X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;ds=sidebyside;f=ir%2Fbe%2Fia32%2Fia32_transform.c;h=263afec4a43adfa73b90d6b494d78fc01a7a9987;hb=8a16899bb7e9c4c04b39f202b71e47949c9bfd8a;hp=a3f10cab1908bacf065d34a8393aa19915936ef5;hpb=2b6d7097c5370a0c7b8e2477d9fe26342b878c8b;p=libfirm diff --git a/ir/be/ia32/ia32_transform.c b/ir/be/ia32/ia32_transform.c index a3f10cab1..263afec4a 100644 --- a/ir/be/ia32/ia32_transform.c +++ b/ir/be/ia32/ia32_transform.c @@ -187,7 +187,7 @@ static ir_node *get_symconst_base(void) { ir_graph *irg = current_ir_graph; - if (be_get_irg_options(irg)->pic) { + if (be_options.pic) { const arch_env_t *arch_env = be_get_irg_arch_env(irg); return arch_env->impl->get_pic_base(irg); } @@ -449,7 +449,7 @@ ir_entity *ia32_gen_fp_known_const(ia32_known_const_t kct) case 0: mode = mode_Iu; break; case 1: mode = mode_Lu; break; case 2: mode = mode_F; break; - default: panic("internal compiler error (ia32_gen_fp_known_const)"); + default: panic("internal compiler error"); } tv = new_tarval_from_str(cnst_str, strlen(cnst_str), mode); @@ -662,12 +662,6 @@ static int is_downconv(const ir_node *node) if (!is_Conv(node)) return 0; - /* we only want to skip the conv when we're the only user - * (because this test is used in the context of address-mode selection - * and we don't want to use address mode for multiple users) */ - if (get_irn_n_edges(node) > 1) - return 0; - src_mode = get_irn_mode(get_Conv_op(node)); dest_mode = get_irn_mode(node); return @@ -679,8 +673,15 @@ static int is_downconv(const ir_node *node) /** Skip all Down-Conv's on a given node and return the resulting node. */ ir_node *ia32_skip_downconv(ir_node *node) { - while (is_downconv(node)) + while (is_downconv(node)) { + /* we only want to skip the conv when we're the only user + * (because this test is used in the context of address-mode selection + * and we don't want to use address mode for multiple users) */ + if (get_irn_n_edges(node) > 1) + break; + node = get_Conv_op(node); + } return node; } @@ -1017,6 +1018,13 @@ static ir_node *skip_float_upconv(ir_node *node) return node; } +static void check_x87_floatmode(ir_mode *mode) +{ + if (mode != ia32_mode_E) { + panic("ia32: x87 only supports x86 extended float mode"); + } +} + /** * Construct a standard binary operation, set AM and immediate if required. * @@ -1038,6 +1046,9 @@ static ir_node *gen_binop_x87_float(ir_node *node, ir_node *op1, ir_node *op2, /* All operations are considered commutative, because there are reverse * variants */ match_flags_t flags = match_commutative | match_am; + ir_mode *mode + = is_Div(node) ? get_Div_resmode(node) : get_irn_mode(node); + check_x87_floatmode(mode); op1 = skip_float_upconv(op1); op2 = skip_float_upconv(op2); @@ -1907,6 +1918,7 @@ static ir_node *gen_Minus(ir_node *node) set_ia32_op_type(new_node, ia32_AddrModeS); set_ia32_ls_mode(new_node, mode); } else { + check_x87_floatmode(mode); new_node = new_bd_ia32_vfchs(dbgi, block, new_op); } } else { @@ -1963,6 +1975,7 @@ static ir_node *create_float_abs(dbg_info *dbgi, ir_node *block, ir_node *op, /* TODO, implement -Abs case */ assert(!negate); } else { + check_x87_floatmode(mode); new_node = new_bd_ia32_vfabs(dbgi, new_block, new_op); SET_IA32_ORIG_NODE(new_node, node); if (negate) { @@ -2693,14 +2706,6 @@ static ir_node *gen_general_Store(ir_node *node) addr.mem = be_transform_node(mem); if (mode_is_float(mode)) { - /* Convs (and strict-Convs) before stores are unnecessary if the mode - is the same. */ - while (is_Conv(val) && mode == get_irn_mode(val)) { - ir_node *op = get_Conv_op(val); - if (!mode_is_float(get_irn_mode(op))) - break; - val = op; - } new_val = be_transform_node(val); if (ia32_cg_config.use_sse2) { new_node = new_bd_ia32_xStore(dbgi, new_block, addr.base, @@ -2710,26 +2715,19 @@ static ir_node *gen_general_Store(ir_node *node) addr.index, addr.mem, new_val, mode); } } else if (!ia32_cg_config.use_sse2 && is_float_to_int_conv(val)) { - val = get_Conv_op(val); - - /* TODO: is this optimisation still necessary at all (middleend)? */ - /* We can skip ALL float->float up-Convs (and strict-up-Convs) before - * stores. */ - while (is_Conv(val)) { - ir_node *op = get_Conv_op(val); - if (!mode_is_float(get_irn_mode(op))) - break; - if (get_mode_size_bits(get_irn_mode(op)) > get_mode_size_bits(get_irn_mode(val))) - break; - val = op; - } + val = get_Conv_op(val); new_val = be_transform_node(val); new_node = gen_vfist(dbgi, new_block, addr.base, addr.index, addr.mem, new_val); } else { + unsigned dest_bits = get_mode_size_bits(mode); + while (is_downconv(val) + && get_mode_size_bits(get_irn_mode(val)) >= dest_bits) { + val = get_Conv_op(val); + } new_val = create_immediate_or_transform(val, 0); assert(mode != mode_b); - if (get_mode_size_bits(mode) == 8) { + if (dest_bits == 8) { new_node = new_bd_ia32_Store8Bit(dbgi, new_block, addr.base, addr.index, addr.mem, new_val); } else { @@ -2855,8 +2853,10 @@ static ir_node *create_Fucom(ir_node *node) ir_node *left = get_Cmp_left(node); ir_node *new_left = be_transform_node(left); ir_node *right = get_Cmp_right(node); + ir_mode *cmp_mode = get_irn_mode(left); ir_node *new_right; ir_node *new_node; + check_x87_floatmode(cmp_mode); if (ia32_cg_config.use_fucomi) { new_right = be_transform_node(right); @@ -2908,85 +2908,18 @@ static ir_node *create_Ucomi(ir_node *node) return new_node; } -/** - * returns true if it is assured, that the upper bits of a node are "clean" - * which means for a 16 or 8 bit value, that the upper bits in the register - * are 0 for unsigned and a copy of the last significant bit for signed - * numbers. - */ -static bool upper_bits_clean(ir_node *transformed_node, ir_mode *mode) +static bool ia32_mux_upper_bits_clean(const ir_node *node, ir_mode *mode) { - assert(ia32_mode_needs_gp_reg(mode)); - if (get_mode_size_bits(mode) >= 32) - return true; - - if (is_Proj(transformed_node)) - return upper_bits_clean(get_Proj_pred(transformed_node), mode); - - switch (get_ia32_irn_opcode(transformed_node)) { - case iro_ia32_Conv_I2I: - case iro_ia32_Conv_I2I8Bit: { - ir_mode *smaller_mode = get_ia32_ls_mode(transformed_node); - if (mode_is_signed(smaller_mode) != mode_is_signed(mode)) - return false; - if (get_mode_size_bits(smaller_mode) > get_mode_size_bits(mode)) - return false; - - return true; - } - - case iro_ia32_Shr: - if (mode_is_signed(mode)) { - return false; /* TODO handle signed modes */ - } else { - ir_node *right = get_irn_n(transformed_node, n_ia32_Shr_count); - if (is_ia32_Immediate(right) || is_ia32_Const(right)) { - const ia32_immediate_attr_t *attr - = get_ia32_immediate_attr_const(right); - if (attr->symconst == 0 && - (unsigned)attr->offset >= 32 - get_mode_size_bits(mode)) { - return true; - } - } - return upper_bits_clean(get_irn_n(transformed_node, n_ia32_Shr_val), mode); - } - - case iro_ia32_Sar: - /* TODO too conservative if shift amount is constant */ - return upper_bits_clean(get_irn_n(transformed_node, n_ia32_Sar_val), mode); - - case iro_ia32_And: - if (!mode_is_signed(mode)) { - return - upper_bits_clean(get_irn_n(transformed_node, n_ia32_And_right), mode) || - upper_bits_clean(get_irn_n(transformed_node, n_ia32_And_left), mode); - } - /* TODO if one is known to be zero extended, then || is sufficient */ - /* FALLTHROUGH */ - case iro_ia32_Or: - case iro_ia32_Xor: - return - upper_bits_clean(get_irn_n(transformed_node, n_ia32_binary_right), mode) && - upper_bits_clean(get_irn_n(transformed_node, n_ia32_binary_left), mode); - - case iro_ia32_Const: - case iro_ia32_Immediate: { - const ia32_immediate_attr_t *attr = - get_ia32_immediate_attr_const(transformed_node); - if (mode_is_signed(mode)) { - long shifted = attr->offset >> (get_mode_size_bits(mode) - 1); - return shifted == 0 || shifted == -1; - } else { - unsigned long shifted = (unsigned long)attr->offset; - shifted >>= get_mode_size_bits(mode)-1; - shifted >>= 1; - return shifted == 0; - } - } - - default: - return false; + ir_node *mux_true = get_Mux_true(node); + ir_node *mux_false = get_Mux_false(node); + ir_mode *mux_mode = get_irn_mode(node); + /* mux nodes which get transformed to the set instruction are not clean */ + if (is_Const(mux_true) && is_Const(mux_false) + && get_mode_size_bits(mux_mode) == 8) { + return false; } + return be_upper_bits_clean(mux_true, mode) + && be_upper_bits_clean(mux_false, mode); } /** @@ -3034,8 +2967,9 @@ static ir_node *gen_Cmp(ir_node *node) match_am_and_immediates | match_immediate); /* use 32bit compare mode if possible since the opcode is smaller */ - if (upper_bits_clean(am.new_op1, cmp_mode) && - upper_bits_clean(am.new_op2, cmp_mode)) { + if (am.op_type == ia32_Normal && + be_upper_bits_clean(and_left, cmp_mode) && + be_upper_bits_clean(and_right, cmp_mode)) { cmp_mode = mode_is_signed(cmp_mode) ? mode_Is : mode_Iu; } @@ -3056,8 +2990,9 @@ static ir_node *gen_Cmp(ir_node *node) match_16bit_am | match_am_and_immediates | match_immediate); /* use 32bit compare mode if possible since the opcode is smaller */ - if (upper_bits_clean(am.new_op1, cmp_mode) && - upper_bits_clean(am.new_op2, cmp_mode)) { + if (am.op_type == ia32_Normal && + be_upper_bits_clean(left, cmp_mode) && + be_upper_bits_clean(right, cmp_mode)) { cmp_mode = mode_is_signed(cmp_mode) ? mode_Is : mode_Iu; } @@ -3252,7 +3187,6 @@ enum setcc_transform_insn { SETCC_TR_NOT, SETCC_TR_AND, SETCC_TR_SET, - SETCC_TR_SBB, }; typedef struct setcc_transform { @@ -3588,9 +3522,6 @@ static ir_node *gen_Mux(ir_node *node) case SETCC_TR_SET: new_node = create_set_32bit(dbgi, new_block, flags, res.cc, node); break; - case SETCC_TR_SBB: - new_node = new_bd_ia32_Sbb0(dbgi, new_block, flags); - break; default: panic("unknown setcc transform"); } @@ -3654,9 +3585,9 @@ static ir_node *gen_x87_fp_to_gp(ir_node *node) } /** - * Creates a x87 strict Conv by placing a Store and a Load + * Creates a x87 Conv by placing a Store and a Load */ -static ir_node *gen_x87_strict_conv(ir_mode *tgt_mode, ir_node *node) +static ir_node *gen_x87_conv(ir_mode *tgt_mode, ir_node *node) { ir_node *block = get_nodes_block(node); ir_graph *irg = get_Block_irg(block); @@ -3737,7 +3668,7 @@ static ir_node *gen_x87_gp_to_fp(ir_node *node, ir_mode *src_mode) /* first convert to 32 bit signed if necessary */ if (get_mode_size_bits(src_mode) < 32) { - if (!upper_bits_clean(new_op, src_mode)) { + if (!be_upper_bits_clean(op, src_mode)) { new_op = create_Conv_I2I(dbgi, block, noreg_GP, noreg_GP, nomem, new_op, src_mode); SET_IA32_ORIG_NODE(new_op, node); } @@ -3800,16 +3731,11 @@ static ir_node *create_I2I_Conv(ir_mode *src_mode, ir_mode *tgt_mode, { ir_node *new_block = be_transform_node(block); ir_node *new_node; - ir_mode *smaller_mode; ia32_address_mode_t am; ia32_address_t *addr = &am.addr; (void) node; - if (get_mode_size_bits(src_mode) < get_mode_size_bits(tgt_mode)) { - smaller_mode = src_mode; - } else { - smaller_mode = tgt_mode; - } + assert(get_mode_size_bits(src_mode) < get_mode_size_bits(tgt_mode)); #ifdef DEBUG_libfirm if (is_Const(op)) { @@ -3818,25 +3744,19 @@ static ir_node *create_I2I_Conv(ir_mode *src_mode, ir_mode *tgt_mode, } #endif + if (be_upper_bits_clean(op, src_mode)) { + return be_transform_node(op); + } + match_arguments(&am, block, NULL, op, NULL, match_am | match_8bit_am | match_16bit_am); - if (upper_bits_clean(am.new_op2, smaller_mode)) { - /* unnecessary conv. in theory it shouldn't have been AM */ - assert(is_ia32_NoReg_GP(addr->base)); - assert(is_ia32_NoReg_GP(addr->index)); - assert(is_NoMem(addr->mem)); - assert(am.addr.offset == 0); - assert(am.addr.symconst_ent == NULL); - return am.new_op2; - } - new_node = create_Conv_I2I(dbgi, new_block, addr->base, addr->index, - addr->mem, am.new_op2, smaller_mode); + addr->mem, am.new_op2, src_mode); set_am_attributes(new_node, &am); /* match_arguments assume that out-mode = in-mode, this isn't true here * so fix it */ - set_ia32_ls_mode(new_node, smaller_mode); + set_ia32_ls_mode(new_node, src_mode); SET_IA32_ORIG_NODE(new_node, node); new_node = fix_mem_proj(new_node, &am); return new_node; @@ -3869,17 +3789,10 @@ static ir_node *gen_Conv(ir_node *node) } if (src_mode == tgt_mode) { - if (get_Conv_strict(node)) { - if (ia32_cg_config.use_sse2) { - /* when we are in SSE mode, we can kill all strict no-op conversion */ - return be_transform_node(op); - } - } else { - /* this should be optimized already, but who knows... */ - DEBUG_ONLY(ir_fprintf(stderr, "Debug warning: conv %+F is pointless\n", node);) + /* this should be optimized already, but who knows... */ + DEBUG_ONLY(ir_fprintf(stderr, "Debug warning: conv %+F is pointless\n", node);) DB((dbg, LEVEL_1, "killed Conv(mode, mode) ...")); - return be_transform_node(op); - } + return be_transform_node(op); } if (mode_is_float(src_mode)) { @@ -3893,21 +3806,14 @@ static ir_node *gen_Conv(ir_node *node) nomem, new_op); set_ia32_ls_mode(res, tgt_mode); } else { - if (get_Conv_strict(node)) { - /* if fp_no_float_fold is not set then we assume that we - * don't have any float operations in a non - * mode_float_arithmetic mode and can skip strict upconvs */ - if (src_bits < tgt_bits) { - DB((dbg, LEVEL_1, "killed Conv(float, float) ...")); - return new_op; - } else { - res = gen_x87_strict_conv(tgt_mode, new_op); - SET_IA32_ORIG_NODE(get_Proj_pred(res), node); - return res; - } + if (src_bits < tgt_bits) { + DB((dbg, LEVEL_1, "killed Conv(float, float) ...")); + return new_op; + } else { + res = gen_x87_conv(tgt_mode, new_op); + SET_IA32_ORIG_NODE(get_Proj_pred(res), node); + return res; } - DB((dbg, LEVEL_1, "killed Conv(float, float) ...")); - return new_op; } } else { /* ... to int */ @@ -3935,10 +3841,10 @@ static ir_node *gen_Conv(ir_node *node) unsigned float_mantissa = get_mode_mantissa_size(tgt_mode); res = gen_x87_gp_to_fp(node, src_mode); - /* we need a strict-Conv, if the int mode has more bits than the + /* we need a float-conv, if the int mode has more bits than the * float mantissa */ if (float_mantissa < int_mantissa) { - res = gen_x87_strict_conv(tgt_mode, res); + res = gen_x87_conv(tgt_mode, res); SET_IA32_ORIG_NODE(get_Proj_pred(res), node); } return res; @@ -3950,7 +3856,7 @@ static ir_node *gen_Conv(ir_node *node) return be_transform_node(op); } else { /* to int */ - if (src_bits == tgt_bits) { + if (src_bits >= tgt_bits) { DB((dbg, LEVEL_1, "omitting unnecessary Conv(%+F, %+F) ...", src_mode, tgt_mode)); return be_transform_node(op); @@ -4288,7 +4194,7 @@ static ir_node *gen_ia32_l_LLtoFloat(ir_node *node) ir_node *mem_high; if (ia32_cg_config.use_sse2) { - panic("ia32_l_LLtoFloat not implemented for SSE2"); + panic("not implemented for SSE2"); } /* do a store */ @@ -4617,7 +4523,7 @@ static ir_node *gen_Proj_Store(ir_node *node) if (pn == pn_Store_M) { return new_pred; } - panic("exception control flow for gen_float_const_Store not implemented yet"); + panic("exception control flow not implemented yet"); } else if (get_ia32_op_type(new_pred) == ia32_AddrModeD) { /* destination address mode */ if (pn == pn_Store_M) { @@ -4778,7 +4684,7 @@ static ir_node *gen_be_Call(ir_node *node) /* special case for PIC trampoline calls */ old_no_pic_adjust = ia32_no_pic_adjust; - ia32_no_pic_adjust = be_get_irg_options(current_ir_graph)->pic; + ia32_no_pic_adjust = be_options.pic; match_arguments(&am, src_block, NULL, src_ptr, src_mem, match_am | match_immediate); @@ -5134,14 +5040,14 @@ static ir_node *gen_parity(ir_node *node) * operations) */ ir_node *count = ia32_create_Immediate(NULL, 0, 16); - ir_node *shr = new_bd_ia32_Shr(dbgi, new_block, new_param, count); - ir_node *xor = new_bd_ia32_Xor(dbgi, new_block, noreg_GP, noreg_GP, nomem, - shr, new_param); - ir_node *xor2 = new_bd_ia32_XorHighLow(dbgi, new_block, xor); + ir_node *shr = new_bd_ia32_Shr(dbgi, new_block, new_param, count); + ir_node *xorn = new_bd_ia32_Xor(dbgi, new_block, noreg_GP, noreg_GP, nomem, + shr, new_param); + ir_node *xor2 = new_bd_ia32_XorHighLow(dbgi, new_block, xorn); ir_node *flags; - set_ia32_ls_mode(xor, mode_Iu); - set_ia32_commutative(xor); + set_ia32_ls_mode(xorn, mode_Iu); + set_ia32_commutative(xorn); set_irn_mode(xor2, mode_T); flags = new_r_Proj(xor2, mode_Iu, pn_ia32_XorHighLow_flags); @@ -5471,7 +5377,7 @@ static ir_node *gen_Builtin(ir_node *node) case ir_bk_inner_trampoline: return gen_inner_trampoline(node); } - panic("Builtin %s not implemented in IA32", get_builtin_kind_name(kind)); + panic("Builtin %s not implemented", get_builtin_kind_name(kind)); } /** @@ -5515,7 +5421,7 @@ static ir_node *gen_Proj_Builtin(ir_node *proj) return get_Tuple_pred(new_node, 0); } } - panic("Builtin %s not implemented in IA32", get_builtin_kind_name(kind)); + panic("Builtin %s not implemented", get_builtin_kind_name(kind)); } static ir_node *gen_be_IncSP(ir_node *node) @@ -5595,16 +5501,6 @@ static ir_node *gen_Proj_be_Call(ir_node *node) return res; } -/** - * Transform the Projs from a Cmp. - */ -static ir_node *gen_Proj_Cmp(ir_node *node) -{ - /* this probably means not all mode_b nodes were lowered... */ - panic("trying to directly transform Proj_Cmp %+F (mode_b not lowered?)", - node); -} - static ir_node *gen_Proj_ASM(ir_node *node) { ir_mode *mode = get_irn_mode(node); @@ -5654,8 +5550,6 @@ static ir_node *gen_Proj(ir_node *node) return gen_Proj_be_AddSP(node); case beo_Call: return gen_Proj_be_Call(node); - case iro_Cmp: - return gen_Proj_Cmp(node); case iro_Start: proj = get_Proj_proj(node); switch (proj) { @@ -5755,6 +5649,8 @@ static void register_transformers(void) be_set_transform_function(op_Switch, gen_Switch); be_set_transform_function(op_SymConst, gen_SymConst); be_set_transform_function(op_Unknown, ia32_gen_Unknown); + + be_set_upper_bits_clean_function(op_Mux, ia32_mux_upper_bits_clean); } /** @@ -5792,7 +5688,6 @@ static void postprocess_fp_call_results(void) for (j = get_method_n_ress(mtp) - 1; j >= 0; --j) { ir_type *res_tp = get_method_res_type(mtp, j); ir_node *res, *new_res; - const ir_edge_t *edge, *next; ir_mode *res_mode; if (! is_atomic_type(res_tp)) { @@ -5809,7 +5704,7 @@ static void postprocess_fp_call_results(void) new_res = NULL; /* now patch the users */ - foreach_out_edge_safe(res, edge, next) { + foreach_out_edge_safe(res, edge) { ir_node *succ = get_edge_src_irn(edge); /* ignore Keeps */