[TTFS] clean interior-point kernels to improve inference (#228)

- use @inbounds @simd in a systematic way before for loops
- fix small broadcast calls in kernels.jl
- rewrite broadcast operators in print_init to ease compiler job

src/IPM/callbacks.jl

@@ -13,14 +13,15 @@ function eval_grad_f_wrapper!(solver::MadNLPSolver, f::Vector{T},x::Vector{T}) w
     nlp = solver.nlp
     cnt = solver.cnt
     @trace(solver.logger,"Evaluating objective gradient.")
+    obj_scaling = solver.obj_scale[] * (get_minimize(nlp) ? one(T) : -one(T))
     x_nlpmodel = _madnlp_unsafe_wrap(x, get_nvar(nlp))
     f_nlpmodel = _madnlp_unsafe_wrap(f, get_nvar(nlp))
     cnt.eval_function_time += @elapsed grad!(
         nlp,
         x_nlpmodel,
         f_nlpmodel
     )
-    f.*=solver.obj_scale[] * (get_minimize(nlp) ? 1. : -1.)
+    _scal!(obj_scaling, f)
     cnt.obj_grad_cnt+=1
     cnt.obj_grad_cnt==1 && (is_valid(f)  || throw(InvalidNumberException(:grad)))
     return f
@@ -38,8 +39,8 @@ function eval_cons_wrapper!(solver::MadNLPSolver, c::Vector{T},x::Vector{T}) whe
         c_nlpmodel
     )
     view(c,solver.ind_ineq).-=view(x,get_nvar(nlp)+1:solver.n)
-    c.-=solver.rhs
-    c.*=solver.con_scale
+    c .-= solver.rhs
+    c .*= solver.con_scale
     cnt.con_cnt+=1
     cnt.con_cnt==2 && (is_valid(c) || throw(InvalidNumberException(:cons)))
     return c
@@ -124,3 +125,4 @@ function eval_lag_hess_wrapper!(solver::MadNLPSolver, kkt::AbstractDenseKKTSyste
     cnt.lag_hess_cnt==1 && (is_valid(hess) || throw(InvalidNumberException(:hess)))
     return hess
 end
+

src/IPM/factorization.jl

@@ -72,7 +72,7 @@ function solve_refine_wrapper!(
     xy = view(full(x), kkt.ind_eq_shifted)
     xz = view(full(x), kkt.ind_ineq_shifted)
 
-    v_c .= 0.0
+    fill!(v_c, zero(T))
     v_c[kkt.ind_ineq] .= (Σs .* bz .+ α .* bs) ./ α.^2
     jtprod!(jv_t, kkt, v_c)
     # init right-hand-side