Skip to content
GitLab
Commit 176780f1 authored by Heiko Becker's avatar Heiko Becker
Browse files

Try to prove copnstant case, fix machine epsilon. Still not working...

parent bf0162f3
Hide whitespace changes
Inline Side-by-side
coq/exps.v
View file @ 176780f1
......@@ -32,7 +32,7 @@ Inductive exp (V:Type): Type :=
Define the machine epsilon for floating point operations.
FIXME: Currently set to 1.0 instead of the concrete value!
**)
Definition machineEpsilon:R := realFromNum 59604645 7 8.
Definition machineEpsilon:R := realFromNum 2220446049250313 15 16.
(**
Define a perturbation function to ease writing of basic definitions
**)
......
coq/simple_doppler.v
View file @ 176780f1
Require Import Coq.Reals.Reals Interval.Interval_tactic.
Require Import Coq.Reals.Reals Interval.Interval_tactic Coq.micromega.Psatz.
Require Import Daisy.abs_err Daisy.daisy_lang Daisy.newIntervalArith Daisy.exps Daisy.realConversion.
(*
Daisy output:
[ Info ]
[ Info ] Starting specs preprocessing phase
[ Info ] Finished specs preprocessing phase
[ Info ]
[ Info ]
[ Info ] Starting range-error phase
[ Info ] Machine epsilon 5.9604645E-8
[ Info ] 331.4: [331.4, 331.4], 2.8421709430404007e-14
[ Info ] u: [-100.0, 100.0], 7.105427357601002e-15
[ Info ] (331.4 + u): [231.4, 431.4],6.394884621840902e-14
[ Info ] Finished range-error phase
[ Info ]
[ Info ] Starting info phase
[ Info ] doppler
[ Info ] abs-error: 6.394884621840902e-14, range: [231.4, 431.4],
[ Info ] rel-error: 2.7635629307869066E-16
[ Info ] Finished info phase
[ Info ] time:
[ Info ] info: 8 ms, rangeError: 81 ms, analysis: 14 ms, frontend: 2547 ms,
def doppler(u: Real): Real = {
require(-100.0 <= u && u <= 100)
331.4 + u
}
[ Info ]
[ Info ] Starting specs preprocessing phase
[ Info ] Finished specs preprocessing phase
[ Info ]
[ Info ]
[ Info ] Starting range-error phase
[ Info ] Machine epsilon 2.220446049250313E-16
[ Info ] 331.4: [331.4, 331.4],7.358558207215537e-14
[ Info ] u: [-100.0, 100.0],2.220446049250313e-14
[ Info ] (331.4 + u): [231.4, 431.4],1.9158008512931703e-13
[ Info ] Finished range-error phase
[ Info ]
[ Info ] Starting info phase
[ Info ] doppler
[ Info ] abs-error: 1.9158008512931703e-13, range: [231.4, 431.4],
[ Info ] rel-error: 8.279173946815775E-16
[ Info ] Finished info phase
[ Info ] time:
[ Info ] info: 7 ms, rangeError: 58 ms, analysis: 8 ms, frontend: 2167 ms,
*)
Definition u:nat := 1.
(** 1655/5 = 331; 0,4 = 2/5 **)
......@@ -41,11 +32,11 @@ Definition valCst:exp R := Const cst1.
Definition valCstAddVarU:exp R := Binop Plus valCst varU.
(** Error values **)
Definition errCst1 := realFromNum 28421709430404007 16 14.
Definition errVaru := realFromNum 7105427357601002 15 15.
Definition errCst1 := realFromNum 7358558207215537 15 14.
Definition errVaru := realFromNum 2220446049250313 15 14.
Definition lowerBoundAddUCst:R := - (1157) / 5.
Definition upperBoundAddUCst:R := 2157 / 5.
Definition errAddUCst := realFromNum 6394884621840902 15 14.
Definition errAddUCst := realFromNum 19158008512931703 16 13.
(** The added assertion becomes the precondition for us **)
Definition precondition := fun env:nat -> R => (- 100 <= env u)%R /\ (env u <= 100)%R.
......@@ -62,7 +53,7 @@ Inductive absEnv : abs_env :=
Goal forall (cenv:nat->R) (vR:R) (vF:R),
precondition cenv ->
eval_exp (0)%R cenv valCstAddVarU vR ->
eval_exp m_eps cenv valCstAddVarU vF ->
eval_exp machineEpsilon cenv valCstAddVarU vF ->
AbsErrExp valCstAddVarU absEnv errAddUCst /\ (Rabs (vR - vF) <= errAddUCst)%R.
Proof.
intros cenv vR vF precond eval_real eval_float.
......@@ -76,10 +67,35 @@ Proof.
+ unfold valCst.
apply (AbsErrConst cst1 (mkInterval cst1 cst1) errCst1); [constructor | ].
unfold isSoundErr; simpl.
admit. (** Should be easy once we have the real epsilon **)
unfold errCst1, cst1, machineEpsilon.
unfold realFromNum.
unfold negPow.
unfold Rabs.
assert (1657 / 5 - 2220446049250313 * (1 / 10 ^ 15) * (1 / 2 ^ 16) >= 0)%R by interval.
assert (1657 / 5 + 2220446049250313 * (1 / 10 ^ 15) * (1 / 2 ^ 16) >= 0)%R by interval.
destruct Rcase_abs.
* exfalso.
apply Rlt_not_le in r.
apply r.
apply Rge_le in H; auto.
* destruct Rcase_abs.
{ exfalso.
apply Rlt_not_le in r0.
apply r0.
apply Rge_le in H0; auto.
}
{
assert ((1657 / 5 - 2220446049250313 * (1 / 10 ^ 15) * (1 / 2 ^ 16)) <= 1657 / 5 + 2220446049250313 * (1 / 10 ^ 15) * (1 / 2 ^ 16))%R
by interval.
assert (Rmax (1657 / 5 - 2220446049250313 * (1 / 10 ^ 15) * (1 / 2 ^ 16))
(1657 / 5 + 2220446049250313 * (1 / 10 ^ 15) * (1 / 2 ^ 16)) = (1657 / 5 + 2220446049250313 * (1 / 10 ^ 15) * (1 / 2 ^ 16)))%R by (apply Rmax_right; auto).
rewrite H2.
field_simplify.
(** TODO: Flaw in computation. Wolfram Alpha says that this inequation is simply false **)
admit.
+ apply (AbsErrVar u (mkInterval (- 100) (100)) errVaru); [constructor | ].
unfold isSoundErr; simpl.
unfold m_eps, errVaru.
unfold machineEpsilon, errVaru.
unfold realFromNum.
unfold negPow.
simpl.
......@@ -131,4 +147,4 @@ Proof.
as floatAddErrContained
by (apply additionIsValid; auto).
admit.
Admitted.
\ No newline at end of file
Admitted.
src/main/scala/daisy/analysis/RangeErrorPhase.scala
View file @ 176780f1
......@@ -40,7 +40,9 @@ object RangeErrorPhase extends DaisyPhase with ErrorFunctions {
override def run(ctx: Context, prg: Program): (Context, Program) = {
reporter = ctx.reporter
reporter.info(s"\nStarting $name")
reporter.info(s"Machine epsilon ${math.ulp(1.0f)/2}")
val eps = java.lang.Math.ulp(1.0)
reporter.info(s"Machine epsilon $eps")
val timer = ctx.timers.rangeError.start
......
src/main/scala/daisy/utils/FinitePrecision.scala
View file @ 176780f1
......@@ -42,7 +42,8 @@ object FinitePrecision {
}
def absRoundoff(r: Rational): Rational = {
// PERFORMANCE: this may not be the fastest way
Rational.fromDouble(math.ulp(Rational.abs(r).floatValue)/2)
Rational.fromDouble(math.ulp(1.0))*Rational.abs(r)
//Rational.fromDouble(math.ulp(Rational.abs(r).floatValue)/2)
}
}
......@@ -58,7 +59,8 @@ object FinitePrecision {
}
def absRoundoff(r: Rational): Rational = {
Rational.fromDouble(math.ulp(Rational.abs(r).doubleValue)/2)
Rational.fromDouble(math.ulp(1.0))*Rational.abs(r)
//Rational.fromDouble(math.ulp(Rational.abs(r).doubleValue)/2)
}
}
......@@ -103,4 +105,4 @@ object FinitePrecision {
val minNormal: Rational = zero // dummy
}*/
}
\ No newline at end of file
}
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment