- Added static const int DUCK_DELAY_FRAMES = 30 to define the delay (1 second at 30 FPS).

- Added `int duckFrameCounter` to track frames the duck button is held.
  - Added `bool isDucking` to track crouch state and prevent re-triggering.
  - Updated `CPad` constructor to initialize `duckFrameCounter` and `isDucking`.

liberty/Makefile

@@ -15,16 +15,16 @@ IS_MAC := $(shell uname -s | grep -i "darwin" > /dev/null && echo "yes" || echo
 
 MOD_NAME?=
 
-GTA_DIR?=../../gta3
+GTA_DIR?=../../liberty
 GTA_GAMEFILES_LOOSE_DIR?=../gamefiles/liberty
-GTA_MOD_DIR?=../../gta3_mod$(MOD_NAME)
+GTA_MOD_DIR?=../../liberty_mod$(MOD_NAME)
 GTA_MOD_IMG_DIR?=$(GTA_MOD_DIR)/img
 GTA_MOD_CUTS_DIR=$(GTA_MOD_DIR)/cuts
 GTA_MOD_SFX_DIR?=$(GTA_MOD_DIR)/sfx
 GTA_MOD_LOOSE_DIR?=$(GTA_MOD_DIR)/loose
 
 REPACK_DIR?=repack-data
-REPACK_GTA_DIR?=$(REPACK_DIR)/gta3
+REPACK_GTA_DIR?=$(REPACK_DIR)/liberty
 REPACK_IMG_ORIG_DIR?=$(REPACK_DIR)/img-orig
 REPACK_IMG_DC_DIR?=$(REPACK_DIR)/img-dc
 REPACK_SFX_ORIG_DIR?=$(REPACK_DIR)/sfx-orig
@@ -183,6 +183,14 @@ $(REPACK_GTA_DIR)/GTA3SF8.b: assets/GTA3SF8.b
 	mkdir -p $(@D)
 	cp $< $@
 
+$(REPACK_GTA_DIR)/0GDTEX.PVR: assets/0GDTEX.PVR
+	mkdir -p $(@D)
+	cp $< $@
+
+$(REPACK_GTA_DIR)/%.ico: assets/%.ico
+	mkdir -p $(@D)
+	cp $< $@
+
 1ST_READ.BIN: $(TARGET)
 	rm -f $(TARGET).bin
 	rm -f 1ST_READ.BIN
@@ -198,20 +206,20 @@ IP.BIN:
 	rm -f IP.BIN
 	$(KOS_BASE)/utils/makeip/makeip ip.txt IP.BIN
 
-$(PROJECT_NAME).iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b
+$(PROJECT_NAME).iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	rm -f $(PROJECT_NAME).iso
 	rm -f $(REPACK_GTA_DIR)/1ST_READ.BIN
 	cp 1ST_READ.BIN $(REPACK_GTA_DIR)
 	mkisofs -C 0,11702 -V $(PROJECT_NAME) -G IP.BIN -r -J -l -o $(PROJECT_NAME).iso $(REPACK_GTA_DIR)
 
-$(PROJECT_NAME)-no-repack.iso: IP.BIN 1ST_READ.BIN $(REPACK_GTA_DIR)/GTA3SF8.b
+$(PROJECT_NAME)-no-repack.iso: IP.BIN 1ST_READ.BIN $(REPACK_GTA_DIR)/GTA3SF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	rm -f $(PROJECT_NAME)-no-repack.iso
 	rm -f $(REPACK_GTA_DIR)/1ST_READ.BIN
 	cp 1ST_READ.BIN $(REPACK_GTA_DIR)
 	mkisofs -C 0,11702 -V $(PROJECT_NAME) -G IP.BIN -r -J -l -o $(PROJECT_NAME)-no-repack.iso $(REPACK_GTA_DIR)
 
 
-$(PROJECT_NAME).ds.iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b
+$(PROJECT_NAME).ds.iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	rm -f $(PROJECT_NAME).ds.iso
 	rm -f $(REPACK_GTA_DIR)/1ST_READ.BIN
 	cp $(TARGET).bin $(REPACK_GTA_DIR)/1ST_READ.BIN
@@ -222,7 +230,7 @@ $(PROJECT_NAME).ds.iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_
 	$(KOS_BASE)/utils/scramble/scramble $(TARGET)-prebuilt.bin 1ST_READ_PREBUILT.BIN
 	mkdir -p $(REPACK_GTA_DIR)
 
-$(PROJECT_NAME)-prebuilt.iso: IP.BIN 1ST_READ_PREBUILT.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b
+$(PROJECT_NAME)-prebuilt.iso: IP.BIN 1ST_READ_PREBUILT.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	rm -f $(REPACK_GTA_DIR)/1ST_READ.BIN
 	cp 1ST_READ_PREBUILT.BIN $(REPACK_GTA_DIR)/1ST_READ.BIN
 	mkisofs -C 0,11702 -V $(PROJECT_NAME) -G IP.BIN -r -J -l -o $(PROJECT_NAME).iso $(REPACK_GTA_DIR)
@@ -241,15 +249,15 @@ $(PROJECT_NAME)-prebuilt.cdi: $(PROJECT_NAME)-prebuilt.iso
 	rm 1ST_READ_PREBUILT.BIN
 	@echo && echo && echo "*** CDI Baked Successfully ($@) ***" && echo && echo
 else
-$(PROJECT_NAME).cdi: $(TARGET) $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b
+$(PROJECT_NAME).cdi: $(TARGET) $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	mkdcdisc -e $(TARGET) -o $(PROJECT_NAME).cdi -d $(REPACK_GTA_DIR)/ $(MKDCDISC_PAD_OPTION) -n $(PROJECT_NAME) -a $(TEAM_NAME) -s $(DISC_SERIAL) -r $(RELEASE_DATE)
 	@echo && echo && echo "*** CDI Baked Successfully ($@) ***" && echo && echo
 
-$(PROJECT_NAME)-no-repack.cdi: $(TARGET) $(REPACK_GTA_DIR)/GTA3SF8.b
+$(PROJECT_NAME)-no-repack.cdi: $(TARGET) $(REPACK_GTA_DIR)/GTA3SF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	mkdcdisc -e $(TARGET) -o $(PROJECT_NAME)-no-repack.cdi -d $(REPACK_GTA_DIR)/ $(MKDCDISC_PAD_OPTION) -n $(PROJECT_NAME) -a $(TEAM_NAME) -s $(DISC_SERIAL) -r $(RELEASE_DATE)
 	@echo && echo && echo "*** CDI Baked Successfully ($@) ***" && echo && echo
 
-$(PROJECT_NAME)-prebuilt.cdi: $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b
+$(PROJECT_NAME)-prebuilt.cdi: $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTA3SF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	mkdcdisc -e $(TARGET) -o $(PROJECT_NAME).cdi -d $(REPACK_GTA_DIR)/ $(MKDCDISC_PAD_OPTION) -n $(PROJECT_NAME) -a $(TEAM_NAME) -s $(DISC_SERIAL) -r $(RELEASE_DATE)
 	@echo && echo && echo "*** CDI Baked Successfully ($@) ***" && echo && echo
 endif
@@ -310,7 +318,10 @@ streamheaderpack: ../src/tools/streamheaderpack.cpp
 
 TXD_OPTS_fonts = 256 256
 TXD_OPTS_hud = 128 128
-TXD_OPTS_menu = 512 512
+TXD_OPTS_menu = 512 512 \
+	--delete-tex mapTop03 --delete-tex mapTop02 --delete-tex mapTop01 \
+	--delete-tex mapMid03 --delete-tex mapMid02 --delete-tex mapMid01 \
+	--delete-tex mapBot03 --delete-tex mapBot02 --delete-tex mapBot01
 TXD_OPTS_LOADSC0 = 512 512
 TXD_OPTS_LOADSC1 = 512 512
 TXD_OPTS_LOADSC10 = 512 512

miami/Makefile

@@ -118,6 +118,7 @@ OBJS_NO_FAST_MATH = \
 	../src/miami/core/Cam.o \
 	../src/miami/core/Camera.o \
 	../src/miami/vehicles/Bike.o \
+	../src/miami/vehicles/Boat.o \
 	../src/miami/renderer/Particle.o
 
 KOS_CPPFLAGS += -fbuiltin -ffast-math -ffp-contract=fast \
@@ -182,7 +183,15 @@ $(TARGET): $(OBJS)
 run: $(TARGET)
 	$(KOS_LOADER) $(TARGET)
 
-$(REPACK_GTA_DIR)/GTAVCSF8.b: GTAVCSF8.b
+$(REPACK_GTA_DIR)/GTAVCSF8.b: assets/GTAVCSF8.b
+	mkdir -p $(@D)
+	cp $< $@
+
+$(REPACK_GTA_DIR)/0GDTEX.PVR: assets/0GDTEX.PVR
+	mkdir -p $(@D)
+	cp $< $@
+
+$(REPACK_GTA_DIR)/%.ico: assets/%.ico
 	mkdir -p $(@D)
 	cp $< $@
 
@@ -201,20 +210,20 @@ IP.BIN:
 	rm -f IP.BIN
 	$(KOS_BASE)/utils/makeip/makeip ip.txt IP.BIN
 
-$(PROJECT_NAME).iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b
+$(PROJECT_NAME).iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	rm -f $(PROJECT_NAME).iso
 	rm -f $(REPACK_GTA_DIR)/1ST_READ.BIN
 	cp 1ST_READ.BIN $(REPACK_GTA_DIR)
 	mkisofs -C 0,11702 -V $(PROJECT_NAME) -G IP.BIN -r -J -l -o $(PROJECT_NAME).iso $(REPACK_GTA_DIR)
 
-$(PROJECT_NAME)-no-repack.iso: IP.BIN 1ST_READ.BIN $(REPACK_GTA_DIR)/GTAVCSF8.b
+$(PROJECT_NAME)-no-repack.iso: IP.BIN 1ST_READ.BIN $(REPACK_GTA_DIR)/GTAVCSF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	rm -f $(PROJECT_NAME)-no-repack.iso
 	rm -f $(REPACK_GTA_DIR)/1ST_READ.BIN
 	cp 1ST_READ.BIN $(REPACK_GTA_DIR)
 	mkisofs -C 0,11702 -V $(PROJECT_NAME) -G IP.BIN -r -J -l -o $(PROJECT_NAME)-no-repack.iso $(REPACK_GTA_DIR)
 
 
-$(PROJECT_NAME).ds.iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b
+$(PROJECT_NAME).ds.iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	rm -f $(PROJECT_NAME).ds.iso
 	rm -f $(REPACK_GTA_DIR)/1ST_READ.BIN
 	cp $(TARGET).bin $(REPACK_GTA_DIR)/1ST_READ.BIN
@@ -225,7 +234,7 @@ $(PROJECT_NAME).ds.iso: IP.BIN 1ST_READ.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_
 	$(KOS_BASE)/utils/scramble/scramble $(TARGET)-prebuilt.bin 1ST_READ_PREBUILT.BIN
 	mkdir -p $(REPACK_GTA_DIR)
 
-$(PROJECT_NAME)-prebuilt.iso: IP.BIN 1ST_READ_PREBUILT.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b
+$(PROJECT_NAME)-prebuilt.iso: IP.BIN 1ST_READ_PREBUILT.BIN $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	rm -f $(REPACK_GTA_DIR)/1ST_READ.BIN
 	cp 1ST_READ_PREBUILT.BIN $(REPACK_GTA_DIR)/1ST_READ.BIN
 	mkisofs -C 0,11702 -V $(PROJECT_NAME) -G IP.BIN -r -J -l -o $(PROJECT_NAME).iso $(REPACK_GTA_DIR)
@@ -244,15 +253,15 @@ $(PROJECT_NAME)-prebuilt.cdi: $(PROJECT_NAME)-prebuilt.iso
 	rm 1ST_READ_PREBUILT.BIN
 	@echo && echo && echo "*** CDI Baked Successfully ($@) ***" && echo && echo
 else
-$(PROJECT_NAME).cdi: $(TARGET) $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b
+$(PROJECT_NAME).cdi: $(TARGET) $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	mkdcdisc -e $(TARGET) -o $(PROJECT_NAME).cdi -d $(REPACK_GTA_DIR)/ $(MKDCDISC_PAD_OPTION) -n $(PROJECT_NAME) -a $(TEAM_NAME) -s $(DISC_SERIAL) -r $(RELEASE_DATE)
 	@echo && echo && echo "*** CDI Baked Successfully ($@) ***" && echo && echo
 
-$(PROJECT_NAME)-no-repack.cdi: $(TARGET) $(REPACK_GTA_DIR)/GTAVCSF8.b
+$(PROJECT_NAME)-no-repack.cdi: $(TARGET) $(REPACK_GTA_DIR)/GTAVCSF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	mkdcdisc -e $(TARGET) -o $(PROJECT_NAME)-no-repack.cdi -d $(REPACK_GTA_DIR)/ $(MKDCDISC_PAD_OPTION) -n $(PROJECT_NAME) -a $(TEAM_NAME) -s $(DISC_SERIAL) -r $(RELEASE_DATE)
 	@echo && echo && echo "*** CDI Baked Successfully ($@) ***" && echo && echo
 
-$(PROJECT_NAME)-prebuilt.cdi: $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b
+$(PROJECT_NAME)-prebuilt.cdi: $(REPACK_DIR)/repacked $(REPACK_GTA_DIR)/GTAVCSF8.b $(REPACK_GTA_DIR)/0GDTEX.PVR $(REPACK_GTA_DIR)/settings.ico
 	mkdcdisc -e $(TARGET) -o $(PROJECT_NAME).cdi -d $(REPACK_GTA_DIR)/ $(MKDCDISC_PAD_OPTION) -n $(PROJECT_NAME) -a $(TEAM_NAME) -s $(DISC_SERIAL) -r $(RELEASE_DATE)
 	@echo && echo && echo "*** CDI Baked Successfully ($@) ***" && echo && echo
 endif

src/liberty/animation/AnimManager.cpp

@@ -11,6 +11,8 @@
 #include "AnimBlendAssocGroup.h"
 #include "AnimManager.h"
 
+void* re3StreamingAlloc(size_t size);
+
 CAnimBlock CAnimManager::ms_aAnimBlocks[NUMANIMBLOCKS];
 CAnimBlendHierarchy CAnimManager::ms_aAnimations[NUMANIMATIONS];
 int32 CAnimManager::ms_numAnimBlocks;
@@ -837,7 +839,7 @@ CAnimManager::LoadAnimFile(int fd, bool compress)
 			uint16_t flags;
 			CFileMgr::Read(fd, (char*)&flags, sizeof(flags));
 
-			seq->keyFrames = RwMalloc(dataSize);
+			seq->keyFrames = re3StreamingAlloc(dataSize);
 			assert(seq->keyFrames);
 			CFileMgr::Read(fd, (char*)seq->keyFrames, dataSize - sizeof(flags));
 			seq->type = flags;

src/liberty/audio/sampman.h

@@ -173,6 +173,7 @@ public:
 	char  GetCDAudioDriveLetter  (void);
 	
 	void UpdateEffectsVolume(void);
+	void UpdateStreamsVolume(void);
 
 #ifdef DC_SH4
 	void UpdateChannelVolume(uint32 nChannel);

src/liberty/audio/sampman_dc.cpp

@@ -11,6 +11,8 @@
 #include "crossplatform.h"
 
 #if !defined(AUDIO_OAL) &&  !defined(AUDIO_MSS)
+#define syncf(...) // dbglog(DBG_CRITICAL, __VA_ARGS__)
+#define streamf(...) // dbglog(DBG_CRITICAL, __VA_ARGS__)
 #define verbosef(...) // dbglog(DBG_CRITICAL, __VA_ARGS__)
 #define debugf(...) // dbglog(DBG_CRITICAL, __VA_ARGS__)
 
@@ -47,6 +49,8 @@
 /* Quick access to the AICA channels */
 #define AICA_CHANNEL(x)     (AICA_MEM_CHANNELS + (x) * sizeof(aica_channel_t))
 
+static uint32_t chn_version[64];
+
 int aica_play_chn(int chn, int size, uint32_t aica_buffer, int fmt, int vol, int pan, int loop, int freq) {
 	// assert(size <= 65534);
 	// We gotta fix this at some point
@@ -70,6 +74,7 @@ int aica_play_chn(int chn, int size, uint32_t aica_buffer, int fmt, int vol, int
     chan->freq = freq;
     chan->vol = vol;
     chan->pan = pan;
+	chan->version = ++chn_version[chn];
 	snd_sh4_to_aica(tmp, cmd->size);
     return chn;
 }
@@ -175,6 +180,12 @@ file_t fdPedSfx;
 volatile uint32 nPedSfxReqReadId = 1;
 volatile uint32 nPedSfxReqNextId = 1;
 
+// this is very wasteful and temporary
+#define BANK_STAGE_SIZE 16 * 2048
+static  uint8_t stagingBufferBank[BANK_STAGE_SIZE] __attribute__((aligned(32)));
+std::mutex stagingBufferMtx;
+
+
 static int32 DCStreamedLength[TOTAL_STREAMED_SOUNDS];
 
 struct WavHeader {
@@ -357,6 +368,13 @@ cSampleManager::Initialise(void)
 					if (channels[i].ch != -1) {
 						assert(channels[i].nSfx != -1);
 
+						uint32_t channel_version = g2_read_32(SPU_RAM_UNCACHED_BASE + AICA_CHANNEL(channels[i].ch) + offsetof(aica_channel_t, version));
+
+						if (chn_version[channels[i].ch] != channel_version) {
+							syncf("Stream version missmatch, skipping update. expected %d got %d\n", chn_version[channels[i].ch], channel_version);
+							continue;
+						}
+
 						uint16_t channel_pos = (g2_read_32(SPU_RAM_UNCACHED_BASE + AICA_CHANNEL(channels[i].ch) + offsetof(aica_channel_t, pos)) & 0xffff);
 						// verbosef("Channel %d pos: %d\n", i, channel_pos);
 						if (!channels[i].loop) {
@@ -383,21 +401,28 @@ cSampleManager::Initialise(void)
 				{
 					std::lock_guard<std::mutex> lk(streams[i].mtx);
 					if (streams[i].playing) {
+						uint32_t channel_version = g2_read_32(SPU_RAM_UNCACHED_BASE + AICA_CHANNEL(streams[i].mapped_ch[0]) + offsetof(aica_channel_t, version));
+
+						if (chn_version[streams[i].mapped_ch[0]] != channel_version) {
+							syncf("Stream version missmatch, skipping update. expected %d got %d\n", chn_version[streams[i].mapped_ch[0]], channel_version);
+							continue;
+						}
+
 						// get channel pos
 						uint32_t channel_pos = g2_read_32(SPU_RAM_UNCACHED_BASE + AICA_CHANNEL(streams[i].mapped_ch[0]) + offsetof(aica_channel_t, pos)) & 0xffff;
 						uint32_t logical_pos = channel_pos;
 						if (logical_pos > STREAM_CHANNEL_SAMPLE_COUNT/2) {
 							logical_pos -= STREAM_CHANNEL_SAMPLE_COUNT/2;
 						}
-						verbosef("Stream %d pos: %d, log: %d, rem: %d\n", i, channel_pos, logical_pos, streams[i].played_samples);
+						streamf("Stream %d pos: %d, log: %d, played: %d\n", i, channel_pos, logical_pos, streams[i].played_samples);
 			
-						bool can_refill = (streams[i].played_samples + STREAM_CHANNEL_SAMPLE_COUNT/2) < streams[i].total_samples;
-						bool can_fetch = (streams[i].played_samples + STREAM_CHANNEL_SAMPLE_COUNT/2 + STREAM_CHANNEL_SAMPLE_COUNT/2 + STREAM_CHANNEL_SAMPLE_COUNT/2) < streams[i].total_samples;
+						bool can_refill = (streams[i].played_samples + STREAM_CHANNEL_SAMPLE_COUNT/2 + (!streams[i].first_refill)*STREAM_CHANNEL_SAMPLE_COUNT/2) < streams[i].total_samples;
+						bool can_fetch = (streams[i].played_samples + STREAM_CHANNEL_SAMPLE_COUNT/2 + STREAM_CHANNEL_SAMPLE_COUNT/2 + (!streams[i].first_refill)*STREAM_CHANNEL_SAMPLE_COUNT/2) < streams[i].total_samples;
 						// copy over data if needed from staging
 						if (channel_pos >= STREAM_CHANNEL_SAMPLE_COUNT/2 && !streams[i].next_is_upper_half) {
 							streams[i].next_is_upper_half = true;
 							if (can_refill) { // could we need a refill?
-								verbosef("Filling channel %d with lower half\n", i);
+								streamf("Filling channel %d with lower half\n", i);
 								// fill lower half
 								spu_memload(streams[i].aica_buffers[0], streams[i].buffer, STREAM_CHANNEL_BUFFER_SIZE/2);
 								if (streams[i].stereo) {
@@ -413,7 +438,7 @@ cSampleManager::Initialise(void)
 						} else if (channel_pos < STREAM_CHANNEL_SAMPLE_COUNT/2 && streams[i].next_is_upper_half) {
 							streams[i].next_is_upper_half = false;
 							if (can_refill) { // could we need a refill?
-								verbosef("Filling channel %d with upper half\n", i);
+								streamf("Filling channel %d with upper half\n", i);
 								// fill upper half
 								spu_memload(streams[i].aica_buffers[0] + STREAM_CHANNEL_BUFFER_SIZE/2, streams[i].buffer, STREAM_CHANNEL_BUFFER_SIZE/2);
 								if (streams[i].stereo) {
@@ -433,7 +458,7 @@ cSampleManager::Initialise(void)
 						// if end of file, stop
 						if ((streams[i].played_samples + logical_pos) > streams[i].total_samples) {
 							// stop channel
-							debugf("Auto stopping stream: %d -> {%d, %d}, %d total\n", i, streams[i].mapped_ch[0], streams[i].mapped_ch[1], streams[i].total_samples);
+							streamf("Auto stopping stream: %d -> {%d, %d}, %d total\n", i, streams[i].mapped_ch[0], streams[i].mapped_ch[1], streams[i].total_samples);
 							aica_stop_chn(streams[i].mapped_ch[0]);
 							aica_stop_chn(streams[i].mapped_ch[1]);
 							streams[i].playing = false;
@@ -441,7 +466,7 @@ cSampleManager::Initialise(void)
 					}
 					
 					if (do_read) {
-						debugf("Queueing stream read: %d, file: %d, buffer: %p, size: %d, tell: %d\n", i, streams[i].fd, streams[i].buffer, do_read, fs_tell(streams[i].fd));
+						streamf("Queueing stream read: %d, file: %d, buffer: %p, size: %d, file_offset: %d\n", i, streams[i].fd, streams[i].buffer, do_read, streams[i].file_offset);
 						CdStreamQueueAudioRead(streams[i].fd, streams[i].buffer, do_read, streams[i].file_offset);
 						streams[i].file_offset += do_read;
 					}
@@ -481,6 +506,7 @@ cSampleManager::Initialise(void)
 void
 cSampleManager::Terminate(void)
 {
+	CdStreamDiscardAudioRead(fdPedSfx);
 	fs_close(fdPedSfx);
 }
 
@@ -519,6 +545,7 @@ void
 cSampleManager::SetMusicMasterVolume(uint8 nVolume)
 {
 	m_nMusicVolume = nVolume;
+	UpdateStreamsVolume();
 }
 
 void
@@ -532,6 +559,7 @@ void
 cSampleManager::SetMusicFadeVolume(uint8 nVolume)
 {
 	m_nMusicFadeVolume = nVolume;
+	UpdateStreamsVolume();
 }
 
 void
@@ -568,26 +596,29 @@ cSampleManager::LoadSampleBank(uint8 nBank)
 		// TODO: Split per-bank sfx file
 		int fd = fs_open(SampleBankDataFilename, O_RDONLY);
 		assert(fd >= 0);
-		// this is very wasteful and temporary
-		void* stagingBuffer = memalign(32, 32 * 2048);
-		assert(stagingBuffer != 0);
+		
+		
+		{
+			std::lock_guard lk(stagingBufferMtx); // for stagingBufferBank
+		
+			void* stagingBuffer = stagingBufferBank;
 
-		// Ideally, we'd suspend the CdStream thingy here or read via that instead
-		uintptr_t loadOffset = bank.base;
-		fs_seek(fd, fileStart, SEEK_SET);
+			// Ideally, we'd suspend the CdStream thingy here or read via that instead
+			uintptr_t loadOffset = bank.base;
+			fs_seek(fd, fileStart, SEEK_SET);
 
-		while (fileSize > 0) {
-			size_t readSize = fileSize > 32 * 2048 ? 32 * 2048 : fileSize;
-			int rs = fs_read(fd, stagingBuffer, readSize);
-			debugf("Read %d bytes, expected %d\n", rs, readSize);
-			assert(rs == readSize);
-			spu_memload(loadOffset, stagingBuffer, readSize);
-			loadOffset += readSize;
-			fileSize -= readSize;
-			debugf("Loaded %d bytes, %d remaining\n", readSize, fileSize);
+			while (fileSize > 0) {
+				size_t readSize = fileSize > sizeof(stagingBufferBank) ? sizeof(stagingBufferBank) : fileSize;
+				int rs = fs_read(fd, stagingBuffer, readSize);
+				debugf("Read %d bytes, expected %d\n", rs, readSize);
+				assert(rs == readSize);
+				spu_memload(loadOffset, stagingBuffer, readSize);
+				loadOffset += readSize;
+				fileSize -= readSize;
+				debugf("Loaded %d bytes, %d remaining\n", readSize, fileSize);
+			}
 		}
 		fs_close(fd);
-		free(stagingBuffer);
 		
 
 		for (int nSfx = BankStartOffset[nBank]; nSfx < BankStartOffset[nBank+1]; nSfx++) {
@@ -728,23 +759,27 @@ cSampleManager::LoadPedComment(uint32 nComment)
 
 	assert(m_aSamples[nComment].nByteSize <= PED_BLOCKSIZE_ADPCM);
 
-	CdStreamQueueAudioRead(nComment, (void*)nPedSlotSfxAddr[nCurrentPedSlot], m_aSamples[nComment].nByteSize, m_aSamples[nComment].nFileOffset, [](AudioReadCmd* cmd) {
+	CdStreamQueueAudioRead(fdPedSfx, (void*)nPedSlotSfxAddr[nCurrentPedSlot], m_aSamples[nComment].nByteSize, m_aSamples[nComment].nFileOffset, [](AudioReadCmd* cmd) {
 		debugf("Loading ped comment %d, offset: %d, size: %d\n", nComment, m_aSamples[nComment].nFileOffset, m_aSamples[nComment].nByteSize);
-		fs_seek(fdPedSfx, cmd->seek, SEEK_SET);
+		assert(fs_seek(fdPedSfx, cmd->seek, SEEK_SET) == cmd->seek);
 
 
 		// TODO: When we can dma directly to AICA, we can use this instead
 		// fs_read(fdPedSfx, SPU_BASE_U8 + (uintptr_t)cmd->dest, cmd->size);
 
-		void* stagingBuffer = memalign(32, cmd->size);
-		assert(stagingBuffer != 0);
-		debugf("Allocated %d bytes at %p\n", cmd->size, stagingBuffer);
-		int rs = fs_read(fdPedSfx, stagingBuffer, cmd->size);
-		debugf("Read %d bytes, expected %d\n", rs, cmd->size);
-		assert(rs == cmd->size);
-
-		spu_memload((uintptr_t)cmd->dest, stagingBuffer, cmd->size);
-		free(stagingBuffer);
+		assert(cmd->size < sizeof(stagingBufferBank));
+		{
+			std::lock_guard lk(stagingBufferMtx); // for stagingBufferBank
+			void* stagingBuffer = stagingBufferBank;
+			assert(stagingBuffer != 0);
+			debugf("Allocated %d bytes at %p\n", cmd->size, stagingBuffer);
+			int rs = fs_read(fdPedSfx, stagingBuffer, cmd->size);
+			debugf("Read %d bytes, expected %d\n", rs, cmd->size);
+			assert(rs == cmd->size);
+	
+			spu_memload((uintptr_t)cmd->dest, stagingBuffer, cmd->size);
+		}
+		
 		nPedSfxReqReadId = nPedSfxReqReadId + 1;
 	});
 
@@ -1014,7 +1049,7 @@ cSampleManager::PreloadStreamedFile(uint8 nFile, uint8 nStream, uint32_t seek_by
 {
 	ASSERT( nStream < MAX_STREAMS );
 	file_t f = fs_open(DCStreamedNameTable[nFile], O_RDONLY);
-	debugf("PreloadStreamedFile(%p, %d, %d) is %s\n", f, nFile, nStream, DCStreamedNameTable[nFile]);
+	streamf("PreloadStreamedFile(%p, %d, %d) is %s\n", f, nFile, nStream, DCStreamedNameTable[nFile]);
 	assert(f >= 0 );
 	WavHeader hdr;
 	assert(fs_read(f, &hdr, sizeof(hdr)) == sizeof(hdr));
@@ -1037,13 +1072,13 @@ cSampleManager::PreloadStreamedFile(uint8 nFile, uint8 nStream, uint32_t seek_by
 		streams[nStream].next_is_upper_half = true;
 		streams[nStream].first_refill = true;
 
-		debugf("PreloadStreamedFile: %s: stream: %d, freq: %d, chans: %d, byte size: %d, played samples: %d\n", DCStreamedNameTable[nFile], nStream, hdr.samplesPerSec, hdr.numOfChan, hdr.dataSize, streams[nStream].played_samples);
+		streamf("PreloadStreamedFile: %s: stream: %d, freq: %d, chans: %d, byte size: %d, played samples: %d\n", DCStreamedNameTable[nFile], nStream, hdr.samplesPerSec, hdr.numOfChan, hdr.dataSize, streams[nStream].played_samples);
 	
 
 		// How to avoid the lock?
 		if (seek_bytes_aligned) {
 			streams[nStream].played_samples = seek_bytes_aligned * (streams[nStream].stereo ? 1 : 2);
-			debugf("Seeking aligned to: %d, played_samples: %d\n", seek_bytes_aligned, streams[nStream].played_samples);
+			streamf("Seeking aligned to: %d, played_samples: %d\n", seek_bytes_aligned, streams[nStream].played_samples);
 			fs_seek(streams[nStream].fd, 2048 + seek_bytes_aligned, SEEK_SET);
 		} else {
 			fs_seek(f, 2048, SEEK_SET);
@@ -1072,7 +1107,7 @@ cSampleManager::PreloadStreamedFile(uint8 nFile, uint8 nStream, uint32_t seek_by
 		streams[nStream].file_offset = fs_tell(f);
 	}
 
-	verbosef("PreloadStreamedFile: %p %d - %s, %d, %d, \n", f, nFile, DCStreamedNameTable[nFile], streams[nStream].rate, streams[nStream].stereo);
+	streamf("PreloadStreamedFile: %p %d - %s, %d, %d, \n", f, nFile, DCStreamedNameTable[nFile], streams[nStream].rate, streams[nStream].stereo);
 }
 
 // we can't really pause a stream, so we just make it go very slow with zero volume
@@ -1195,12 +1230,29 @@ cSampleManager::GetStreamedFilePosition(uint8 nStream)
 	return rv;
 }
 
+
+static uint8 nStreamVolumes[MAX_STREAMS];
+static uint8 nStreamPans[MAX_STREAMS];
+static uint8 nStreamEffect[MAX_STREAMS];
+void
+cSampleManager::UpdateStreamsVolume(void)
+{
+	if(_bSampmanInitialised) {
+		for (int nStream = 0; nStream < MAX_STREAMS; nStream++) {
+			SetStreamedVolumeAndPan(nStreamVolumes[nStream], nStreamPans[nStream], nStreamEffect[nStream], nStream);
+		}
+	}
+}
+
 void
 cSampleManager::SetStreamedVolumeAndPan(uint8 nVolume, uint8 nPan, uint8 nEffectFlag, uint8 nStream)
 {
 	ASSERT( nStream < MAX_STREAMS );
 	if (nVolume > MAX_VOLUME)
 		nVolume = MAX_VOLUME;
+	nStreamVolumes[nStream] = nVolume;
+	nStreamPans[nStream] = nPan;
+	nStreamEffect[nStream] = nEffectFlag;
 	nVolume = linearlize_volume(nVolume); //nVolume * 255 / MAX_VOLUME;
 	nVolume = m_nMusicFadeVolume * nVolume * m_nMusicVolume >> 14;
 	if (streams[nStream].vol != nVolume || streams[nStream].nPan != nPan) {
@@ -1268,6 +1320,8 @@ cSampleManager::InitialiseSampleBanks(void)
 		assert(m_aSamples[nComment].nByteSize <= PED_BLOCKSIZE_ADPCM);
 	}
 
+	assert(PED_BLOCKSIZE_ADPCM <= BANK_STAGE_SIZE);
+
 	LoadSampleBank(SFX_BANK_0);
 	
 	return TRUE;

src/liberty/collision/ColLine.h

@@ -1,6 +1,6 @@
 #pragma once
 
-struct CColLine
+struct alignas(8) CColLine
 {
 	// NB: this has to be compatible with two CVuVectors
 	CVector p0;

src/liberty/collision/ColModel.cpp

@@ -2,6 +2,9 @@
 #include "ColModel.h"
 #include "Game.h"
 #include "MemoryHeap.h"
+#include "Collision.h"
+
+void* re3StreamingAlloc(size_t size);
 
 CColModel::CColModel(void)
 {
@@ -22,12 +25,12 @@ CColModel::CColModel(void)
 CColModel::~CColModel(void)
 {
 	RemoveCollisionVolumes();
-	RemoveTrianglePlanes();
 }
 
 void
 CColModel::RemoveCollisionVolumes(void)
 {
+	CCollision::RemoveTrianglePlanes(this);
 	if(ownsCollisionVolumes){
 		RwFree(spheres);
 		RwFree(lines);
@@ -93,6 +96,8 @@ CColModel::operator=(const CColModel &other)
 	int i;
 	int numVerts;
 
+	CCollision::RemoveTrianglePlanes(this);
+
 	boundingSphere = other.boundingSphere;
 	boundingBox = other.boundingBox;
 
@@ -163,7 +168,7 @@ CColModel::operator=(const CColModel &other)
 		if(vertices)
 			RwFree(vertices);
 		if(numVerts){
-			vertices = (CompressedVector*)RwMalloc(numVerts*sizeof(CompressedVector));
+			vertices = (CompressedVector*)re3StreamingAlloc(numVerts*sizeof(CompressedVector));
 			for(i = 0; i < numVerts; i++)
 				vertices[i] = other.vertices[i];
 		}
@@ -173,7 +178,7 @@ CColModel::operator=(const CColModel &other)
 			numTriangles = other.numTriangles;
 			if(triangles)
 				RwFree(triangles);
-			triangles = (CColTriangle*)RwMalloc(numTriangles*sizeof(CColTriangle));
+			triangles = (CColTriangle*)re3StreamingAlloc(numTriangles*sizeof(CColTriangle));
 		}
 		for(i = 0; i < numTriangles; i++)
 			triangles[i] = other.triangles[i];

src/liberty/collision/ColPoint.h

@@ -1,6 +1,6 @@
 #pragma once
 
-struct CColPoint
+struct alignas(8) CColPoint
 {
 	CVector point;
 	int pad1;

src/liberty/collision/ColSphere.h

@@ -2,12 +2,17 @@
 
 #include "SurfaceTable.h"
 
-struct CColSphere
+struct alignas(8) CColSphere
 {
 	// NB: this has to be compatible with a CVuVector
 	CVector center;
 	float radius;
 	uint8 surface;
 	uint8 piece;
+	void Set(float radius, uint8 surf = SURFACE_DEFAULT, uint8 piece = 0) {
+		this->radius = radius;
+		this->surface = surf;
+		this->piece = piece;
+	}
 	void Set(float radius, const CVector &center, uint8 surf = SURFACE_DEFAULT, uint8 piece = 0);
 };

src/liberty/collision/Collision.cpp

@@ -23,6 +23,10 @@
 #include "Collision.h"
 #include "Frontend.h"
 
+#ifdef DC_SH4
+#include "VuCollision.h"
+#endif
+
 #ifdef VU_COLLISION
 #include "VuCollision.h"
 
@@ -737,7 +741,12 @@ CCollision::TestLineOfSight(const CColLine &line, const CMatrix &matrix, CColMod
 
 	// transform line to model space
 	Invert(matrix, matTransform);
-	CColLine newline(matTransform * line.p0, matTransform * line.p1);
+	CColLine newline;
+#ifndef DC_SH4
+	newline.Set(matTransform * line.p0, matTransform * line.p1);
+#else
+	TransformPoints(reinterpret_cast<CVuVector*>(&newline), 2, matTransform, &line.p0, sizeof(CColLine)/2);
+#endif
 
 	// If we don't intersect with the bounding box, no chance on the rest
 	if(!TestLineBox(newline, model.boundingBox))
@@ -1577,7 +1586,12 @@ CCollision::ProcessLineOfSight(const CColLine &line,
 
 	// transform line to model space
 	Invert(matrix, matTransform);
-	CColLine newline(matTransform * line.p0, matTransform * line.p1);
+	CColLine newline;
+#ifdef DC_SH4
+	TransformPoints(reinterpret_cast<CVuVector*>(&newline), 2, matTransform, &line.p0, sizeof(CColLine)/2);
+#else
+	newline.Set(matTransform * line.p0, matTransform * line.p1);
+#endif
 
 	// If we don't intersect with the bounding box, no chance on the rest
 	if(!TestLineBox(newline, model.boundingBox))
@@ -1601,8 +1615,18 @@ CCollision::ProcessLineOfSight(const CColLine &line,
 	}
 
 	if(coldist < mindist){
+#ifndef DC_SH4
 		point.point = matrix * point.point;
 		point.normal = Multiply3x3(matrix, point.normal);
+#else
+		mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&matrix)));
+		mat_trans_single3_nodiv(point.point.x,
+		                        point.point.y,
+		                        point.point.z);
+		mat_trans_normal3(point.normal.x,
+		                  point.normal.y,
+		                  point.normal.z);
+#endif
 		mindist = coldist;
 		return true;
 	}
@@ -1739,9 +1763,14 @@ CCollision::ProcessVerticalLine(const CColLine &line,
 
 	// transform line to model space
 	// Why does the game seem to do this differently than above?
-	CColLine newline(MultiplyInverse(matrix, line.p0), MultiplyInverse(matrix, line.p1));
-	newline.p1.x = newline.p0.x;
-	newline.p1.y = newline.p0.y;
+	CMatrix matTransform;
+	Invert(matrix, matTransform);
+	CColLine newline;
+#ifndef DC_SH4
+	newline.Set(matTransform * line.p0, matTransform * line.p1);
+#else
+	TransformPoints(reinterpret_cast<CVuVector*>(&newline), 2, matTransform, &line.p0, sizeof(CColLine)/2);
+#endif
 
 	if(!TestVerticalLineBox(newline, model.boundingBox))
 		return false;
@@ -1765,14 +1794,31 @@ CCollision::ProcessVerticalLine(const CColLine &line,
 	}
 
 	if(coldist < mindist){
+#ifndef DC_SH4
 		point.point = matrix * point.point;
 		point.normal = Multiply3x3(matrix, point.normal);
+#else
+		mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&matrix)));
+		mat_trans_single3_nodiv(point.point.x,
+		                        point.point.y,
+		                        point.point.z);
+		mat_trans_normal3(point.normal.x,
+		                  point.normal.y,
+		                  point.normal.z);
+#endif
 		if(TempStoredPoly.valid && poly){
 			*poly = TempStoredPoly;
+#ifndef DC_SH4
 			poly->verts[0] = matrix * poly->verts[0];
 			poly->verts[1] = matrix * poly->verts[1];
 			poly->verts[2] = matrix * poly->verts[2];
+#else
+			mat_trans_single3_nodiv(poly->verts[0].x, poly->verts[0].y, poly->verts[0].z);
+			mat_trans_single3_nodiv(poly->verts[1].x, poly->verts[1].y, poly->verts[1].z);
+			mat_trans_single3_nodiv(poly->verts[2].x, poly->verts[2].y, poly->verts[2].z);
+#endif
 		}
+
 		mindist = coldist;
 		return true;
 	}
@@ -2123,25 +2169,60 @@ CCollision::ProcessColModels(const CMatrix &matrixA, CColModel &modelA,
 	assert(modelA.numLines <= MAXNUMLINES);
 
 	// From model A space to model B space
-	matAB = Invert(matrixB, matAB);
+	Invert(matrixB, matAB);
+#ifndef DC_SH4
 	matAB *= matrixA;
+#else
+	mat_load(reinterpret_cast<const matrix_t*>(&matAB));
+	mat_apply(reinterpret_cast<const matrix_t*>(&matrixA));
+#endif
 
 	CColSphere bsphereAB;	// bounding sphere of A in B space
 	bsphereAB.radius = modelA.boundingSphere.radius;
+#ifndef DC_SH4
 	bsphereAB.center = matAB * modelA.boundingSphere.center;
+#else
+	mat_trans_single3_nodiv_nomod(modelA.boundingSphere.center.x,
+	                              modelA.boundingSphere.center.y,
+	                              modelA.boundingSphere.center.z,
+	                              bsphereAB.center.x,
+	                              bsphereAB.center.y,
+	                              bsphereAB.center.z);
+#endif
 	if(!TestSphereBox(bsphereAB, modelB.boundingBox))
 		return 0;
-	// B to A space
-	matBA = Invert(matrixA, matBA);
-	matBA *= matrixB;
 
 	// transform modelA's spheres and lines to B space
 	for(i = 0; i < modelA.numSpheres; i++){
 		CColSphere &s = modelA.spheres[i];
+#ifndef DC_SH4
 		aSpheresA[i].Set(s.radius, matAB * s.center, s.surface, s.piece);
+#else
+		auto &d = aSpheresA[i];
+		mat_trans_single3_nodiv_nomod(s.center.x, s.center.y, s.center.z,
+		                              d.center.x, d.center.y, d.center.z);
+		d.Set(s.radius, s.surface, s.piece);
+#endif
 	}
-	for(i = 0; i < modelA.numLines; i++)
+
+	for(i = 0; i < modelA.numLines; i++) {
+#ifndef DC_SH4
 		aLinesA[i].Set(matAB * modelA.lines[i].p0, matAB * modelA.lines[i].p1);
+#else
+		mat_trans_single3_nodiv_nomod(modelA.lines[i].p0.x,
+		                              modelA.lines[i].p0.y,
+		                              modelA.lines[i].p0.z,
+		                              aLinesA[i].p0.x,
+		                              aLinesA[i].p0.y,
+		                              aLinesA[i].p0.z);
+		mat_trans_single3_nodiv_nomod(modelA.lines[i].p1.x,
+		                              modelA.lines[i].p1.y,
+		                              modelA.lines[i].p1.z,
+		                              aLinesA[i].p1.x,
+		                              aLinesA[i].p1.y,
+		                              aLinesA[i].p1.z);
+#endif
+	}
 
 	// Test them against model B's bounding volumes
 	int numSpheresA = 0;
@@ -2160,9 +2241,26 @@ CCollision::ProcessColModels(const CMatrix &matrixA, CColModel &modelA,
 	int numSpheresB = 0;
 	int numBoxesB = 0;
 	int numTrianglesB = 0;
+	// B to A space
+	Invert(matrixA, matBA);
+#ifndef DC_SH4
+	matBA *= matrixB;
+#else
+	mat_load(reinterpret_cast<const matrix_t*>(&matBA));
+	mat_apply(reinterpret_cast<const matrix_t*>(&matrixB));
+#endif
 	for(i = 0; i < modelB.numSpheres; i++){
 		s.radius = modelB.spheres[i].radius;
+#ifndef DC_SH4
 		s.center = matBA * modelB.spheres[i].center;
+#else
+		mat_trans_single3_nodiv_nomod(modelB.spheres[i].center.x,
+		                              modelB.spheres[i].center.y,
+		                              modelB.spheres[i].center.z,
+		                              s.center.x,
+		                              s.center.y,
+		                              s.center.z);
+#endif
 		if(TestSphereBox(s, modelA.boundingBox))
 			aSphereIndicesB[numSpheresB++] = i;
 	}
@@ -2209,9 +2307,22 @@ CCollision::ProcessColModels(const CMatrix &matrixA, CColModel &modelA,
 		if(hasCollided)
 			numCollisions++;
 	}
+
+#ifdef DC_SH4
+	mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&matrixB)));
+#endif
 	for(i = 0; i < numCollisions; i++){
+#ifndef DC_SH4
 		spherepoints[i].point = matrixB * spherepoints[i].point;
 		spherepoints[i].normal = Multiply3x3(matrixB, spherepoints[i].normal);
+#else
+		mat_trans_single3_nodiv(spherepoints[i].point.x,
+		                        spherepoints[i].point.y,
+		                        spherepoints[i].point.z);
+		mat_trans_normal3(spherepoints[i].normal.x,
+		                  spherepoints[i].normal.y,
+		                  spherepoints[i].normal.z);
+#endif
 	}
 
 	// And the same thing for the lines in A
@@ -2242,8 +2353,17 @@ CCollision::ProcessColModels(const CMatrix &matrixA, CColModel &modelA,
 	for(i = 0; i < numLinesA; i++)
 		if(aCollided[i]){
 			j = aLineIndicesA[i];
+#ifndef DC_SH4
 			linepoints[j].point = matrixB * linepoints[j].point;
 			linepoints[j].normal = Multiply3x3(matrixB, linepoints[j].normal);
+#else
+			mat_trans_single3_nodiv(linepoints[j].point.x,
+			                        linepoints[j].point.y,
+			                        linepoints[j].point.z);
+			mat_trans_normal3(linepoints[j].normal.x,
+			                  linepoints[j].normal.y,
+			                  linepoints[j].normal.z);
+#endif
 		}
 
 	return numCollisions;	// sphere collisions
@@ -2287,6 +2407,15 @@ CCollision::DistToLine(const CVector *l0, const CVector *l1, const CVector *poin
 	return (*point - closest).Magnitude();
 }
 
+void
+CCollision::RemoveTrianglePlanes(CColModel *model)
+{
+	if(model->trianglePlanes){
+		ms_colModelCache.Remove(model->GetLinkPtr());
+		model->RemoveTrianglePlanes();
+	}
+}
+
 void
 CCollision::CalculateTrianglePlanes(CColModel *model)
 {

src/liberty/collision/Collision.h

@@ -41,6 +41,7 @@ public:
 	static void DrawColModel(const CMatrix &mat, const CColModel &colModel);
 	static void DrawColModel_Coloured(const CMatrix &mat, const CColModel &colModel, int32 id);
 
+	static void RemoveTrianglePlanes(CColModel *model);
 	static void CalculateTrianglePlanes(CColModel *model);
 
 	// all these return true if there's a collision

src/liberty/core/CdStreamDC.cpp

@@ -479,6 +479,7 @@ RemoveFirstInQueue(Queue *queue)
 }
 
 std::vector<AudioReadCmd> pendingAudioReads;
+volatile int pendingAudioRead_fd = -1;
 #if !defined(DC_SH4)
 std::mutex pendingAudioReadsMutex;
 #endif
@@ -487,8 +488,9 @@ void CdStreamQueueAudioRead(int fd, void* pBuffer, size_t bytes, size_t seek, st
 	AudioReadCmd cmd = { pBuffer, fd, bytes, seek};
 	if (!callback) {
 		cmd.callback = [](AudioReadCmd* cmd){
-			lseek(cmd->fd, cmd->seek, SEEK_SET);
-			read(cmd->fd, cmd->dest, cmd->size);
+			assert(pendingAudioRead_fd == cmd->fd);
+			assert(lseek(cmd->fd, cmd->seek, SEEK_SET) == cmd->seek);
+			assert(read(cmd->fd, cmd->dest, cmd->size) == cmd->size);
 		};
 	} else {
 		cmd.callback = callback;
@@ -500,7 +502,7 @@ void CdStreamQueueAudioRead(int fd, void* pBuffer, size_t bytes, size_t seek, st
 		auto mask = irq_disable();
 		#endif
 		for (auto it = pendingAudioReads.rbegin(); it != pendingAudioReads.rend(); ++it) {
-			if (it->fd == -1 || it->fd == fd) {
+			if (it->fd == -1) {
 				*it = cmd;
 				goto out;
 			}
@@ -532,6 +534,12 @@ void CdStreamDiscardAudioRead(int fd) {
 	#if defined(DC_SH4)
 	irq_restore(mask);
 	#endif
+
+	while (pendingAudioRead_fd == fd) {
+		#if defined(DC_SH4)
+		thd_pass();
+		#endif
+	}
 }
 
 AudioReadCmd CdStreamNextAudioRead() {
@@ -548,6 +556,8 @@ AudioReadCmd CdStreamNextAudioRead() {
 			break;
 		}
 	}
+	assert(pendingAudioRead_fd == -1);
+	pendingAudioRead_fd = cmd.fd;
 	#if defined(DC_SH4)
 	irq_restore(mask);
 	#endif
@@ -568,6 +578,7 @@ int read_loop(int fd, void* pBuffer, size_t bytes) {
 		auto cmd = CdStreamNextAudioRead();
 		while (cmd.fd != -1) {
 			cmd.callback(&cmd);
+			pendingAudioRead_fd = -1;
 			cmd = CdStreamNextAudioRead();
 		}
 	}
@@ -584,6 +595,7 @@ void *CdStreamThread(void *param)
 		auto cmd = CdStreamNextAudioRead();
 		while (cmd.fd != -1) {
 			cmd.callback(&cmd);
+			pendingAudioRead_fd = -1;
 			cmd = CdStreamNextAudioRead();
 		}
 

src/liberty/core/FileLoader.cpp

@@ -28,6 +28,8 @@
 
 #include <kos/dbglog.h>
 
+void* re3StreamingAlloc(size_t size);
+
 char CFileLoader::ms_line[256];
 
 const char*
@@ -221,7 +223,7 @@ CFileLoader::LoadCollisionFile(const char *filename)
 
 		mi = CModelInfo::GetModelInfo(modelname, nil);
 		if(mi){
-			if(mi->GetColModel()){
+			if(mi->GetColModel() && mi->DoesOwnColModel()){
 				LoadCollisionModel(work_buff+24, *mi->GetColModel(), modelname);
 			}else{
 				CColModel *model = new CColModel;
@@ -255,6 +257,24 @@ CFileLoader::LoadCollisionModel(uint8 *buf, CColModel &model, char *modelname)
 	model.boundingBox.max.z = *(float*)(buf+36);
 	model.numSpheres = *(int16*)(buf+40);
 	buf += 44;
+	if (model.spheres) {
+		RwFree(model.spheres);
+	}
+	if (model.lines) {
+		RwFree(model.lines);
+	}
+	if (model.boxes) {
+		RwFree(model.boxes);
+	}
+	if (model.vertices) {
+		RwFree(model.vertices);
+	}
+	if (model.triangles) {
+		RwFree(model.triangles);
+	}
+	if (model.trianglePlanes) {
+		CCollision::RemoveTrianglePlanes(&model);
+	}
 	if(model.numSpheres > 0){
 		model.spheres = (CColSphere*)RwMalloc(model.numSpheres*sizeof(CColSphere));
 		REGISTER_MEMPTR(&model.spheres);
@@ -292,7 +312,7 @@ CFileLoader::LoadCollisionModel(uint8 *buf, CColModel &model, char *modelname)
 	int32 numVertices = *(int16*)buf;
 	buf += 4;
 	if(numVertices > 0){
-		model.vertices = (CompressedVector*)RwMalloc(numVertices*sizeof(CompressedVector));
+		model.vertices = (CompressedVector*)re3StreamingAlloc(numVertices*sizeof(CompressedVector));
 		REGISTER_MEMPTR(&model.vertices);
 		for(i = 0; i < numVertices; i++){
 			model.vertices[i].SetFixed(*(int16*)buf, *(int16*)(buf+2), *(int16*)(buf+4));
@@ -304,7 +324,7 @@ CFileLoader::LoadCollisionModel(uint8 *buf, CColModel &model, char *modelname)
 	model.numTriangles = *(int16*)buf;
 	buf += 4;
 	if(model.numTriangles > 0){
-		model.triangles = (CColTriangle*)RwMalloc(model.numTriangles*sizeof(CColTriangle));
+		model.triangles = (CColTriangle*)re3StreamingAlloc(model.numTriangles*sizeof(CColTriangle));
 		REGISTER_MEMPTR(&model.triangles);
 		for(i = 0; i < model.numTriangles; i++){
 			model.triangles[i].Set(model.vertices, *(uint16*)buf, *(uint16*)(buf+2), *(uint16*)(buf+4), buf[6], buf[7]);
@@ -672,118 +692,383 @@ char *DoubleSidedNames[] = {
 	"sbwy_tunl_cstm2",
 	"sbwy_tunl_cstm1",
 	"tenmnt6ad",
+	"LUIGICLUBout",
+	"redlightbuild03",
+	"dinerind",
+	"dinerind02",
+	"franksclb02",
+	"glassfx1",
+	"lswank_inside",
+	"luigiineerclub",
+	"swank_inside",
+	"train_rust",
+    "glassfx2",
+    "glassfx3",
+    "glassfx4",
+    "glassfx55",
+    "glassfx_composh",
+    "glassfxsub1",
+    "glassfxsub2",
 	""
 
 };
 char *TreeNames[] = {
-	"coast_treepatch",
-	"comparknewtrees",
-	"comtreepatchprk",
-	"condotree01",
-	"condotree1",
-	"indatree03",
-	"indtreepatch5",
-	"indtreepatch06f",
-	"new_carprktrees",
-	"new_carprktrees4",
-	"newcoasttrees1",
-	"newcoasttrees2",
-	"newcoasttrees3",
-	"newtreepatch_sub",
-	"newtrees1_sub",
-	"newunitrepatch",
-	"pinetree_narrow",
-	"pinetree_wide",
-	"treencom2",
-	"treepatch",
-	"treepatch01_sub",
-	"treepatch02_sub",
-	"treepatch2",
-	"treepatch2b",
-	"treepatch03",
-	"treepatch03_sub",
-	"treepatch04_sub",
-	"treepatch05_sub",
-	"treepatch06_sub",
-	"treepatch07_sub",
-	"treepatch08_sub",
-	"treepatch09_sub",
-	"treepatch10_sub",
-	"treepatch11_sub",
-	"treepatch12_sub",
-	"treepatch13_sub",
-	"treepatch14_sub",
-	"treepatch15_sub",
-	"treepatch16_sub",
-	"treepatch17_sub",
-	"treepatch18_sub",
-	"treepatch19_sub",
-	"treepatch20_sub",
-	"treepatch21_sub",
-	"treepatch22_sub",
-	"treepatch23_sub",
-	"treepatch24_sub",
-	"treepatch25_sub",
-	"treepatch26_sub",
-	"treepatch27_sub",
-	"treepatch28_sub",
-	"treepatch29_sub",
-	"treepatch30_sub",
-	"treepatch31_sub",
-	"treepatch32_sub",
-	"treepatch33_sub",
-	"treepatch34_sub",
-	"treepatch35_sub",
-	"treepatch69",
-	"treepatch152_sub",
-	"treepatch153_sub",
-	"treepatch171_sub",
-	"treepatch172_sub",
-	"treepatch173_sub",
-	"treepatch212_sub",
-	"treepatch213_sub",
-	"treepatch214_sub",
-	"treepatcha",
-	"treepatchb",
-	"treepatchcomtop1",
-	"treepatchd",
-	"treepatche",
-	"treepatchh",
-	"treepatchindaa2",
-	"treepatchindnew",
-	"treepatchindnew2",
-	"treepatchk",
-	"treepatchkb4",
-	"treepatchkb5",
-	"treepatchkb6",
-	"treepatchkb7",
-	"treepatchkb9",
-	"treepatchl",
-	"treepatchm",
-	"treepatchnew_sub",
-	"treepatchttwrs",
-	"treesuni1",
-	"trepatchindaa1",
-	"veg_bush2",
-	"veg_bush14",
-	"veg_tree1",
-	"veg_tree3",
-	"veg_treea1",
-	"veg_treea3",
-	"veg_treeb1",
-	"veg_treenew01",
-	"veg_treenew03",
-	"veg_treenew05",
-	"veg_treenew06",
-	"veg_treenew08",
-	"veg_treenew09",
-	"veg_treenew10",
-	"veg_treenew16",
-	"veg_treenew17",
-	"vegclubtree01",
-	"vegclubtree02",
-	"vegclubtree03",
-	"vegpathtree",
+	"Dam_pod1",
+	"building_fucked",
+	"ch_roof_kb",
+	"chtwn_fmrkt",
+	"hghwysgn01",
+	"hghwysgn02",
+	"hghwysgn_sub",
+	"ind_land037",
+	"ind_land125ind",
+	"indbilbridge1",
+	"industpatch05",
+	"iten_club01",
+	"iten_details7",
+	"iten_washline02",
+	"iten_washline03",
+	"mak_Watertank",
+	"mak_Watertank2",
+	"mak_Watertank3",
+	"pharmas",
+	"rd_Road1A5",
+	"rdsign02bk",
+	"rdsign06",
+	"rdsign14",
+	"rdsign15",
+	"rdsign17",
+	"rdsign18bk",
+	"rdsign19",
+	"rdsign19",
+	"wire_shed",	
+    "Dam_pod2",
+    "Dineradam",
+    "Gdyn_barrier17",
+    "LODky_skyscrp6",
+    "LODt_skyscrp1",
+    "LODy_skyscrp23",
+    "LODy_skyscrp26b",
+    "Pumpfirescape",
+    "Pumphouse",
+    "airtower1",
+    "airtower2",
+    "bar_barrier10",
+    "bar_barrier10b",
+    "bar_barrier12",
+    "bar_barrier16",
+    "bar_barriergate1",
+    "barrierturn",
+    "billboard01",
+    "billboard02",
+    "billboard03",
+    "carlift01",
+    "carlift02",
+    "carparkfence",
+    "casino_garden",
+    "chinabanner1",
+    "chinabanner2",
+    "chinabanner3",
+    "chinabanner4",
+    "coast_treepatch",
+    "comparknewtrees",
+    "comtreepatchprk",
+    "condo_ivy",
+    "condotree01",
+    "condotree1",
+    "cons_buid02",
+    "cranebasea",
+    "cranebaseb",
+    "cranesmalltop",
+    "cranetopa",
+    "cranetopb",
+    "csky_skyscrp23",
+    "csky_skyscrp26b",
+    "damfence01",
+    "damfence02",
+    "damfence03",
+    "damfence04",
+    "damfence05",
+    "damfence06",
+    "damfence07",
+    "damfence08",
+    "damfencing",
+    "dinerind",
+    "dinerind02",
+    "dinersign",
+    "doc_crane_cab",
+    "doc_crane_leggs",
+    "doc_shedbig1",
+    "doc_shedbig12",
+    "doc_shedbig13",
+    "doc_shedbig2",
+    "doc_shedbig3",
+    "doc_shedbig31",
+    "doc_shedbig4",
+    "fencesmallb",
+    "firescapa1",
+    "firescapb1",
+    "foundation",
+    "glassfx1",
+    "glassfx2",
+    "glassfx3",
+    "glassfx4",
+    "glassfx55",
+    "glassfx_composh",
+    "glassfxsub1",
+    "glassfxsub2",
+    "helix_barrier",
+    "ind_customroad001",
+    "ind_customroad002",
+    "ind_customroad003",
+    "ind_customroad004",
+    "ind_customroad005",
+    "ind_customroad006",
+    "ind_customroad008",
+    "ind_customroad009",
+    "ind_customroad010",
+    "ind_customroad011",
+    "ind_customroad012",
+    "ind_customroad013",
+    "ind_customroad014",
+    "ind_customroad015",
+    "ind_customroad016",
+    "ind_customroad017",
+    "ind_customroad018",
+    "ind_customroad019",
+    "ind_customroad020",
+    "ind_customroad021",
+    "ind_customroad022",
+    "ind_customroad023",
+    "ind_customroad024",
+    "ind_customroad025",
+    "ind_customroad026",
+    "ind_customroad028",
+    "ind_customroad029",
+    "ind_customroad033",
+    "ind_customroad034",
+    "ind_customroad035",
+    "ind_customroad036",
+    "ind_customroad037",
+    "ind_customroad038",
+    "ind_customroad039",
+    "ind_customroad040",
+    "ind_customroad041",
+    "ind_customroad042",
+    "ind_customroad043",
+    "ind_customroad044",
+    "ind_customroad045",
+    "ind_customroad046",
+    "ind_customroad047",
+    "ind_customroad048",
+    "ind_customroad049",
+    "ind_customroad050",
+    "ind_customroad051",
+    "ind_customroad052",
+    "ind_customroad053",
+    "ind_customroad054",
+    "ind_customroad055",
+    "ind_customroad056",
+    "ind_customroad057",
+    "ind_customroad058",
+    "ind_customroad059",
+    "ind_customroad060",
+    "ind_customroad061",
+    "ind_customroad062",
+    "ind_customroad063",
+    "ind_customroad064",
+    "ind_customroad065",
+    "ind_customroad066",
+    "ind_customroad067",
+    "ind_customroad068",
+    "ind_customroad069",
+    "ind_customroad070",
+    "ind_customroad071",
+    "ind_customroad072",
+    "ind_customroad073",
+    "ind_customroad074",
+    "ind_customroad075",
+    "ind_customroad076",
+    "ind_customroad077",
+    "ind_customroad078",
+    "ind_customroad079",
+    "ind_customroad080",
+    "ind_customroad081",
+    "ind_customroad082",
+    "ind_customroad083",
+    "ind_customroad084",
+    "ind_customroad085",
+    "ind_customroad086",
+    "ind_customroad087",
+    "ind_customroad088",
+    "ind_customroad089",
+    "ind_customroad090",
+    "ind_customroad091",
+    "ind_customroad093",
+    "ind_customroad094",
+    "ind_customroad095",
+    "ind_customroad096",
+    "ind_customroad097",
+    "ind_customroad098",
+    "ind_customroad099",
+    "ind_customroad0bb",
+    "indatree03",
+    "indhelix_barrier",
+    "indjunk",
+    "indtreepatch06f",
+    "indtreepatch5",
+    "iten_washline01",
+    "kmricndo01",
+    "kmricndo02",
+    "lhouse_barrier1",
+    "lhouse_barrier2",
+    "lhouse_barrier3",
+    "mak_billboard",
+    "mak_billboardsrvc",
+    "mscp_barrier",
+    "mscp_barrier01",
+    "mscp_barriersup",
+    "nbbridgerda",
+    "nbbridgerdb",
+    "new_carprktrees",
+    "new_carprktrees4",
+    "newairportwall1",
+    "newairportwall2",
+    "newairportwall3",
+    "newairportwall4",
+    "newairportwall5",
+    "newairportwall6",
+    "newcoasttrees1",
+    "newcoasttrees2",
+    "newcoasttrees3",
+    "newtreepatch_sub",
+    "newtrees1_sub",
+    "newunitrepatch",
+    "nrailsteps",
+    "nrailstepswest",
+    "overpassind",
+    "pinetree_narrow",
+    "pinetree_wide",
+    "plnt_chimgrad",
+    "plnt_pipepart01",
+    "plnt_pipjoin4way04",
+    "plnt_pylon01",
+    "rd_Road1A10",
+    "salvsdetail",
+    "salvstrans",
+    "skyscrapenew",
+    "skyscrpunbuilt2",
+    "sub_billboard1",
+    "sub_tripbboard",
+    "subfraightback02",
+    "subfraightback03",
+    "subfraightback04",
+    "subsign1",
+    "tall_fence",
+    "tcsky_skyscrp6",
+    "telepole",
+    "telepole01",
+    "telepole02",
+    "telepole03",
+    "telepole04",
+    "tenkb_builds01",
+    "tenkb_builds03",
+    "tenkb_builds04",
+    "tenkb_builds05",
+    "tenkb_builds06",
+    "tenkb_builds11",
+    "trainstairst",
+    "treencom2",
+    "treepatch",
+    "treepatch01_sub",
+    "treepatch02_sub",
+    "treepatch03",
+    "treepatch03_sub",
+    "treepatch04_sub",
+    "treepatch05_sub",
+    "treepatch06_sub",
+    "treepatch07_sub",
+    "treepatch08_sub",
+    "treepatch09_sub",
+    "treepatch10_sub",
+    "treepatch11_sub",
+    "treepatch12_sub",
+    "treepatch13_sub",
+    "treepatch14_sub",
+    "treepatch152_sub",
+    "treepatch153_sub",
+    "treepatch15_sub",
+    "treepatch16_sub",
+    "treepatch171_sub",
+    "treepatch172_sub",
+    "treepatch173_sub",
+    "treepatch17_sub",
+    "treepatch18_sub",
+    "treepatch19_sub",
+    "treepatch2",
+    "treepatch20_sub",
+    "treepatch212_sub",
+    "treepatch213_sub",
+    "treepatch214_sub",
+    "treepatch21_sub",
+    "treepatch22_sub",
+    "treepatch23_sub",
+    "treepatch24_sub",
+    "treepatch25_sub",
+    "treepatch26_sub",
+    "treepatch27_sub",
+    "treepatch28_sub",
+    "treepatch29_sub",
+    "treepatch2b",
+    "treepatch30_sub",
+    "treepatch31_sub",
+    "treepatch32_sub",
+    "treepatch33_sub",
+    "treepatch34_sub",
+    "treepatch35_sub",
+    "treepatch69",
+    "treepatcha",
+    "treepatchb",
+    "treepatchcomtop1",
+    "treepatchd",
+    "treepatche",
+    "treepatchh",
+    "treepatchindaa2",
+    "treepatchindnew",
+    "treepatchindnew2",
+    "treepatchk",
+    "treepatchkb4",
+    "treepatchkb5",
+    "treepatchkb6",
+    "treepatchkb7",
+    "treepatchkb9",
+    "treepatchl",
+    "treepatchm",
+    "treepatchnew_sub",
+    "treepatchttwrs",
+    "treesuni1",
+    "trepatchindaa1",
+    "veg_bush14",
+    "veg_bush2",
+    "veg_tree1",
+    "veg_tree3",
+    "veg_treea1",
+    "veg_treea3",
+    "veg_treeb1",
+    "veg_treenew01",
+    "veg_treenew03",
+    "veg_treenew05",
+    "veg_treenew06",
+    "veg_treenew08",
+    "veg_treenew09",
+    "veg_treenew10",
+    "veg_treenew16",
+    "veg_treenew17",
+    "vegclubtree01",
+    "vegclubtree02",
+    "vegclubtree03",
+    "vegpathtree",
+    "wlst_skyscrp1",
 	""
 };
 char *OptimizedNames[] = {

src/liberty/core/Streaming.cpp

@@ -1170,6 +1170,24 @@ bool re3EmergencyRemoveModel() {
 	return usedmem != CStreaming::ms_memoryUsed;
 }
 
+void* re3StreamingAlloc(size_t size) {
+	auto rv = RwMalloc(size);
+
+	while (rv == nil) {
+		if (re3RemoveLeastUsedModel()) {
+			rv = RwMalloc(size);
+			continue;
+		}
+		if (re3EmergencyRemoveModel()) {
+			rv = RwMalloc(size);
+			continue;
+		}
+		return nil;
+	}
+
+	return rv;
+}
+
 bool
 CStreaming::RemoveLeastUsedModel(void)
 {

src/liberty/extras/ini.h

@@ -94,6 +94,10 @@
 #include <sys/stat.h>
 #include <cctype>
 
+#ifdef DC_SH4
+#include <dc/vmu_pkg.h>
+#endif
+
 namespace mINI
 {
 	namespace INIStringUtil
@@ -335,6 +339,7 @@ namespace mINI
 	private:
 		std::ifstream fileReadStream;
 		T_LineDataPtr lineData;
+		std::size_t start_offt;
 
 		T_LineData readFile()
 		{
@@ -343,7 +348,7 @@ namespace mINI
 			fileContents.resize(fileReadStream.tellg());
 			fileReadStream.seekg(0, std::ios::beg);
 			std::size_t fileSize = fileContents.size();
-			fileReadStream.read(&fileContents[0], fileSize);
+			fileReadStream.read(const_cast<char *>(fileContents.c_str()), fileSize);
 			fileReadStream.close();
 			T_LineData output;
 			if (fileSize == 0)
@@ -352,7 +357,19 @@ namespace mINI
 			}
 			std::string buffer;
 			buffer.reserve(50);
-			for (std::size_t i = 0; i < fileSize; ++i)
+#ifdef DC_SH4
+			{
+				vmu_pkg_t vmu_pkg;
+				if(vmu_pkg_parse(reinterpret_cast<uint8*>(const_cast<int8*>(fileContents.c_str())), &vmu_pkg) != 0) {
+					// If we failed to parse, we assume it's raw with no VMS header.
+					start_offt = 0;
+				} else {
+					start_offt = reinterpret_cast<unsigned int>(vmu_pkg.data)
+							   - reinterpret_cast<unsigned int>(fileContents.c_str());
+				}
+			}
+#endif
+			for (std::size_t i = start_offt; i < fileSize; ++i)
 			{
 				char& c = fileContents[i];
 				if (c == '\n')
@@ -378,6 +395,8 @@ namespace mINI
 			{
 				lineData = std::make_shared<T_LineData>();
 			}
+
+			start_offt = 0;
 		}
 		~INIReader() { }
 
@@ -426,22 +445,71 @@ namespace mINI
 	{
 	private:
 		std::ofstream fileWriteStream;
-
+		std::stringstream memStream;
+		inline static std::ios_base::iostate lastError_;
 	public:
 		bool prettyPrint = false;
 
-		INIGenerator(std::string const& filename)
+		INIGenerator(std::string const& filename,  bool prettyPrint_=false)
+			: fileWriteStream(filename, std::ios::out | std::ios::binary),
+			  prettyPrint(prettyPrint_)
 		{
-			fileWriteStream.open(filename, std::ios::out | std::ios::binary);
+			lastError_ = fileWriteStream.rdstate();
+		}
+
+		~INIGenerator()
+		{
+			if(!fileWriteStream.good()) {
+				return;
+			}
+
+			std::string str = memStream.str();
+			const char *buf = str.c_str();
+			int buf_size = memStream.tellp();
+
+#ifdef DC_SH4
+			uint8_t *data;
+			uint8_t icon_buf[512 * 1];
+			vmu_pkg_t vmu_pkg = {
+				.desc_short = "DCA-L Config",
+				.desc_long = "DCA-L Settings File",
+				.app_id = "The Gang",
+				.icon_cnt = 1,
+				.icon_anim_speed = 0,
+				.data_len = buf_size,
+				.icon_data = icon_buf,
+				.data = reinterpret_cast<const uint8_t*>(buf),
+			};
+
+			if (vmu_pkg_load_icon(&vmu_pkg, "settings.ico") < 0) {
+				vmu_pkg.icon_cnt = 0;
+			}
+
+			if(vmu_pkg_build(&vmu_pkg, &data, &buf_size) < 0) {
+				lastError_ = std::ios_base::badbit;
+				return;
+			}
+
+			buf = reinterpret_cast<char*>(data);
+#endif
+
+			fileWriteStream.write(buf, buf_size);
+			lastError_ = fileWriteStream.rdstate();
+
+#ifdef DC_SH4
+			// Must free the internal buffer allocated by vmu_pkg_build().
+			free(data);
+#endif
+		}
+
+		bool operator<<(const std::string& str)
+		{
+			memStream << str;
+			return true;
 		}
-		~INIGenerator() { }
 
 		bool operator<<(INIStructure const& data)
 		{
-			if (!fileWriteStream.is_open())
-			{
-				return false;
-			}
 			if (!data.size())
 			{
 				return true;
@@ -451,13 +519,13 @@ namespace mINI
 			{
 				auto const& section = it->first;
 				auto const& collection = it->second;
-				fileWriteStream
+				memStream
 					<< "["
 					<< section
 					<< "]";
 				if (collection.size())
 				{
-					fileWriteStream << INIStringUtil::endl;
+					memStream << INIStringUtil::endl;
 					auto it2 = collection.begin();
 					for (;;)
 					{
@@ -465,7 +533,7 @@ namespace mINI
 						INIStringUtil::replace(key, "=", "\\=");
 						auto value = it2->second;
 						INIStringUtil::trim(value);
-						fileWriteStream
+						memStream
 							<< key
 							<< ((prettyPrint) ? " = " : "=")
 							<< value;
@@ -473,21 +541,29 @@ namespace mINI
 						{
 							break;
 						}
-						fileWriteStream << INIStringUtil::endl;
+						memStream << INIStringUtil::endl;
 					}
 				}
 				if (++it == data.end())
 				{
 					break;
 				}
-				fileWriteStream << INIStringUtil::endl;
+				memStream << INIStringUtil::endl;
 				if (prettyPrint)
 				{
-					fileWriteStream << INIStringUtil::endl;
+					memStream << INIStringUtil::endl;
 				}
 			}
 			return true;
 		}
+
+		operator bool() const
+		{
+			return fileWriteStream.rdstate() == fileWriteStream.goodbit;
+		}
+
+		static std::ios_base::iostate lastError() { return lastError_; }
+		static bool wasGood() { return lastError_ == std::ios_base::goodbit; }
 	};
 
 	class INIWriter
@@ -667,12 +743,15 @@ namespace mINI
 		{
 			struct stat buf;
 			bool fileExists = (stat(filename.c_str(), &buf) == 0);
+
 			if (!fileExists)
 			{
-				INIGenerator generator(filename);
-				generator.prettyPrint = prettyPrint;
-				return generator << data;
+				if(INIGenerator generator(filename, prettyPrint); generator) {
+					generator << data;
+				}
+				return INIGenerator::wasGood();
 			}
+
 			INIStructure originalData;
 			T_LineDataPtr lineData;
 			bool readSuccess = false;
@@ -688,23 +767,21 @@ namespace mINI
 				return false;
 			}
 			T_LineData output = getLazyOutput(lineData, data, originalData);
-			std::ofstream fileWriteStream(filename, std::ios::out | std::ios::binary);
-			if (fileWriteStream.is_open())
 			{
-				if (output.size())
+				if (INIGenerator generator(filename, prettyPrint); generator && output.size())
 				{
 					auto line = output.begin();
 					for (;;)
 					{
-						fileWriteStream << *line;
+						generator << *line;
 						if (++line == output.end())
 						{
 							break;
 						}
-						fileWriteStream << INIStringUtil::endl;
+						generator << INIStringUtil::endl;
 					}
 				}
-				return true;
+				return INIGenerator::wasGood();
 			}
 			return false;
 		}
@@ -741,9 +818,10 @@ namespace mINI
 			{
 				return false;
 			}
-			INIGenerator generator(filename);
-			generator.prettyPrint = pretty;
-			return generator << data;
+			if(INIGenerator generator(filename, pretty); generator) {
+				generator << data;
+			}
+			return INIGenerator::wasGood();
 		}
 		bool write(INIStructure& data, bool pretty = false) const
 		{

src/liberty/math/Matrix.cpp

@@ -1,11 +1,5 @@
 #include "common.h"
 
-CMatrix::CMatrix(void)
-{
-	m_attachment = nil;
-	m_hasRwMatrix = false;
-}
-
 CMatrix::CMatrix(CMatrix const &m)
 {
 	m_attachment = nil;
@@ -510,14 +504,9 @@ operator*(const CMatrix &m1, const CMatrix &m2)
 {
 	// TODO: VU0 code
 	CMatrix out;
-#if defined(RW_DC) && 0 // THIS IS BROKEN, 4th element shouldn't be processed
-#	ifdef DC_SH4
-	MATH_Load_Matrix_Product(reinterpret_cast<const matrix_t *>(&m1), reinterpret_cast<const matrix_t *>(&m2));
-
-#	elif defined(RW_DC)
-	mat_load(reinterpret_cast<const matrix_t *>(&m2));
-	mat_apply(reinterpret_cast<const matrix_t *>(&m1));
-#	endif
+#if defined(RW_DC)
+	mat_load(reinterpret_cast<const matrix_t *>(&m1));
+	mat_apply(reinterpret_cast<const matrix_t *>(&m2));
 	mat_store(reinterpret_cast<matrix_t *>(&out));
 #else
 	out.rx = m1.rx * m2.rx + m1.fx * m2.ry + m1.ux * m2.rz;

src/liberty/math/Matrix.h

@@ -1,6 +1,6 @@
 #pragma once
 
-class CMatrix
+class alignas(8) CMatrix
 {
 public:
 	union
@@ -8,17 +8,17 @@ public:
 		alignas(8) float f[4][4];
 		struct alignas(8)
 		{
-			float rx, ry, rz, rw;
-			float fx, fy, fz, fw;
-			float ux, uy, uz, uw;
-			float px, py, pz, pw;
+			float rx, ry, rz, rw=0.0f;
+			float fx, fy, fz, fw=0.0f;
+			float ux, uy, uz, uw=0.0f;
+			float px, py, pz, pw=1.0f;
 		};
 	};
 
-	RwMatrix *m_attachment;
-	bool m_hasRwMatrix;	// are we the owner?
+	RwMatrix *m_attachment = nil;
+	bool m_hasRwMatrix = false;	// are we the owner?
 
-	CMatrix(void);
+	CMatrix(void) {}
 	CMatrix(CMatrix const &m);
 	CMatrix(RwMatrix *matrix, bool owner = false);
 	CMatrix(float scale){

src/liberty/math/VuVector.h

@@ -45,7 +45,7 @@ __always_inline void TransformPoint(CVuVector &out, const CMatrix &mat, const CV
 		": : "r" (&out) , "r" (&mat) ,"r" (&in): "memory");
 #elif defined(DC_SH4)
 	mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&mat)));
-	mat_trans_nodiv_nomod(in.x, in.y, in.z, out.x, out.y, out.z, out.y);
+	mat_trans_single3_nodiv_nomod(in.x, in.y, in.z, out.x, out.y, out.z);
 #else
 	out = mat * in;
 #endif
@@ -72,7 +72,7 @@ __always_inline void TransformPoint(CVuVector &out, const CMatrix &mat, const Rw
 		": : "r" (&out) , "r" (&mat) ,"r" (&in): "memory");
 #elif defined(DC_SH4)
 	mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&mat)));
-	mat_trans_nodiv_nomod(in.x, in.y, in.z, out.x, out.y, out.z, out.y);
+	mat_trans_single3_nodiv_nomod(in.x, in.y, in.z, out.x, out.y, out.z);
 #else
 	out = mat * in;
 #endif
@@ -80,7 +80,6 @@ __always_inline void TransformPoint(CVuVector &out, const CMatrix &mat, const Rw
 
 __always_inline void TransformPoints(CVuVector *out, int n, const CMatrix &mat, const RwV3d *in, int stride)
 {
-	assert(false);
 #ifdef GTA_PS2
 	__asm__ __volatile__("\n\
 		paddub	$3,%4,$0\n\
@@ -110,9 +109,11 @@ __always_inline void TransformPoints(CVuVector *out, int n, const CMatrix &mat,
 		": : "r" (out) , "r" (n), "r" (&mat), "r" (in), "r" (stride): "memory");
 #elif defined(DC_SH4)
     mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&mat)));
-	mat_transform(reinterpret_cast<vector_t *>(const_cast<RwV3d *>(in)),
-	              reinterpret_cast<vector_t *>(out), 
-				  n, stride - sizeof(vector_t));
+	while(n--) {
+		mat_trans_single3_nodiv_nomod(in->x, in->y, in->z, out->x, out->y, out->z);
+		in = reinterpret_cast<const RwV3d *>(reinterpret_cast<const uint8_t *>(in) + stride);
+		++out;
+	}
 #else
 	while(n--){
 		*out = mat * *in;

src/liberty/modelinfo/BaseModelInfo.cpp

@@ -38,6 +38,14 @@ CBaseModelInfo::DeleteCollisionModel(void)
 	}
 }
 
+void CBaseModelInfo::SetColModel(CColModel *col, bool owns) {
+	if (m_bOwnsColModel) {
+		delete m_colModel;
+	}
+	m_colModel = col;
+	m_bOwnsColModel = owns;
+}
+
 void
 CBaseModelInfo::AddRef(void)
 {

src/liberty/modelinfo/BaseModelInfo.h

@@ -56,8 +56,7 @@ public:
 	}
 	char *GetModelName(void) { return m_name; }
 	void SetModelName(const char *name) { strncpy(m_name, name, MAX_MODEL_NAME); }
-	void SetColModel(CColModel *col, bool owns = false){
-		m_colModel = col; m_bOwnsColModel = owns; }
+	void SetColModel(CColModel *col, bool owns = false);
 	CColModel *GetColModel(void) { return m_colModel; }
 	bool DoesOwnColModel(void) { return m_bOwnsColModel; }
 	void DeleteCollisionModel(void);

src/liberty/objects/CutsceneHead.cpp

@@ -197,6 +197,10 @@ CCutsceneHead::PlayAnimation(const char *animName)
 		RwStreamSkip(stream, offset*2048);
 		if(RwStreamFindChunk(stream, rwID_HANIMANIMATION, nil, nil)){
 			anim = RpHAnimAnimationStreamRead(stream);
+			if (hier->interpolator->currentAnim) {
+				RpHAnimAnimationDestroy(hier->interpolator->currentAnim);
+				hier->interpolator->currentAnim = nil;
+			}
 			RpHAnimHierarchySetCurrentAnim(hier, anim);
 		}
 

src/liberty/renderer/Hud.cpp

@@ -39,7 +39,11 @@
 #define AMMO_X 66.0f
 #define HEALTH_X 110.0f
 #define STARS_X 60.0f
+#ifdef RW_DC
+#define ZONE_Y 64.0f
+#else
 #define ZONE_Y 30.0f
+#endif
 #define VEHICLE_Y 55.0f
 #define CLOCK_X 111.0f
 #define SUBS_Y 68.0f

src/liberty/renderer/Weather.h

@@ -58,7 +58,11 @@ public:
 };
 
 enum {
+#ifdef RW_DC
+	NUM_RAIN_STREAKS = 10
+#elif
 	NUM_RAIN_STREAKS = 35
+#endif
 };
 
 struct tRainStreak

src/liberty/save/PCSave.cpp

@@ -17,6 +17,8 @@
 
 #include "vmu/vmu.h"
 
+void* re3StreamingAlloc(size_t size);
+
 const char* _psGetUserFilesFolder();
 
 C_PcSave PcSaveHelper;
@@ -93,16 +95,17 @@ uint32_t C_PcSave::PcClassLoadRoutine(int32 file, uint8 *data) {
 		return size;
 	} else {
 		size &= ~0x80000000;
-		uint8* compressed = (uint8*)malloc(size);
+		uint8* compressed = (uint8*)re3StreamingAlloc(size);
+		assert(compressed);
 		err = CFileMgr::Read(file, (const char*)compressed, size) != size;
 		if (err || CFileMgr::GetErrorReadWrite(file)) {
-			free(compressed);
+			RwFree(compressed);
 			return 0;
 		}
 
 		lzo_uint decompressed_size = 0;
 		auto crv = lzo1x_decompress(compressed, size, data, &decompressed_size, NULL);
-		free(compressed);
+		RwFree(compressed);
 		if (crv != LZO_E_OK) {
 			return 0;
 		}
@@ -117,31 +120,37 @@ uint32_t C_PcSave::PcClassLoadRoutine(int32 file, uint8 *data) {
 bool
 C_PcSave::PcClassSaveRoutine(int32 file, uint8 *data, uint32 size)
 {
-	void* wrkmem = malloc(LZO1X_1_MEM_COMPRESS);
-	uint8* compressed = (uint8*)malloc(size*2);
+	void* wrkmem = re3StreamingAlloc(LZO1X_1_MEM_COMPRESS);
+	assert(wrkmem);
+	uint8* compressed = (uint8*)re3StreamingAlloc(size*2);
+	assert(compressed);
 	lzo_uint compressed_size;
 	int crv = lzo1x_1_compress(data, size, compressed, &compressed_size, wrkmem);
-	free(wrkmem);
+	RwFree(wrkmem);
+
+	if (crv == LZO_E_OK && compressed_size >= size) {
+		crv = LZO_E_NOT_COMPRESSIBLE;
+	}
 
 	if (crv == LZO_E_OK) {
 		uint32_t compressed_size32 = compressed_size | 0x80000000;
 		bool err = CFileMgr::Write(file, (const char*)&compressed_size32, sizeof(compressed_size32)) != sizeof(compressed_size32);
 		if (err || CFileMgr::GetErrorReadWrite(file)) {
-			free(compressed);
+			RwFree(compressed);
 			nErrorCode = SAVESTATUS_ERR_SAVE_WRITE;
 			strncpy(SaveFileNameJustSaved, ValidSaveName, sizeof(ValidSaveName) - 1);
 			return false;
 		}
 
 		err = CFileMgr::Write(file, (const char*)compressed, compressed_size) != compressed_size;
-		free(compressed);
+		RwFree(compressed);
 		if (err || CFileMgr::GetErrorReadWrite(file)) {
 			nErrorCode = SAVESTATUS_ERR_SAVE_WRITE;
 			strncpy(SaveFileNameJustSaved, ValidSaveName, sizeof(ValidSaveName) - 1);
 			return false;
 		}
 	} else if (crv == LZO_E_NOT_COMPRESSIBLE) {
-		free(compressed);
+		RwFree(compressed);
 		uint32_t compressed_size32 = size;
 		bool err = CFileMgr::Write(file, (const char*)&compressed_size32, sizeof(compressed_size32)) != sizeof(compressed_size32);
 		if (err || CFileMgr::GetErrorReadWrite(file)) {
@@ -156,7 +165,7 @@ C_PcSave::PcClassSaveRoutine(int32 file, uint8 *data, uint32 size)
 			return false;
 		}
 	} else {
-		free(compressed);
+		RwFree(compressed);
 		return false;
 	}
 

src/miami/animation/AnimManager.cpp

@@ -12,6 +12,8 @@
 #include "AnimManager.h"
 #include "Streaming.h"
 
+void* re3StreamingAlloc(size_t size);
+
 CAnimBlock CAnimManager::ms_aAnimBlocks[NUMANIMBLOCKS];
 CAnimBlendHierarchy CAnimManager::ms_aAnimations[NUMANIMATIONS];
 int32 CAnimManager::ms_numAnimBlocks;
@@ -1312,7 +1314,7 @@ CAnimManager::LoadAnimFile(RwStream *stream, bool compress, char (*uncompressedA
 			uint16_t flags;
 			RwStreamRead(stream, &flags, sizeof(flags));
 
-			seq->keyFrames = RwMalloc(dataSize);
+			seq->keyFrames = re3StreamingAlloc(dataSize);
 			assert(seq->keyFrames);
 			RwStreamRead(stream, seq->keyFrames, dataSize - sizeof(flags));
 			seq->type = flags;

src/miami/animation/CutsceneMgr.cpp

@@ -419,7 +419,8 @@ CCutsceneMgr::DeleteCutsceneData(void)
 		CBaseModelInfo *minfo = CModelInfo::GetModelInfo(i);
 		CColModel *colModel = minfo->GetColModel();
 		if (colModel != &CTempColModels::ms_colModelPed1) {
-			delete colModel;
+			// no need to delete anymore, SetColModel will do it (~skmp)
+			//delete colModel;
 			minfo->SetColModel(&CTempColModels::ms_colModelPed1);
 		}
 	}

src/miami/audio/AudioLogic.cpp

@@ -5220,6 +5220,10 @@ cAudioManager::ProcessPedOneShots(cPedParams &params)
 						if (iSound > 60)
 							iSound = 21;
 					}
+					// In some cases the left and right channels have different loop points
+					// This looks like a data file issue where the left and right channels have different loop points
+					// This is a hot fix to always have the correct loop point for each channel individually
+					SET_LOOP_OFFSETS(m_sQueueSample.m_nSampleIndex)
 					AddSampleToRequestedQueue();
 				}
 			}

src/miami/audio/sampman.h

@@ -198,6 +198,7 @@ public:
 	char  GetCDAudioDriveLetter  (void);
 	
 	void UpdateEffectsVolume(void);
+	void UpdateStreamsVolume(void);
 
 #ifdef DC_SH4
 	void UpdateChannelVolume(uint32 nChannel);

src/miami/audio/sampman_dc.cpp

@@ -12,6 +12,8 @@
 #include "thread/thread.h"
 
 #if !defined(AUDIO_OAL) &&  !defined(AUDIO_MSS)
+#define syncf(...) // dbglog(DBG_CRITICAL, __VA_ARGS__)
+#define streamf(...) // dbglog(DBG_CRITICAL, __VA_ARGS__)
 #define verbosef(...) // dbglog(DBG_CRITICAL, __VA_ARGS__)
 #define debugf(...) // dbglog(DBG_CRITICAL, __VA_ARGS__)
 
@@ -46,6 +48,8 @@
 /* Quick access to the AICA channels */
 #define AICA_CHANNEL(x)     (AICA_MEM_CHANNELS + (x) * sizeof(aica_channel_t))
 
+static uint32_t chn_version[64];
+
 int aica_play_chn(int chn, int size, uint32_t aica_buffer, int fmt, int vol, int pan, int loop, int freq) {
 	// assert(size <= 65534);
 	// We gotta fix this at some point
@@ -69,6 +73,7 @@ int aica_play_chn(int chn, int size, uint32_t aica_buffer, int fmt, int vol, int
     chan->freq = freq;
     chan->vol = vol;
     chan->pan = pan;
+	chan->version = ++chn_version[chn];
 	snd_sh4_to_aica(tmp, cmd->size);
     return chn;
 }
@@ -182,6 +187,12 @@ uintptr_t gPlayerTalkData = 0;
 uint32 gPlayerTalkReqId = 0;
 #endif
 
+// this is very wasteful and temporary
+#define BANK_STAGE_SIZE 16 * 2048
+static  uint8_t stagingBufferBank[BANK_STAGE_SIZE] __attribute__((aligned(32)));
+std::mutex stagingBufferMtx;
+
+
 static int32 DCStreamedLength[TOTAL_STREAMED_SOUNDS];
 
 struct WavHeader {
@@ -364,6 +375,12 @@ cSampleManager::Initialise(void)
 					if (channels[i].ch != -1) {
 						assert(channels[i].nSfx != -1);
 
+						uint32_t channel_version = g2_read_32(SPU_RAM_UNCACHED_BASE + AICA_CHANNEL(channels[i].ch) + offsetof(aica_channel_t, version));
+
+						if (chn_version[channels[i].ch] != channel_version) {
+							syncf("SFX version missmatch, skipping update. expected %d got %d\n", chn_version[channels[i].ch], channel_version);
+							continue;
+						}
 						uint16_t channel_pos = (g2_read_32(SPU_RAM_UNCACHED_BASE + AICA_CHANNEL(channels[i].ch) + offsetof(aica_channel_t, pos)) & 0xffff);
 						// verbosef("Channel %d pos: %d\n", i, channel_pos);
 						if (!channels[i].loop) {
@@ -390,21 +407,27 @@ cSampleManager::Initialise(void)
 				{
 					std::lock_guard<std::mutex> lk(streams[i].mtx);
 					if (streams[i].playing) {
+						uint32_t channel_version = g2_read_32(SPU_RAM_UNCACHED_BASE + AICA_CHANNEL(streams[i].mapped_ch[0]) + offsetof(aica_channel_t, version));
+
+						if (chn_version[streams[i].mapped_ch[0]] != channel_version) {
+							syncf("Stream version missmatch, skipping update. expected %d got %d\n", chn_version[streams[i].mapped_ch[0]], channel_version);
+							continue;
+						}
 						// get channel pos
 						uint32_t channel_pos = g2_read_32(SPU_RAM_UNCACHED_BASE + AICA_CHANNEL(streams[i].mapped_ch[0]) + offsetof(aica_channel_t, pos)) & 0xffff;
 						uint32_t logical_pos = channel_pos;
 						if (logical_pos > STREAM_CHANNEL_SAMPLE_COUNT/2) {
 							logical_pos -= STREAM_CHANNEL_SAMPLE_COUNT/2;
 						}
-						verbosef("Stream %d pos: %d, log: %d, rem: %d\n", i, channel_pos, logical_pos, streams[i].played_samples);
+						streamf("Stream %d pos: %d, log: %d, played: %d\n", i, channel_pos, logical_pos, streams[i].played_samples);
 			
-						bool can_refill = (streams[i].played_samples + STREAM_CHANNEL_SAMPLE_COUNT/2) < streams[i].total_samples;
-						bool can_fetch = (streams[i].played_samples + STREAM_CHANNEL_SAMPLE_COUNT/2 + STREAM_CHANNEL_SAMPLE_COUNT/2 + STREAM_CHANNEL_SAMPLE_COUNT/2) < streams[i].total_samples;
+						bool can_refill = (streams[i].played_samples + STREAM_CHANNEL_SAMPLE_COUNT/2 + (!streams[i].first_refill)*STREAM_CHANNEL_SAMPLE_COUNT/2) < streams[i].total_samples;
+						bool can_fetch = (streams[i].played_samples + STREAM_CHANNEL_SAMPLE_COUNT/2 + STREAM_CHANNEL_SAMPLE_COUNT/2 + (!streams[i].first_refill)*STREAM_CHANNEL_SAMPLE_COUNT/2) < streams[i].total_samples;
 						// copy over data if needed from staging
 						if (channel_pos >= STREAM_CHANNEL_SAMPLE_COUNT/2 && !streams[i].next_is_upper_half) {
 							streams[i].next_is_upper_half = true;
 							if (can_refill) { // could we need a refill?
-								verbosef("Filling channel %d with lower half\n", i);
+								streamf("Filling channel %d with lower half\n", i);
 								// fill lower half
 								spu_memload(streams[i].aica_buffers[0], streams[i].buffer, STREAM_CHANNEL_BUFFER_SIZE/2);
 								if (streams[i].stereo) {
@@ -420,7 +443,7 @@ cSampleManager::Initialise(void)
 						} else if (channel_pos < STREAM_CHANNEL_SAMPLE_COUNT/2 && streams[i].next_is_upper_half) {
 							streams[i].next_is_upper_half = false;
 							if (can_refill) { // could we need a refill?
-								verbosef("Filling channel %d with upper half\n", i);
+								streamf("Filling channel %d with upper half\n", i);
 								// fill upper half
 								spu_memload(streams[i].aica_buffers[0] + STREAM_CHANNEL_BUFFER_SIZE/2, streams[i].buffer, STREAM_CHANNEL_BUFFER_SIZE/2);
 								if (streams[i].stereo) {
@@ -440,7 +463,7 @@ cSampleManager::Initialise(void)
 						// if end of file, stop
 						if ((streams[i].played_samples + logical_pos) > streams[i].total_samples) {
 							// stop channel
-							debugf("Auto stopping stream: %d -> {%d, %d}, %d total\n", i, streams[i].mapped_ch[0], streams[i].mapped_ch[1], streams[i].total_samples);
+							streamf("Auto stopping stream: %d -> {%d, %d}, %d total\n", i, streams[i].mapped_ch[0], streams[i].mapped_ch[1], streams[i].total_samples);
 							aica_stop_chn(streams[i].mapped_ch[0]);
 							aica_stop_chn(streams[i].mapped_ch[1]);
 							streams[i].playing = false;
@@ -448,7 +471,7 @@ cSampleManager::Initialise(void)
 					}
 					
 					if (do_read) {
-						debugf("Queueing stream read: %d, file: %d, buffer: %p, size: %d, tell: %d\n", i, streams[i].fd, streams[i].buffer, do_read, fs_tell(streams[i].fd));
+						streamf("Queueing stream read: %d, file: %d, buffer: %p, size: %d, file_offset: %d\n", i, streams[i].fd, streams[i].buffer, do_read, streams[i].file_offset);
 						CdStreamQueueAudioRead(streams[i].fd, streams[i].buffer, do_read, streams[i].file_offset);
 						streams[i].file_offset += do_read;
 					}
@@ -488,6 +511,7 @@ cSampleManager::Initialise(void)
 void
 cSampleManager::Terminate(void)
 {
+	CdStreamDiscardAudioRead(fdPedSfx);
 	fs_close(fdPedSfx);
 }
 
@@ -514,7 +538,6 @@ cSampleManager::UpdateEffectsVolume(void)
 	}
 }
 
-
 void
 cSampleManager::SetEffectsMasterVolume(uint8 nVolume)
 {
@@ -526,6 +549,7 @@ void
 cSampleManager::SetMusicMasterVolume(uint8 nVolume)
 {
 	m_nMusicVolume = nVolume;
+	UpdateStreamsVolume();
 }
 
 void
@@ -545,6 +569,7 @@ void
 cSampleManager::SetMusicFadeVolume(uint8 nVolume)
 {
 	m_nMusicFadeVolume = nVolume;
+	UpdateStreamsVolume();
 }
 
 void
@@ -581,26 +606,29 @@ cSampleManager::LoadSampleBank(uint8 nBank)
 		// TODO: Split per-bank sfx file
 		int fd = fs_open(SampleBankDataFilename, O_RDONLY);
 		assert(fd >= 0);
-		// this is very wasteful and temporary
-		void* stagingBuffer = memalign(32, 8 * 2048);
-		assert(stagingBuffer != 0);
-
-		// Ideally, we'd suspend the CdStream thingy here or read via that instead
-		uintptr_t loadOffset = bank.base;
+		
 		fs_seek(fd, fileStart, SEEK_SET);
+		{ 
+			std::lock_guard lk(stagingBufferMtx); // for stagingBufferBank
+		
+			void* stagingBuffer = stagingBufferBank;
+			assert(stagingBuffer != 0);
 
-		while (fileSize > 0) {
-			size_t readSize = fileSize > 8 * 2048 ? 8 * 2048 : fileSize;
-			int rs = fs_read(fd, stagingBuffer, readSize);
-			debugf("Read %d bytes, expected %d\n", rs, readSize);
-			assert(rs == readSize);
-			spu_memload(loadOffset, stagingBuffer, readSize);
-			loadOffset += readSize;
-			fileSize -= readSize;
-			debugf("Loaded %d bytes, %d remaining\n", readSize, fileSize);
+			// Ideally, we'd suspend the CdStream thingy here or read via that instead
+			uintptr_t loadOffset = bank.base;
+
+			while (fileSize > 0) {
+				size_t readSize = fileSize > sizeof(stagingBufferBank) ? sizeof(stagingBufferBank) : fileSize;
+				int rs = fs_read(fd, stagingBuffer, readSize);
+				debugf("Read %d bytes, expected %d\n", rs, readSize);
+				assert(rs == readSize);
+				spu_memload(loadOffset, stagingBuffer, readSize);
+				loadOffset += readSize;
+				fileSize -= readSize;
+				debugf("Loaded %d bytes, %d remaining\n", readSize, fileSize);
+			}
 		}
 		fs_close(fd);
-		free(stagingBuffer);
 		
 
 		for (int nSfx = BankStartOffset[nBank]; nSfx < BankStartOffset[nBank+1]; nSfx++) {
@@ -687,21 +715,25 @@ cSampleManager::LoadMissionAudio(uint8 nSlot, uint32 nSample)
 	ASSERT(nSample < TOTAL_AUDIO_SAMPLES);
 	
 	debugf("Loading mission audio comment %d, offset: %d, size: %d\n", nSample, m_aSamples[nSample].nFileOffset, m_aSamples[nSample].nByteSize);
-	CdStreamQueueAudioRead(nSample, (void*)gPlayerTalkData, m_aSamples[nSample].nByteSize, m_aSamples[nSample].nFileOffset, [](AudioReadCmd* cmd) {
-		fs_seek(fdPedSfx, cmd->seek, SEEK_SET);
+	CdStreamQueueAudioRead(fdPedSfx, (void*)gPlayerTalkData, m_aSamples[nSample].nByteSize, m_aSamples[nSample].nFileOffset, [](AudioReadCmd* cmd) {
+		assert(fs_seek(fdPedSfx, cmd->seek, SEEK_SET) == cmd->seek);
 
 		// TODO: When we can dma directly to AICA, we can use this instead
 		// fs_read(fdPedSfx, SPU_BASE_U8 + (uintptr_t)cmd->dest, cmd->size);
 
-		void* stagingBuffer = memalign(32, cmd->size);
-		assert(stagingBuffer != 0);
-		debugf("Allocated %d bytes at %p\n", cmd->size, stagingBuffer);
-		int rs = fs_read(fdPedSfx, stagingBuffer, cmd->size);
-		debugf("Read %d bytes, expected %d\n", rs, cmd->size);
-		assert(rs == cmd->size);
-
-		spu_memload((uintptr_t)cmd->dest, stagingBuffer, cmd->size);
-		free(stagingBuffer);
+		assert(cmd->size < sizeof(stagingBufferBank));
+		{
+			std::lock_guard lk(stagingBufferMtx); // for stagingBufferBank
+			void* stagingBuffer = stagingBufferBank;
+			assert(stagingBuffer != 0);
+			debugf("Allocated %d bytes at %p\n", cmd->size, stagingBuffer);
+			int rs = fs_read(fdPedSfx, stagingBuffer, cmd->size);
+			debugf("Read %d bytes, expected %d\n", rs, cmd->size);
+			assert(rs == cmd->size);
+	
+			spu_memload((uintptr_t)cmd->dest, stagingBuffer, cmd->size);
+		}
+		
 		nPedSfxReqReadId = nPedSfxReqReadId + 1;
 	});
 	
@@ -780,22 +812,26 @@ cSampleManager::LoadPedComment(uint32 nComment)
 	assert(m_aSamples[nComment].nByteSize <= pedBlocksizeMax);
 
 	debugf("Loading ped comment %d, offset: %d, size: %d\n", nComment, m_aSamples[nComment].nFileOffset, m_aSamples[nComment].nByteSize);
-	CdStreamQueueAudioRead(nComment, (void*)nPedSlotSfxAddr[nCurrentPedSlot], m_aSamples[nComment].nByteSize, m_aSamples[nComment].nFileOffset, [](AudioReadCmd* cmd) {
-		fs_seek(fdPedSfx, cmd->seek, SEEK_SET);
+	CdStreamQueueAudioRead(fdPedSfx, (void*)nPedSlotSfxAddr[nCurrentPedSlot], m_aSamples[nComment].nByteSize, m_aSamples[nComment].nFileOffset, [](AudioReadCmd* cmd) {
+		assert(fs_seek(fdPedSfx, cmd->seek, SEEK_SET) == cmd->seek);
 
 
 		// TODO: When we can dma directly to AICA, we can use this instead
 		// fs_read(fdPedSfx, SPU_BASE_U8 + (uintptr_t)cmd->dest, cmd->size);
 
-		void* stagingBuffer = memalign(32, cmd->size);
-		assert(stagingBuffer != 0);
-		debugf("Allocated %d bytes at %p\n", cmd->size, stagingBuffer);
-		int rs = fs_read(fdPedSfx, stagingBuffer, cmd->size);
-		debugf("Read %d bytes, expected %d\n", rs, cmd->size);
-		assert(rs == cmd->size);
+		assert(cmd->size < sizeof(stagingBufferBank));
+		{
+			std::lock_guard lk(stagingBufferMtx); // for stagingBufferBank
+			void* stagingBuffer = stagingBufferBank;
+			assert(stagingBuffer != 0);
+			debugf("Allocated %d bytes at %p\n", cmd->size, stagingBuffer);
+			int rs = fs_read(fdPedSfx, stagingBuffer, cmd->size);
+			debugf("Read %d bytes, expected %d\n", rs, cmd->size);
+			assert(rs == cmd->size);
+
+			spu_memload((uintptr_t)cmd->dest, stagingBuffer, cmd->size);
+		}
 
-		spu_memload((uintptr_t)cmd->dest, stagingBuffer, cmd->size);
-		free(stagingBuffer);
 		nPedSfxReqReadId = nPedSfxReqReadId + 1;
 	});
 
@@ -1081,7 +1117,7 @@ cSampleManager::PreloadStreamedFile(uint32 nFile, uint8 nStream, uint32_t seek_b
 	ASSERT( nFile < TOTAL_STREAMED_SOUNDS );
 
 	file_t f = fs_open(DCStreamedNameTable[nFile], O_RDONLY);
-	debugf("PreloadStreamedFile(%p, %d, %d) is %s\n", f, nFile, nStream, DCStreamedNameTable[nFile]);
+	streamf("PreloadStreamedFile(%p, %d, %d) is %s\n", f, nFile, nStream, DCStreamedNameTable[nFile]);
 	assert(f >= 0 );
 	WavHeader hdr;
 	assert(fs_read(f, &hdr, sizeof(hdr)) == sizeof(hdr));
@@ -1113,13 +1149,13 @@ cSampleManager::PreloadStreamedFile(uint32 nFile, uint8 nStream, uint32_t seek_b
 		streams[nStream].next_is_upper_half = true;
 		streams[nStream].first_refill = true;
 
-		debugf("PreloadStreamedFile: %s: stream: %d, freq: %d, chans: %d, byte size: %d, played samples: %d\n", DCStreamedNameTable[nFile], nStream, hdr.samplesPerSec, hdr.numOfChan, hdr.dataSize, streams[nStream].played_samples);
+		streamf("PreloadStreamedFile: %s: stream: %d, freq: %d, chans: %d, byte size: %d, played samples: %d\n", DCStreamedNameTable[nFile], nStream, hdr.samplesPerSec, hdr.numOfChan, hdr.dataSize, streams[nStream].played_samples);
 	
 
 		// How to avoid the lock?
 		if (seek_bytes_aligned) {
 			streams[nStream].played_samples = seek_bytes_aligned * (streams[nStream].stereo ? 1 : 2);
-			debugf("Seeking aligned to: %d, played_samples: %d\n", seek_bytes_aligned, streams[nStream].played_samples);
+			streamf("Seeking aligned to: %d, played_samples: %d\n", seek_bytes_aligned, streams[nStream].played_samples);
 			fs_seek(streams[nStream].fd, 2048 + seek_bytes_aligned, SEEK_SET);
 		} else {
 			fs_seek(f, 2048, SEEK_SET);
@@ -1148,7 +1184,7 @@ cSampleManager::PreloadStreamedFile(uint32 nFile, uint8 nStream, uint32_t seek_b
 		streams[nStream].file_offset = fs_tell(f);
 	}
 
-	verbosef("PreloadStreamedFile: %p %d - %s, %d, %d, \n", f, nFile, DCStreamedNameTable[nFile], streams[nStream].rate, streams[nStream].stereo);
+	streamf("PreloadStreamedFile: %p %d - %s, %d, %d, \n", f, nFile, DCStreamedNameTable[nFile], streams[nStream].rate, streams[nStream].stereo);
 }
 
 // we can't really pause a stream, so we just make it go very slow with zero volume
@@ -1275,12 +1311,28 @@ cSampleManager::GetStreamedFilePosition(uint8 nStream)
 	return rv;
 }
 
+static uint8 nStreamVolumes[MAX_STREAMS];
+static uint8 nStreamPans[MAX_STREAMS];
+static uint8 nStreamEffect[MAX_STREAMS];
+void
+cSampleManager::UpdateStreamsVolume(void)
+{
+	if(_bSampmanInitialised) {
+		for (int nStream = 0; nStream < MAX_STREAMS; nStream++) {
+			SetStreamedVolumeAndPan(nStreamVolumes[nStream], nStreamPans[nStream], nStreamEffect[nStream], nStream);
+		}
+	}
+}
+
 void
 cSampleManager::SetStreamedVolumeAndPan(uint8 nVolume, uint8 nPan, uint8 nEffectFlag, uint8 nStream)
 {
 	ASSERT( nStream < MAX_STREAMS );
 	if (nVolume > MAX_VOLUME)
 		nVolume = MAX_VOLUME;
+	nStreamVolumes[nStream] = nVolume;
+	nStreamPans[nStream] = nPan;
+	nStreamEffect[nStream] = nEffectFlag;
 	nVolume = linearlize_volume(nVolume); //nVolume * 255 / MAX_VOLUME;
 	nVolume = m_nMusicFadeVolume * nVolume * m_nMusicVolume >> 14;
 	if (streams[nStream].vol != nVolume || streams[nStream].nPan != nPan) {
@@ -1349,16 +1401,21 @@ cSampleManager::InitialiseSampleBanks(void)
 	for (uint32 nComment = SAMPLEBANK_PED_START; nComment <= SAMPLEBANK_PED_END; nComment++) {
 		pedBlocksizeMax = Max(pedBlocksizeMax, m_aSamples[nComment].nByteSize);
 	}
+	assert(pedBlocksizeMax <= BANK_STAGE_SIZE);
 	debugf("Max ped comment size: %d\n", pedBlocksizeMax);
 
 #ifdef FIX_BUGS
 
 	// Find biggest player comment
 	uint32 nMaxPlayerSize = 0;
-	for (uint32 i = PLAYER_COMMENTS_START; i <= PLAYER_COMMENTS_END; i++)
+	for (uint32 i = PLAYER_COMMENTS_START; i <= PLAYER_COMMENTS_END; i++) {
 		nMaxPlayerSize = Max(nMaxPlayerSize, m_aSamples[i].nByteSize);
+	}
 
 	debugf("Max player comment size: %d\n", nMaxPlayerSize);
+
+	assert(nMaxPlayerSize < sizeof(stagingBufferBank));
+
 	gPlayerTalkData = snd_mem_malloc(nMaxPlayerSize);
 	ASSERT(gPlayerTalkData != 0);
 

src/miami/collision/ColLine.h

@@ -1,6 +1,6 @@
 #pragma once
 
-struct CColLine
+struct alignas(8) CColLine
 {
 	// NB: this has to be compatible with two CVuVectors
 	CVector p0;

src/miami/collision/ColModel.cpp

@@ -5,6 +5,8 @@
 #include "MemoryHeap.h"
 #include "Pools.h"
 
+void* re3StreamingAlloc(size_t size);
+
 CColModel::CColModel(void)
 {
 	numSpheres = 0;
@@ -43,13 +45,13 @@ CColModel::operator delete(void *p, size_t) throw()
 void
 CColModel::RemoveCollisionVolumes(void)
 {
+	CCollision::RemoveTrianglePlanes(this);
 	if(ownsCollisionVolumes){
 		RwFree(spheres);
 		RwFree(lines);
 		RwFree(boxes);
 		RwFree(vertices);
 		RwFree(triangles);
-		CCollision::RemoveTrianglePlanes(this);
 	}
 	numSpheres = 0;
 	numLines = 0;
@@ -109,6 +111,8 @@ CColModel::operator=(const CColModel &other)
 	int i;
 	int numVerts;
 
+	CCollision::RemoveTrianglePlanes(this);
+
 	boundingSphere = other.boundingSphere;
 	boundingBox = other.boundingBox;
 
@@ -179,7 +183,7 @@ CColModel::operator=(const CColModel &other)
 		if(vertices)
 			RwFree(vertices);
 		if(numVerts){
-			vertices = (CompressedVector*)RwMalloc(numVerts*sizeof(CompressedVector));
+			vertices = (CompressedVector*)re3StreamingAlloc(numVerts*sizeof(CompressedVector));
 			for(i = 0; i < numVerts; i++)
 				vertices[i] = other.vertices[i];
 		}
@@ -189,7 +193,7 @@ CColModel::operator=(const CColModel &other)
 			numTriangles = other.numTriangles;
 			if(triangles)
 				RwFree(triangles);
-			triangles = (CColTriangle*)RwMalloc(numTriangles*sizeof(CColTriangle));
+			triangles = (CColTriangle*)re3StreamingAlloc(numTriangles*sizeof(CColTriangle));
 		}
 		for(i = 0; i < numTriangles; i++)
 			triangles[i] = other.triangles[i];

src/miami/collision/ColPoint.h

@@ -1,6 +1,6 @@
 #pragma once
 
-struct CColPoint
+struct alignas(8) CColPoint
 {
 	CVector point;
 	int pad1;

src/miami/collision/ColSphere.h

@@ -2,7 +2,7 @@
 
 #include "SurfaceTable.h"
 
-struct CSphere
+struct alignas(8) CSphere
 {
 	// NB: this has to be compatible with a CVuVector
 	CVector center;
@@ -15,6 +15,11 @@ struct CColSphere : public CSphere
 	uint8 surface;
 	uint8 piece;
 
+	void Set(float radius, uint8 surf = SURFACE_DEFAULT, uint8 piece = 0) {
+		this->radius = radius;
+		this->surface = surf;
+		this->piece = piece;
+	}
 	void Set(float radius, const CVector &center, uint8 surf, uint8 piece);
 	bool IntersectRay(CVector const &from, CVector const &dir, CVector &entry, CVector &exit);
 	using CSphere::Set;

src/miami/collision/Collision.cpp

@@ -24,6 +24,10 @@
 #include "Camera.h"
 #include "ColStore.h"
 
+#ifdef DC_SH4
+#include "VuCollision.h"
+#endif
+
 #ifdef VU_COLLISION
 #include "VuCollision.h"
 
@@ -572,7 +576,12 @@ CCollision::TestLineOfSight(const CColLine &line, const CMatrix &matrix, CColMod
 
 	// transform line to model space
 	Invert(matrix, matTransform);
-	CColLine newline(matTransform * line.p0, matTransform * line.p1);
+	CColLine newline;
+#ifndef DC_SH4
+	newline.Set(matTransform * line.p0, matTransform * line.p1);
+#else
+	TransformPoints(reinterpret_cast<CVuVector*>(&newline), 2, matTransform, &line.p0, sizeof(CColLine)/2);
+#endif
 
 	// If we don't intersect with the bounding box, no chance on the rest
 	if(!TestLineBox(newline, model.boundingBox))
@@ -1428,7 +1437,12 @@ CCollision::ProcessLineOfSight(const CColLine &line,
 
 	// transform line to model space
 	Invert(matrix, matTransform);
-	CColLine newline(matTransform * line.p0, matTransform * line.p1);
+	CColLine newline;
+#ifdef DC_SH4
+	TransformPoints(reinterpret_cast<CVuVector*>(&newline), 2, matTransform, &line.p0, sizeof(CColLine)/2);
+#else
+	newline.Set(matTransform * line.p0, matTransform * line.p1);
+#endif
 
 	// If we don't intersect with the bounding box, no chance on the rest
 	if(!TestLineBox(newline, model.boundingBox))
@@ -1455,8 +1469,18 @@ CCollision::ProcessLineOfSight(const CColLine &line,
 	}
 
 	if(coldist < mindist){
+#ifndef DC_SH4
 		point.point = matrix * point.point;
 		point.normal = Multiply3x3(matrix, point.normal);
+#else
+		mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&matrix)));
+		mat_trans_single3_nodiv(point.point.x,
+		                        point.point.y,
+		                        point.point.z);
+		mat_trans_normal3(point.normal.x,
+		                  point.normal.y,
+		                  point.normal.z);
+#endif
 		mindist = coldist;
 		return true;
 	}
@@ -1593,7 +1617,14 @@ CCollision::ProcessVerticalLine(const CColLine &line,
 
 	// transform line to model space
 	// Why does the game seem to do this differently than above?
-	CColLine newline(MultiplyInverse(matrix, line.p0), MultiplyInverse(matrix, line.p1));
+	CMatrix matTransform;
+	Invert(matrix, matTransform);
+	CColLine newline;
+#ifndef DC_SH4
+	newline.Set(matTransform * line.p0, matTransform * line.p1);
+#else
+	TransformPoints(reinterpret_cast<CVuVector*>(&newline), 2, matTransform, &line.p0, sizeof(CColLine)/2);
+#endif
 
 	if(!TestLineBox(newline, model.boundingBox))
 		return false;
@@ -1618,13 +1649,29 @@ CCollision::ProcessVerticalLine(const CColLine &line,
 	}
 
 	if(coldist < mindist){
+#ifndef DC_SH4
 		point.point = matrix * point.point;
 		point.normal = Multiply3x3(matrix, point.normal);
+#else
+		mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&matrix)));
+		mat_trans_single3_nodiv(point.point.x,
+		                        point.point.y,
+		                        point.point.z);
+		mat_trans_normal3(point.normal.x,
+		                  point.normal.y,
+		                  point.normal.z);
+#endif
 		if(TempStoredPoly.valid && poly){
 			*poly = TempStoredPoly;
+#ifndef DC_SH4
 			poly->verts[0] = matrix * poly->verts[0];
 			poly->verts[1] = matrix * poly->verts[1];
 			poly->verts[2] = matrix * poly->verts[2];
+#else
+			mat_trans_single3_nodiv(poly->verts[0].x, poly->verts[0].y, poly->verts[0].z);
+			mat_trans_single3_nodiv(poly->verts[1].x, poly->verts[1].y, poly->verts[1].z);
+			mat_trans_single3_nodiv(poly->verts[2].x, poly->verts[2].y, poly->verts[2].z);
+#endif
 		}
 		mindist = coldist;
 		return true;
@@ -1976,26 +2023,60 @@ CCollision::ProcessColModels(const CMatrix &matrixA, CColModel &modelA,
 	assert(modelA.numLines <= MAXNUMLINES);
 
 	// From model A space to model B space
-	matAB = Invert(matrixB, matAB);
+	Invert(matrixB, matAB);
+#ifndef DC_SH4
 	matAB *= matrixA;
+#else
+	mat_load(reinterpret_cast<const matrix_t*>(&matAB));
+	mat_apply(reinterpret_cast<const matrix_t*>(&matrixA));
+#endif
 
 	CColSphere bsphereAB;	// bounding sphere of A in B space
 	bsphereAB.radius = modelA.boundingSphere.radius;
+#ifndef DC_SH4
 	bsphereAB.center = matAB * modelA.boundingSphere.center;
+#else
+	mat_trans_single3_nodiv_nomod(modelA.boundingSphere.center.x,
+	                              modelA.boundingSphere.center.y,
+	                              modelA.boundingSphere.center.z,
+	                              bsphereAB.center.x,
+	                              bsphereAB.center.y,
+	                              bsphereAB.center.z);
+#endif
 	if(!TestSphereBox(bsphereAB, modelB.boundingBox))
 		return 0;
-	// B to A space
-	matBA = Invert(matrixA, matBA);
-	matBA *= matrixB;
 
 	// transform modelA's spheres and lines to B space
 	for(i = 0; i < modelA.numSpheres; i++){
 		CColSphere &s = modelA.spheres[i];
+#ifndef DC_SH4
 		aSpheresA[i].Set(s.radius, matAB * s.center, s.surface, s.piece);
+#else
+		auto &d = aSpheresA[i];
+		mat_trans_single3_nodiv_nomod(s.center.x, s.center.y, s.center.z,
+		                              d.center.x, d.center.y, d.center.z);
+		d.Set(s.radius, s.surface, s.piece);
+#endif
 	}
-	for(i = 0; i < modelA.numLines; i++)
-		aLinesA[i].Set(matAB * modelA.lines[i].p0, matAB * modelA.lines[i].p1);
 
+	for(i = 0; i < modelA.numLines; i++) {
+#ifndef DC_SH4
+		aLinesA[i].Set(matAB * modelA.lines[i].p0, matAB * modelA.lines[i].p1);
+#else
+		mat_trans_single3_nodiv_nomod(modelA.lines[i].p0.x,
+		                              modelA.lines[i].p0.y,
+		                              modelA.lines[i].p0.z,
+		                              aLinesA[i].p0.x,
+		                              aLinesA[i].p0.y,
+		                              aLinesA[i].p0.z);
+		mat_trans_single3_nodiv_nomod(modelA.lines[i].p1.x,
+		                              modelA.lines[i].p1.y,
+		                              modelA.lines[i].p1.z,
+		                              aLinesA[i].p1.x,
+		                              aLinesA[i].p1.y,
+		                              aLinesA[i].p1.z);
+#endif
+	}
 	// Test them against model B's bounding volumes
 	int numSpheresA = 0;
 	int numLinesA = 0;
@@ -2013,9 +2094,26 @@ CCollision::ProcessColModels(const CMatrix &matrixA, CColModel &modelA,
 	int numSpheresB = 0;
 	int numBoxesB = 0;
 	int numTrianglesB = 0;
+	// B to A space
+	Invert(matrixA, matBA);
+#ifndef DC_SH4
+	matBA *= matrixB;
+#else
+	mat_load(reinterpret_cast<const matrix_t*>(&matBA));
+	mat_apply(reinterpret_cast<const matrix_t*>(&matrixB));
+#endif
 	for(i = 0; i < modelB.numSpheres; i++){
 		s.radius = modelB.spheres[i].radius;
+#ifndef DC_SH4
 		s.center = matBA * modelB.spheres[i].center;
+#else
+		mat_trans_single3_nodiv_nomod(modelB.spheres[i].center.x,
+		                              modelB.spheres[i].center.y,
+		                              modelB.spheres[i].center.z,
+		                              s.center.x,
+		                              s.center.y,
+		                              s.center.z);
+#endif
 		if(TestSphereBox(s, modelA.boundingBox))
 			aSphereIndicesB[numSpheresB++] = i;
 	}
@@ -2062,9 +2160,22 @@ CCollision::ProcessColModels(const CMatrix &matrixA, CColModel &modelA,
 		if(hasCollided)
 			numCollisions++;
 	}
+
+#ifdef DC_SH4
+	mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&matrixB)));
+#endif
 	for(i = 0; i < numCollisions; i++){
+#ifndef DC_SH4
 		spherepoints[i].point = matrixB * spherepoints[i].point;
 		spherepoints[i].normal = Multiply3x3(matrixB, spherepoints[i].normal);
+#else
+		mat_trans_single3_nodiv(spherepoints[i].point.x,
+		                        spherepoints[i].point.y,
+		                        spherepoints[i].point.z);
+		mat_trans_normal3(spherepoints[i].normal.x,
+		                  spherepoints[i].normal.y,
+		                  spherepoints[i].normal.z);
+#endif
 	}
 
 	// And the same thing for the lines in A
@@ -2095,8 +2206,17 @@ CCollision::ProcessColModels(const CMatrix &matrixA, CColModel &modelA,
 	for(i = 0; i < numLinesA; i++)
 		if(aCollided[i]){
 			j = aLineIndicesA[i];
+#ifndef DC_SH4
 			linepoints[j].point = matrixB * linepoints[j].point;
 			linepoints[j].normal = Multiply3x3(matrixB, linepoints[j].normal);
+#else
+			mat_trans_single3_nodiv(linepoints[j].point.x,
+			                        linepoints[j].point.y,
+			                        linepoints[j].point.z);
+			mat_trans_normal3(linepoints[j].normal.x,
+			                  linepoints[j].normal.y,
+			                  linepoints[j].normal.z);
+#endif
 		}
 
 	return numCollisions;	// sphere collisions

src/miami/core/CdStreamDC.cpp

@@ -479,6 +479,7 @@ RemoveFirstInQueue(Queue *queue)
 }
 
 std::vector<AudioReadCmd> pendingAudioReads;
+volatile int pendingAudioRead_fd = -1;
 #if !defined(DC_SH4)
 std::mutex pendingAudioReadsMutex;
 #endif
@@ -487,8 +488,9 @@ void CdStreamQueueAudioRead(int fd, void* pBuffer, size_t bytes, size_t seek, st
 	AudioReadCmd cmd = { pBuffer, fd, bytes, seek};
 	if (!callback) {
 		cmd.callback = [](AudioReadCmd* cmd){
-			lseek(cmd->fd, cmd->seek, SEEK_SET);
-			read(cmd->fd, cmd->dest, cmd->size);
+			assert(pendingAudioRead_fd == cmd->fd);
+			assert(lseek(cmd->fd, cmd->seek, SEEK_SET) == cmd->seek);
+			assert(read(cmd->fd, cmd->dest, cmd->size) == cmd->size);
 		};
 	} else {
 		cmd.callback = callback;
@@ -500,7 +502,7 @@ void CdStreamQueueAudioRead(int fd, void* pBuffer, size_t bytes, size_t seek, st
 		auto mask = irq_disable();
 		#endif
 		for (auto it = pendingAudioReads.rbegin(); it != pendingAudioReads.rend(); ++it) {
-			if (it->fd == -1 || it->fd == fd) {
+			if (it->fd == -1) {
 				*it = cmd;
 				goto out;
 			}
@@ -532,6 +534,12 @@ void CdStreamDiscardAudioRead(int fd) {
 	#if defined(DC_SH4)
 	irq_restore(mask);
 	#endif
+
+	while (pendingAudioRead_fd == fd) {
+		#if defined(DC_SH4)
+		thd_pass();
+		#endif
+	}
 }
 
 AudioReadCmd CdStreamNextAudioRead() {
@@ -548,6 +556,8 @@ AudioReadCmd CdStreamNextAudioRead() {
 			break;
 		}
 	}
+	assert(pendingAudioRead_fd == -1);
+	pendingAudioRead_fd = cmd.fd;
 	#if defined(DC_SH4)
 	irq_restore(mask);
 	#endif
@@ -568,6 +578,7 @@ int read_loop(int fd, void* pBuffer, size_t bytes) {
 		auto cmd = CdStreamNextAudioRead();
 		while (cmd.fd != -1) {
 			cmd.callback(&cmd);
+			pendingAudioRead_fd = -1;
 			cmd = CdStreamNextAudioRead();
 		}
 	}
@@ -584,6 +595,7 @@ void *CdStreamThread(void *param)
 		auto cmd = CdStreamNextAudioRead();
 		while (cmd.fd != -1) {
 			cmd.callback(&cmd);
+			pendingAudioRead_fd = -1;
 			cmd = CdStreamNextAudioRead();
 		}
 

src/miami/core/FileLoader.cpp

@@ -30,6 +30,8 @@
 #include "ColStore.h"
 #include "Occlusion.h"
 
+void* re3StreamingAlloc(size_t size);
+
 char CFileLoader::ms_line[256];
 
 const char*
@@ -303,6 +305,24 @@ CFileLoader::LoadCollisionModel(uint8 *buf, CColModel &model, char *modelname)
 	model.boundingBox.max.z = *(float*)(buf+36);
 	model.numSpheres = *(int16*)(buf+40);
 	buf += 44;
+	if (model.spheres) {
+		RwFree(model.spheres);
+	}
+	if (model.lines) {
+		RwFree(model.lines);
+	}
+	if (model.boxes) {
+		RwFree(model.boxes);
+	}
+	if (model.vertices) {
+		RwFree(model.vertices);
+	}
+	if (model.triangles) {
+		RwFree(model.triangles);
+	}
+	if (model.trianglePlanes) {
+		CCollision::RemoveTrianglePlanes(&model);
+	}
 	if(model.numSpheres > 0){
 		model.spheres = (CColSphere*)RwMalloc(model.numSpheres*sizeof(CColSphere));
 		REGISTER_MEMPTR(&model.spheres);
@@ -360,7 +380,7 @@ CFileLoader::LoadCollisionModel(uint8 *buf, CColModel &model, char *modelname)
 	model.numTriangles = *(int16*)buf;
 	buf += 4;
 	if(model.numTriangles > 0){
-		model.triangles = (CColTriangle*)RwMalloc(model.numTriangles*sizeof(CColTriangle));
+		model.triangles = (CColTriangle*)re3StreamingAlloc(model.numTriangles*sizeof(CColTriangle));
 		REGISTER_MEMPTR(&model.triangles);
 		for(i = 0; i < model.numTriangles; i++){
 			model.triangles[i].Set(*(uint16*)buf, *(uint16*)(buf+2), *(uint16*)(buf+4), buf[6]);
@@ -698,269 +718,671 @@ CFileLoader::LoadObjectTypes(const char *filename)
 //find . -type f -name "*rail*.dff" ! -name "*shad*"
 //find . -type f -name "*veg*.dff" ! -name "*shad*"
 char *AlphaTestNames[] = {
-	"xpolytrees2_dt",
-	"gf_treesfw3_01",
-	"ci_trees05",
-	"xpolytrees3_dt",
-	"LODrtrees5",
-	"LODrtrees2",
-	"nbt_mansiontrees04",
-	"malltreereflect2",
-	"veg_treeb1",
-	"LODrtrees4",
-	"Streetlamp2",
-	"nbt_mansiontrees02",
-	"LODmrgtrees4",
-	"gf_tree2_02",
-	"malltreereflect",
-	"nbt_hoteltrees02",
-	"veg_treea1",
-	"Streetlamp1",
-	"gf_treesfw5_01",
-	"nbt_mansiontrees05",
-	"malltrees01",
-	"xpolytrees4_dt",
-	"nbt_mansiontrees03",
-	"xpolytrees1_dt",
-	"hoteltrees01",
-	"hoteltrees03",
-	"gf_treesfw1_01",
-	"veg_tree3",
-	"ap_treeshot1_02",
-	"hoteltrees02",
-	"ci_trees03",
-	"veg_treea3",
-	"mc_treesfw3_01",
-	"ap_treesfw1_01",
-	"nbt_hoteltrees01",
-	"ci_trees04",
-	"gf_treesfw2_01",
-	"LODrtrees3",
-	"ci_trees1",
-	"mc_treesfw1_01",
-	"LODrtrees1",
-	"gf_treesfw4_01",
-	"nbt_mansiontrees01",
-	"hoteltrees06",
-	"mn_treesis_int",
-	"hoteltrees05",
-	"mc_treesfw2_01",
-	"gf_tree1_01",
-	"mn_treesisl05",
-	"doublestreetlght1",
-	"ci_trees02",
-	"bank_palms09",
-	"veg_palmkb3",
-	"lw_palm1",
-	"veg_palmkb5",
-	"bank_palms21",
-	"veg_palmkb1",
-	"veg_palmbig14",
-	"veg_palmkb14",
-	"veg_palm02",
-	"veg_palmkb2",
-	"veg_palmkb9",
-	"bank_palms28",
-	"veg_palmkb7",
-	"bank_palms23",
-	"bank_palms32",
-	"veg_palmkb8",
-	"bank_palms02",
-	"bank_palms33",
-	"veg_palmkbb11",
-	"veg_palmkb4",
-	"veg_palm01",
-	"veg_palm03",
-	"veg_palmkb13",
-	"veg_palmkb10",
-	"veg_palm04",
-	"gf_tennisfence",
-	"marinafence04",
-	"marinafence01",
-	"b_hse_pierfence",
-	"ap_wallfence5",
-	"dockfence",
-	"dtn_hospital_fence",
-	"mc_fence1",
-	"od_clubfence_dy",
-	"ap_wallfence1",
-	"const_woodfence30",
-	"bb_fence1",
-	"ap_seafences",
-	"marinafence03",
-	"nbthotel08fence02",
-	"dt_sheraton_fence",
-	"dt_scabby_fence04",
-	"nbt_barfence01",
-	"ap_wallfence2",
-	"od_clevelfence",
-	"mc_fence2",
-	"ci_fence2",
-	"dt_scabby_fence03",
-	"mansionfence",
-	"nbt_hotel07fence",
-	"marinafence02",
-	"fence01",
-	"compound_fence",
-	"ap_wallfence7",
-	"fencehaitism",
-	"od_clubfence_nt",
-	"marinafence05",
-	"ap_wallfence6",
-	"Mansion2_C_fence",
-	"dt_scabby_fence07",
-	"kickfence",
-	"pw_backfence1",
-	"mc_fence3",
-	"washfence1",
-	"LODst_woodfence30",
-	"washfence2",
-	"carparkfence0",
-	"ap_wallfence3",
-	"fencesmallb",
-	"ap_wallfence4",
-	"fencehaiti",
-	"stationfence",
-	"dt_scabby_fence01",
-	"pw_bayfence2_01",
-	"lha_carfence",
-	"tall_fence",
-	"ap_seaplanfence1",
-	"LODse_pierfence",
-	"marinaveg3",
-	"ml_vegbits05",
-	"svegrgedoor",
-	"ml_vegbits04",
-	"nrth3veg35",
-	"nrth3veg05",
-	"ht_veg01_nt",
-	"nbeachvegy1",
-	"odnvegbush1",
-	"washvegy4",
-	"ht_veg02_nt",
-	"wshotelveg1",
-	"dtn_veg3",
-	"nrth7veg09",
-	"washvegy238",
-	"veg_palwee02",
-	"nrth3veg50",
-	"washpshoutveg",
-	"nrth4veg08",
-	"washvegy3",
-	"washvegy241",
-	"washvegy2",
-	"washvegy239",
-	"odnvegbush2b",
-	"veg_gaz",
-	"nrth1veg21",
-	"ht_veg04_dy",
-	"ml_vegbits03",
-	"nbeachvegy2",
-	"nrth3veg16",
-	"ht_veg02_dy",
-	"nrth4veg05",
-	"ht_veg04_nt",
-	"veg_palwee01",
-	"ht_veg01_dy",
-	"vegeha1",
-	"veged",
-	"veg_ivy_balcny_kb3",
-	"nrth1veg42",
-	"marinaveg1",
-	"nrth3veg25",
-	"veg_fern_balcny_kb1",
-	"spad_veg1",
-	"washvegy2413",
-	"marinaveg2",
-	"washvegy240",
-	"vegetationb",
-	"nrth1veg37",
-	"ml_vegbits01",
-	"veged01",
-	"dtn_veg4",
-	"washvegy242",
-	"veg_palwee03",
-	"nrth3veg59",
-	"nrth4veg212",
-	"nrth4veg21",
-	"nrth7veg",
-	"washvegy237",
-	"nrth3veg08",
-	"nrth4veg09",
-	"veg_fern_balcny_kb2",
-	"ml_vegbits02",
-	"odnvegbush2",
-	"vegetationb03",
-	"veged02",
-	"kb_planterbox",
-	"plants05b",
-	"nbdecoshplants",
-	"ap_planters2_01",
-	"ml_planterbed",
-	"marinaplanter1",
-	"washskyplant1",
-	"dzplant",
-	"ci_planter2",
-	"ap_planters1_01",
-	"gf_planters4",
-	"washskyplant2",
-	"kb_planter+bush",
-	"marinaplanter2",
-	"ci_planter1",
-	"plants04",
-	"gf_planters2",
-	"kb_planterbush2",
-	"plants01",
-	"kb_planter+bush2",
-	"plants05",
-	"ci_busht_04",
-	"starsbush2",
-	"mansbushes2",
-	"odnvegbush1",
-	"odrv_bushes",
-	"ci_busht_02",
-	"mlmallbush",
-	"mansbushes",
-	"nbw_bush02",
-	"ci_busht_08",
-	"nbw_bush01",
-	"odnvegbush2b",
-	"ap_carbush2_01",
-	"new_bushtest42",
-	"new_bushsm",
-	"mallbushs",
-	"odrv_bushes01",
-	"kb_planter+bush",
-	"ci_busht_06",
-	"beach_bush08s",
-	"beach_bush06s",
-	"new_bushtest",
-	"doontoon66_bushes01",
-	"beach_bush04",
-	"starbitbush",
-	"beach_bush02",
-	"lhavnew_bush",
-	"kb_planterbush2",
-	"ci_busht_11",
-	"odnvegbush2",
-	"kb_planter+bush2",
-	"mallbushdense",
-	"dk_rail06",
-	"cl_railing",
-	"docks10rail",
-	"cl_railingb",
-	"dk_rail05",
-	"shpfrnts03rail01",
-	"dk_rail02",
-	"nbeachbit03rails",
-	"ci_mans1rail",
-	"dk_rail07",
-	"dk_rail01",
-	"ci_mans2rail1",
-	"ci_mans1rail1",
-	"dk_rail04",
-	"shpfrnts03rail02",
-	"dk_rail03",	
+    "BillBd04",
+    "BillBd1",
+    "BillBd2",
+    "BillBd3",
+    "LODk_grassarea",
+    "LODmrgtrees4",
+    "LODo_polgrnda10",
+    "LODo_polgrnda12",
+    "LODo_polgrnda13",
+    "LODo_polgrnda14",
+    "LODo_polgrnda16",
+    "LODo_polgrnda17",
+    "LODo_polgrnda7",
+    "LODo_polgrnda8",
+    "LODo_polgrnda9",
+    "LODrtrees1",
+    "LODrtrees2",
+    "LODrtrees3",
+    "LODrtrees4",
+    "LODrtrees5",
+    "LODse_pierfence",
+    "LODst_woodfence30",
+    "Man3_base_hedge",
+    "Mansion2_C_fence",
+    "Rose_Bush01",
+    "Rose_Bush03",
+    "Starhouse",
+    "Streetlamp1",
+    "Streetlamp2",
+    "ap_blastdef_01",
+    "ap_blastdef_03",
+    "ap_boardshad1",
+    "ap_carbush2_01",
+    "ap_planters1_01",
+    "ap_planters2_01",
+    "ap_radar1_01",
+    "ap_seafences",
+    "ap_seaplanfence1",
+    "ap_stairsout02",
+    "ap_stairsout03",
+    "ap_stairsout04",
+    "ap_stairsout1",
+    "ap_subfraightback01",
+    "ap_subfraightback02",
+    "ap_subfraightback04",
+    "ap_treesfw1_01",
+    "ap_treeshot1_02",
+    "ap_wallfence1",
+    "ap_wallfence2",
+    "ap_wallfence3",
+    "ap_wallfence4",
+    "ap_wallfence5",
+    "ap_wallfence6",
+    "ap_wallfence7",
+    "b_hse_pierfence",
+    "bank_palms02",
+    "bank_palms09",
+    "bank_palms21",
+    "bank_palms23",
+    "bank_palms28",
+    "bank_palms32",
+    "bank_palms33",
+    "basketballcourt04",
+    "basketballcourt05",
+    "bb_fence1",
+    "beach_bush02",
+    "beach_bush04",
+    "beach_bush06s",
+    "beach_bush08s",
+    "biggrass2",
+    "bldigste1mesh",
+    "boatcranelg0",
+    "build1",
+    "buldingsite1",
+    "carparkfence0",
+    "chandelier",
+    "ci_astagelights",
+    "ci_astudganwal",
+    "ci_boards1",
+    "ci_boards2",
+    "ci_busht_02",
+    "ci_busht_04",
+    "ci_busht_06",
+    "ci_busht_08",
+    "ci_busht_11",
+    "ci_dstudiogantry",
+    "ci_fence2",
+    "ci_mans1rail",
+    "ci_mans1rail1",
+    "ci_mans2rail1",
+    "ci_planter1",
+    "ci_planter2",
+    "ci_pornsetxtra",
+    "ci_shipset",
+    "ci_stairsout01",
+    "ci_trees02",
+    "ci_trees03",
+    "ci_trees04",
+    "ci_trees05",
+    "ci_trees1",
+    "cl_curtains",
+    "cl_lightsuprts",
+    "cl_railing",
+    "cl_railingb",
+    "cl_recessedlights1",
+    "club_exterior01",
+    "club_exterior02",
+    "club_exterior06",
+    "club_exterior07",
+    "club_exterior08",
+    "club_exterior09",
+    "compound_fence",
+    "concerth.col",
+    "concerth02",
+    "concerth03",
+    "concerth04",
+    "concerth05",
+    "concerth06",
+    "concerth07",
+    "concerth08",
+    "concerth09",
+    "concerth10",
+    "concerth11",
+    "concerth12",
+    "concerth27",
+    "concerth29",
+    "concerth48",
+    "concerth49",
+    "concerth50",
+    "concerth51",
+    "concerth52",
+    "concerth54",
+    "concerth56",
+    "const_woodfence30",
+    "cranebasea0",
+    "crgoshp010",
+    "crocneon",
+    "deco_polgrnda10",
+    "deco_polgrnda12",
+    "deco_polgrnda13",
+    "deco_polgrnda14",
+    "deco_polgrnda16",
+    "deco_polgrnda17",
+    "deco_polgrnda7",
+    "deco_polgrnda8",
+    "deco_polgrnda9",
+    "dk_cargoshp01",
+    "dk_cargoshp03",
+    "dk_cargoshp04",
+    "dk_cargoshp05",
+    "dk_cargoshp10",
+    "dk_cargoshp100",
+    "dk_cargoshp12",
+    "dk_cargoshp24",
+    "dk_cargoshp25",
+    "dk_cargoshp28",
+    "dk_cargoshp31",
+    "dk_cargoshp32",
+    "dk_cargoshp35",
+    "dk_cargoshp40",
+    "dk_cargoshp41",
+    "dk_cargoshp47",
+    "dk_cargoshp50",
+    "dk_cargoshp51",
+    "dk_cargoshp53",
+    "dk_cargoshp54",
+    "dk_cargoshp64",
+    "dk_cargoshp65",
+    "dk_cargoshp66",
+    "dk_cargoshp68",
+    "dk_cargoshp70",
+    "dk_cargoshp71",
+    "dk_cargoshp72",
+    "dk_cargoshp73",
+    "dk_cargoshp76",
+    "dk_cargoshp95",
+    "dk_rail01",
+    "dk_rail02",
+    "dk_rail03",
+    "dk_rail03",	
+    "dk_rail04",
+    "dk_rail05",
+    "dk_rail06",
+    "dk_rail07",
+    "doc_crane_cab0",
+    "doc_crane_cab01",
+    "doc_crane_cab02",
+    "doc_crane_cab03",
+    "doc_crane_cab04",
+    "doc_craneeggs04",
+    "dock_grassarea",
+    "dock_props01",
+    "dock_props02",
+    "dockcranescale0",
+    "dockcranescale01",
+    "dockfence",
+    "dockgate01",
+    "dockgate02",
+    "dockgrass",
+    "docks10",
+    "docks10rail",
+    "docksware01",
+    "doontoon18",
+    "doontoon25_alfas",
+    "doontoon32_alfa",
+    "doontoon66_bushes01",
+    "doontoon76",
+    "doublestreetlght1",
+    "drive_flowers06",
+    "dt_compound_alfas",
+    "dt_scabby_fence01",
+    "dt_scabby_fence03",
+    "dt_scabby_fence04",
+    "dt_scabby_fence07",
+    "dt_sheraton_fence",
+    "dt_stadium__alfa1",
+    "dtn_cablesa",
+    "dtn_cablesb",
+    "dtn_cablesc",
+    "dtn_cablesd",
+    "dtn_cablese",
+    "dtn_cablesf",
+    "dtn_hospital_fence",
+    "dtn_veg3",
+    "dtn_veg4",
+    "dts_telwire1",
+    "dts_telwire2",
+    "dts_telwire4",
+    "dts_telwire9",
+    "dzplant",
+    "fence01",
+    "fencehaiti",
+    "fencehaitism",
+    "fencesmallb",
+    "gf_drdivide1_01",
+    "gf_drivingrange1",
+    "gf_golfwall",
+    "gf_picket1_01",
+    "gf_planters2",
+    "gf_planters4",
+    "gf_stadlight1_02",
+    "gf_tennisfence",
+    "gf_tree1_01",
+    "gf_tree2_02",
+    "gf_treesfw1_01",
+    "gf_treesfw2_01",
+    "gf_treesfw3_01",
+    "gf_treesfw4_01",
+    "gf_treesfw5_01",
+    "grassbitsjm",
+    "grasspatch",
+    "haitinwire11",
+    "haitinwire22",
+    "haitinwire23",
+    "havanahouse12b",
+    "hoteltrees01",
+    "hoteltrees02",
+    "hoteltrees03",
+    "hoteltrees05",
+    "hoteltrees06",
+    "ht_veg01_dy",
+    "ht_veg01_nt",
+    "ht_veg02_dy",
+    "ht_veg02_nt",
+    "ht_veg04_dy",
+    "ht_veg04_nt",
+    "kb_planter",
+    "kb_planter+bush",
+    "kb_planter+bush2",
+    "kb_planterbox",
+    "kb_planterbush2",
+    "kickfence",
+    "labiggrass",
+    "labiggrass01",
+    "labiggrass02",
+    "lh_haiblockc3",
+    "lha_carfence",
+    "lhavnew_bush",
+    "lhbasket",
+    "lhbillboard05xx",
+    "lhroofst01",
+    "lhroofst02",
+    "lhroofst03",
+    "lhroofst04",
+    "lhroofst07",
+    "lhroofst08",
+    "lhroofst09",
+    "lhroofst10",
+    "lhroofst11",
+    "lhroofst12",
+    "lhroofst14",
+    "lhroofst16",
+    "lhroofst17",
+    "lhroofst18",
+    "lhroofst19",
+    "lhroofstuffg",
+    "lhsteps3",
+    "lithavabit01",
+    "lithavabit03",
+    "lithavabit04",
+    "lithavabit05",
+    "lithavabit06",
+    "lithavabit07",
+    "lithavabit08",
+    "lithavabit09",
+    "lithavabit10",
+    "lithavabit11",
+    "lithavabit12",
+    "littlehacoast02",
+    "littlehacoast05",
+    "lw_palm1",
+    "mallbushdense",
+    "mallbushs",
+    "malltreereflect",
+    "malltreereflect2",
+    "malltrees01",
+    "man_build01",
+    "man_build01_al",
+    "man_build02",
+    "man_build02_al",
+    "man_build03",
+    "man_build03_al",
+    "man_build04",
+    "man_build04_fnce",
+    "man_build05",
+    "man_build05_al",
+    "man_build_int",
+    "man_chandeliers",
+    "man_f_chandelier",
+    "man_lights_pool",
+    "man_pool",
+    "man_safenewmny",
+    "man_scar_curtains",
+    "man_scar_props206",
+    "man_sdr_folge",
+    "man_sdr_rug",
+    "man_twr_stairs",
+    "man_twr_stairsB",
+    "man_xchandeliers",
+    "mansbushes",
+    "mansbushes2",
+    "mansion1b",
+    "mansionfence",
+    "marinafence01",
+    "marinafence02",
+    "marinafence03",
+    "marinafence04",
+    "marinafence05",
+    "marinaplanter1",
+    "marinaplanter2",
+    "marinaveg1",
+    "marinaveg2",
+    "marinaveg3",
+    "mc_fence1",
+    "mc_fence2",
+    "mc_fence3",
+    "mc_largebuild",
+    "mc_tower_01",
+    "mc_tower_02",
+    "mc_tower_03",
+    "mc_treesfw1_01",
+    "mc_treesfw2_01",
+    "mc_treesfw3_01",
+    "mc_wall",
+    "mc_wall2",
+    "mc_wall3",
+    "ml_planterbed",
+    "ml_vegbits01",
+    "ml_vegbits02",
+    "ml_vegbits03",
+    "ml_vegbits04",
+    "ml_vegbits05",
+    "mlmallbush",
+    "mn_treesis_int",
+    "mn_treesisl05",
+    "nbchflw01",
+    "nbchflw02",
+    "nbdcprmndfnc1",
+    "nbdcprmndfnc2",
+    "nbdecoshplants",
+    "nbeachbed01",
+    "nbeachbed02",
+    "nbeachbit03rails",
+    "nbeachvegy1",
+    "nbeachvegy2",
+    "nbecland12",
+    "nbecland12bball",
+    "nbecland12gates",
+    "nbt_balcony02",
+    "nbt_balcony03",
+    "nbt_balcony05",
+    "nbt_barfence01",
+    "nbt_hotel06balcony",
+    "nbt_hotel07fence",
+    "nbt_hoteltrees01",
+    "nbt_hoteltrees02",
+    "nbt_mansion01",
+    "nbt_mansion02",
+    "nbt_mansion03",
+    "nbt_mansion04",
+    "nbt_mansion0402",
+    "nbt_mansiontrees01",
+    "nbt_mansiontrees02",
+    "nbt_mansiontrees03",
+    "nbt_mansiontrees04",
+    "nbt_mansiontrees05",
+    "nbt_rooftopstart",
+    "nbthotel08fence02",
+    "nbw_bush01",
+    "nbw_bush02",
+    "nbwbusshy2",
+    "new_bushsm",
+    "new_bushtest",
+    "new_bushtest42",
+    "newbuild01",
+    "newbuild02",
+    "nhaitiplnt14",
+    "nhaitiplnt15",
+    "nrth1veg21",
+    "nrth1veg37",
+    "nrth1veg42",
+    "nrth3veg05",
+    "nrth3veg08",
+    "nrth3veg16",
+    "nrth3veg25",
+    "nrth3veg35",
+    "nrth3veg50",
+    "nrth3veg59",
+    "nrth4veg05",
+    "nrth4veg08",
+    "nrth4veg09",
+    "nrth4veg21",
+    "nrth4veg212",
+    "nrth7veg",
+    "nrth7veg09",
+    "nt_cablebox1_01",
+    "nt_cablebox2_01",
+    "nt_cablebox3_01",
+    "nt_cablebox4_01",
+    "nt_cablebox5_01",
+    "nt_cablebox6_01",
+    "od_alleys1_01_dy",
+    "od_alleys1_01_nt",
+    "od_alleys1b_01_dy",
+    "od_alleys1b_01_nt",
+    "od_alleys2_01_dy",
+    "od_alleys2_01_nt",
+    "od_alleys3_01_dy",
+    "od_alleys3_01_nt",
+    "od_alleys3b_01_dy",
+    "od_alleys3b_01_nt",
+    "od_bighotsign01",
+    "od_clevelfence",
+    "od_clubfence_dy",
+    "od_clubfence_nt",
+    "oddoorway",
+    "oddoorway2",
+    "oddoorwayS02",
+    "oddoorwayS1",
+    "odnvegbush1",
+    "odnvegbush2",
+    "odnvegbush2b",
+    "odnwiret",
+    "odrampbit",
+    "odrroofst02",
+    "odrroofst06",
+    "odrroofst07",
+    "odrv_bushes",
+    "odrv_bushes01",
+    "pierentrance02",
+    "plants01",
+    "plants04",
+    "plants05",
+    "plants05b",
+    "pw_backfence1",
+    "pw_bayfence2_01",
+    "rustship_structure0",
+    "scrapgate",
+    "sexgarden2",
+    "shipstairs",
+    "shpfrnts03rail01",
+    "shpfrnts03rail02",
+    "sjmgrass",
+    "sjmgrass2",
+    "spad_veg1",
+    "starbbnet02",
+    "starbitbush",
+    "starbits05",
+    "starbits06",
+    "starbits13",
+    "starbits13b",
+    "starbits13c",
+    "starbits17",
+    "starbits20",
+    "starbits22",
+    "starbits24",
+    "starbits45",
+    "starbits46",
+    "starbits49",
+    "starbits50",
+    "starbits51",
+    "starbits52",
+    "starbits54",
+    "starbits55",
+    "starbits56",
+    "starbits57",
+    "starbits58",
+    "starbits59",
+    "starblocks04",
+    "stardepot_FENCE",
+    "starfbed02",
+    "starfbed03",
+    "starfbed04",
+    "starfbed05",
+    "starfbed06",
+    "starsbush2",
+    "starwaterfrnt08",
+    "stationfence",
+    "stripinter",
+    "stripinterupside",
+    "striplights",
+    "svegrgedoor",
+    "tall_fence",
+    "telewire02",
+    "telewire03",
+    "telewire04",
+    "telgrphpole02",
+    "tesad06",
+    "tesad06sas",
+    "tesad06ss",
+    "veg_fern_balcny_kb1",
+    "veg_fern_balcny_kb2",
+    "veg_gaz",
+    "veg_ivy_balcny_kb3",
+    "veg_palm01",
+    "veg_palm02",
+    "veg_palm03",
+    "veg_palm04",
+    "veg_palmbig14",
+    "veg_palmkb1",
+    "veg_palmkb10",
+    "veg_palmkb13",
+    "veg_palmkb14",
+    "veg_palmkb1pERC",
+    "veg_palmkb2",
+    "veg_palmkb3",
+    "veg_palmkb4",
+    "veg_palmkb5",
+    "veg_palmkb7",
+    "veg_palmkb8",
+    "veg_palmkb9",
+    "veg_palmkbb11",
+    "veg_palwee01",
+    "veg_palwee02",
+    "veg_palwee03",
+    "veg_tree3",
+    "veg_treea1",
+    "veg_treea3",
+    "veg_treeb1",
+    "veged",
+    "veged01",
+    "veged02",
+    "vegeha1",
+    "vegetationb",
+    "vegetationb03",
+    "washbuild003",
+    "washbuild005",
+    "washbuild013",
+    "washbuild014",
+    "washbuild015",
+    "washbuild017",
+    "washbuild018",
+    "washbuild020",
+    "washbuild021",
+    "washbuild025",
+    "washbuild025neon",
+    "washbuild030",
+    "washbuild033",
+    "washbuild039",
+    "washbuild040",
+    "washbuild041",
+    "washbuild043",
+    "washbuild050",
+    "washbuild05067",
+    "washbuild058",
+    "washbuild062",
+    "washbuild063",
+    "washbuild066",
+    "washbuild068",
+    "washbuild069",
+    "washbuild071",
+    "washbuild072",
+    "washbuild073",
+    "washbuild074",
+    "washbuild075",
+    "washbuild081",
+    "washbuild0812",
+    "washbuild111",
+    "washbuild112",
+    "washbuild114",
+    "washbuild116",
+    "washbuild119",
+    "washbuild120",
+    "washbuild121",
+    "washbuild184",
+    "washbuild187",
+    "washbuild191",
+    "washbuild192",
+    "washbuild194",
+    "washbuild195",
+    "washbuild198",
+    "washbuild199",
+    "washbuild200",
+    "washbuild201",
+    "washbuild202",
+    "washbuild203",
+    "washbuild206",
+    "washbuild210",
+    "washbuild213",
+    "washbuild214",
+    "washbuild215",
+    "washbuild216",
+    "washbuild217",
+    "washbuild231",
+    "washelipad",
+    "washfence1",
+    "washfence2",
+    "washpshoutdet",
+    "washpshoutveg",
+    "washskyplant1",
+    "washskyplant2",
+    "washundermall1",
+    "washvegy2",
+    "washvegy237",
+    "washvegy238",
+    "washvegy239",
+    "washvegy240",
+    "washvegy241",
+    "washvegy2413",
+    "washvegy242",
+    "washvegy3",
+    "washvegy4",
+    "woshmallglas1",
+    "woshmallglas2",
+    "wsgbuildwl97",
+    "wshbuildws42",
+    "wshotelsign35",
+    "wshotelveg1",
+    "wshscarland226",
+    "wshtelgrphcabl01",
+    "wshtelgrphcabl02",
+    "wshtelgrphcabl03",
+    "wshtelgrphcabl04",
+    "wshtelgrphcabl05",
+    "wshtelgrphcabl06",
+    "wshtelgrphcabl07",
+    "wshtelgrphcabl08",
+    "wshtelgrphcabl09",
+    "wshtelgrphcabl10",
+    "wshtelgrphcabl11",
+    "xpolytrees1_dt",
+    "xpolytrees2_dt",
+    "xpolytrees3_dt",
+    "xpolytrees4_dt",
+
 	"",
 };
 

src/miami/core/Frontend.cpp

@@ -68,6 +68,9 @@ const CRGBA SCROLLBAR_COLOR = LABEL_COLOR;
 #define MAX_VISIBLE_OPTION_ON_SCREEN (hasNativeList(m_nCurrScreen) ? MAX_VISIBLE_LIST_ROW : MAX_VISIBLE_OPTION)
 #define SCREEN_HAS_AUTO_SCROLLBAR (m_nTotalListRow > MAX_VISIBLE_OPTION && !hasNativeList(m_nCurrScreen))
 
+#define MIN_DRAWDIST 0.5f // default was 0.925f
+#define MAX_DRAWDIST 1.8f
+
 int GetOptionCount(int screen)
 {
 	int i = 0;
@@ -729,10 +732,10 @@ CMenuManager::CheckSliderMovement(int value)
 		break;
 	case MENUACTION_DRAWDIST:
 		if(value > 0)
-			m_PrefsLOD += ((1.8f - 0.925f) / MENUSLIDER_LOGICAL_BARS);
+			m_PrefsLOD += ((MAX_DRAWDIST - MIN_DRAWDIST) / MENUSLIDER_LOGICAL_BARS);
 		else
-			m_PrefsLOD -= ((1.8f - 0.925f) / MENUSLIDER_LOGICAL_BARS);
-		m_PrefsLOD = Clamp(m_PrefsLOD, 0.925f, 1.8f);
+			m_PrefsLOD -= ((MAX_DRAWDIST - MIN_DRAWDIST) / MENUSLIDER_LOGICAL_BARS);
+		m_PrefsLOD = Clamp(m_PrefsLOD, MIN_DRAWDIST, MAX_DRAWDIST);
 		CRenderer::ms_lodDistScale = m_PrefsLOD;
 		break;
 
@@ -1562,7 +1565,7 @@ CMenuManager::DrawStandardMenus(bool activeScreen)
 							ProcessSlider(m_PrefsBrightness / 384.0f, SLIDER_Y(70.0f), HOVEROPTION_INCREASE_BRIGHTNESS, HOVEROPTION_DECREASE_BRIGHTNESS, SCREEN_WIDTH, true);
 							break;
 						case MENUACTION_DRAWDIST:
-							ProcessSlider((m_PrefsLOD - 0.925f) / 0.875f, SLIDER_Y(99.0f), HOVEROPTION_INCREASE_DRAWDIST, HOVEROPTION_DECREASE_DRAWDIST, SCREEN_WIDTH, true);
+							ProcessSlider((m_PrefsLOD - MIN_DRAWDIST) / (MAX_DRAWDIST - MIN_DRAWDIST), SLIDER_Y(99.0f), HOVEROPTION_INCREASE_DRAWDIST, HOVEROPTION_DECREASE_DRAWDIST, SCREEN_WIDTH, true);
 							break;
 						case MENUACTION_MUSICVOLUME:
 							if(m_nPrefsAudio3DProviderIndex != NO_AUDIO_PROVIDER)

src/miami/core/Pad.cpp

@@ -2026,6 +2026,10 @@ void CPad::Update(int16 pad)
 				NewState.RightStickY	= 0;
 				NewState.RightShock		= 0;
 				NewState.LeftShoulder1  = 0;
+				NewState.Cross          = 0;      // Added for safety
+                NewState.Triangle       = 0;      // Added for safety
+                NewState.Circle         = 0;      // Added for safety
+                NewState.Square         = 0;      // Added for safety
 			}
 			else
 			{
@@ -2048,8 +2052,13 @@ void CPad::Update(int16 pad)
 				NewState.RightStickX	= state->joy2x;
 				NewState.RightStickY	= state->joy2y;
 				NewState.RightShock		= state->dpad_left;
-
 				NewState.LeftShoulder1  = (state->rtrig > 128 && state->ltrig > 128) ? 255 : 0;
+				// Add PS2-style A and B mappings, This may be the only solution for purchase items issue ?... 
+				//Further more we can add other PS2 specific mapping here as certain `bool CPad::` functions are not available for mapping down below. 
+				NewState.Cross          = NewState.A;      // A -> Cross (accept/purchase)
+				NewState.Triangle       = NewState.B;      // B -> Triangle (exit)
+				NewState.Circle         = NewState.X;      // Optional: X -> Circle
+				NewState.Square         = NewState.Y;      // Optional: Y -> Square
 			}
 
 		} 
@@ -2076,6 +2085,10 @@ void CPad::Update(int16 pad)
 			NewState.RightStickY	= 0;
 			NewState.RightShock		= 0;
 			NewState.LeftShoulder1  = 0;
+			NewState.Cross          = 0;      // Added for safety
+			NewState.Triangle       = 0;      // Added for safety
+			NewState.Circle         = 0;      // Added for safety
+			NewState.Square         = 0;      // Added for safety
 		}
 
 		// if (old_contMaple == nullptr && contMaple != nullptr)
@@ -2132,12 +2145,12 @@ void CPad::Update(int16 pad)
 	if ( JustOutOfFrontend != 0 )
 		--JustOutOfFrontend;
 		
-#ifdef RW_DC
+
 	//auto old_contMaple = contMaple;
 	//auto n_dev = maple_enum_count();
 	contMaple = maple_enum_type(0, MAPLE_FUNC_CONTROLLER);
 	state = (cont_state_t *)maple_dev_status(contMaple);
-#endif
+
 }
 
 void CPad::DoCheats(void)
@@ -3515,16 +3528,15 @@ bool CPad::CycleCameraModeJustDown(void)
 	bool result;
 	switch (CURMODE)
 	{
-		case 0:
-		case 2:
-		case 3:
+		case 0: //audible feedback when changing camera ?
+
 		{
-			result = !!(NewState.Select && !OldState.Select);
+			result = !!(NewState.DPadUp && !OldState.DPadUp);
 
 			break;
 		}
 
-		case 1:
+		case 1: //audible feedback when changing camera ?
 		{
 			result = !!(NewState.DPadUp && !OldState.DPadUp);
 
@@ -3541,9 +3553,9 @@ bool CPad::CycleCameraModeJustDown(void)
 	{
 		switch (CURMODE)
 		{
-			case 1:
+			case 1: //audible feedback when changing camera ?
 			{
-				result = !!(NewState.DPadDown && !OldState.DPadDown);
+				result = !!(NewState.DPadUp && !OldState.DPadUp);
 				break;
 			}
 			default:
@@ -3869,36 +3881,49 @@ bool CPad::CollectPickupJustDown(void)
 	return false;
 }
 
-bool CPad::DuckJustDown(void)
+bool CPad::DuckJustDown(void) 
 {
-	if (ArePlayerControlsDisabled())
+	if (ArePlayerControlsDisabled()) {
+		duckFrameCounter = 0;
 		return false;
+	}
+
 #ifdef RW_DC
-	switch (CPad::GetPad(0)->Mode)
-	{
-		case 0:	//Xbox Mode
-			if (CPad::GetPad(0)->IsDualAnalog)
-			{
-				return !!(NewState.X&& !OldState.X);
-			}
-			else
-			{
-				return !!(NewState.X&& !OldState.X);
-			}
-		case 1:	//PS2 Mode
-			if (CPad::GetPad(0)->IsDualAnalog)
-			{
-				return !!(NewState.X&& !OldState.X);
-			}
-			else
-			{
-				return !!(NewState.X&& !OldState.X);
-			}
+	bool buttonPressed = false;
+	switch (CPad::GetPad(0)->Mode) {
+		case 0: // Xbox Mode
+		buttonPressed = NewState.X;
+			break;
+		case 1: // PS2 Mode
+			buttonPressed = NewState.X;
+			break;
+	}
+
+	if (buttonPressed) {
+		duckFrameCounter++;
+		if (duckFrameCounter >= DUCK_DELAY_FRAMES && !isDucking) {
+			isDucking = true;
+			return true; // Trigger ducking after delay
+		}
+	} else {
+		duckFrameCounter = 0;
+		isDucking = false;
 	}
 
 	return false;
 #else
-	return !!(NewState.LeftShock && !OldState.LeftShock);
+	bool buttonPressed = NewState.Square;
+	if (buttonPressed) {
+		duckFrameCounter++;
+		if (duckFrameCounter >= DUCK_DELAY_FRAMES && !isDucking) {
+			isDucking = true;
+			return true;
+		}
+	} else {
+		duckFrameCounter = 0;
+		isDucking = false;
+	}
+	return false;
 #endif
 }
 

src/miami/core/Pad.h

@@ -173,11 +173,26 @@ public:
 	uint32 CameraJustUpTime;
 	uint32 CameraLastPressed;
 	bool CameraIsDoublePressed;
+	static const int DUCK_DELAY_FRAMES = 30; // 1 second at 30 FPS
+	int duckFrameCounter;					// Frames the duck button has been held
+	bool isDucking;							// Tracks if duck action is active
 
 #ifdef DETECT_PAD_INPUT_SWITCH
-	static bool IsAffectedByController;
+    static bool IsAffectedByController;
 #endif
-	CPad() { }
+    
+    CPad() : duckFrameCounter(0), isDucking(false), // New variables initialized
+			SteeringLeftRightBuffer{0}, DrunkDrivingBufferUsed(0), Phase(0), Mode(0), 
+			ShakeDur(0), DisablePlayerControls(0), ShakeFreq(0), iCurrHornHistory(0), 
+			JustOutOfFrontend(0), bApplyBrakes(0), LastTimeTouched(0), AverageWeapon(0), 
+			AverageEntries(0), IsKeyboardMouse(false), IsDualAnalog(false), 
+			CameraJustDown(false), CameraJustUp(false), CameraJustUpTime(0), 
+			CameraLastPressed(0), CameraIsDoublePressed(false) {
+		Clear(false);
+		memset(bHornHistory, 0, sizeof(bHornHistory));
+		memset(CheatString, 0, sizeof(CheatString));
+	}
+
 	~CPad() { }
 
 	static bool bDisplayNoControllerMessage;

src/miami/core/Streaming.cpp

@@ -1386,6 +1386,24 @@ bool re3EmergencyRemoveModel() {
 	return usedmem != CStreaming::ms_memoryUsed;
 }
 
+void* re3StreamingAlloc(size_t size) {
+	auto rv = RwMalloc(size);
+
+	while (rv == nil) {
+		if (re3RemoveLeastUsedModel()) {
+			rv = RwMalloc(size);
+			continue;
+		}
+		if (re3EmergencyRemoveModel()) {
+			rv = RwMalloc(size);
+			continue;
+		}
+		return nil;
+	}
+
+	return rv;
+}
+
 bool
 CStreaming::RemoveLeastUsedModel(uint32 excludeMask)
 {

src/miami/core/common.h

@@ -7,6 +7,8 @@
 #pragma warning(disable: 4838)  // narrowing conversion
 #pragma warning(disable: 4996)  // POSIX names
 
+#include "src/common_defines.h"
+
 #ifdef __MWERKS__
 #define __STDC_LIMIT_MACROS // so we get UINT32_MAX etc
 #endif

src/miami/core/re3.cpp

@@ -59,6 +59,8 @@
 
 #include <list>
 
+#include "vmu/vmu.h"
+
 #ifdef RWLIBS
 extern "C" int vsprintf(char* const _Buffer, char const* const _Format, va_list  _ArgList);
 #endif
@@ -195,7 +197,13 @@ CustomFrontendOptionsPopulate(void)
 #define MINI_CASE_SENSITIVE
 #include "ini.h"
 
-mINI::INIFile ini("reVC.ini");
+mINI::INIFile ini(
+#ifdef DC_SIM
+	"reVC.ini"
+#else
+	"/vmu/" VMU_DEFAULT_PATH "/reVCini"
+#endif
+);
 mINI::INIStructure cfg;
 
 bool ReadIniIfExists(const char *cat, const char *key, uint32 *out)
@@ -483,13 +491,19 @@ void SaveINIControllerSettings()
 #endif
 	StoreIni("Controller", "PadButtonsInited", ControlsManager.ms_padButtonsInited);
 
-	ini.write(cfg);
+	{
+		RAIIVmuBeep(VMU_DEFAULT_PATH, 1.0f);
+		ini.write(cfg);
+	}
 }
 
 bool LoadINISettings()
 {
-	if (!ini.read(cfg))
-		return false;
+	{
+		RAIIVmuBeep(VMU_DEFAULT_PATH, 1.0f);
+		if (!ini.read(cfg))
+			return false;
+	}
 
 #ifdef IMPROVED_VIDEOMODE
 	ReadIniIfExists("VideoMode", "Width", &FrontEndMenuManager.m_nPrefsWidth);

src/miami/extras/ini.h

@@ -94,6 +94,10 @@
 #include <sys/stat.h>
 #include <cctype>
 
+#ifdef DC_SH4
+#include <dc/vmu_pkg.h>
+#endif
+
 namespace mINI
 {
 	namespace INIStringUtil
@@ -335,6 +339,7 @@ namespace mINI
 	private:
 		std::ifstream fileReadStream;
 		T_LineDataPtr lineData;
+		std::size_t start_offt;
 
 		T_LineData readFile()
 		{
@@ -343,7 +348,7 @@ namespace mINI
 			fileContents.resize(fileReadStream.tellg());
 			fileReadStream.seekg(0, std::ios::beg);
 			std::size_t fileSize = fileContents.size();
-			fileReadStream.read(&fileContents[0], fileSize);
+			fileReadStream.read(const_cast<char *>(fileContents.c_str()), fileSize);
 			fileReadStream.close();
 			T_LineData output;
 			if (fileSize == 0)
@@ -352,7 +357,19 @@ namespace mINI
 			}
 			std::string buffer;
 			buffer.reserve(50);
-			for (std::size_t i = 0; i < fileSize; ++i)
+#ifdef DC_SH4
+			{
+				vmu_pkg_t vmu_pkg;
+				if(vmu_pkg_parse(reinterpret_cast<uint8*>(const_cast<int8*>(fileContents.c_str())), &vmu_pkg) != 0) {
+					// If we failed to parse, we assume it's raw with no VMS header.
+					start_offt = 0;
+				} else {
+					start_offt = reinterpret_cast<unsigned int>(vmu_pkg.data)
+							   - reinterpret_cast<unsigned int>(fileContents.c_str());
+				}
+			}
+#endif
+			for (std::size_t i = start_offt; i < fileSize; ++i)
 			{
 				char& c = fileContents[i];
 				if (c == '\n')
@@ -378,6 +395,8 @@ namespace mINI
 			{
 				lineData = std::make_shared<T_LineData>();
 			}
+
+			start_offt = 0;
 		}
 		~INIReader() { }
 
@@ -426,22 +445,71 @@ namespace mINI
 	{
 	private:
 		std::ofstream fileWriteStream;
-
+		std::stringstream memStream;
+		inline static std::ios_base::iostate lastError_;
 	public:
 		bool prettyPrint = false;
 
-		INIGenerator(std::string const& filename)
+		INIGenerator(std::string const& filename,  bool prettyPrint_=false)
+			: fileWriteStream(filename, std::ios::out | std::ios::binary),
+			  prettyPrint(prettyPrint_)
 		{
-			fileWriteStream.open(filename, std::ios::out | std::ios::binary);
+			lastError_ = fileWriteStream.rdstate();
+		}
+
+		~INIGenerator()
+		{
+			if(!fileWriteStream.good()) {
+				return;
+			}
+
+			std::string str = memStream.str();
+			const char *buf = str.c_str();
+			int buf_size = memStream.tellp();
+
+#ifdef DC_SH4
+			uint8_t *data;
+			uint8_t icon_buf[512 * 1];
+			vmu_pkg_t vmu_pkg = {
+				.desc_short = "DCA-M Config",
+				.desc_long = "DCA-M Settings File",
+				.app_id = "The Gang",
+				.icon_cnt = 1,
+				.icon_anim_speed = 0,
+				.data_len = buf_size,
+				.icon_data = icon_buf,
+				.data = reinterpret_cast<const uint8_t*>(buf),
+			};
+
+			if (vmu_pkg_load_icon(&vmu_pkg, "settings.ico") < 0) {
+				vmu_pkg.icon_cnt = 0;
+			}
+
+			if(vmu_pkg_build(&vmu_pkg, &data, &buf_size) < 0) {
+				lastError_ = std::ios_base::badbit;
+				return;
+			}
+
+			buf = reinterpret_cast<char*>(data);
+#endif
+
+			fileWriteStream.write(buf, buf_size);
+			lastError_ = fileWriteStream.rdstate();
+
+#ifdef DC_SH4
+			// Must free the internal buffer allocated by vmu_pkg_build().
+			free(data);
+#endif
+		}
+
+		bool operator<<(const std::string& str)
+		{
+			memStream << str;
+			return true;
 		}
-		~INIGenerator() { }
 
 		bool operator<<(INIStructure const& data)
 		{
-			if (!fileWriteStream.is_open())
-			{
-				return false;
-			}
 			if (!data.size())
 			{
 				return true;
@@ -451,13 +519,13 @@ namespace mINI
 			{
 				auto const& section = it->first;
 				auto const& collection = it->second;
-				fileWriteStream
+				memStream
 					<< "["
 					<< section
 					<< "]";
 				if (collection.size())
 				{
-					fileWriteStream << INIStringUtil::endl;
+					memStream << INIStringUtil::endl;
 					auto it2 = collection.begin();
 					for (;;)
 					{
@@ -465,7 +533,7 @@ namespace mINI
 						INIStringUtil::replace(key, "=", "\\=");
 						auto value = it2->second;
 						INIStringUtil::trim(value);
-						fileWriteStream
+						memStream
 							<< key
 							<< ((prettyPrint) ? " = " : "=")
 							<< value;
@@ -473,21 +541,29 @@ namespace mINI
 						{
 							break;
 						}
-						fileWriteStream << INIStringUtil::endl;
+						memStream << INIStringUtil::endl;
 					}
 				}
 				if (++it == data.end())
 				{
 					break;
 				}
-				fileWriteStream << INIStringUtil::endl;
+				memStream << INIStringUtil::endl;
 				if (prettyPrint)
 				{
-					fileWriteStream << INIStringUtil::endl;
+					memStream << INIStringUtil::endl;
 				}
 			}
 			return true;
 		}
+
+		operator bool() const
+		{
+			return fileWriteStream.rdstate() == fileWriteStream.goodbit;
+		}
+
+		static std::ios_base::iostate lastError() { return lastError_; }
+		static bool wasGood() { return lastError_ == std::ios_base::goodbit; }
 	};
 
 	class INIWriter
@@ -667,12 +743,15 @@ namespace mINI
 		{
 			struct stat buf;
 			bool fileExists = (stat(filename.c_str(), &buf) == 0);
+
 			if (!fileExists)
 			{
-				INIGenerator generator(filename);
-				generator.prettyPrint = prettyPrint;
-				return generator << data;
+				if(INIGenerator generator(filename, prettyPrint); generator) {
+					generator << data;
+				}
+				return INIGenerator::wasGood();
 			}
+
 			INIStructure originalData;
 			T_LineDataPtr lineData;
 			bool readSuccess = false;
@@ -688,23 +767,21 @@ namespace mINI
 				return false;
 			}
 			T_LineData output = getLazyOutput(lineData, data, originalData);
-			std::ofstream fileWriteStream(filename, std::ios::out | std::ios::binary);
-			if (fileWriteStream.is_open())
 			{
-				if (output.size())
+				if (INIGenerator generator(filename, prettyPrint); generator && output.size())
 				{
 					auto line = output.begin();
 					for (;;)
 					{
-						fileWriteStream << *line;
+						generator << *line;
 						if (++line == output.end())
 						{
 							break;
 						}
-						fileWriteStream << INIStringUtil::endl;
+						generator << INIStringUtil::endl;
 					}
 				}
-				return true;
+				return INIGenerator::wasGood();
 			}
 			return false;
 		}
@@ -741,9 +818,10 @@ namespace mINI
 			{
 				return false;
 			}
-			INIGenerator generator(filename);
-			generator.prettyPrint = pretty;
-			return generator << data;
+			if(INIGenerator generator(filename, pretty); generator) {
+				generator << data;
+			}
+			return INIGenerator::wasGood();
 		}
 		bool write(INIStructure& data, bool pretty = false) const
 		{

src/miami/math/Matrix.cpp

@@ -1,11 +1,5 @@
 #include "common.h"
 
-CMatrix::CMatrix(void)
-{
-	m_attachment = nil;
-	m_hasRwMatrix = false;
-}
-
 CMatrix::CMatrix(CMatrix const &m)
 {
 	m_attachment = nil;
@@ -434,6 +428,11 @@ operator*(const CMatrix &m1, const CMatrix &m2)
 {
 	// TODO: VU0 code
 	CMatrix out;
+#if defined(RW_DC)
+	mat_load(reinterpret_cast<const matrix_t *>(&m1));
+	mat_apply(reinterpret_cast<const matrix_t *>(&m2));
+	mat_store(reinterpret_cast<matrix_t *>(&out));
+#else
 	out.rx = m1.rx * m2.rx + m1.fx * m2.ry + m1.ux * m2.rz;
 	out.ry = m1.ry * m2.rx + m1.fy * m2.ry + m1.uy * m2.rz;
 	out.rz = m1.rz * m2.rx + m1.fz * m2.ry + m1.uz * m2.rz;
@@ -446,6 +445,7 @@ operator*(const CMatrix &m1, const CMatrix &m2)
 	out.px = m1.rx * m2.px + m1.fx * m2.py + m1.ux * m2.pz + m1.px;
 	out.py = m1.ry * m2.px + m1.fy * m2.py + m1.uy * m2.pz + m1.py;
 	out.pz = m1.rz * m2.px + m1.fz * m2.py + m1.uz * m2.pz + m1.pz;
+#endif
 	return out;
 }
 

src/miami/math/Matrix.h

@@ -1,6 +1,6 @@
 #pragma once
 
-class CMatrix
+class alignas(8) CMatrix
 {
 public:
 #ifdef GTA_PS2
@@ -23,18 +23,18 @@ public:
 		float f[4][4];
 		struct
 		{
-			float rx, ry, rz, rw;
-			float fx, fy, fz, fw;
-			float ux, uy, uz, uw;
-			float px, py, pz, pw;
+			float rx, ry, rz, rw=0.0f;
+			float fx, fy, fz, fw=0.0f;
+			float ux, uy, uz, uw=0.0f;
+			float px, py, pz, pw=1.0f;
 		};
 	};
 
-	RwMatrix *m_attachment;
-	bool m_hasRwMatrix;	// are we the owner?
+	RwMatrix *m_attachment=nil;
+	bool m_hasRwMatrix=false;	// are we the owner?
 #endif
 
-	CMatrix(void);
+	CMatrix(void) {}
 	CMatrix(CMatrix const &m);
 	CMatrix(RwMatrix *matrix, bool owner = false);
 	CMatrix(float scale){

src/miami/math/VuVector.h

@@ -1,6 +1,14 @@
 #pragma once
 
-class TYPEALIGN(16) CVuVector : public CVector
+#include "maths.h"
+
+#ifdef RW_DC
+#define VECTOR_ALIGN 8
+#else
+#define VECTOR_ALIGN 16
+#endif
+
+class TYPEALIGN(VECTOR_ALIGN) CVuVector : public CVector
 {
 public:
 	float w;
@@ -26,7 +34,123 @@ public:
 	// TODO: operator-
 };
 
-void TransformPoint(CVuVector &out, const CMatrix &mat, const CVuVector &in);
-void TransformPoint(CVuVector &out, const CMatrix &mat, const RwV3d &in);
-void TransformPoints(CVuVector *out, int n, const CMatrix &mat, const RwV3d *in, int stride);
-void TransformPoints(CVuVector *out, int n, const CMatrix &mat, const CVuVector *in);
+__always_inline void TransformPoint(CVuVector &out, const CMatrix &mat, const CVuVector &in)
+{
+#ifdef GTA_PS2
+	__asm__ __volatile__("\n\
+		lqc2    vf01,0x0(%2)\n\
+		lqc2    vf02,0x0(%1)\n\
+		lqc2    vf03,0x10(%1)\n\
+		lqc2    vf04,0x20(%1)\n\
+		lqc2    vf05,0x30(%1)\n\
+		vmulax.xyz	ACC,   vf02,vf01\n\
+		vmadday.xyz	ACC,   vf03,vf01\n\
+		vmaddaz.xyz	ACC,   vf04,vf01\n\
+		vmaddw.xyz	vf06,vf05,vf00\n\
+		sqc2    vf06,0x0(%0)\n\
+		": : "r" (&out) , "r" (&mat) ,"r" (&in): "memory");
+#elif defined(DC_SH4)
+	mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&mat)));
+	mat_trans_single3_nodiv_nomod(in.x, in.y, in.z, out.x, out.y, out.z);
+#else
+	out = mat * in;
+#endif
+}
+
+__always_inline void TransformPoint(CVuVector &out, const CMatrix &mat, const RwV3d &in)
+{
+#ifdef GTA_PS2
+	__asm__ __volatile__("\n\
+		ldr	$8,0x0(%2)\n\
+		ldl	$8,0x7(%2)\n\
+		lw	$9,0x8(%2)\n\
+		pcpyld	$10,$9,$8\n\
+		qmtc2	$10,vf01\n\
+		lqc2    vf02,0x0(%1)\n\
+		lqc2    vf03,0x10(%1)\n\
+		lqc2    vf04,0x20(%1)\n\
+		lqc2    vf05,0x30(%1)\n\
+		vmulax.xyz	ACC,   vf02,vf01\n\
+		vmadday.xyz	ACC,   vf03,vf01\n\
+		vmaddaz.xyz	ACC,   vf04,vf01\n\
+		vmaddw.xyz	vf06,vf05,vf00\n\
+		sqc2    vf06,0x0(%0)\n\
+		": : "r" (&out) , "r" (&mat) ,"r" (&in): "memory");
+#elif defined(DC_SH4)
+	mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&mat)));
+	mat_trans_single3_nodiv_nomod(in.x, in.y, in.z, out.x, out.y, out.z);
+#else
+	out = mat * in;
+#endif
+}
+
+__always_inline void TransformPoints(CVuVector *out, int n, const CMatrix &mat, const RwV3d *in, int stride)
+{
+#ifdef GTA_PS3
+	__asm__ __volatile__("\n\
+		paddub	$3,%4,$0\n\
+		lqc2    vf02,0x0(%2)\n\
+		lqc2    vf03,0x10(%2)\n\
+		lqc2    vf04,0x20(%2)\n\
+		lqc2    vf05,0x30(%2)\n\
+		ldr	$8,0x0(%3)\n\
+		ldl	$8,0x7(%3)\n\
+		lw	$9,0x8(%3)\n\
+		pcpyld	$10,$9,$8\n\
+		qmtc2	$10,vf01\n\
+	1:	vmulax.xyz	ACC,   vf02,vf01\n\
+		vmadday.xyz	ACC,   vf03,vf01\n\
+		vmaddaz.xyz	ACC,   vf04,vf01\n\
+		vmaddw.xyz	vf06,vf05,vf00\n\
+		add	%3,%3,$3\n\
+		ldr	$8,0x0(%3)\n\
+		ldl	$8,0x7(%3)\n\
+		lw	$9,0x8(%3)\n\
+		pcpyld	$10,$9,$8\n\
+		qmtc2	$10,vf01\n\
+		addi	%1,%1,-1\n\
+		addiu	%0,%0,0x10\n\
+		sqc2    vf06,-0x10(%0)\n\
+		bnez	%1,1b\n\
+		": : "r" (out) , "r" (n), "r" (&mat), "r" (in), "r" (stride): "memory");
+#elif defined(DC_SH4)
+    mat_load(reinterpret_cast<matrix_t *>(const_cast<CMatrix *>(&mat)));
+	while(n--) {
+		mat_trans_single3_nodiv_nomod(in->x, in->y, in->z, out->x, out->y, out->z);
+		in = reinterpret_cast<const RwV3d *>(reinterpret_cast<const uint8_t *>(in) + stride);
+		++out;
+	}
+#else
+	while(n--){
+		*out = mat * *in;
+		in = (RwV3d*)((uint8*)in + stride);
+		out++;
+	}
+#endif
+}
+
+__always_inline void TransformPoints(CVuVector *out, int n, const CMatrix &mat, const CVuVector *in)
+{
+#ifdef GTA_PS2
+	__asm__ __volatile__("\n\
+		lqc2    vf02,0x0(%2)\n\
+		lqc2    vf03,0x10(%2)\n\
+		lqc2    vf04,0x20(%2)\n\
+		lqc2    vf05,0x30(%2)\n\
+		lqc2    vf01,0x0(%3)\n\
+		nop\n\
+	1:	vmulax.xyz	ACC,   vf02,vf01\n\
+		vmadday.xyz	ACC,   vf03,vf01\n\
+		vmaddaz.xyz	ACC,   vf04,vf01\n\
+		vmaddw.xyz	vf06,vf05,vf00\n\
+		lqc2	vf01,0x10(%3)\n\
+		addiu	%3,%3,0x10\n\
+		addi	%1,%1,-1\n\
+		addiu	%0,%0,0x10\n\
+		sqc2    vf06,-0x10(%0)\n\
+		bnez	%1,1b\n\
+		": : "r" (out) , "r" (n), "r" (&mat) ,"r" (in): "memory");
+#else
+    TransformPoints(out, n, mat, in, sizeof(CVuVector));
+#endif
+}

src/miami/math/math.cpp

@@ -3,116 +3,3 @@
 #include "VuVector.h"
 
 // TODO: move more stuff into here
-
-
-void TransformPoint(CVuVector &out, const CMatrix &mat, const CVuVector &in)
-{
-#ifdef GTA_PS2
-	__asm__ __volatile__("\n\
-		lqc2    vf01,0x0(%2)\n\
-		lqc2    vf02,0x0(%1)\n\
-		lqc2    vf03,0x10(%1)\n\
-		lqc2    vf04,0x20(%1)\n\
-		lqc2    vf05,0x30(%1)\n\
-		vmulax.xyz	ACC,   vf02,vf01\n\
-		vmadday.xyz	ACC,   vf03,vf01\n\
-		vmaddaz.xyz	ACC,   vf04,vf01\n\
-		vmaddw.xyz	vf06,vf05,vf00\n\
-		sqc2    vf06,0x0(%0)\n\
-		": : "r" (&out) , "r" (&mat) ,"r" (&in): "memory");
-#else
-	out = mat * in;
-#endif
-}
-
-void TransformPoint(CVuVector &out, const CMatrix &mat, const RwV3d &in)
-{
-#ifdef GTA_PS2
-	__asm__ __volatile__("\n\
-		ldr	$8,0x0(%2)\n\
-		ldl	$8,0x7(%2)\n\
-		lw	$9,0x8(%2)\n\
-		pcpyld	$10,$9,$8\n\
-		qmtc2	$10,vf01\n\
-		lqc2    vf02,0x0(%1)\n\
-		lqc2    vf03,0x10(%1)\n\
-		lqc2    vf04,0x20(%1)\n\
-		lqc2    vf05,0x30(%1)\n\
-		vmulax.xyz	ACC,   vf02,vf01\n\
-		vmadday.xyz	ACC,   vf03,vf01\n\
-		vmaddaz.xyz	ACC,   vf04,vf01\n\
-		vmaddw.xyz	vf06,vf05,vf00\n\
-		sqc2    vf06,0x0(%0)\n\
-		": : "r" (&out) , "r" (&mat) ,"r" (&in): "memory");
-#else
-	out = mat * in;
-#endif
-}
-
-void TransformPoints(CVuVector *out, int n, const CMatrix &mat, const RwV3d *in, int stride)
-{
-#ifdef GTA_PS3
-	__asm__ __volatile__("\n\
-		paddub	$3,%4,$0\n\
-		lqc2    vf02,0x0(%2)\n\
-		lqc2    vf03,0x10(%2)\n\
-		lqc2    vf04,0x20(%2)\n\
-		lqc2    vf05,0x30(%2)\n\
-		ldr	$8,0x0(%3)\n\
-		ldl	$8,0x7(%3)\n\
-		lw	$9,0x8(%3)\n\
-		pcpyld	$10,$9,$8\n\
-		qmtc2	$10,vf01\n\
-	1:	vmulax.xyz	ACC,   vf02,vf01\n\
-		vmadday.xyz	ACC,   vf03,vf01\n\
-		vmaddaz.xyz	ACC,   vf04,vf01\n\
-		vmaddw.xyz	vf06,vf05,vf00\n\
-		add	%3,%3,$3\n\
-		ldr	$8,0x0(%3)\n\
-		ldl	$8,0x7(%3)\n\
-		lw	$9,0x8(%3)\n\
-		pcpyld	$10,$9,$8\n\
-		qmtc2	$10,vf01\n\
-		addi	%1,%1,-1\n\
-		addiu	%0,%0,0x10\n\
-		sqc2    vf06,-0x10(%0)\n\
-		bnez	%1,1b\n\
-		": : "r" (out) , "r" (n), "r" (&mat), "r" (in), "r" (stride): "memory");
-#else
-	while(n--){
-		*out = mat * *in;
-		in = (RwV3d*)((uint8*)in + stride);
-		out++;
-	}
-#endif
-}
-
-void TransformPoints(CVuVector *out, int n, const CMatrix &mat, const CVuVector *in)
-{
-#ifdef GTA_PS2
-	__asm__ __volatile__("\n\
-		lqc2    vf02,0x0(%2)\n\
-		lqc2    vf03,0x10(%2)\n\
-		lqc2    vf04,0x20(%2)\n\
-		lqc2    vf05,0x30(%2)\n\
-		lqc2    vf01,0x0(%3)\n\
-		nop\n\
-	1:	vmulax.xyz	ACC,   vf02,vf01\n\
-		vmadday.xyz	ACC,   vf03,vf01\n\
-		vmaddaz.xyz	ACC,   vf04,vf01\n\
-		vmaddw.xyz	vf06,vf05,vf00\n\
-		lqc2	vf01,0x10(%3)\n\
-		addiu	%3,%3,0x10\n\
-		addi	%1,%1,-1\n\
-		addiu	%0,%0,0x10\n\
-		sqc2    vf06,-0x10(%0)\n\
-		bnez	%1,1b\n\
-		": : "r" (out) , "r" (n), "r" (&mat) ,"r" (in): "memory");
-#else
-	while(n--){
-		*out = mat * *in;
-		in++;
-		out++;
-	}
-#endif
-}

src/miami/math/maths.h

@@ -1,5 +1,32 @@
 #pragma once
 
+#include "src/common_defines.h"
+
+#include <dc/matrix.h>
+
+#ifdef DC_SH4
+
+#define mat_trans_nodiv_nomod(x, y, z, x2, y2, z2, w2) do { \
+        register float __x __asm__("fr12") = (x); \
+        register float __y __asm__("fr13") = (y); \
+        register float __z __asm__("fr14") = (z); \
+        register float __w __asm__("fr15") = 1.0f; \
+        __asm__ __volatile__( "ftrv  xmtrx, fv12\n" \
+                              : "=f" (__x), "=f" (__y), "=f" (__z), "=f" (__w) \
+                              : "0" (__x), "1" (__y), "2" (__z), "3" (__w) ); \
+        x2 = __x; y2 = __y; z2 = __z; w2 = __w; \
+    } while(false)
+
+#else
+
+#define mat_trans_nodiv_nomod(x_, y_, z_, x2, y2, z2, w2) do { \
+		vector_t tmp = { x_, y_, z_, 1.0f }; \
+		mat_transform(&tmp, &tmp, 1, 0); \
+		x2 = tmp.x; y2 = tmp.y; z2 = tmp.z; w2 = tmp.w; \
+	} while(false)
+#endif
+
+
 // wrapper around float versions of functions
 // in gta they are in CMaths but that makes the code rather noisy
 

src/miami/modelinfo/BaseModelInfo.cpp

@@ -40,6 +40,14 @@ CBaseModelInfo::DeleteCollisionModel(void)
 	}
 }
 
+void CBaseModelInfo::SetColModel(CColModel *col, bool owns) {
+	if (m_bOwnsColModel) {
+		delete m_colModel;
+	}
+	m_colModel = col;
+	m_bOwnsColModel = owns;
+}
+
 void
 CBaseModelInfo::AddRef(void)
 {

src/miami/modelinfo/BaseModelInfo.h

@@ -52,8 +52,7 @@ public:
 	bool IsClump(void) { return m_type == MITYPE_CLUMP || m_type == MITYPE_PED || m_type == MITYPE_VEHICLE;	}
 	char *GetModelName(void) { return m_name; }
 	void SetModelName(const char *name) { strncpy(m_name, name, MAX_MODEL_NAME); }
-	void SetColModel(CColModel *col, bool owns = false){
-		m_colModel = col; m_bOwnsColModel = owns; }
+	void SetColModel(CColModel *col, bool owns = false);
 	CColModel *GetColModel(void) { return m_colModel; }
 	bool DoesOwnColModel(void) { return m_bOwnsColModel; }
 	void DeleteCollisionModel(void);

src/miami/renderer/ShadowCamera.cpp

@@ -271,13 +271,13 @@ CShadowCamera::InvertRaster()
 	RwIm2DVertexSetIntRGBA     (&vx[1], 255, 255, 255, 255);
 
 	RwIm2DVertexSetScreenX     (&vx[2], crw);
-	RwIm2DVertexSetScreenY     (&vx[2], 0.0f);
+	RwIm2DVertexSetScreenY     (&vx[2], crh);
 	RwIm2DVertexSetScreenZ     (&vx[2], RwIm2DGetNearScreenZ());
 	RwIm2DVertexSetRecipCameraZ(&vx[2], recipZ);
 	RwIm2DVertexSetIntRGBA     (&vx[2], 255, 255, 255, 255);
 
 	RwIm2DVertexSetScreenX     (&vx[3], crw);
-	RwIm2DVertexSetScreenY     (&vx[3], crh);
+	RwIm2DVertexSetScreenY     (&vx[3], 0.0f);
 	RwIm2DVertexSetScreenZ     (&vx[3], RwIm2DGetNearScreenZ());
 	RwIm2DVertexSetRecipCameraZ(&vx[3], recipZ);
 	RwIm2DVertexSetIntRGBA     (&vx[3], 255, 255, 255, 255);
@@ -289,7 +289,7 @@ CShadowCamera::InvertRaster()
 	RwRenderStateSet(rwRENDERSTATESRCBLEND,          (void *)rwBLENDINVDESTCOLOR);
 	RwRenderStateSet(rwRENDERSTATEDESTBLEND,         (void *)rwBLENDZERO);
 
-	RwIm2DRenderPrimitive(rwPRIMTYPETRISTRIP, vx, 4);
+	RwIm2DRenderPrimitive(rwPRIMTYPETRIFAN, vx, 4);
 
 	RwRenderStateSet(rwRENDERSTATEZTESTENABLE,       (void *)TRUE);
 	RwRenderStateSet(rwRENDERSTATESRCBLEND,          (void *)rwBLENDSRCALPHA);

src/miami/renderer/Weather.h

@@ -60,7 +60,11 @@ public:
 };
 
 enum {
+#ifdef RW_DC
+	NUM_RAIN_STREAKS = 10
+#elif
 	NUM_RAIN_STREAKS = 35
+#endif
 };
 
 struct tRainStreak

src/miami/rw/RwHelper.cpp

@@ -384,23 +384,23 @@ RwBool Im2DRenderQuad(RwReal x1, RwReal y1, RwReal x2, RwReal y2, RwReal z, RwRe
     RwIm2DVertexSetU(&vx[1], uvOffset, recipCamZ);
     RwIm2DVertexSetV(&vx[1], 1.0f + uvOffset, recipCamZ);
 
-    RwIm2DVertexSetScreenX(&vx[2], x2);
-    RwIm2DVertexSetScreenY(&vx[2], y1);
+	RwIm2DVertexSetScreenX(&vx[2], x2);
+    RwIm2DVertexSetScreenY(&vx[2], y2);
     RwIm2DVertexSetScreenZ(&vx[2], z);
     RwIm2DVertexSetIntRGBA(&vx[2], 255, 255, 255, 255);
     RwIm2DVertexSetRecipCameraZ(&vx[2], recipCamZ);
     RwIm2DVertexSetU(&vx[2], 1.0f + uvOffset, recipCamZ);
-    RwIm2DVertexSetV(&vx[2], uvOffset, recipCamZ);
-
+    RwIm2DVertexSetV(&vx[2], 1.0f + uvOffset, recipCamZ);
+	
     RwIm2DVertexSetScreenX(&vx[3], x2);
-    RwIm2DVertexSetScreenY(&vx[3], y2);
+    RwIm2DVertexSetScreenY(&vx[3], y1);
     RwIm2DVertexSetScreenZ(&vx[3], z);
     RwIm2DVertexSetIntRGBA(&vx[3], 255, 255, 255, 255);
     RwIm2DVertexSetRecipCameraZ(&vx[3], recipCamZ);
     RwIm2DVertexSetU(&vx[3], 1.0f + uvOffset, recipCamZ);
-    RwIm2DVertexSetV(&vx[3], 1.0f + uvOffset, recipCamZ);
+    RwIm2DVertexSetV(&vx[3], uvOffset, recipCamZ);
 
-    RwIm2DRenderPrimitive(rwPRIMTYPETRISTRIP, vx, 4);
+    RwIm2DRenderPrimitive(rwPRIMTYPETRIFAN, vx, 4);
 
     return TRUE;
 }

src/miami/save/PCSave.cpp

@@ -17,6 +17,8 @@
 
 #include "vmu/vmu.h"
 
+void* re3StreamingAlloc(size_t size);
+
 const char* _psGetUserFilesFolder();
 
 C_PcSave PcSaveHelper;
@@ -76,31 +78,37 @@ C_PcSave::SaveSlot(int32 slot)
 bool
 C_PcSave::PcClassSaveRoutine(int32 file, uint8 *data, uint32 size)
 {
-	void* wrkmem = malloc(LZO1X_1_MEM_COMPRESS);
-	uint8* compressed = (uint8*)malloc(size*2);
+	void* wrkmem = re3StreamingAlloc(LZO1X_1_MEM_COMPRESS);
+	assert(wrkmem);
+	uint8* compressed = (uint8*)re3StreamingAlloc(size*2);
+	assert(compressed);
 	lzo_uint compressed_size;
 	int crv = lzo1x_1_compress(data, size, compressed, &compressed_size, wrkmem);
-	free(wrkmem);
+	RwFree(wrkmem);
+
+	if (crv == LZO_E_OK && compressed_size >= size) {
+		crv = LZO_E_NOT_COMPRESSIBLE;
+	}
 
 	if (crv == LZO_E_OK) {
 		uint32_t compressed_size32 = compressed_size | 0x80000000;
 		bool err = CFileMgr::Write(file, (const char*)&compressed_size32, sizeof(compressed_size32)) != sizeof(compressed_size32);
 		if (err || CFileMgr::GetErrorReadWrite(file)) {
-			free(compressed);
+			RwFree(compressed);
 			nErrorCode = SAVESTATUS_ERR_SAVE_WRITE;
 			strncpy(SaveFileNameJustSaved, ValidSaveName, sizeof(ValidSaveName) - 1);
 			return false;
 		}
 
 		err = CFileMgr::Write(file, (const char*)compressed, compressed_size) != compressed_size;
-		free(compressed);
+		RwFree(compressed);
 		if (err || CFileMgr::GetErrorReadWrite(file)) {
 			nErrorCode = SAVESTATUS_ERR_SAVE_WRITE;
 			strncpy(SaveFileNameJustSaved, ValidSaveName, sizeof(ValidSaveName) - 1);
 			return false;
 		}
 	} else if (crv == LZO_E_NOT_COMPRESSIBLE) {
-		free(compressed);
+		RwFree(compressed);
 		uint32_t compressed_size32 = size;
 		bool err = CFileMgr::Write(file, (const char*)&compressed_size32, sizeof(compressed_size32)) != sizeof(compressed_size32);
 		if (err || CFileMgr::GetErrorReadWrite(file)) {
@@ -115,7 +123,7 @@ C_PcSave::PcClassSaveRoutine(int32 file, uint8 *data, uint32 size)
 			return false;
 		}
 	} else {
-		free(compressed);
+		RwFree(compressed);
 		return false;
 	}
 
@@ -153,16 +161,17 @@ uint32_t C_PcSave::PcClassLoadRoutine(int32 file, uint8 *data) {
 		return size;
 	} else {
 		size &= ~0x80000000;
-		uint8* compressed = (uint8*)malloc(size);
+		uint8* compressed = (uint8*)re3StreamingAlloc(size);
+		assert(compressed);
 		err = CFileMgr::Read(file, (const char*)compressed, size) != size;
 		if (err || CFileMgr::GetErrorReadWrite(file)) {
-			free(compressed);
+			RwFree(compressed);
 			return 0;
 		}
 
 		lzo_uint decompressed_size = 0;
 		auto crv = lzo1x_decompress(compressed, size, data, &decompressed_size, NULL);
-		free(compressed);
+		RwFree(compressed);
 		if (crv != LZO_E_OK) {
 			return 0;
 		}

src/tools/texconv.cpp

@@ -53,6 +53,9 @@ uint32_t pvr_map32(uint32_t offset32) {return 0;}
 void Hackpresent() { }
 void re3RemoveLeastUsedModel() { assert(false); }
 void re3EmergencyRemoveModel() { assert(false); }
+void* re3StreamingAlloc(size_t sz) {
+	return RwMalloc(sz);
+}
 void RwTexDictionaryGtaStreamRead1(rw::Stream*){ assert(false); }
 void RwTexDictionaryGtaStreamRead2(rw::Stream*, rw::TexDictionary*) { assert(false);  }
 void pvr_ta_data(void* data, int size) {

vendor/emu/emu/window.cpp

@@ -125,6 +125,8 @@ void x11_window_create()
     x11_win = (void*)x11Window;
     x11_vis = (void*)x11Visual->visual;
 
+    delete x11Visual;
+
     x11_window_set_text("GTA3dc");
 }
 

vendor/librw/src/anim.cpp

@@ -221,6 +221,7 @@ AnimInterpolator::setCurrentAnim(Animation *anim)
 {
 	int32 i;
 	AnimInterpolatorInfo *interpInfo = anim->interpInfo;
+	assert(this->currentAnim == nil || this->currentAnim  == anim);
 	this->currentAnim = anim;
 	this->currentTime = 0.0f;
 	int32 maxkf = this->maxInterpKeyFrameSize;