
0xAA	2863311530

---

From Int64 (ll2mp)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4294967296	100.00%	0.00e+00	0.00e+00	48
TOTAL	4294967296	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	184.6	-0.31x	1.00x	1370.5	-0.38x	-1.00x
ev/clk/SM	0.61	-0.31x	1.00x	4.71	-0.38x	-1.00x
clk/ev	294.5	-0.35x	1.00x	76.6	-0.49x	-1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	0
other	6
total	6

Special Values Table:

Input	Value
0	0
+1	1
-1	-1
+2	2
-2	-2
+MAX	9223372036854775807
-MAX	-9223372036854775808
+Mx-1	9223372036854775806
-Mx+1	-9223372036854775806
+half	4611686018427387903
-half	-4611686018427387904
+100	100
-100	-100
+1T	1000000000000
-1T	-1000000000000
~MAX	9223372036854775792

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16777216	100.00%	0.00e+00	0.00e+00	106
TOTAL	16777216	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	591.5	6.60x	-1.00x	3990.5	4.68x	-0.99x
ev/clk/SM	1.97	6.60x	-1.00x	13.72	4.68x	-0.99x
clk/ev	107.5	4.34x	1.00x	28.8	3.72x	-0.99x

SASS Instructions:

Class	Count
fp32	0
fp64	3
other	4
total	7

Special Values Table:

Input	Value
0	0
+1	1
-1	-1
+2	2
-2	-2
+MAX	9223372036854775807
-MAX	-9223372036854775808
+Mx-1	9223372036854775806
-Mx+1	-9223372036854775806
+half	4611686018427387903
-half	-4611686018427387904
+100	100
-100	-100
+1T	1000000000000
-1T	-1000000000000
~MAX	9223372036854775792

---

From UInt64 (ull2mp)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4294967296	100.00%	0.00e+00	0.00e+00	48
TOTAL	4294967296	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	184.7	-0.31x	-1.00x	1371.0	-0.38x	-1.00x
ev/clk/SM	0.61	-0.31x	-1.00x	4.71	-0.38x	-1.00x
clk/ev	294.6	-0.35x	-1.00x	76.8	-0.49x	-1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	0
other	6
total	6

Special Values Table:

Input	Value
0	0
1	1
2	2
MAX	18446744073709551615
Mx-1	18446744073709551614
half	9223372036854775807
hlf+1	9223372036854775808
100	100
1T	1000000000000
MSB	9223372036854775808
~MSB	9223372036854775807
hi32	18446744069414584320
lo32	4294967295
0xAA	12297829382473034410
0x55	6148914691236517205
~MAX	18446744073709551360

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16777216	100.00%	0.00e+00	0.00e+00	106
TOTAL	16777216	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	591.4	6.60x	-1.00x	3992.9	4.69x	-0.99x
ev/clk/SM	1.97	6.60x	-1.00x	13.73	4.69x	-0.99x
clk/ev	107.2	4.36x	1.00x	28.5	3.77x	1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	3
other	4
total	7

Special Values Table:

Input	Value
0	0
1	1
2	2
MAX	18446744073709551615
Mx-1	18446744073709551614
half	9223372036854775807
hlf+1	9223372036854775808
100	100
1T	1000000000000
MSB	9223372036854775808
~MSB	9223372036854775807
hi32	18446744069414584320
lo32	4294967295
0xAA	12297829382473034410
0x55	6148914691236517205
~MAX	18446744073709551360

---

To Native Float (mp2fp)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	4278190075	100.00%	0.00e+00	0.00e+00	53
input special	3	7e-08%	–	–	–
TOTAL	4278190078	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	734.1	1.24x	4.09x	714.9	-0.17x	-0.32x
ev/clk/SM	2.44	1.24x	4.09x	2.46	-0.17x	-0.32x
clk/ev	219.4	-0.48x	1.09x	152.5	-0.17x	-0.28x

SASS Instructions:

Class	Count
fp32	7
fp64	3
other	30
total	40

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	16769024	99.95%	0.00e+00	0.00e+00	113
input special	8192	0.05%	–	–	–
TOTAL	16777216	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	12.7	-0.48x	-1.00x	77.1	-0.16x	1.00x
ev/clk/SM	0.04	-0.48x	-1.00x	0.27	-0.16x	1.00x
clk/ev	3635.9	-0.44x	-1.00x	943.6	-0.26x	-1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	86
other	1081
total	1167

Special Values Table:

Input	Value
-INF	-inf
-maxN	-1.8e+308
-1	-1
-minN	-2.2e-308
-maxD	-2.2e-308
-minD	-4.9e-324
-0	-0
+0	+0
+minD	4.9e-324
+maxD	2.2e-308
+minN	2.2e-308
+1	1
+maxN	1.8e+308
+INF	+inf
QNAN	nan

---

From Native Float (fp2mp)

Type: fp32mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	532651806	22.09%	1.00e-13	1.60e-17	43
output special	1879048192	77.91%	–	–	–
unclassified	24802	1e-03%	2.97e-09	1.13e-12	28
TOTAL	2411724800	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp32	vs fp64	B200	vs fp32	vs fp64
GFLOPS	1290.1	2.18x	9.67x	1247.5	-0.32x	-0.90x
ev/clk/SM	4.29	2.18x	9.67x	4.29	-0.32x	-0.90x
clk/ev	86.8	1.22x	3.78x	85.4	-0.38x	-0.88x

SASS Instructions:

Class	Count
fp32	6
fp64	1
other	18
total	25

Special Values Table:

Input	Value
-INF	-inf
-maxN	-3.4e+38
-1	-1
-minN	-1.2e-38
-maxD	-1.2e-38
-minD	-1.4e-45
-0	-0
+0	+0
+minD	1.4e-45
+maxD	1.2e-38
+minN	1.2e-38
+1	1
+maxN	3.4e+38
+INF	+inf
QNAN	nan

Type: fp64mp2

Accuracy: def (mid)

Measured Accuracy:

Class	Count	Percent	Max RelErr	Avg RelErr	Bits
normal (OK)	1047552	11.75%	0.00e+00	0.00e+00	106
input special	7864320	88.24%	–	–	–
output denormal	255	3e-03%	0.00e+00	0.00e+00	0
TOTAL	8912127	100.00%

Measured Performance:

Metric	B300 SXM6 AC	vs fp64	vs fp128	B200	vs fp64	vs fp128
GFLOPS	9.5	-0.31x	1.00x	66.7	-0.24x	1.00x
ev/clk/SM	0.03	-0.31x	1.00x	0.23	-0.24x	1.00x
clk/ev	4219.9	-0.29x	-1.00x	895.4	-0.25x	1.00x

SASS Instructions:

Class	Count
fp32	0
fp64	102
other	1177
total	1279

Special Values Table:

Input	Value
-INF	-inf
-maxN	-1.8e+308
-1	-1
-minN	-2.2e-308
-maxD	-2.2e-308
-minD	-4.9e-324
-0	-0
+0	+0
+minD	4.9e-324
+maxD	2.2e-308
+minN	2.2e-308
+1	1
+maxN	1.8e+308
+INF	+inf
QNAN	nan

---

Appendix: Legends

Measured Accuracy Legend

Pattern Dataset

Accuracy measurements use the pattern dataset, which provides exhaustive bit-pattern coverage across the IEEE-754 floating-point representation space. The dataset is designed to systematically test:

• Sign bit: Both positive and negative values

• Exponent field: Full range from denormals through maximum normal values

• Mantissa bits: Both most significant and least significant bits

For multi-precision types (fp32mp2, fp64mp2), both the high and low components are generated with coordinated exponents to ensure proper normalization (|lo| < ulp(hi)/2).

Sample distribution:

Region	Coverage
Denormal inputs	Systematically tested via exponent field patterns
Near-denormal (smallest normal binade)	Covered by exponent boundary patterns
Normal range	Dense coverage with sign/exponent/mantissa bit variations
Near-infinity (largest normal binade)	Covered by exponent boundary patterns
Special values (INF, NaN)	Tested separately in the Special Values Table

For binary functions, bits are divided equally between arguments. For example, with 32-bit rigor and a binary function, each argument receives 16 bits of pattern control, yielding ~4 billion test combinations (2^32 total samples).

Note: For unary functions with a single 32-bit argument (e.g., ``int2mp``, ``uint2mp``), the pattern dataset provides exhaustive coverage of all 2^32 possible input values.

Classification Categories

The accuracy classification table shows only deviations from expected behavior: cases that exceed the warning threshold or have special value mismatches. If only normal (OK) is shown, all test cases passed within acceptable accuracy bounds.

Category	Description
normal (OK)	Result within warning threshold (acceptable accuracy)
output special	Output special value mismatch (INF or NaN)
input special	Special input (INF or NaN) causes result mismatch
output denormal	Output is denormal, accuracy loss
input denormal	Denormal input causes accuracy loss
output near denormal	Output is in smallest normal binade and exceeds warning threshold
input near denormal	Input is in smallest normal binade and result exceeds warning threshold
output near inf	Output is in largest normal binade and exceeds warning threshold
input near inf	Input is in largest normal binade and result exceeds warning threshold
cancellation	Cancellation case: result is much smaller than inputs
unclassified	Warning without identified cause (between warning and error threshold)
error (FAIL)	Exceeds error threshold without any mitigating factor

Table columns:

Column	Description
Count	Number of test cases in this category
Percent	Percentage of total test cases
Max RelErr	Maximum relative error observed in this category
Avg RelErr	Average relative error in this category
Bits	Minimum correct mantissa bits (derived from max relative error)

Special Values Legend (Floating Point)

Symbol	Description
-INF	Negative infinity
+INF	Positive infinity
-maxN	Negative maximum normal (largest finite negative)
+maxN	Positive maximum normal (largest finite positive)
-minN	Negative minimum normal (smallest normal negative)
+minN	Positive minimum normal (smallest normal positive)
-maxD	Negative maximum denormal
+maxD	Positive maximum denormal
-minD	Negative minimum denormal (smallest representable negative)
+minD	Positive minimum denormal (smallest representable positive)
-0	Negative zero
+0	Positive zero
-1	Negative one
+1	Positive one
QNAN	Quiet NaN (Not a Number)
nan	Result is NaN

Special Values Legend (Integer Conversions)

Symbol	Description
-INF	Negative infinity (saturates to min integer)
+INF	Positive infinity (saturates to max integer)
-LONG	Minimum 64-bit signed integer (-2^63)
+LONG	Maximum 64-bit signed integer (2^63-1)
-INT	Minimum 32-bit signed integer (-2^31)
+INT	Maximum 32-bit signed integer (2^31-1)
+UINT	Maximum 32-bit unsigned integer (2^32-1)
+ULONG	Maximum 64-bit unsigned integer (2^64-1)
QNAN	Quiet NaN (converts to 0)

Performance Metrics Legend

Metric	Type	Description
GFLOPS	Throughput	Giga floating-point operations per second
ev/clk/SM	Throughput	Evaluations per clock cycle per SM
clk/ev	Latency	Clock cycles per evaluation
vs fp32	Ratio	Performance ratio compared to native fp32 operations
vs fp64	Ratio	Performance ratio compared to native fp64 operations (for fp32mp2) or reference fp64 (for fp64mp2)
vs fp128	Ratio	Performance ratio compared to reference fp128 operations (for fp64mp2)

Note: Negative ratios indicate slower performance than the baseline.

SASS Instructions Legend

Counts come from the generated SASS (cuobjdump --dump-sass). Only the primary <op>_device_impl function body, up to (but not including) its RET epilogue, is counted; NOP scheduling slots, trailing self-loop BRA padding, and any helper functions emitted in the same dump (libdevice slowpaths, outlined denormal handlers, etc.) are excluded — their cost is included only when they’re actually called.

Class	Mnemonics	Notes
fp32	FADD, FMUL, FFMA, FSET, FSETP, FMNMX, FCMP, FCHK, FRND, FSWZADD, MUFU.{RCP,SQRT,RSQ,SIN,COS,EX2,LG2}, plus conversion family (F2F, F2I, I2F, I2FP, F2FP, F2IP, F2DP, D2FP) when the precision suffix is F32 (e.g. F2I.F32.TRUNC, I2FP.F32.S32)	Single-precision IEEE-754 ops. MUFU is the Multi-Function Unit (transcendentals & reciprocals). FSEL is not in this class — despite the F prefix it’s a predicated 32-bit register MOV with no FP semantics, freely emitted by ptxas in pure bit-manipulation code, so it falls into other.
fp64	DADD, DMUL, DFMA, DSET, DSETP, DMNMX, DCMP, F2D, D2F, I2D, UI2D, D2I, D2UI, MUFU.*64H (e.g. MUFU.RCP64H, MUFU.RSQ64H — high-half helpers of fp64 transcendental sequences), plus any conversion-family op with F64 in the suffix (e.g. F2I.F64.TRUNC, I2F.F64.S32, F2F.F64.F32, I2FP.F64.S32)	Double-precision IEEE-754 ops. F32↔F64 conversions are counted as fp64 since they exercise the fp64 datapath.
other	everything else (integer ALU, memory, control flow, predicate/uniform, fp16 H*2, FSEL, etc.)	Scheduling slots (NOP) are excluded entirely.

Note: SASS counts are taken from the accuracy run directory (the ``–acc`` argument) and reflect the GPU architecture that directory was built for. Different GPUs may compile to different instruction counts; the spec shows one representative count per (function, type, method).

SASS Instructions Summary

Per-function instruction counts (NOPs excluded) for every (type, method) combination that exposes a dedicated implementation. Empty cells (—) mean either:

• the combination wasn’t built / tested, or

• the combination is a wrapper over higher-precision system math — specifically the fp64mp2 path of exp/log/pow/sin/cos/tanh/cbrt/rcbrt/ erf/erfc/boys_f0/normcdfinv/floor/ceil/ round/trunc, which falls back to system fp64 / reference fp128 math and drags in arbitrary libm sub-routines whose count isn’t comparable to the dedicated implementations (see Mathematical Functions).

Function	fp32mp2									fp64mp2
	low			def			high			low			def			high
	fp32	fp64	other	fp32	fp64	other	fp32	fp64	other	fp32	fp64	other	fp32	fp64	other	fp32	fp64	other
add	8	0	1	11	0	0	20	0	0	0	8	4	0	11	0	0	20	0
sub	8	0	1	11	0	0	20	0	0	0	8	4	0	11	0	0	20	0
mul	5	0	1	9	0	0	9	0	0	0	5	2	0	9	0	0	9	0
div	6	0	1	13	0	0	23	0	23	1	11	27	1	18	25	1	28	49
acc	7	0	1	10	0	0	13	0	0	0	7	2	0	10	0	0	13	0
fma	16	0	1	19	0	0	37	0	0	0	16	2	0	19	0	0	37	0
mad	13	0	0	16	0	0	29	0	0	0	13	0	0	16	0	0	29	0
sqrt	—	—	—	18	0	1	—	—	—	—	—	—	0	23	25	—	—	—
rsqrt	—	—	—	17	0	0	—	—	—	—	—	—	0	22	23	—	—	—
cbrt	—	—	—	75	0	27	—	—	—	—	—	—	—	—	—	—	—	—
rcbrt	—	—	—	95	0	35	—	—	—	—	—	—	—	—	—	—	—	—
pow	—	—	—	305	0	69	—	—	—	—	—	—	—	—	—	—	—	—
exp	—	—	—	136	0	14	—	—	—	—	—	—	—	—	—	—	—	—
exp2	—	—	—	126	0	21	—	—	—	—	—	—	—	—	—	—	—	—
exp10	—	—	—	198	0	21	—	—	—	—	—	—	—	—	—	—	—	—
expm1	—	—	—	285	0	28	—	—	—	—	—	—	—	—	—	—	—	—
log	—	—	—	135	0	20	—	—	—	—	—	—	—	—	—	—	—	—
log2	—	—	—	145	0	21	—	—	—	—	—	—	—	—	—	—	—	—
log10	—	—	—	145	0	21	—	—	—	—	—	—	—	—	—	—	—	—
log1p	—	—	—	288	0	52	—	—	—	—	—	—	—	—	—	—	—	—
sin	—	—	—	226	0	227	—	—	—	—	—	—	—	—	—	—	—	—
cos	—	—	—	226	0	230	—	—	—	—	—	—	—	—	—	—	—	—
tan	—	—	—	242	0	229	—	—	—	—	—	—	—	—	—	—	—	—
sinpi	—	—	—	269	0	73	—	—	—	—	—	—	—	—	—	—	—	—
cospi	—	—	—	265	0	66	—	—	—	—	—	—	—	—	—	—	—	—
asin	—	—	—	432	0	9	—	—	—	—	—	—	—	—	—	—	—	—
acos	—	—	—	454	0	10	—	—	—	—	—	—	—	—	—	—	—	—
atan	—	—	—	264	0	6	—	—	—	—	—	—	—	—	—	—	—	—
atan2	—	—	—	299	0	46	—	—	—	—	—	—	—	—	—	—	—	—
sinh	—	—	—	288	0	23	—	—	—	—	—	—	—	—	—	—	—	—
cosh	—	—	—	160	0	18	—	—	—	—	—	—	—	—	—	—	—	—
tanh	—	—	—	310	0	28	—	—	—	—	—	—	—	—	—	—	—	—
asinh	—	—	—	521	0	88	—	—	—	—	—	—	—	—	—	—	—	—
acosh	—	—	—	517	0	85	—	—	—	—	—	—	—	—	—	—	—	—
atanh	—	—	—	463	0	68	—	—	—	—	—	—	—	—	—	—	—	—
erf	—	—	—	560	0	22	—	—	—	—	—	—	—	—	—	—	—	—
erfc	—	—	—	525	0	20	—	—	—	—	—	—	—	—	—	—	—	—
normcdfinv	—	—	—	735	0	36	—	—	—	—	—	—	—	—	—	—	—	—
boys_f0	—	—	—	749	0	11	—	—	—	—	—	—	—	—	—	—	—	—
floor	—	—	—	28	0	13	—	—	—	—	—	—	—	—	—	—	—	—
ceil	—	—	—	28	0	13	—	—	—	—	—	—	—	—	—	—	—	—
round	—	—	—	71	0	29	—	—	—	—	—	—	—	—	—	—	—	—
trunc	—	—	—	35	0	23	—	—	—	—	—	—	—	—	—	—	—	—
fmod	—	—	—	41	0	235	—	—	—	—	—	—	—	—	—	—	—	—
remainder	—	—	—	65	0	275	—	—	—	—	—	—	—	—	—	—	—	—
ldexp	—	—	—	22	9	52	—	—	—	—	—	—	—	—	—	—	—	—
scalbn	—	—	—	22	9	52	—	—	—	—	—	—	—	—	—	—	—	—
scalbln	—	—	—	22	9	52	—	—	—	—	—	—	0	14	22	—	—	—
frexp	—	—	—	13	2	25	—	—	—	—	—	—	0	11	28	—	—	—
eq	—	—	—	2	0	1	—	—	—	—	—	—	0	2	1	—	—	—
ne	—	—	—	2	0	1	—	—	—	—	—	—	0	2	1	—	—	—
lt	—	—	—	3	0	2	—	—	—	—	—	—	0	3	2	—	—	—
le	—	—	—	3	0	2	—	—	—	—	—	—	0	3	2	—	—	—
gt	—	—	—	3	0	2	—	—	—	—	—	—	0	3	2	—	—	—
ge	—	—	—	3	0	2	—	—	—	—	—	—	0	3	2	—	—	—
mp2int	—	—	—	2	0	5	—	—	—	—	—	—	0	5	3	—	—	—
mp2uint	—	—	—	2	0	5	—	—	—	—	—	—	0	5	3	—	—	—
mp2ll	—	—	—	2	0	7	—	—	—	—	—	—	0	5	4	—	—	—
mp2ull	—	—	—	2	0	7	—	—	—	—	—	—	0	5	4	—	—	—
int2mp	—	—	—	2	0	3	—	—	—	—	—	—	0	1	1	—	—	—
uint2mp	—	—	—	2	0	3	—	—	—	—	—	—	0	1	1	—	—	—
ll2mp	—	—	—	0	0	6	—	—	—	—	—	—	0	3	4	—	—	—
ull2mp	—	—	—	0	0	6	—	—	—	—	—	—	0	3	4	—	—	—
mp2fp	—	—	—	7	3	30	—	—	—	—	—	—	0	86	1081	—	—	—
fp2mp	—	—	—	6	1	18	—	—	—	—	—	—	0	102	1177	—	—	—