X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=cpu%2Fdrc%2Femit_x86.c;h=5362d9bd0e0b7a297b0bfc4f8b05afa7af050a81;hb=898d51a7fd1c090799ac88869fa0daf096fd702f;hp=225735921fe05f885a3d2734e9b903d56ec27e56;hpb=a2b8c5a54568093b247ced39f0754cbb30324830;p=picodrive.git diff --git a/cpu/drc/emit_x86.c b/cpu/drc/emit_x86.c index 2257359..5362d9b 100644 --- a/cpu/drc/emit_x86.c +++ b/cpu/drc/emit_x86.c @@ -1,4 +1,10 @@ /* + * Basic macros to emit x86 instructions and some utils + * Copyright (C) 2008,2009,2010 notaz + * + * This work is licensed under the terms of MAME license. + * See COPYING file in the top-level directory. + * * note: * temp registers must be eax-edx due to use of SETcc and r/w 8/16. * note about silly things like emith_eor_r_r_r: @@ -10,6 +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 @@ -47,15 +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) @@ -63,10 +70,13 @@ 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_OP_MODRM(op,mod,r,rm) { \ +#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); \ -} +} while (0) #define JMP8_POS(ptr) \ ptr = tcache_ptr; \ @@ -84,6 +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) @@ -112,34 +127,45 @@ 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_eor_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) { \ + emith_add_r_r(d, s1); \ + } else { \ + emith_move_r_r(d, s1); \ + emith_add_r_r(d, s2); \ + } \ +} while (0) + +#define emith_eor_r_r_r(d, s1, s2) do { \ if (d == s1) { \ emith_eor_r_r(d, s2); \ } else if (d == s2) { \ @@ -148,29 +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)) @@ -195,7 +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) \ @@ -214,20 +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) @@ -257,26 +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_and_r_r_imm(d, s, imm) { \ +#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) 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) @@ -303,10 +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)) @@ -332,13 +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); \ @@ -364,7 +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 */ @@ -376,20 +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 @@ -418,53 +465,53 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI }; #define is_abcdx(r) (xAX <= (r) && (r) <= xDX) -#define emith_read_op_8_16(op, r, rs, offs) do { \ +#define emith_read_r_r_offs(r, rs, offs) \ + emith_deref_op(0x8b, r, rs, offs) + +#define emith_write_r_r_offs(r, rs, offs) \ + emith_deref_op(0x89, r, rs, offs) + +// note: don't use prefixes on this +#define emith_read8_r_r_offs(r, rs, offs) do { \ int r_ = r; \ if (!is_abcdx(r)) \ r_ = rcache_get_tmp(); \ - emith_deref_op(op, r_, rs, offs); \ + emith_deref_op(0x8a, r_, rs, offs); \ if ((r) != r_) { \ emith_move_r_r(r, r_); \ rcache_free_tmp(r_); \ } \ } while (0) -#define emith_write_op_8_16(op, r, rs, offs) do { \ +#define emith_write8_r_r_offs(r, rs, offs) do {\ int r_ = r; \ if (!is_abcdx(r)) { \ r_ = rcache_get_tmp(); \ emith_move_r_r(r_, r); \ } \ - emith_deref_op(op, r_, rs, offs); \ + emith_deref_op(0x88, r_, rs, offs); \ if ((r) != r_) \ rcache_free_tmp(r_); \ } while (0) -#define emith_read_r_r_offs(r, rs, offs) \ - emith_deref_op(0x8b, r, rs, offs) - -#define emith_write_r_r_offs(r, rs, offs) \ - emith_deref_op(0x89, r, rs, offs) - -#define emith_read8_r_r_offs(r, rs, offs) \ - emith_read_op_8_16(0x8a, r, rs, offs) - -#define emith_write8_r_r_offs(r, rs, offs) \ - emith_write_op_8_16(0x88, r, rs, offs) - -#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_op_8_16(0x8b, r, rs, offs); \ -} + 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_read_op_8_16(0x89, r, rs, offs); \ -} + 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) @@ -487,7 +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); \ } @@ -495,30 +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); \ } @@ -529,10 +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) @@ -540,10 +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() @@ -572,6 +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 @@ -579,20 +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; \ @@ -600,7 +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); \ @@ -616,16 +706,33 @@ enum { xAX = 0, xCX, xDX, xBX, xSP, xBP, xSI, xDI }; emith_ret(); \ } -// assumes EBX is free temporary -#define emith_sh2_wcall(a, tab, ret_ptr) { \ +#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_ctx_read(arg2_, offsetof(SH2, is_slave)); \ - emith_push_imm((long)(ret_ptr)); \ - 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() { \ @@ -686,8 +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_); \ }