From e05b81fc5b3f640496795ced5d893ece4cc51c2d Mon Sep 17 00:00:00 2001
From: notaz <notasas@gmail.com>
Date: Tue, 29 Dec 2009 22:43:10 +0000
Subject: [PATCH] 32x: drc: inline dispatcher and irq handling; do write-caused
 irqs

git-svn-id: file:///home/notaz/opt/svn/PicoDrive@849 be3aeb3a-fb24-0410-a615-afba39da0efa
---
 cpu/drc/emit_arm.c      |  75 +++++--
 cpu/drc/emit_x86.c      |  75 ++++++-
 cpu/sh2/compiler.c      | 441 +++++++++++++++++++++++++++-------------
 cpu/sh2/mame/sh2pico.c  |  10 +-
 cpu/sh2/sh2.c           |   2 -
 cpu/sh2/sh2.h           |  23 ++-
 pico/32x/32x.c          |  15 +-
 pico/32x/memory.c       | 136 +++++++------
 pico/pico_int.h         |   2 +
 platform/linux/Makefile |   2 +-
 10 files changed, 535 insertions(+), 246 deletions(-)

diff --git a/cpu/drc/emit_arm.c b/cpu/drc/emit_arm.c
index 8b6af69..6a115e3 100644
--- a/cpu/drc/emit_arm.c
+++ b/cpu/drc/emit_arm.c
@@ -155,6 +155,10 @@
 #define EOP_C_AM2_IMM(cond,u,b,l,rn,rd,offset_12) \
 	EMIT(((cond)<<28) | 0x05000000 | ((u)<<23) | ((b)<<22) | ((l)<<20) | ((rn)<<16) | ((rd)<<12) | (offset_12))
 
+#define EOP_C_AM2_REG(cond,u,b,l,rn,rd,shift_imm,shift_op,rm) \
+	EMIT(((cond)<<28) | 0x07000000 | ((u)<<23) | ((b)<<22) | ((l)<<20) | ((rn)<<16) | ((rd)<<12) | \
+		((shift_imm)<<7) | ((shift_op)<<5) | (rm))
+
 /* addressing mode 3 */
 #define EOP_C_AM3(cond,u,r,l,rn,rd,s,h,immed_reg) \
 	EMIT(((cond)<<28) | 0x01000090 | ((u)<<23) | ((r)<<22) | ((l)<<20) | ((rn)<<16) | ((rd)<<12) | \
@@ -165,12 +169,16 @@
 #define EOP_C_AM3_REG(cond,u,l,rn,rd,s,h,rm)       EOP_C_AM3(cond,u,0,l,rn,rd,s,h,rm)
 
 /* ldr and str */
+#define EOP_LDR_IMM2(cond,rd,rn,offset_12) EOP_C_AM2_IMM(cond,1,0,1,rn,rd,offset_12)
+
 #define EOP_LDR_IMM(   rd,rn,offset_12) EOP_C_AM2_IMM(A_COND_AL,1,0,1,rn,rd,offset_12)
 #define EOP_LDR_NEGIMM(rd,rn,offset_12) EOP_C_AM2_IMM(A_COND_AL,0,0,1,rn,rd,offset_12)
 #define EOP_LDR_SIMPLE(rd,rn)           EOP_C_AM2_IMM(A_COND_AL,1,0,1,rn,rd,0)
 #define EOP_STR_IMM(   rd,rn,offset_12) EOP_C_AM2_IMM(A_COND_AL,1,0,0,rn,rd,offset_12)
 #define EOP_STR_SIMPLE(rd,rn)           EOP_C_AM2_IMM(A_COND_AL,1,0,0,rn,rd,0)
 
+#define EOP_LDR_REG_LSL(cond,rd,rn,rm,shift_imm) EOP_C_AM2_REG(cond,1,0,1,rn,rd,shift_imm,A_AM1_LSL,rm)
+
 #define EOP_LDRH_IMM(   rd,rn,offset_8)  EOP_C_AM3_IMM(A_COND_AL,1,1,rn,rd,0,1,offset_8)
 #define EOP_LDRH_SIMPLE(rd,rn)           EOP_C_AM3_IMM(A_COND_AL,1,1,rn,rd,0,1,0)
 #define EOP_LDRH_REG(   rd,rn,rm)        EOP_C_AM3_REG(A_COND_AL,1,1,rn,rd,0,1,rm)
@@ -192,8 +200,6 @@
 #define EOP_C_BX(cond,rm) \
 	EMIT(((cond)<<28) | 0x012fff10 | (rm))
 
-#define EOP_BX(rm) EOP_C_BX(A_COND_AL,rm)
-
 #define EOP_C_B_PTR(ptr,cond,l,signed_immed_24) \
 	EMIT_PTR(ptr, ((cond)<<28) | 0x0a000000 | ((l)<<24) | (signed_immed_24))
 
@@ -232,6 +238,7 @@
 #define EOP_MSR_REG(rm)       EOP_C_MSR_REG(A_COND_AL,rm)
 
 
+// XXX: AND, RSB, *C, MVN will break if 1 insn is not enough
 static void emith_op_imm2(int cond, int s, int op, int rd, int rn, unsigned int imm)
 {
 	int ror2;
@@ -253,10 +260,9 @@ static void emith_op_imm2(int cond, int s, int op, int rd, int rn, unsigned int
 
 		EOP_C_DOP_IMM(cond, op, s, rn, rd, ror2 & 0x0f, v & 0xff);
 
-		if (op == A_OP_MOV) {
+		if (op == A_OP_MOV)
 			op = A_OP_ORR;
-			rn = rd;
-		}
+		rn = rd;
 	}
 }
 
@@ -461,6 +467,12 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_and_r_r_imm(d, s, imm) \
 	emith_op_imm2(A_COND_AL, 0, A_OP_AND, d, s, imm)
 
+#define emith_add_r_r_imm(d, s, imm) \
+	emith_op_imm2(A_COND_AL, 0, A_OP_ADD, d, s, imm)
+
+#define emith_sub_r_r_imm(d, s, imm) \
+	emith_op_imm2(A_COND_AL, 0, A_OP_SUB, d, s, imm)
+
 #define emith_neg_r_r(d, s) \
 	EOP_RSB_IMM(d, s, 0, 0)
 
@@ -583,18 +595,12 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_pass_arg_imm(arg, imm) \
 	emith_move_r_imm(arg, imm)
 
-#define emith_call_cond(cond, target) \
-	emith_xbranch(cond, target, 1)
+#define emith_jump(target) \
+	emith_jump_cond(A_COND_AL, target)
 
 #define emith_jump_cond(cond, target) \
 	emith_xbranch(cond, target, 0)
 
-#define emith_call(target) \
-	emith_call_cond(A_COND_AL, target)
-
-#define emith_jump(target) \
-	emith_jump_cond(A_COND_AL, target)
-
 #define emith_jump_patchable(cond) \
 	emith_jump_cond(cond, 0)
 
@@ -604,8 +610,37 @@ static int emith_xbranch(int cond, void *target, int is_call)
 	*ptr_ = (*ptr_ & 0xff000000) | (val & 0x00ffffff); \
 } while (0)
 
+#define emith_jump_reg_c(cond, r) \
+	EOP_C_BX(cond, r)
+
 #define emith_jump_reg(r) \
-	EOP_BX(r)
+	emith_jump_reg_c(A_COND_AL, r)
+
+#define emith_jump_ctx_c(cond, offs) \
+	EOP_LDR_IMM2(cond,15,CONTEXT_REG,offs)
+
+#define emith_jump_ctx(offs) \
+	emith_jump_ctx_c(A_COND_AL, offs)
+
+#define emith_call_cond(cond, target) \
+	emith_xbranch(cond, target, 1)
+
+#define emith_call(target) \
+	emith_call_cond(A_COND_AL, target)
+
+#define emith_call_ctx(offs) { \
+	emith_move_r_r(14, 15); \
+	emith_jump_ctx(offs); \
+}
+
+#define emith_ret_c(cond) \
+	emith_jump_reg_c(cond, 14)
+
+#define emith_ret() \
+	emith_ret_c(A_COND_AL)
+
+#define emith_ret_to_ctx(offs) \
+	emith_ctx_write(14, offs)
 
 /* SH2 drc specific */
 #define emith_sh2_drc_entry() \
@@ -614,6 +649,18 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_sh2_drc_exit() \
 	EOP_LDMFD_SP(A_R4M|A_R5M|A_R6M|A_R7M|A_R8M|A_R9M|A_R10M|A_R11M|A_R15M)
 
+#define emith_sh2_wcall(a, tab, ret_ptr) { \
+	int val_ = (char *)(ret_ptr) - (char *)tcache_ptr - 2*4; \
+	if (val_ >= 0) \
+		emith_add_r_r_imm(14, 15, val_); \
+	else if (val_ < 0) \
+		emith_sub_r_r_imm(14, 15, -val_); \
+	emith_lsr(12, a, SH2_WRITE_SHIFT); \
+	EOP_LDR_REG_LSL(A_COND_AL,12,tab,12,2); \
+	emith_ctx_read(2, offsetof(SH2, is_slave)); \
+	emith_jump_reg(12); \
+}
+
 #define emith_sh2_dtbf_loop() { \
 	int cr, rn;                                                          \
 	int tmp_ = rcache_get_tmp();                                         \
diff --git a/cpu/drc/emit_x86.c b/cpu/drc/emit_x86.c
index 247d1d6..02ad79c 100644
--- a/cpu/drc/emit_x86.c
+++ b/cpu/drc/emit_x86.c
@@ -239,7 +239,17 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 	emith_bic_r_imm(r, imm); \
 }
 
+#define emith_jump_reg_c(cond, r) emith_jump_reg(r)
+#define emith_jump_ctx_c(cond, offs) emith_jump_ctx(offs)
+#define emith_ret_c(cond) emith_ret()
+
 // _r_r_imm
+#define emith_add_r_r_imm(d, s, imm) { \
+	if (d != s) \
+		emith_move_r_r(d, s); \
+	emith_add_r_imm(d, imm); \
+}
+
 #define emith_and_r_r_imm(d, s, imm) { \
 	if (d != s) \
 		emith_move_r_r(d, s); \
@@ -279,6 +289,11 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define emith_push(r) \
 	EMIT_OP(0x50 + (r))
 
+#define emith_push_imm(imm) { \
+	EMIT_OP(0x68); \
+	EMIT(imm, u32); \
+}
+
 #define emith_pop(r) \
 	EMIT_OP(0x58 + (r))
 
@@ -376,29 +391,41 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define emith_rolcf emith_rolc
 #define emith_rorcf emith_rorc
 
-// XXX: offs is 8bit only
-#define emith_ctx_read(r, offs) do { \
-	EMIT_OP_MODRM(0x8b, 1, r, xBP); \
-	EMIT(offs, u8); 	/* mov tmp, [ebp+#offs] */ \
+#define emith_ctx_op(op, r, offs) do { \
+	/* mov r <-> [ebp+#offs] */ \
+	if ((offs) >= 0x80) { \
+		EMIT_OP_MODRM(op, 2, r, xBP); \
+		EMIT(offs, u32); \
+	} else { \
+		EMIT_OP_MODRM(op, 1, r, xBP); \
+		EMIT(offs, u8); \
+	} \
 } while (0)
 
+#define emith_ctx_read(r, offs) \
+	emith_ctx_op(0x8b, r, offs)
+
+#define emith_ctx_write(r, offs) \
+	emith_ctx_op(0x89, r, offs)
+
 #define emith_ctx_read_multiple(r, offs, cnt, tmpr) do { \
 	int r_ = r, offs_ = offs, cnt_ = cnt;     \
 	for (; cnt_ > 0; r_++, offs_ += 4, cnt_--) \
 		emith_ctx_read(r_, offs_);        \
 } while (0)
 
-#define emith_ctx_write(r, offs) do { \
-	EMIT_OP_MODRM(0x89, 1, r, xBP); \
-	EMIT(offs, u8); 	/* mov [ebp+#offs], tmp */ \
-} while (0)
-
 #define emith_ctx_write_multiple(r, offs, cnt, tmpr) do { \
 	int r_ = r, offs_ = offs, cnt_ = cnt;     \
 	for (; cnt_ > 0; r_++, offs_ += 4, cnt_--) \
 		emith_ctx_write(r_, offs_);       \
 } while (0)
 
+// assumes EBX is free
+#define emith_ret_to_ctx(offs) { \
+	emith_pop(xBX); \
+	emith_ctx_write(xBX, offs); \
+}
+
 #define emith_jump(ptr) { \
 	u32 disp = (u32)(ptr) - ((u32)tcache_ptr + 5); \
 	EMIT_OP(0xe9); \
@@ -429,9 +456,25 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define emith_call_cond(cond, ptr) \
 	emith_call(ptr)
 
+#define emith_call_reg(r) \
+	EMIT_OP_MODRM(0xff, 3, 2, r)
+
+#define emith_call_ctx(offs) { \
+	EMIT_OP_MODRM(0xff, 2, 2, xBP); \
+	EMIT(offs, u32); \
+}
+
+#define emith_ret() \
+	EMIT_OP(0xc3)
+
 #define emith_jump_reg(r) \
 	EMIT_OP_MODRM(0xff, 3, 4, r)
 
+#define emith_jump_ctx(offs) { \
+	EMIT_OP_MODRM(0xff, 2, 4, xBP); \
+	EMIT(offs, u32); \
+}
+
 #define EMITH_JMP_START(cond) { \
 	u8 *cond_ptr; \
 	JMP8_POS(cond_ptr)
@@ -476,7 +519,19 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 	emith_pop(xSI);         \
 	emith_pop(xBP);         \
 	emith_pop(xBX);         \
-	EMIT_OP(0xc3); /* ret */\
+	emith_ret();            \
+}
+
+// assumes EBX is free temporary
+#define emith_sh2_wcall(a, tab, ret_ptr) { \
+	int arg2_; \
+	host_arg2reg(arg2_, 2); \
+	emith_lsr(xBX, a, SH2_WRITE_SHIFT); \
+	EMIT_OP_MODRM(0x8b, 0, xBX, 4); \
+	EMIT_SIB(2, xBX, tab); /* mov ebx, [tab + ebx * 4] */ \
+	emith_ctx_read(arg2_, offsetof(SH2, is_slave)); \
+	emith_push_imm((long)(ret_ptr)); \
+	emith_jump_reg(xBX); \
 }
 
 #define emith_sh2_dtbf_loop() { \
diff --git a/cpu/sh2/compiler.c b/cpu/sh2/compiler.c
index 618b497..ee84364 100644
--- a/cpu/sh2/compiler.c
+++ b/cpu/sh2/compiler.c
@@ -41,6 +41,16 @@ static char sh2dasm_buff[64];
 #else
 #define do_host_disasm(x)
 #endif
+
+#if (DRC_DEBUG & 4)
+static void REGPARM(3) *sh2_drc_announce_entry(void *block, SH2 *sh2, u32 sr)
+{
+  if (block != NULL)
+    dbg(4, "= %csh2 enter %08x %p, c=%d", sh2->is_slave ? 's' : 'm',
+      sh2->pc, block, (signed int)sr >> 12);
+  return block;
+}
+#endif
 // } debug
 
 #define BLOCK_CYCLE_LIMIT 100
@@ -101,7 +111,7 @@ static temp_reg_t reg_temp[] = {
   {  3, },
 };
 
-#else
+#elif defined(__i386__)
 #include "../drc/emit_x86.c"
 
 static const int reg_map_g2h[] = {
@@ -121,6 +131,8 @@ static temp_reg_t reg_temp[] = {
   { xDX, },
 };
 
+#else
+#error unsupported arch
 #endif
 
 #define T	0x00000001
@@ -130,6 +142,7 @@ static temp_reg_t reg_temp[] = {
 #define M	0x00000200
 #define T_save	0x00000800
 
+#define I_SHIFT 4
 #define Q_SHIFT 8
 #define M_SHIFT 9
 
@@ -159,12 +172,16 @@ static void **hash_table;
 #define HASH_FUNC(hash_tab, addr) \
   ((block_desc **)(hash_tab))[(addr) & HASH_MASK]
 
-static void REGPARM(2) (*sh2_drc_entry)(const void *block, SH2 *sh2);
-static void (*sh2_drc_exit)(void);
+static void REGPARM(1) (*sh2_drc_entry)(SH2 *sh2);
+static void            (*sh2_drc_dispatcher)(void);
+static void            (*sh2_drc_exit)(void);
+static void            (*sh2_drc_test_irq)(void);
+static void REGPARM(2) (*sh2_drc_write8)(u32 a, u32 d);
+static void REGPARM(2) (*sh2_drc_write8_slot)(u32 a, u32 d);
+static void REGPARM(2) (*sh2_drc_write16)(u32 a, u32 d);
+static void REGPARM(2) (*sh2_drc_write16_slot)(u32 a, u32 d);
 
-// tmp
 extern void REGPARM(2) sh2_do_op(SH2 *sh2, int opcode);
-static void REGPARM(1) sh2_test_irq(SH2 *sh2);
 
 static void flush_tcache(int tcid)
 {
@@ -484,9 +501,14 @@ static int emit_memhandler_read(int size)
   emith_move_r_r(ctxr, CONTEXT_REG);
   switch (size) {
   case 0: // 8
+    // must writeback cycles for poll detection stuff
+    if (reg_map_g2h[SHR_SR] != -1)
+      emith_ctx_write(reg_map_g2h[SHR_SR], SHR_SR * 4);
     emith_call(p32x_sh2_read8);
     break;
   case 1: // 16
+    if (reg_map_g2h[SHR_SR] != -1)
+      emith_ctx_write(reg_map_g2h[SHR_SR], SHR_SR * 4);
     emith_call(p32x_sh2_read16);
     break;
   case 2: // 32
@@ -498,19 +520,32 @@ static int emit_memhandler_read(int size)
   return rcache_get_tmp_arg(0);
 }
 
-static void emit_memhandler_write(int size)
+static void emit_memhandler_write(int size, u32 pc, int delay)
 {
   int ctxr;
   host_arg2reg(ctxr, 2);
-  emith_move_r_r(ctxr, CONTEXT_REG);
   switch (size) {
   case 0: // 8
-    emith_call(p32x_sh2_write8);
+    // XXX: consider inlining sh2_drc_write8
+    if (delay) {
+      emith_call(sh2_drc_write8_slot);
+    } else {
+      emit_move_r_imm32(SHR_PC, pc);
+      rcache_clean();
+      emith_call(sh2_drc_write8);
+    }
     break;
   case 1: // 16
-    emith_call(p32x_sh2_write16);
+    if (delay) {
+      emith_call(sh2_drc_write16_slot);
+    } else {
+      emit_move_r_imm32(SHR_PC, pc);
+      rcache_clean();
+      emith_call(sh2_drc_write16);
+    }
     break;
   case 2: // 32
+    emith_move_r_r(ctxr, CONTEXT_REG);
     emith_call(p32x_sh2_write32);
     break;
   }
@@ -528,19 +563,6 @@ static int emit_indirect_indexed_read(int rx, int ry, int size)
   return emit_memhandler_read(size);
 }
 
-// tmp_wr -> @(Rx,Ry)
-static void emit_indirect_indexed_write(int tmp_wr, int rx, int ry, int size)
-{
-  int a0, t;
-  rcache_clean();
-  t = rcache_get_tmp_arg(1);
-  emith_move_r_r(t, tmp_wr);
-  a0 = rcache_get_reg_arg(0, rx);
-  t  = rcache_get_reg(ry, RC_GR_READ);
-  emith_add_r_r(a0, t);
-  emit_memhandler_write(size);
-}
-
 // read @Rn, @rm
 static void emit_indirect_read_double(u32 *rnr, u32 *rmr, int rn, int rm, int size)
 {
@@ -593,27 +615,67 @@ static void emit_do_static_regs(int is_write, int tmpr)
   }
 }
 
-static void sh2_generate_utils(void)
+static void emit_block_entry(void)
 {
-  int ctx, blk, tmp;
+  int arg0, arg1, arg2;
 
-  host_arg2reg(blk, 0);
-  host_arg2reg(ctx, 1);
-  host_arg2reg(tmp, 2);
+  host_arg2reg(arg0, 0);
+  host_arg2reg(arg1, 1);
+  host_arg2reg(arg2, 2);
 
-  // sh2_drc_entry(void *block, SH2 *sh2)
-  sh2_drc_entry = (void *)tcache_ptr;
-  emith_sh2_drc_entry();
-  emith_move_r_r(CONTEXT_REG, ctx); // move ctx, arg1
-  emit_do_static_regs(0, tmp);
-  emith_jump_reg(blk); // jump arg0
+#if (DRC_DEBUG & 4)
+  emith_move_r_r(arg1, CONTEXT_REG);
+  emith_move_r_r(arg2, rcache_get_reg(SHR_SR, RC_GR_READ));
+  emith_call(sh2_drc_announce_entry);
+  rcache_invalidate();
+#endif
+  emith_tst_r_r(arg0, arg0);
+  EMITH_SJMP_START(DCOND_EQ);
+  emith_jump_reg_c(DCOND_NE, arg0);
+  EMITH_SJMP_END(DCOND_EQ);
+}
 
-  // sh2_drc_exit(void)
-  sh2_drc_exit = (void *)tcache_ptr;
-  emit_do_static_regs(1, tmp);
-  emith_sh2_drc_exit();
+static void REGPARM(3) *lookup_block(u32 pc, int is_slave, int *tcache_id)
+{
+  block_desc *bd = NULL;
+  void *block = NULL;
+  *tcache_id = 0;
+
+  // we have full block id tables for data_array and RAM
+  // BIOS goes to data_array table too
+  if ((pc & 0xe0000000) == 0xc0000000 || (pc & ~0xfff) == 0) {
+    int blkid = Pico32xMem->drcblk_da[is_slave][(pc & 0xfff) >> SH2_DRCBLK_DA_SHIFT];
+    *tcache_id = 1 + is_slave;
+    if (blkid & 1) {
+      bd = &block_tables[*tcache_id][blkid >> 1];
+      block = bd->tcache_ptr;
+    }
+  }
+  // RAM
+  else if ((pc & 0xc6000000) == 0x06000000) {
+    int blkid = Pico32xMem->drcblk_ram[(pc & 0x3ffff) >> SH2_DRCBLK_RAM_SHIFT];
+    if (blkid & 1) {
+      bd = &block_tables[0][blkid >> 1];
+      block = bd->tcache_ptr;
+    }
+  }
+  // ROM
+  else if ((pc & 0xc6000000) == 0x02000000) {
+    bd = HASH_FUNC(hash_table, pc);
 
-  rcache_invalidate();
+    if (bd != NULL) {
+      if (bd->addr == pc)
+        block = bd->tcache_ptr;
+      else
+        block = dr_find_block(bd, pc);
+    }
+  }
+
+#if (DRC_DEBUG & 1)
+  if (bd != NULL)
+    bd->refcount++;
+#endif
+  return block;
 }
 
 #define DELAYED_OP \
@@ -628,6 +690,12 @@ static void sh2_generate_utils(void)
   drcf.use_saved_t = 1;        \
 }
 
+#define FLUSH_CYCLES(sr) \
+  if (cycles > 0) { \
+    emith_sub_r_imm(sr, cycles << 12); \
+    cycles = 0; \
+  }
+
 #define CHECK_UNHANDLED_BITS(mask) { \
   if ((op & (mask)) != 0) \
     goto default_; \
@@ -651,7 +719,7 @@ static void sh2_generate_utils(void)
 #define OP_FLAGS(pc) op_flags[((pc) - base_pc) / 2]
 #define OF_DELAY_OP (1 << 0)
 
-static void *sh2_translate(SH2 *sh2, int tcache_id)
+static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
 {
   // XXX: maybe use structs instead?
   void *branch_target_ptr[MAX_LOCAL_BRANCHES];
@@ -768,8 +836,7 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
         int blkid;
 
         sr = rcache_get_reg(SHR_SR, RC_GR_RMW);
-        emith_sub_r_imm(sr, cycles << 12);
-        cycles = 0;
+        FLUSH_CYCLES(sr);
         rcache_flush();
         do_host_disasm(tcache_id);
 
@@ -863,8 +930,12 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
       case 0x04: // MOV.B Rm,@(R0,Rn)   0000nnnnmmmm0100
       case 0x05: // MOV.W Rm,@(R0,Rn)   0000nnnnmmmm0101
       case 0x06: // MOV.L Rm,@(R0,Rn)   0000nnnnmmmm0110
-        tmp = rcache_get_reg(GET_Rm(), RC_GR_READ);
-        emit_indirect_indexed_write(tmp, SHR_R0, GET_Rn(), op & 3);
+        rcache_clean();
+        tmp  = rcache_get_reg_arg(1, GET_Rm());
+        tmp2 = rcache_get_reg_arg(0, SHR_R0);
+        tmp3 = rcache_get_reg(GET_Rn(), RC_GR_READ);
+        emith_add_r_r(tmp2, tmp3);
+        emit_memhandler_write(op & 3, pc, drcf.delayed_op);
         goto end_op;
       case 0x07:
         // MUL.L     Rm,Rn      0000nnnnmmmm0111
@@ -954,9 +1025,8 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
           emit_move_r_imm32(SHR_PC, pc - 2);
           tmp = rcache_get_reg(SHR_SR, RC_GR_RMW);
           emith_clear_msb(tmp, tmp, 20); // clear cycles
-          drcf.test_irq = 1;
           cycles = 1;
-          break;
+          goto end_op;
         case 2: // RTE        0000000000101011
           DELAYED_OP;
           rcache_clean();
@@ -1036,7 +1106,7 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
       tmp  = rcache_get_reg_arg(0, GET_Rn());
       tmp2 = rcache_get_reg_arg(1, GET_Rm());
       emith_add_r_imm(tmp, (op & 0x0f) * 4);
-      emit_memhandler_write(2);
+      emit_memhandler_write(2, pc, drcf.delayed_op);
       goto end_op;
 
     case 0x02:
@@ -1048,7 +1118,7 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
         rcache_clean();
         rcache_get_reg_arg(0, GET_Rn());
         rcache_get_reg_arg(1, GET_Rm());
-        emit_memhandler_write(op & 3);
+        emit_memhandler_write(op & 3, pc, drcf.delayed_op);
         goto end_op;
       case 0x04: // MOV.B Rm,@–Rn       0010nnnnmmmm0100
       case 0x05: // MOV.W Rm,@–Rn       0010nnnnmmmm0101
@@ -1058,7 +1128,7 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
         rcache_clean();
         rcache_get_reg_arg(0, GET_Rn());
         rcache_get_reg_arg(1, GET_Rm());
-        emit_memhandler_write(op & 3);
+        emit_memhandler_write(op & 3, pc, drcf.delayed_op);
         goto end_op;
       case 0x07: // DIV0S Rm,Rn         0010nnnnmmmm0111
         sr   = rcache_get_reg(SHR_SR, RC_GR_RMW);
@@ -1380,8 +1450,8 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
         rcache_get_reg_arg(0, GET_Rn());
         tmp3 = rcache_get_reg_arg(1, tmp);
         if (tmp == SHR_SR)
-          emith_clear_msb(tmp3, tmp3, 20); // reserved bits defined by ISA as 0
-        emit_memhandler_write(2);
+          emith_clear_msb(tmp3, tmp3, 22); // reserved bits defined by ISA as 0
+        emit_memhandler_write(2, pc, drcf.delayed_op);
         goto end_op;
       case 0x04:
       case 0x05:
@@ -1541,7 +1611,7 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
           emith_move_r_r(tmp2, tmp);
           rcache_free_tmp(tmp);
           rcache_get_reg_arg(0, GET_Rn());
-          emit_memhandler_write(0);
+          emit_memhandler_write(0, pc, drcf.delayed_op);
           cycles += 3;
           break;
         default:
@@ -1720,7 +1790,7 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
         tmp2 = rcache_get_reg_arg(1, SHR_R0);
         tmp3 = (op & 0x100) >> 8;
         emith_add_r_imm(tmp, (op & 0x0f) << tmp3);
-        emit_memhandler_write(tmp3);
+        emit_memhandler_write(tmp3, pc, drcf.delayed_op);
         goto end_op;
       case 0x0400: // MOV.B @(disp,Rm),R0  10000100mmmmdddd
       case 0x0500: // MOV.W @(disp,Rm),R0  10000101mmmmdddd
@@ -1824,7 +1894,7 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
         tmp2 = rcache_get_reg_arg(1, SHR_R0);
         tmp3 = (op & 0x300) >> 8;
         emith_add_r_imm(tmp, (op & 0xff) << tmp3);
-        emit_memhandler_write(tmp3);
+        emit_memhandler_write(tmp3, pc, drcf.delayed_op);
         goto end_op;
       case 0x0400: // MOV.B @(disp,GBR),R0   11000100dddddddd
       case 0x0500: // MOV.W @(disp,GBR),R0   11000101dddddddd
@@ -1850,12 +1920,12 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
         emith_add_r_imm(tmp, 4);
         tmp = rcache_get_reg_arg(1, SHR_SR);
         emith_clear_msb(tmp, tmp, 22);
-        emit_memhandler_write(2);
+        emit_memhandler_write(2, pc, drcf.delayed_op);
         // push PC
         rcache_get_reg_arg(0, SHR_SP);
         tmp = rcache_get_tmp_arg(1);
         emith_move_r_imm(tmp, pc);
-        emit_memhandler_write(2);
+        emit_memhandler_write(2, pc, drcf.delayed_op);
         // obtain new PC
         tmp = rcache_get_reg_arg(0, SHR_VBR);
         emith_add_r_imm(tmp, (op & 0xff) * 4);
@@ -1918,7 +1988,7 @@ static void *sh2_translate(SH2 *sh2, int tcache_id)
         tmp3 = rcache_get_reg_arg(0, SHR_GBR);
         tmp4 = rcache_get_reg(SHR_R0, RC_GR_READ);
         emith_add_r_r(tmp3, tmp4);
-        emit_memhandler_write(0);
+        emit_memhandler_write(0, pc, drcf.delayed_op);
         cycles += 2;
         goto end_op;
       }
@@ -1962,8 +2032,7 @@ end_op:
     if (branch_patch_cond != -1 && drcf.delayed_op != 2) {
       sr = rcache_get_reg(SHR_SR, RC_GR_RMW);
       // handle cycles
-      emith_sub_r_imm(sr, cycles << 12);
-      cycles = 0;
+      FLUSH_CYCLES(sr);
       rcache_clean();
 
       if (drcf.use_saved_t)
@@ -1983,8 +2052,16 @@ end_op:
       }
     }
     // test irq?
-    if (drcf.test_irq && drcf.delayed_op != 2)
-      break;
+    // XXX: delay slots..
+    if (drcf.test_irq && drcf.delayed_op != 2) {
+      if (!drcf.delayed_op)
+        emit_move_r_imm32(SHR_PC, pc);
+      sr = rcache_get_reg(SHR_SR, RC_GR_RMW);
+      FLUSH_CYCLES(sr);
+      rcache_flush();
+      emith_call(sh2_drc_test_irq);
+      drcf.test_irq = 0;
+    }
     if (drcf.delayed_op == 1)
       break;
 
@@ -1995,19 +2072,13 @@ end_op:
   if (!drcf.delayed_op)
     emit_move_r_imm32(SHR_PC, pc);
 
-  if (drcf.test_irq) {
-    rcache_flush();
-    emith_pass_arg_r(0, CONTEXT_REG);
-    emith_call(sh2_test_irq);
-  }
-
 end_block_btf:
   this_block->end_addr = pc;
 
   tmp = rcache_get_reg(SHR_SR, RC_GR_RMW);
-  emith_sub_r_imm(tmp, cycles << 12);
+  FLUSH_CYCLES(tmp);
   rcache_flush();
-  emith_jump(sh2_drc_exit);
+  emith_jump(sh2_drc_dispatcher);
 
   // link local branches
   for (i = 0; i < branch_patch_count; i++) {
@@ -2023,7 +2094,7 @@ end_block_btf:
       target = tcache_ptr;
       emit_move_r_imm32(SHR_PC, branch_patch_pc[i]);
       rcache_flush();
-      emith_jump(sh2_drc_exit);
+      emith_jump(sh2_drc_dispatcher);
     }
     emith_jump_patch(branch_patch_ptr[i], target);
   }
@@ -2085,67 +2156,165 @@ unimplemented:
 */
 }
 
-void __attribute__((noinline)) sh2_drc_dispatcher(SH2 *sh2)
+static void sh2_generate_utils(void)
 {
-  // TODO: need to handle self-caused interrupts
-  sh2_test_irq(sh2);
+  int arg0, arg1, arg2, sr, tmp;
+  void *sh2_drc_write_end, *sh2_drc_write_slot_end;
 
-  while (((signed int)sh2->sr >> 12) > 0)
-  {
-    void *block = NULL;
-    block_desc *bd = NULL;
-    int tcache_id = 0;
-
-    // FIXME: must avoid doing it so often..
-    //sh2_test_irq(sh2);
-
-    // we have full block id tables for data_array and RAM
-    // BIOS goes to data_array table too
-    if ((sh2->pc & 0xe0000000) == 0xc0000000 || (sh2->pc & ~0xfff) == 0) {
-      int blkid = Pico32xMem->drcblk_da[sh2->is_slave][(sh2->pc & 0xfff) >> SH2_DRCBLK_DA_SHIFT];
-      tcache_id = 1 + sh2->is_slave;
-      if (blkid & 1) {
-        bd = &block_tables[tcache_id][blkid >> 1];
-        block = bd->tcache_ptr;
-      }
-    }
-    // RAM
-    else if ((sh2->pc & 0xc6000000) == 0x06000000) {
-      int blkid = Pico32xMem->drcblk_ram[(sh2->pc & 0x3ffff) >> SH2_DRCBLK_RAM_SHIFT];
-      if (blkid & 1) {
-        bd = &block_tables[tcache_id][blkid >> 1];
-        block = bd->tcache_ptr;
-      }
-    }
-    // ROM
-    else if ((sh2->pc & 0xc6000000) == 0x02000000) {
-      bd = HASH_FUNC(hash_table, sh2->pc);
-
-      if (bd != NULL) {
-        if (bd->addr == sh2->pc)
-          block = bd->tcache_ptr;
-        else
-          block = dr_find_block(bd, sh2->pc);
-      }
-    }
+  host_arg2reg(arg0, 0);
+  host_arg2reg(arg1, 1);
+  host_arg2reg(arg2, 2);
+  emith_move_r_r(arg0, arg0); // nop
 
-    if (block == NULL)
-      block = sh2_translate(sh2, tcache_id);
-    if (block == NULL) {
-      // sh2_translate failed, possibly tcache overflow, clean up and try again
-      flush_tcache(tcache_id);
-      block = sh2_translate(sh2, tcache_id);
-    }
+  // sh2_drc_exit(void)
+  sh2_drc_exit = (void *)tcache_ptr;
+  emit_do_static_regs(1, arg2);
+  emith_sh2_drc_exit();
 
-    dbg(4, "= %csh2 enter %08x %p, c=%d", sh2->is_slave ? 's' : 'm',
-      sh2->pc, block, (signed int)sh2->sr >> 12);
-#if (DRC_DEBUG & 1)
-    if (bd != NULL)
-      bd->refcount++;
+  // sh2_drc_dispatcher(void)
+  sh2_drc_dispatcher = (void *)tcache_ptr;
+  sr = rcache_get_reg(SHR_SR, RC_GR_READ);
+  emith_cmp_r_imm(sr, 0);
+  emith_jump_cond(DCOND_LT, sh2_drc_exit);
+  rcache_invalidate();
+  emith_ctx_read(arg0, SHR_PC * 4);
+  emith_ctx_read(arg1, offsetof(SH2, is_slave));
+  emith_add_r_r_imm(arg2, CONTEXT_REG, offsetof(SH2, drc_tmp));
+  emith_call(lookup_block);
+  emit_block_entry();
+  // lookup failed, call sh2_translate()
+  emith_move_r_r(arg0, CONTEXT_REG);
+  emith_ctx_read(arg1, offsetof(SH2, drc_tmp)); // tcache_id
+  emith_call(sh2_translate);
+  emit_block_entry();
+  // sh2_translate() failed, flush cache and retry
+  emith_ctx_read(arg0, offsetof(SH2, drc_tmp));
+  emith_call(flush_tcache);
+  emith_move_r_r(arg0, CONTEXT_REG);
+  emith_ctx_read(arg1, offsetof(SH2, drc_tmp));
+  emith_call(sh2_translate);
+  emit_block_entry();
+  // XXX: can't translate, fail
+  emith_call(exit);
+
+  // sh2_drc_test_irq(void)
+  // assumes it's called from main function (may jump to dispatcher)
+  sh2_drc_test_irq = (void *)tcache_ptr;
+  emith_ctx_read(arg1, offsetof(SH2, pending_level));
+  sr = rcache_get_reg(SHR_SR, RC_GR_READ);
+  emith_lsr(arg0, sr, I_SHIFT);
+  emith_and_r_imm(arg0, 0x0f);
+  emith_cmp_r_r(arg1, arg0); // pending_level > ((sr >> 4) & 0x0f)?
+  EMITH_SJMP_START(DCOND_GT);
+  emith_ret_c(DCOND_LE);     // nope, return
+  EMITH_SJMP_END(DCOND_GT);
+  // adjust SP
+  tmp = rcache_get_reg(SHR_SP, RC_GR_RMW);
+  emith_sub_r_imm(tmp, 4*2);
+  rcache_clean();
+  // push SR
+  tmp = rcache_get_reg_arg(0, SHR_SP);
+  emith_add_r_imm(tmp, 4);
+  tmp = rcache_get_reg_arg(1, SHR_SR);
+  emith_clear_msb(tmp, tmp, 22);
+  emith_move_r_r(arg2, CONTEXT_REG);
+  emith_call(p32x_sh2_write32);
+  rcache_invalidate();
+  // push PC
+  rcache_get_reg_arg(0, SHR_SP);
+  emith_ctx_read(arg1, SHR_PC * 4);
+  emith_move_r_r(arg2, CONTEXT_REG);
+  emith_call(p32x_sh2_write32);
+  rcache_invalidate();
+  // update I, cycles, do callback
+  emith_ctx_read(arg1, offsetof(SH2, pending_level));
+  sr = rcache_get_reg(SHR_SR, RC_GR_RMW);
+  emith_bic_r_imm(sr, I);
+  emith_or_r_r_lsl(sr, arg1, I_SHIFT);
+  emith_sub_r_imm(sr, 13 << 12); // at least 13 cycles
+  rcache_flush();
+  emith_move_r_r(arg0, CONTEXT_REG);
+  emith_call_ctx(offsetof(SH2, irq_callback)); // vector = sh2->irq_callback(sh2, level);
+  // obtain new PC
+  emith_lsl(arg0, arg0, 2);
+  emith_ctx_read(arg1, SHR_VBR * 4);
+  emith_add_r_r(arg0, arg1);
+  emit_memhandler_read(2);
+  emith_ctx_write(arg0, SHR_PC * 4);
+#ifdef __i386__
+  emith_add_r_imm(xSP, 4); // fix stack
+#endif
+  emith_jump(sh2_drc_dispatcher);
+  rcache_invalidate();
+
+  // sh2_drc_entry(SH2 *sh2)
+  sh2_drc_entry = (void *)tcache_ptr;
+  emith_sh2_drc_entry();
+  emith_move_r_r(CONTEXT_REG, arg0); // move ctx, arg0
+  emit_do_static_regs(0, arg2);
+  emith_call(sh2_drc_test_irq);
+  emith_jump(sh2_drc_dispatcher);
+
+  // write-caused irq detection
+  sh2_drc_write_end = tcache_ptr;
+  emith_tst_r_r(arg0, arg0);
+  EMITH_SJMP_START(DCOND_NE);
+  emith_jump_ctx_c(DCOND_EQ, offsetof(SH2, drc_tmp)); // return
+  EMITH_SJMP_END(DCOND_NE);
+  // since PC is up to date, jump to it's block instead of returning
+  emith_call(sh2_drc_test_irq);
+  emith_jump_ctx(offsetof(SH2, drc_tmp));
+
+  // write-caused irq detection for writes in delay slot
+  sh2_drc_write_slot_end = tcache_ptr;
+  emith_tst_r_r(arg0, arg0);
+  EMITH_SJMP_START(DCOND_NE);
+  emith_jump_ctx_c(DCOND_EQ, offsetof(SH2, drc_tmp));
+  EMITH_SJMP_END(DCOND_NE);
+  // just burn cycles to get back to dispatcher after branch is handled
+  sr = rcache_get_reg(SHR_SR, RC_GR_RMW);
+  emith_ctx_write(sr, offsetof(SH2, irq_cycles));
+  emith_clear_msb(sr, sr, 20); // clear cycles
+  rcache_flush();
+  emith_jump_ctx(offsetof(SH2, drc_tmp));
+
+  // sh2_drc_write8(u32 a, u32 d)
+  sh2_drc_write8 = (void *)tcache_ptr;
+  emith_ret_to_ctx(offsetof(SH2, drc_tmp));
+  emith_ctx_read(arg2, offsetof(SH2, write8_tab));
+  emith_sh2_wcall(arg0, arg2, sh2_drc_write_end);
+
+  // sh2_drc_write16(u32 a, u32 d)
+  sh2_drc_write16 = (void *)tcache_ptr;
+  emith_ret_to_ctx(offsetof(SH2, drc_tmp));
+  emith_ctx_read(arg2, offsetof(SH2, write16_tab));
+  emith_sh2_wcall(arg0, arg2, sh2_drc_write_end);
+
+  // sh2_drc_write8_slot(u32 a, u32 d)
+  sh2_drc_write8_slot = (void *)tcache_ptr;
+  emith_ret_to_ctx(offsetof(SH2, drc_tmp));
+  emith_ctx_read(arg2, offsetof(SH2, write8_tab));
+  emith_sh2_wcall(arg0, arg2, sh2_drc_write_slot_end);
+
+  // sh2_drc_write16_slot(u32 a, u32 d)
+  sh2_drc_write16_slot = (void *)tcache_ptr;
+  emith_ret_to_ctx(offsetof(SH2, drc_tmp));
+  emith_ctx_read(arg2, offsetof(SH2, write16_tab));
+  emith_sh2_wcall(arg0, arg2, sh2_drc_write_slot_end);
+
+  rcache_invalidate();
+#if (DRC_DEBUG & 2)
+  host_dasm_new_symbol(sh2_drc_entry);
+  host_dasm_new_symbol(sh2_drc_dispatcher);
+  host_dasm_new_symbol(sh2_drc_exit);
+  host_dasm_new_symbol(sh2_drc_test_irq);
+  host_dasm_new_symbol(sh2_drc_write_end);
+  host_dasm_new_symbol(sh2_drc_write_slot_end);
+  host_dasm_new_symbol(sh2_drc_write8);
+  host_dasm_new_symbol(sh2_drc_write8_slot);
+  host_dasm_new_symbol(sh2_drc_write16);
+  host_dasm_new_symbol(sh2_drc_write16_slot);
 #endif
-    sh2_drc_entry(block, sh2);
-    dbg(4, "= leave %p", block);
-  }
 }
 
 static void sh2_smc_rm_block(u16 *drcblk, u16 *p, block_desc *btab, u32 a)
@@ -2193,6 +2362,7 @@ void sh2_drc_wcheck_da(unsigned int a, int val, int cpuid)
 
 void sh2_execute(SH2 *sh2c, int cycles)
 {
+  int ret_cycles;
   sh2 = sh2c; // XXX
 
   sh2c->cycles_aim += cycles;
@@ -2203,23 +2373,14 @@ void sh2_execute(SH2 *sh2c, int cycles)
   // others are usual SH2 flags
   sh2c->sr &= 0x3f3;
   sh2c->sr |= cycles << 12;
-  sh2_drc_dispatcher(sh2c);
+  sh2_drc_entry(sh2c);
 
-  sh2c->cycles_done += cycles - ((signed int)sh2c->sr >> 12);
-}
+  // TODO: irq cycles
+  ret_cycles = (signed int)sh2c->sr >> 12;
+  if (ret_cycles > 0)
+    printf("warning: drc returned with cycles: %d\n", ret_cycles);
 
-static void REGPARM(1) sh2_test_irq(SH2 *sh2)
-{
-  if (sh2->pending_level > ((sh2->sr >> 4) & 0x0f))
-  {
-    if (sh2->pending_irl > sh2->pending_int_irq)
-      sh2_do_irq(sh2, sh2->pending_irl, 64 + sh2->pending_irl/2);
-    else {
-      sh2_do_irq(sh2, sh2->pending_int_irq, sh2->pending_int_vector);
-      sh2->pending_int_irq = 0; // auto-clear
-      sh2->pending_level = sh2->pending_irl;
-    }
-  }
+  sh2c->cycles_done += cycles - ret_cycles;
 }
 
 #if (DRC_DEBUG & 1)
diff --git a/cpu/sh2/mame/sh2pico.c b/cpu/sh2/mame/sh2pico.c
index 167f8cc..a65638b 100644
--- a/cpu/sh2/mame/sh2pico.c
+++ b/cpu/sh2/mame/sh2pico.c
@@ -77,13 +77,9 @@ void sh2_execute(SH2 *sh2_, int cycles)
 		/* FIXME: Darxide doesn't like this */
 		if (sh2->test_irq && !sh2->delay && sh2->pending_level > ((sh2->sr >> 4) & 0x0f))
 		{
-			if (sh2->pending_irl > sh2->pending_int_irq)
-				sh2_do_irq(sh2, sh2->pending_irl, 64 + sh2->pending_irl/2);
-			else {
-				sh2_do_irq(sh2, sh2->pending_int_irq, sh2->pending_int_vector);
-				sh2->pending_int_irq = 0; // auto-clear
-				sh2->pending_level = sh2->pending_irl;
-			}
+			int level = sh2->pending_level;
+			int vector = sh2->irq_callback(sh2, level);
+			sh2_do_irq(sh2, level, vector);
 			sh2->test_irq = 0;
 		}
 
diff --git a/cpu/sh2/sh2.c b/cpu/sh2/sh2.c
index d8d82ed..f0bd907 100644
--- a/cpu/sh2/sh2.c
+++ b/cpu/sh2/sh2.c
@@ -36,8 +36,6 @@ void sh2_reset(SH2 *sh2)
 
 void sh2_do_irq(SH2 *sh2, int level, int vector)
 {
-	sh2->irq_callback(sh2->is_slave, level);
-
 	sh2->r[15] -= 4;
 	p32x_sh2_write32(sh2->r[15], sh2->sr, sh2);	/* push SR onto stack */
 	sh2->r[15] -= 4;
diff --git a/cpu/sh2/sh2.h b/cpu/sh2/sh2.h
index 9d6f0d2..074b142 100644
--- a/cpu/sh2/sh2.h
+++ b/cpu/sh2/sh2.h
@@ -1,6 +1,12 @@
 #ifndef __SH2_H__
 #define __SH2_H__
 
+#if !defined(REGPARM) && defined(__i386__) 
+#define REGPARM(x) __attribute__((regparm(x)))
+#else
+#define REGPARM(x)
+#endif
+
 // registers - matches structure order
 typedef enum {
   SHR_R0 = 0, SHR_SP = 15,
@@ -8,7 +14,7 @@ typedef enum {
   SHR_GBR, SHR_VBR, SHR_MACH, SHR_MACL,
 } sh2_reg_e;
 
-typedef struct
+typedef struct SH2_
 {
 	unsigned int	r[16];		// 00
 	unsigned int	pc;		// 40
@@ -26,6 +32,7 @@ typedef struct
 
 	// drc stuff
 	int		drc_tmp;	// 70
+	int		irq_cycles;
 
 	// interpreter stuff
 	int		icount;		// cycles left in current timeslice
@@ -37,7 +44,7 @@ typedef struct
 	int	pending_irl;
 	int	pending_int_irq;	// internal irq
 	int	pending_int_vector;
-	void	(*irq_callback)(int id, int level);
+	int	REGPARM(2) (*irq_callback)(struct SH2_ *sh2, int level);
 	int	is_slave;
 
 	unsigned int	cycles_aim;	// subtract sh2_icount to get global counter
@@ -57,17 +64,11 @@ void sh2_execute(SH2 *sh2, int cycles);
 
 // pico memhandlers
 // XXX: move somewhere else
-#if !defined(REGPARM) && defined(__i386__) 
-#define REGPARM(x) __attribute__((regparm(x)))
-#else
-#define REGPARM(x)
-#endif
-
 unsigned int REGPARM(2) p32x_sh2_read8(unsigned int a, SH2 *sh2);
 unsigned int REGPARM(2) p32x_sh2_read16(unsigned int a, SH2 *sh2);
 unsigned int REGPARM(2) p32x_sh2_read32(unsigned int a, SH2 *sh2);
-void REGPARM(3) p32x_sh2_write8(unsigned int a, unsigned int d, SH2 *sh2);
-void REGPARM(3) p32x_sh2_write16(unsigned int a, unsigned int d, SH2 *sh2);
-void REGPARM(3) p32x_sh2_write32(unsigned int a, unsigned int d, SH2 *sh2);
+int REGPARM(3) p32x_sh2_write8 (unsigned int a, unsigned int d, SH2 *sh2);
+int REGPARM(3) p32x_sh2_write16(unsigned int a, unsigned int d, SH2 *sh2);
+int REGPARM(3) p32x_sh2_write32(unsigned int a, unsigned int d, SH2 *sh2);
 
 #endif /* __SH2_H__ */
diff --git a/pico/32x/32x.c b/pico/32x/32x.c
index ba89d98..323b0b6 100644
--- a/pico/32x/32x.c
+++ b/pico/32x/32x.c
@@ -4,10 +4,19 @@
 struct Pico32x Pico32x;
 SH2 sh2s[2];
 
-static void sh2_irq_cb(int id, int level)
+static int REGPARM(2) sh2_irq_cb(SH2 *sh2, int level)
 {
-  // diagnostic for now
-  elprintf(EL_32X, "%csh2 ack %d @ %08x", id ? 's' : 'm', level, sh2_pc(id));
+  if (sh2->pending_irl > sh2->pending_int_irq) {
+    elprintf(EL_32X, "%csh2 ack/irl %d @ %08x",
+      sh2->is_slave ? 's' : 'm', level, sh2->pc);
+    return 64 + sh2->pending_irl / 2;
+  } else {
+    elprintf(EL_32X, "%csh2 ack/int %d/%d @ %08x",
+      sh2->is_slave ? 's' : 'm', level, sh2->pending_int_vector, sh2->pc);
+    sh2->pending_int_irq = 0; // auto-clear
+    sh2->pending_level = sh2->pending_irl;
+    return sh2->pending_int_vector;
+  }
 }
 
 void p32x_update_irls(void)
diff --git a/pico/32x/memory.c b/pico/32x/memory.c
index eaf1e3a..7104cef 100644
--- a/pico/32x/memory.c
+++ b/pico/32x/memory.c
@@ -523,7 +523,7 @@ static u32 sh2_peripheral_read32(u32 a, int id)
   return d;
 }
 
-static void sh2_peripheral_write8(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_peripheral_write8(u32 a, u32 d, int id)
 {
   u8 *r = (void *)Pico32xMem->sh2_peri_regs[id];
   elprintf(EL_32X, "%csh2 peri w8  [%08x]       %02x @%06x", id ? 's' : 'm', a, d, sh2_pc(id));
@@ -540,11 +540,13 @@ static void sh2_peripheral_write8(u32 a, u32 d, int id)
       int vector = PREG8(oregs, 0x63) & 0x7f;
       elprintf(EL_32X, "%csh2 SCI recv irq (%d, %d)", (id ^ 1) ? 's' : 'm', level, vector);
       sh2_internal_irq(&sh2s[id ^ 1], level, vector);
+      return 1;
     }
   }
+  return 0;
 }
 
-static void sh2_peripheral_write16(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_peripheral_write16(u32 a, u32 d, int id)
 {
   u16 *r = (void *)Pico32xMem->sh2_peri_regs[id];
   elprintf(EL_32X, "%csh2 peri w16 [%08x]     %04x @%06x", id ? 's' : 'm', a, d, sh2_pc(id));
@@ -559,10 +561,11 @@ static void sh2_peripheral_write16(u32 a, u32 d, int id)
     }
     if ((d & 0xff00) == 0x5a00) // WTCNT
       PREG8(r, 0x81) = d;
-    return;
+    return 0;
   }
 
   r[(a / 2) ^ 1] = d;
+  return 0;
 }
 
 static void sh2_peripheral_write32(u32 a, u32 d, int id)
@@ -1016,52 +1019,55 @@ static u32 sh2_read16_da(u32 a, int id)
   return ((u16 *)Pico32xMem->data_array[id])[(a & 0xfff) / 2];
 }
 
-static void sh2_write_ignore(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write_ignore(u32 a, u32 d, int id)
 {
+  return 0;
 }
 
 // write8
-static void sh2_write8_unmapped(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write8_unmapped(u32 a, u32 d, int id)
 {
   elprintf(EL_UIO, "%csh2 unmapped w8  [%08x]       %02x @%06x",
     id ? 's' : 'm', a, d & 0xff, sh2_pc(id));
+  return 0;
 }
 
-static void sh2_write8_cs0(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write8_cs0(u32 a, u32 d, int id)
 {
   elprintf(EL_32X, "%csh2 w8  [%08x]       %02x @%06x",
     id ? 's' : 'm', a, d & 0xff, sh2_pc(id));
 
   if ((a & 0x3ff00) == 0x4100) {
     p32x_vdp_write8(a, d);
-    return;
+    return 0;
   }
 
   if ((a & 0x3ff00) == 0x4000) {
     p32x_sh2reg_write8(a, d, id);
-    return;
+    return 1;
   }
 
-  sh2_write8_unmapped(a, d, id);
+  return sh2_write8_unmapped(a, d, id);
 }
 
 #define sh2_write8_dramN(n) \
   if (!(a & 0x20000) || d) { \
     u8 *dram = (u8 *)Pico32xMem->dram[n]; \
     dram[(a & 0x1ffff) ^ 1] = d; \
-  }
+  } \
+  return 0;
 
-static void sh2_write8_dram0(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write8_dram0(u32 a, u32 d, int id)
 {
   sh2_write8_dramN(0);
 }
 
-static void sh2_write8_dram1(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write8_dram1(u32 a, u32 d, int id)
 {
   sh2_write8_dramN(1);
 }
 
-static void sh2_write8_sdram(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write8_sdram(u32 a, u32 d, int id)
 {
   u32 a1 = a & 0x3ffff;
 #ifdef DRC_SH2
@@ -1070,9 +1076,10 @@ static void sh2_write8_sdram(u32 a, u32 d, int id)
     sh2_drc_wcheck_ram(a, t, id);
 #endif
   Pico32xMem->sdram[a1 ^ 1] = d;
+  return 0;
 }
 
-static void sh2_write8_da(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write8_da(u32 a, u32 d, int id)
 {
   u32 a1 = a & 0xfff;
 #ifdef DRC_SH2
@@ -1081,16 +1088,18 @@ static void sh2_write8_da(u32 a, u32 d, int id)
     sh2_drc_wcheck_da(a, t, id);
 #endif
   Pico32xMem->data_array[id][a1 ^ 1] = d;
+  return 0;
 }
 
 // write16
-static void sh2_write16_unmapped(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write16_unmapped(u32 a, u32 d, int id)
 {
   elprintf(EL_UIO, "%csh2 unmapped w16 [%08x]     %04x @%06x",
     id ? 's' : 'm', a, d & 0xffff, sh2_pc(id));
+  return 0;
 }
 
-static void sh2_write16_cs0(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write16_cs0(u32 a, u32 d, int id)
 {
   if (((EL_LOGMASK & EL_PWM) || (a & 0x30) != 0x30)) // hide PWM
     elprintf(EL_32X, "%csh2 w16 [%08x]     %04x @%06x",
@@ -1099,45 +1108,46 @@ static void sh2_write16_cs0(u32 a, u32 d, int id)
   if ((a & 0x3ff00) == 0x4100) {
     sh2_poll[id].cnt = 0; // for poll before VDP accesses
     p32x_vdp_write16(a, d);
-    return;
+    return 0;
   }
 
   if ((a & 0x3fe00) == 0x4200) {
     Pico32xMem->pal[(a & 0x1ff) / 2] = d;
     Pico32x.dirty_pal = 1;
-    return;
+    return 0;
   }
 
   if ((a & 0x3ff00) == 0x4000) {
     p32x_sh2reg_write16(a, d, id);
-    return;
+    return 1;
   }
 
-  sh2_write16_unmapped(a, d, id);
+  return sh2_write16_unmapped(a, d, id);
 }
 
 #define sh2_write16_dramN(n) \
   u16 *pd = &Pico32xMem->dram[n][(a & 0x1ffff) / 2]; \
   if (!(a & 0x20000)) { \
     *pd = d; \
-    return; \
+    return 0; \
   } \
   /* overwrite */ \
   if (!(d & 0xff00)) d |= *pd & 0xff00; \
   if (!(d & 0x00ff)) d |= *pd & 0x00ff; \
-  *pd = d
+  *pd = d; \
+  return 0
 
-static void sh2_write16_dram0(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write16_dram0(u32 a, u32 d, int id)
 {
   sh2_write16_dramN(0);
 }
 
-static void sh2_write16_dram1(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write16_dram1(u32 a, u32 d, int id)
 {
   sh2_write16_dramN(1);
 }
 
-static void sh2_write16_sdram(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write16_sdram(u32 a, u32 d, int id)
 {
   u32 a1 = a & 0x3ffff;
 #ifdef DRC_SH2
@@ -1146,9 +1156,10 @@ static void sh2_write16_sdram(u32 a, u32 d, int id)
     sh2_drc_wcheck_ram(a, t, id);
 #endif
   ((u16 *)Pico32xMem->sdram)[a1 / 2] = d;
+  return 0;
 }
 
-static void sh2_write16_da(u32 a, u32 d, int id)
+static int REGPARM(3) sh2_write16_da(u32 a, u32 d, int id)
 {
   u32 a1 = a & 0xfff;
 #ifdef DRC_SH2
@@ -1157,6 +1168,7 @@ static void sh2_write16_da(u32 a, u32 d, int id)
     sh2_drc_wcheck_da(a, t, id);
 #endif
   ((u16 *)Pico32xMem->data_array[id])[a1 / 2] = d;
+  return 0;
 }
 
 
@@ -1165,18 +1177,21 @@ typedef struct {
   u32 mask;
 } sh2_memmap;
 
-typedef u32  (sh2_read_handler)(u32 a, int id);
-typedef void (sh2_write_handler)(u32 a, u32 d, int id);
+typedef u32 (sh2_read_handler)(u32 a, int id);
+typedef int REGPARM(3) (sh2_write_handler)(u32 a, u32 d, int id);
 
-#define SH2MAP_ADDR2OFFS(a) \
-  (((a >> 25) & 3) | ((a >> 27) & 0x1c))
+#define SH2MAP_ADDR2OFFS_R(a) \
+  ((((a) >> 25) & 3) | (((a) >> 27) & 0x1c))
+
+#define SH2MAP_ADDR2OFFS_W(a) \
+  ((u32)(a) >> SH2_WRITE_SHIFT)
 
 u32 REGPARM(2) p32x_sh2_read8(u32 a, SH2 *sh2)
 {
   const sh2_memmap *sh2_map = sh2->read8_map;
   uptr p;
 
-  sh2_map += SH2MAP_ADDR2OFFS(a);
+  sh2_map += SH2MAP_ADDR2OFFS_R(a);
   p = sh2_map->addr;
   if (map_flag_set(p))
     return ((sh2_read_handler *)(p << 1))(a, sh2->is_slave);
@@ -1189,7 +1204,7 @@ u32 REGPARM(2) p32x_sh2_read16(u32 a, SH2 *sh2)
   const sh2_memmap *sh2_map = sh2->read16_map;
   uptr p;
 
-  sh2_map += SH2MAP_ADDR2OFFS(a);
+  sh2_map += SH2MAP_ADDR2OFFS_R(a);
   p = sh2_map->addr;
   if (map_flag_set(p))
     return ((sh2_read_handler *)(p << 1))(a, sh2->is_slave);
@@ -1204,7 +1219,7 @@ u32 REGPARM(2) p32x_sh2_read32(u32 a, SH2 *sh2)
   u32 offs;
   uptr p;
 
-  offs = SH2MAP_ADDR2OFFS(a);
+  offs = SH2MAP_ADDR2OFFS_R(a);
   sh2_map += offs;
   p = sh2_map->addr;
   if (!map_flag_set(p)) {
@@ -1220,40 +1235,42 @@ u32 REGPARM(2) p32x_sh2_read32(u32 a, SH2 *sh2)
   return (handler(a, sh2->is_slave) << 16) | handler(a + 2, sh2->is_slave);
 }
 
-void REGPARM(3) p32x_sh2_write8(u32 a, u32 d, SH2 *sh2)
+// return nonzero if write potentially causes an interrupt (used by drc)
+int REGPARM(3) p32x_sh2_write8(u32 a, u32 d, SH2 *sh2)
 {
   const void **sh2_wmap = sh2->write8_tab;
   sh2_write_handler *wh;
 
-  wh = sh2_wmap[SH2MAP_ADDR2OFFS(a)];
-  wh(a, d, sh2->is_slave);
+  wh = sh2_wmap[SH2MAP_ADDR2OFFS_W(a)];
+  return wh(a, d, sh2->is_slave);
 }
 
-void REGPARM(3) p32x_sh2_write16(u32 a, u32 d, SH2 *sh2)
+int REGPARM(3) p32x_sh2_write16(u32 a, u32 d, SH2 *sh2)
 {
   const void **sh2_wmap = sh2->write16_tab;
   sh2_write_handler *wh;
 
-  wh = sh2_wmap[SH2MAP_ADDR2OFFS(a)];
-  wh(a, d, sh2->is_slave);
+  wh = sh2_wmap[SH2MAP_ADDR2OFFS_W(a)];
+  return wh(a, d, sh2->is_slave);
 }
 
-void REGPARM(3) p32x_sh2_write32(u32 a, u32 d, SH2 *sh2)
+int REGPARM(3) p32x_sh2_write32(u32 a, u32 d, SH2 *sh2)
 {
   const void **sh2_wmap = sh2->write16_tab;
   sh2_write_handler *handler;
   u32 offs;
 
-  offs = SH2MAP_ADDR2OFFS(a);
+  offs = SH2MAP_ADDR2OFFS_W(a);
 
-  if (offs == 0x1f) {
+  if (offs == SH2MAP_ADDR2OFFS_W(0xffffc000)) {
     sh2_peripheral_write32(a, d, sh2->is_slave);
-    return;
+    return 0;
   }
 
   handler = sh2_wmap[offs];
   handler(a, d >> 16, sh2->is_slave);
   handler(a + 2, d, sh2->is_slave);
+  return 0;
 }
 
 // -----------------------------------------------------------------
@@ -1380,7 +1397,7 @@ static void get_bios(void)
 
 static sh2_memmap sh2_read8_map[0x20], sh2_read16_map[0x20];
 // for writes we are using handlers only
-static void *sh2_write8_map[0x20], *sh2_write16_map[0x20];
+static sh2_write_handler *sh2_write8_map[0x80], *sh2_write16_map[0x80];
 
 void Pico32xSwapDRAM(int b)
 {
@@ -1393,8 +1410,8 @@ void Pico32xSwapDRAM(int b)
   sh2_read8_map[2].addr   = sh2_read8_map[6].addr   =
   sh2_read16_map[2].addr  = sh2_read16_map[6].addr  = MAP_MEMORY(Pico32xMem->dram[b]);
 
-  sh2_write8_map[2]  = sh2_write8_map[6]  = b ? sh2_write8_dram1 : sh2_write8_dram0;
-  sh2_write16_map[2] = sh2_write16_map[6] = b ? sh2_write16_dram1 : sh2_write16_dram0;
+  sh2_write8_map[0x04/2]  = sh2_write8_map[0x24/2]  = b ? sh2_write8_dram1 : sh2_write8_dram0;
+  sh2_write16_map[0x04/2] = sh2_write16_map[0x24/2] = b ? sh2_write16_dram1 : sh2_write16_dram0;
 }
 
 void PicoMemSetup32x(void)
@@ -1448,24 +1465,27 @@ void PicoMemSetup32x(void)
 
   // SH2 maps: A31,A30,A29,CS1,CS0
   // all unmapped by default
-  for (i = 0; i < 0x20; i++) {
+  for (i = 0; i < ARRAY_SIZE(sh2_read8_map); i++) {
     sh2_read8_map[i].addr   = MAP_HANDLER(sh2_read8_unmapped);
     sh2_read16_map[i].addr  = MAP_HANDLER(sh2_read16_unmapped);
+  }
+
+  for (i = 0; i < ARRAY_SIZE(sh2_write8_map); i++) {
     sh2_write8_map[i]       = sh2_write8_unmapped;
     sh2_write16_map[i]      = sh2_write16_unmapped;
   }
 
   // "purge area"
-  for (i = 0x08; i <= 0x0b; i++) {
-    sh2_write8_map[i]       =
-    sh2_write16_map[i]      = sh2_write_ignore;
+  for (i = 0x40; i <= 0x5f; i++) {
+    sh2_write8_map[i >> 1]  =
+    sh2_write16_map[i >> 1] = sh2_write_ignore;
   }
 
   // CS0
   sh2_read8_map[0].addr   = sh2_read8_map[4].addr   = MAP_HANDLER(sh2_read8_cs0);
   sh2_read16_map[0].addr  = sh2_read16_map[4].addr  = MAP_HANDLER(sh2_read16_cs0);
-  sh2_write8_map[0]       = sh2_write8_map[4]       = sh2_write8_cs0;
-  sh2_write16_map[0]      = sh2_write16_map[4]      = sh2_write16_cs0;
+  sh2_write8_map[0x00/2]  = sh2_write8_map[0x20/2]  = sh2_write8_cs0;
+  sh2_write16_map[0x00/2] = sh2_write16_map[0x20/2] = sh2_write16_cs0;
   // CS1 - ROM
   sh2_read8_map[1].addr   = sh2_read8_map[5].addr   =
   sh2_read16_map[1].addr  = sh2_read16_map[5].addr  = MAP_MEMORY(Pico.rom);
@@ -1477,20 +1497,20 @@ void PicoMemSetup32x(void)
   // CS3 - SDRAM
   sh2_read8_map[3].addr   = sh2_read8_map[7].addr   =
   sh2_read16_map[3].addr  = sh2_read16_map[7].addr  = MAP_MEMORY(Pico32xMem->sdram);
-  sh2_write8_map[3]       = sh2_write8_map[7]       = sh2_write8_sdram;
-  sh2_write16_map[3]      = sh2_write16_map[7]      = sh2_write16_sdram;
+  sh2_write8_map[0x06/2]  = sh2_write8_map[0x26/2]  = sh2_write8_sdram;
+  sh2_write16_map[0x06/2] = sh2_write16_map[0x26/2] = sh2_write16_sdram;
   sh2_read8_map[3].mask   = sh2_read8_map[7].mask   =
   sh2_read16_map[3].mask  = sh2_read16_map[7].mask  = 0x03ffff;
   // SH2 data array
   sh2_read8_map[0x18].addr   = MAP_HANDLER(sh2_read8_da);
   sh2_read16_map[0x18].addr  = MAP_HANDLER(sh2_read16_da);
-  sh2_write8_map[0x18]       = sh2_write8_da;
-  sh2_write16_map[0x18]      = sh2_write16_da;
+  sh2_write8_map[0xc0/2]     = sh2_write8_da;
+  sh2_write16_map[0xc0/2]    = sh2_write16_da;
   // SH2 IO
   sh2_read8_map[0x1f].addr   = MAP_HANDLER(sh2_peripheral_read8);
   sh2_read16_map[0x1f].addr  = MAP_HANDLER(sh2_peripheral_read16);
-  sh2_write8_map[0x1f]       = sh2_peripheral_write8;
-  sh2_write16_map[0x1f]      = sh2_peripheral_write16;
+  sh2_write8_map[0xff/2]     = sh2_peripheral_write8;
+  sh2_write16_map[0xff/2]    = sh2_peripheral_write16;
 
   // map DRAM area, both 68k and SH2
   Pico32xSwapDRAM(1);
diff --git a/pico/pico_int.h b/pico/pico_int.h
index 68a77c0..2f4d1d3 100644
--- a/pico/pico_int.h
+++ b/pico/pico_int.h
@@ -469,6 +469,8 @@ typedef struct
 #define SH2_DRCBLK_RAM_SHIFT 1
 #define SH2_DRCBLK_DA_SHIFT  1
 
+#define SH2_WRITE_SHIFT 25
+
 struct Pico32x
 {
   unsigned short regs[0x20];
diff --git a/platform/linux/Makefile b/platform/linux/Makefile
index 250f824..47a8dcc 100644
--- a/platform/linux/Makefile
+++ b/platform/linux/Makefile
@@ -68,7 +68,7 @@ vpath %.s = ../..
 vpath %.S = ../..
 vpath %.asm = ../..
 
-DIRS += platform/linux
+DIRS += platform/linux zlib unzip
 
 all: mkdirs PicoDrive
 
-- 
2.39.2