From: kub <derkub@gmail.com>
Date: Mon, 3 Oct 2022 08:41:37 +0000 (+0000)
Subject: core, add TH pad pin latency (regression after 5 min)
X-Git-Tag: v2.00~286
X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=e05680a2b9bd5e079aa29a0c39c03cc9b90c5911;p=picodrive.git

core, add TH pad pin latency (regression after 5 min)
---

diff --git a/pico/memory.c b/pico/memory.c
index 860dfed1..2ca6f2c2 100644
--- a/pico/memory.c
+++ b/pico/memory.c
@@ -328,7 +328,11 @@ static NOINLINE u32 port_read(int i)
   // disables output before doing TH-low read, so don't emulate it for TH.
   // Decap Attack reportedly doesn't work on Nomad but works on must
   // other MD revisions (different pull-up strength?).
-  u32 mask = 0x3f | (padTHLatency[i] < SekCyclesDone() ? 0x40 : 0);
+  u32 mask = 0x3f;
+  if (CYCLES_GE(SekCyclesDone(), padTHLatency[i])) {
+    mask |= 0x40;
+    padTHLatency[i] = SekCyclesDone();
+  }
   out |= mask & ~ctrl_reg;
 
   in = port_readers[i](i, out);
@@ -407,10 +411,13 @@ NOINLINE void io_ports_write(u32 a, u32 d)
       Pico.m.padTHPhase[a - 1]++;
   }
 
-  // after pulling down TH there's a latency before the new value can be read
-  // back as input (see Decap Attack, not in Samurai Showdown, 32x WWF Raw)
-  if (a >= 4 && a <= 5 && !(d & 0x40))
-    padTHLatency[a - 4] = SekCyclesDone() + 25;
+  // after switching TH to input there's a latency before the pullup value is 
+  // read back as input (see Decap Attack, not in Samurai Showdown, 32x WWF Raw)
+  if (a >= 4 && a <= 5) {
+    padTHLatency[a - 4] = SekCyclesDone(); // if output, effective immediately
+    if ((PicoMem.ioports[a] & 0x40) && !(d & 0x40))
+      padTHLatency[a - 4] += 25; // latency after switching to input
+  }
 
   // certain IO ports can be used as RAM
   PicoMem.ioports[a] = d;