X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=tools%2Ftranslate.c;h=99d83e9e18936cbcd751a763e25b735747647abb;hb=f124ee13ba892739bb8164e6be9c8d3942a28744;hp=cca300f1759fde203fe33e48d798e9402aece34d;hpb=7a3c55553ae875cfc821e347867e08266a03f13c;p=ia32rtools.git diff --git a/tools/translate.c b/tools/translate.c index cca300f..99d83e9 100644 --- a/tools/translate.c +++ b/tools/translate.c @@ -18,6 +18,7 @@ #define _GNU_SOURCE #include #include +#include #include #include #include @@ -140,6 +141,7 @@ enum op_op { OP_FLDc, OP_FST, OP_FIST, + OP_FABS, OP_FADD, OP_FDIV, OP_FMUL, @@ -222,10 +224,10 @@ struct parsed_op { unsigned char pfo; unsigned char pfo_inv; unsigned char operand_cnt; - unsigned char p_argnum; // arg push: altered before call arg # + unsigned char p_argnum; // arg push: call's saved arg # unsigned char p_arggrp; // arg push: arg group # for above unsigned char p_argpass;// arg push: arg of host func - short p_argnext;// arg push: same arg pushed elsewhere or -1 + short pad; int regmask_src; // all referensed regs int regmask_dst; int pfomask; // flagop: parsed_flag_op that can't be delayed @@ -287,6 +289,7 @@ enum sct_func_attr { SCTFA_RM_REGS = (1 << 2), // don't emit regs (mask) SCTFA_NOWARN = (1 << 3), // don't try to detect problems SCTFA_ARGFRAME = (1 << 4), // copy all args to a struct, in order + SCTFA_UA_FLOAT = (1 << 5), // emit float i/o helpers for alignemnt }; enum x87_const { @@ -308,7 +311,6 @@ enum segment { SEG_GS, }; -// note: limited to 32k due to p_argnext #define MAX_OPS 4096 #define MAX_ARG_GRP 2 @@ -1112,6 +1114,7 @@ static const struct { { "fstp", OP_FST, 1, 1, OPF_FPOP }, { "fist", OP_FIST, 1, 1, OPF_FINT }, { "fistp", OP_FIST, 1, 1, OPF_FPOP|OPF_FINT }, + { "fabs", OP_FABS, 0, 0, 0 }, { "fadd", OP_FADD, 0, 2, 0 }, { "faddp", OP_FADD, 0, 2, OPF_FPOP }, { "fdiv", OP_FDIV, 0, 2, 0 }, @@ -1506,6 +1509,7 @@ static void parse_op(struct parsed_op *op, char words[16][256], int wordc) case OP_FISUB: case OP_FIDIVR: case OP_FISUBR: + case OP_FABS: case OP_FCHS: case OP_FCOS: case OP_FSIN: @@ -1935,6 +1939,9 @@ static int parse_stack_esp_offset(struct parsed_op *po, return 0; } +// returns g_func_pp arg number if arg is accessed +// -1 otherwise (stack vars, va_list) +// note: 'popr' must be from 'po', not some other op static int stack_frame_access(struct parsed_op *po, struct parsed_opr *popr, char *buf, size_t buf_size, const char *name, const char *cast, int is_src, int is_lea) @@ -1943,7 +1950,7 @@ static int stack_frame_access(struct parsed_op *po, const char *prefix = ""; const char *bp_arg = NULL; char ofs_reg[16] = { 0, }; - char argname[8]; + char argname[8], buf2[32]; int i, arg_i, arg_s; int unaligned = 0; int stack_ra = 0; @@ -1965,13 +1972,20 @@ static int stack_frame_access(struct parsed_op *po, arg_i = (offset - stack_ra - 4) / 4; if (arg_i < 0 || arg_i >= g_func_pp->argc_stack) { - if (g_func_pp->is_vararg - && arg_i == g_func_pp->argc_stack && is_lea) - { - // should be va_list - if (cast[0] == 0) - cast = "(u32)"; - snprintf(buf, buf_size, "%sap", cast); + if (g_func_pp->is_vararg && arg_i >= g_func_pp->argc_stack) { + // vararg access - messy and non-portable, + // but works with gcc on both x86 and ARM + if (arg_i == g_func_pp->argc_stack) + // should be va_list + snprintf(buf2, sizeof(buf2), "*(u32 *)&ap"); + else + snprintf(buf2, sizeof(buf2), "(*(u32 *)&ap + %u)", + (arg_i - g_func_pp->argc_stack) * 4); + + if (is_lea) + snprintf(buf, buf_size, "%s%s", cast, buf2); + else + snprintf(buf, buf_size, "%s*(u32 *)%s", cast, buf2); return -1; } ferr(po, "offset 0x%x (%s,%d) doesn't map to any arg\n", @@ -2397,6 +2411,20 @@ static char *out_src_opr_u32(char *buf, size_t buf_size, return out_src_opr(buf, buf_size, po, popr, NULL, 0); } +// do we need a helper func to perform a float i/o? +static int float_opr_needs_helper(struct parsed_op *po, + struct parsed_opr *popr) +{ + if (!(g_sct_func_attr & SCTFA_UA_FLOAT)) + return 0; + if (popr->type != OPT_REGMEM) + return 0; + if (is_stack_access(po, popr)) + return 0; + + return 1; +} + static char *out_opr_float(char *buf, size_t buf_size, struct parsed_op *po, struct parsed_opr *popr, int is_src, int need_float_stack) @@ -2450,7 +2478,10 @@ static char *out_opr_float(char *buf, size_t buf_size, break; } out_src_opr(tmp, sizeof(tmp), po, popr, "", 1); - snprintf(buf, buf_size, "*(%s *)(%s)", cast, tmp); + if (is_src && float_opr_needs_helper(po, popr)) + snprintf(buf, buf_size, "%s_load(%s)", cast, tmp); + else + snprintf(buf, buf_size, "*(%s *)(%s)", cast, tmp); break; case OPT_CONST: @@ -3630,8 +3661,8 @@ static const struct parsed_proto *try_recover_pp( } static void scan_for_call_type(int i, const struct parsed_opr *opr, - int magic, const struct parsed_proto **pp_found, int *pp_i, - int *multi) + int magic, int is_call_op, const struct parsed_proto **pp_found, + int *pp_i, int *multi) { const struct parsed_proto *pp = NULL; struct parsed_op *po; @@ -3644,7 +3675,8 @@ static void scan_for_call_type(int i, const struct parsed_opr *opr, lr = &g_label_refs[i]; for (; lr != NULL; lr = lr->next) { check_i(&ops[i], lr->i); - scan_for_call_type(lr->i, opr, magic, pp_found, pp_i, multi); + scan_for_call_type(lr->i, opr, magic, is_call_op, + pp_found, pp_i, multi); } if (i > 0 && LAST_OP(i - 1)) return; @@ -3689,29 +3721,30 @@ static void scan_for_call_type(int i, const struct parsed_opr *opr, if (i == g_func_pp->argc) return; pp = g_func_pp->arg[i].pp; - if (pp == NULL) - ferr(po, "icall: arg%d (%s) is not a fptr?\n", - i + 1, g_func_pp->arg[i].reg); + if (pp == NULL) { + if (is_call_op) + ferr(po, "icall: arg%d (%s) is not a fptr?\n", + i + 1, g_func_pp->arg[i].reg); + return; + } check_func_pp(po, pp, "icall reg-arg"); } else - pp = try_recover_pp(po, opr, 1, NULL); + pp = try_recover_pp(po, opr, is_call_op, NULL); if (*pp_found != NULL && pp != NULL && *pp_found != pp) { - if (!IS((*pp_found)->ret_type.name, pp->ret_type.name) - || (*pp_found)->is_stdcall != pp->is_stdcall - //|| (*pp_found)->is_fptr != pp->is_fptr - || (*pp_found)->argc != pp->argc - || (*pp_found)->argc_reg != pp->argc_reg - || (*pp_found)->argc_stack != pp->argc_stack) - { + if (pp_cmp_func(*pp_found, pp)) { + if (pp_i != NULL && *pp_i != -1) + fnote(&ops[*pp_i], "(other ref)\n"); ferr(po, "icall: parsed_proto mismatch\n"); } - *multi = 1; + if (multi != NULL) + *multi = 1; } if (pp != NULL) { *pp_found = pp; - *pp_i = po - ops; + if (pp_i != NULL) + *pp_i = po - ops; } } @@ -3915,7 +3948,8 @@ static void resolve_branches_parse_calls(int opcnt) po->operand_cnt = 0; po->regmask_src = pseudo_ops[l].regmask_src; po->regmask_dst = pseudo_ops[l].regmask_dst; - po->flags = pseudo_ops[l].flags; + po->flags &= OPF_TAIL; + po->flags |= pseudo_ops[l].flags; po->flags |= po->regmask_dst ? OPF_DATA : 0; break; } @@ -3976,7 +4010,8 @@ static void resolve_branches_parse_calls(int opcnt) if (po->bt_i != -1 || (po->flags & OPF_RMD)) continue; - if (po->operand[0].type == OPT_LABEL) + if (po->operand[0].type == OPT_LABEL + || po->operand[0].type == OPT_REG) // assume tail call goto tailcall; @@ -4254,20 +4289,72 @@ static void check_simple_sequence(int opcnt, int *fsz) *fsz += seq_len * 4; } +static int scan_prologue_ecx(int i, int opcnt, int flags_set, + int limit, int *ecx_push_out) +{ + const struct parsed_proto *pp; + int ecx_push = 0, other_push = 0; + int ret; + + while (limit > 0 && ops[i].op == OP_PUSH + && IS(opr_name(&ops[i], 0), "ecx")) + { + ops[i].flags |= flags_set; + ecx_push++; + i++; + limit--; + } + + ret = i; + if (ecx_push == 0 || flags_set != 0) + goto out; + + // check if some of the pushes aren't really call args + for (; i < opcnt; i++) { + if (i > 0 && g_labels[i] != NULL) + break; + if (ops[i].flags & (OPF_JMP|OPF_TAIL)) + break; + if (ops[i].op == OP_PUSH) + other_push++; + } + + if (ops[i].op != OP_CALL) + goto out; + + pp = ops[i].pp; + if (pp == NULL && ops[i].operand[0].type == OPT_LABEL) + pp = proto_parse(g_fhdr, opr_name(&ops[i], 0), 1); + if (pp == NULL) + goto out; + + ferr_assert(&ops[i], ecx_push + other_push >= pp->argc_stack); + if (other_push < pp->argc_stack) + ecx_push -= pp->argc_stack - other_push; + +out: + if (ecx_push_out != NULL) + *ecx_push_out = ecx_push; + return ret; +} + static int scan_prologue(int i, int opcnt, int *ecx_push, int *esp_sub) { const char *name; int j, len, ret; + int ecx_tmp = 0; for (; i < opcnt; i++) if (!(ops[i].flags & OPF_DONE)) break; - while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) { - ops[i].flags |= OPF_RMD | OPF_DONE | OPF_NOREGS; - g_stack_fsz += 4; - (*ecx_push)++; - i++; + ret = scan_prologue_ecx(i, opcnt, 0, 4, &ecx_tmp); + if (ecx_tmp > 0) { + scan_prologue_ecx(i, opcnt, OPF_RMD | OPF_DONE | OPF_NOREGS, + ecx_tmp, NULL); + g_stack_fsz += 4 * ecx_tmp; + *ecx_push += ecx_tmp; + i = ret; } for (; i < opcnt; i++) { @@ -4455,39 +4542,6 @@ static void scan_prologue_epilogue(int opcnt, int *stack_align) check_simple_sequence(opcnt, &push_fsz); i = scan_prologue(0, opcnt, &ecx_push, &esp_sub); - if (ecx_push && !esp_sub) { - // could actually be args for a call.. - for (; i < opcnt; i++) - if (ops[i].op != OP_PUSH) - break; - - if (ops[i].op == OP_CALL && ops[i].operand[0].type == OPT_LABEL) { - const struct parsed_proto *pp; - pp = proto_parse(g_fhdr, opr_name(&ops[i], 0), 1); - j = pp ? pp->argc_stack : 0; - while (i > 0 && j > 0) { - i--; - if (ops[i].op == OP_PUSH) { - ops[i].flags &= ~(OPF_RMD | OPF_DONE | OPF_NOREGS); - j--; - } - } - if (j != 0) - ferr(&ops[i], "unhandled prologue\n"); - - // recheck - i = ecx_push = 0; - g_stack_fsz = g_seh_size; - while (ops[i].op == OP_PUSH && IS(opr_name(&ops[i], 0), "ecx")) { - if (!(ops[i].flags & OPF_RMD)) - break; - g_stack_fsz += 4; - ecx_push++; - i++; - } - } - } - found = 0; if (ecx_push || esp_sub) { @@ -4653,6 +4707,17 @@ static int resolve_origin(int i, const struct parsed_opr *opr, } } +static int resolve_origin_reg(int i, int reg, int magic, int *op_i, + int *is_caller) +{ + struct parsed_opr opr = OPR_INIT(OPT_REG, OPLM_DWORD, reg); + + *op_i = -1; + if (is_caller != NULL) + *is_caller = 0; + return resolve_origin(i, &opr, magic, op_i, is_caller); +} + // find an instruction that previously referenced opr // if multiple results are found - fail // *op_i must be set to -1 by the caller @@ -4926,9 +4991,8 @@ static int resolve_used_bits(int i, int opcnt, int reg, } static const struct parsed_proto *resolve_deref(int i, int magic, - struct parsed_opr *opr, int level) + const struct parsed_opr *opr, int level) { - struct parsed_opr opr_s = OPR_INIT(OPT_REG, OPLM_DWORD, 0); const struct parsed_proto *pp = NULL; int from_caller = 0; char s_reg[4]; @@ -4950,8 +5014,7 @@ static const struct parsed_proto *resolve_deref(int i, int magic, if (reg < 0) return NULL; - opr_s.reg = reg; - ret = resolve_origin(i, &opr_s, i + magic, &j, NULL); + ret = resolve_origin_reg(i, reg, i + magic, &j, NULL); if (ret != 1) return NULL; @@ -4966,8 +5029,7 @@ static const struct parsed_proto *resolve_deref(int i, int magic, ops[j].operand[1].name); if (reg < 0) return NULL; - opr_s.reg = reg; - ret = resolve_origin(j, &opr_s, j + magic, &k, NULL); + ret = resolve_origin_reg(j, reg, j + magic, &k, NULL); if (ret != 1) return NULL; j = k; @@ -5017,32 +5079,34 @@ static const struct parsed_proto *resolve_deref(int i, int magic, return proto_lookup_struct(g_fhdr, pp->type.name, offset); } -static const struct parsed_proto *resolve_icall(int i, int opcnt, +static const struct parsed_proto *resolve_func_ptr(int i, int opcnt, + int is_call_op, const struct parsed_opr *opr, int *pp_i, int *multi_src) { const struct parsed_proto *pp = NULL; int search_advice = 0; - *multi_src = 0; - *pp_i = -1; + if (multi_src != NULL) + *multi_src = 0; + if (pp_i != NULL) + *pp_i = -1; - switch (ops[i].operand[0].type) { + switch (opr->type) { case OPT_REGMEM: // try to resolve struct member calls - pp = resolve_deref(i, i + opcnt * 19, &ops[i].operand[0], 0); + pp = resolve_deref(i, i + opcnt * 19, opr, 0); if (pp != NULL) break; // fallthrough case OPT_LABEL: case OPT_OFFSET: - pp = try_recover_pp(&ops[i], &ops[i].operand[0], - 1, &search_advice); + pp = try_recover_pp(&ops[i], opr, is_call_op, &search_advice); if (!search_advice) break; // fallthrough default: - scan_for_call_type(i, &ops[i].operand[0], i + opcnt * 9, &pp, - pp_i, multi_src); + scan_for_call_type(i, opr, i + opcnt * 9, is_call_op, + &pp, pp_i, multi_src); break; } @@ -5103,7 +5167,8 @@ static struct parsed_proto *process_call(int i, int opcnt) if (pp == NULL) { // indirect call - pp_c = resolve_icall(i, opcnt, &call_i, &multipath); + pp_c = resolve_func_ptr(i, opcnt, 1, &ops[i].operand[0], + &call_i, &multipath); if (pp_c != NULL) { if (!pp_c->is_func && !pp_c->is_fptr) ferr(po, "call to non-func: %s\n", pp_c->name); @@ -5211,13 +5276,107 @@ out: return pp; } +static void check_fptr_args(int i, int opcnt, struct parsed_proto *pp) +{ + struct parsed_opr s_opr = OPR_INIT(OPT_REG, OPLM_DWORD, 0); + const struct parsed_proto *pp_arg, *pp_cmp; + const struct parsed_op *po_a; + const char *s_reg; + int pp_cmp_i; + int arg, reg; + int bad = 0; + int j; + + for (arg = 0; arg < pp->argc; arg++) { + pp_cmp = NULL; + pp_cmp_i = -1; + + pp_arg = pp->arg[arg].pp; + if (pp_arg == NULL || !pp_arg->is_func) + continue; + + s_reg = pp->arg[arg].reg; + if (s_reg != NULL) { + reg = char_array_i(regs_r32, ARRAY_SIZE(regs_r32), s_reg); + ferr_assert(&ops[i], reg >= 0); + s_opr.reg = reg; + scan_for_call_type(i, &s_opr, i + arg + opcnt * 28, 0, + &pp_cmp, &pp_cmp_i, NULL); + if (pp_cmp != NULL && !pp_compatible_func(pp_arg, pp_cmp)) { + bad = 1; + if (pp_cmp_i >= 0) + fnote(&ops[pp_cmp_i], "(referenced here)\n"); + } + } + else { + for (j = 0; j < pp->arg[arg].push_ref_cnt; j++) { + po_a = pp->arg[arg].push_refs[j]; + if (po_a == NULL || po_a->op != OP_PUSH) + continue; + pp_cmp = resolve_func_ptr(po_a - ops, opcnt, 0, + &po_a->operand[0], &pp_cmp_i, NULL); + if (pp_cmp != NULL && !pp_compatible_func(pp_arg, pp_cmp)) { + bad = 1; + if (pp_cmp_i < 0) + pp_cmp_i = po_a - ops; + if (pp_cmp_i >= 0) + fnote(&ops[pp_cmp_i], "(referenced here)\n"); + } + } + } + + if (bad) + ferr(&ops[i], "incompatible fptr arg %d\n", arg + 1); + } +} + +static void pp_insert_reg_arg(struct parsed_proto *pp, const char *reg) +{ + int i; + + for (i = 0; i < pp->argc; i++) + if (pp->arg[i].reg == NULL) + break; + + if (pp->argc_stack) + memmove(&pp->arg[i + 1], &pp->arg[i], + sizeof(pp->arg[0]) * pp->argc_stack); + memset(&pp->arg[i], 0, sizeof(pp->arg[i])); + pp->arg[i].reg = strdup(reg); + pp->arg[i].type.name = strdup("int"); + pp->argc++; + pp->argc_reg++; +} + +static void pp_insert_stack_args(struct parsed_proto *pp, int count) +{ + int a; + + pp->argc += count; + pp->argc_stack += count; + + for (a = 0; a < pp->argc; a++) + if (pp->arg[a].type.name == NULL) + pp->arg[a].type.name = strdup("int"); +} + +static void pp_add_push_ref(struct parsed_proto *pp, + int arg, struct parsed_op *po) +{ + pp->arg[arg].push_refs = realloc(pp->arg[arg].push_refs, + (pp->arg[arg].push_ref_cnt + 1) + * sizeof(pp->arg[arg].push_refs[0])); + ferr_assert(po, pp->arg[arg].push_refs != NULL); + pp->arg[arg].push_refs[pp->arg[arg].push_ref_cnt++] = po; +} + static void mark_float_arg(struct parsed_op *po, struct parsed_proto *pp, int arg, int *regmask_ffca) { - po->p_argnext = -1; + ferr_assert(po, pp->arg[arg].push_ref_cnt == 0); + pp_add_push_ref(pp, arg, po); + po->p_argnum = arg + 1; - ferr_assert(po, pp->arg[arg].datap == NULL); - pp->arg[arg].datap = po; po->flags |= OPF_DONE | OPF_FARGNR | OPF_FARG; if (regmask_ffca != NULL) *regmask_ffca |= 1 << arg; @@ -5287,9 +5446,9 @@ static int collect_call_args_no_push(int i, struct parsed_proto *pp, for (arg = base_arg; arg < pp->argc; arg++) { ferr_assert(&ops[i], pp->arg[arg].reg == NULL); - po = pp->arg[arg].datap; - if (po == NULL) - ferr(&ops[i], "arg %d/%d not found\n", arg, pp->argc); + if (pp->arg[arg].push_ref_cnt != 1) + ferr(&ops[i], "arg %d/%d not found or bad\n", arg, pp->argc); + po = pp->arg[arg].push_refs[0]; if (po->operand[0].lmod == OPLM_QWORD) arg++; } @@ -5297,8 +5456,8 @@ static int collect_call_args_no_push(int i, struct parsed_proto *pp, return 0; } -static int collect_call_args_early(int i, struct parsed_proto *pp, - int *regmask, int *regmask_ffca) +static int collect_call_args_early(int i, int opcnt, + struct parsed_proto *pp, int *regmask, int *regmask_ffca) { struct parsed_op *po; int arg, ret; @@ -5369,8 +5528,7 @@ static int collect_call_args_early(int i, struct parsed_proto *pp, if (ops[j].op == OP_PUSH) { - ops[j].p_argnext = -1; - ferr_assert(&ops[j], pp->arg[arg].datap == NULL); + int ref_handled = 0; k = check_for_stp(j + 1, i); if (k != -1) { @@ -5381,15 +5539,18 @@ static int collect_call_args_early(int i, struct parsed_proto *pp, if (!pp->arg[arg].type.is_float) ferr(&ops[i], "arg %d should be float\n", arg + 1); mark_float_arg(&ops[k], pp, arg, regmask_ffca); + ref_handled = 1; } } - if (pp->arg[arg].datap == NULL) { - pp->arg[arg].datap = &ops[j]; - if (regmask != NULL && ops[j].operand[0].type == OPT_REG) - *regmask |= 1 << ops[j].operand[0].reg; + if (!ref_handled) { + ferr_assert(&ops[j], pp->arg[arg].push_ref_cnt == 0); + pp_add_push_ref(pp, arg, &ops[j]); } + if (regmask != NULL && ops[j].operand[0].type == OPT_REG) + *regmask |= 1 << ops[j].operand[0].reg; + ops[j].flags |= OPF_RMD | OPF_DONE | OPF_FARGNR | OPF_FARG; ops[j].flags &= ~OPF_RSAVE; @@ -5400,41 +5561,56 @@ static int collect_call_args_early(int i, struct parsed_proto *pp, } } + if (!g_header_mode) + check_fptr_args(i, opcnt, pp); + return 0; } -static int sync_argnum(struct parsed_op *po, int argnum) +// ensure all s_a* numbers match for a given func arg in all branches +// returns 1 if any changes were made, 0 if not +static int sync_argnum(struct parsed_proto *pp, int arg, + int *argnum, int *arggrp) { struct parsed_op *po_tmp; + int changed = 0; + int i; // see if other branches don't have higher argnum - for (po_tmp = po; po_tmp != NULL; ) { - if (argnum < po_tmp->p_argnum) - argnum = po_tmp->p_argnum; - // note: p_argnext is active on current collect_call_args only - po_tmp = po_tmp->p_argnext >= 0 ? &ops[po_tmp->p_argnext] : NULL; + for (i = 0; i < pp->arg[arg].push_ref_cnt; i++) { + po_tmp = pp->arg[arg].push_refs[i]; + if (*argnum < po_tmp->p_argnum) + *argnum = po_tmp->p_argnum; + if (*arggrp < po_tmp->p_arggrp) + *arggrp = po_tmp->p_arggrp; } // make all argnums consistent - for (po_tmp = po; po_tmp != NULL; ) { - if (po_tmp->p_argnum != 0) - po_tmp->p_argnum = argnum; - po_tmp = po_tmp->p_argnext >= 0 ? &ops[po_tmp->p_argnext] : NULL; + for (i = 0; i < pp->arg[arg].push_ref_cnt; i++) { + po_tmp = pp->arg[arg].push_refs[i]; + if (po_tmp->p_argnum == 0) + continue; + if (po_tmp->p_argnum != *argnum || po_tmp->p_arggrp != *arggrp) { + po_tmp->p_argnum = *argnum; + po_tmp->p_arggrp = *arggrp; + changed = 1; + } } - return argnum; + return changed; } static int collect_call_args_r(struct parsed_op *po, int i, - struct parsed_proto *pp, int *regmask, int *arg_grp, - int arg, int argnum, int magic, int need_op_saving, int may_reuse) + struct parsed_proto *pp, int *regmask, + int arg, int argnum, int magic, + int skip, int need_op_saving, int may_reuse) { struct parsed_proto *pp_tmp; - struct parsed_op *po_tmp; struct label_ref *lr; int need_to_save_current; int arg_grp_current = 0; int save_args_seen = 0; + int dummy = 0; int ret = 0; int reg; char buf[32]; @@ -5471,8 +5647,8 @@ static int collect_call_args_r(struct parsed_op *po, int i, check_i(&ops[j], lr->i); if ((ops[lr->i].flags & (OPF_JMP|OPF_CJMP)) != OPF_JMP) may_reuse = 1; - ret = collect_call_args_r(po, lr->i, pp, regmask, arg_grp, - arg, argnum, magic, need_op_saving, may_reuse); + ret = collect_call_args_r(po, lr->i, pp, regmask, + arg, argnum, magic, skip, need_op_saving, may_reuse); if (ret < 0) return ret; } @@ -5486,8 +5662,8 @@ static int collect_call_args_r(struct parsed_op *po, int i, continue; } need_op_saving = 1; - ret = collect_call_args_r(po, lr->i, pp, regmask, arg_grp, - arg, argnum, magic, need_op_saving, may_reuse); + ret = collect_call_args_r(po, lr->i, pp, regmask, + arg, argnum, magic, skip, need_op_saving, may_reuse); if (ret < 0) return ret; } @@ -5505,6 +5681,8 @@ static int collect_call_args_r(struct parsed_op *po, int i, if (may_reuse && pp_tmp->argc_stack > 0) ferr(po, "arg collect %d/%d hit '%s' with %d stack args\n", arg, pp->argc, opr_name(&ops[j], 0), pp_tmp->argc_stack); + if (!pp_tmp->is_unresolved) + skip = pp_tmp->argc_stack; } // esp adjust of 0 means we collected it before else if (ops[j].op == OP_ADD && ops[j].operand[0].reg == xSP @@ -5533,19 +5711,19 @@ static int collect_call_args_r(struct parsed_op *po, int i, may_reuse = 1; } + else if (ops[j].op == OP_PUSH && skip > 0) { + // XXX: might want to rm OPF_FARGNR and only use this + skip--; + } else if (ops[j].op == OP_PUSH && !(ops[j].flags & (OPF_FARGNR|OPF_DONE))) { if (pp->is_unresolved && (ops[j].flags & OPF_RMD)) break; - ops[j].p_argnext = -1; - po_tmp = pp->arg[arg].datap; - if (po_tmp != NULL) - ops[j].p_argnext = po_tmp - ops; - pp->arg[arg].datap = &ops[j]; + pp_add_push_ref(pp, arg, &ops[j]); - argnum = sync_argnum(&ops[j], argnum); + sync_argnum(pp, arg, &argnum, &dummy); need_to_save_current = 0; reg = -1; @@ -5627,6 +5805,7 @@ static int collect_call_args_r(struct parsed_op *po, int i, if (pp->arg[arg].is_saved) { ops[j].flags &= ~OPF_RMD; ops[j].p_argnum = argnum; + ops[j].p_arggrp = arg_grp_current; } // tracking reg usage @@ -5658,48 +5837,28 @@ static int collect_call_args_r(struct parsed_op *po, int i, return -1; } - if (arg_grp_current > *arg_grp) - *arg_grp = arg_grp_current; - return arg; } -static int collect_call_args(struct parsed_op *po, int i, +static int collect_call_args(struct parsed_op *po, int i, int opcnt, struct parsed_proto *pp, int *regmask, int magic) { - // arg group is for cases when pushes for - // multiple funcs are going on - struct parsed_op *po_tmp; - int arg_grp = 0; int ret; - int a; - ret = collect_call_args_r(po, i, pp, regmask, &arg_grp, - 0, 1, magic, 0, 0); + ret = collect_call_args_r(po, i, pp, regmask, 0, 1, magic, + 0, 0, 0); if (ret < 0) return ret; - if (pp->is_unresolved) { - pp->argc += ret; - pp->argc_stack += ret; - for (a = 0; a < pp->argc; a++) - if (pp->arg[a].type.name == NULL) - pp->arg[a].type.name = strdup("int"); - } + if (pp->is_unresolved) + pp_insert_stack_args(pp, ret); - if (arg_grp != 0) { - // propagate arg_grp - for (a = 0; a < pp->argc; a++) { - if (pp->arg[a].reg != NULL) - continue; + // note: p_argnum, p_arggrp will be propagated in a later pass, + // look for sync_argnum() (p_arggrp is for cases when mixed pushes + // for multiple funcs are going on) - po_tmp = pp->arg[a].datap; - while (po_tmp != NULL) { - po_tmp->p_arggrp = arg_grp; - po_tmp = po_tmp->p_argnext >= 0 ? &ops[po_tmp->p_argnext] : NULL; - } - } - } + if (!g_header_mode) + check_fptr_args(i, opcnt, pp); return ret; } @@ -5906,24 +6065,6 @@ static void reg_use_pass(int i, int opcnt, unsigned char *cbits, } } -static void pp_insert_reg_arg(struct parsed_proto *pp, const char *reg) -{ - int i; - - for (i = 0; i < pp->argc; i++) - if (pp->arg[i].reg == NULL) - break; - - if (pp->argc_stack) - memmove(&pp->arg[i + 1], &pp->arg[i], - sizeof(pp->arg[0]) * pp->argc_stack); - memset(&pp->arg[i], 0, sizeof(pp->arg[i])); - pp->arg[i].reg = strdup(reg); - pp->arg[i].type.name = strdup("int"); - pp->argc++; - pp->argc_reg++; -} - static void output_std_flag_z(FILE *fout, struct parsed_op *po, int *pfomask, const char *dst_opr_text) { @@ -6137,7 +6278,8 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) if (pp != NULL) { if (!(po->flags & OPF_ATAIL)) { // since we know the args, try to collect them - ret = collect_call_args_early(i, pp, ®mask, ®mask_ffca); + ret = collect_call_args_early(i, opcnt, pp, + ®mask, ®mask_ffca); if (ret != 0) pp = NULL; } @@ -6182,7 +6324,7 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) if (!pp->is_unresolved && !(po->flags & OPF_ATAIL)) { // since we know the args, collect them - collect_call_args(po, i, pp, ®mask, i + opcnt * 2); + collect_call_args(po, i, opcnt, pp, ®mask, i + opcnt * 2); } // for unresolved, collect after other passes } @@ -6342,19 +6484,24 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) if (pp->is_unresolved) { int regmask_stack = 0; - collect_call_args(po, i, pp, ®mask, i + opcnt * 2); - // this is pretty rough guess: - // see ecx and edx were pushed (and not their saved versions) - for (arg = 0; arg < pp->argc; arg++) { - if (pp->arg[arg].reg != NULL && !pp->arg[arg].is_saved) - continue; + if ((po->flags & OPF_TAIL) && g_func_pp->is_stdcall) + pp_insert_stack_args(pp, g_func_pp->argc_stack); + else { + collect_call_args(po, i, opcnt, pp, ®mask, i + opcnt * 2); + + // this is pretty rough guess: + // see ecx and edx were pushed (and not their saved versions) + for (arg = 0; arg < pp->argc; arg++) { + if (pp->arg[arg].reg != NULL && !pp->arg[arg].is_saved) + continue; - tmp_op = pp->arg[arg].datap; - if (tmp_op == NULL) - ferr(po, "parsed_op missing for arg%d\n", arg); - if (tmp_op->operand[0].type == OPT_REG) - regmask_stack |= 1 << tmp_op->operand[0].reg; + if (pp->arg[arg].push_ref_cnt == 0) + ferr(po, "parsed_op missing for arg%d\n", arg); + tmp_op = pp->arg[arg].push_refs[0]; + if (tmp_op->operand[0].type == OPT_REG) + regmask_stack |= 1 << tmp_op->operand[0].reg; + } } // quick dumb check for potential reg-args @@ -6497,7 +6644,41 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) } } - // pass8: final adjustments + // pass8: sync all push arg numbers + // some calls share args and not all of them + // (there's only partial intersection) + do { + int changed, argnum, arggrp; + + found = 0; + for (i = 0; i < opcnt; i++) + { + po = &ops[i]; + if ((po->flags & (OPF_RMD|OPF_DONE)) || po->op != OP_CALL) + continue; + + pp = po->pp; + arggrp = 0; + do { + changed = 0; + for (arg = argnum = 0; arg < pp->argc; arg++) { + if (pp->arg[arg].reg != NULL) + continue; + if (pp->arg[arg].is_saved) + changed |= sync_argnum(pp, arg, &argnum, &arggrp); + argnum++; + } + found |= changed; + } + while (changed); + + if (argnum > 32) + ferr(po, "too many args or looping in graph\n"); + } + } + while (found); + + // pass9: final adjustments for (i = 0; i < opcnt; i++) { po = &ops[i]; @@ -6861,7 +7042,10 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) || (tmp_op && (tmp_op->op == OP_AND || tmp_op->op == OP_OR)) )) { - out_src_opr_u32(buf3, sizeof(buf3), po, last_arith_dst); + struct parsed_op *po_arith = (void *)((char *)last_arith_dst + - offsetof(struct parsed_op, operand[0])); + ferr_assert(po, &ops[po_arith - ops] == po_arith); + out_src_opr_u32(buf3, sizeof(buf3), po_arith, last_arith_dst); out_test_for_cc(buf1, sizeof(buf1), po, po->pfo, po->pfo_inv, last_arith_dst->lmod, buf3); is_delayed = 1; @@ -7732,9 +7916,12 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) } // stack arg - tmp_op = pp->arg[arg].datap; - if (tmp_op == NULL) + if (pp->arg[arg].push_ref_cnt == 0) ferr(po, "parsed_op missing for arg%d\n", arg); + if (pp->arg[arg].push_ref_cnt > 1) + ferr_assert(po, pp->arg[arg].is_saved); + tmp_op = pp->arg[arg].push_refs[0]; + ferr_assert(po, tmp_op != NULL); if (tmp_op->flags & OPF_VAPUSH) { fprintf(fout, "ap"); @@ -7751,7 +7938,9 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) ferr_assert(po, cast[0] == 0); out_src_opr(buf1, sizeof(buf1), tmp_op, &tmp_op->operand[0], cast, 0); - tmp_op = pp->arg[++arg].datap; + arg++; + ferr_assert(po, pp->arg[arg].push_ref_cnt == 1); + tmp_op = pp->arg[arg].push_refs[0]; ferr_assert(po, tmp_op != NULL); out_src_opr(buf2, sizeof(buf2), tmp_op, &tmp_op->operand[0], cast, 0); @@ -7834,6 +8023,7 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) break; case OP_RET: + do_tail: if (g_func_pp->is_vararg) fprintf(fout, " va_end(ap);\n"); if (g_func_pp->has_retreg) { @@ -8000,19 +8190,27 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) break; case OP_FST: + dead_dst = 0; if (po->flags & OPF_FARG) { // store to stack as func arg - snprintf(buf1, sizeof(buf1), "fs_%d", po->p_argnum); - dead_dst = 0; + fprintf(fout, " fs_%d = %s;", po->p_argnum, float_st0); + } + else if (po->operand[0].type == OPT_REG + && po->operand[0].reg == xST0) + { + dead_dst = 1; + } + else if (float_opr_needs_helper(po, &po->operand[0])) { + out_src_opr(buf1, sizeof(buf1), po, &po->operand[0], "", 1); + fprintf(fout, " %s_store(%s, %s);", + po->operand[0].lmod == OPLM_QWORD ? "double" : "float", + float_st0, buf1); } else { out_dst_opr_float(buf1, sizeof(buf1), po, &po->operand[0], need_float_stack); - dead_dst = po->operand[0].type == OPT_REG - && po->operand[0].reg == xST0; - } - if (!dead_dst) fprintf(fout, " %s = %s;", buf1, float_st0); + } if (po->flags & OPF_FSHIFT) { if (need_float_stack) fprintf(fout, " f_stp++;"); @@ -8038,6 +8236,11 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) strcat(g_comment, " fist"); break; + case OP_FABS: + fprintf(fout, " %s = fabs%s(%s);", float_st0, + need_double ? "" : "f", float_st0); + break; + case OP_FADD: case OP_FDIV: case OP_FMUL: @@ -8249,7 +8452,7 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) fprintf(fout, " f_st0 = f_st1;"); } strcat(g_comment, " ftol"); - break; + goto tail_check; case OPP_CIPOW: if (need_float_stack) { @@ -8262,7 +8465,7 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) need_double ? "" : "f"); } strcat(g_comment, " CIpow"); - break; + goto tail_check; case OPP_ABORT: fprintf(fout, " do_skip_code_abort();"); @@ -8273,6 +8476,14 @@ static void gen_func(FILE *fout, FILE *fhdr, const char *funcn, int opcnt) fprintf(fout, " do_emms();"); break; + tail_check: + if (po->flags & OPF_TAIL) { + fprintf(fout, "\n"); + strcat(g_comment, " tail"); + goto do_tail; + } + break; + default: no_output = 1; ferr(po, "unhandled op type %d, flags %x\n", @@ -8569,6 +8780,9 @@ static void gen_hdr_dep_pass(int i, int opcnt, unsigned char *cbits, if (g_bp_frame && !(po->flags & OPF_EBP_S)) dep->regmask_live |= 1 << xBP; } + if ((po->flags & OPF_TAIL) && po->pp != NULL + && po->pp->is_stdcall) + fp->is_stdcall = 1; } else if (po->op == OP_RET) { if (po->operand_cnt > 0) { @@ -8586,10 +8800,9 @@ static void gen_hdr_dep_pass(int i, int opcnt, unsigned char *cbits, ret = 1; } else { - struct parsed_opr opr = OPR_INIT(OPT_REG, OPLM_DWORD, xAX); j = -1; from_caller = 0; - ret = resolve_origin(i, &opr, i + opcnt * 4, &j, &from_caller); + ret = resolve_origin_reg(i, xAX, i + opcnt * 4, &j, &from_caller); } if (ret != 1 && from_caller) { @@ -8740,7 +8953,7 @@ static void gen_hdr(const char *funcn, int opcnt) if (pp != NULL) { if (!(po->flags & OPF_ATAIL)) // since we know the args, try to collect them - if (collect_call_args_early(i, pp, NULL, NULL) != 0) + if (collect_call_args_early(i, opcnt, pp, NULL, NULL) != 0) pp = NULL; } @@ -8785,7 +8998,7 @@ static void gen_hdr(const char *funcn, int opcnt) if (!pp->is_unresolved && !(po->flags & OPF_ATAIL)) { // since we know the args, collect them - ret = collect_call_args(po, i, pp, ®mask_dummy, + ret = collect_call_args(po, i, opcnt, pp, ®mask_dummy, i + opcnt * 1); } if (!(po->flags & OPF_TAIL) @@ -8972,9 +9185,7 @@ static void output_hdr_fp(FILE *fout, const struct func_prototype *fp, fp->pp ? fp->pp->ret_type.name : fp->has_ret64 ? "__int64" : fp->has_ret ? "int" : "void"); - if (regmask_dep && (fp->is_stdcall || fp->argc_stack > 0) - && (regmask_dep & ~mxCX) == 0) - { + if (regmask_dep == mxCX && fp->is_stdcall && fp->argc_stack > 0) { fprintf(fout, "/*__thiscall*/ "); argc_normal++; regmask_dep = 0; @@ -9675,6 +9886,7 @@ int main(int argc, char *argv[]) "rm_regmask", "nowarn", "argframe", + "align_float", }; // parse manual attribute-list comment