`(exp)` uses unexpectedly megatons of gas

[CryptoPascal31]

Issue description

When used with an argument with many decimals, (exp) uses billions of gas

Steps to reproduce

pact> (env-gasmodel "table")
"Set gas model to table-based cost model"
pact> (env-gaslimit 1000000)
"Set gas limit to 1000000"
pact> (env-gas 0)
"Set gas to 0"
pact> (exp -0.00369328591312029)
(interactive):1:0: Gas limit "1000000"exceeded: 18446744073709552
 1 | (exp -0.00369328591312029)
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^

Expected Behavior

I would expect something more reasonable.

Debug Information

This one is ok:

pact> (env-gasmodel "table")
"Set gas model to table-based cost model"
pact> (env-gaslimit 10)
"Set gas limit to 10"
pact> (env-gas 0)
"Set gas to 0"
pact> (^ (exp 1.0) -0.00369328591312029)
0.99631352587875243731474483865895308554172515869140625
pact> (env-gas)
3

Note: I have the same beahviour on-chain and in REPL

[CryptoPascal31]

@jmcardon I did some more investigation on this.

pact-5/pact/Pact/Core/Gas/TableGasModel.hs

Line 200 in db3149b

transExpCost !power = MilliGas total

In your gas computation, you assume that the exponentiation is done by squaring with 2n multiplication with growing operands... 😱 This ends up to an amazing complexity... which makes impossibles every exponentiation with more than 3 or 4 digits.

Fortunately, the MPFR used under the hood is a lot smarter ...

And according to this: https://www.mpfr.org/algorithms.pdf paragraph 4.4..
It uses an approach with a complexity of $O(n^{1/3} . M(n))$ 😄