The following figure shows the baseline latency model for the crossbar module.
Figure 1. Crossbar Model Baseline Latency
The preceding figure shows the baseline latency:
- T_AW = T_AR = 2 cycles of
aclk, for the forward propagation ofaw/arvalid, provided there are no pending conditions that would inhibit granting arbitration (such as a higher-priority request). Each arbitration also causes 2 bubble cycles, resulting in 3 cycles (minimum) between successive arbitrations by the same SI slot. - T_WC = 1 cycle of
aclk. - T_W = 1 cycle of
aclk, with no bubble cycles (supports continuous back-to-back data transmission). - T_R = 1 or 2 cycles of
aclk, with no bubble cycles (supports continuous back-to-back data transmission). The second latency cycle occurs when there is a re-arbitration (when the requesting MI slot is different than the last granted MI slot) following an idle cycle. While the same MI slot propagates back-to-back data, or while multiple MI slots continuously interleave data, latency through the R-channel arbiter is 1 cycle. - T_B (B-channel latency, not shown) = 1 or 2
cycles of
aclk. (Same as for T_R.)