Browse Source

drc: revive x86 dynarec, support x86-64

notaz 6 years ago
parent
commit
898d51a7fd
10 changed files with 285 additions and 123 deletions
  1. 6 2
      Makefile
  2. 2 0
      Makefile.libretro
  3. 2 0
      cpu/drc/cmn.h
  4. 13 0
      cpu/drc/emit_arm.c
  5. 176 80
      cpu/drc/emit_x86.c
  6. 75 37
      cpu/sh2/compiler.c
  7. 8 1
      cpu/sh2/sh2.c
  8. 1 1
      cpu/sh2/sh2.h
  9. 1 1
      platform/libpicofe
  10. 1 1
      platform/linux/emu.c

+ 6 - 2
Makefile

@@ -44,6 +44,9 @@ asm_mix ?= 1
 else # if not arm
 use_fame ?= 1
 use_cz80 ?= 1
+ifneq (,$(findstring 86,$(ARCH)))
+use_sh2drc ?= 1
+endif
 endif
 
 -include Makefile.local
@@ -221,8 +224,9 @@ pico/carthw_cfg.c: pico/carthw.cfg
 	tools/make_carthw_c $< $@
 
 # random deps
-pico/carthw/svp/compiler.o : cpu/drc/emit_$(ARCH).c
-cpu/sh2/compiler.o : cpu/drc/emit_$(ARCH).c
+pico/carthw/svp/compiler.o : cpu/drc/emit_arm.c
+cpu/sh2/compiler.o : cpu/drc/emit_arm.c
+cpu/sh2/compiler.o : cpu/drc/emit_x86.c
 cpu/sh2/mame/sh2pico.o : cpu/sh2/mame/sh2.c
 pico/pico.o pico/cd/mcd.o pico/32x/32x.o : pico/pico_cmn.c pico/pico_int.h
 pico/memory.o pico/cd/memory.o pico/32x/memory.o : pico/pico_int.h pico/memory.h

+ 2 - 0
Makefile.libretro

@@ -52,6 +52,7 @@ ifeq ($(platform), unix)
 	SHARED := -shared
 	DONT_COMPILE_IN_ZLIB = 1
 	CFLAGS += -DFAMEC_NO_GOTOS
+	use_sh2drc = 1
 
 # Portable Linux
 else ifeq ($(platform), linux-portable)
@@ -62,6 +63,7 @@ else ifeq ($(platform), linux-portable)
 	LIBM :=
 	DONT_COMPILE_IN_ZLIB = 1
 	CFLAGS += -DFAMEC_NO_GOTOS
+	use_sh2drc = 1
 
 # OS X
 else ifeq ($(platform), osx)

+ 2 - 0
cpu/drc/cmn.h

@@ -1,7 +1,9 @@
 typedef unsigned char  u8;
 typedef signed char    s8;
 typedef unsigned short u16;
+typedef signed short   s16;
 typedef unsigned int   u32;
+typedef signed int     s32;
 
 #define DRC_TCACHE_SIZE         (2*1024*1024)
 

+ 13 - 0
cpu/drc/emit_arm.c

@@ -6,6 +6,7 @@
  * See COPYING file in the top-level directory.
  */
 #define CONTEXT_REG 11
+#define RET_REG     0
 
 // XXX: tcache_ptr type for SVP and SH2 compilers differs..
 #define EMIT_PTR(ptr, x) \
@@ -369,6 +370,9 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define EMITH_NOTHING1(cond) \
 	(void)(cond)
 
+#define EMITH_SJMP_DECL_()
+#define EMITH_SJMP_START_(cond)	EMITH_NOTHING1(cond)
+#define EMITH_SJMP_END_(cond)	EMITH_NOTHING1(cond)
 #define EMITH_SJMP_START(cond)	EMITH_NOTHING1(cond)
 #define EMITH_SJMP_END(cond)	EMITH_NOTHING1(cond)
 #define EMITH_SJMP3_START(cond)	EMITH_NOTHING1(cond)
@@ -378,6 +382,9 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_move_r_r(d, s) \
 	EOP_MOV_REG_SIMPLE(d, s)
 
+#define emith_move_r_r_ptr(d, s) \
+	emith_move_r_r(d, s)
+
 #define emith_mvn_r_r(d, s) \
 	EOP_MVN_REG(A_COND_AL,0,d,s,A_AM1_LSL,0)
 
@@ -521,6 +528,9 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_add_r_r_imm(d, s, imm) \
 	emith_op_imm2(A_COND_AL, 0, A_OP_ADD, d, s, imm)
 
+#define emith_add_r_r_ptr_imm(d, s, imm) \
+	emith_add_r_r_imm(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)
 
@@ -611,6 +621,9 @@ static int emith_xbranch(int cond, void *target, int is_call)
 #define emith_ctx_read(r, offs) \
 	emith_read_r_r_offs(r, CONTEXT_REG, offs)
 
+#define emith_ctx_read_ptr(r, offs) \
+	emith_ctx_read(r, offs)
+
 #define emith_ctx_write(r, offs) \
 	EOP_STR_IMM(r, CONTEXT_REG, offs)
 

+ 176 - 80
cpu/drc/emit_x86.c

@@ -16,6 +16,7 @@
 enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 
 #define CONTEXT_REG xBP
+#define RET_REG     xAX
 
 #define ICOND_JO  0x00
 #define ICOND_JNO 0x01
@@ -53,15 +54,15 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define EMIT_PTR(ptr, val, type) \
 	*(type *)(ptr) = val
 
-#define EMIT(val, type) { \
+#define EMIT(val, type) do { \
 	EMIT_PTR(tcache_ptr, val, type); \
 	tcache_ptr += sizeof(type); \
-}
+} while (0)
 
-#define EMIT_OP(op) { \
+#define EMIT_OP(op) do { \
 	COUNT_OP; \
 	EMIT(op, u8); \
-}
+} while (0)
 
 #define EMIT_MODRM(mod,r,rm) \
 	EMIT(((mod)<<6) | ((r)<<3) | (rm), u8)
@@ -69,6 +70,9 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define EMIT_SIB(scale,index,base) \
 	EMIT(((scale)<<6) | ((index)<<3) | (base), u8)
 
+#define EMIT_REX(w,r,x,b) \
+	EMIT(0x40 | ((w)<<3) | ((r)<<2) | ((x)<<1) | (b), u8)
+
 #define EMIT_OP_MODRM(op,mod,r,rm) do { \
 	EMIT_OP(op); \
 	EMIT_MODRM(mod, r, rm); \
@@ -90,6 +94,11 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define emith_move_r_r(dst, src) \
 	EMIT_OP_MODRM(0x8b, 3, dst, src)
 
+#define emith_move_r_r_ptr(dst, src) do { \
+	EMIT_REX_FOR_PTR(); \
+	EMIT_OP_MODRM(0x8b, 3, dst, src); \
+} while (0)
+
 #define emith_add_r_r(d, s) \
 	EMIT_OP_MODRM(0x01, 3, s, d)
 
@@ -118,34 +127,34 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 	EMIT_OP_MODRM(0x39, 3, s, d)
 
 // fake teq - test equivalence - get_flags(d ^ s)
-#define emith_teq_r_r(d, s) { \
+#define emith_teq_r_r(d, s) do { \
 	emith_push(d); \
 	emith_eor_r_r(d, s); \
 	emith_pop(d); \
-}
+} while (0)
 
-#define emith_mvn_r_r(d, s) { \
+#define emith_mvn_r_r(d, s) do { \
 	if (d != s) \
 		emith_move_r_r(d, s); \
 	EMIT_OP_MODRM(0xf7, 3, 2, d); /* NOT d */ \
-}
+} while (0)
 
-#define emith_negc_r_r(d, s) { \
+#define emith_negc_r_r(d, s) do { \
 	int tmp_ = rcache_get_tmp(); \
 	emith_move_r_imm(tmp_, 0); \
 	emith_sbc_r_r(tmp_, s); \
 	emith_move_r_r(d, tmp_); \
 	rcache_free_tmp(tmp_); \
-}
+} while (0)
 
-#define emith_neg_r_r(d, s) { \
+#define emith_neg_r_r(d, s) do { \
 	if (d != s) \
 		emith_move_r_r(d, s); \
 	EMIT_OP_MODRM(0xf7, 3, 3, d); /* NEG d */ \
-}
+} while (0)
 
 // _r_r_r
-#define emith_add_r_r_r(d, s1, s2) { \
+#define emith_add_r_r_r(d, s1, s2) do { \
 	if (d == s1) { \
 		emith_add_r_r(d, s2); \
 	} else if (d == s2) { \
@@ -154,9 +163,9 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 		emith_move_r_r(d, s1); \
 		emith_add_r_r(d, s2); \
 	} \
-}
+} while (0)
 
-#define emith_eor_r_r_r(d, s1, s2) { \
+#define emith_eor_r_r_r(d, s1, s2) do { \
 	if (d == s1) { \
 		emith_eor_r_r(d, s2); \
 	} else if (d == s2) { \
@@ -165,29 +174,29 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 		emith_move_r_r(d, s1); \
 		emith_eor_r_r(d, s2); \
 	} \
-}
+} while (0)
 
 // _r_r_shift
-#define emith_or_r_r_lsl(d, s, lslimm) { \
+#define emith_or_r_r_lsl(d, s, lslimm) do { \
 	int tmp_ = rcache_get_tmp(); \
 	emith_lsl(tmp_, s, lslimm); \
 	emith_or_r_r(d, tmp_); \
 	rcache_free_tmp(tmp_); \
-}
+} while (0)
 
 // d != s
-#define emith_eor_r_r_lsr(d, s, lsrimm) { \
+#define emith_eor_r_r_lsr(d, s, lsrimm) do { \
 	emith_push(s); \
 	emith_lsr(s, s, lsrimm); \
 	emith_eor_r_r(d, s); \
 	emith_pop(s); \
-}
+} while (0)
 
 // _r_imm
-#define emith_move_r_imm(r, imm) { \
+#define emith_move_r_imm(r, imm) do { \
 	EMIT_OP(0xb8 + (r)); \
 	EMIT(imm, u32); \
-}
+} while (0)
 
 #define emith_move_r_imm_s8(r, imm) \
 	emith_move_r_imm(r, (u32)(signed int)(signed char)(imm))
@@ -212,7 +221,14 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define emith_and_r_imm(r, imm) \
 	emith_arith_r_imm(4, r, imm)
 
-#define emith_sub_r_imm(r, imm) \
+/* used for sub cycles after test, so retain flags with lea */
+#define emith_sub_r_imm(r, imm) do { \
+	assert(r != xSP); \
+	EMIT_OP_MODRM(0x8d, 2, r, r); \
+	EMIT(-(s32)(imm), s32); \
+} while (0)
+
+#define emith_subf_r_imm(r, imm) \
 	emith_arith_r_imm(5, r, imm)
 
 #define emith_eor_r_imm(r, imm) \
@@ -231,20 +247,20 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 	emith_arith_r_imm(4, r, ~(imm))
 
 // fake conditionals (using SJMP instead)
-#define emith_move_r_imm_c(cond, r, imm) { \
+#define emith_move_r_imm_c(cond, r, imm) do { \
 	(void)(cond); \
 	emith_move_r_imm(r, imm); \
-}
+} while (0)
 
-#define emith_add_r_imm_c(cond, r, imm) { \
+#define emith_add_r_imm_c(cond, r, imm) do { \
 	(void)(cond); \
 	emith_add_r_imm(r, imm); \
-}
+} while (0)
 
-#define emith_sub_r_imm_c(cond, r, imm) { \
+#define emith_sub_r_imm_c(cond, r, imm) do { \
 	(void)(cond); \
 	emith_sub_r_imm(r, imm); \
-}
+} while (0)
 
 #define emith_or_r_imm_c(cond, r, imm) \
 	emith_or_r_imm(r, imm)
@@ -274,26 +290,40 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #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); \
-}
+// _r_r_imm - use lea
+#define emith_add_r_r_imm(d, s, imm) do { \
+	assert(s != xSP); \
+	EMIT_OP_MODRM(0x8d, 2, d, s); /* lea */ \
+	EMIT(imm, s32); \
+} while (0)
+
+#define emith_add_r_r_ptr_imm(d, s, imm) do { \
+	if (s != xSP) { \
+		EMIT_REX_FOR_PTR(); \
+		EMIT_OP_MODRM(0x8d, 2, d, s); /* lea */ \
+	} \
+	else { \
+		if (d != s) \
+			emith_move_r_r_ptr(d, s); \
+		EMIT_REX_FOR_PTR(); \
+		EMIT_OP_MODRM(0x81, 3, 0, d); /* add */ \
+	} \
+	EMIT(imm, s32); \
+} while (0)
 
-#define emith_and_r_r_imm(d, s, imm) { \
+#define emith_and_r_r_imm(d, s, imm) do { \
 	if (d != s) \
 		emith_move_r_r(d, s); \
 	emith_and_r_imm(d, imm); \
-}
+} while (0)
 
 // shift
-#define emith_shift(op, d, s, cnt) { \
+#define emith_shift(op, d, s, cnt) do { \
 	if (d != s) \
 		emith_move_r_r(d, s); \
 	EMIT_OP_MODRM(0xc1, 3, op, d); \
 	EMIT(cnt, u8); \
-}
+} while (0)
 
 #define emith_lsl(d, s, cnt) \
 	emith_shift(4, d, s, cnt)
@@ -320,10 +350,10 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define emith_push(r) \
 	EMIT_OP(0x50 + (r))
 
-#define emith_push_imm(imm) { \
+#define emith_push_imm(imm) do { \
 	EMIT_OP(0x68); \
 	EMIT(imm, u32); \
-}
+} while (0)
 
 #define emith_pop(r) \
 	EMIT_OP(0x58 + (r))
@@ -349,13 +379,14 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 	emith_asr(d, d, 32 - (bits)); \
 }
 
-#define emith_setc(r) { \
+#define emith_setc(r) do { \
+	assert(is_abcdx(r)); \
 	EMIT_OP(0x0f); \
 	EMIT_OP_MODRM(0x92, 3, 0, r); /* SETC r */ \
-}
+} while (0)
 
 // XXX: stupid mess
-#define emith_mul_(op, dlo, dhi, s1, s2) { \
+#define emith_mul_(op, dlo, dhi, s1, s2) do { \
 	int rmr; \
 	if (dlo != xAX && dhi != xAX) \
 		emith_push(xAX); \
@@ -381,7 +412,7 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 		emith_pop(xDX); \
 	if (dlo != xAX && dhi != xAX) \
 		emith_pop(xAX); \
-}
+} while (0)
 
 #define emith_mul_u64(dlo, dhi, s1, s2) \
 	emith_mul_(4, dlo, dhi, s1, s2) /* MUL */
@@ -393,20 +424,19 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 	emith_mul_(4, d, -1, s1, s2)
 
 // (dlo,dhi) += signed(s1) * signed(s2)
-#define emith_mula_s64(dlo, dhi, s1, s2) { \
+#define emith_mula_s64(dlo, dhi, s1, s2) do { \
 	emith_push(dhi); \
 	emith_push(dlo); \
 	emith_mul_(5, dlo, dhi, s1, s2); \
 	EMIT_OP_MODRM(0x03, 0, dlo, 4); \
-	EMIT_SIB(0, 4, 4); /* add dlo, [esp] */ \
+	EMIT_SIB(0, 4, 4); /* add dlo, [xsp] */ \
 	EMIT_OP_MODRM(0x13, 1, dhi, 4); \
 	EMIT_SIB(0, 4, 4); \
-	EMIT(4, u8); /* adc dhi, [esp+4] */ \
-	emith_add_r_imm(xSP, 4*2); \
-}
+	EMIT(sizeof(void *), u8); /* adc dhi, [xsp+{4,8}] */ \
+	emith_add_r_r_ptr_imm(xSP, xSP, sizeof(void *) * 2); \
+} while (0)
 
 // "flag" instructions are the same
-#define emith_subf_r_imm emith_sub_r_imm
 #define emith_addf_r_r   emith_add_r_r
 #define emith_subf_r_r   emith_sub_r_r
 #define emith_adcf_r_r   emith_adc_r_r
@@ -464,19 +494,24 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 		rcache_free_tmp(r_); \
 } while (0)
 
-#define emith_read16_r_r_offs(r, rs, offs) { \
+#define emith_read16_r_r_offs(r, rs, offs) do { \
 	EMIT(0x66, u8); /* operand override */ \
 	emith_read_r_r_offs(r, rs, offs); \
-}
+} while (0)
 
-#define emith_write16_r_r_offs(r, rs, offs) { \
+#define emith_write16_r_r_offs(r, rs, offs) do { \
 	EMIT(0x66, u8); \
 	emith_write_r_r_offs(r, rs, offs); \
-}
+} while (0)
 
 #define emith_ctx_read(r, offs) \
 	emith_read_r_r_offs(r, CONTEXT_REG, offs)
 
+#define emith_ctx_read_ptr(r, offs) do { \
+	EMIT_REX_FOR_PTR(); \
+	emith_deref_op(0x8b, r, CONTEXT_REG, offs); \
+} while (0)
+
 #define emith_ctx_write(r, offs) \
 	emith_write_r_r_offs(r, CONTEXT_REG, offs)
 
@@ -499,7 +534,7 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 }
 
 #define emith_jump(ptr) { \
-	u32 disp = (u32)(ptr) - ((u32)tcache_ptr + 5); \
+	u32 disp = (u8 *)(ptr) - ((u8 *)tcache_ptr + 5); \
 	EMIT_OP(0xe9); \
 	EMIT(disp, u32); \
 }
@@ -507,30 +542,30 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define emith_jump_patchable(target) \
 	emith_jump(target)
 
-#define emith_jump_cond(cond, ptr) { \
-	u32 disp = (u32)(ptr) - ((u32)tcache_ptr + 6); \
+#define emith_jump_cond(cond, ptr) do { \
+	u32 disp = (u8 *)(ptr) - ((u8 *)tcache_ptr + 6); \
 	EMIT(0x0f, u8); \
 	EMIT_OP(0x80 | (cond)); \
 	EMIT(disp, u32); \
-}
+} while (0)
 
 #define emith_jump_cond_patchable(cond, target) \
 	emith_jump_cond(cond, target)
 
 #define emith_jump_patch(ptr, target) do { \
-	u32 disp_ = (u32)(target) - ((u32)(ptr) + 4); \
+	u32 disp_ = (u8 *)(target) - ((u8 *)(ptr) + 4); \
 	u32 offs_ = (*(u8 *)(ptr) == 0x0f) ? 2 : 1; \
 	EMIT_PTR((u8 *)(ptr) + offs_, disp_ - offs_, u32); \
 } while (0)
 
 #define emith_jump_at(ptr, target) { \
-	u32 disp_ = (u32)(target) - ((u32)(ptr) + 5); \
+	u32 disp_ = (u8 *)(target) - ((u8 *)(ptr) + 5); \
 	EMIT_PTR(ptr, 0xe9, u8); \
 	EMIT_PTR((u8 *)(ptr) + 1, disp_, u32); \
 }
 
 #define emith_call(ptr) { \
-	u32 disp = (u32)(ptr) - ((u32)tcache_ptr + 5); \
+	u32 disp = (u8 *)(ptr) - ((u8 *)tcache_ptr + 5); \
 	EMIT_OP(0xe8); \
 	EMIT(disp, u32); \
 }
@@ -541,10 +576,10 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define emith_call_reg(r) \
 	EMIT_OP_MODRM(0xff, 3, 2, r)
 
-#define emith_call_ctx(offs) { \
+#define emith_call_ctx(offs) do { \
 	EMIT_OP_MODRM(0xff, 2, 2, CONTEXT_REG); \
 	EMIT(offs, u32); \
-}
+} while (0)
 
 #define emith_ret() \
 	EMIT_OP(0xc3)
@@ -552,10 +587,10 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define emith_jump_reg(r) \
 	EMIT_OP_MODRM(0xff, 3, 4, r)
 
-#define emith_jump_ctx(offs) { \
+#define emith_jump_ctx(offs) do { \
 	EMIT_OP_MODRM(0xff, 2, 4, CONTEXT_REG); \
 	EMIT(offs, u32); \
-}
+} while (0)
 
 #define emith_push_ret()
 
@@ -584,6 +619,15 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 
 // "simple" jump (no more then a few insns)
 // ARM will use conditional instructions here
+#define EMITH_SJMP_DECL_() \
+	u8 *cond_ptr
+
+#define EMITH_SJMP_START_(cond) \
+	JMP8_POS(cond_ptr)
+
+#define EMITH_SJMP_END_(cond) \
+	JMP8_EMIT(cond, cond_ptr)
+
 #define EMITH_SJMP_START EMITH_JMP_START
 #define EMITH_SJMP_END EMITH_JMP_END
 
@@ -591,20 +635,55 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 #define EMITH_SJMP3_MID EMITH_JMP3_MID
 #define EMITH_SJMP3_END EMITH_JMP3_END
 
-#define emith_pass_arg_r(arg, reg) { \
+#define emith_pass_arg_r(arg, reg) do { \
 	int rd = 7; \
 	host_arg2reg(rd, arg); \
-	emith_move_r_r(rd, reg); \
-}
+	emith_move_r_r_ptr(rd, reg); \
+} while (0)
 
-#define emith_pass_arg_imm(arg, imm) { \
+#define emith_pass_arg_imm(arg, imm) do { \
 	int rd = 7; \
 	host_arg2reg(rd, arg); \
 	emith_move_r_imm(rd, imm); \
-}
+} while (0)
 
 #define host_instructions_updated(base, end)
 
+#ifdef __x86_64__
+
+#define PTR_SCALE 3
+#define NA_TMP_REG xCX // non-arg tmp from reg_temp[]
+
+#define EMIT_REX_FOR_PTR() \
+	EMIT_REX(1,0,0,0)
+
+#define host_arg2reg(rd, arg) \
+	switch (arg) { \
+	case 0: rd = xDI; break; \
+	case 1: rd = xSI; break; \
+	case 2: rd = xDX; break; \
+	}
+
+#define emith_sh2_drc_entry() { \
+	emith_push(xBX); \
+	emith_push(xBP); \
+	emith_push(xSI); /* to align */ \
+}
+
+#define emith_sh2_drc_exit() {  \
+	emith_pop(xSI); \
+	emith_pop(xBP); \
+	emith_pop(xBX); \
+	emith_ret(); \
+}
+
+#else
+
+#define PTR_SCALE 2
+#define NA_TMP_REG xBX // non-arg tmp from reg_temp[]
+
+#define EMIT_REX_FOR_PTR()
+
 #define host_arg2reg(rd, arg) \
 	switch (arg) { \
 	case 0: rd = xAX; break; \
@@ -612,7 +691,6 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 	case 2: rd = xCX; break; \
 	}
 
-/* SH2 drc specific */
 #define emith_sh2_drc_entry() { \
 	emith_push(xBX);        \
 	emith_push(xBP);        \
@@ -628,15 +706,33 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 	emith_ret();            \
 }
 
-// assumes EBX is free temporary
+#endif
+
+#define emith_save_caller_regs(mask) do { \
+	if ((mask) & (1 << xAX)) emith_push(xAX); \
+	if ((mask) & (1 << xCX)) emith_push(xCX); \
+	if ((mask) & (1 << xDX)) emith_push(xDX); \
+	if ((mask) & (1 << xSI)) emith_push(xSI); \
+	if ((mask) & (1 << xDI)) emith_push(xDI); \
+} while (0)
+
+#define emith_restore_caller_regs(mask) do { \
+	if ((mask) & (1 << xDI)) emith_pop(xDI); \
+	if ((mask) & (1 << xSI)) emith_pop(xSI); \
+	if ((mask) & (1 << xDX)) emith_pop(xDX); \
+	if ((mask) & (1 << xCX)) emith_pop(xCX); \
+	if ((mask) & (1 << xAX)) emith_pop(xAX); \
+} while (0)
+
 #define emith_sh2_wcall(a, tab) { \
 	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_move_r_r(arg2_, CONTEXT_REG); \
-	emith_jump_reg(xBX); \
+	emith_lsr(NA_TMP_REG, a, SH2_WRITE_SHIFT); \
+	EMIT_REX_FOR_PTR(); \
+	EMIT_OP_MODRM(0x8b, 0, NA_TMP_REG, 4); \
+	EMIT_SIB(PTR_SCALE, NA_TMP_REG, tab); /* mov tmp, [tab + tmp * {4,8}] */ \
+	emith_move_r_r_ptr(arg2_, CONTEXT_REG); \
+	emith_jump_reg(NA_TMP_REG); \
 }
 
 #define emith_sh2_dtbf_loop() { \
@@ -697,8 +793,8 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI };
 	JMP8_EMIT(ICOND_JE, jmp0); /* do_sub: */  \
 	emith_sub_r_r(rn, rm);                    \
 	JMP8_EMIT_NC(jmp1);      /* done: */      \
-	emith_setc(tmp_);                         \
-	EMIT_OP_MODRM(0x31, 3, tmp_, sr); /* T = Q1 ^ Q2 */ \
+	emith_adc_r_r(tmp_, tmp_);                \
+	emith_eor_r_r(sr, tmp_);                  \
 	rcache_free_tmp(tmp_);                    \
 }
 

+ 75 - 37
cpu/sh2/compiler.c

@@ -326,6 +326,27 @@ static temp_reg_t reg_temp[] = {
   { xDX, },
 };
 
+#elif defined(__x86_64__)
+#include "../drc/emit_x86.c"
+
+static const int reg_map_g2h[] = {
+  -1, -1, -1, -1,
+  -1, -1, -1, -1,
+  -1, -1, -1, -1,
+  -1, -1, -1, -1,
+  -1, -1, -1, xBX,
+  -1, -1, -1, -1,
+};
+
+// ax, cx, dx are usually temporaries by convention
+static temp_reg_t reg_temp[] = {
+  { xAX, },
+  { xCX, },
+  { xDX, },
+  { xSI, },
+  { xDI, },
+};
+
 #else
 #error unsupported arch
 #endif
@@ -919,13 +940,12 @@ do_alloc:
   return tr->hreg;
 }
 
-static int rcache_get_arg_id(int arg)
+static int rcache_get_hr_id(int hr)
 {
-  int i, r = 0;
-  host_arg2reg(r, arg);
+  int i;
 
   for (i = 0; i < ARRAY_SIZE(reg_temp); i++)
-    if (reg_temp[i].hreg == r)
+    if (reg_temp[i].hreg == hr)
       break;
 
   if (i == ARRAY_SIZE(reg_temp)) // can't happen
@@ -938,7 +958,7 @@ static int rcache_get_arg_id(int arg)
     gconst_check_evict(reg_temp[i].greg);
   }
   else if (reg_temp[i].type == HR_TEMP) {
-    printf("arg %d reg %d already used, aborting\n", arg, r);
+    printf("host reg %d already used, aborting\n", hr);
     exit(1);
   }
 
@@ -948,6 +968,13 @@ static int rcache_get_arg_id(int arg)
   return i;
 }
 
+static int rcache_get_arg_id(int arg)
+{
+  int r = 0;
+  host_arg2reg(r, arg);
+  return rcache_get_hr_id(r);
+}
+
 // get a reg to be used as function arg
 static int rcache_get_tmp_arg(int arg)
 {
@@ -957,6 +984,15 @@ static int rcache_get_tmp_arg(int arg)
   return reg_temp[id].hreg;
 }
 
+// ... as return value after a call
+static int rcache_get_tmp_ret(void)
+{
+  int id = rcache_get_hr_id(RET_REG);
+  reg_temp[id].type = HR_TEMP;
+
+  return reg_temp[id].hreg;
+}
+
 // same but caches a reg. RC_GR_READ only.
 static int rcache_get_reg_arg(int arg, sh2_reg_e r)
 {
@@ -1104,8 +1140,8 @@ static int emit_get_rbase_and_offs(u32 a, u32 *offs)
 
   // XXX: could use some related reg
   hr = rcache_get_tmp();
-  emith_ctx_read(hr, poffs);
-  emith_add_r_imm(hr, a & mask & ~0xff);
+  emith_ctx_read_ptr(hr, poffs);
+  emith_add_r_r_ptr_imm(hr, hr, a & mask & ~0xff);
   *offs = a & 0xff; // XXX: ARM oriented..
   return hr;
 }
@@ -1154,7 +1190,7 @@ static int emit_memhandler_read_(int size, int ram_check)
     emith_ctx_write(reg_map_g2h[SHR_SR], SHR_SR * 4);
 
   arg1 = rcache_get_tmp_arg(1);
-  emith_move_r_r(arg1, CONTEXT_REG);
+  emith_move_r_r_ptr(arg1, CONTEXT_REG);
 
 #if 0 // can't do this because of unmapped reads
  // ndef PDB_NET
@@ -1208,8 +1244,7 @@ static int emit_memhandler_read_(int size, int ram_check)
   if (reg_map_g2h[SHR_SR] != -1)
     emith_ctx_read(reg_map_g2h[SHR_SR], SHR_SR * 4);
 
-  // assuming arg0 and retval reg matches
-  return rcache_get_tmp_arg(0);
+  return rcache_get_tmp_ret();
 }
 
 static int emit_memhandler_read(int size)
@@ -1279,7 +1314,7 @@ static void emit_memhandler_write(int size)
     emith_call(sh2_drc_write16);
     break;
   case 2: // 32
-    emith_move_r_r(ctxr, CONTEXT_REG);
+    emith_move_r_r_ptr(ctxr, CONTEXT_REG);
     emith_call(sh2_drc_write32);
     break;
   }
@@ -1351,26 +1386,23 @@ static void emit_do_static_regs(int is_write, int tmpr)
   }
 }
 
+/* just after lookup function, jump to address returned */
 static void emit_block_entry(void)
 {
-  int arg0;
-
-  host_arg2reg(arg0, 0);
-
 #if (DRC_DEBUG & 8) || defined(PDB)
   int arg1, arg2;
   host_arg2reg(arg1, 1);
   host_arg2reg(arg2, 2);
 
   emit_do_static_regs(1, arg2);
-  emith_move_r_r(arg1, CONTEXT_REG);
+  emith_move_r_r_ptr(arg1, CONTEXT_REG);
   emith_move_r_r(arg2, rcache_get_reg(SHR_SR, RC_GR_READ));
   emith_call(sh2_drc_log_entry);
   rcache_invalidate();
 #endif
-  emith_tst_r_r(arg0, arg0);
+  emith_tst_r_r(RET_REG, RET_REG);
   EMITH_SJMP_START(DCOND_EQ);
-  emith_jump_reg_c(DCOND_NE, arg0);
+  emith_jump_reg_c(DCOND_NE, RET_REG);
   EMITH_SJMP_END(DCOND_EQ);
 }
 
@@ -2703,14 +2735,18 @@ end_op:
       struct op_data *opd_b =
         (op_flags[i] & OF_DELAY_OP) ? &ops[i-1] : opd;
       u32 target_pc = opd_b->imm;
-      int cond = -1;
+      int cond = -1, ncond = -1;
       void *target = NULL;
+      EMITH_SJMP_DECL_();
 
       sr = rcache_get_reg(SHR_SR, RC_GR_RMW);
       FLUSH_CYCLES(sr);
+      rcache_clean();
 
-      if (opd_b->op != OP_BRANCH)
+      if (opd_b->op != OP_BRANCH) {
         cond = (opd_b->op == OP_BRANCH_CF) ? DCOND_EQ : DCOND_NE;
+        ncond = (opd_b->op == OP_BRANCH_CF) ? DCOND_NE : DCOND_EQ;
+      }
       if (cond != -1) {
         int ctaken = (op_flags[i] & OF_DELAY_OP) ? 1 : 2;
 
@@ -2719,9 +2755,9 @@ end_op:
         else
           emith_tst_r_imm(sr, T);
 
+        EMITH_SJMP_START_(ncond);
         emith_sub_r_imm_c(cond, sr, ctaken<<12);
       }
-      rcache_clean();
 
 #if LINK_BRANCHES
       if (find_in_array(branch_target_pc, branch_target_count, target_pc) >= 0)
@@ -2750,8 +2786,10 @@ end_op:
           return NULL;
       }
 
-      if (cond != -1)
+      if (cond != -1) {
         emith_jump_cond_patchable(cond, target);
+        EMITH_SJMP_END_(ncond);
+      }
       else {
         emith_jump_patchable(target);
         rcache_invalidate();
@@ -2906,18 +2944,18 @@ static void sh2_generate_utils(void)
   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_add_r_r_ptr_imm(arg2, CONTEXT_REG, offsetof(SH2, drc_tmp));
   emith_call(dr_lookup_block);
   emit_block_entry();
   // lookup failed, call sh2_translate()
-  emith_move_r_r(arg0, CONTEXT_REG);
+  emith_move_r_r_ptr(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_move_r_r_ptr(arg0, CONTEXT_REG);
   emith_ctx_read(arg1, offsetof(SH2, drc_tmp));
   emith_call(sh2_translate);
   emit_block_entry();
@@ -2944,13 +2982,13 @@ static void sh2_generate_utils(void)
   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_move_r_r_ptr(arg2, CONTEXT_REG);
   emith_call(p32x_sh2_write32); // XXX: use sh2_drc_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_move_r_r_ptr(arg2, CONTEXT_REG);
   emith_call(p32x_sh2_write32);
   rcache_invalidate();
   // update I, cycles, do callback
@@ -2960,16 +2998,16 @@ static void sh2_generate_utils(void)
   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_move_r_r_ptr(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_lsl(arg0, RET_REG, 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
+  tmp = emit_memhandler_read(2);
+  emith_ctx_write(tmp, SHR_PC * 4);
+#if defined(__i386__) || defined(__x86_64__)
+  emith_add_r_r_ptr_imm(xSP, xSP, sizeof(void *)); // fix stack
 #endif
   emith_jump(sh2_drc_dispatcher);
   rcache_invalidate();
@@ -2977,19 +3015,19 @@ static void sh2_generate_utils(void)
   // 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
+  emith_move_r_r_ptr(CONTEXT_REG, arg0); // move ctx, arg0
   emit_do_static_regs(0, arg2);
   emith_call(sh2_drc_test_irq);
   emith_jump(sh2_drc_dispatcher);
 
   // sh2_drc_write8(u32 a, u32 d)
   sh2_drc_write8 = (void *)tcache_ptr;
-  emith_ctx_read(arg2, offsetof(SH2, write8_tab));
+  emith_ctx_read_ptr(arg2, offsetof(SH2, write8_tab));
   emith_sh2_wcall(arg0, arg2);
 
   // sh2_drc_write16(u32 a, u32 d)
   sh2_drc_write16 = (void *)tcache_ptr;
-  emith_ctx_read(arg2, offsetof(SH2, write16_tab));
+  emith_ctx_read_ptr(arg2, offsetof(SH2, write16_tab));
   emith_sh2_wcall(arg0, arg2);
 
 #ifdef PDB_NET
@@ -3015,7 +3053,7 @@ static void sh2_generate_utils(void)
     emith_ctx_read(arg2, offsetof(SH2, pdb_io_csum[1]));  \
     emith_adc_r_imm(arg2, 0x01000000);                    \
     emith_ctx_write(arg2, offsetof(SH2, pdb_io_csum[1])); \
-    emith_move_r_r(arg2, CONTEXT_REG);                    \
+    emith_move_r_r_ptr(arg2, CONTEXT_REG);                \
     emith_jump(func); \
     func = tmp; \
   }

+ 8 - 1
cpu/sh2/sh2.c

@@ -237,7 +237,7 @@ static void dump_regs(SH2 *sh2)
 	printf("%csh2 SR:      %03x  PR: %08x\n", csh2, sh2->sr, sh2->pr);
 }
 
-void do_sh2_cmp(SH2 *current)
+void REGPARM(1) do_sh2_cmp(SH2 *current)
 {
 	static int current_slave;
 	static u32 current_val;
@@ -251,6 +251,13 @@ void do_sh2_cmp(SH2 *current)
 	int cycles;
 	int i, ret;
 
+#if 0
+	sr = current->sr;
+	current->sr &= 0x3f3;
+	do_sh2_trace(current, (signed int)sr >> 12);
+	current->sr = sr;
+	return;
+#endif
 	sh2ref[1].is_slave = 1;
 
 	while (1) {

+ 1 - 1
cpu/sh2/sh2.h

@@ -118,7 +118,7 @@ void REGPARM(3) p32x_sh2_write32(unsigned int a, unsigned int d, SH2 *sh2);
 // debug
 #ifdef DRC_CMP
 void do_sh2_trace(SH2 *current, int cycles);
-void do_sh2_cmp(SH2 *current);
+void REGPARM(1) do_sh2_cmp(SH2 *current);
 #endif
 
 #endif /* __SH2_H__ */

+ 1 - 1
platform/libpicofe

@@ -1 +1 @@
-Subproject commit f287890d65ad36ca75bb71d05745693ae78b1490
+Subproject commit 21082d0b2b9910727770674cef9b68b9e97a3155

+ 1 - 1
platform/linux/emu.c

@@ -29,7 +29,7 @@ void pemu_prep_defconfig(void)
 
 void pemu_validate_config(void)
 {
-#ifndef __arm__
+#if !defined(__arm__) && !defined(__i386__) && !defined(__x86_64__)
 	PicoIn.opt &= ~POPT_EN_DRC;
 #endif
 }