@@ -111,6 +111,7 @@
 	int32_t shifterOperand;
 	int32_t shifterCarryOut;
 
+	uint32_t currentPC;
 	enum ExecutionMode executionMode;
 	enum PrivilegeMode privilegeMode;
 

@@ -194,7 +194,8 @@
 void ARMStep(struct ARMCore* cpu) {
 	// TODO
 	uint32_t opcode;
-	ARMInstruction instruction = _ARMLoadInstructionARM(cpu->memory, cpu->gprs[ARM_PC] - WORD_SIZE_ARM, &opcode);
+	cpu->currentPC = cpu->gprs[ARM_PC] - WORD_SIZE_ARM;
+	ARMInstruction instruction = _ARMLoadInstructionARM(cpu->memory, cpu->currentPC, &opcode);
 	cpu->gprs[ARM_PC] += WORD_SIZE_ARM;
 
 	int condition = opcode >> 28;

@@ -5,10 +5,9 @@
 static const ThumbInstruction _thumbTable[0x400];
 
 void ThumbStep(struct ARMCore* cpu) {
-	uint32_t address = cpu->gprs[ARM_PC];
-	cpu->gprs[ARM_PC] = address + WORD_SIZE_THUMB;
-	address -= WORD_SIZE_THUMB;
-	uint16_t opcode = ((uint16_t*) cpu->memory->activeRegion)[(address & cpu->memory->activeMask) >> 1];
+	cpu->currentPC = cpu->gprs[ARM_PC] - WORD_SIZE_THUMB;
+	cpu->gprs[ARM_PC] += WORD_SIZE_THUMB;
+	uint16_t opcode = ((uint16_t*) cpu->memory->activeRegion)[(cpu->currentPC & cpu->memory->activeMask) >> 1];
 	ThumbInstruction instruction = _thumbTable[opcode >> 6];
 	instruction(cpu, opcode);
 }

@@ -67,6 +67,7 @@ memory->d.activeMask = 0;
 	memory->d.setActiveRegion = GBASetActiveRegion;
 	memory->d.activePrefetchCycles32 = 0;
 	memory->d.activePrefetchCycles16 = 0;
+	memory->biosPrefetch = 0;
 	memory->d.waitMultiple = GBAWaitMultiple;
 }
 
@@ -79,6 +80,9 @@
 static void GBASetActiveRegion(struct ARMMemory* memory, uint32_t address) {
 	struct GBAMemory* gbaMemory = (struct GBAMemory*) memory;
 
+	if (gbaMemory->activeRegion == REGION_BIOS) {
+		gbaMemory->biosPrefetch = memory->load32(memory, gbaMemory->p->cpu.currentPC + WORD_SIZE_ARM * 2, 0);
+	}
 	gbaMemory->activeRegion = address >> BASE_OFFSET;
 	memory->activePrefetchCycles32 = gbaMemory->waitstatesPrefetch32[gbaMemory->activeRegion];
 	memory->activePrefetchCycles16 = gbaMemory->waitstatesPrefetch16[gbaMemory->activeRegion];
@@ -121,7 +125,15 @@ int wait = 0;
 
 	switch (address & ~OFFSET_MASK) {
 	case BASE_BIOS:
-		GBALog(gbaMemory->p, GBA_LOG_STUB, "Unimplemented memory Load32: 0x%08X", address);
+		if (gbaMemory->p->cpu.currentPC >> BASE_OFFSET == REGION_BIOS) {
+			if (address < hleBiosLength) {
+				value = gbaMemory->bios[address >> 2];
+			} else {
+				value = 0;
+			}
+		} else {
+			value = gbaMemory->biosPrefetch;
+		}
 		break;
 	case BASE_WORKING_RAM:
 		value = gbaMemory->wram[(address & (SIZE_WORKING_RAM - 1)) >> 2];
@@ -158,6 +170,12 @@ GBALog(gbaMemory->p, GBA_LOG_STUB, "Unimplemented memory Load32: 0x%08X", address);
 		break;
 	default:
 		GBALog(gbaMemory->p, GBA_LOG_GAME_ERROR, "Bad memory Load32: 0x%08X", address);
+		if (gbaMemory->p->cpu.executionMode == MODE_ARM) {
+			value = memory->load32(memory, gbaMemory->p->cpu.currentPC + WORD_SIZE_ARM * 2, 0);
+		} else {
+			value = memory->load16(memory, gbaMemory->p->cpu.currentPC + WORD_SIZE_THUMB * 2, 0);
+			value |= value << 16;
+		}
 		break;
 	}
 
@@ -225,6 +243,7 @@ GBALog(gbaMemory->p, GBA_LOG_STUB, "Unimplemented memory Load16: 0x%08X", address);
 		break;
 	default:
 		GBALog(gbaMemory->p, GBA_LOG_GAME_ERROR, "Bad memory Load16: 0x%08X", address);
+		value = memory->load16(memory, gbaMemory->p->cpu.currentPC + (gbaMemory->p->cpu.executionMode == MODE_ARM ? WORD_SIZE_ARM : WORD_SIZE_THUMB) * 2, 0);
 		break;
 	}
 
@@ -292,6 +311,7 @@ }
 		break;
 	default:
 		GBALog(gbaMemory->p, GBA_LOG_GAME_ERROR, "Bad memory Load8: 0x%08x", address);
+		value = memory->load16(memory, gbaMemory->p->cpu.currentPC + (gbaMemory->p->cpu.executionMode == MODE_ARM ? WORD_SIZE_ARM : WORD_SIZE_THUMB) * 2, 0) >> ((address & 1) << 3);
 		break;
 	}
 

@@ -122,6 +122,7 @@ char waitstatesSeq16[256];
 	char waitstatesPrefetch32[256];
 	char waitstatesPrefetch16[256];
 	int activeRegion;
+	uint32_t biosPrefetch;
 
 	struct GBADMA dma[4];
 };