diff --git a/target-sh4/translate.c b/target-sh4/translate.c index 5cf2abf696..be3b29ecfe 100644 --- a/target-sh4/translate.c +++ b/target-sh4/translate.c @@ -57,18 +57,38 @@ enum { BS_EXCP = 3, /* We reached an exception condition */ }; -static TCGv cpu_env, cpu_T[2]; +/* global register indexes */ +static TCGv cpu_env; +static TCGv cpu_gregs[24]; + +/* dyngen register indexes */ +static TCGv cpu_T[2]; #include "gen-icount.h" static void sh4_translate_init(void) { + int i; static int done_init = 0; + static const char * const gregnames[24] = { + "R0_BANK0", "R1_BANK0", "R2_BANK0", "R3_BANK0", + "R4_BANK0", "R5_BANK0", "R6_BANK0", "R7_BANK0", + "R8", "R9", "R10", "R11", "R12", "R13", "R14", "R15", + "R0_BANK1", "R1_BANK1", "R2_BANK1", "R3_BANK1", + "R4_BANK1", "R5_BANK1", "R6_BANK1", "R7_BANK1" + }; + if (done_init) return; + cpu_env = tcg_global_reg_new(TCG_TYPE_PTR, TCG_AREG0, "env"); - cpu_T[0] = tcg_global_reg_new(TCG_TYPE_TL, TCG_AREG1, "T0"); - cpu_T[1] = tcg_global_reg_new(TCG_TYPE_TL, TCG_AREG2, "T1"); + cpu_T[0] = tcg_global_reg_new(TCG_TYPE_I32, TCG_AREG1, "T0"); + cpu_T[1] = tcg_global_reg_new(TCG_TYPE_I32, TCG_AREG2, "T1"); + + for (i = 0; i < 24; i++) + cpu_gregs[i] = tcg_global_mem_new(TCG_TYPE_I32, TCG_AREG0, + offsetof(CPUState, gregs[i]), + gregnames[i]); /* register helpers */ #undef DEF_HELPER @@ -78,24 +98,6 @@ static void sh4_translate_init(void) done_init = 1; } -/* General purpose registers moves. */ -static inline void gen_movl_imm_rN(uint32_t arg, int reg) -{ - TCGv tmp = tcg_const_i32(arg); - tcg_gen_st_i32(tmp, cpu_env, offsetof(CPUState, gregs[reg])); - tcg_temp_free(tmp); -} - -static always_inline void gen_movl_T_rN (TCGv t, int reg) -{ - tcg_gen_st_i32(t, cpu_env, offsetof(CPUState, gregs[reg])); -} - -static always_inline void gen_movl_rN_T (TCGv t, int reg) -{ - tcg_gen_ld_i32(t, cpu_env, offsetof(CPUState, gregs[reg])); -} - #ifdef CONFIG_USER_ONLY #define GEN_OP_LD(width, reg) \ @@ -340,29 +342,29 @@ void _decode_opc(DisasContext * ctx) switch (ctx->opcode & 0xf000) { case 0x1000: /* mov.l Rm,@(disp,Rn) */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_addl_imm_T1(B3_0 * 4); gen_op_stl_T0_T1(ctx); return; case 0x5000: /* mov.l @(disp,Rm),Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_addl_imm_T0(B3_0 * 4); gen_op_ldl_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0xe000: /* mov #imm,Rn */ - gen_movl_imm_rN(B7_0s, REG(B11_8)); + tcg_gen_movi_i32(cpu_gregs[REG(B11_8)], B7_0s); return; case 0x9000: /* mov.w @(disp,PC),Rn */ tcg_gen_movi_i32(cpu_T[0], ctx->pc + 4 + B7_0 * 2); gen_op_ldw_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0xd000: /* mov.l @(disp,PC),Rn */ tcg_gen_movi_i32(cpu_T[0], (ctx->pc + 4 + B7_0 * 4) & ~3); gen_op_ldl_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x7000: /* add #imm,Rn */ gen_op_add_imm_rN(B7_0s, REG(B11_8)); @@ -382,324 +384,324 @@ void _decode_opc(DisasContext * ctx) switch (ctx->opcode & 0xf00f) { case 0x6003: /* mov Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x2000: /* mov.b Rm,@Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_stb_T0_T1(ctx); return; case 0x2001: /* mov.w Rm,@Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_stw_T0_T1(ctx); return; case 0x2002: /* mov.l Rm,@Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_stl_T0_T1(ctx); return; case 0x6000: /* mov.b @Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldb_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x6001: /* mov.w @Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldw_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x6002: /* mov.l @Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldl_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x2004: /* mov.b Rm,@-Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_dec1_rN(REG(B11_8)); /* modify register status */ - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_inc1_rN(REG(B11_8)); /* recover register status */ gen_op_stb_T0_T1(ctx); /* might cause re-execution */ gen_op_dec1_rN(REG(B11_8)); /* modify register status */ return; case 0x2005: /* mov.w Rm,@-Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_dec2_rN(REG(B11_8)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_inc2_rN(REG(B11_8)); gen_op_stw_T0_T1(ctx); gen_op_dec2_rN(REG(B11_8)); return; case 0x2006: /* mov.l Rm,@-Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_dec4_rN(REG(B11_8)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_inc4_rN(REG(B11_8)); gen_op_stl_T0_T1(ctx); gen_op_dec4_rN(REG(B11_8)); return; case 0x6004: /* mov.b @Rm+,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldb_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); if ( B11_8 != B7_4 ) gen_op_inc1_rN(REG(B7_4)); return; case 0x6005: /* mov.w @Rm+,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldw_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); if ( B11_8 != B7_4 ) gen_op_inc2_rN(REG(B7_4)); return; case 0x6006: /* mov.l @Rm+,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldl_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); if ( B11_8 != B7_4 ) gen_op_inc4_rN(REG(B7_4)); return; case 0x0004: /* mov.b Rm,@(R0,Rn) */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_add_rN_T1(REG(0)); gen_op_stb_T0_T1(ctx); return; case 0x0005: /* mov.w Rm,@(R0,Rn) */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_add_rN_T1(REG(0)); gen_op_stw_T0_T1(ctx); return; case 0x0006: /* mov.l Rm,@(R0,Rn) */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_add_rN_T1(REG(0)); gen_op_stl_T0_T1(ctx); return; case 0x000c: /* mov.b @(R0,Rm),Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_add_rN_T0(REG(0)); gen_op_ldb_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x000d: /* mov.w @(R0,Rm),Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_add_rN_T0(REG(0)); gen_op_ldw_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x000e: /* mov.l @(R0,Rm),Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_add_rN_T0(REG(0)); gen_op_ldl_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x6008: /* swap.b Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_swapb_T0(); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x6009: /* swap.w Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_swapw_T0(); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x200d: /* xtrct Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_xtrct_T0_T1(); - gen_movl_T_rN(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[1]); return; case 0x300c: /* add Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_add_T0_rN(REG(B11_8)); return; case 0x300e: /* addc Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_addc_T0_T1(); - gen_movl_T_rN(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[1]); return; case 0x300f: /* addv Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_addv_T0_T1(); - gen_movl_T_rN(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[1]); return; case 0x2009: /* and Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_and_T0_rN(REG(B11_8)); return; case 0x3000: /* cmp/eq Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_cmp_eq_T0_T1(); return; case 0x3003: /* cmp/ge Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_cmp_ge_T0_T1(); return; case 0x3007: /* cmp/gt Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_cmp_gt_T0_T1(); return; case 0x3006: /* cmp/hi Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_cmp_hi_T0_T1(); return; case 0x3002: /* cmp/hs Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_cmp_hs_T0_T1(); return; case 0x200c: /* cmp/str Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_cmp_str_T0_T1(); return; case 0x2007: /* div0s Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_div0s_T0_T1(); return; case 0x3004: /* div1 Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_div1_T0_T1(); - gen_movl_T_rN(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[1]); return; case 0x300d: /* dmuls.l Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_dmulsl_T0_T1(); return; case 0x3005: /* dmulu.l Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_dmulul_T0_T1(); return; case 0x600e: /* exts.b Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); tcg_gen_andi_i32(cpu_T[0], cpu_T[0], 0xff); tcg_gen_ext8s_i32(cpu_T[0], cpu_T[0]); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x600f: /* exts.w Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); tcg_gen_andi_i32(cpu_T[0], cpu_T[0], 0xffff); tcg_gen_ext16s_i32(cpu_T[0], cpu_T[0]); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x600c: /* extu.b Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); tcg_gen_andi_i32(cpu_T[0], cpu_T[0], 0xff); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x600d: /* extu.w Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); tcg_gen_andi_i32(cpu_T[0], cpu_T[0], 0xffff); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x000f: /* mac.l @Rm+,@Rn+ */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_ldl_T0_T0(ctx); tcg_gen_mov_i32(cpu_T[0], cpu_T[1]); - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldl_T0_T0(ctx); gen_op_macl_T0_T1(); gen_op_inc4_rN(REG(B11_8)); gen_op_inc4_rN(REG(B7_4)); return; case 0x400f: /* mac.w @Rm+,@Rn+ */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_ldl_T0_T0(ctx); tcg_gen_mov_i32(cpu_T[0], cpu_T[1]); - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldl_T0_T0(ctx); gen_op_macw_T0_T1(); gen_op_inc2_rN(REG(B11_8)); gen_op_inc2_rN(REG(B7_4)); return; case 0x0007: /* mul.l Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_mull_T0_T1(); return; case 0x200f: /* muls.w Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); tcg_gen_andi_i32(cpu_T[0], cpu_T[0], 0xffff); tcg_gen_ext16s_i32(cpu_T[0], cpu_T[0]); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); tcg_gen_andi_i32(cpu_T[1], cpu_T[1], 0xffff); tcg_gen_ext16s_i32(cpu_T[1], cpu_T[1]); gen_op_mulsw_T0_T1(); return; case 0x200e: /* mulu.w Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); tcg_gen_andi_i32(cpu_T[0], cpu_T[0], 0xffff); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); tcg_gen_andi_i32(cpu_T[1], cpu_T[1], 0xffff); gen_op_muluw_T0_T1(); return; case 0x600b: /* neg Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_neg_T0(); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x600a: /* negc Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_negc_T0(); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x6007: /* not Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_not_T0(); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x200b: /* or Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_or_T0_rN(REG(B11_8)); return; case 0x400c: /* shad Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_shad_T0_T1(); - gen_movl_T_rN(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[1]); return; case 0x400d: /* shld Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_shld_T0_T1(); - gen_movl_T_rN(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[1]); return; case 0x3008: /* sub Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_sub_T0_rN(REG(B11_8)); return; case 0x300a: /* subc Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_subc_T0_T1(); - gen_movl_T_rN(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[1]); return; case 0x300b: /* subv Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_subv_T0_T1(); - gen_movl_T_rN(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[1]); return; case 0x2008: /* tst Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_tst_T0_T1(); return; case 0x200a: /* xor Rm,Rn */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_xor_T0_rN(REG(B11_8)); return; case 0xf00c: /* fmov {F,D,X}Rm,{F,D,X}Rn - FPSCR: Nothing */ @@ -714,33 +716,33 @@ void _decode_opc(DisasContext * ctx) case 0xf00a: /* fmov {F,D,X}Rm,@Rn - FPSCR: Nothing */ if (ctx->fpscr & FPSCR_SZ) { gen_op_fmov_drN_DT0(XREG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_stfq_DT0_T1(ctx); } else { gen_op_fmov_frN_FT0(FREG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_stfl_FT0_T1(ctx); } return; case 0xf008: /* fmov @Rm,{F,D,X}Rn - FPSCR: Nothing */ if (ctx->fpscr & FPSCR_SZ) { - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldfq_T0_DT0(ctx); gen_op_fmov_DT0_drN(XREG(B11_8)); } else { - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldfl_T0_FT0(ctx); gen_op_fmov_FT0_frN(FREG(B11_8)); } return; case 0xf009: /* fmov @Rm+,{F,D,X}Rn - FPSCR: Nothing */ if (ctx->fpscr & FPSCR_SZ) { - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldfq_T0_DT0(ctx); gen_op_fmov_DT0_drN(XREG(B11_8)); gen_op_inc8_rN(REG(B7_4)); } else { - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_ldfl_T0_FT0(ctx); gen_op_fmov_FT0_frN(FREG(B11_8)); gen_op_inc4_rN(REG(B7_4)); @@ -750,14 +752,14 @@ void _decode_opc(DisasContext * ctx) if (ctx->fpscr & FPSCR_SZ) { gen_op_dec8_rN(REG(B11_8)); gen_op_fmov_drN_DT0(XREG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_inc8_rN(REG(B11_8)); gen_op_stfq_DT0_T1(ctx); gen_op_dec8_rN(REG(B11_8)); } else { gen_op_dec4_rN(REG(B11_8)); gen_op_fmov_frN_FT0(FREG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_inc4_rN(REG(B11_8)); gen_op_stfl_FT0_T1(ctx); gen_op_dec4_rN(REG(B11_8)); @@ -765,12 +767,12 @@ void _decode_opc(DisasContext * ctx) return; case 0xf006: /* fmov @(R0,Rm),{F,D,X}Rm - FPSCR: Nothing */ if (ctx->fpscr & FPSCR_SZ) { - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_add_rN_T0(REG(0)); gen_op_ldfq_T0_DT0(ctx); gen_op_fmov_DT0_drN(XREG(B11_8)); } else { - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_add_rN_T0(REG(0)); gen_op_ldfl_T0_FT0(ctx); gen_op_fmov_FT0_frN(FREG(B11_8)); @@ -779,12 +781,12 @@ void _decode_opc(DisasContext * ctx) case 0xf007: /* fmov {F,D,X}Rn,@(R0,Rn) - FPSCR: Nothing */ if (ctx->fpscr & FPSCR_SZ) { gen_op_fmov_drN_DT0(XREG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_add_rN_T1(REG(0)); gen_op_stfq_DT0_T1(ctx); } else { gen_op_fmov_frN_FT0(FREG(B7_4)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_add_rN_T1(REG(0)); gen_op_stfl_FT0_T1(ctx); } @@ -841,7 +843,7 @@ void _decode_opc(DisasContext * ctx) gen_op_and_imm_rN(B7_0, REG(0)); return; case 0xcd00: /* and.b #imm,@(R0,GBR) */ - gen_movl_rN_T(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); gen_op_addl_GBR_T0(); tcg_gen_mov_i32(cpu_T[0], cpu_T[1]); gen_op_ldub_T0_T0(ctx); @@ -871,81 +873,81 @@ void _decode_opc(DisasContext * ctx) ctx->flags |= DELAY_SLOT_CONDITIONAL; return; case 0x8800: /* cmp/eq #imm,R0 */ - gen_movl_rN_T(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); gen_op_cmp_eq_imm_T0(B7_0s); return; case 0xc400: /* mov.b @(disp,GBR),R0 */ gen_op_stc_gbr_T0(); gen_op_addl_imm_T0(B7_0); gen_op_ldb_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_gregs[REG(0)], cpu_T[0]); return; case 0xc500: /* mov.w @(disp,GBR),R0 */ gen_op_stc_gbr_T0(); gen_op_addl_imm_T0(B7_0 * 2); gen_op_ldw_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_gregs[REG(0)], cpu_T[0]); return; case 0xc600: /* mov.l @(disp,GBR),R0 */ gen_op_stc_gbr_T0(); gen_op_addl_imm_T0(B7_0 * 4); gen_op_ldl_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_gregs[REG(0)], cpu_T[0]); return; case 0xc000: /* mov.b R0,@(disp,GBR) */ gen_op_stc_gbr_T0(); gen_op_addl_imm_T0(B7_0); tcg_gen_mov_i32(cpu_T[0], cpu_T[1]); - gen_movl_rN_T(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); gen_op_stb_T0_T1(ctx); return; case 0xc100: /* mov.w R0,@(disp,GBR) */ gen_op_stc_gbr_T0(); gen_op_addl_imm_T0(B7_0 * 2); tcg_gen_mov_i32(cpu_T[0], cpu_T[1]); - gen_movl_rN_T(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); gen_op_stw_T0_T1(ctx); return; case 0xc200: /* mov.l R0,@(disp,GBR) */ gen_op_stc_gbr_T0(); gen_op_addl_imm_T0(B7_0 * 4); tcg_gen_mov_i32(cpu_T[0], cpu_T[1]); - gen_movl_rN_T(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); gen_op_stl_T0_T1(ctx); return; case 0x8000: /* mov.b R0,@(disp,Rn) */ - gen_movl_rN_T(cpu_T[0], REG(0)); - gen_movl_rN_T(cpu_T[1], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B7_4)]); gen_op_addl_imm_T1(B3_0); gen_op_stb_T0_T1(ctx); return; case 0x8100: /* mov.w R0,@(disp,Rn) */ - gen_movl_rN_T(cpu_T[0], REG(0)); - gen_movl_rN_T(cpu_T[1], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B7_4)]); gen_op_addl_imm_T1(B3_0 * 2); gen_op_stw_T0_T1(ctx); return; case 0x8400: /* mov.b @(disp,Rn),R0 */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_addl_imm_T0(B3_0); gen_op_ldb_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_gregs[REG(0)], cpu_T[0]); return; case 0x8500: /* mov.w @(disp,Rn),R0 */ - gen_movl_rN_T(cpu_T[0], REG(B7_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B7_4)]); gen_op_addl_imm_T0(B3_0 * 2); gen_op_ldw_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_gregs[REG(0)], cpu_T[0]); return; case 0xc700: /* mova @(disp,PC),R0 */ - gen_movl_imm_rN(((ctx->pc & 0xfffffffc) + 4 + B7_0 * 4) & ~3, - REG(0)); + tcg_gen_movi_i32(cpu_gregs[REG(0)], + ((ctx->pc & 0xfffffffc) + 4 + B7_0 * 4) & ~3); return; case 0xcb00: /* or #imm,R0 */ gen_op_or_imm_rN(B7_0, REG(0)); return; case 0xcf00: /* or.b #imm,@(R0,GBR) */ - gen_movl_rN_T(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); gen_op_addl_GBR_T0(); tcg_gen_mov_i32(cpu_T[0], cpu_T[1]); gen_op_ldub_T0_T0(ctx); @@ -961,7 +963,7 @@ void _decode_opc(DisasContext * ctx) gen_op_tst_imm_rN(B7_0, REG(0)); return; case 0xcc00: /* tst.b #imm,@(R0,GBR) */ - gen_movl_rN_T(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); gen_op_addl_GBR_T0(); gen_op_ldub_T0_T0(ctx); gen_op_tst_imm_T0(B7_0); @@ -970,7 +972,7 @@ void _decode_opc(DisasContext * ctx) gen_op_xor_imm_rN(B7_0, REG(0)); return; case 0xce00: /* xor.b #imm,@(R0,GBR) */ - gen_movl_rN_T(cpu_T[0], REG(0)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); gen_op_addl_GBR_T0(); tcg_gen_mov_i32(cpu_T[0], cpu_T[1]); gen_op_ldub_T0_T0(ctx); @@ -981,23 +983,23 @@ void _decode_opc(DisasContext * ctx) switch (ctx->opcode & 0xf08f) { case 0x408e: /* ldc Rm,Rn_BANK */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); - gen_movl_T_rN(cpu_T[0], ALTREG(B6_4)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); + tcg_gen_mov_i32(cpu_gregs[ALTREG(B6_4)], cpu_T[0]); return; case 0x4087: /* ldc.l @Rm+,Rn_BANK */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_ldl_T0_T0(ctx); - gen_movl_T_rN(cpu_T[0], ALTREG(B6_4)); + tcg_gen_mov_i32(cpu_gregs[ALTREG(B6_4)], cpu_T[0]); gen_op_inc4_rN(REG(B11_8)); return; case 0x0082: /* stc Rm_BANK,Rn */ - gen_movl_rN_T(cpu_T[0], ALTREG(B6_4)); - gen_movl_T_rN(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[ALTREG(B6_4)]); + tcg_gen_mov_i32(cpu_gregs[REG(B11_8)], cpu_T[0]); return; case 0x4083: /* stc.l Rm_BANK,@-Rn */ gen_op_dec4_rN(REG(B11_8)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); - gen_movl_rN_T(cpu_T[0], ALTREG(B6_4)); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[ALTREG(B6_4)]); gen_op_inc4_rN(REG(B11_8)); gen_op_stl_T0_T1(ctx); gen_op_dec4_rN(REG(B11_8)); @@ -1006,48 +1008,48 @@ void _decode_opc(DisasContext * ctx) switch (ctx->opcode & 0xf0ff) { case 0x0023: /* braf Rn */ - CHECK_NOT_DELAY_SLOT gen_movl_rN_T(cpu_T[0], REG(B11_8)); + CHECK_NOT_DELAY_SLOT tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_braf_T0(ctx->pc + 4); ctx->flags |= DELAY_SLOT; ctx->delayed_pc = (uint32_t) - 1; return; case 0x0003: /* bsrf Rn */ - CHECK_NOT_DELAY_SLOT gen_movl_rN_T(cpu_T[0], REG(B11_8)); + CHECK_NOT_DELAY_SLOT tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_bsrf_T0(ctx->pc + 4); ctx->flags |= DELAY_SLOT; ctx->delayed_pc = (uint32_t) - 1; return; case 0x4015: /* cmp/pl Rn */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_cmp_pl_T0(); return; case 0x4011: /* cmp/pz Rn */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_cmp_pz_T0(); return; case 0x4010: /* dt Rn */ gen_op_dt_rN(REG(B11_8)); return; case 0x402b: /* jmp @Rn */ - CHECK_NOT_DELAY_SLOT gen_movl_rN_T(cpu_T[0], REG(B11_8)); + CHECK_NOT_DELAY_SLOT tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_jmp_T0(); ctx->flags |= DELAY_SLOT; ctx->delayed_pc = (uint32_t) - 1; return; case 0x400b: /* jsr @Rn */ - CHECK_NOT_DELAY_SLOT gen_movl_rN_T(cpu_T[0], REG(B11_8)); + CHECK_NOT_DELAY_SLOT tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_jsr_T0(ctx->pc + 4); ctx->flags |= DELAY_SLOT; ctx->delayed_pc = (uint32_t) - 1; return; #define LDST(reg,ldnum,ldpnum,ldop,stnum,stpnum,stop,extrald) \ case ldnum: \ - gen_movl_rN_T (cpu_T[0], REG(B11_8)); \ + tcg_gen_mov_i32 (cpu_T[0], cpu_gregs[REG(B11_8)]); \ gen_op_##ldop##_T0_##reg (); \ extrald \ return; \ case ldpnum: \ - gen_movl_rN_T (cpu_T[0], REG(B11_8)); \ + tcg_gen_mov_i32 (cpu_T[0], cpu_gregs[REG(B11_8)]); \ gen_op_ldl_T0_T0 (ctx); \ gen_op_inc4_rN (REG(B11_8)); \ gen_op_##ldop##_T0_##reg (); \ @@ -1055,12 +1057,12 @@ void _decode_opc(DisasContext * ctx) return; \ case stnum: \ gen_op_##stop##_##reg##_T0 (); \ - gen_movl_T_rN (cpu_T[0], REG(B11_8)); \ + tcg_gen_mov_i32 (cpu_gregs[REG(B11_8)], cpu_T[0]); \ return; \ case stpnum: \ gen_op_##stop##_##reg##_T0 (); \ gen_op_dec4_rN (REG(B11_8)); \ - gen_movl_rN_T (cpu_T[1], REG(B11_8)); \ + tcg_gen_mov_i32 (cpu_T[1], cpu_gregs[REG(B11_8)]); \ gen_op_inc4_rN (REG(B11_8)); \ gen_op_stl_T0_T1 (ctx); \ gen_op_dec4_rN (REG(B11_8)); \ @@ -1079,23 +1081,23 @@ void _decode_opc(DisasContext * ctx) LDST(fpscr, 0x406a, 0x4066, lds, 0x006a, 0x4062, sts, ctx->bstate = BS_STOP;) case 0x00c3: /* movca.l R0,@Rm */ - gen_movl_rN_T(cpu_T[0], REG(0)); - gen_movl_rN_T(cpu_T[1], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(0)]); + tcg_gen_mov_i32(cpu_T[1], cpu_gregs[REG(B11_8)]); gen_op_stl_T0_T1(ctx); return; case 0x0029: /* movt Rn */ gen_op_movt_rN(REG(B11_8)); return; case 0x0093: /* ocbi @Rn */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_ldl_T0_T0(ctx); return; case 0x00a3: /* ocbp @Rn */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_ldl_T0_T0(ctx); return; case 0x00b3: /* ocbwb @Rn */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); gen_op_ldl_T0_T0(ctx); return; case 0x0083: /* pref @Rn */ @@ -1141,7 +1143,7 @@ void _decode_opc(DisasContext * ctx) gen_op_shlr16_Rn(REG(B11_8)); return; case 0x401b: /* tas.b @Rn */ - gen_movl_rN_T(cpu_T[0], REG(B11_8)); + tcg_gen_mov_i32(cpu_T[0], cpu_gregs[REG(B11_8)]); tcg_gen_mov_i32(cpu_T[0], cpu_T[1]); gen_op_ldub_T0_T0(ctx); gen_op_cmp_eq_imm_T0(0);