Understanding the formation and evolution of giant planets (>= 1 M-Jup) at wide orbital separation (>= 5 AU) is one of the goals of direct imaging. Over the past 15 yr, many surveys have placed strong constraints on the occurrence rate of wide-orbit giants, mostly based on non-detections, but very few have tried to make a direct link with planet formation theories. In the present work, we combine the results of our previously published VLT/NaCo large program with the results of 12 past imaging surveys to constitute a statistical sample of 199 FGK stars within 100 pc, including three stars with sub-stellar companions. Using Monte Carlo simulations and assuming linear flat distributions for the mass and semi-major axis of planets, we estimate the sub-stellar companion frequency to be within 0.75-5.70% at the 68% confidence level (CL) within 20-300 AU and 0.5-75 M-Jup, which is compatible with previously published results. We also compare our results with the predictions of state-of-the-art population synthesis models based on the gravitational instability (GI) formation scenario with and without scattering. We estimate that in both the scattered and non-scattered populations, we would be able to detect more than 30% of companions in the 1-75 M-Jup range (95% CL). With the three sub-stellar detections in our sample, we estimate the fraction of stars that host a planetary system formed by GI to be within 1.0-8.6% (95% CL). We also conclude that even though GI is not common, it predicts a mass distribution of wide-orbit massive companions that is much closer to what is observed than what the core accretion scenario predicts. Finally, we associate the present paper with the release of the Direct Imaging Virtual Archive (DIVA), a public database that aims at gathering the results of past, present, and future direct imaging surveys.