//
// 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
// a bit before writing a result. I.e., if the result of an operation leaves a zero in
// the carry flag, you don't have to zero gpu_flag_c before you can write that zero!
static uint8 gpu_flag_z, gpu_flag_n, gpu_flag_c;
-static uint32 gpu_reg_bank_0[32];
-static uint32 gpu_reg_bank_1[32];
+uint32 gpu_reg_bank_0[32];
+uint32 gpu_reg_bank_1[32];
static uint32 * gpu_reg;
static uint32 * gpu_alternate_reg;
uint32 GPUReadLong(uint32 offset, uint32 who/*=UNKNOWN*/)
{
if (offset >= 0xF02000 && offset <= 0xF020FF)
- WriteLog("GPU: ReadLong--Attempt to read from GPU register file by %s!\n", whoName[who]);
+ {
+ WriteLog("GPU: ReadLong--Attempt to read from GPU register file (%X) by %s!\n", offset, whoName[who]);
+ uint32 reg = (offset & 0xFC) >> 2;
+ return (reg < 32 ? gpu_reg_bank_0[reg] : gpu_reg_bank_1[reg - 32]);
+ }
// if ((offset >= GPU_WORK_RAM_BASE) && (offset < GPU_WORK_RAM_BASE + 0x1000))
if ((offset >= GPU_WORK_RAM_BASE) && (offset <= GPU_WORK_RAM_BASE + 0x0FFC))
{
//WriteLog("[GPU W16:%08X,%04X]", offset, data);
uint32 old_data = GPUReadLong(offset & 0xFFFFFFC, who);
+
if (offset & 0x02)
old_data = (old_data & 0xFFFF0000) | (data & 0xFFFF);
else
old_data = (old_data & 0x0000FFFF) | ((data & 0xFFFF) << 16);
+
GPUWriteLong(offset & 0xFFFFFFC, old_data, who);
}
+
return;
}
else if ((offset == GPU_WORK_RAM_BASE + 0x0FFF) || (GPU_CONTROL_RAM_BASE + 0x1F))
case 0x00:
{
bool IMASKCleared = (gpu_flags & IMASK) && !(data & IMASK);
- gpu_flags = data;
+ // NOTE: According to the JTRM, writing a 1 to IMASK has no effect; only the
+ // IRQ logic can set it. So we mask it out here to prevent problems...
+ gpu_flags = data & (~IMASK);
gpu_flag_z = gpu_flags & ZERO_FLAG;
gpu_flag_c = (gpu_flags & CARRY_FLAG) >> 1;
gpu_flag_n = (gpu_flags & NEGA_FLAG) >> 2;
{
//WriteLog("asked to perform a single step (single step is %senabled)\n",(data&0x8)?"":"not ");
}
+
gpu_control = (gpu_control & 0xF7C0) | (data & (~0xF7C0));
// if gpu wasn't running but is now running, execute a few cycles
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)
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)
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);
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);
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)
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)
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)
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)
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 $12345678!
+*/
static void gpu_opcode_load(void)
{
#ifdef GPU_DIS_LOAD
if (doGPUDis)
WriteLog("%06X: LOAD (R%02u), R%02u [NCZ:%u%u%u, R%02u=%08X, R%02u=%08X] -> ", gpu_pc-2, IMM_1, IMM_2, gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_1, RM, IMM_2, RN);
#endif
+#ifdef GPU_CORRECT_ALIGNMENT
+ uint32 mask[4] = { 0x00000000, 0xFF000000, 0xFFFF0000, 0xFFFFFF00 };
+// if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+ RN = GPUReadLong(RM & 0xFFFFFFFC, GPU);
+// RN = GPUReadLong(RM & 0x00FFFFFC, GPU);
+// else
+// RN = GPUReadLong(RM, GPU);
+ // Simulate garbage in unaligned reads...
+//seems that this behavior is different in GPU mem vs. main mem...
+// if ((RM < 0xF03000) || (RM > 0xF0BFFF))
+// RN |= mask[RM & 0x03];
+#else
RN = GPUReadLong(RM, GPU);
+#endif
#ifdef GPU_DIS_LOAD
if (doGPUDis)
WriteLog("[NCZ:%u%u%u, R%02u=%08X]\n", gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN);
static void gpu_opcode_loadp(void)
{
+#ifdef GPU_CORRECT_ALIGNMENT
+ if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+ {
+ gpu_hidata = GPUReadLong((RM & 0xFFFFFFF8) + 0, GPU);
+ RN = GPUReadLong((RM & 0xFFFFFFF8) + 4, GPU);
+ }
+ else
+ {
+ gpu_hidata = GPUReadLong(RM + 0, GPU);
+ RN = GPUReadLong(RM + 4, GPU);
+ }
+#else
gpu_hidata = GPUReadLong(RM + 0, GPU);
RN = GPUReadLong(RM + 4, GPU);
+#endif
}
static void gpu_opcode_load_r14_indexed(void)
if (doGPUDis)
WriteLog("%06X: LOAD (R14+$%02X), R%02u [NCZ:%u%u%u, R14+$%02X=%08X, R%02u=%08X] -> ", gpu_pc-2, gpu_convert_zero[IMM_1] << 2, IMM_2, gpu_flag_n, gpu_flag_c, gpu_flag_z, gpu_convert_zero[IMM_1] << 2, gpu_reg[14]+(gpu_convert_zero[IMM_1] << 2), IMM_2, RN);
#endif
+#ifdef GPU_CORRECT_ALIGNMENT
+ uint32 address = gpu_reg[14] + (gpu_convert_zero[IMM_1] << 2);
+
+ if ((RM >= 0xF03000) && (RM <= 0xF03FFF))
+ RN = GPUReadLong(address & 0xFFFFFFFC, GPU);
+ else
+ RN = GPUReadLong(address, GPU);
+#else
RN = GPUReadLong(gpu_reg[14] + (gpu_convert_zero[IMM_1] << 2), GPU);
+#endif
#ifdef GPU_DIS_LOAD14I
if (doGPUDis)
WriteLog("[NCZ:%u%u%u, R%02u=%08X]\n", gpu_flag_n, gpu_flag_c, gpu_flag_z, IMM_2, RN);
if (doGPUDis)
WriteLog("%06X: LOAD (R15+$%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);