X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fassem_arm64.c;h=271bee5807e69d7663b20503ad0126c800fa5e99;hb=9b495f6ec3f28cf5ed1d41f6af16a9967fcf3e64;hp=3a88f9efc78594b45aed93866d5da9840fc20d95;hpb=104df9d3b15f92d5c73d2d6beb6f01f0cc158e03;p=pcsx_rearmed.git

diff --git a/libpcsxcore/new_dynarec/assem_arm64.c b/libpcsxcore/new_dynarec/assem_arm64.c
index 3a88f9ef..271bee58 100644
--- a/libpcsxcore/new_dynarec/assem_arm64.c
+++ b/libpcsxcore/new_dynarec/assem_arm64.c
@@ -23,15 +23,13 @@
 #include "pcnt.h"
 #include "arm_features.h"
 
-#define unused __attribute__((unused))
-
 void do_memhandler_pre();
 void do_memhandler_post();
 
 /* Linker */
 static void set_jump_target(void *addr, void *target)
 {
-  u_int *ptr = addr;
+  u_int *ptr = NDRC_WRITE_OFFSET(addr);
   intptr_t offset = (u_char *)target - (u_char *)addr;
 
   if ((*ptr&0xFC000000) == 0x14000000) { // b
@@ -45,7 +43,7 @@ static void set_jump_target(void *addr, void *target)
     // should only happen when jumping to an already compiled block (see add_jump_out)
     // a workaround would be to do a trampoline jump via a stub at the end of the block
     assert(-1048576 <= offset && offset < 1048576);
-    *ptr=(*ptr&0xFF00000F)|(((offset>>2)&0x7ffff)<<5);
+    *ptr=(*ptr&0xFF00001F)|(((offset>>2)&0x7ffff)<<5);
   }
   else if((*ptr&0x9f000000)==0x10000000) { // adr
     // generated by do_miniht_insert
@@ -142,7 +140,7 @@ static unused const char *condname[16] = {
 
 static void output_w32(u_int word)
 {
-  *((u_int *)out) = word;
+  *((u_int *)NDRC_WRITE_OFFSET(out)) = word;
   out += 4;
 }
 
@@ -399,6 +397,27 @@ static void emit_movimm(u_int imm, u_int rt)
   }
 }
 
+static void emit_movimm64(uint64_t imm, u_int rt)
+{
+  u_int shift, op, imm16, insns = 0;
+  for (shift = 0; shift < 4; shift++) {
+    imm16 = (imm >> shift * 16) & 0xffff;
+    if (!imm16)
+      continue;
+    op = insns ? 0xf2800000 : 0xd2800000;
+    assem_debug("mov%c %s,#%#x", insns ? 'k' : 'z', regname64[rt], imm16);
+    if (shift)
+      assem_debug(",lsl #%u", shift * 16);
+    assem_debug("\n");
+    output_w32(op | (shift << 21) | imm16_rd(imm16, rt));
+    insns++;
+  }
+  if (!insns) {
+    assem_debug("movz %s,#0\n", regname64[rt]);
+    output_w32(0xd2800000 | imm16_rd(0, rt));
+  }
+}
+
 static void emit_readword(void *addr, u_int rt)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
@@ -598,6 +617,10 @@ static void emit_addimm_s(u_int s, u_int is64, u_int rs, uintptr_t imm, u_int rt
 
 static void emit_addimm(u_int rs, uintptr_t imm, u_int rt)
 {
+  if (imm == 0) {
+    emit_mov(rs, rt);
+    return;
+  }
   emit_addimm_s(0, 0, rs, imm, rt);
 }
 
@@ -838,6 +861,12 @@ static void emit_cmp(u_int rs,u_int rt)
   output_w32(0x6b000000 | rm_rn_rd(rt, rs, WZR));
 }
 
+static void emit_cmpcs(u_int rs,u_int rt)
+{
+  assem_debug("ccmp %s,%s,#0,cs\n",regname[rs],regname[rt]);
+  output_w32(0x7a400000 | (COND_CS << 12) | rm_rn_rd(rt, rs, 0));
+}
+
 static void emit_set_gz32(u_int rs, u_int rt)
 {
   //assem_debug("set_gz32\n");
@@ -961,9 +990,11 @@ static void emit_cb(u_int isnz, u_int is64, const void *a, u_int r)
   output_w32(0x34000000 | is64 | isnz | imm19_rt(offset, r));
 }
 
-static unused void emit_cbz(const void *a, u_int r)
+static void *emit_cbz(u_int r, const void *a)
 {
+  void *ret = out;
   emit_cb(0, 0, a, r);
+  return ret;
 }
 
 static void emit_jmpreg(u_int r)
@@ -1171,14 +1202,11 @@ static void emit_clz(u_int rs, u_int rt)
 }
 
 // special case for checking invalid_code
-static void emit_cmpmem_indexedsr12_reg(u_int rbase, u_int r, u_int imm)
+static void emit_ldrb_indexedsr12_reg(u_int rbase, u_int r, u_int rt)
 {
-  host_tempreg_acquire();
-  emit_shrimm(r, 12, HOST_TEMPREG);
-  assem_debug("ldrb %s,[%s,%s,uxtw]\n",regname[HOST_TEMPREG],regname64[rbase],regname[HOST_TEMPREG]);
-  output_w32(0x38604800 | rm_rn_rd(HOST_TEMPREG, rbase, HOST_TEMPREG));
-  emit_cmpimm(HOST_TEMPREG, imm);
-  host_tempreg_release();
+  emit_shrimm(r, 12, rt);
+  assem_debug("ldrb %s,[%s,%s,uxtw]\n",regname[rt],regname64[rbase],regname[rt]);
+  output_w32(0x38604800 | rm_rn_rd(rt, rbase, rt));
 }
 
 // special for loadlr_assemble, rs2 is destroyed
@@ -1323,16 +1351,7 @@ static void emit_movimm_from64(u_int rs_val, u_int rs, uintptr_t rt_val, u_int r
   }
   // just move the whole thing. At least on Linux all addresses
   // seem to be 48bit, so 3 insns - not great not terrible
-  assem_debug("movz %s,#%#lx\n", regname64[rt], rt_val & 0xffff);
-  output_w32(0xd2800000 | imm16_rd(rt_val & 0xffff, rt));
-  assem_debug("movk %s,#%#lx,lsl #16\n", regname64[rt], (rt_val >> 16) & 0xffff);
-  output_w32(0xf2a00000 | imm16_rd((rt_val >> 16) & 0xffff, rt));
-  assem_debug("movk %s,#%#lx,lsl #32\n", regname64[rt], (rt_val >> 32) & 0xffff);
-  output_w32(0xf2c00000 | imm16_rd((rt_val >> 32) & 0xffff, rt));
-  if (rt_val >> 48) {
-    assem_debug("movk %s,#%#lx,lsl #48\n", regname64[rt], (rt_val >> 48) & 0xffff);
-    output_w32(0xf2e00000 | imm16_rd((rt_val >> 48) & 0xffff, rt));
-  }
+  emit_movimm64(rt_val, rt);
 }
 
 // trashes x2
@@ -1507,8 +1526,13 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
   emit_addimm(cc<0?2:cc,adj,2);
   if(is_dynamic) {
     uintptr_t l1 = ((uintptr_t *)mem_rtab)[addr>>12] << 1;
-    emit_adrp((void *)l1, 1);
-    emit_addimm64(1, l1 & 0xfff, 1);
+    intptr_t offset = (l1 & ~0xfffl) - ((intptr_t)out & ~0xfffl);
+    if (-4294967296l <= offset && offset < 4294967296l) {
+      emit_adrp((void *)l1, 1);
+      emit_addimm64(1, l1 & 0xfff, 1);
+    }
+    else
+      emit_movimm64(l1, 1);
   }
   else
     emit_far_call(do_memhandler_pre);
@@ -1892,7 +1916,7 @@ static void do_miniht_insert(u_int return_address,u_int rt,int temp) {
   emit_writeword(rt,&mini_ht[(return_address&0xFF)>>3][0]);
 }
 
-static void clear_cache_arm64(char *start, char *end)
+static unused void clear_cache_arm64(char *start, char *end)
 {
   // Don't rely on GCC's __clear_cache implementation, as it caches
   // icache/dcache cache line sizes, that can vary between cores on
@@ -1937,7 +1961,7 @@ static void clear_cache_arm64(char *start, char *end)
 static void arch_init(void)
 {
   uintptr_t diff = (u_char *)&ndrc->tramp.f - (u_char *)&ndrc->tramp.ops;
-  struct tramp_insns *ops = ndrc->tramp.ops;
+  struct tramp_insns *ops = NDRC_WRITE_OFFSET(ndrc->tramp.ops);
   size_t i;
   assert(!(diff & 3));
   start_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops));