Made some minor tweaks to SIF to test some theories and fix some broken-looking code -- please report any regressions. Also commented some of SIF and IPU stuff. :)

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@3812 96395faa-99c1-11dd-bbfe-3dabce05a288

common/include/Pcsx2Types.h

@@ -161,7 +161,7 @@ union u128
 
 	bool operator!=( const u128& right ) const
 	{
-		return (lo != right.lo) && (hi != right.hi);
+		return (lo != right.lo) || (hi != right.hi);
 	}
 
 	// In order for the following ToString() and WriteTo methods to be available, you must
@@ -206,7 +206,7 @@ struct s128
 
 	bool operator!=( const s128& right ) const
 	{
-		return (lo != right.lo) && (hi != right.hi);
+		return (lo != right.lo) || (hi != right.hi);
 	}
 };
 

pcsx2/Dmac.h

@@ -92,6 +92,8 @@ union tDMA_TAG {
 	};
 	u32 _u32;
 
+	tDMA_TAG() {}
+
 	tDMA_TAG(u32 val) { _u32 = val; }
 	u16 upper() const { return (_u32 >> 16); }
 	u16 lower() const { return (u16)_u32; }

pcsx2/IPU/IPU.cpp

@@ -206,7 +206,7 @@ __fi u32 ipuRead32(u32 mem)
 			if (!ipuRegs.ctrl.BUSY)
 				IPU_LOG("read32: IPU_CTRL=0x%08X", ipuRegs.ctrl._u32);
 
-			return ipuRegs.ctrl._u32;
+		return ipuRegs.ctrl._u32;
 		}		
 
 		ipucase(IPU_BP): // IPU_BP
@@ -218,7 +218,7 @@ __fi u32 ipuRead32(u32 mem)
 			ipuRegs.ipubp |= g_BP.FP << 16;
 
 			IPU_LOG("read32: IPU_BP=0x%08X", ipuRegs.ipubp);
-			return ipuRegs.ipubp;
+		return ipuRegs.ipubp;
 		}
 
 		default:
@@ -448,6 +448,14 @@ static bool __fastcall ipuVDEC(u32 val)
 				jNO_DEFAULT
 			}
 
+			// HACK ATTACK!  This code OR's the MPEG decoder's bitstream position into the upper
+			// 16 bits of DATA; which really doesn't make sense since (a) we already rewound the bits
+			// back into the IPU internal buffer above, and (b) the IPU doesn't have an MPEG internal
+			// 32-bit decoder buffer of its own anyway.  Furthermore, setting the upper 16 bits to
+			// any value other than zero appears to work fine.  When set to zero, however, FMVs run
+			// very choppy (basically only decoding/updating every 30th frame or so). So yeah,
+			// someone with knowledge on the subject please feel free to explain this one. :) --air
+
 			ipuRegs.cmd.DATA &= 0xFFFF;
 			ipuRegs.cmd.DATA |= 0x10000;
 
@@ -466,6 +474,7 @@ static bool __fastcall ipuVDEC(u32 val)
 			IPU_LOG("VDEC command data 0x%x(0x%x). Skip 0x%X bits/Table=%d (%s), pct %d",
 			        ipuRegs.cmd.DATA, ipuRegs.cmd.DATA >> 16, val & 0x3f, (val >> 26) & 3, (val >> 26) & 1 ?
 			        ((val >> 26) & 2 ? "DMV" : "MBT") : (((val >> 26) & 2 ? "MC" : "MBAI")), ipuRegs.ctrl.PCT);
+
 			return true;
 
 		jNO_DEFAULT
@@ -488,8 +497,6 @@ static __fi bool ipuFDEC(u32 val)
 
 static bool ipuSETIQ(u32 val)
 {
-	int i;
-
 	if ((val >> 27) & 1)
 	{
 		u8 (&niq)[64] = decoder.niq;
@@ -500,7 +507,7 @@ static bool ipuSETIQ(u32 val)
 		}
 
 		IPU_LOG("Read non-intra quantization matrix from FIFO.");
-		for (i = 0; i < 8; i++)
+		for (uint i = 0; i < 8; i++)
 		{
 			IPU_LOG("%02X %02X %02X %02X %02X %02X %02X %02X",
 			        niq[i * 8 + 0], niq[i * 8 + 1], niq[i * 8 + 2], niq[i * 8 + 3],
@@ -517,7 +524,7 @@ static bool ipuSETIQ(u32 val)
 		}
 
 		IPU_LOG("Read intra quantization matrix from FIFO.");
-		for (i = 0; i < 8; i++)
+		for (uint i = 0; i < 8; i++)
 		{
 			IPU_LOG("%02X %02X %02X %02X %02X %02X %02X %02X",
 			        iq[i * 8 + 0], iq[i * 8 + 1], iq[i * 8 + 2], iq[i *8 + 3],
@@ -539,7 +546,7 @@ static bool ipuSETVQ(u32 val)
 	    "%02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d\n"
 	    "%02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d\n"
 	    "%02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d\n"
-	    "%02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d",
+		"%02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d %02d:%02d:%02d",
 	    vqclut[0] >> 10, (vqclut[0] >> 5) & 0x1F, vqclut[0] & 0x1F,
 	    vqclut[1] >> 10, (vqclut[1] >> 5) & 0x1F, vqclut[1] & 0x1F,
 	    vqclut[2] >> 10, (vqclut[2] >> 5) & 0x1F, vqclut[2] & 0x1F,

pcsx2/IopDma.cpp

@@ -195,14 +195,8 @@ void psxDma9(u32 madr, u32 bcr, u32 chcr)
 	SIF_LOG("IOP: dmaSIF0 chcr = %lx, madr = %lx, bcr = %lx, tadr = %lx",	chcr, madr, bcr, HW_DMA9_TADR);
 
 	sif0.iop.busy = true;
-	psHu32(SBUS_F240) |= 0x2000;
 
-	/*if (sif0.ee.busy)
+	SIF0Dma();
-	{*/
-		SIF0Dma();
-		psHu32(SBUS_F240) &= ~0x20;
-		psHu32(SBUS_F240) &= ~0x2000;
-	//}
 }
 
 void psxDma10(u32 madr, u32 bcr, u32 chcr)
@@ -210,15 +204,8 @@ void psxDma10(u32 madr, u32 bcr, u32 chcr)
 	SIF_LOG("IOP: dmaSIF1 chcr = %lx, madr = %lx, bcr = %lx",	chcr, madr, bcr);
 
 	sif1.iop.busy = true;
-	psHu32(SBUS_F240) |= 0x4000;
 
-	/*if (sif1.ee.busy)
+	SIF1Dma();
-	{*/
-		SIF1Dma();
-		psHu32(SBUS_F240) &= ~0x40;
-		psHu32(SBUS_F240) &= ~0x100;
-		psHu32(SBUS_F240) &= ~0x4000;
-	//}
 }
 
 /* psxDma11 & psxDma 12 are in IopSio2.cpp, along with the appropriate interrupt functions. */

pcsx2/Sif.h

@@ -22,13 +22,17 @@ static DMACh& sif0dma = (DMACh&)eeHw[0xc000];
 static DMACh& sif1dma = (DMACh&)eeHw[0xc400];
 static DMACh& sif2dma = (DMACh&)eeHw[0xc800];
 
-
+// Despite its name, this is actually the IOP's DMAtag, which itself also contains
+// the EE's DMAtag in its upper 64 bits.  Note that only the lower 24 bits of 'data' is
+// the IOP's chain transfer address (loaded into MADR).  Bits 30 and 31 are transfer stop
+// bits of some sort.
 struct sifData
 {
 	s32 data;
 	s32 words;
-	s32 count; // I think this is unused.
+
-	s32 addr; // This too.
+	tDMA_TAG	tag_lo;		// EE DMA tag
+	tDMA_TAG	tag_hi;		// EE DMA tag
 };
 
 struct sifFifo
@@ -100,6 +104,7 @@ struct sif_ee
 
 	s32 cycles;
 };
+
 struct sif_iop
 {
 	bool end;
@@ -110,6 +115,7 @@ struct sif_iop
 	s32 counter; // Used to keep track of how much is left in IOP.
 	struct sifData data; // Only used in IOP.
 };
+
 struct _sif
 {
 	sifFifo fifo; // Used in both.

pcsx2/Sif0.cpp

@@ -69,8 +69,8 @@ static __fi bool WriteIOPtoFifo()
 
 	SIF_LOG("Write IOP to Fifo: +++++++++++ %lX of %lX", writeSize, sif0.iop.counter);
 
-	sif0.fifo.write((u32*)iopPhysMem(hw_dma(9).madr), writeSize);
+	sif0.fifo.write((u32*)iopPhysMem(hw_dma9.madr), writeSize);
-	hw_dma(9).madr += writeSize << 2;
+	hw_dma9.madr += writeSize << 2;
 
 	// iop is 1/8th the clock rate of the EE and psxcycles is in words (not quadwords).
 	sif0.iop.cycles += (writeSize >> 2)/* * BIAS*/;		// fixme : should be >> 4
@@ -83,13 +83,13 @@ static __fi bool WriteIOPtoFifo()
 static __fi bool ProcessEETag()
 {
 	static __aligned16 u32 tag[4];
+	tDMA_TAG& ptag(*(tDMA_TAG*)tag);
 
 	sif0.fifo.read((u32*)&tag[0], 4); // Tag
 	SIF_LOG("SIF0 EE read tag: %x %x %x %x", tag[0], tag[1], tag[2], tag[3]);
 
-	sif0dma.unsafeTransfer(((tDMA_TAG*)(tag)));
+	sif0dma.unsafeTransfer(&ptag);
 	sif0dma.madr = tag[1];
-	tDMA_TAG ptag(tag[0]);
 
 	SIF_LOG("SIF0 EE dest chain tag madr:%08X qwc:%04X id:%X irq:%d(%08X_%08X)",
 		sif0dma.madr, sif0dma.qwc, ptag.ID, ptag.IRQ, tag[1], tag[0]);
@@ -102,16 +102,12 @@ static __fi bool ProcessEETag()
 
 	switch (ptag.ID)
 	{
-		case TAG_REFE:
+		case TAG_CNT:	break;
-			sif0.ee.end = true;
-			if (dmacRegs.ctrl.STS != NO_STS)
-				dmacRegs.stadr.ADDR = sif0dma.madr + (sif0dma.qwc * 16);
-				break;
 
-		case TAG_REFS:
+		case TAG_CNTS:
 			if (dmacRegs.ctrl.STS != NO_STS)
 				dmacRegs.stadr.ADDR = sif0dma.madr + (sif0dma.qwc * 16);
-				break;
+			break;
 
 		case TAG_END:
 			sif0.ee.end = true;
@@ -120,22 +116,29 @@ static __fi bool ProcessEETag()
 	return true;
 }
 
-// Read Fifo into an iop tag, and transfer it to hw_dma(9). And presumably process it.
+// Read Fifo into an iop tag, and transfer it to hw_dma9. And presumably process it.
 static __fi bool ProcessIOPTag()
 {
-	// Process DMA tag at hw_dma(9).tadr
+	// Process DMA tag at hw_dma9.tadr
-	sif0.iop.data = *(sifData *)iopPhysMem(hw_dma(9).tadr);
+	sif0.iop.data = *(sifData *)iopPhysMem(hw_dma9.tadr);
 	sif0.iop.data.words = (sif0.iop.data.words + 3) & 0xfffffffc; // Round up to nearest 4.
-	sif0.fifo.write((u32*)iopPhysMem(hw_dma(9).tadr + 8), 4);
 
-	hw_dma(9).tadr += 16; ///hw_dma(9).madr + 16 + sif0.sifData.words << 2;
+	// send the EE's side of the DMAtag.  The tag is only 64 bits, with the upper 64 bits
+	// ignored by the EE.
 
-	// We're only copying the first 24 bits.
+	sif0.fifo.write((u32*)iopPhysMem(hw_dma9.tadr + 8), 2);
-	hw_dma(9).madr = sif0data & 0xFFFFFF;
+	sif0.fifo.writePos = (sif0.fifo.writePos + 2) & (FIFO_SIF_W - 1);		// iggy on the upper 64.
+	sif0.fifo.size += 2;
+
+	hw_dma9.tadr += 16; ///hw_dma9.madr + 16 + sif0.sifData.words << 2;
+
+	// We're only copying the first 24 bits.  Bits 30 and 31 (checked below) are Stop/IRQ bits.
+	hw_dma9.madr = sif0data & 0xFFFFFF;
 	sif0.iop.counter = sif0words;
 
+	// IOP tags have an IRQ bit and an End of Transfer bit:
 	if (sif0tag.IRQ  || (sif0tag.ID & 4)) sif0.iop.end = true;
-	SIF_LOG("SIF0 IOP Tag: madr=%lx, tadr=%lx, counter=%lx (%08X_%08X)", hw_dma(9).madr, hw_dma(9).tadr, sif0.iop.counter, sif0words, sif0data);
+	SIF_LOG("SIF0 IOP Tag: madr=%lx, tadr=%lx, counter=%lx (%08X_%08X)", hw_dma9.madr, hw_dma9.tadr, sif0.iop.counter, sif0words, sif0data);
 
 	return true;
 }
@@ -265,7 +268,7 @@ static __fi void HandleIOPTransfer()
 		}
 		else
 		{
-			// Read Fifo into an iop tag, and transfer it to hw_dma(9).
+			// Read Fifo into an iop tag, and transfer it to hw_dma9.
 			// And presumably process it.
 			ProcessIOPTag();
 		}
@@ -282,6 +285,9 @@ static __fi void HandleIOPTransfer()
 
 static __fi void Sif0End()
 {
+	psHu32(SBUS_F240) &= ~0x20;
+	psHu32(SBUS_F240) &= ~0x2000;
+
 	SIF_LOG("SIF0 DMA end...");
 }
 
@@ -351,8 +357,5 @@ __fi void dmaSIF0()
 	{
         //hwIntcIrq(INTC_SBUS); // not sure, so let's not
 		SIF0Dma();
-		// Do we really want to mess with the SIF flags like that? Nah.
-		//psHu32(SBUS_F240) &= ~0x20;
-		//psHu32(SBUS_F240) &= ~0x2000;
 	}
 }

pcsx2/Sif1.cpp

@@ -68,9 +68,9 @@ static __fi bool WriteFifoToIOP()
 
 	SIF_LOG("Sif 1 IOP doing transfer %04X to %08X", readSize, HW_DMA10_MADR);
 
-	sif1.fifo.read((u32*)iopPhysMem(hw_dma(10).madr), readSize);
+	sif1.fifo.read((u32*)iopPhysMem(hw_dma10.madr), readSize);
-	psxCpu->Clear(hw_dma(10).madr, readSize);
+	psxCpu->Clear(hw_dma10.madr, readSize);
-	hw_dma(10).madr += readSize << 2;
+	hw_dma10.madr += readSize << 2;
 	sif1.iop.cycles += readSize >> 2;		// fixme: should be >> 4
 	sif1.iop.counter -= readSize;
 
@@ -151,7 +151,7 @@ static __fi bool SIFIOPReadTag()
 		sif1data & 0xffffff, sif1words, sif1tag.ID, sif1tag.IRQ);
 
 	// Only use the first 24 bits.
-	hw_dma(10).madr = sif1data & 0xffffff;
+	hw_dma10.madr = sif1data & 0xffffff;
 
 	sif1.iop.counter = sif1words;
 	if (sif1tag.IRQ  || (sif1tag.ID & 4)) sif1.iop.end = true;
@@ -276,6 +276,9 @@ static __fi void HandleIOPTransfer()
 
 static __fi void Sif1End()
 {
+	psHu32(SBUS_F240) &= ~0x40;
+	psHu32(SBUS_F240) &= ~0x4000;
+
 	SIF_LOG("SIF1 DMA end...");
 }
 
@@ -349,9 +352,5 @@ __fi void dmaSIF1()
 	if (sif1.iop.busy)
 	{
 		SIF1Dma();
-		// Do we really want to mess with the SIF flags like that? Nah.
-		//psHu32(SBUS_F240) &= ~0x40;
-		//psHu32(SBUS_F240) &= ~0x100;
-		//psHu32(SBUS_F240) &= ~0x4000;
 	}
 }

pcsx2/windows/VCprojects/IopSif.cpp

@@ -17,20 +17,20 @@ s32 PrepareEERead()
 {
 	static __aligned16 u32 tag[4];
 
-	// Process DMA tag at hw_dma(9).tadr
+	// Process DMA tag at hw_dma9.tadr
-	sif0.iop.data = *(sifData *)iopPhysMem(hw_dma(9).tadr);
+	sif0.iop.data = *(sifData *)iopPhysMem(hw_dma9.tadr);
 	sif0.iop.data.words = (sif0.iop.data.words + 3) & 0xfffffffc; // Round up to nearest 4.
-	memcpy(tag, (u32*)iopPhysMem(hw_dma(9).tadr + 8), 16);
+	memcpy(tag, (u32*)iopPhysMem(hw_dma9.tadr + 8), 16);
 
-	hw_dma(9).tadr += 16; ///hw_dma(9).madr + 16 + sif0.sifData.words << 2;
+	hw_dma9.tadr += 16; ///hw_dma9.madr + 16 + sif0.sifData.words << 2;
 
 	// We're only copying the first 24 bits.
-	hw_dma(9).madr = sif0data & 0xFFFFFF;
+	hw_dma9.madr = sif0data & 0xFFFFFF;
 	sif0.iop.counter = sif0words;
 
 	if (sif0tag.IRQ  || (sif0tag.ID & 4)) sif0.iop.end = true;
 	SIF_LOG("SIF0 IOP to EE Tag: madr=%lx, tadr=%lx, counter=%lx (%08X_%08X)"
-		"\n\tread tag: %x %x %x %x", hw_dma(9).madr, hw_dma(9).tadr, sif0.iop.counter, sif0words, sif0data,
+		"\n\tread tag: %x %x %x %x", hw_dma9.madr, hw_dma9.tadr, sif0.iop.counter, sif0words, sif0data,
 		tag[0], tag[1], tag[2], tag[3]);
 
 	sif0dma.unsafeTransfer(((tDMA_TAG*)(tag)));
@@ -98,7 +98,7 @@ s32 DoSifRead(u32 iopAvailable)
 		return false;
 	}
 
-	memcpy((u32*)ptag, (u32*)iopPhysMem(hw_dma(9).madr), transferSizeBytes);
+	memcpy((u32*)ptag, (u32*)iopPhysMem(hw_dma9.madr), transferSizeBytes);
 
 	// Clearing handled by vtlb memory protection and manual blocks.
 	//Cpu->Clear(sif0dma.madr, readSize*4);