YES(?, 28*a + 10)

Initial complexity problem:
1:	T:
		(1, 1)    evalspeedpldi2start(a, b, c) -> evalspeedpldi2entryin(a, b, c)
		(?, 1)    evalspeedpldi2entryin(a, b, c) -> evalspeedpldi2bb5in(b, 0, a) [ a >= 0 /\ b >= 1 ]
		(?, 1)    evalspeedpldi2entryin(a, b, c) -> evalspeedpldi2returnin(a, b, c) [ 0 >= a + 1 ]
		(?, 1)    evalspeedpldi2entryin(a, b, c) -> evalspeedpldi2returnin(a, b, c) [ 0 >= b ]
		(?, 1)    evalspeedpldi2bb5in(a, b, c) -> evalspeedpldi2bb2in(a, b, c) [ c >= 1 ]
		(?, 1)    evalspeedpldi2bb5in(a, b, c) -> evalspeedpldi2returnin(a, b, c) [ 0 >= c ]
		(?, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb3in(a, b, c) [ a >= b + 1 ]
		(?, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb5in(a, 0, c) [ b >= a ]
		(?, 1)    evalspeedpldi2bb3in(a, b, c) -> evalspeedpldi2bb5in(a, b + 1, c - 1)
		(?, 1)    evalspeedpldi2returnin(a, b, c) -> evalspeedpldi2stop(a, b, c)
	start location:	evalspeedpldi2start
	leaf cost:	0

Repeatedly removing leaves of the complexity graph in problem 1 produces the following problem:
2:	T:
		(1, 1)    evalspeedpldi2start(a, b, c) -> evalspeedpldi2entryin(a, b, c)
		(?, 1)    evalspeedpldi2entryin(a, b, c) -> evalspeedpldi2bb5in(b, 0, a) [ a >= 0 /\ b >= 1 ]
		(?, 1)    evalspeedpldi2bb5in(a, b, c) -> evalspeedpldi2bb2in(a, b, c) [ c >= 1 ]
		(?, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb3in(a, b, c) [ a >= b + 1 ]
		(?, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb5in(a, 0, c) [ b >= a ]
		(?, 1)    evalspeedpldi2bb3in(a, b, c) -> evalspeedpldi2bb5in(a, b + 1, c - 1)
	start location:	evalspeedpldi2start
	leaf cost:	4

Repeatedly propagating knowledge in problem 2 produces the following problem:
3:	T:
		(1, 1)    evalspeedpldi2start(a, b, c) -> evalspeedpldi2entryin(a, b, c)
		(1, 1)    evalspeedpldi2entryin(a, b, c) -> evalspeedpldi2bb5in(b, 0, a) [ a >= 0 /\ b >= 1 ]
		(?, 1)    evalspeedpldi2bb5in(a, b, c) -> evalspeedpldi2bb2in(a, b, c) [ c >= 1 ]
		(?, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb3in(a, b, c) [ a >= b + 1 ]
		(?, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb5in(a, 0, c) [ b >= a ]
		(?, 1)    evalspeedpldi2bb3in(a, b, c) -> evalspeedpldi2bb5in(a, b + 1, c - 1)
	start location:	evalspeedpldi2start
	leaf cost:	4

Applied AI with 'oct' on problem 3 to obtain the following invariants:
  For symbol evalspeedpldi2bb2in: X_3 - 1 >= 0 /\ X_2 + X_3 - 1 >= 0 /\ X_1 + X_3 - 2 >= 0 /\ X_2 >= 0 /\ X_1 + X_2 - 1 >= 0 /\ X_1 - 1 >= 0
  For symbol evalspeedpldi2bb3in: X_3 - 1 >= 0 /\ X_2 + X_3 - 1 >= 0 /\ X_1 + X_3 - 2 >= 0 /\ X_1 - X_2 - 1 >= 0 /\ X_2 >= 0 /\ X_1 + X_2 - 1 >= 0 /\ X_1 - 1 >= 0
  For symbol evalspeedpldi2bb5in: X_3 >= 0 /\ X_2 + X_3 >= 0 /\ X_1 + X_3 - 1 >= 0 /\ X_2 >= 0 /\ X_1 + X_2 - 1 >= 0 /\ X_1 - 1 >= 0


This yielded the following problem:
4:	T:
		(?, 1)    evalspeedpldi2bb3in(a, b, c) -> evalspeedpldi2bb5in(a, b + 1, c - 1) [ c - 1 >= 0 /\ b + c - 1 >= 0 /\ a + c - 2 >= 0 /\ a - b - 1 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 ]
		(?, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb5in(a, 0, c) [ c - 1 >= 0 /\ b + c - 1 >= 0 /\ a + c - 2 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 /\ b >= a ]
		(?, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb3in(a, b, c) [ c - 1 >= 0 /\ b + c - 1 >= 0 /\ a + c - 2 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 /\ a >= b + 1 ]
		(?, 1)    evalspeedpldi2bb5in(a, b, c) -> evalspeedpldi2bb2in(a, b, c) [ c >= 0 /\ b + c >= 0 /\ a + c - 1 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 /\ c >= 1 ]
		(1, 1)    evalspeedpldi2entryin(a, b, c) -> evalspeedpldi2bb5in(b, 0, a) [ a >= 0 /\ b >= 1 ]
		(1, 1)    evalspeedpldi2start(a, b, c) -> evalspeedpldi2entryin(a, b, c)
	start location:	evalspeedpldi2start
	leaf cost:	4

A polynomial rank function with
	Pol(evalspeedpldi2bb3in) = 4*V_2 + 7*V_3 - 1
	Pol(evalspeedpldi2bb5in) = 4*V_2 + 7*V_3 + 1
	Pol(evalspeedpldi2bb2in) = 4*V_2 + 7*V_3
	Pol(evalspeedpldi2entryin) = 7*V_1 + 1
	Pol(evalspeedpldi2start) = 7*V_1 + 1
orients all transitions weakly and the transitions
	evalspeedpldi2bb5in(a, b, c) -> evalspeedpldi2bb2in(a, b, c) [ c >= 0 /\ b + c >= 0 /\ a + c - 1 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 /\ c >= 1 ]
	evalspeedpldi2bb3in(a, b, c) -> evalspeedpldi2bb5in(a, b + 1, c - 1) [ c - 1 >= 0 /\ b + c - 1 >= 0 /\ a + c - 2 >= 0 /\ a - b - 1 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 ]
	evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb5in(a, 0, c) [ c - 1 >= 0 /\ b + c - 1 >= 0 /\ a + c - 2 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 /\ b >= a ]
	evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb3in(a, b, c) [ c - 1 >= 0 /\ b + c - 1 >= 0 /\ a + c - 2 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 /\ a >= b + 1 ]
strictly and produces the following problem:
5:	T:
		(7*a + 1, 1)    evalspeedpldi2bb3in(a, b, c) -> evalspeedpldi2bb5in(a, b + 1, c - 1) [ c - 1 >= 0 /\ b + c - 1 >= 0 /\ a + c - 2 >= 0 /\ a - b - 1 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 ]
		(7*a + 1, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb5in(a, 0, c) [ c - 1 >= 0 /\ b + c - 1 >= 0 /\ a + c - 2 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 /\ b >= a ]
		(7*a + 1, 1)    evalspeedpldi2bb2in(a, b, c) -> evalspeedpldi2bb3in(a, b, c) [ c - 1 >= 0 /\ b + c - 1 >= 0 /\ a + c - 2 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 /\ a >= b + 1 ]
		(7*a + 1, 1)    evalspeedpldi2bb5in(a, b, c) -> evalspeedpldi2bb2in(a, b, c) [ c >= 0 /\ b + c >= 0 /\ a + c - 1 >= 0 /\ b >= 0 /\ a + b - 1 >= 0 /\ a - 1 >= 0 /\ c >= 1 ]
		(1, 1)          evalspeedpldi2entryin(a, b, c) -> evalspeedpldi2bb5in(b, 0, a) [ a >= 0 /\ b >= 1 ]
		(1, 1)          evalspeedpldi2start(a, b, c) -> evalspeedpldi2entryin(a, b, c)
	start location:	evalspeedpldi2start
	leaf cost:	4

Complexity upper bound 28*a + 10

Time: 0.374 sec (SMT: 0.353 sec)