From: notaz <notasas@gmail.com>
Date: Thu, 5 Feb 2015 23:29:00 +0000 (+0200)
Subject: Merge branch 'upstream' into libretro
X-Git-Tag: r24l~889^2
X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=commitdiff_plain;h=2af7e74feec67c0a798ac190774cb4b3e925be76;hp=cebb70f24ab54693b12aaa27b85a52689e26e1ff

Merge branch 'upstream' into libretro
---

diff --git a/.travis.yml b/.travis.yml
new file mode 100644
index 00000000..7c4eafbd
--- /dev/null
+++ b/.travis.yml
@@ -0,0 +1,8 @@
+language: cpp
+compiler:
+  - gcc
+  - clang
+before_install:
+  - sudo apt-get update -qq
+  - sudo apt-get install -y libsdl1.2-dev libasound2-dev libpng-dev libz-dev
+script: ./configure && make
diff --git a/README b/README.md
similarity index 83%
rename from README
rename to README.md
index 23ae219f..99644108 100644
--- a/README
+++ b/README.md
@@ -1,12 +1,16 @@
 PCSX-ReARMed - yet another PCSX fork
--------------------------------
+====================================
+
+[![Build Status](https://travis-ci.org/notaz/pcsx_rearmed.svg?branch=master)](https://travis-ci.org/notaz/pcsx_rearmed)
+
+*see [readme.txt](readme.txt) for more complete documentation*
 
 PCSX ReARMed is yet another PCSX fork based on the PCSX-Reloaded project,
 which itself contains code from PCSX, PCSX-df and PCSX-Revolution. This
 version is ARM architecture oriented and features MIPS->ARM recompiler by
 Ari64, NEON GTE code and more performance improvements. It was created for
 Pandora handheld, but should be usable on other devices after some code
-adjustments (N900, GPH Wiz/Caanoo versions are also available).
+adjustments (N900, GPH Wiz/Caanoo, PlayBook versions are also available).
 
 PCSX ReARMed features ARM NEON GPU by Exophase, that in many cases produces
 pixel perfect graphics at very high performance. There is also Una-i's GPU
@@ -14,7 +18,7 @@ plugin from PCSX4ALL project, and traditional P.E.Op.S. one.
 
 
 PCSX-Reloaded
--------------------------------
+=============
 
 PCSX-Reloaded is a forked version of the dead PCSX emulator, with a nicer
 interface and several improvements to stability and functionality.
diff --git a/frontend/libpicofe b/frontend/libpicofe
index 515ac0b9..9fec8a91 160000
--- a/frontend/libpicofe
+++ b/frontend/libpicofe
@@ -1 +1 @@
-Subproject commit 515ac0b9d2c4d45a465335d54b8c49830914fcea
+Subproject commit 9fec8a91c9b19856ac0b51de53b847b38ed8dc61
diff --git a/frontend/libretro.c b/frontend/libretro.c
index 38514b5c..acc64a12 100644
--- a/frontend/libretro.c
+++ b/frontend/libretro.c
@@ -283,7 +283,7 @@ void retro_get_system_info(struct retro_system_info *info)
 {
 	memset(info, 0, sizeof(*info));
 	info->library_name = "PCSX-ReARMed";
-	info->library_version = "r21";
+	info->library_version = "r22";
 	info->valid_extensions = "bin|cue|img|mdf|pbp|toc|cbn|m3u";
 	info->need_fullpath = true;
 }
@@ -1255,6 +1255,7 @@ void retro_init(void)
 	cycle_multiplier = 200;
 #endif
 	pl_rearmed_cbs.gpu_peops.iUseDither = 1;
+	spu_config.iUseFixedUpdates = 1;
 
 	McdDisable[0] = 0;
 	McdDisable[1] = 1;
diff --git a/frontend/plat_omap.c b/frontend/plat_omap.c
index 4e3ea795..f25f31cc 100644
--- a/frontend/plat_omap.c
+++ b/frontend/plat_omap.c
@@ -111,10 +111,11 @@ void *plat_gvideo_set_mode(int *w_in, int *h_in, int *bpp)
 		}
 	}
 
-	vout_fbdev_clear(layer_fb);
 	buf = vout_fbdev_resize(layer_fb, w, h, *bpp,
 		l, r, t, b, 3);
 
+	vout_fbdev_clear(layer_fb);
+
 	omap_enable_layer(1);
 
 	return buf;
diff --git a/frontend/plugin_lib.c b/frontend/plugin_lib.c
index 56da763e..ab4d4152 100644
--- a/frontend/plugin_lib.c
+++ b/frontend/plugin_lib.c
@@ -34,6 +34,8 @@
 #include "../libpcsxcore/psxmem_map.h"
 #include "../plugins/dfinput/externals.h"
 
+#define HUD_HEIGHT 10
+
 int in_type1, in_type2;
 int in_a1[2] = { 127, 127 }, in_a2[2] = { 127, 127 };
 int in_adev[2] = { -1, -1 }, in_adev_axis[2][2] = {{ 0, 1 }, { 0, 1 }};
@@ -101,20 +103,20 @@ static void hud_printf(void *fb, int w, int x, int y, const char *texto, ...)
 
 static void print_msg(int h, int border)
 {
-	hud_print(pl_vout_buf, pl_vout_w, border + 2, h - 10, hud_msg);
+	hud_print(pl_vout_buf, pl_vout_w, border + 2, h - HUD_HEIGHT, hud_msg);
 }
 
 static void print_fps(int h, int border)
 {
-	hud_printf(pl_vout_buf, pl_vout_w, border + 2, h - 10,
+	hud_printf(pl_vout_buf, pl_vout_w, border + 2, h - HUD_HEIGHT,
 		"%2d %4.1f", pl_rearmed_cbs.flips_per_sec,
 		pl_rearmed_cbs.vsps_cur);
 }
 
 static void print_cpu_usage(int w, int h, int border)
 {
-	hud_printf(pl_vout_buf, pl_vout_w, pl_vout_w - border - 28, h - 10,
-		"%3d", pl_rearmed_cbs.cpu_usage);
+	hud_printf(pl_vout_buf, pl_vout_w, pl_vout_w - border - 28,
+		h - HUD_HEIGHT, "%3d", pl_rearmed_cbs.cpu_usage);
 }
 
 // draw 192x8 status of 24 sound channels
@@ -126,7 +128,7 @@ static __attribute__((noinline)) void draw_active_chans(int vout_w, int vout_h)
 
 	static const unsigned short colors[2] = { 0x1fe3, 0x0700 };
 	unsigned short *dest = (unsigned short *)pl_vout_buf +
-		vout_w * (vout_h - 10) + vout_w / 2 - 192/2;
+		vout_w * (vout_h - HUD_HEIGHT) + vout_w / 2 - 192/2;
 	unsigned short *d, p;
 	int c, x, y;
 
diff --git a/include/arm_features.h b/include/arm_features.h
index fdec5229..bb48a2bf 100644
--- a/include/arm_features.h
+++ b/include/arm_features.h
@@ -3,7 +3,7 @@
 
 #if defined(__ARM_ARCH_7__) || defined(__ARM_ARCH_7A__) \
  || defined(__ARM_ARCH_7R__) || defined(__ARM_ARCH_7M__) \
- || defined(__ARM_ARCH_7EM__)
+ || defined(__ARM_ARCH_7EM__) || defined(__ARM_ARCH_7S__)
 
 #define HAVE_ARMV7
 #define HAVE_ARMV6
diff --git a/libpcsxcore/cdrom.c b/libpcsxcore/cdrom.c
index 3a2eb1f6..556b512f 100644
--- a/libpcsxcore/cdrom.c
+++ b/libpcsxcore/cdrom.c
@@ -1413,13 +1413,15 @@ void psxDma3(u32 madr, u32 bcr, u32 chcr) {
 			psxCpu->Clear(madr, cdsize / 4);
 			pTransfer += cdsize;
 
-
-			// burst vs normal
 			if( chcr == 0x11400100 ) {
+				HW_DMA3_MADR = SWAPu32(madr + cdsize);
 				CDRDMA_INT( (cdsize/4) / 4 );
 			}
 			else if( chcr == 0x11000000 ) {
-				CDRDMA_INT( (cdsize/4) * 1 );
+				// CDRDMA_INT( (cdsize/4) * 1 );
+				// halted
+				psxRegs.cycle += (cdsize/4) * 24/2;
+				CDRDMA_INT(16);
 			}
 			return;
 
diff --git a/libpcsxcore/gte_arm.S b/libpcsxcore/gte_arm.S
index 3ef876dd..d6fc1ab8 100644
--- a/libpcsxcore/gte_arm.S
+++ b/libpcsxcore/gte_arm.S
@@ -11,7 +11,7 @@
 .align 2
 
 .macro sgnxt16 rd rs
-#ifdef HAVE_ARMV7
+#ifdef HAVE_ARMV6
     sxth     \rd, \rs
 #else
     lsl      \rd, \rs, #16
@@ -28,7 +28,7 @@
 .endm
 
 .macro ssatx rd wr bit
-#ifdef HAVE_ARMV7
+#ifdef HAVE_ARMV6
     ssat     \rd, #\bit, \rd
 #else
     cmp      \rd, \wr
@@ -52,7 +52,7 @@
 .endm
 
 .macro usat16_ rd rs
-#ifdef HAVE_ARMV7
+#ifdef HAVE_ARMV6
     usat     \rd, #16, \rs
 #else
     subs     \rd, \rs, #0
diff --git a/libpcsxcore/mdec.c b/libpcsxcore/mdec.c
index bed4e53b..d6c7ab68 100644
--- a/libpcsxcore/mdec.c
+++ b/libpcsxcore/mdec.c
@@ -32,7 +32,7 @@
  * 320x240x16@60Hz => 9.216 MB/s
  * so 2.0 to 4.0 should be fine.
  */
-#define MDEC_BIAS 2.0f
+#define MDEC_BIAS 2
 
 #define DSIZE			8
 #define DSIZE2			(DSIZE * DSIZE)
@@ -545,15 +545,13 @@ void psxDma1(u32 adr, u32 bcr, u32 chcr) {
 	int blk[DSIZE2 * 6];
 	u8 * image;
 	int size;
-	int dmacnt;
+	u32 words;
 
 	if (chcr != 0x01000200) return;
 
-	size = (bcr >> 16) * (bcr & 0xffff);
+	words = (bcr >> 16) * (bcr & 0xffff);
 	/* size in byte */
-	size *= 4;
-	/* I guess the memory speed is limitating */
-	dmacnt = size;
+	size = words * 4;
 
 	if (!(mdec.reg1 & MDEC1_BUSY)) {
 		/* add to pending */
@@ -625,7 +623,7 @@ void psxDma1(u32 adr, u32 bcr, u32 chcr) {
 	}
 	
 	/* define the power of mdec */
-	MDECOUTDMA_INT((int) ((dmacnt* MDEC_BIAS)));
+	MDECOUTDMA_INT(words * MDEC_BIAS);
 	}
 }
 
diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c
index 9ee832ef..6853e9e5 100644
--- a/libpcsxcore/new_dynarec/assem_arm.c
+++ b/libpcsxcore/new_dynarec/assem_arm.c
@@ -1285,7 +1285,7 @@ void emit_andimm(int rs,int imm,int rt)
     assem_debug("bic %s,%s,#%d\n",regname[rt],regname[rs],imm);
     output_w32(0xe3c00000|rd_rn_rm(rt,rs,0)|armval);
   }else if(imm==65535) {
-    #ifndef HAVE_ARMV7
+    #ifndef HAVE_ARMV6
     assem_debug("bic %s,%s,#FF000000\n",regname[rt],regname[rs]);
     output_w32(0xe3c00000|rd_rn_rm(rt,rs,0)|0x4FF);
     assem_debug("bic %s,%s,#00FF0000\n",regname[rt],regname[rt]);
@@ -1418,7 +1418,7 @@ void emit_shrdimm(int rs,int rs2,u_int imm,int rt)
 
 void emit_signextend16(int rs,int rt)
 {
-  #ifndef HAVE_ARMV7
+  #ifndef HAVE_ARMV6
   emit_shlimm(rs,16,rt);
   emit_sarimm(rt,16,rt);
   #else
@@ -1429,7 +1429,7 @@ void emit_signextend16(int rs,int rt)
 
 void emit_signextend8(int rs,int rt)
 {
-  #ifndef HAVE_ARMV7
+  #ifndef HAVE_ARMV6
   emit_shlimm(rs,24,rt);
   emit_sarimm(rt,24,rt);
   #else
diff --git a/libpcsxcore/new_dynarec/emu_if.c b/libpcsxcore/new_dynarec/emu_if.c
index 092c8ae9..22db5d11 100644
--- a/libpcsxcore/new_dynarec/emu_if.c
+++ b/libpcsxcore/new_dynarec/emu_if.c
@@ -378,7 +378,7 @@ static void ari64_clear(u32 addr, u32 size)
 {
 	u32 start, end, main_ram;
 
-	size *= 4; /* PCSX uses DMA units */
+	size *= 4; /* PCSX uses DMA units (words) */
 
 	evprintf("ari64_clear %08x %04x\n", addr, size);
 
diff --git a/libpcsxcore/ppf.c b/libpcsxcore/ppf.c
index 0d80107b..edebdd03 100644
--- a/libpcsxcore/ppf.c
+++ b/libpcsxcore/ppf.c
@@ -347,8 +347,10 @@ int LoadSBI(const char *fname, int sector_count) {
 		return -1;
 
 	sbi_sectors = calloc(1, sector_count / 8);
-	if (sbi_sectors == NULL)
+	if (sbi_sectors == NULL) {
+		fclose(sbihandle);
 		return -1;
+	}
 
 	// 4-byte SBI header
 	fread(buffer, 1, 4, sbihandle);
diff --git a/libpcsxcore/psxdma.c b/libpcsxcore/psxdma.c
index ff7d6a3b..03ee5639 100644
--- a/libpcsxcore/psxdma.c
+++ b/libpcsxcore/psxdma.c
@@ -37,7 +37,7 @@ void spuInterrupt() {
 
 void psxDma4(u32 madr, u32 bcr, u32 chcr) { // SPU
 	u16 *ptr;
-	u32 size;
+	u32 words;
 
 	switch (chcr) {
 		case 0x01000201: //cpu to spu transfer
@@ -51,8 +51,10 @@ void psxDma4(u32 madr, u32 bcr, u32 chcr) { // SPU
 #endif
 				break;
 			}
-			SPU_writeDMAMem(ptr, (bcr >> 16) * (bcr & 0xffff) * 2, psxRegs.cycle);
-			SPUDMA_INT((bcr >> 16) * (bcr & 0xffff) / 2);
+			words = (bcr >> 16) * (bcr & 0xffff);
+			SPU_writeDMAMem(ptr, words * 2, psxRegs.cycle);
+			HW_DMA4_MADR = SWAPu32(madr + words * 4);
+			SPUDMA_INT(words / 2);
 			return;
 
 		case 0x01000200: //spu to cpu transfer
@@ -66,10 +68,13 @@ void psxDma4(u32 madr, u32 bcr, u32 chcr) { // SPU
 #endif
 				break;
 			}
-			size = (bcr >> 16) * (bcr & 0xffff) * 2;
-			SPU_readDMAMem(ptr, size, psxRegs.cycle);
-			psxCpu->Clear(madr, size);
-			break;
+			words = (bcr >> 16) * (bcr & 0xffff);
+			SPU_readDMAMem(ptr, words * 2, psxRegs.cycle);
+			psxCpu->Clear(madr, words);
+
+			HW_DMA4_MADR = SWAPu32(madr + words * 4);
+			SPUDMA_INT(words / 2);
+			return;
 
 #ifdef PSXDMA_LOG
 		default:
@@ -124,6 +129,7 @@ static u32 gpuDmaChainSize(u32 addr) {
 
 void psxDma2(u32 madr, u32 bcr, u32 chcr) { // GPU
 	u32 *ptr;
+	u32 words;
 	u32 size;
 
 	switch (chcr) {
@@ -139,12 +145,14 @@ void psxDma2(u32 madr, u32 bcr, u32 chcr) { // GPU
 				break;
 			}
 			// BA blocks * BS words (word = 32-bits)
-			size = (bcr >> 16) * (bcr & 0xffff);
-			GPU_readDataMem(ptr, size);
-			psxCpu->Clear(madr, size);
+			words = (bcr >> 16) * (bcr & 0xffff);
+			GPU_readDataMem(ptr, words);
+			psxCpu->Clear(madr, words);
+
+			HW_DMA2_MADR = SWAPu32(madr + words * 4);
 
 			// already 32-bit word size ((size * 4) / 4)
-			GPUDMA_INT(size / 4);
+			GPUDMA_INT(words / 4);
 			return;
 
 		case 0x01000201: // mem2vram
@@ -159,11 +167,13 @@ void psxDma2(u32 madr, u32 bcr, u32 chcr) { // GPU
 				break;
 			}
 			// BA blocks * BS words (word = 32-bits)
-			size = (bcr >> 16) * (bcr & 0xffff);
-			GPU_writeDataMem(ptr, size);
+			words = (bcr >> 16) * (bcr & 0xffff);
+			GPU_writeDataMem(ptr, words);
+
+			HW_DMA2_MADR = SWAPu32(madr + words * 4);
 
 			// already 32-bit word size ((size * 4) / 4)
-			GPUDMA_INT(size / 4);
+			GPUDMA_INT(words / 4);
 			return;
 
 		case 0x01000401: // dma chain
@@ -175,7 +185,11 @@ void psxDma2(u32 madr, u32 bcr, u32 chcr) { // GPU
 			if ((int)size <= 0)
 				size = gpuDmaChainSize(madr);
 			HW_GPU_STATUS &= ~PSXGPU_nBUSY;
-			
+
+			// we don't emulate progress, just busy flag and end irq,
+			// so pretend we're already at the last block
+			HW_DMA2_MADR = SWAPu32(0xffffff);
+
 			// Tekken 3 = use 1.0 only (not 1.5x)
 
 			// Einhander = parse linked list in pieces (todo)
@@ -206,7 +220,7 @@ void gpuInterrupt() {
 }
 
 void psxDma6(u32 madr, u32 bcr, u32 chcr) {
-	u32 size;
+	u32 words;
 	u32 *mem = (u32 *)PSXM(madr);
 
 #ifdef PSXDMA_LOG
@@ -224,7 +238,7 @@ void psxDma6(u32 madr, u32 bcr, u32 chcr) {
 		}
 
 		// already 32-bit size
-		size = bcr;
+		words = bcr;
 
 		while (bcr--) {
 			*mem-- = SWAP32((madr - 4) & 0xffffff);
@@ -232,7 +246,10 @@ void psxDma6(u32 madr, u32 bcr, u32 chcr) {
 		}
 		mem++; *mem = 0xffffff;
 
-		GPUOTCDMA_INT(size);
+		//GPUOTCDMA_INT(size);
+		// halted
+		psxRegs.cycle += words;
+		GPUOTCDMA_INT(16);
 		return;
 	}
 #ifdef PSXDMA_LOG
diff --git a/plugins/dfsound/Makefile.c64p b/plugins/dfsound/Makefile.c64p
index 45fe76a4..5b977810 100644
--- a/plugins/dfsound/Makefile.c64p
+++ b/plugins/dfsound/Makefile.c64p
@@ -5,7 +5,9 @@ endif
 include $(C64_TOOLS_DSP_ROOT)/install.mk
 
 TARGET_BASENAME = pcsxr_spu
-OPTFLAGS += -O2
+OPTFLAGS += -o3
+CFLAGS += -mo
+#CFLAGS += -k -mw
 CFLAGS += -DNO_OS -DWANT_THREAD_CODE
 
 OBJ = \
diff --git a/plugins/dfsound/dma.c b/plugins/dfsound/dma.c
index 4798a19b..43019f1d 100644
--- a/plugins/dfsound/dma.c
+++ b/plugins/dfsound/dma.c
@@ -71,6 +71,7 @@ void CALLBACK SPUwriteDMA(unsigned short val)
 
  spu.spuAddr += 2;
  spu.spuAddr &= 0x7fffe;
+ spu.bMemDirty = 1;
 }
 
 ////////////////////////////////////////////////////////////////////////
@@ -83,6 +84,7 @@ void CALLBACK SPUwriteDMAMem(unsigned short *pusPSXMem, int iSize,
  int i;
  
  do_samples_if_needed(cycles, 1);
+ spu.bMemDirty = 1;
 
  if(spu.spuAddr + iSize*2 < 0x80000)
   {
diff --git a/plugins/dfsound/externals.h b/plugins/dfsound/externals.h
index f6fc4409..2db75ac2 100644
--- a/plugins/dfsound/externals.h
+++ b/plugins/dfsound/externals.h
@@ -162,14 +162,8 @@ typedef struct
 
  int dirty;          // registers changed
 
- // normalized offsets
- int nIIR_DEST_A0, nIIR_DEST_A1, nIIR_DEST_B0, nIIR_DEST_B1,
- 	nACC_SRC_A0, nACC_SRC_A1, nACC_SRC_B0, nACC_SRC_B1, 
-	nIIR_SRC_A0, nIIR_SRC_A1, nIIR_SRC_B0, nIIR_SRC_B1,
-	nACC_SRC_C0, nACC_SRC_C1, nACC_SRC_D0, nACC_SRC_D1,
-	nMIX_DEST_A0, nMIX_DEST_A1, nMIX_DEST_B0, nMIX_DEST_B1;
  // MIX_DEST_xx - FB_SRC_x
- int nFB_SRC_A0, nFB_SRC_A1, nFB_SRC_B0, nFB_SRC_B1;
+ int FB_SRC_A0, FB_SRC_A1, FB_SRC_B0, FB_SRC_B1;
 } REVERBInfo;
 
 ///////////////////////////////////////////////////////////
@@ -195,6 +189,7 @@ typedef struct
  int             decode_dirty_ch;
  unsigned int    bSpuInit:1;
  unsigned int    bSPUIsOpen:1;
+ unsigned int    bMemDirty:1;          // had external write to SPU RAM
 
  unsigned int    dwNoiseVal;           // global noise generator
  unsigned int    dwNoiseCount;
diff --git a/plugins/dfsound/freeze.c b/plugins/dfsound/freeze.c
index 5b4267bf..43e6535c 100644
--- a/plugins/dfsound/freeze.c
+++ b/plugins/dfsound/freeze.c
@@ -231,6 +231,8 @@ long CALLBACK SPUfreeze(uint32_t ulFreezeMode, SPUFreeze_t * pF,
 
  if(!pF) return 0;                                     // first check
 
+ do_samples(cycles, 1);
+
  if(ulFreezeMode)                                      // info or save?
   {//--------------------------------------------------//
    if(ulFreezeMode==1)                                 
@@ -242,8 +244,6 @@ long CALLBACK SPUfreeze(uint32_t ulFreezeMode, SPUFreeze_t * pF,
 
    if(ulFreezeMode==2) return 1;                       // info mode? ok, bye
                                                        // save mode:
-   do_samples(cycles, 1);
-
    memcpy(pF->cSPURam,spu.spuMem,0x80000);             // copy common infos
    memcpy(pF->cSPUPort,spu.regArea,0x200);
 
@@ -279,6 +279,7 @@ long CALLBACK SPUfreeze(uint32_t ulFreezeMode, SPUFreeze_t * pF,
 
  memcpy(spu.spuMem,pF->cSPURam,0x80000);               // get ram
  memcpy(spu.regArea,pF->cSPUPort,0x200);
+ spu.bMemDirty = 1;
 
  if(pF->xaS.nsamples<=4032)                            // start xa again
   SPUplayADPCMchannel(&pF->xaS);
diff --git a/plugins/dfsound/registers.c b/plugins/dfsound/registers.c
index 4588fa7c..91bcaf86 100644
--- a/plugins/dfsound/registers.c
+++ b/plugins/dfsound/registers.c
@@ -145,17 +145,6 @@ void CALLBACK SPUwriteRegister(unsigned long reg, unsigned short val,
       break;
     //-------------------------------------------------//
     case H_SPUReverbAddr:
-      if(val==0xFFFF || val<=0x200)
-       {spu.rvb->StartAddr=spu.rvb->CurrAddr=0;}
-      else
-       {
-        const long iv=(unsigned long)val<<2;
-        if(spu.rvb->StartAddr!=iv)
-         {
-          spu.rvb->StartAddr=(unsigned long)val<<2;
-          spu.rvb->CurrAddr=spu.rvb->StartAddr;
-         }
-       }
       goto rvbd;
     //-------------------------------------------------//
     case H_SPUirqAddr:
@@ -246,38 +235,38 @@ void CALLBACK SPUwriteRegister(unsigned long reg, unsigned short val,
       ReverbOn(16,24,val);
       break;
     //-------------------------------------------------//
-    case H_Reverb+0   : spu.rvb->FB_SRC_A=val*4;         goto rvbd;
-    case H_Reverb+2   : spu.rvb->FB_SRC_B=val*4;         goto rvbd;
-    case H_Reverb+4   : spu.rvb->IIR_ALPHA=(short)val;   goto rvbd;
-    case H_Reverb+6   : spu.rvb->ACC_COEF_A=(short)val;  goto rvbd;
-    case H_Reverb+8   : spu.rvb->ACC_COEF_B=(short)val;  goto rvbd;
-    case H_Reverb+10  : spu.rvb->ACC_COEF_C=(short)val;  goto rvbd;
-    case H_Reverb+12  : spu.rvb->ACC_COEF_D=(short)val;  goto rvbd;
-    case H_Reverb+14  : spu.rvb->IIR_COEF=(short)val;    goto rvbd;
-    case H_Reverb+16  : spu.rvb->FB_ALPHA=(short)val;    goto rvbd;
-    case H_Reverb+18  : spu.rvb->FB_X=(short)val;        goto rvbd;
-    case H_Reverb+20  : spu.rvb->IIR_DEST_A0=val*4;      goto rvbd;
-    case H_Reverb+22  : spu.rvb->IIR_DEST_A1=val*4;      goto rvbd;
-    case H_Reverb+24  : spu.rvb->ACC_SRC_A0=val*4;       goto rvbd;
-    case H_Reverb+26  : spu.rvb->ACC_SRC_A1=val*4;       goto rvbd;
-    case H_Reverb+28  : spu.rvb->ACC_SRC_B0=val*4;       goto rvbd;
-    case H_Reverb+30  : spu.rvb->ACC_SRC_B1=val*4;       goto rvbd;
-    case H_Reverb+32  : spu.rvb->IIR_SRC_A0=val*4;       goto rvbd;
-    case H_Reverb+34  : spu.rvb->IIR_SRC_A1=val*4;       goto rvbd;
-    case H_Reverb+36  : spu.rvb->IIR_DEST_B0=val*4;      goto rvbd;
-    case H_Reverb+38  : spu.rvb->IIR_DEST_B1=val*4;      goto rvbd;
-    case H_Reverb+40  : spu.rvb->ACC_SRC_C0=val*4;       goto rvbd;
-    case H_Reverb+42  : spu.rvb->ACC_SRC_C1=val*4;       goto rvbd;
-    case H_Reverb+44  : spu.rvb->ACC_SRC_D0=val*4;       goto rvbd;
-    case H_Reverb+46  : spu.rvb->ACC_SRC_D1=val*4;       goto rvbd;
-    case H_Reverb+48  : spu.rvb->IIR_SRC_B1=val*4;       goto rvbd;
-    case H_Reverb+50  : spu.rvb->IIR_SRC_B0=val*4;       goto rvbd;
-    case H_Reverb+52  : spu.rvb->MIX_DEST_A0=val*4;      goto rvbd;
-    case H_Reverb+54  : spu.rvb->MIX_DEST_A1=val*4;      goto rvbd;
-    case H_Reverb+56  : spu.rvb->MIX_DEST_B0=val*4;      goto rvbd;
-    case H_Reverb+58  : spu.rvb->MIX_DEST_B1=val*4;      goto rvbd;
-    case H_Reverb+60  : spu.rvb->IN_COEF_L=(short)val;   goto rvbd;
-    case H_Reverb+62  : spu.rvb->IN_COEF_R=(short)val;   goto rvbd;
+    case H_Reverb+0   : goto rvbd;
+    case H_Reverb+2   : goto rvbd;
+    case H_Reverb+4   : spu.rvb->IIR_ALPHA=(short)val;   break;
+    case H_Reverb+6   : spu.rvb->ACC_COEF_A=(short)val;  break;
+    case H_Reverb+8   : spu.rvb->ACC_COEF_B=(short)val;  break;
+    case H_Reverb+10  : spu.rvb->ACC_COEF_C=(short)val;  break;
+    case H_Reverb+12  : spu.rvb->ACC_COEF_D=(short)val;  break;
+    case H_Reverb+14  : spu.rvb->IIR_COEF=(short)val;    break;
+    case H_Reverb+16  : spu.rvb->FB_ALPHA=(short)val;    break;
+    case H_Reverb+18  : spu.rvb->FB_X=(short)val;        break;
+    case H_Reverb+20  : goto rvbd;
+    case H_Reverb+22  : goto rvbd;
+    case H_Reverb+24  : goto rvbd;
+    case H_Reverb+26  : goto rvbd;
+    case H_Reverb+28  : goto rvbd;
+    case H_Reverb+30  : goto rvbd;
+    case H_Reverb+32  : goto rvbd;
+    case H_Reverb+34  : goto rvbd;
+    case H_Reverb+36  : goto rvbd;
+    case H_Reverb+38  : goto rvbd;
+    case H_Reverb+40  : goto rvbd;
+    case H_Reverb+42  : goto rvbd;
+    case H_Reverb+44  : goto rvbd;
+    case H_Reverb+46  : goto rvbd;
+    case H_Reverb+48  : goto rvbd;
+    case H_Reverb+50  : goto rvbd;
+    case H_Reverb+52  : goto rvbd;
+    case H_Reverb+54  : goto rvbd;
+    case H_Reverb+56  : goto rvbd;
+    case H_Reverb+58  : goto rvbd;
+    case H_Reverb+60  : spu.rvb->IN_COEF_L=(short)val;   break;
+    case H_Reverb+62  : spu.rvb->IN_COEF_R=(short)val;   break;
    }
  return;
 
diff --git a/plugins/dfsound/reverb.c b/plugins/dfsound/reverb.c
index 2ff6edc8..ec570fb3 100644
--- a/plugins/dfsound/reverb.c
+++ b/plugins/dfsound/reverb.c
@@ -50,16 +50,16 @@ INLINE int rvb2ram_offs(int curr, int space, int iOff)
 
 // get_buffer content helper: takes care about wraps
 #define g_buffer(var) \
- ((int)(signed short)spu.spuMem[rvb2ram_offs(curr_addr, space, rvb->n##var)])
+ ((int)(signed short)spu.spuMem[rvb2ram_offs(curr_addr, space, rvb->var)])
 
 // saturate iVal and store it as var
 #define s_buffer(var, iVal) \
  ssat32_to_16(iVal); \
- spu.spuMem[rvb2ram_offs(curr_addr, space, rvb->n##var)] = iVal
+ spu.spuMem[rvb2ram_offs(curr_addr, space, rvb->var)] = iVal
 
 #define s_buffer1(var, iVal) \
  ssat32_to_16(iVal); \
- spu.spuMem[rvb2ram_offs(curr_addr, space, rvb->n##var + 1)] = iVal
+ spu.spuMem[rvb2ram_offs(curr_addr, space, rvb->var + 1)] = iVal
 
 ////////////////////////////////////////////////////////////////////////
 
@@ -174,43 +174,54 @@ static void MixREVERB_off(int *SSumLR, int ns_to, int curr_addr)
 static void REVERBPrep(void)
 {
  REVERBInfo *rvb = spu.rvb;
- int space = 0x40000 - rvb->StartAddr;
- int t;
- #define prep_offs(v) \
-   t = rvb->v; \
+ int space, t;
+
+ t = spu.regArea[(H_SPUReverbAddr - 0xc00) >> 1];
+ if (t == 0xFFFF || t <= 0x200)
+  spu.rvb->StartAddr = spu.rvb->CurrAddr = 0;
+ else if (spu.rvb->StartAddr != (t << 2))
+  spu.rvb->StartAddr = spu.rvb->CurrAddr = t << 2;
+
+ space = 0x40000 - rvb->StartAddr;
+
+ #define prep_offs(v, r) \
+   t = spu.regArea[(0x1c0 + r) >> 1] * 4; \
    while (t >= space) \
      t -= space; \
-   rvb->n##v = t
- #define prep_offs2(d, v1, v2) \
-   t = rvb->v1 - rvb->v2; \
+   rvb->v = t
+ #define prep_offs2(d, r1, r2) \
+   t = spu.regArea[(0x1c0 + r1) >> 1] * 4; \
+   t -= spu.regArea[(0x1c0 + r2) >> 1] * 4; \
+   while (t < 0) \
+     t += space; \
    while (t >= space) \
      t -= space; \
-   rvb->n##d = t
-
- prep_offs(IIR_SRC_A0);
- prep_offs(IIR_SRC_A1);
- prep_offs(IIR_SRC_B0);
- prep_offs(IIR_SRC_B1);
- prep_offs(IIR_DEST_A0);
- prep_offs(IIR_DEST_A1);
- prep_offs(IIR_DEST_B0);
- prep_offs(IIR_DEST_B1);
- prep_offs(ACC_SRC_A0);
- prep_offs(ACC_SRC_A1);
- prep_offs(ACC_SRC_B0);
- prep_offs(ACC_SRC_B1);
- prep_offs(ACC_SRC_C0);
- prep_offs(ACC_SRC_C1);
- prep_offs(ACC_SRC_D0);
- prep_offs(ACC_SRC_D1);
- prep_offs(MIX_DEST_A0);
- prep_offs(MIX_DEST_A1);
- prep_offs(MIX_DEST_B0);
- prep_offs(MIX_DEST_B1);
- prep_offs2(FB_SRC_A0, MIX_DEST_A0, FB_SRC_A);
- prep_offs2(FB_SRC_A1, MIX_DEST_A1, FB_SRC_A);
- prep_offs2(FB_SRC_B0, MIX_DEST_B0, FB_SRC_B);
- prep_offs2(FB_SRC_B1, MIX_DEST_B1, FB_SRC_B);
+   rvb->d = t
+
+ prep_offs(IIR_SRC_A0, 32);
+ prep_offs(IIR_SRC_A1, 34);
+ prep_offs(IIR_SRC_B0, 36);
+ prep_offs(IIR_SRC_B1, 38);
+ prep_offs(IIR_DEST_A0, 20);
+ prep_offs(IIR_DEST_A1, 22);
+ prep_offs(IIR_DEST_B0, 36);
+ prep_offs(IIR_DEST_B1, 38);
+ prep_offs(ACC_SRC_A0, 24);
+ prep_offs(ACC_SRC_A1, 26);
+ prep_offs(ACC_SRC_B0, 28);
+ prep_offs(ACC_SRC_B1, 30);
+ prep_offs(ACC_SRC_C0, 40);
+ prep_offs(ACC_SRC_C1, 42);
+ prep_offs(ACC_SRC_D0, 44);
+ prep_offs(ACC_SRC_D1, 46);
+ prep_offs(MIX_DEST_A0, 52);
+ prep_offs(MIX_DEST_A1, 54);
+ prep_offs(MIX_DEST_B0, 56);
+ prep_offs(MIX_DEST_B1, 58);
+ prep_offs2(FB_SRC_A0, 52, 0);
+ prep_offs2(FB_SRC_A1, 54, 0);
+ prep_offs2(FB_SRC_B0, 56, 2);
+ prep_offs2(FB_SRC_B1, 58, 2);
 
 #undef prep_offs
 #undef prep_offs2
diff --git a/plugins/dfsound/spu.c b/plugins/dfsound/spu.c
index 8681d354..c6a1bf5e 100644
--- a/plugins/dfsound/spu.c
+++ b/plugins/dfsound/spu.c
@@ -818,12 +818,8 @@ static void do_channels(int ns_to)
   }
 
   if (spu.rvb->StartAddr) {
-   if (do_rvb) {
-    if (unlikely(spu.rvb->dirty))
-     REVERBPrep();
-
+   if (do_rvb)
     REVERBDo(spu.SSumLR, RVB, ns_to, spu.rvb->CurrAddr);
-   }
 
    spu.rvb->CurrAddr += ns_to / 2;
    while (spu.rvb->CurrAddr >= 0x40000)
@@ -846,6 +842,7 @@ static struct spu_worker {
    unsigned int i_ready;
    unsigned int i_reaped;
    unsigned int last_boot_cnt; // dsp
+   unsigned int ram_dirty;
   };
   // aligning for C64X_DSP
   unsigned int _pad0[128/4];
@@ -970,11 +967,8 @@ static void queue_channel_work(int ns_to, unsigned int silentch)
 
  work->rvb_addr = 0;
  if (spu.rvb->StartAddr) {
-  if (spu_config.iUseReverb) {
-   if (unlikely(spu.rvb->dirty))
-    REVERBPrep();
+  if (spu_config.iUseReverb)
    work->rvb_addr = spu.rvb->CurrAddr;
-  }
 
   spu.rvb->CurrAddr += ns_to / 2;
   while (spu.rvb->CurrAddr >= 0x40000)
@@ -989,9 +983,9 @@ static void do_channel_work(struct work_item *work)
 {
  unsigned int mask;
  unsigned int decode_dirty_ch = 0;
+ const SPUCHAN *s_chan;
  int *SB, sinc, spos, sbpos;
  int d, ch, ns_to;
- SPUCHAN *s_chan;
 
  ns_to = work->ns_to;
 
@@ -1057,8 +1051,12 @@ static void sync_worker_thread(int force)
  struct work_item *work;
  int done, used_space;
 
+ // rvb offsets will change, thread may be using them
+ force |= spu.rvb->dirty && spu.rvb->StartAddr;
+
  done = thread_get_i_done() - worker->i_reaped;
  used_space = worker->i_ready - worker->i_reaped;
+
  //printf("done: %d use: %d dsp: %u/%u\n", done, used_space,
  //  worker->boot_cnt, worker->i_done);
 
@@ -1150,6 +1148,9 @@ void do_samples(unsigned int cycles_to, int do_direct)
      }
    }
 
+  if (unlikely(spu.rvb->dirty))
+   REVERBPrep();
+
   if (do_direct || worker == NULL || !spu_config.iUseThread) {
    do_channels(ns_to);
    do_samples_finish(spu.SSumLR, ns_to, silentch, spu.decode_pos);
@@ -1252,7 +1253,7 @@ void schedule_next_irq(void)
 
 void CALLBACK SPUasync(unsigned int cycle, unsigned int flags)
 {
- do_samples(cycle, 0);
+ do_samples(cycle, spu_config.iUseFixedUpdates);
 
  if (spu.spuCtrl & CTRL_IRQ)
   schedule_next_irq();
diff --git a/plugins/dfsound/spu_c64x.c b/plugins/dfsound/spu_c64x.c
index be10a6bb..e0aa0da2 100644
--- a/plugins/dfsound/spu_c64x.c
+++ b/plugins/dfsound/spu_c64x.c
@@ -47,6 +47,25 @@ static struct {
  unsigned int req_sent:1;
 } f;
 
+static noinline void dsp_fault(void)
+{
+ dsp_msg_t msg;
+
+ f.dsp_cache_inv_virt(worker, sizeof(*worker));
+ printf("dsp crash/fault/corruption:\n");
+ printf("state rdy/reap/done: %u %u %u\n",
+   worker->i_ready, worker->i_reaped, worker->i_done);
+ printf("active/boot: %u %u\n",
+   worker->active, worker->boot_cnt);
+
+ if (f.req_sent) {
+  f.dsp_rpc_recv(&msg);
+  f.req_sent = 0;
+ }
+ f.dsp_logbuf_print();
+ spu_config.iUseThread = 0;
+}
+
 static void thread_work_start(void)
 {
  struct region_mem *mem;
@@ -84,9 +103,11 @@ static void thread_work_start(void)
 
  f.dsp_cache_inv_virt(&worker->i_done, 64);
  worker->last_boot_cnt = worker->boot_cnt;
+ worker->ram_dirty = spu.bMemDirty;
+ spu.bMemDirty = 0;
 
  mem = (void *)f.region.virt_addr;
- memcpy(&mem->spu_config, &spu_config, sizeof(mem->spu_config));
+ memcpy(&mem->in.spu_config, &spu_config, sizeof(mem->in.spu_config));
 
  DSP_MSG_INIT(&msg, f.compid, CCMD_DOIT, f.region.phys_addr, 0);
  ret = f.dsp_rpc_send(&msg);
@@ -97,6 +118,11 @@ static void thread_work_start(void)
   return;
  }
  f.req_sent = 1;
+
+#if 0
+ f.dsp_rpc_recv(&msg);
+ f.req_sent = 0;
+#endif
 }
 
 static int thread_get_i_done(void)
@@ -110,6 +136,11 @@ static void thread_work_wait_sync(struct work_item *work, int force)
  int limit = 1000;
  int ns_to;
 
+ if ((unsigned int)(worker->i_done - worker->i_reaped) > WORK_MAXCNT) {
+  dsp_fault();
+  return;
+ }
+
  while (worker->i_done == worker->i_reaped && limit-- > 0) {
   if (!f.req_sent) {
    printf("dsp: req not sent?\n");
@@ -222,7 +253,7 @@ static void init_spu_thread(void)
  }
  mem = (void *)f.region.virt_addr;
 
- memcpy(&mem->spu_config, &spu_config, sizeof(mem->spu_config));
+ memcpy(&mem->in.spu_config, &spu_config, sizeof(mem->in.spu_config));
 
  DSP_MSG_INIT(&init_msg, f.compid, CCMD_INIT, f.region.phys_addr, 0);
  ret = f.dsp_rpc(&init_msg, &msg_in);
@@ -301,4 +332,23 @@ static void exit_spu_thread(void)
  worker = NULL;
 }
 
+/* debug: "access" shared mem from gdb */
+#if 0
+struct region_mem *dbg_dsp_mem;
+
+void dbg_dsp_mem_update(void)
+{
+ struct region_mem *mem;
+
+ if (dbg_dsp_mem == NULL)
+  dbg_dsp_mem = malloc(sizeof(*dbg_dsp_mem));
+ if (dbg_dsp_mem == NULL)
+  return;
+
+ mem = (void *)f.region.virt_addr;
+ f.dsp_cache_inv_virt(mem, sizeof(*mem));
+ memcpy(dbg_dsp_mem, mem, sizeof(*dbg_dsp_mem));
+}
+#endif
+
 // vim:shiftwidth=1:expandtab
diff --git a/plugins/dfsound/spu_c64x.h b/plugins/dfsound/spu_c64x.h
index 8210e63a..7c4d5659 100644
--- a/plugins/dfsound/spu_c64x.h
+++ b/plugins/dfsound/spu_c64x.h
@@ -15,10 +15,10 @@ struct region_mem {
   // these are not to be modified by DSP
   SPUCHAN s_chan[24 + 1];
   REVERBInfo rvb;
+  SPUConfig spu_config;
  } in;
  int _pad2[128/4 - ((sizeof(struct spu_in) / 4) & (128/4 - 1))];
  struct spu_worker worker;
- SPUConfig spu_config;
  // init/debug
  int sizeof_region_mem;
  int offsetof_s_chan1;
diff --git a/plugins/dfsound/spu_c64x_dspcode.c b/plugins/dfsound/spu_c64x_dspcode.c
index b0352a9f..570da5ed 100644
--- a/plugins/dfsound/spu_c64x_dspcode.c
+++ b/plugins/dfsound/spu_c64x_dspcode.c
@@ -37,13 +37,40 @@ struct out_driver *out_current;
 void SetupSound(void) {}
 
 
+static void enable_l2_cache(void)
+{
+ volatile uint32_t *L2CFG = (volatile uint32_t *)0x01840000;
+ uint32_t *MARi = (void *)0x01848000;
+ int i;
+
+ // program Memory Attribute Registers
+ // (old c64_tools has the defaults messed up)
+ // 00000000-0fffffff - not configurable
+ // 10000000-7fffffff - system
+ for (i = 0x10; i < 0x80; i++)
+  MARi[i] = 0;
+ // 80000000-9fffffff - RAM
+ for ( ; i < 0xa0; i++)
+  MARi[i] = 1;
+ // 0xa00000-ffffffff - reserved, etc
+ for ( ; i < 0x100; i++)
+  MARi[i] = 0;
+
+ // enable L2 (1 for 32k, 2 for 64k)
+ if (!(*L2CFG & 2)) {
+  *L2CFG = 2;
+  // wait the for the write
+  *L2CFG;
+ }
+}
+
 static void invalidate_cache(struct work_item *work)
 {
  // see comment in writeout_cache()
  //syscalls.cache_inv(work, offsetof(typeof(*work), SSumLR), 1);
- syscalls.cache_inv(spu.s_chan, sizeof(spu.s_chan[0]) * 24, 0);
+ syscalls.cache_inv(spu.s_chan, sizeof(spu.s_chan[0]) * 24, 1);
  syscalls.cache_inv(work->SSumLR,
-   sizeof(work->SSumLR[0]) * 2 * work->ns_to, 0);
+   sizeof(work->SSumLR[0]) * 2 * work->ns_to, 1);
 }
 
 static void writeout_cache(struct work_item *work)
@@ -53,7 +80,7 @@ static void writeout_cache(struct work_item *work)
  syscalls.cache_wb(work->SSumLR, sizeof(work->SSumLR[0]) * 2 * ns_to, 1);
  // have to invalidate now, otherwise there is a race between
  // DSP evicting dirty lines and ARM writing new data to this area
- syscalls.cache_inv(work, offsetof(typeof(*work), SSumLR), 0);
+ syscalls.cache_inv(work, offsetof(typeof(*work), SSumLR), 1);
 }
 
 static void do_processing(void)
@@ -112,6 +139,7 @@ static unsigned int exec(dsp_component_cmd_t cmd,
 
  switch (cmd) {
   case CCMD_INIT:
+   enable_l2_cache();
    InitADSR();
 
    spu.spuMemC = mem->spu_ram;
@@ -119,7 +147,7 @@ static unsigned int exec(dsp_component_cmd_t cmd,
    spu.s_chan = mem->in.s_chan;
    spu.rvb = &mem->in.rvb;
    worker = &mem->worker;
-   memcpy(&spu_config, &mem->spu_config, sizeof(spu_config));
+   memcpy(&spu_config, &mem->in.spu_config, sizeof(spu_config));
 
    mem->sizeof_region_mem = sizeof(*mem);
    mem->offsetof_s_chan1 = offsetof(typeof(*mem), in.s_chan[1]);
@@ -131,16 +159,18 @@ static unsigned int exec(dsp_component_cmd_t cmd,
   case CCMD_DOIT:
    worker->active = ACTIVE_CNT;
    worker->boot_cnt++;
-   syscalls.cache_wb(&worker->i_done, 64, 1);
-   memcpy(&spu_config, &mem->spu_config, sizeof(spu_config));
+   syscalls.cache_inv(worker, 128, 1);
+   syscalls.cache_wb(&worker->i_done, 128, 1);
+   memcpy(&spu_config, &mem->in.spu_config, sizeof(spu_config));
+
+   if (worker->ram_dirty)
+    // it's faster to do it all than just a 512k buffer
+    syscalls.cache_wbInvAll();
 
    do_processing();
 
-   // c64_tools lib does BCACHE_wbInvAll() when it receives mailbox irq,
-   // but invalidate anyway in case c64_tools is ever fixed..
-   // XXX edit: don't bother as reverb is not handled, will fix if needed
-   //syscalls.cache_inv(mem, sizeof(mem->spu_ram) + sizeof(mem->SB), 0);
-   //syscalls.cache_inv(&mem->in, sizeof(mem->in), 0);
+   syscalls.cache_inv(&mem->SB, sizeof(mem->SB), 0);
+   syscalls.cache_inv(&mem->in, sizeof(mem->in), 0);
    break;
 
   default:
diff --git a/plugins/dfsound/spu_config.h b/plugins/dfsound/spu_config.h
index fce1cdaa..3e88a2c2 100644
--- a/plugins/dfsound/spu_config.h
+++ b/plugins/dfsound/spu_config.h
@@ -8,6 +8,7 @@ typedef struct
  int        iUseInterpolation;
  int        iTempo;
  int        iUseThread;
+ int        iUseFixedUpdates;  // output fixed number of samples/frame
 
  // status
  int        iThreadAvail;
diff --git a/plugins/gpu_unai/gpu.cpp b/plugins/gpu_unai/gpu.cpp
index d5096170..1552bed9 100644
--- a/plugins/gpu_unai/gpu.cpp
+++ b/plugins/gpu_unai/gpu.cpp
@@ -432,7 +432,7 @@ void  GPU_readDataMem(u32* dmaAddress, s32 dmaCount)
 	{
 		if ((&pvram[px])>(VIDEO_END)) pvram-=512*1024;
 		// lower 16 bit
-		u32 data = (unsigned long)pvram[px];
+		u32 data = pvram[px];
 
 		if (++px>=x_end) 
 		{
@@ -442,7 +442,7 @@ void  GPU_readDataMem(u32* dmaAddress, s32 dmaCount)
 
 		if ((&pvram[px])>(VIDEO_END)) pvram-=512*1024;
 		// higher 16 bit (always, even if it's an odd width)
-		data |= (unsigned long)(pvram[px])<<16;
+		data |= (u32)(pvram[px])<<16;
 		
 		*dmaAddress++ = data;
 
diff --git a/plugins/gpu_unai/gpu_command.h b/plugins/gpu_unai/gpu_command.h
index 049b1462..d6e7a742 100644
--- a/plugins/gpu_unai/gpu_command.h
+++ b/plugins/gpu_unai/gpu_command.h
@@ -21,9 +21,9 @@
 ///////////////////////////////////////////////////////////////////////////////
 INLINE void gpuSetTexture(u16 tpage)
 {
-	long tp;
-	long tx, ty;
-	GPU_GP1 = (GPU_GP1 & ~0x7FF) | (tpage & 0x7FF);
+	u32 tp;
+	u32 tx, ty;
+	GPU_GP1 = (GPU_GP1 & ~0x1FF) | (tpage & 0x1FF);
 
 	TextureWindow[0]&= ~TextureWindow[2];
 	TextureWindow[1]&= ~TextureWindow[3];
@@ -31,6 +31,7 @@ INLINE void gpuSetTexture(u16 tpage)
 	tp = (tpage >> 7) & 3;
 	tx = (tpage & 0x0F) << 6;
 	ty = (tpage & 0x10) << 4;
+	if (tp == 3) tp = 2;
 
 	tx += (TextureWindow[0] >> (2 - tp));
 	ty += TextureWindow[1];
@@ -437,8 +438,8 @@ void gpuSendPacketFunction(const int PRIM)
 		case 0xE5:
 			{
 				const u32 temp = PacketBuffer.U4[0];
-				DrawingOffset[0] = ((long)temp<<(32-11))>>(32-11);
-				DrawingOffset[1] = ((long)temp<<(32-22))>>(32-11);
+				DrawingOffset[0] = ((s32)temp<<(32-11))>>(32-11);
+				DrawingOffset[1] = ((s32)temp<<(32-22))>>(32-11);
 				//isSkip = false;
 				DO_LOG(("DrawingOffset(0x%x)\n",PRIM));
 			}
diff --git a/plugins/gpu_unai/gpu_raster_line.h b/plugins/gpu_unai/gpu_raster_line.h
index 4edfa06f..fc59b797 100644
--- a/plugins/gpu_unai/gpu_raster_line.h
+++ b/plugins/gpu_unai/gpu_raster_line.h
@@ -26,7 +26,7 @@
 #define GPU_DIGITS  16
 #define GPU_DIGITSC (GPU_DIGITS+3)
 
-INLINE long GPU_DIV(long rs, long rt)
+INLINE s32 GPU_DIV(s32 rs, s32 rt)
 {
 	return rt ? (rs / rt) : (0);
 }
diff --git a/plugins/gpu_unai/gpulib_if.cpp b/plugins/gpu_unai/gpulib_if.cpp
index 0d506bc7..646b0f22 100644
--- a/plugins/gpu_unai/gpulib_if.cpp
+++ b/plugins/gpu_unai/gpulib_if.cpp
@@ -490,8 +490,8 @@ int do_cmd_list(unsigned int *list, int list_len, int *last_cmd)
       }
       case 0xE5: {
         const u32 temp = PacketBuffer.U4[0];
-        DrawingOffset[0] = ((long)temp<<(32-11))>>(32-11);
-        DrawingOffset[1] = ((long)temp<<(32-22))>>(32-11);
+        DrawingOffset[0] = ((s32)temp<<(32-11))>>(32-11);
+        DrawingOffset[1] = ((s32)temp<<(32-22))>>(32-11);
         gpu.ex_regs[5] = temp;
         break;
       }
diff --git a/plugins/gpulib/gpu.c b/plugins/gpulib/gpu.c
index 0e6fe63d..125bd89b 100644
--- a/plugins/gpulib/gpu.c
+++ b/plugins/gpulib/gpu.c
@@ -21,7 +21,6 @@
 #define unlikely(x)
 #define preload(...)
 #define noinline
-#error huh
 #endif
 
 #define gpu_log(fmt, ...) \
@@ -522,7 +521,6 @@ void GPUwriteData(uint32_t data)
 long GPUdmaChain(uint32_t *rambase, uint32_t start_addr)
 {
   uint32_t addr, *list, ld_addr = 0;
-  uint32_t *llist_entry = NULL;
   int len, left, count;
   long cpu_cycles = 0;
 
@@ -531,15 +529,6 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr)
   if (unlikely(gpu.cmd_len > 0))
     flush_cmd_buffer();
 
-  // ff7 sends it's main list twice, detect this
-  if (*gpu.state.frame_count == gpu.state.last_list.frame &&
-      *gpu.state.hcnt - gpu.state.last_list.hcnt <= 1 &&
-       gpu.state.last_list.cycles > 2048)
-  {
-    llist_entry = rambase + (gpu.state.last_list.addr & 0x1fffff) / 4;
-    *llist_entry |= 0x800000;
-  }
-
   log_io("gpu_dma_chain\n");
   addr = start_addr & 0xffffff;
   for (count = 0; (addr & 0x800000) == 0; count++)
@@ -586,9 +575,6 @@ long GPUdmaChain(uint32_t *rambase, uint32_t start_addr)
     }
   }
 
-  if (llist_entry)
-    *llist_entry &= ~0x800000;
-
   gpu.state.last_list.frame = *gpu.state.frame_count;
   gpu.state.last_list.hcnt = *gpu.state.hcnt;
   gpu.state.last_list.cycles = cpu_cycles;
diff --git a/plugins/gpulib/vout_pl.c b/plugins/gpulib/vout_pl.c
index 541b5e0b..a9437cb1 100644
--- a/plugins/gpulib/vout_pl.c
+++ b/plugins/gpulib/vout_pl.c
@@ -89,11 +89,12 @@ void vout_blank(void)
 {
   int w = gpu.screen.hres;
   int h = gpu.screen.h;
+
+  check_mode_change(0);
   if (gpu.state.enhancement_active) {
     w *= 2;
     h *= 2;
   }
-  check_mode_change(0);
   cbs->pl_vout_flip(NULL, 1024, gpu.status.rgb24, w, h);
 }
 
diff --git a/readme.txt b/readme.txt
index 435a682d..55e1feb8 100644
--- a/readme.txt
+++ b/readme.txt
@@ -113,6 +113,14 @@ the main menu where it is possible to enable/disable individual cheats.
 Changelog
 ---------
 
+r22 (2015-02-05)
+* general: fixed a race condition/crash in threaded SPU mode
+* pandora: C64x: fixed compatibility with newer c64_tools, enabled L2 cache
+* frontend: fixed control config corruption on load for devices that are
+  disconnected on startup
+* some dma accuracy improvements, might fix occasional glitches in ff7
+* ARMv6 build and the dynarec now make use of available instructions (gizmo98)
+
 r21 (2015-01-12)
 + general: added ability to run SPU emulation on a separate thread, enabled it
   by default when multicore CPU is detected. Significant effort was made to