X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=plugins%2Fdfsound%2Fspu_c64x_dspcode.c;h=570da5eda44921d4ac9fbe93708e29f655619190;hp=97d3028079129f0b2e35fbe7a7e6deb0f2be93d0;hb=16f3ca666fb090dcb9ac0b399b767e4ed0aabece;hpb=3bd31caf9e9f5ddab2bf4fbdb5a129f4972c45f3

diff --git a/plugins/dfsound/spu_c64x_dspcode.c b/plugins/dfsound/spu_c64x_dspcode.c
index 97d30280..570da5ed 100644
--- a/plugins/dfsound/spu_c64x_dspcode.c
+++ b/plugins/dfsound/spu_c64x_dspcode.c
@@ -31,31 +31,62 @@
 /* dummy deps, some bloat but avoids ifdef hell in SPU code.. */
 static void thread_work_start(void) {}
 static void thread_work_wait_sync(struct work_item *work, int force) {}
+static void thread_sync_caches(void) {}
 static int  thread_get_i_done(void) { return 0; }
 struct out_driver *out_current;
 void SetupSound(void) {}
 
 
+static void enable_l2_cache(void)
+{
+ volatile uint32_t *L2CFG = (volatile uint32_t *)0x01840000;
+ uint32_t *MARi = (void *)0x01848000;
+ int i;
+
+ // program Memory Attribute Registers
+ // (old c64_tools has the defaults messed up)
+ // 00000000-0fffffff - not configurable
+ // 10000000-7fffffff - system
+ for (i = 0x10; i < 0x80; i++)
+  MARi[i] = 0;
+ // 80000000-9fffffff - RAM
+ for ( ; i < 0xa0; i++)
+  MARi[i] = 1;
+ // 0xa00000-ffffffff - reserved, etc
+ for ( ; i < 0x100; i++)
+  MARi[i] = 0;
+
+ // enable L2 (1 for 32k, 2 for 64k)
+ if (!(*L2CFG & 2)) {
+  *L2CFG = 2;
+  // wait the for the write
+  *L2CFG;
+ }
+}
+
 static void invalidate_cache(struct work_item *work)
 {
- syscalls.cache_inv(work, offsetof(typeof(*work), RVB), 1);
- syscalls.cache_inv(spu.s_chan, sizeof(spu.s_chan[0]) * 24, 0);
+ // see comment in writeout_cache()
+ //syscalls.cache_inv(work, offsetof(typeof(*work), SSumLR), 1);
+ syscalls.cache_inv(spu.s_chan, sizeof(spu.s_chan[0]) * 24, 1);
  syscalls.cache_inv(work->SSumLR,
-   sizeof(work->SSumLR[0]) * 2 * work->ns_to, 0);
+   sizeof(work->SSumLR[0]) * 2 * work->ns_to, 1);
 }
 
 static void writeout_cache(struct work_item *work)
 {
  int ns_to = work->ns_to;
 
- syscalls.cache_wb(work->RVB, sizeof(work->RVB[0]) * 2 * ns_to, 1);
  syscalls.cache_wb(work->SSumLR, sizeof(work->SSumLR[0]) * 2 * ns_to, 1);
+ // have to invalidate now, otherwise there is a race between
+ // DSP evicting dirty lines and ARM writing new data to this area
+ syscalls.cache_inv(work, offsetof(typeof(*work), SSumLR), 1);
 }
 
 static void do_processing(void)
 {
+ int left, dirty = 0, had_rvb = 0;
  struct work_item *work;
- int left, dirty = 0;
 
  while (worker->active)
  {
@@ -70,6 +101,8 @@ static void do_processing(void)
 
    work = &worker->i[worker->i_done & WORK_I_MASK];
    invalidate_cache(work);
+   had_rvb |= work->rvb_addr;
+   spu.spuCtrl = work->ctrl;
    do_channel_work(work);
    writeout_cache(work);
 
@@ -82,6 +115,11 @@ static void do_processing(void)
   if (dirty) {
    syscalls.cache_wb(spu.spuMemC + 0x800, 0x800, 1);
    syscalls.cache_wb(spu.SB, sizeof(spu.SB[0]) * SB_SIZE * 24, 1);
+   if (had_rvb) {
+    left = 0x40000 - spu.rvb->StartAddr;
+    syscalls.cache_wb(spu.spuMem + spu.rvb->StartAddr, left * 2, 1);
+    had_rvb = 0;
+   }
    dirty = 0;
    continue;
   }
@@ -101,16 +139,18 @@ static unsigned int exec(dsp_component_cmd_t cmd,
 
  switch (cmd) {
   case CCMD_INIT:
+   enable_l2_cache();
    InitADSR();
 
    spu.spuMemC = mem->spu_ram;
    spu.SB = mem->SB;
-   spu.s_chan = mem->s_chan;
+   spu.s_chan = mem->in.s_chan;
+   spu.rvb = &mem->in.rvb;
    worker = &mem->worker;
-   memcpy(&spu_config, &mem->spu_config, sizeof(spu_config));
+   memcpy(&spu_config, &mem->in.spu_config, sizeof(spu_config));
 
    mem->sizeof_region_mem = sizeof(*mem);
-   mem->offsetof_s_chan1 = offsetof(typeof(*mem), s_chan[1]);
+   mem->offsetof_s_chan1 = offsetof(typeof(*mem), in.s_chan[1]);
    mem->offsetof_spos_3_20 = offsetof(typeof(*mem), worker.i[3].ch[20]);
    // seems to be unneeded, no write-alloc? but just in case..
    syscalls.cache_wb(&mem->sizeof_region_mem, 3 * 4, 1);
@@ -119,14 +159,18 @@ static unsigned int exec(dsp_component_cmd_t cmd,
   case CCMD_DOIT:
    worker->active = ACTIVE_CNT;
    worker->boot_cnt++;
-   syscalls.cache_wb(&worker->i_done, 64, 1);
-   memcpy(&spu_config, &mem->spu_config, sizeof(spu_config));
+   syscalls.cache_inv(worker, 128, 1);
+   syscalls.cache_wb(&worker->i_done, 128, 1);
+   memcpy(&spu_config, &mem->in.spu_config, sizeof(spu_config));
+
+   if (worker->ram_dirty)
+    // it's faster to do it all than just a 512k buffer
+    syscalls.cache_wbInvAll();
 
    do_processing();
 
-   // c64_tools lib does BCACHE_wbInvAll() when it receives mailbox irq,
-   // but invalidate anyway in case c64_tools is ever fixed..
-   syscalls.cache_inv(mem, sizeof(mem->spu_ram) + sizeof(mem->SB), 0);
+   syscalls.cache_inv(&mem->SB, sizeof(mem->SB), 0);
+   syscalls.cache_inv(&mem->in, sizeof(mem->in), 0);
    break;
 
   default: