sh2 drc, several bug fixes

cpu/drc/emit_arm.c

@@ -1,7 +1,7 @@
 /*
  * Basic macros to emit ARM instructions and some utils
  * Copyright (C) 2008,2009,2010 notaz
- * Copyright (C) 2019 kub
+ * Copyright (C) 2019-2024 kub
  *
  * This work is licensed under the terms of MAME license.
  * See COPYING file in the top-level directory.
@@ -1196,7 +1196,7 @@ static inline void emith_pool_adjust(int tcache_offs, int move_offs)
 
 #define emith_jump_at(ptr, target) do { \
 	u32 *ptr_ = (u32 *)ptr; \
-	u32 val_ = (u32 *)(target) - (u32 *)(ptr) - 2; \
+	u32 val_ = (u32 *)(target) - ptr_ - 2; \
 	EOP_C_B_PTR(ptr_, A_COND_AL, 0, val_ & 0xffffff); \
 } while (0)
 #define emith_jump_at_size() 4

cpu/drc/emit_arm64.c

@@ -1,6 +1,6 @@
 /*
  * Basic macros to emit ARM A64 instructions and some utils
- * Copyright (C) 2019 kub
+ * Copyright (C) 2019-2024 kub
  *
  * This work is licensed under the terms of MAME license.
  * See COPYING file in the top-level directory.

cpu/drc/emit_mips.c

@@ -1,6 +1,6 @@
 /*
  * Basic macros to emit MIPS32/MIPS64 Release 1 or 2 instructions and some utils
- * Copyright (C) 2019 kub
+ * Copyright (C) 2019-2024 kub
  *
  * This work is licensed under the terms of MAME license.
  * See COPYING file in the top-level directory.
@@ -1671,12 +1671,20 @@ static NOINLINE void host_instructions_updated(void *base, void *end, int force)
 	asm volatile(
 	"	rdhwr	%2, $1;"
 	"	bal	0f;"			// needed to allow for jr.hb:
+#if _MIPS_SZPTR == 64
+	"0:	daddiu	$ra, $ra, 3f-0b;"	//   set ra to insn after jr.hb
+#else
 	"0:	addiu	$ra, $ra, 3f-0b;"	//   set ra to insn after jr.hb
+#endif
 	"	beqz	%2, 3f;"
 
 	"1:	synci	0(%0);"
 	"	sltu	%3, %0, %1;"
+#if _MIPS_SZPTR == 64
+	"	daddu	%0, %0, %2;"
+#else
 	"	addu	%0, %0, %2;"
+#endif
 	"	bnez	%3, 1b;"
 
 	"	sync;"

cpu/drc/emit_ppc.c

@@ -1,6 +1,6 @@
 /*
  * Basic macros to emit PowerISA 2.03 64 bit instructions and some utils
- * Copyright (C) 2020 kub
+ * Copyright (C) 2020-2024 kub
  *
  * This work is licensed under the terms of MAME license.
  * See COPYING file in the top-level directory.

cpu/drc/emit_riscv.c

@@ -1,6 +1,6 @@
 /*
  * Basic macros to emit RISC-V RV64IM instructions and some utils
- * Copyright (C) 2019 kub
+ * Copyright (C) 2019-2024 kub
  *
  * This work is licensed under the terms of MAME license.
  * See COPYING file in the top-level directory.
@@ -710,9 +710,9 @@ static void emith_move_imm(int r, uintptr_t imm)
 	if (lui >> 12) {
 		EMIT(R5_MOVT_IMM(r, lui));
 		if (imm & 0xfff)
-			EMIT(R5_ADD_IMM(r, r, imm));
+			EMIT(R5_ADDW_IMM(r, r, imm));
 	} else
-		EMIT(R5_ADD_IMM(r, Z0, imm));
+		EMIT(R5_ADDW_IMM(r, Z0, imm));
 }
 
 static void emith_move_ptr_imm(int r, uintptr_t imm)

cpu/drc/emit_x86.c

@@ -1,7 +1,7 @@
 /*
  * Basic macros to emit x86 instructions and some utils
  * Copyright (C) 2008,2009,2010 notaz
- * Copyright (C) 2019 kub
+ * Copyright (C) 2019-2024 kub
  *
  * This work is licensed under the terms of MAME license.
  * See COPYING file in the top-level directory.
@@ -1365,7 +1365,8 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI,	// x86-64,i386 common
 	/* overflow if top 17 bits of MACH aren't all 1 or 0 */ \
 	/* to check: add MACH >> 31 to MACH >> 15. this is 0 if no overflow */ \
 	emith_asr(rn, mh, 15);                    \
-	emith_addf_r_r_r_lsr(rn, rn, mh, 31);     \
+	emith_lsr(rm, mh, 31);                    \
+	emith_addf_r_r(rn, rm);                   \
 	EMITH_SJMP_START(DCOND_EQ); /* sum != 0 -> -ovl */ \
 	emith_move_r_imm_c(DCOND_NE, ml, 0x00000000); \
 	emith_move_r_imm_c(DCOND_NE, mh, 0x00008000); \

cpu/sh2/compiler.c

@@ -1,7 +1,7 @@
 /*
  * SH2 recompiler
  * (C) notaz, 2009,2010,2013
- * (C) kub, 2018,2019,2020
+ * (C) kub, 2018-2024
  *
  * This work is licensed under the terms of MAME license.
  * See COPYING file in the top-level directory.
@@ -2610,7 +2610,8 @@ static uptr split_address(uptr la, uptr mask, s32 *offs)
 #ifdef __arm__
   // arm32 offset has an add/sub flag and an unsigned 8 bit value, which only
   // allows values of [-255...255]. the value -256 thus can't be used.
-  if (*offs + sign == 0) {
+  if (*offs < 0) {  // TODO not working at all with negative offsets on ARM?
+  //if (*offs == -sign) {
     la -= sign;
     *offs += sign;
   }
@@ -2631,7 +2632,7 @@ static int emit_get_rbase_and_offs(SH2 *sh2, sh2_reg_e r, int rmode, s32 *offs)
   // is r constant and points to a memory region?
   if (! gconst_get(r, &a))
     return -1;
-  poffs = dr_ctx_get_mem_ptr(sh2, a, &mask);
+  poffs = dr_ctx_get_mem_ptr(sh2, a + *offs, &mask);
   if (poffs == -1)
     return -1;
 
@@ -3244,10 +3245,11 @@ static void emit_branch_linkage_code(SH2 *sh2, struct block_desc *block, int tca
 }
 
 #define FLUSH_CYCLES(sr) \
-  if (cycles > 0) { \
+  if (cycles > 0) \
     emith_sub_r_imm(sr, cycles << 12); \
-    cycles = 0; \
-  }
+  else if (cycles < 0) /* may happen after a branch not taken */ \
+    emith_add_r_imm(sr, -cycles << 12); \
+  cycles = 0; \
 
 static void *dr_get_pc_base(u32 pc, SH2 *sh2);
 static void sh2_smc_rm_blocks(u32 a, int len, int tcache_id, int free);
@@ -3960,10 +3962,10 @@ static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
 #if DIV_OPTIMIZER
           if (div(opd).div1 == 16 && div(opd).ro == div(opd).rn) {
             // divide 32/16
-            tmp = rcache_get_tmp_arg(1);
-            emith_add_r_r_ptr_imm(tmp, CONTEXT_REG, offsetof(SH2, drc_tmp));
             rcache_get_reg_arg(0, div(opd).rn, NULL);
             rcache_get_reg_arg(2, div(opd).rm, NULL);
+            tmp = rcache_get_tmp_arg(1);
+            emith_add_r_r_ptr_imm(tmp, CONTEXT_REG, offsetof(SH2, drc_tmp));
             rcache_invalidate_tmp();
             emith_abicall(sh2_drc_divu32);
             tmp = rcache_get_tmp_ret();
@@ -3979,16 +3981,17 @@ static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
             emith_or_r_r_r(sr, sr, tmp3);         // T
             rcache_free_tmp(tmp3);
             skip_op = div(opd).div1 + div(opd).rotcl;
+            cycles += skip_op;
           }
           else if (div(opd).div1 == 32 && div(opd).ro != div(opd).rn) {
             // divide 64/32
             tmp4 = rcache_get_reg(div(opd).ro, RC_GR_READ, NULL);
             emith_ctx_write(tmp4, offsetof(SH2, drc_tmp));
             rcache_free(tmp4);
-            tmp = rcache_get_tmp_arg(1);
-            emith_add_r_r_ptr_imm(tmp, CONTEXT_REG, offsetof(SH2, drc_tmp));
             rcache_get_reg_arg(0, div(opd).rn, NULL);
             rcache_get_reg_arg(2, div(opd).rm, NULL);
+            tmp = rcache_get_tmp_arg(1);
+            emith_add_r_r_ptr_imm(tmp, CONTEXT_REG, offsetof(SH2, drc_tmp));
             rcache_invalidate_tmp();
             emith_abicall(sh2_drc_divu64);
             tmp = rcache_get_tmp_ret();
@@ -4004,6 +4007,7 @@ static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
             emith_or_r_r_lsl(sr, tmp3, Q_SHIFT);
             rcache_free_tmp(tmp3);
             skip_op = div(opd).div1 + div(opd).rotcl;
+            cycles += skip_op;
           }
 #endif
           break;
@@ -4085,13 +4089,12 @@ static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
 #if DIV_OPTIMIZER
         if (div(opd).div1 == 16 && div(opd).ro == div(opd).rn) {
           // divide 32/16
-          tmp = rcache_get_tmp_arg(1);
-          emith_add_r_r_ptr_imm(tmp, CONTEXT_REG, offsetof(SH2, drc_tmp));
-          rcache_get_reg_arg(0, div(opd).rn, NULL);
+          tmp = rcache_get_reg_arg(0, div(opd).rn, NULL);
           tmp2 = rcache_get_reg_arg(2, div(opd).rm, NULL);
-          tmp3 = rcache_get_tmp();
+          tmp3 = rcache_get_tmp_arg(1);
           emith_lsr(tmp3, tmp2, 31);
           emith_or_r_r_lsl(sr, tmp3, M_SHIFT);        // M = Rm[31]
+          emith_add_r_r_ptr_imm(tmp3, CONTEXT_REG, offsetof(SH2, drc_tmp));
           rcache_invalidate_tmp();
           emith_abicall(sh2_drc_divs32);
           tmp = rcache_get_tmp_ret();
@@ -4108,6 +4111,7 @@ static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
           emith_or_r_r_r(sr, sr, tmp3);               // T
           rcache_free_tmp(tmp3);
           skip_op = div(opd).div1 + div(opd).rotcl;
+          cycles += skip_op;
         }
         else if (div(opd).div1 == 32 && div(opd).ro != div(opd).rn) {
           // divide 64/32
@@ -4138,6 +4142,7 @@ static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
           emith_or_r_r_lsl(sr, tmp3, Q_SHIFT);        // Q = !Ro[0]^M
           rcache_free_tmp(tmp3);
           skip_op = div(opd).div1 + div(opd).rotcl;
+          cycles += skip_op;
         } else
 #endif
         {
@@ -5113,7 +5118,7 @@ static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
         emith_move_r_imm_s8_patch(rtsadd, tcache_ptr - (u8 *)rtsret);
 #endif
 
-      // branch not taken, correct cycle count
+      // branch not taken, correct cycle count (now, cycles < 0)
       if (ctaken)
         cycles -= ctaken;
       // set T bit to reflect branch not taken for OP_BRANCH_CT/CF
@@ -5243,10 +5248,6 @@ static void REGPARM(2) *sh2_translate(SH2 *sh2, int tcache_id)
  printf("~~~\n");
 */
 
-#if (DRC_DEBUG)
-  fflush(stdout);
-#endif
-
   return block_entry_ptr;
 }
 
@@ -5675,8 +5676,9 @@ static void sh2_smc_rm_blocks(u32 a, int len, int tcache_id, int free)
     a += rest, len -= rest;
   } while (len > 0);
 
-  if (!removed && len <= 4) {
-    dbg(2, "rm_blocks called @%08x, no work?", _a);
+  if (!removed) {
+    if (len <= 4)
+      dbg(2, "rm_blocks called @%08x, no work?", _a);
     return;
   }
 
@@ -5984,7 +5986,6 @@ int sh2_drc_init(SH2 *sh2)
     // disasm the utils
     tcache_dsm_ptrs[0] = tcache;
     do_host_disasm(0);
-    fflush(stdout);
 #endif
 #if (DRC_DEBUG & 1)
     hash_collisions = 0;

cpu/sh2/mame/sh2pico.c

@@ -1,7 +1,7 @@
 #include "../sh2.h"
 
 #ifdef DRC_CMP
-#include "../compiler.c"
+#include "../compiler.h"
 #define BUSY_LOOP_HACKS 0
 #else
 #define BUSY_LOOP_HACKS 1

pico/32x/sh2soc.c

@@ -435,24 +435,25 @@ void REGPARM(3) sh2_peripheral_write32(u32 a, u32 d, SH2 *sh2)
   old = r[a / 4];
   r[a / 4] = d;
 
+  // TODO: DRC doesn't correctly extend 'd' parameter register to 64bit :-/
   switch (a) {
     // division unit (TODO: verify):
     case 0x104: // DVDNT: divident L, starts divide
       elprintf_sh2(sh2, EL_32XP, "divide %08x / %08x",
-        d, r[0x100 / 4]);
+        r[0x104 / 4], r[0x100 / 4]);
       if (r[0x100 / 4]) {
         signed int divisor = r[0x100 / 4];
-                       r[0x118 / 4] = r[0x110 / 4] = (signed int)d % divisor;
-        r[0x104 / 4] = r[0x11c / 4] = r[0x114 / 4] = (signed int)d / divisor;
+                       r[0x118 / 4] = r[0x110 / 4] = (signed int)r[0x104 / 4] % divisor;
+        r[0x104 / 4] = r[0x11c / 4] = r[0x114 / 4] = (signed int)r[0x104 / 4] / divisor;
       }
       else
         r[0x110 / 4] = r[0x114 / 4] = r[0x118 / 4] = r[0x11c / 4] = 0; // ?
       break;
     case 0x114:
       elprintf_sh2(sh2, EL_32XP, "divide %08x%08x / %08x @%08x",
-        r[0x110 / 4], d, r[0x100 / 4], sh2_pc(sh2));
+        r[0x110 / 4], r[0x114 / 4], r[0x100 / 4], sh2_pc(sh2));
       if (r[0x100 / 4]) {
-        signed long long divident = (signed long long)r[0x110 / 4] << 32 | d;
+        signed long long divident = (signed long long)r[0x110 / 4] << 32 | r[0x114 / 4];
         signed int divisor = r[0x100 / 4];
         // XXX: undocumented mirroring to 0x118,0x11c?
         r[0x118 / 4] = r[0x110 / 4] = divident % divisor;

platform/libretro/libretro.c

@@ -2542,7 +2542,7 @@ void retro_init(void)
       | POPT_EN_MCD_PCM|POPT_EN_MCD_CDDA|POPT_EN_MCD_GFX
       | POPT_EN_32X|POPT_EN_PWM
       | POPT_ACC_SPRITES|POPT_DIS_32C_BORDER;
-#ifdef __arm__
+#ifdef DRC_SH2
 #ifdef _3DS
    if (ctr_svchack_successful)
 #endif

platform/linux/emu.c

@@ -36,7 +36,7 @@ void pemu_prep_defconfig(void)
 
 void pemu_validate_config(void)
 {
-#if !defined(__arm__) && !defined(__aarch64__) && !defined(__mips__) && !defined(__riscv__) &&  !defined(__riscv) && !defined(__powerpc__) && !defined(__ppc__)  && !defined(__PPC__) && !defined(__i386__) && !defined(__x86_64__)
+#if !defined(DRC_SH2)
 	PicoIn.opt &= ~POPT_EN_DRC;
 #endif
 }