]> Shamusworld >> Repos - virtualjaguar/blobdiff - src/gpu.cpp
Remove OP dump to fix lockups when exiting VJ.
[virtualjaguar] / src / gpu.cpp
index 4892b7aa536d4bee7d790435bacfd929b621fa28..e1fdd41e8751be298c26804e4a198a37de0c4f7d 100644 (file)
@@ -5,14 +5,15 @@
 //
 // Originally by David Raingeard (Cal2)
 // GCC/SDL port by Niels Wagenaar (Linux/WIN32) and Caz (BeOS)
-// Cleanups, endian wrongness, and bad ASM amelioration by James L. Hammons
+// Cleanups, endian wrongness, and bad ASM amelioration by James Hammons
 // (C) 2010 Underground Software
 //
-// JLH = James L. Hammons <jlhamm@acm.org>
+// JLH = James Hammons <jlhamm@acm.org>
 //
 // Who  When        What
 // ---  ----------  -------------------------------------------------------------
 // JLH  01/16/2010  Created this log ;-)
+// JLH  11/26/2011  Added fixes for LOAD/STORE alignment issues
 
 //
 // Note: Endian wrongness probably stems from the MAME origins of this emu and
 #include "jagdasm.h"
 #include "jaguar.h"
 #include "log.h"
-#include "m68k.h"
+#include "m68000/m68kinterface.h"
 //#include "memory.h"
 #include "tom.h"
 
+
+// Seems alignment in loads & stores was off...
+#define GPU_CORRECT_ALIGNMENT
 //#define GPU_DEBUG
 
 // For GPU dissasembly...
 
+#if 0
 #define GPU_DIS_ABS
 #define GPU_DIS_ADD
 #define GPU_DIS_ADDC
 #define GPU_DIS_SUBQT
 #define GPU_DIS_XOR
 
-bool doGPUDis = false;
-//bool doGPUDis = true;
-//*/
+//bool doGPUDis = false;
+bool doGPUDis = true;
+#endif
+
 /*
 GPU opcodes use (BIOS flying ATARI logo):
 +                    add 357416
@@ -739,10 +745,11 @@ WriteLog("GPU: %s setting GPU PC to %08X %s\n", whoName[who], gpu_pc, (GPU_RUNNI
 //WriteLog("GPU->CPU interrupt\n");
                                if (TOMIRQEnabled(IRQ_GPU))
                                {
-                                       if ((TOMIRQEnabled(IRQ_GPU)) && (JaguarInterruptHandlerIsValid(64)))
+//This is the programmer's responsibility, to make sure the handler is valid, not ours!
+//                                     if ((TOMIRQEnabled(IRQ_GPU))// && (JaguarInterruptHandlerIsValid(64)))
                                        {
                                                TOMSetPendingGPUInt();
-                                               m68k_set_irq(7);                        // Set 68000 NMI
+                                               m68k_set_irq(2);                        // Set 68000 IPL 2
                                                GPUReleaseTimeslice();
                                        }
                                }
@@ -1724,7 +1731,16 @@ static void gpu_opcode_store_r14_indexed(void)
        if (doGPUDis)
                WriteLog("%06X: STORE  R%02u, (R14+$%02X) [NCZ:%u%u%u, R%02u=%08X, R14+$%02X=%08X]\n", gpu_pc-2, IMM_2, gpu_convert_zero[IMM_1] << 2, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN, gpu_convert_zero[IMM_1] << 2, gpu_reg[14]+(gpu_convert_zero[IMM_1] << 2));
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       uint32 address = gpu_reg[14] + (gpu_convert_zero[IMM_1] << 2);
+       
+       if (address >= 0xF03000 && address <= 0xF03FFF)
+               GPUWriteLong(address & 0xFFFFFFFC, RN, GPU);
+       else
+               GPUWriteLong(address, RN, GPU);
+#else
        GPUWriteLong(gpu_reg[14] + (gpu_convert_zero[IMM_1] << 2), RN, GPU);
+#endif
 }
 
 static void gpu_opcode_store_r15_indexed(void)
@@ -1733,7 +1749,16 @@ static void gpu_opcode_store_r15_indexed(void)
        if (doGPUDis)
                WriteLog("%06X: STORE  R%02u, (R15+$%02X) [NCZ:%u%u%u, R%02u=%08X, R15+$%02X=%08X]\n", gpu_pc-2, IMM_2, gpu_convert_zero[IMM_1] << 2, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN, gpu_convert_zero[IMM_1] << 2, gpu_reg[15]+(gpu_convert_zero[IMM_1] << 2));
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       uint32 address = gpu_reg[15] + (gpu_convert_zero[IMM_1] << 2);
+
+       if (address >= 0xF03000 && address <= 0xF03FFF)
+               GPUWriteLong(address & 0xFFFFFFFC, RN, GPU);
+       else
+               GPUWriteLong(address, RN, GPU);
+#else
        GPUWriteLong(gpu_reg[15] + (gpu_convert_zero[IMM_1] << 2), RN, GPU);
+#endif
 }
 
 static void gpu_opcode_load_r14_ri(void)
@@ -1742,7 +1767,16 @@ static void gpu_opcode_load_r14_ri(void)
        if (doGPUDis)
                WriteLog("%06X: LOAD   (R14+R%02u), R%02u [NCZ:%u%u%u, R14+R%02u=%08X, R%02u=%08X] -> ", gpu_pc-2, IMM_1, IMM_2, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_1, RM+gpu_reg[14], IMM_2, RN);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       uint32 address = gpu_reg[14] + RM;
+
+       if (address >= 0xF03000 && address <= 0xF03FFF)
+               RN = GPUReadLong(address & 0xFFFFFFFC, GPU);
+       else
+               RN = GPUReadLong(address, GPU);
+#else
        RN = GPUReadLong(gpu_reg[14] + RM, GPU);
+#endif
 #ifdef GPU_DIS_LOAD14R
        if (doGPUDis)
                WriteLog("[NCZ:%u%u%u, R%02u=%08X]\n", gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN);
@@ -1755,7 +1789,16 @@ static void gpu_opcode_load_r15_ri(void)
        if (doGPUDis)
                WriteLog("%06X: LOAD   (R15+R%02u), R%02u [NCZ:%u%u%u, R15+R%02u=%08X, R%02u=%08X] -> ", gpu_pc-2, IMM_1, IMM_2, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_1, RM+gpu_reg[15], IMM_2, RN);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       uint32 address = gpu_reg[15] + RM;
+
+       if (address >= 0xF03000 && address <= 0xF03FFF)
+               RN = GPUReadLong(address & 0xFFFFFFFC, GPU);
+       else
+               RN = GPUReadLong(address, GPU);
+#else
        RN = GPUReadLong(gpu_reg[15] + RM, GPU);
+#endif
 #ifdef GPU_DIS_LOAD15R
        if (doGPUDis)
                WriteLog("[NCZ:%u%u%u, R%02u=%08X]\n", gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN);
@@ -1768,7 +1811,16 @@ static void gpu_opcode_store_r14_ri(void)
        if (doGPUDis)
                WriteLog("%06X: STORE  R%02u, (R14+R%02u) [NCZ:%u%u%u, R%02u=%08X, R14+R%02u=%08X]\n", gpu_pc-2, IMM_2, IMM_1, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN, IMM_1, RM+gpu_reg[14]);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       uint32 address = gpu_reg[14] + RM;
+
+       if (address >= 0xF03000 && address <= 0xF03FFF)
+               GPUWriteLong(address & 0xFFFFFFFC, RN, GPU);
+       else
+               GPUWriteLong(address, RN, GPU);
+#else
        GPUWriteLong(gpu_reg[14] + RM, RN, GPU);
+#endif
 }
 
 static void gpu_opcode_store_r15_ri(void)
@@ -1777,7 +1829,16 @@ static void gpu_opcode_store_r15_ri(void)
        if (doGPUDis)
                WriteLog("%06X: STORE  R%02u, (R15+R%02u) [NCZ:%u%u%u, R%02u=%08X, R15+R%02u=%08X]\n", gpu_pc-2, IMM_2, IMM_1, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN, IMM_1, RM+gpu_reg[15]);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT_STORE
+       uint32 address = gpu_reg[15] + RM;
+
+       if (address >= 0xF03000 && address <= 0xF03FFF)
+               GPUWriteLong(address & 0xFFFFFFFC, RN, GPU);
+       else
+               GPUWriteLong(address, RN, GPU);
+#else
        GPUWriteLong(gpu_reg[15] + RM, RN, GPU);
+#endif
 }
 
 static void gpu_opcode_nop(void)
@@ -1827,10 +1888,17 @@ static void gpu_opcode_storew(void)
        if (doGPUDis)
                WriteLog("%06X: STOREW R%02u, (R%02u) [NCZ:%u%u%u, R%02u=%08X, R%02u=%08X]\n", gpu_pc-2, IMM_2, IMM_1, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN, IMM_1, RM);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+               GPUWriteLong(RM & 0xFFFFFFFE, RN & 0xFFFF, GPU);
+       else
+               JaguarWriteWord(RM, RN, GPU);
+#else
        if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
                GPUWriteLong(RM, RN & 0xFFFF, GPU);
        else
                JaguarWriteWord(RM, RN, GPU);
+#endif
 }
 
 static void gpu_opcode_store(void)
@@ -1839,13 +1907,33 @@ static void gpu_opcode_store(void)
        if (doGPUDis)
                WriteLog("%06X: STORE  R%02u, (R%02u) [NCZ:%u%u%u, R%02u=%08X, R%02u=%08X]\n", gpu_pc-2, IMM_2, IMM_1, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN, IMM_1, RM);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+               GPUWriteLong(RM & 0xFFFFFFFC, RN, GPU);
+       else
+               GPUWriteLong(RM, RN, GPU);
+#else
        GPUWriteLong(RM, RN, GPU);
+#endif
 }
 
 static void gpu_opcode_storep(void)
 {
+#ifdef GPU_CORRECT_ALIGNMENT
+       if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+       {
+               GPUWriteLong((RM & 0xFFFFFFF8) + 0, gpu_hidata, GPU);
+               GPUWriteLong((RM & 0xFFFFFFF8) + 4, RN, GPU);
+       }
+       else
+       {
+               GPUWriteLong(RM + 0, gpu_hidata, GPU);
+               GPUWriteLong(RM + 4, RN, GPU);
+       }
+#else
        GPUWriteLong(RM + 0, gpu_hidata, GPU);
        GPUWriteLong(RM + 4, RN, GPU);
+#endif
 }
 
 static void gpu_opcode_loadb(void)
@@ -1870,23 +1958,60 @@ static void gpu_opcode_loadw(void)
        if (doGPUDis)
                WriteLog("%06X: LOADW  (R%02u), R%02u [NCZ:%u%u%u, R%02u=%08X, R%02u=%08X] -> ", gpu_pc-2, IMM_1, IMM_2, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_1, RM, IMM_2, RN);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+               RN = GPUReadLong(RM & 0xFFFFFFFE, GPU) & 0xFFFF;
+       else
+               RN = JaguarReadWord(RM, GPU);
+#else
        if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
                RN = GPUReadLong(RM, GPU) & 0xFFFF;
        else
                RN = JaguarReadWord(RM, GPU);
+#endif
 #ifdef GPU_DIS_LOADW
        if (doGPUDis)
                WriteLog("[NCZ:%u%u%u, R%02u=%08X]\n", gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN);
 #endif
 }
 
+// According to the docs, & "Do The Same", this address is long aligned...
+// So let's try it:
+// And it works!!! Need to fix all instances...
+// Also, Power Drive Rally seems to contradict the idea that only LOADs in
+// the $F03000-$F03FFF range are aligned...
+#warning "!!! Alignment issues, need to find definitive final word on this !!!"
+/*
+Preliminary testing on real hardware seems to confirm that something strange goes on
+with unaligned reads in main memory. When the address is off by 1, the result is the
+same as the long address with the top byte replaced by something. So if the read is
+from $401, and $400 has 12 34 56 78, the value read will be $nn345678, where nn is a currently unknown vlaue.
+When the address is off by 2, the result would be $nnnn5678, where nnnn is unknown.
+When the address is off by 3, the result would be $nnnnnn78, where nnnnnn is unknown.
+It may be that the "unknown" values come from the prefetch queue, but not sure how
+to test that. They seem to be stable, though, which would indicate such a mechanism.
+Sometimes, however, the off by 2 case returns $12345678!
+*/
 static void gpu_opcode_load(void)
 {
 #ifdef GPU_DIS_LOAD
        if (doGPUDis)
                WriteLog("%06X: LOAD   (R%02u), R%02u [NCZ:%u%u%u, R%02u=%08X, R%02u=%08X] -> ", gpu_pc-2, IMM_1, IMM_2, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_1, RM, IMM_2, RN);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       uint32 mask[4] = { 0x00000000, 0xFF000000, 0xFFFF0000, 0xFFFFFF00 };
+//     if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+               RN = GPUReadLong(RM & 0xFFFFFFFC, GPU);
+//             RN = GPUReadLong(RM & 0x00FFFFFC, GPU);
+//     else
+//             RN = GPUReadLong(RM, GPU);
+       // Simulate garbage in unaligned reads...
+//seems that this behavior is different in GPU mem vs. main mem...
+//     if ((RM < 0xF03000) || (RM > 0xF0BFFF))
+//             RN |= mask[RM & 0x03];
+#else
        RN = GPUReadLong(RM, GPU);
+#endif
 #ifdef GPU_DIS_LOAD
        if (doGPUDis)
                WriteLog("[NCZ:%u%u%u, R%02u=%08X]\n", gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN);
@@ -1895,8 +2020,21 @@ static void gpu_opcode_load(void)
 
 static void gpu_opcode_loadp(void)
 {
+#ifdef GPU_CORRECT_ALIGNMENT
+       if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+       {
+               gpu_hidata = GPUReadLong((RM & 0xFFFFFFF8) + 0, GPU);
+               RN                 = GPUReadLong((RM & 0xFFFFFFF8) + 4, GPU);
+       }
+       else
+       {
+               gpu_hidata = GPUReadLong(RM + 0, GPU);
+               RN                 = GPUReadLong(RM + 4, GPU);
+       }
+#else
        gpu_hidata = GPUReadLong(RM + 0, GPU);
        RN                 = GPUReadLong(RM + 4, GPU);
+#endif
 }
 
 static void gpu_opcode_load_r14_indexed(void)
@@ -1905,7 +2043,16 @@ static void gpu_opcode_load_r14_indexed(void)
        if (doGPUDis)
                WriteLog("%06X: LOAD   (R14+$%02X), R%02u [NCZ:%u%u%u, R14+$%02X=%08X, R%02u=%08X] -> ", gpu_pc-2, gpu_convert_zero[IMM_1] << 2, IMM_2, gpu_flag_n, gpu_flag_c, gpu_flag_z, gpu_convert_zero[IMM_1] << 2, gpu_reg[14]+(gpu_convert_zero[IMM_1] << 2), IMM_2, RN);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       uint32 address = gpu_reg[14] + (gpu_convert_zero[IMM_1] << 2);
+
+       if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+               RN = GPUReadLong(address & 0xFFFFFFFC, GPU);
+       else
+               RN = GPUReadLong(address, GPU);
+#else
        RN = GPUReadLong(gpu_reg[14] + (gpu_convert_zero[IMM_1] << 2), GPU);
+#endif
 #ifdef GPU_DIS_LOAD14I
        if (doGPUDis)
                WriteLog("[NCZ:%u%u%u, R%02u=%08X]\n", gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN);
@@ -1918,7 +2065,16 @@ static void gpu_opcode_load_r15_indexed(void)
        if (doGPUDis)
                WriteLog("%06X: LOAD   (R15+$%02X), R%02u [NCZ:%u%u%u, R15+$%02X=%08X, R%02u=%08X] -> ", gpu_pc-2, gpu_convert_zero[IMM_1] << 2, IMM_2, gpu_flag_n, gpu_flag_c, gpu_flag_z, gpu_convert_zero[IMM_1] << 2, gpu_reg[15]+(gpu_convert_zero[IMM_1] << 2), IMM_2, RN);
 #endif
+#ifdef GPU_CORRECT_ALIGNMENT
+       uint32 address = gpu_reg[15] + (gpu_convert_zero[IMM_1] << 2);
+
+       if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+               RN = GPUReadLong(address & 0xFFFFFFFC, GPU);
+       else
+               RN = GPUReadLong(address, GPU);
+#else
        RN = GPUReadLong(gpu_reg[15] + (gpu_convert_zero[IMM_1] << 2), GPU);
+#endif
 #ifdef GPU_DIS_LOAD15I
        if (doGPUDis)
                WriteLog("[NCZ:%u%u%u, R%02u=%08X]\n", gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN);