X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;ds=sidebyside;f=libpcsxcore%2Fnew_dynarec%2Fassem_arm.c;h=ed00103fd192701fa960905858719c6064440713;hb=d1e4ebd9988a9a5d9fb38b89f19e24b9ab6029d7;hp=d0b1bd7941ee5fd10a7644ffc90f0f321bac1d68;hpb=be516ebe45e48044b599e9d9f9f2d296c3f3ee62;p=pcsx_rearmed.git

diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c
index d0b1bd79..ed00103f 100644
--- a/libpcsxcore/new_dynarec/assem_arm.c
+++ b/libpcsxcore/new_dynarec/assem_arm.c
@@ -49,14 +49,6 @@ u_char translation_cache[1 << TARGET_SIZE_2] __attribute__((aligned(4096)));
 #pragma GCC diagnostic ignored "-Wunused-but-set-variable"
 #endif
 
-extern int cycle_count;
-extern int last_count;
-extern int pcaddr;
-extern int pending_exception;
-extern int branch_target;
-extern uint64_t readmem_dword;
-extern u_int mini_ht[32][2];
-
 void indirect_jump_indexed();
 void indirect_jump();
 void do_interrupt();
@@ -289,7 +281,6 @@ static int isclean(void *addr)
   if((*ptr&0xFF000000)!=0xeb000000) ptr++;
   if((*ptr&0xFF000000)!=0xeb000000) return 1; // bl instruction
   if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code) return 0;
-  if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code_vm) return 0;
   if((int)ptr+((*ptr<<8)>>6)+8==(int)verify_code_ds) return 0;
   return 1;
 }
@@ -423,17 +414,23 @@ static void genimm_checked(u_int imm,u_int *encoded)
 
 static u_int genjmp(u_int addr)
 {
-  int offset=addr-(int)out-8;
-  if(offset<-33554432||offset>=33554432) {
-    if (addr>2) {
-      SysPrintf("genjmp: out of range: %08x\n", offset);
-      exit(1);
-    }
+  if (addr < 3) return 0; // a branch that will be patched later
+  int offset = addr-(int)out-8;
+  if (offset < -33554432 || offset >= 33554432) {
+    SysPrintf("genjmp: out of range: %08x\n", offset);
+    abort();
     return 0;
   }
   return ((u_int)offset>>2)&0xffffff;
 }
 
+static unused void emit_breakpoint(void)
+{
+  assem_debug("bkpt #0\n");
+  //output_w32(0xe1200070);
+  output_w32(0xe7f001f0);
+}
+
 static void emit_mov(int rs,int rt)
 {
   assem_debug("mov %s,%s\n",regname[rt],regname[rs]);
@@ -452,12 +449,6 @@ static void emit_add(int rs1,int rs2,int rt)
   output_w32(0xe0800000|rd_rn_rm(rt,rs1,rs2));
 }
 
-static void emit_adds(int rs1,int rs2,int rt)
-{
-  assem_debug("adds %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
-  output_w32(0xe0900000|rd_rn_rm(rt,rs1,rs2));
-}
-
 static void emit_adcs(int rs1,int rs2,int rt)
 {
   assem_debug("adcs %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
@@ -546,12 +537,15 @@ static void emit_loadreg(int r, int hr)
   if((r&63)==0)
     emit_zeroreg(hr);
   else {
-    int addr=((int)reg)+((r&63)<<REG_SHIFT)+((r&64)>>4);
-    if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4);
-    if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4);
-    if(r==CCREG) addr=(int)&cycle_count;
-    if(r==CSREG) addr=(int)&Status;
-    if(r==INVCP) addr=(int)&invc_ptr;
+    int addr = (int)&psxRegs.GPR.r[r];
+    switch (r) {
+    //case HIREG: addr = &hi; break;
+    //case LOREG: addr = &lo; break;
+    case CCREG: addr = (int)&cycle_count; break;
+    case CSREG: addr = (int)&Status; break;
+    case INVCP: addr = (int)&invc_ptr; break;
+    default: assert(r < 34); break;
+    }
     u_int offset = addr-(u_int)&dynarec_local;
     assert(offset<4096);
     assem_debug("ldr %s,fp+%d\n",regname[hr],offset);
@@ -566,10 +560,13 @@ static void emit_storereg(int r, int hr)
     assert(0);
     return;
   }
-  int addr=((int)reg)+((r&63)<<REG_SHIFT)+((r&64)>>4);
-  if((r&63)==HIREG) addr=(int)&hi+((r&64)>>4);
-  if((r&63)==LOREG) addr=(int)&lo+((r&64)>>4);
-  if(r==CCREG) addr=(int)&cycle_count;
+  int addr = (int)&psxRegs.GPR.r[r];
+  switch (r) {
+  //case HIREG: addr = &hi; break;
+  //case LOREG: addr = &lo; break;
+  case CCREG: addr = (int)&cycle_count; break;
+  default: assert(r < 34); break;
+  }
   u_int offset = addr-(u_int)&dynarec_local;
   assert(offset<4096);
   assem_debug("str %s,fp+%d\n",regname[hr],offset);
@@ -722,23 +719,6 @@ static void emit_addnop(u_int r)
   output_w32(0xe2800000|rd_rn_rm(r,r,0));
 }
 
-static void emit_adcimm(u_int rs,int imm,u_int rt)
-{
-  u_int armval;
-  genimm_checked(imm,&armval);
-  assem_debug("adc %s,%s,#%d\n",regname[rt],regname[rs],imm);
-  output_w32(0xe2a00000|rd_rn_rm(rt,rs,0)|armval);
-}
-
-static void emit_addimm64_32(int rsh,int rsl,int imm,int rth,int rtl)
-{
-  // TODO: if(genimm(imm,&armval)) ...
-  // else
-  emit_movimm(imm,HOST_TEMPREG);
-  emit_adds(HOST_TEMPREG,rsl,rtl);
-  emit_adcimm(rsh,0,rth);
-}
-
 static void emit_andimm(int rs,int imm,int rt)
 {
   u_int armval;
@@ -1049,7 +1029,7 @@ static void emit_set_if_carry32(int rs1, int rs2, int rt)
 static void emit_call(const void *a_)
 {
   int a = (int)a_;
-  assem_debug("bl %x (%x+%x)%s\n",a,(int)out,a-(int)out-8,func_name(a));
+  assem_debug("bl %x (%x+%x)%s\n",a,(int)out,a-(int)out-8,func_name(a_));
   u_int offset=genjmp(a);
   output_w32(0xeb000000|offset);
 }
@@ -1057,7 +1037,7 @@ static void emit_call(const void *a_)
 static void emit_jmp(const void *a_)
 {
   int a = (int)a_;
-  assem_debug("b %x (%x+%x)%s\n",a,(int)out,a-(int)out-8,func_name(a));
+  assem_debug("b %x (%x+%x)%s\n",a,(int)out,a-(int)out-8,func_name(a_));
   u_int offset=genjmp(a);
   output_w32(0xea000000|offset);
 }
@@ -1070,56 +1050,63 @@ static void emit_jne(const void *a_)
   output_w32(0x1a000000|offset);
 }
 
-static void emit_jeq(int a)
+static void emit_jeq(const void *a_)
 {
+  int a = (int)a_;
   assem_debug("beq %x\n",a);
   u_int offset=genjmp(a);
   output_w32(0x0a000000|offset);
 }
 
-static void emit_js(int a)
+static void emit_js(const void *a_)
 {
+  int a = (int)a_;
   assem_debug("bmi %x\n",a);
   u_int offset=genjmp(a);
   output_w32(0x4a000000|offset);
 }
 
-static void emit_jns(int a)
+static void emit_jns(const void *a_)
 {
+  int a = (int)a_;
   assem_debug("bpl %x\n",a);
   u_int offset=genjmp(a);
   output_w32(0x5a000000|offset);
 }
 
-static void emit_jl(int a)
+static void emit_jl(const void *a_)
 {
+  int a = (int)a_;
   assem_debug("blt %x\n",a);
   u_int offset=genjmp(a);
   output_w32(0xba000000|offset);
 }
 
-static void emit_jge(int a)
+static void emit_jge(const void *a_)
 {
+  int a = (int)a_;
   assem_debug("bge %x\n",a);
   u_int offset=genjmp(a);
   output_w32(0xaa000000|offset);
 }
 
-static void emit_jno(int a)
+static void emit_jno(const void *a_)
 {
+  int a = (int)a_;
   assem_debug("bvc %x\n",a);
   u_int offset=genjmp(a);
   output_w32(0x7a000000|offset);
 }
 
-static void emit_jc(int a)
+static void emit_jc(const void *a_)
 {
+  int a = (int)a_;
   assem_debug("bcs %x\n",a);
   u_int offset=genjmp(a);
   output_w32(0x2a000000|offset);
 }
 
-static void emit_jcc(void *a_)
+static void emit_jcc(const void *a_)
 {
   int a = (int)a_;
   assem_debug("bcc %x\n",a);
@@ -1400,21 +1387,6 @@ static void emit_rsbimm(int rs, int imm, int rt)
   output_w32(0xe2600000|rd_rn_rm(rt,rs,0)|armval);
 }
 
-// Load 2 immediates optimizing for small code size
-static void emit_mov2imm_compact(int imm1,u_int rt1,int imm2,u_int rt2)
-{
-  emit_movimm(imm1,rt1);
-  u_int armval;
-  if(genimm(imm2-imm1,&armval)) {
-    assem_debug("add %s,%s,#%d\n",regname[rt2],regname[rt1],imm2-imm1);
-    output_w32(0xe2800000|rd_rn_rm(rt2,rt1,0)|armval);
-  }else if(genimm(imm1-imm2,&armval)) {
-    assem_debug("sub %s,%s,#%d\n",regname[rt2],regname[rt1],imm1-imm2);
-    output_w32(0xe2400000|rd_rn_rm(rt2,rt1,0)|armval);
-  }
-  else emit_movimm(imm2,rt2);
-}
-
 // Conditionally select one of two immediates, optimizing for small code size
 // This will only be called if HAVE_CMOV_IMM is defined
 static void emit_cmov2imm_e_ne_compact(int imm1,int imm2,u_int rt)
@@ -1597,7 +1569,8 @@ static void literal_pool_jumpover(int n)
   set_jump_target(jaddr, out);
 }
 
-static void emit_extjump2(u_char *addr, int target, void *linker)
+// parsed by get_pointer, find_extjump_insn
+static void emit_extjump2(u_char *addr, u_int target, void *linker)
 {
   u_char *ptr=(u_char *)addr;
   assert((ptr[3]&0x0e)==0xa);
@@ -1620,14 +1593,11 @@ static void emit_extjump2(u_char *addr, int target, void *linker)
   emit_jmp(linker);
 }
 
-static void emit_extjump(void *addr, int target)
-{
-  emit_extjump2(addr, target, dyna_linker);
-}
-
-static void emit_extjump_ds(void *addr, int target)
+static void check_extjump2(void *src)
 {
-  emit_extjump2(addr, target, dyna_linker_ds);
+  u_int *ptr = src;
+  assert((ptr[1] & 0x0fff0000) == 0x059f0000); // ldr rx, [pc, #ofs]
+  (void)ptr;
 }
 
 // put rt_val into rt, potentially making use of rs with value rs_val
@@ -1674,23 +1644,6 @@ static int is_similar_value(u_int v1,u_int v2)
   return 0;
 }
 
-// trashes r2
-static void pass_args(int a0, int a1)
-{
-  if(a0==1&&a1==0) {
-    // must swap
-    emit_mov(a0,2); emit_mov(a1,1); emit_mov(2,0);
-  }
-  else if(a0!=0&&a1==0) {
-    emit_mov(a1,1);
-    if (a0>=0) emit_mov(a0,0);
-  }
-  else {
-    if(a0>=0&&a0!=0) emit_mov(a0,0);
-    if(a1>=0&&a1!=1) emit_mov(a1,1);
-  }
-}
-
 static void mov_loadtype_adj(enum stub_type type,int rs,int rt)
 {
   switch(type) {
@@ -1787,40 +1740,14 @@ static void do_readstub(int n)
   emit_jmp(stubs[n].retaddr); // return address
 }
 
-// return memhandler, or get directly accessable address and return 0
-static void *get_direct_memhandler(void *table,u_int addr,enum stub_type type,u_int *addr_host)
-{
-  u_int l1,l2=0;
-  l1=((u_int *)table)[addr>>12];
-  if((l1&(1<<31))==0) {
-    u_int v=l1<<1;
-    *addr_host=v+addr;
-    return NULL;
-  }
-  else {
-    l1<<=1;
-    if(type==LOADB_STUB||type==LOADBU_STUB||type==STOREB_STUB)
-      l2=((u_int *)l1)[0x1000/4 + 0x1000/2 + (addr&0xfff)];
-    else if(type==LOADH_STUB||type==LOADHU_STUB||type==STOREH_STUB)
-      l2=((u_int *)l1)[0x1000/4 + (addr&0xfff)/2];
-    else
-      l2=((u_int *)l1)[(addr&0xfff)/4];
-    if((l2&(1<<31))==0) {
-      u_int v=l2<<1;
-      *addr_host=v+(addr&0xfff);
-      return NULL;
-    }
-    return (void *)(l2<<1);
-  }
-}
-
 static void inline_readstub(enum stub_type type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
 {
   int rs=get_reg(regmap,target);
   int rt=get_reg(regmap,target);
   if(rs<0) rs=get_reg(regmap,-1);
   assert(rs>=0);
-  u_int host_addr=0,is_dynamic,far_call=0;
+  u_int is_dynamic,far_call=0;
+  uintptr_t host_addr = 0;
   void *handler;
   int cc=get_reg(regmap,CCREG);
   if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj+1),cc,target?rs:-1,rt))
@@ -1983,7 +1910,7 @@ static void inline_writestub(enum stub_type type, int i, u_int addr, signed char
   int rt=get_reg(regmap,target);
   assert(rs>=0);
   assert(rt>=0);
-  u_int host_addr=0;
+  uintptr_t host_addr = 0;
   void *handler = get_direct_memhandler(mem_wtab, addr, type, &host_addr);
   if (handler == NULL) {
     if(addr!=host_addr)
@@ -2101,36 +2028,28 @@ static void do_unalignedwritestub(int n)
 #endif
 }
 
-static void do_invstub(int n)
+// this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr
+static void do_dirty_stub_emit_args(u_int arg0)
 {
-  literal_pool(20);
-  u_int reglist=stubs[n].a;
-  set_jump_target(stubs[n].addr, out);
-  save_regs(reglist);
-  if(stubs[n].b!=0) emit_mov(stubs[n].b,0);
-  emit_call(&invalidate_addr);
-  restore_regs(reglist);
-  emit_jmp(stubs[n].retaddr); // return address
+  #ifndef HAVE_ARMV7
+  emit_loadlp((int)source, 1);
+  emit_loadlp((int)copy, 2);
+  emit_loadlp(slen*4, 3);
+  #else
+  emit_movw(((u_int)source)&0x0000FFFF, 1);
+  emit_movw(((u_int)copy)&0x0000FFFF, 2);
+  emit_movt(((u_int)source)&0xFFFF0000, 1);
+  emit_movt(((u_int)copy)&0xFFFF0000, 2);
+  emit_movw(slen*4, 3);
+  #endif
+  emit_movimm(arg0, 0);
 }
 
-void *do_dirty_stub(int i)
+static void *do_dirty_stub(int i)
 {
   assem_debug("do_dirty_stub %x\n",start+i*4);
-  u_int addr=(u_int)source;
-  // Careful about the code output here, verify_dirty needs to parse it.
-  #ifndef HAVE_ARMV7
-  emit_loadlp(addr,1);
-  emit_loadlp((int)copy,2);
-  emit_loadlp(slen*4,3);
-  #else
-  emit_movw(addr&0x0000FFFF,1);
-  emit_movw(((u_int)copy)&0x0000FFFF,2);
-  emit_movt(addr&0xFFFF0000,1);
-  emit_movt(((u_int)copy)&0xFFFF0000,2);
-  emit_movw(slen*4,3);
-  #endif
-  emit_movimm(start+i*4,0);
-  emit_call((int)start<(int)0xC0000000?&verify_code:&verify_code_vm);
+  do_dirty_stub_emit_args(start + i*4);
+  emit_call(verify_code);
   void *entry = out;
   load_regs_entry(i);
   if (entry == out)
@@ -2141,20 +2060,8 @@ void *do_dirty_stub(int i)
 
 static void do_dirty_stub_ds()
 {
-  // Careful about the code output here, verify_dirty needs to parse it.
-  #ifndef HAVE_ARMV7
-  emit_loadlp((int)start<(int)0xC0000000?(int)source:(int)start,1);
-  emit_loadlp((int)copy,2);
-  emit_loadlp(slen*4,3);
-  #else
-  emit_movw(((int)start<(int)0xC0000000?(u_int)source:(u_int)start)&0x0000FFFF,1);
-  emit_movw(((u_int)copy)&0x0000FFFF,2);
-  emit_movt(((int)start<(int)0xC0000000?(u_int)source:(u_int)start)&0xFFFF0000,1);
-  emit_movt(((u_int)copy)&0xFFFF0000,2);
-  emit_movw(slen*4,3);
-  #endif
-  emit_movimm(start+1,0);
-  emit_call(&verify_code_ds);
+  do_dirty_stub_emit_args(start + 1);
+  emit_call(verify_code_ds);
 }
 
 /* Special assem */
@@ -2276,7 +2183,7 @@ static void loadlr_assemble_arm(int i,struct regstat *i_regs)
   int offset;
   void *jaddr=0;
   int memtarget=0,c=0;
-  int fastload_reg_override=0;
+  int fastio_reg_override=-1;
   u_int hr,reglist=0;
   tl=get_reg(i_regs->regmap,rt1[i]);
   s=get_reg(i_regs->regmap,rs1[i]);
@@ -2304,12 +2211,13 @@ static void loadlr_assemble_arm(int i,struct regstat *i_regs)
     }else{
       emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR
     }
-    jaddr=emit_fastpath_cmp_jump(i,temp2,&fastload_reg_override);
+    jaddr=emit_fastpath_cmp_jump(i,temp2,&fastio_reg_override);
   }
   else {
     if(ram_offset&&memtarget) {
+      host_tempreg_acquire();
       emit_addimm(temp2,ram_offset,HOST_TEMPREG);
-      fastload_reg_override=HOST_TEMPREG;
+      fastio_reg_override=HOST_TEMPREG;
     }
     if (opcode[i]==0x22||opcode[i]==0x26) {
       emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
@@ -2320,8 +2228,9 @@ static void loadlr_assemble_arm(int i,struct regstat *i_regs)
   if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR
     if(!c||memtarget) {
       int a=temp2;
-      if(fastload_reg_override) a=fastload_reg_override;
+      if(fastio_reg_override>=0) a=fastio_reg_override;
       emit_readword_indexed(0,a,temp2);
+      if(fastio_reg_override==HOST_TEMPREG) host_tempreg_release();
       if(jaddr) add_stub_r(LOADW_STUB,jaddr,out,i,temp2,i_regs,ccadj[i],reglist);
     }
     else
@@ -2563,7 +2472,7 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs)
         emit_negmi(quotient,quotient); // .. quotient and ..
         emit_negmi(remainder,remainder); // .. remainder for div0 case (will be negated back after jump)
         emit_movs(d2,HOST_TEMPREG);
-        emit_jeq((int)out+52); // Division by zero
+        emit_jeq(out+52); // Division by zero
         emit_negsmi(HOST_TEMPREG,HOST_TEMPREG);
 #ifdef HAVE_ARMV5
         emit_clz(HOST_TEMPREG,quotient);
@@ -2572,7 +2481,7 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs)
         emit_movimm(0,quotient);
         emit_addpl_imm(quotient,1,quotient);
         emit_lslpls_imm(HOST_TEMPREG,1,HOST_TEMPREG);
-        emit_jns((int)out-2*4);
+        emit_jns(out-2*4);
 #endif
         emit_orimm(quotient,1<<31,quotient);
         emit_shr(quotient,quotient,quotient);
@@ -2599,7 +2508,7 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs)
         emit_mov(d1,remainder);
         emit_movimm(0xffffffff,quotient); // div0 case
         emit_test(d2,d2);
-        emit_jeq((int)out+40); // Division by zero
+        emit_jeq(out+40); // Division by zero
 #ifdef HAVE_ARMV5
         emit_clz(d2,HOST_TEMPREG);
         emit_movimm(1<<31,quotient);
@@ -2608,7 +2517,7 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs)
         emit_movimm(0,HOST_TEMPREG);
         emit_addpl_imm(HOST_TEMPREG,1,HOST_TEMPREG);
         emit_lslpls_imm(d2,1,d2);
-        emit_jns((int)out-2*4);
+        emit_jns(out-2*4);
         emit_movimm(1<<31,quotient);
 #endif
         emit_shr(quotient,HOST_TEMPREG,quotient);
@@ -2635,6 +2544,11 @@ static void multdiv_assemble_arm(int i,struct regstat *i_regs)
 }
 #define multdiv_assemble multdiv_assemble_arm
 
+static void do_jump_vaddr(int rs)
+{
+  emit_jmp(jump_vaddr_reg[rs]);
+}
+
 static void do_preload_rhash(int r) {
   // Don't need this for ARM.  On x86, this puts the value 0xf8 into the
   // register.  On ARM the hash can be done with a single instruction (below)
@@ -2657,11 +2571,11 @@ static void do_miniht_jump(int rs,int rh,int ht) {
   emit_cmp(rh,rs);
   emit_ldreq_indexed(ht,4,15);
   #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
-  emit_mov(rs,7);
-  emit_jmp(jump_vaddr_reg[7]);
-  #else
-  emit_jmp(jump_vaddr_reg[rs]);
+  if(rs!=7)
+    emit_mov(rs,7);
+  rs=7;
   #endif
+  do_jump_vaddr(rs);
 }
 
 static void do_miniht_insert(u_int return_address,int rt,int temp) {