X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=ir%2Fbe%2Fia32%2Fia32_spec.pl;h=0249eddf416d5bf509fb45fe11d24807a7763b17;hb=eb593ca265a302849042866b57a3cf7a6e9495ab;hp=59d8692bc6821d7868d01959b3a579597bd18072;hpb=3c80f7a89f47a90db39bb261f573f34cb244556f;p=libfirm diff --git a/ir/be/ia32/ia32_spec.pl b/ir/be/ia32/ia32_spec.pl index 59d8692bc..0249eddf4 100644 --- a/ir/be/ia32/ia32_spec.pl +++ b/ir/be/ia32/ia32_spec.pl @@ -2,7 +2,10 @@ # $Id$ # This is the specification for the ia32 assembler Firm-operations +use File::Basename; + $new_emit_syntax = 1; +my $myname = $0; # the cpu architecture (ia32, ia64, mips, sparc, ppc, ...) $arch = "ia32"; @@ -13,23 +16,28 @@ $arch = "ia32"; # %nodes = ( # # => { -# "op_flags" => "N|L|C|X|I|F|Y|H|c|K", -# "irn_flags" => "R|N|I|S" -# "arity" => "0|1|2|3 ... |variable|dynamic|any", -# "state" => "floats|pinned|mem_pinned|exc_pinned", -# "args" => [ -# { "type" => "type 1", "name" => "name 1" }, -# { "type" => "type 2", "name" => "name 2" }, +# op_flags => "N|L|C|X|I|F|Y|H|c|K", +# irn_flags => "R|N|I|S" +# arity => "0|1|2|3 ... |variable|dynamic|any", +# state => "floats|pinned|mem_pinned|exc_pinned", +# args => [ +# { type => "type 1", name => "name 1" }, +# { type => "type 2", name => "name 2" }, # ... # ], -# "comment" => "any comment for constructor", -# "reg_req" => { "in" => [ "reg_class|register" ], "out" => [ "reg_class|register|in_rX" ] }, -# "cmp_attr" => "c source code for comparing node attributes", -# "emit" => "emit code with templates", -# "attr" => "attitional attribute arguments for constructor" -# "init_attr" => "emit attribute initialization template" -# "rd_constructor" => "c source code which constructs an ir_node" -# "latency" => "latency of this operation (can be float)" +# comment => "any comment for constructor", +# reg_req => { in => [ "reg_class|register" ], out => [ "reg_class|register|in_rX" ] }, +# cmp_attr => "c source code for comparing node attributes", +# outs => { "out1", "out2" } # optional, creates pn_op_out1, ... consts +# ins => { "in1", "in2" } # optional, creates n_op_in1, ... consts +# mode => "mode_Iu" # optional, predefines the mode +# emit => "emit code with templates", +# attr => "attitional attribute arguments for constructor", +# init_attr => "emit attribute initialization template", +# rd_constructor => "c source code which constructs an ir_node", +# hash_func => "name of the hash function for this operation", +# latency => "latency of this operation (can be float)" +# attr_type => "name of the attribute struct", # }, # # ... # (all nodes you need to describe) @@ -48,6 +56,8 @@ $arch = "ia32"; # H irop_flag_highlevel # c irop_flag_constlike # K irop_flag_keep +# NB irop_flag_dump_noblock +# NI irop_flag_dump_noinput # # irn_flags: special node flags, OPTIONAL (default is 0) # following irn_flags are supported: @@ -92,8 +102,6 @@ $arch = "ia32"; # # NOTE: rd_constructor and args are only optional if and only if arity is 0,1,2 or 3 # -# latency: the latency of the operation, default is 1 -# # register types: # 0 - no special type @@ -102,110 +110,121 @@ $arch = "ia32"; # 4 - ignore (do not assign this register) # 8 - emitter can choose an arbitrary register of this class # 16 - the register is a virtual one +# 32 - register represents a state # NOTE: Last entry of each class is the largest Firm-Mode a register can hold %reg_classes = ( - "gp" => [ - { "name" => "eax", "type" => 1 }, - { "name" => "edx", "type" => 1 }, - { "name" => "ebx", "type" => 2 }, - { "name" => "ecx", "type" => 1 }, - { "name" => "esi", "type" => 2 }, - { "name" => "edi", "type" => 2 }, - { "name" => "ebp", "type" => 2 }, - { "name" => "esp", "type" => 4 }, - { "name" => "gp_NOREG", "type" => 4 | 8 | 16 }, # we need a dummy register for NoReg nodes - { "name" => "gp_UKNWN", "type" => 4 | 8 | 16 }, # we need a dummy register for Unknown nodes - { "mode" => "mode_Iu" } + gp => [ + { name => "edx", type => 1 }, + { name => "ecx", type => 1 }, + { name => "eax", type => 1 }, + { name => "ebx", type => 2 }, + { name => "esi", type => 2 }, + { name => "edi", type => 2 }, + { name => "ebp", type => 2 }, + { name => "esp", type => 4 }, + { name => "gp_NOREG", type => 4 | 8 | 16 }, # we need a dummy register for NoReg nodes + { name => "gp_UKNWN", type => 4 | 8 | 16 }, # we need a dummy register for Unknown nodes + { mode => "mode_Iu" } + ], + mmx => [ + { name => "mm0", type => 4 }, + { name => "mm1", type => 4 }, + { name => "mm2", type => 4 }, + { name => "mm3", type => 4 }, + { name => "mm4", type => 4 }, + { name => "mm5", type => 4 }, + { name => "mm6", type => 4 }, + { name => "mm7", type => 4 }, + { mode => "mode_E", flags => "manual_ra" } ], - "xmm" => [ - { "name" => "xmm0", "type" => 1 }, - { "name" => "xmm1", "type" => 1 }, - { "name" => "xmm2", "type" => 1 }, - { "name" => "xmm3", "type" => 1 }, - { "name" => "xmm4", "type" => 1 }, - { "name" => "xmm5", "type" => 1 }, - { "name" => "xmm6", "type" => 1 }, - { "name" => "xmm7", "type" => 1 }, - { "name" => "xmm_NOREG", "type" => 4 | 8 | 16 }, # we need a dummy register for NoReg nodes - { "name" => "xmm_UKNWN", "type" => 4 | 8 | 16 }, # we need a dummy register for Unknown nodes - { "mode" => "mode_E" } + xmm => [ + { name => "xmm0", type => 1 }, + { name => "xmm1", type => 1 }, + { name => "xmm2", type => 1 }, + { name => "xmm3", type => 1 }, + { name => "xmm4", type => 1 }, + { name => "xmm5", type => 1 }, + { name => "xmm6", type => 1 }, + { name => "xmm7", type => 1 }, + { name => "xmm_NOREG", type => 4 | 16 }, # we need a dummy register for NoReg nodes + { name => "xmm_UKNWN", type => 4 | 8 | 16}, # we need a dummy register for Unknown nodes + { mode => "mode_E" } ], - "vfp" => [ - { "name" => "vf0", "type" => 1 | 16 }, - { "name" => "vf1", "type" => 1 | 16 }, - { "name" => "vf2", "type" => 1 | 16 }, - { "name" => "vf3", "type" => 1 | 16 }, - { "name" => "vf4", "type" => 1 | 16 }, - { "name" => "vf5", "type" => 1 | 16 }, - { "name" => "vf6", "type" => 1 | 16 }, - { "name" => "vf7", "type" => 1 | 16 }, - { "name" => "vfp_NOREG", "type" => 4 | 8 | 16 }, # we need a dummy register for NoReg nodes - { "name" => "vfp_UKNWN", "type" => 4 | 8 | 16 }, # we need a dummy register for Unknown nodes - { "mode" => "mode_E" } + vfp => [ + { name => "vf0", type => 1 }, + { name => "vf1", type => 1 }, + { name => "vf2", type => 1 }, + { name => "vf3", type => 1 }, + { name => "vf4", type => 1 }, + { name => "vf5", type => 1 }, + { name => "vf6", type => 1 }, + { name => "vf7", type => 1 }, + { name => "vfp_NOREG", type => 4 | 8 | 16 }, # we need a dummy register for NoReg nodes + { name => "vfp_UKNWN", type => 4 | 8 | 16 }, # we need a dummy register for Unknown nodes + { mode => "mode_E" } ], - "st" => [ - { "name" => "st0", "type" => 1 }, - { "name" => "st1", "type" => 1 }, - { "name" => "st2", "type" => 1 }, - { "name" => "st3", "type" => 1 }, - { "name" => "st4", "type" => 1 }, - { "name" => "st5", "type" => 1 }, - { "name" => "st6", "type" => 1 }, - { "name" => "st7", "type" => 1 }, - { "mode" => "mode_E" } + st => [ + { name => "st0", realname => "st", type => 4 }, + { name => "st1", realname => "st(1)", type => 4 }, + { name => "st2", realname => "st(2)", type => 4 }, + { name => "st3", realname => "st(3)", type => 4 }, + { name => "st4", realname => "st(4)", type => 4 }, + { name => "st5", realname => "st(5)", type => 4 }, + { name => "st6", realname => "st(6)", type => 4 }, + { name => "st7", realname => "st(7)", type => 4 }, + { mode => "mode_E", flags => "manual_ra" } ], - "fp_cw" => [ # the floating point control word - { "name" => "fpcw", "type" => 0 }, - { "mode" => "mode_Hu" }, + fp_cw => [ # the floating point control word + { name => "fpcw", type => 4|32 }, + { mode => "mode_fpcw", flags => "manual_ra|state" } + ], + flags => [ + { name => "eflags", type => 0 }, + { mode => "mode_Iu", flags => "manual_ra" } ], ); # %reg_classes %cpu = ( - "GP" => [ 1, "GP_EAX", "GP_EBX", "GP_ECX", "GP_EDX", "GP_ESI", "GP_EDI", "GP_EBP" ], - "SSE" => [ 1, "SSE_XMM0", "SSE_XMM1", "SSE_XMM2", "SSE_XMM3", "SSE_XMM4", "SSE_XMM5", "SSE_XMM6", "SSE_XMM7" ], - "VFP" => [ 1, "VFP_VF0", "VFP_VF1", "VFP_VF2", "VFP_VF3", "VFP_VF4", "VFP_VF5", "VFP_VF6", "VFP_VF7" ], - "BRANCH" => [ 1, "BRANCH1", "BRANCH2" ], + GP => [ 1, "GP_EAX", "GP_EBX", "GP_ECX", "GP_EDX", "GP_ESI", "GP_EDI", "GP_EBP" ], + SSE => [ 1, "SSE_XMM0", "SSE_XMM1", "SSE_XMM2", "SSE_XMM3", "SSE_XMM4", "SSE_XMM5", "SSE_XMM6", "SSE_XMM7" ], + VFP => [ 1, "VFP_VF0", "VFP_VF1", "VFP_VF2", "VFP_VF3", "VFP_VF4", "VFP_VF5", "VFP_VF6", "VFP_VF7" ], + BRANCH => [ 1, "BRANCH1", "BRANCH2" ], ); # %cpu %vliw = ( - "bundle_size" => 1, - "bundels_per_cycle" => 1 + bundle_size => 1, + bundels_per_cycle => 1 ); # vliw %emit_templates = ( - "S1" => "${arch}_emit_source_register(env, node, 0);", - "S2" => "${arch}_emit_source_register(env, node, 1);", - "S3" => "${arch}_emit_source_register(env, node, 2);", - "S4" => "${arch}_emit_source_register(env, node, 3);", - "S5" => "${arch}_emit_source_register(env, node, 4);", - "S6" => "${arch}_emit_source_register(env, node, 5);", - "D1" => "${arch}_emit_dest_register(env, node, 0);", - "D2" => "${arch}_emit_dest_register(env, node, 1);", - "D3" => "${arch}_emit_dest_register(env, node, 2);", - "D4" => "${arch}_emit_dest_register(env, node, 3);", - "D5" => "${arch}_emit_dest_register(env, node, 4);", - "D6" => "${arch}_emit_dest_register(env, node, 5);", - "A1" => "${arch}_emit_in_node_name(env, node, 0);", - "A2" => "${arch}_emit_in_node_name(env, node, 1);", - "A3" => "${arch}_emit_in_node_name(env, node, 2);", - "A4" => "${arch}_emit_in_node_name(env, node, 3);", - "A5" => "${arch}_emit_in_node_name(env, node, 4);", - "A6" => "${arch}_emit_in_node_name(env, node, 5);", - "X1" => "${arch}_emit_x87_name(env, node, 0);", - "X2" => "${arch}_emit_x87_name(env, node, 1);", - "X3" => "${arch}_emit_x87_name(env, node, 2);", - "C" => "${arch}_emit_immediate(env, node);", - "SE" => "${arch}_emit_extend_suffix(env, get_ia32_ls_mode(node));", - "ME" => "if(get_mode_size_bits(get_ia32_ls_mode(node)) != 32)\n - ${arch}_emit_mode_suffix(env, get_ia32_ls_mode(node));", - "M" => "${arch}_emit_mode_suffix(env, get_ia32_ls_mode(node));", - "XM" => "${arch}_emit_x87_mode_suffix(env, node);", - "XXM" => "${arch}_emit_xmm_mode_suffix(env, node);", - "XSD" => "${arch}_emit_xmm_mode_suffix_s(env, node);", - "AM" => "${arch}_emit_am(env, node);", - "unop" => "${arch}_emit_unop(env, node);", - "binop" => "${arch}_emit_binop(env, node);", - "x87_binop" => "${arch}_emit_x87_binop(env, node);", + S0 => "${arch}_emit_source_register(node, 0);", + S1 => "${arch}_emit_source_register(node, 1);", + S2 => "${arch}_emit_source_register(node, 2);", + S3 => "${arch}_emit_source_register(node, 3);", + SB1 => "${arch}_emit_8bit_source_register_or_immediate(node, 1);", + SB2 => "${arch}_emit_8bit_source_register_or_immediate(node, 2);", + SB3 => "${arch}_emit_8bit_source_register_or_immediate(node, 3);", + SI1 => "${arch}_emit_source_register_or_immediate(node, 1);", + SI3 => "${arch}_emit_source_register_or_immediate(node, 3);", + D0 => "${arch}_emit_dest_register(node, 0);", + D1 => "${arch}_emit_dest_register(node, 1);", + DB0 => "${arch}_emit_8bit_dest_register(node, 0);", + X0 => "${arch}_emit_x87_register(node, 0);", + X1 => "${arch}_emit_x87_register(node, 1);", + SE => "${arch}_emit_extend_suffix(get_ia32_ls_mode(node));", + ME => "if(get_mode_size_bits(get_ia32_ls_mode(node)) != 32)\n + ia32_emit_mode_suffix(node);", + M => "${arch}_emit_mode_suffix(node);", + XM => "${arch}_emit_x87_mode_suffix(node);", + XXM => "${arch}_emit_xmm_mode_suffix(node);", + XSD => "${arch}_emit_xmm_mode_suffix_s(node);", + AM => "${arch}_emit_am(node);", + unop3 => "${arch}_emit_unop(node, n_ia32_unary_op);", + unop4 => "${arch}_emit_unop(node, n_ia32_binary_right);", + binop => "${arch}_emit_binop(node);", + x87_binop => "${arch}_emit_x87_binop(node);", + CMP0 => "${arch}_emit_cmp_suffix_node(node, 0);", + CMP3 => "${arch}_emit_cmp_suffix_node(node, 3);", ); #--------------------------------------------------# @@ -219,13 +238,120 @@ $arch = "ia32"; # |_| # #--------------------------------------------------# -$default_cmp_attr = "return ia32_compare_attr(attr_a, attr_b);"; +$default_op_attr_type = "ia32_op_attr_t"; +$default_attr_type = "ia32_attr_t"; +$default_copy_attr = "ia32_copy_attr"; + +sub ia32_custom_init_attr { + my $node = shift; + my $name = shift; + my $res = ""; + + if(defined($node->{modified_flags})) { + $res .= "\tset_ia32_flags(res, get_ia32_flags(res) | arch_irn_flags_modify_flags);\n"; + } + if(defined($node->{am})) { + my $am = $node->{am}; + if($am eq "source,unary") { + $res .= "\tset_ia32_am_support(res, ia32_am_Source, ia32_am_unary);"; + } elsif($am eq "source,binary") { + $res .= "\tset_ia32_am_support(res, ia32_am_Source, ia32_am_binary);"; + } elsif($am eq "none") { + # nothing to do + } else { + die("Invalid address mode '$am' specified on op $name"); + } + if($am ne "none") { + if($node->{state} ne "exc_pinned" + and $node->{state} ne "pinned") { + die("AM nodes must have pinned or AM pinned state ($name)"); + } + } + } + return $res; +} +$custom_init_attr_func = \&ia32_custom_init_attr; + +%init_attr = ( + ia32_asm_attr_t => + "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". + "\tinit_ia32_x87_attributes(res);". + "\tinit_ia32_asm_attributes(res);", + ia32_attr_t => + "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);", + ia32_condcode_attr_t => + "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". + "\tinit_ia32_condcode_attributes(res, pnc);", + ia32_copyb_attr_t => + "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". + "\tinit_ia32_copyb_attributes(res, size);", + ia32_immediate_attr_t => + "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". + "\tinit_ia32_immediate_attributes(res, symconst, symconst_sign, offset);", + ia32_x87_attr_t => + "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". + "\tinit_ia32_x87_attributes(res);", +); + +%compare_attr = ( + ia32_asm_attr_t => "ia32_compare_asm_attr", + ia32_attr_t => "ia32_compare_nodes_attr", + ia32_condcode_attr_t => "ia32_compare_condcode_attr", + ia32_copyb_attr_t => "ia32_compare_copyb_attr", + ia32_immediate_attr_t => "ia32_compare_immediate_attr", + ia32_x87_attr_t => "ia32_compare_x87_attr", +); %operands = ( ); +$mode_xmm = "mode_E"; +$mode_gp = "mode_Iu"; +$mode_flags = "mode_Iu"; +$mode_fpcw = "mode_fpcw"; +$status_flags = [ "CF", "PF", "AF", "ZF", "SF", "OF" ]; +$status_flags_wo_cf = [ "PF", "AF", "ZF", "SF", "OF" ]; +$fpcw_flags = [ "FP_IM", "FP_DM", "FP_ZM", "FP_OM", "FP_UM", "FP_PM", + "FP_PC0", "FP_PC1", "FP_RC0", "FP_RC1", "FP_X" ]; + %nodes = ( +Immediate => { + state => "pinned", + op_flags => "c", + irn_flags => "I", + reg_req => { out => [ "gp_NOREG" ] }, + attr => "ir_entity *symconst, int symconst_sign, long offset", + attr_type => "ia32_immediate_attr_t", + hash_func => "ia32_hash_Immediate", + latency => 0, + mode => $mode_gp, +}, + +Asm => { + mode => "mode_T", + arity => "variable", + out_arity => "variable", + attr_type => "ia32_asm_attr_t", + attr => "ident *asm_text, const ia32_asm_reg_t *register_map", + init_attr => "attr->asm_text = asm_text;\n". + "\tattr->register_map = register_map;\n", + latency => 10, + modified_flags => 1, +}, + +# "allocates" a free register +ProduceVal => { + op_flags => "c", + irn_flags => "R", + reg_req => { out => [ "gp" ] }, + emit => "", + units => [ ], + latency => 0, + mode => $mode_gp, + cmp_attr => "return 1;", +}, + #-----------------------------------------------------------------# # _ _ _ # # (_) | | | | # @@ -239,692 +365,1101 @@ $default_cmp_attr = "return ia32_compare_attr(attr_a, attr_b);"; # commutative operations -# NOTE: -# All nodes supporting Addressmode have 5 INs: -# 1 - base r1 == NoReg in case of no AM or no base -# 2 - index r2 == NoReg in case of no AM or no index -# 3 - op1 r3 == always present -# 4 - op2 r4 == NoReg in case of immediate operation -# 5 - mem NoMem in case of no AM otherwise it takes the mem from the Load - -"Add" => { - "irn_flags" => "R", - "comment" => "construct Add: Add(a, b) = Add(b, a) = a + b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. addl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Adc" => { - "comment" => "construct Add with Carry: Adc(a, b) = Add(b, a) = a + b + carry", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. adcl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Add64Bit" => { - "irn_flags" => "R", - "comment" => "construct 64Bit Add: Add(a_l, a_h, b_l, b_h) = a_l + b_l; a_h + b_h + carry", - "arity" => 4, - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp" ], "out" => [ "!in", "!in" ] }, - "emit" => ' -. movl %S1, %D1 -. movl %S2, %D2 -. addl %S3, %D1 -. adcl %S4, %D2 -', - "outs" => [ "low_res", "high_res" ], - "units" => [ "GP" ], -}, - -"l_Add" => { - "op_flags" => "C", - "irn_flags" => "R", - "cmp_attr" => "return 1;", - "comment" => "construct lowered Add: Add(a, b) = Add(b, a) = a + b", - "arity" => 2, -}, - -"l_Adc" => { - "op_flags" => "C", - "cmp_attr" => "return 1;", - "comment" => "construct lowered Add with Carry: Adc(a, b) = Adc(b, a) = a + b + carry", - "arity" => 2, -}, - -"Mul" => { - # we should not rematrialize this node. It produces 2 results and has - # very strict constrains - "comment" => "construct MulS: MulS(a, b) = MulS(b, a) = a * b", - "reg_req" => { "in" => [ "gp", "gp", "eax", "gp", "none" ], "out" => [ "eax", "edx", "none" ] }, - "emit" => '. mull %unop', - "outs" => [ "EAX", "EDX", "M" ], - "latency" => 10, - "units" => [ "GP" ], -}, - -"l_Mul" => { - # we should not rematrialize this node. It produces 2 results and has - # very strict constrains - "op_flags" => "C", - "cmp_attr" => "return 1;", - "comment" => "construct lowered MulS: Mul(a, b) = Mul(b, a) = a * b", - "outs" => [ "EAX", "EDX", "M" ], - "arity" => 2 -}, - -"IMul" => { - "irn_flags" => "R", - "comment" => "construct Mul: Mul(a, b) = Mul(b, a) = a * b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. imull %binop', - "latency" => 5, - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"IMul1OP" => { - "irn_flags" => "R", - "comment" => "construct Mul (1 operand format): Mul(a, b) = Mul(b, a) = a * b", - "reg_req" => { "in" => [ "gp", "gp", "eax", "gp", "none" ], "out" => [ "eax", "edx", "none" ] }, - "emit" => '. imull %unop', - "outs" => [ "EAX", "EDX", "M" ], - "latency" => 5, - "units" => [ "GP" ], -}, - -"l_IMul" => { - "op_flags" => "C", - "cmp_attr" => "return 1;", - "comment" => "construct lowered IMul: IMul(a, b) = IMul(b, a) = a * b", - "arity" => 2 -}, - -"And" => { - "irn_flags" => "R", - "comment" => "construct And: And(a, b) = And(b, a) = a AND b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. andl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Or" => { - "irn_flags" => "R", - "comment" => "construct Or: Or(a, b) = Or(b, a) = a OR b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. orl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Xor" => { - "irn_flags" => "R", - "comment" => "construct Xor: Xor(a, b) = Xor(b, a) = a EOR b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. xorl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"l_Xor" => { - "op_flags" => "C", - "cmp_attr" => "return 1;", - "comment" => "construct lowered Xor: Xor(a, b) = Xor(b, a) = a XOR b", - "arity" => 2 +Add => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ], + out => [ "in_r4 in_r5", "flags", "none" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "res", "flags", "M" ], + emit => '. add%M %binop', + am => "source,binary", + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +AddMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => ". add%M %SI3, %AM", + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +AddMem8Bit => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => ". add%M %SB3, %AM", + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +Adc => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp", "flags" ], + out => [ "in_r4 in_r5", "flags", "none" ] }, + ins => [ "base", "index", "mem", "left", "right", "eflags" ], + outs => [ "res", "flags", "M" ], + emit => '. adc%M %binop', + am => "source,binary", + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +l_Add => { + op_flags => "C", + reg_req => { in => [ "none", "none" ], out => [ "none" ] }, + ins => [ "left", "right" ], +}, + +l_Adc => { + reg_req => { in => [ "none", "none", "none" ], out => [ "none" ] }, + ins => [ "left", "right", "eflags" ], +}, + +Mul => { + # we should not rematrialize this node. It produces 2 results and has + # very strict constraints + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax", "gp" ], + out => [ "eax", "edx", "none" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + emit => '. mul%M %unop4', + outs => [ "res_low", "res_high", "M" ], + am => "source,binary", + latency => 10, + units => [ "GP" ], + modified_flags => $status_flags +}, + +l_Mul => { + # we should not rematrialize this node. It produces 2 results and has + # very strict constraints + op_flags => "C", + cmp_attr => "return 1;", + outs => [ "EAX", "EDX", "M" ], + arity => 2 +}, + +IMul => { + irn_flags => "R", + state => "exc_pinned", + # TODO: adjust out requirements for the 3 operand form + # (no need for should_be_same then) + reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ], + out => [ "in_r4 in_r5", "flags", "none" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "res", "flags", "M" ], + am => "source,binary", + latency => 5, + units => [ "GP" ], + mode => $mode_gp, + modified_flags => $status_flags +}, + +IMul1OP => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax", "gp" ], + out => [ "eax", "edx", "none" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + emit => '. imul%M %unop4', + outs => [ "res_low", "res_high", "M" ], + am => "source,binary", + latency => 5, + units => [ "GP" ], + modified_flags => $status_flags +}, + +l_IMul => { + op_flags => "C", + cmp_attr => "return 1;", + outs => [ "res_low", "res_high", "M" ], + arity => 2 +}, + +And => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ], + out => [ "in_r4 in_r5", "flags", "none" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "res", "flags", "M" ], + op_modes => "commutative | am | immediate | mode_neutral", + am => "source,binary", + emit => '. and%M %binop', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +AndMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => '. and%M %SI3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +AndMem8Bit => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => '. and%M %SB3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +Or => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ], + out => [ "in_r4 in_r5", "flags", "none" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "res", "flags", "M" ], + am => "source,binary", + emit => '. or%M %binop', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +OrMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => '. or%M %SI3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +OrMem8Bit => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => '. or%M %SB3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +Xor => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ], + out => [ "in_r4 in_r5", "flags", "none" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "res", "flags", "M" ], + am => "source,binary", + emit => '. xor%M %binop', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +XorMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => '. xor%M %SI3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +XorMem8Bit => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => '. xor%M %SB3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags }, # not commutative operations -"Sub" => { - "irn_flags" => "R", - "comment" => "construct Sub: Sub(a, b) = a - b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. subl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Sbb" => { - "comment" => "construct Sub with Carry: SubC(a, b) = a - b - carry", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3 !in_r4" ] }, - "emit" => '. sbbl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Sub64Bit" => { - "irn_flags" => "R", - "comment" => "construct 64Bit Sub: Sub(a_l, a_h, b_l, b_h) = a_l - b_l; a_h - b_h - borrow", - "arity" => 4, - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp" ], "out" => [ "!in", "!in" ] }, - "emit" => ' -. movl %S1, %D1 -. movl %S2, %D2 -. subl %S3, %D1 -. sbbl %S4, %D2 -', - "outs" => [ "low_res", "high_res" ], - "units" => [ "GP" ], -}, - -"l_Sub" => { - "irn_flags" => "R", - "cmp_attr" => "return 1;", - "comment" => "construct lowered Sub: Sub(a, b) = a - b", - "arity" => 2, -}, - -"l_Sbb" => { - "cmp_attr" => "return 1;", - "comment" => "construct lowered Sub with Carry: SubC(a, b) = a - b - carry", - "arity" => 2, -}, - -"IDiv" => { - "op_flags" => "F|L", - "state" => "exc_pinned", - "reg_req" => { "in" => [ "gp", "gp", "eax", "edx", "gp", "none" ], "out" => [ "eax", "edx", "none" ] }, - "attr" => "ia32_op_flavour_t dm_flav", - "init_attr" => "attr->data.op_flav = dm_flav;", - "emit" => ". idivl %unop", - "outs" => [ "div_res", "mod_res", "M" ], - "latency" => 25, - "units" => [ "GP" ], -}, - -"Div" => { - "op_flags" => "F|L", - "state" => "exc_pinned", - "reg_req" => { "in" => [ "gp", "gp", "eax", "edx", "gp", "none" ], "out" => [ "eax", "edx", "none" ] }, - "attr" => "ia32_op_flavour_t dm_flav", - "init_attr" => "attr->data.op_flav = dm_flav;", - "emit" => ". divl %unop", - "outs" => [ "div_res", "mod_res", "M" ], - "latency" => 25, - "units" => [ "GP" ], -}, - -"Shl" => { - "irn_flags" => "R", - "comment" => "construct Shl: Shl(a, b) = a << b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx", "none" ], "out" => [ "in_r3 !in_r4" ] }, - "emit" => '. shll %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"l_Shl" => { - "cmp_attr" => "return 1;", - "comment" => "construct lowered Shl: Shl(a, b) = a << b", - "arity" => 2 -}, - -"ShlD" => { - "irn_flags" => "R", - "comment" => "construct ShlD: ShlD(a, b, c) = a, b << count (shift left count bits from b into a)", - # Out requirements is: different from all in - # This is because, out must be different from LowPart and ShiftCount. - # We could say "!ecx !in_r4" but it can occur, that all values live through - # this Shift and the only value dying is the ShiftCount. Then there would be a - # register missing, as result must not be ecx and all other registers are - # occupied. What we should write is "!in_r4 !in_r5", but this is not supported - # (and probably never will). So we create artificial interferences of the result - # with all inputs, so the spiller can always assure a free register. - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "ecx", "none" ], "out" => [ "!in" ] }, - "emit" => -' -if (get_ia32_immop_type(node) == ia32_ImmNone) { - if (get_ia32_op_type(node) == ia32_AddrModeD) { - . shldl %%cl, %S4, %AM - } else { - . shldl %%cl, %S4, %S3 - } -} else { - if (get_ia32_op_type(node) == ia32_AddrModeD) { - . shldl $%C, %S4, %AM - } else { - . shldl $%C, %S4, %S3 - } -} -', - "latency" => 6, - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"l_ShlD" => { - "cmp_attr" => "return 1;", - "comment" => "construct lowered ShlD: ShlD(a, b, c) = a, b << count (shift left count bits from b into a)", - "arity" => 3, -}, - -"Shr" => { - "irn_flags" => "R", - "comment" => "construct Shr: Shr(a, b) = a >> b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx", "none" ], "out" => [ "in_r3 !in_r4" ] }, - "emit" => '. shrl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"l_Shr" => { - "cmp_attr" => "return 1;", - "comment" => "construct lowered Shr: Shr(a, b) = a << b", - "arity" => 2 -}, - -"ShrD" => { - "irn_flags" => "R", - "comment" => "construct ShrD: ShrD(a, b, c) = a, b >> count (shift rigth count bits from a into b)", - # Out requirements is: different from all in - # This is because, out must be different from LowPart and ShiftCount. - # We could say "!ecx !in_r4" but it can occur, that all values live through - # this Shift and the only value dying is the ShiftCount. Then there would be a - # register missing, as result must not be ecx and all other registers are - # occupied. What we should write is "!in_r4 !in_r5", but this is not supported - # (and probably never will). So we create artificial interferences of the result - # with all inputs, so the spiller can always assure a free register. - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "ecx", "none" ], "out" => [ "!in" ] }, - "emit" => ' -if (get_ia32_immop_type(node) == ia32_ImmNone) { - if (get_ia32_op_type(node) == ia32_AddrModeD) { - . shrdl %%cl, %S4, %AM - } else { - . shrdl %%cl, %S4, %S3 - } -} else { - if (get_ia32_op_type(node) == ia32_AddrModeD) { - . shrdl $%C, %S4, %AM - } else { - . shrdl $%C, %S4, %S3 - } -} -', - "latency" => 6, - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"l_ShrD" => { - "cmp_attr" => "return 1;", - "comment" => "construct lowered ShrD: ShrD(a, b, c) = a, b >> count (shift rigth count bits from a into b)", - "arity" => 3 -}, - -"Sar" => { - "irn_flags" => "R", - "comment" => "construct Shrs: Shrs(a, b) = a >> b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx", "none" ], "out" => [ "in_r3 !in_r4" ] }, - "emit" => '. sarl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"l_Sar" => { - "cmp_attr" => "return 1;", - "comment" => "construct lowered Sar: Sar(a, b) = a << b", - "arity" => 2 -}, - -"Ror" => { - "irn_flags" => "R", - "comment" => "construct Ror: Ror(a, b) = a ROR b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx", "none" ], "out" => [ "in_r3 !in_r4" ] }, - "emit" => '. rorl %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Rol" => { - "irn_flags" => "R", - "comment" => "construct Rol: Rol(a, b) = a ROL b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx", "none" ], "out" => [ "in_r3 !in_r4" ] }, - "emit" => '. roll %binop', - "units" => [ "GP" ], - "mode" => "mode_Iu", +Sub => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ], + out => [ "in_r4", "flags", "none" ] }, + ins => [ "base", "index", "mem", "minuend", "subtrahend" ], + outs => [ "res", "flags", "M" ], + am => "source,binary", + emit => '. sub%M %binop', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +SubMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "subtrahend" ], + emit => '. sub%M %SI3, %AM', + units => [ "GP" ], + latency => 1, + mode => 'mode_M', + modified_flags => $status_flags +}, + +SubMem8Bit => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "subtrahend" ], + emit => '. sub%M %SB3, %AM', + units => [ "GP" ], + latency => 1, + mode => 'mode_M', + modified_flags => $status_flags +}, + +Sbb => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp", "flags" ], + out => [ "in_r4 !in_r5", "flags", "none" ] }, + ins => [ "base", "index", "mem", "minuend", "subtrahend", "eflags" ], + outs => [ "res", "flags", "M" ], + am => "source,binary", + emit => '. sbb%M %binop', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +l_Sub => { + reg_req => { in => [ "none", "none" ], out => [ "none" ] }, + ins => [ "minuend", "subtrahend" ], +}, + +l_Sbb => { + reg_req => { in => [ "none", "none", "none" ], out => [ "none" ] }, + ins => [ "minuend", "subtrahend", "eflags" ], +}, + +IDiv => { + op_flags => "F|L", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "eax", "edx" ], + out => [ "eax", "flags", "none", "edx", "none" ] }, + ins => [ "base", "index", "mem", "divisor", "dividend_low", "dividend_high" ], + outs => [ "div_res", "flags", "M", "mod_res", "X_exc" ], + am => "source,unary", + emit => ". idiv%M %unop3", + latency => 25, + units => [ "GP" ], + modified_flags => $status_flags +}, + +Div => { + op_flags => "F|L", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "eax", "edx" ], + out => [ "eax", "flags", "none", "edx", "none" ] }, + ins => [ "base", "index", "mem", "divisor", "dividend_low", "dividend_high" ], + outs => [ "div_res", "flags", "M", "mod_res", "X_exc" ], + am => "source,unary", + emit => ". div%M %unop3", + latency => 25, + units => [ "GP" ], + modified_flags => $status_flags +}, + +Shl => { + irn_flags => "R", + reg_req => { in => [ "gp", "ecx" ], + out => [ "in_r1 !in_r2", "flags" ] }, + ins => [ "val", "count" ], + outs => [ "res", "flags" ], + emit => '. shl %SB1, %S0', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +ShlMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "ecx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "count" ], + emit => '. shl%M %SB3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +l_ShlDep => { + cmp_attr => "return 1;", + ins => [ "val", "count", "dep" ], + arity => 3 +}, + +ShlD => { + irn_flags => "R", + reg_req => { in => [ "gp", "gp", "ecx" ], + out => [ "in_r1 !in_r2 !in_r3", "flags" ] }, + ins => [ "val_high", "val_low", "count" ], + outs => [ "res", "flags" ], + emit => ". shld%M %SB2, %S1, %D0", + latency => 6, + units => [ "GP" ], + mode => $mode_gp, + modified_flags => $status_flags +}, + +l_ShlD => { + cmp_attr => "return 1;", + ins => [ "val_high", "val_low", "count" ], + arity => 3, +}, + +Shr => { + irn_flags => "R", + reg_req => { in => [ "gp", "ecx" ], + out => [ "in_r1 !in_r2", "flags" ] }, + ins => [ "val", "count" ], + outs => [ "res", "flags" ], + emit => '. shr %SB1, %S0', + units => [ "GP" ], + mode => $mode_gp, + latency => 1, + modified_flags => $status_flags +}, + +ShrMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "ecx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "count" ], + emit => '. shr%M %SB3, %AM', + units => [ "GP" ], + mode => "mode_M", + latency => 1, + modified_flags => $status_flags +}, + +l_ShrDep => { + cmp_attr => "return 1;", + ins => [ "val", "count", "dep" ], + arity => 3 +}, + +ShrD => { + irn_flags => "R", + reg_req => { in => [ "gp", "gp", "ecx" ], + out => [ "in_r1 !in_r2 !in_r3", "flags" ] }, + ins => [ "val_high", "val_low", "count" ], + outs => [ "res", "flags" ], + emit => ". shrd%M %SB2, %S1, %D0", + latency => 6, + units => [ "GP" ], + mode => $mode_gp, + modified_flags => $status_flags +}, + +l_ShrD => { + cmp_attr => "return 1;", + arity => 3, + ins => [ "val_high", "val_low", "count" ], +}, + +Sar => { + irn_flags => "R", + reg_req => { in => [ "gp", "ecx" ], + out => [ "in_r1 !in_r2", "flags" ] }, + ins => [ "val", "count" ], + outs => [ "res", "flags" ], + emit => '. sar %SB1, %S0', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +SarMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "ecx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "count" ], + emit => '. sar%M %SB3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +l_SarDep => { + cmp_attr => "return 1;", + ins => [ "val", "count", "dep" ], + arity => 3 +}, + +Ror => { + irn_flags => "R", + reg_req => { in => [ "gp", "ecx" ], + out => [ "in_r1 !in_r2", "flags" ] }, + ins => [ "val", "count" ], + outs => [ "res", "flags" ], + emit => '. ror %SB1, %S0', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +RorMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "ecx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "count" ], + emit => '. ror%M %SB3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +Rol => { + irn_flags => "R", + reg_req => { in => [ "gp", "ecx" ], + out => [ "in_r1 !in_r2", "flags" ] }, + ins => [ "val", "count" ], + outs => [ "res", "flags" ], + emit => '. rol %SB1, %S0', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +RolMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "ecx" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "count" ], + emit => '. rol%M %SB3, %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags }, # unary operations -"Neg" => { - "irn_flags" => "R", - "comment" => "construct Minus: Minus(a) = -a", - "reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. negl %unop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Minus64Bit" => { - "irn_flags" => "R", - "comment" => "construct 64Bit Minus: Minus(a_l, a_h, 0) = 0 - a_l; 0 - a_h - borrow", - "arity" => 4, - "reg_req" => { "in" => [ "gp", "gp", "gp" ], "out" => [ "!in", "!in" ] }, - "emit" => ' -. movl %S1, %D1 -. movl %S1, %D2 -. subl %S2, %D1 -. sbbl %S3, %D2 -', - "outs" => [ "low_res", "high_res" ], - "units" => [ "GP" ], -}, - - -"l_Neg" => { - "cmp_attr" => "return 1;", - "comment" => "construct lowered Minus: Minus(a) = -a", - "arity" => 1, -}, - -"Inc" => { - "irn_flags" => "R", - "comment" => "construct Increment: Inc(a) = a++", - "reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. incl %unop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Dec" => { - "irn_flags" => "R", - "comment" => "construct Decrement: Dec(a) = a--", - "reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. decl %unop', - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Not" => { - "irn_flags" => "R", - "comment" => "construct Not: Not(a) = !a", - "reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. notl %unop', - "units" => [ "GP" ], - "mode" => "mode_Iu", +Neg => { + irn_flags => "R", + reg_req => { in => [ "gp" ], + out => [ "in_r1", "flags" ] }, + emit => '. neg %S0', + ins => [ "val" ], + outs => [ "res", "flags" ], + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + modified_flags => $status_flags +}, + +NegMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "none" ] }, + ins => [ "base", "index", "mem" ], + emit => '. neg%M %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + modified_flags => $status_flags +}, + +Minus64Bit => { + irn_flags => "R", + reg_req => { in => [ "gp", "gp" ], out => [ "in_r1", "in_r2" ] }, + outs => [ "low_res", "high_res" ], + units => [ "GP" ], + latency => 3, + modified_flags => $status_flags +}, + + +Inc => { + irn_flags => "R", + reg_req => { in => [ "gp" ], + out => [ "in_r1", "flags" ] }, + ins => [ "val" ], + outs => [ "res", "flags" ], + emit => '. inc %S0', + units => [ "GP" ], + mode => $mode_gp, + latency => 1, + modified_flags => $status_flags_wo_cf +}, + +IncMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "none" ] }, + ins => [ "base", "index", "mem" ], + emit => '. inc%M %AM', + units => [ "GP" ], + mode => "mode_M", + latency => 1, + modified_flags => $status_flags_wo_cf +}, + +Dec => { + irn_flags => "R", + reg_req => { in => [ "gp" ], + out => [ "in_r1", "flags" ] }, + ins => [ "val" ], + outs => [ "res", "flags" ], + emit => '. dec %S0', + units => [ "GP" ], + mode => $mode_gp, + latency => 1, + modified_flags => $status_flags_wo_cf +}, + +DecMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "none" ] }, + ins => [ "base", "index", "mem" ], + emit => '. dec%M %AM', + units => [ "GP" ], + mode => "mode_M", + latency => 1, + modified_flags => $status_flags_wo_cf +}, + +Not => { + irn_flags => "R", + reg_req => { in => [ "gp" ], + out => [ "in_r1", "flags" ] }, + ins => [ "val" ], + outs => [ "res", "flags" ], + emit => '. not %S0', + units => [ "GP" ], + latency => 1, + mode => $mode_gp, + # no flags modified +}, + +NotMem => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "none" ] }, + ins => [ "base", "index", "mem" ], + emit => '. not%M %AM', + units => [ "GP" ], + latency => 1, + mode => "mode_M", + # no flags modified +}, + +Cmc => { + reg_req => { in => [ "flags" ], out => [ "flags" ] }, + emit => '.cmc', + units => [ "GP" ], + latency => 1, + mode => $mode_flags, + modified_flags => $status_flags +}, + +Stc => { + reg_req => { out => [ "flags" ] }, + emit => '.stc', + units => [ "GP" ], + latency => 1, + mode => $mode_flags, + modified_flags => $status_flags }, # other operations -"CondJmp" => { - "op_flags" => "L|X|Y", - "comment" => "construct conditional jump: CMP A, B && JMPxx LABEL", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ] }, - "outs" => [ "false", "true" ], - "latency" => 3, - "units" => [ "BRANCH" ], -}, - -"TestJmp" => { - "op_flags" => "L|X|Y", - "comment" => "construct conditional jump: TEST A, B && JMPxx LABEL", - "reg_req" => { "in" => [ "gp", "gp" ] }, - "outs" => [ "false", "true" ], - "latency" => 3, - "units" => [ "BRANCH" ], -}, - -"CJmpAM" => { - "op_flags" => "L|X|Y", - "comment" => "construct conditional jump without CMP (replaces CondJmp): JMPxx LABEL", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "none", "none" ] }, - "outs" => [ "false", "true" ], - "units" => [ "BRANCH" ], -}, - -"CJmp" => { - "op_flags" => "L|X|Y", - "comment" => "construct conditional jump without CMP (replaces TestJmp): JMPxx LABEL", - "reg_req" => { "in" => [ "gp", "gp" ] }, - "units" => [ "BRANCH" ], -}, - -"SwitchJmp" => { - "op_flags" => "L|X|Y", - "comment" => "construct switch", - "reg_req" => { "in" => [ "gp" ], "out" => [ "none" ] }, - "latency" => 3, - "units" => [ "BRANCH" ], -}, - -"Const" => { - "op_flags" => "c", - "irn_flags" => "R", - "comment" => "represents an integer constant", - "reg_req" => { "out" => [ "gp" ] }, - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Unknown_GP" => { - "op_flags" => "c", - "irn_flags" => "I", - "comment" => "unknown value", - "reg_req" => { "out" => [ "gp_UKNWN" ] }, - "units" => [], - "emit" => "", - "mode" => "mode_Iu" -}, - -"Unknown_VFP" => { - "op_flags" => "c", - "irn_flags" => "I", - "comment" => "unknown value", - "reg_req" => { "out" => [ "vfp_UKNWN" ] }, - "units" => [], - "emit" => "", - "mode" => "mode_E" -}, - -"Unknown_XMM" => { - "op_flags" => "c", - "irn_flags" => "I", - "comment" => "unknown value", - "reg_req" => { "out" => [ "xmm_UKNWN" ] }, - "units" => [], - "emit" => "", - "mode" => "mode_E" -}, - -"NoReg_GP" => { - "op_flags" => "c", - "irn_flags" => "I", - "comment" => "unknown GP value", - "reg_req" => { "out" => [ "gp_NOREG" ] }, - "units" => [], - "emit" => "", - "mode" => "mode_Iu" -}, - -"NoReg_VFP" => { - "op_flags" => "c", - "irn_flags" => "I", - "comment" => "unknown VFP value", - "reg_req" => { "out" => [ "vfp_NOREG" ] }, - "units" => [], - "emit" => "", - "mode" => "mode_E" -}, - -"NoReg_XMM" => { - "op_flags" => "c", - "irn_flags" => "I", - "comment" => "unknown XMM value", - "reg_req" => { "out" => [ "xmm_NOREG" ] }, - "units" => [], - "emit" => "", - "mode" => "mode_E" -}, - -"ChangeCW" => { - "irn_flags" => "R", - "comment" => "change floating point control word", - "reg_req" => { "out" => [ "fp_cw" ] }, - "mode" => "mode_Hu", - "latency" => 3, - "units" => [ "GP" ], -}, - -"FldCW" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "load floating point control word FldCW(ptr, mem) = LD ptr -> reg", - "reg_req" => { "in" => [ "gp", "gp", "none" ], "out" => [ "fp_cw" ] }, - "latency" => 5, - "emit" => ". fldcw %AM", - "mode" => "mode_Hu", - "units" => [ "GP" ], -}, - -"FstCW" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "store floating point control word: FstCW(ptr, mem) = ST ptr -> reg", - "reg_req" => { "in" => [ "gp", "gp", "fp_cw", "none" ] }, - "latency" => 5, - "emit" => ". fstcw %AM", - "mode" => "mode_M", - "units" => [ "GP" ], -}, - -"Cltd" => { - # we should not rematrialize this node. It produces 2 results and has - # very strict constrains - "comment" => "construct CDQ: sign extend EAX -> EDX:EAX", - "reg_req" => { "in" => [ "gp" ], "out" => [ "eax in_r1", "edx" ] }, - "emit" => '. cltd', - "outs" => [ "EAX", "EDX" ], - "units" => [ "GP" ], +Cmp => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ], + out => [ "flags", "none", "none" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "eflags", "unused", "M" ], + am => "source,binary", + emit => '. cmp%M %binop', + attr => "int ins_permuted, int cmp_unsigned", + init_attr => "attr->data.ins_permuted = ins_permuted;\n". + "\tattr->data.cmp_unsigned = cmp_unsigned;\n", + latency => 1, + units => [ "GP" ], + mode => $mode_flags, + modified_flags => $status_flags +}, + +Cmp8Bit => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx", "eax ebx ecx edx" ] , out => [ "flags" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "eflags" ], + am => "source,binary", + emit => '. cmpb %binop', + attr => "int ins_permuted, int cmp_unsigned", + init_attr => "attr->data.ins_permuted = ins_permuted;\n". + "\tattr->data.cmp_unsigned = cmp_unsigned;\n", + latency => 1, + units => [ "GP" ], + mode => $mode_flags, + modified_flags => $status_flags +}, + +Test => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ] , out => [ "flags" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "eflags" ], + am => "source,binary", + emit => '. test%M %binop', + attr => "int ins_permuted, int cmp_unsigned", + init_attr => "attr->data.ins_permuted = ins_permuted;\n". + "\tattr->data.cmp_unsigned = cmp_unsigned;\n", + latency => 1, + units => [ "GP" ], + mode => $mode_flags, + modified_flags => $status_flags +}, + +Test8Bit => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx", "eax ebx ecx edx" ] , out => [ "flags" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "eflags" ], + am => "source,binary", + emit => '. testb %binop', + attr => "int ins_permuted, int cmp_unsigned", + init_attr => "attr->data.ins_permuted = ins_permuted;\n". + "\tattr->data.cmp_unsigned = cmp_unsigned;\n", + latency => 1, + units => [ "GP" ], + mode => $mode_flags, + modified_flags => $status_flags +}, + +Set => { + #irn_flags => "R", + reg_req => { in => [ "eflags" ], out => [ "eax ebx ecx edx" ] }, + ins => [ "eflags" ], + attr_type => "ia32_condcode_attr_t", + attr => "pn_Cmp pnc, int ins_permuted", + init_attr => "attr->attr.data.ins_permuted = ins_permuted;\n". + "\tset_ia32_ls_mode(res, mode_Bu);\n", + emit => '. set%CMP0 %DB0', + latency => 1, + units => [ "GP" ], + mode => $mode_gp, +}, + +SetMem => { + #irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eflags" ], out => [ "none" ] }, + ins => [ "base", "index", "mem","eflags" ], + attr_type => "ia32_condcode_attr_t", + attr => "pn_Cmp pnc, int ins_permuted", + init_attr => "attr->attr.data.ins_permuted = ins_permuted;\n". + "\tset_ia32_ls_mode(res, mode_Bu);\n", + emit => '. set%CMP3 %AM', + latency => 1, + units => [ "GP" ], + mode => 'mode_M', +}, + +CMov => { + #irn_flags => "R", + # (note: leave the false,true order intact to make it compatible with other + # ia32_binary ops) + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "gp", "eflags" ], out => [ "in_r4 in_r5" ] }, + ins => [ "base", "index", "mem", "val_false", "val_true", "eflags" ], + am => "source,binary", + attr_type => "ia32_condcode_attr_t", + attr => "int ins_permuted, pn_Cmp pnc", + init_attr => "attr->attr.data.ins_permuted = ins_permuted;", + latency => 1, + units => [ "GP" ], + mode => $mode_gp, +}, + +Jcc => { + state => "pinned", + op_flags => "L|X|Y", + reg_req => { in => [ "eflags" ], out => [ "none", "none" ] }, + ins => [ "eflags" ], + outs => [ "false", "true" ], + attr_type => "ia32_condcode_attr_t", + attr => "pn_Cmp pnc", + latency => 2, + units => [ "BRANCH" ], +}, + +SwitchJmp => { + state => "pinned", + op_flags => "L|X|Y", + reg_req => { in => [ "gp" ], out => [ "none" ] }, + mode => "mode_T", + attr_type => "ia32_condcode_attr_t", + attr => "long pnc", + latency => 3, + units => [ "BRANCH" ], + modified_flags => $status_flags, +}, + +IJmp => { + state => "pinned", + op_flags => "X", + reg_req => { in => [ "gp", "gp", "none", "gp" ] }, + ins => [ "base", "index", "mem", "target" ], + am => "source,unary", + emit => '. jmp *%unop3', + latency => 1, + units => [ "BRANCH" ], + mode => "mode_X", +}, + +Const => { + op_flags => "c", + irn_flags => "R", + reg_req => { out => [ "gp" ] }, + units => [ "GP" ], + attr => "ir_entity *symconst, int symconst_sign, long offset", + attr_type => "ia32_immediate_attr_t", + latency => 1, + mode => $mode_gp, +}, + +GetEIP => { + op_flags => "c", + reg_req => { out => [ "gp" ] }, + units => [ "GP" ], + latency => 5, + mode => $mode_gp, + modified_flags => $status_flags, +}, + +Unknown_GP => { + state => "pinned", + op_flags => "c|NB", + irn_flags => "I", + reg_req => { out => [ "gp_UKNWN" ] }, + units => [], + emit => "", + latency => 0, + mode => $mode_gp +}, + +Unknown_VFP => { + state => "pinned", + op_flags => "c|NB", + irn_flags => "I", + reg_req => { out => [ "vfp_UKNWN" ] }, + units => [], + emit => "", + mode => "mode_E", + latency => 0, + attr_type => "ia32_x87_attr_t", +}, + +Unknown_XMM => { + state => "pinned", + op_flags => "c|NB", + irn_flags => "I", + reg_req => { out => [ "xmm_UKNWN" ] }, + units => [], + emit => "", + latency => 0, + mode => $mode_xmm +}, + +NoReg_GP => { + state => "pinned", + op_flags => "c|NB|NI", + irn_flags => "I", + reg_req => { out => [ "gp_NOREG" ] }, + units => [], + emit => "", + latency => 0, + mode => $mode_gp +}, + +NoReg_VFP => { + state => "pinned", + op_flags => "c|NB|NI", + irn_flags => "I", + reg_req => { out => [ "vfp_NOREG" ] }, + units => [], + emit => "", + mode => "mode_E", + latency => 0, + attr_type => "ia32_x87_attr_t", +}, + +NoReg_XMM => { + state => "pinned", + op_flags => "c|NB|NI", + irn_flags => "I", + reg_req => { out => [ "xmm_NOREG" ] }, + units => [], + emit => "", + latency => 0, + mode => "mode_E" +}, + +ChangeCW => { + state => "pinned", + op_flags => "c", + irn_flags => "I", + reg_req => { out => [ "fp_cw" ] }, + mode => $mode_fpcw, + latency => 3, + units => [ "GP" ], + modified_flags => $fpcw_flags +}, + +FldCW => { + op_flags => "L|F", + state => "pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "fp_cw" ] }, + ins => [ "base", "index", "mem" ], + latency => 5, + emit => ". fldcw %AM", + mode => $mode_fpcw, + units => [ "GP" ], + modified_flags => $fpcw_flags +}, + +FnstCW => { + op_flags => "L|F", + state => "pinned", + reg_req => { in => [ "gp", "gp", "none", "fp_cw" ], out => [ "none" ] }, + ins => [ "base", "index", "mem", "fpcw" ], + latency => 5, + emit => ". fnstcw %AM", + mode => "mode_M", + units => [ "GP" ], +}, + +FnstCWNOP => { + op_flags => "L|F", + state => "pinned", + reg_req => { in => [ "fp_cw" ], out => [ "none" ] }, + ins => [ "fpcw" ], + latency => 0, + emit => "", + mode => "mode_M", +}, + +Cltd => { + # we should not rematrialize this node. It has very strict constraints. + reg_req => { in => [ "eax", "edx" ], out => [ "edx" ] }, + ins => [ "val", "clobbered" ], + emit => '. cltd', + latency => 1, + mode => $mode_gp, + units => [ "GP" ], }, # Load / Store - -"Load" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "construct Load: Load(ptr, mem) = LD ptr -> reg", - "reg_req" => { "in" => [ "gp", "gp", "none" ], "out" => [ "gp", "none" ] }, - "latency" => 3, - "emit" => ". mov%SE%ME%.l %AM, %D1", - "outs" => [ "res", "M" ], - "units" => [ "GP" ], -}, - -"l_Load" => { - "op_flags" => "L|F", - "cmp_attr" => "return 1;", - "comment" => "construct lowered Load: Load(ptr, mem) = LD ptr -> reg", - "outs" => [ "res", "M" ], - "arity" => 2, -}, - -"l_Store" => { - "op_flags" => "L|F", - "cmp_attr" => "return 1;", - "state" => "exc_pinned", - "comment" => "construct lowered Store: Store(ptr, val, mem) = ST ptr,val", - "arity" => 3, - "mode" => "mode_M", -}, - -"Store" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "construct Store: Store(ptr, val, mem) = ST ptr,val", - "reg_req" => { "in" => [ "gp", "gp", "gp", "none" ] }, - "emit" => '. mov%M %binop', - "latency" => 3, - "units" => [ "GP" ], - "mode" => "mode_M", -}, - -"Store8Bit" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "construct 8Bit Store: Store(ptr, val, mem) = ST ptr,val", - "reg_req" => { "in" => [ "gp", "gp", "eax ebx ecx edx", "none" ] }, - "emit" => '. mov%M %binop', - "latency" => 3, - "units" => [ "GP" ], - "mode" => "mode_M", -}, - -"Lea" => { - "irn_flags" => "R", - "comment" => "construct Lea: Lea(a,b) = lea [a+b*const+offs] | res = a + b * const + offs with const = 0,1,2,4,8", - "reg_req" => { "in" => [ "gp", "gp" ], "out" => [ "in_r1" ] }, - "emit" => '. leal %AM, %D1', - "latency" => 2, - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Push" => { - "comment" => "push on the stack", - "reg_req" => { "in" => [ "gp", "gp", "gp", "esp", "none" ], "out" => [ "esp", "none" ] }, - "emit" => '. pushl %unop', - "outs" => [ "stack:I|S", "M" ], - "latency" => 3, - "units" => [ "GP" ], -}, - -"Pop" => { - "comment" => "pop a gp register from the stack", - "reg_req" => { "in" => [ "gp", "gp", "esp", "none" ], "out" => [ "esp", "gp", "none" ] }, - "emit" => '. popl %unop', - "outs" => [ "stack:I|S", "res", "M" ], - "latency" => 4, - "units" => [ "GP" ], -}, - -"Enter" => { - "comment" => "create stack frame", - "reg_req" => { "in" => [ "esp" ], "out" => [ "ebp", "esp" ] }, - "emit" => '. enter', - "outs" => [ "frame:I", "stack:I|S", "M" ], - "latency" => 15, - "units" => [ "GP" ], -}, - -"Leave" => { - "comment" => "destroy stack frame", - "reg_req" => { "in" => [ "esp", "ebp" ], "out" => [ "ebp", "esp" ] }, - "emit" => '. leave', - "outs" => [ "frame:I", "stack:I|S" ], - "latency" => 3, - "units" => [ "GP" ], -}, - -"AddSP" => { - "irn_flags" => "I", - "comment" => "allocate space on stack", - "reg_req" => { "in" => [ "gp", "gp", "esp", "gp", "none" ], "out" => [ "in_r3", "none" ] }, - "emit" => '. addl %binop', - "outs" => [ "stack:S", "M" ], - "units" => [ "GP" ], -}, - -"SubSP" => { - "irn_flags" => "I", - "comment" => "free space on stack", - "reg_req" => { "in" => [ "gp", "gp", "esp", "gp", "none" ], "out" => [ "in_r3", "none" ] }, - "emit" => '. subl %binop', - "outs" => [ "stack:S", "M" ], - "units" => [ "GP" ], -}, - -"LdTls" => { - "irn_flags" => "R", - "comment" => "get the TLS base address", - "reg_req" => { "out" => [ "gp" ] }, - "units" => [ "GP" ], +# +# Note that we add additional latency values depending on address mode, so a +# lateny of 0 for load is correct + +Load => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "gp", "none", "none" ] }, + ins => [ "base", "index", "mem" ], + outs => [ "res", "M", "X_exc" ], + latency => 0, + emit => ". mov%SE%ME%.l %AM, %D0", + units => [ "GP" ], +}, + +Store => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "none", "none" ] }, + ins => [ "base", "index", "mem", "val" ], + outs => [ "M", "X_exc" ], + emit => '. mov%M %SI3, %AM', + latency => 2, + units => [ "GP" ], + mode => "mode_M", +}, + +Store8Bit => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx" ], out => ["none", "none" ] }, + ins => [ "base", "index", "mem", "val" ], + outs => [ "M", "X_exc" ], + emit => '. mov%M %SB3, %AM', + latency => 2, + units => [ "GP" ], + mode => "mode_M", +}, + +Lea => { + irn_flags => "R", + reg_req => { in => [ "gp", "gp" ], out => [ "gp" ] }, + ins => [ "base", "index" ], + emit => '. leal %AM, %D0', + latency => 2, + units => [ "GP" ], + mode => $mode_gp, +# lea doesn't modify the flags, but setting this seems advantageous since it +# increases chances that the Lea is transformed back to an Add + modified_flags => 1, +}, + +Push => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp", "esp" ], out => [ "esp", "none" ] }, + ins => [ "base", "index", "mem", "val", "stack" ], + emit => '. push%M %unop3', + outs => [ "stack:I|S", "M" ], + am => "source,unary", + latency => 2, + units => [ "GP" ], +}, + +Pop => { + state => "exc_pinned", + reg_req => { in => [ "none", "esp" ], out => [ "gp", "none", "none", "esp" ] }, + ins => [ "mem", "stack" ], + outs => [ "res", "M", "unused", "stack:I|S" ], + emit => '. pop%M %D0', + latency => 3, # Pop is more expensive than Push on Athlon + units => [ "GP" ], +}, + +PopMem => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "esp" ], out => [ "none", "none", "none", "esp" ] }, + ins => [ "base", "index", "mem", "stack" ], + outs => [ "unused0", "M", "unused1", "stack:I|S" ], + emit => '. pop%M %AM', + latency => 3, # Pop is more expensive than Push on Athlon + units => [ "GP" ], +}, + +Enter => { + reg_req => { in => [ "esp" ], out => [ "ebp", "esp", "none" ] }, + emit => '. enter', + outs => [ "frame:I", "stack:I|S", "M" ], + latency => 15, + units => [ "GP" ], +}, + +Leave => { + reg_req => { in => [ "ebp" ], out => [ "ebp", "esp" ] }, + emit => '. leave', + outs => [ "frame:I", "stack:I|S" ], + latency => 3, + units => [ "GP" ], +}, + +AddSP => { + irn_flags => "I", + state => "pinned", + reg_req => { in => [ "gp", "gp", "none", "esp", "gp" ], out => [ "in_r4", "none" ] }, + ins => [ "base", "index", "mem", "stack", "size" ], + am => "source,binary", + emit => '. addl %binop', + latency => 1, + outs => [ "stack:I|S", "M" ], + units => [ "GP" ], + modified_flags => $status_flags +}, + +SubSP => { +#irn_flags => "I", + state => "pinned", + reg_req => { in => [ "gp", "gp", "none", "esp", "gp" ], out => [ "in_r4", "gp", "none" ] }, + ins => [ "base", "index", "mem", "stack", "size" ], + am => "source,binary", + emit => ". subl %binop\n". + ". movl %%esp, %D1", + latency => 2, + outs => [ "stack:I|S", "addr", "M" ], + units => [ "GP" ], + modified_flags => $status_flags +}, + +RepPrefix => { + op_flags => "K", + state => "pinned", + mode => "mode_M", + emit => ". rep", + latency => 0, +}, + +LdTls => { + irn_flags => "R", + reg_req => { out => [ "gp" ] }, + units => [ "GP" ], + latency => 1, +}, + +Bt => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp" ], out => [ "flags" ] }, + ins => [ "left", "right" ], + emit => '. bt%M %S1, %S0', + units => [ "GP" ], + latency => 1, + mode => $mode_flags, + modified_flags => $status_flags # only CF is set, but the other flags are undefined }, - - #-----------------------------------------------------------------------------# # _____ _____ ______ __ _ _ _ # # / ____/ ____| ____| / _| | | | | | # @@ -934,352 +1469,375 @@ if (get_ia32_immop_type(node) == ia32_ImmNone) { # |_____/_____/|______| |_| |_|\___/ \__,_|\__| |_| |_|\___/ \__,_|\___||___/ # #-----------------------------------------------------------------------------# -# commutative operations - -"xAdd" => { - "irn_flags" => "R", - "comment" => "construct SSE Add: Add(a, b) = Add(b, a) = a + b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. add%XXM %binop', - "latency" => 4, - "units" => [ "SSE" ], - "mode" => "mode_E", -}, - -"xMul" => { - "irn_flags" => "R", - "comment" => "construct SSE Mul: Mul(a, b) = Mul(b, a) = a * b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. mul%XXM %binop', - "latency" => 4, - "units" => [ "SSE" ], - "mode" => "mode_E", -}, - -"xMax" => { - "irn_flags" => "R", - "comment" => "construct SSE Max: Max(a, b) = Max(b, a) = a > b ? a : b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. max%XXM %binop', - "latency" => 2, - "units" => [ "SSE" ], - "mode" => "mode_E", -}, - -"xMin" => { - "irn_flags" => "R", - "comment" => "construct SSE Min: Min(a, b) = Min(b, a) = a < b ? a : b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. min%XXM %binop', - "latency" => 2, - "units" => [ "SSE" ], - "mode" => "mode_E", -}, - -"xAnd" => { - "irn_flags" => "R", - "comment" => "construct SSE And: And(a, b) = a AND b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. andp%XSD %binop', - "latency" => 3, - "units" => [ "SSE" ], - "mode" => "mode_E", -}, - -"xOr" => { - "irn_flags" => "R", - "comment" => "construct SSE Or: Or(a, b) = a OR b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. orp%XSD %binop', - "units" => [ "SSE" ], - "mode" => "mode_E", -}, - -"xXor" => { - "irn_flags" => "R", - "comment" => "construct SSE Xor: Xor(a, b) = a XOR b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. xorp%XSD %binop', - "latency" => 3, - "units" => [ "SSE" ], - "mode" => "mode_E", +# produces a 0/+0.0 +xZero => { + irn_flags => "R", + reg_req => { out => [ "xmm" ] }, + emit => '. xorp%XSD %D0, %D0', + latency => 3, + units => [ "SSE" ], + mode => $mode_xmm +}, + +xPzero => { + irn_flags => "R", + reg_req => { out => [ "xmm" ] }, + emit => '. pxor %D0, %D0', + latency => 3, + units => [ "SSE" ], + mode => $mode_xmm +}, + +# produces all 1 bits +xAllOnes => { + irn_flags => "R", + reg_req => { out => [ "xmm" ] }, + emit => '. pcmpeqb %D0, %D0', + latency => 3, + units => [ "SSE" ], + mode => $mode_xmm +}, + +# integer shift left, dword +xPslld => { + irn_flags => "R", + reg_req => { in => [ "xmm", "xmm" ], out => [ "in_r1 !in_r2" ] }, + emit => '. pslld %SI1, %D0', + latency => 3, + units => [ "SSE" ], + mode => $mode_xmm +}, + +# integer shift left, qword +xPsllq => { + irn_flags => "R", + reg_req => { in => [ "xmm", "xmm" ], out => [ "in_r1 !in_r2" ] }, + emit => '. psllq %SI1, %D0', + latency => 3, + units => [ "SSE" ], + mode => $mode_xmm +}, + +# integer shift right, dword +xPsrld => { + irn_flags => "R", + reg_req => { in => [ "xmm", "xmm" ], out => [ "in_r1 !in_r2" ] }, + emit => '. psrld %SI1, %D0', + latency => 1, + units => [ "SSE" ], + mode => $mode_xmm +}, + +# mov from integer to SSE register +xMovd => { + irn_flags => "R", + reg_req => { in => [ "gp" ], out => [ "xmm" ] }, + emit => '. movd %S0, %D0', + latency => 1, + units => [ "SSE" ], + mode => $mode_xmm }, -# not commutative operations +# commutative operations -"xAndNot" => { - "irn_flags" => "R", - "comment" => "construct SSE AndNot: AndNot(a, b) = a AND NOT b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3 !in_r4" ] }, - "emit" => '. andnp%XSD %binop', - "latency" => 3, - "units" => [ "SSE" ], - "mode" => "mode_E", +xAdd => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4 in_r5" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + am => "source,binary", + emit => '. add%XXM %binop', + latency => 4, + units => [ "SSE" ], + mode => $mode_xmm +}, + +xMul => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4 in_r5" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + am => "source,binary", + emit => '. mul%XXM %binop', + latency => 4, + units => [ "SSE" ], + mode => $mode_xmm +}, + +xMax => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4 in_r5" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + am => "source,binary", + emit => '. max%XXM %binop', + latency => 2, + units => [ "SSE" ], + mode => $mode_xmm +}, + +xMin => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4 in_r5" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + am => "source,binary", + emit => '. min%XXM %binop', + latency => 2, + units => [ "SSE" ], + mode => $mode_xmm +}, + +xAnd => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4 in_r5" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + am => "source,binary", + emit => '. andp%XSD %binop', + latency => 3, + units => [ "SSE" ], + mode => $mode_xmm +}, + +xOr => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4 in_r5" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + am => "source,binary", + emit => '. orp%XSD %binop', + latency => 3, + units => [ "SSE" ], + mode => $mode_xmm +}, + +xXor => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4 in_r5" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + am => "source,binary", + emit => '. xorp%XSD %binop', + latency => 3, + units => [ "SSE" ], + mode => $mode_xmm }, -"xSub" => { - "irn_flags" => "R", - "comment" => "construct SSE Sub: Sub(a, b) = a - b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] }, - "emit" => '. sub%XXM %binop', - "latency" => 4, - "units" => [ "SSE" ], - "mode" => "mode_E", -}, +# not commutative operations -"xDiv" => { - "irn_flags" => "R", - "comment" => "construct SSE Div: Div(a, b) = a / b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3 !in_r4" ] }, - "outs" => [ "res", "M" ], - "emit" => '. div%XXM %binop', - "latency" => 16, - "units" => [ "SSE" ], +xAndNot => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4 !in_r5" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + am => "source,binary", + emit => '. andnp%XSD %binop', + latency => 3, + units => [ "SSE" ], + mode => $mode_xmm +}, + +xSub => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4" ] }, + ins => [ "base", "index", "mem", "minuend", "subtrahend" ], + am => "source,binary", + emit => '. sub%XXM %binop', + latency => 4, + units => [ "SSE" ], + mode => $mode_xmm +}, + +xDiv => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "in_r4 !in_r5", "none" ] }, + ins => [ "base", "index", "mem", "dividend", "divisor" ], + am => "source,binary", + outs => [ "res", "M" ], + emit => '. div%XXM %binop', + latency => 16, + units => [ "SSE" ], }, # other operations -"xCmp" => { - "irn_flags" => "R", - "comment" => "construct SSE Compare: Cmp(a, b) == a = a cmp b", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3 !in_r4" ] }, - "latency" => 3, - "units" => [ "SSE" ], - "mode" => "mode_E", -}, - -"xCondJmp" => { - "op_flags" => "L|X|Y", - "comment" => "construct conditional jump: UCOMIS A, B && JMPxx LABEL", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "none", "none" ] }, - "outs" => [ "false", "true" ], - "latency" => 5, - "units" => [ "SSE" ], -}, - -"xConst" => { - "op_flags" => "c", - "irn_flags" => "R", - "comment" => "represents a SSE constant", - "reg_req" => { "out" => [ "xmm" ] }, - "emit" => '. mov%XXM $%C, %D1', - "latency" => 2, - "units" => [ "SSE" ], - "mode" => "mode_E", +Ucomi => { + irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm", "xmm" ], out => [ "eflags" ] }, + ins => [ "base", "index", "mem", "left", "right" ], + outs => [ "flags" ], + am => "source,binary", + attr => "int ins_permuted", + init_attr => "attr->data.ins_permuted = ins_permuted;", + emit => ' .ucomi%XXM %binop', + latency => 3, + units => [ "SSE" ], + mode => $mode_flags, + modified_flags => 1, }, # Load / Store -"xLoad" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "construct SSE Load: Load(ptr, mem) = LD ptr", - "reg_req" => { "in" => [ "gp", "gp", "none" ], "out" => [ "xmm", "none" ] }, - "emit" => '. mov%XXM %AM, %D1', - "outs" => [ "res", "M" ], - "latency" => 2, - "units" => [ "SSE" ], -}, - -"xStore" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "construct Store: Store(ptr, val, mem) = ST ptr,val", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "none" ] }, - "emit" => '. mov%XXM %binop', - "latency" => 2, - "units" => [ "SSE" ], - "mode" => "mode_M", -}, - -"xStoreSimple" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "construct Store without index: Store(ptr, val, mem) = ST ptr,val", - "reg_req" => { "in" => [ "gp", "xmm", "none" ] }, - "emit" => '. mov%XXM %S2, %AM', - "latency" => 2, - "units" => [ "SSE" ], - "mode" => "mode_M", -}, - -"l_X87toSSE" => { - "op_flags" => "L|F", - "comment" => "construct: transfer a value from x87 FPU into a SSE register", - "cmp_attr" => "return 1;", - "arity" => 3, -}, - -"l_SSEtoX87" => { - "op_flags" => "L|F", - "comment" => "construct: transfer a value from SSE register to x87 FPU", - "cmp_attr" => "return 1;", - "arity" => 3, -}, - -"GetST0" => { - "op_flags" => "L|F", - "irn_flags" => "I", - "state" => "exc_pinned", - "comment" => "store ST0 onto stack", - "reg_req" => { "in" => [ "gp", "gp", "none" ] }, - "emit" => '. fstp%XM %AM', - "latency" => 4, - "units" => [ "SSE" ], - "mode" => "mode_M", -}, - -"SetST0" => { - "op_flags" => "L|F", - "irn_flags" => "I", - "state" => "exc_pinned", - "comment" => "load ST0 from stack", - "reg_req" => { "in" => [ "gp", "none" ], "out" => [ "vf0", "none" ] }, - "emit" => '. fld%M %AM', - "outs" => [ "res", "M" ], - "latency" => 2, - "units" => [ "SSE" ], +xLoad => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "xmm", "none", "none" ] }, + ins => [ "base", "index", "mem" ], + outs => [ "res", "M", "X_exc" ], + emit => '. mov%XXM %AM, %D0', + attr => "ir_mode *load_mode", + init_attr => "attr->ls_mode = load_mode;", + latency => 0, + units => [ "SSE" ], +}, + +xStore => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm" ], out => [ "none", "none" ] }, + ins => [ "base", "index", "mem", "val" ], + outs => [ "M", "X_exc" ], + emit => '. mov%XXM %S3, %AM', + latency => 0, + units => [ "SSE" ], + mode => "mode_M", +}, + +xStoreSimple => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => '. mov%XXM %S3, %AM', + latency => 0, + units => [ "SSE" ], + mode => "mode_M", +}, + +CvtSI2SS => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "xmm" ] }, + ins => [ "base", "index", "mem", "val" ], + am => "source,unary", + emit => '. cvtsi2ss %unop3, %D0', + latency => 2, + units => [ "SSE" ], + mode => $mode_xmm +}, + +CvtSI2SD => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "xmm" ] }, + ins => [ "base", "index", "mem", "val" ], + am => "source,unary", + emit => '. cvtsi2sd %unop3, %D0', + latency => 2, + units => [ "SSE" ], + mode => $mode_xmm +}, + + +l_LLtoFloat => { + op_flags => "L|F", + cmp_attr => "return 1;", + ins => [ "val_high", "val_low" ], +}, + +l_FloattoLL => { + op_flags => "L|F", + cmp_attr => "return 1;", + ins => [ "val" ], + outs => [ "res_high", "res_low" ], }, # CopyB -"CopyB" => { - "op_flags" => "F|H", - "state" => "pinned", - "comment" => "implements a memcopy: CopyB(dst, src, size, mem) == memcpy(dst, src, size)", - "reg_req" => { "in" => [ "edi", "esi", "ecx", "none" ], "out" => [ "edi", "esi", "ecx", "none" ] }, - "outs" => [ "DST", "SRC", "CNT", "M" ], - "units" => [ "GP" ], -}, - -"CopyB_i" => { - "op_flags" => "F|H", - "state" => "pinned", - "comment" => "implements a memcopy: CopyB(dst, src, mem) == memcpy(dst, src, attr(size))", - "reg_req" => { "in" => [ "edi", "esi", "none" ], "out" => [ "edi", "esi", "none" ] }, - "outs" => [ "DST", "SRC", "M" ], - "units" => [ "GP" ], +CopyB => { + op_flags => "F|H", + state => "pinned", + reg_req => { in => [ "edi", "esi", "ecx", "none" ], out => [ "edi", "esi", "ecx", "none" ] }, + outs => [ "DST", "SRC", "CNT", "M" ], + attr_type => "ia32_copyb_attr_t", + attr => "unsigned size", + units => [ "GP" ], + latency => 3, +# we don't care about this flag, so no need to mark this node +# modified_flags => [ "DF" ] +}, + +CopyB_i => { + op_flags => "F|H", + state => "pinned", + reg_req => { in => [ "edi", "esi", "none" ], out => [ "edi", "esi", "none" ] }, + outs => [ "DST", "SRC", "M" ], + attr_type => "ia32_copyb_attr_t", + attr => "unsigned size", + units => [ "GP" ], + latency => 3, +# we don't care about this flag, so no need to mark this node +# modified_flags => [ "DF" ] }, # Conversions -"Conv_I2I" => { - "reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3", "none" ] }, - "comment" => "construct Conv Int -> Int", - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Conv_I2I8Bit" => { - "reg_req" => { "in" => [ "gp", "gp", "eax ebx ecx edx", "none" ], "out" => [ "in_r3", "none" ] }, - "comment" => "construct Conv Int -> Int", - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"Conv_I2FP" => { - "reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "xmm", "none" ] }, - "comment" => "construct Conv Int -> Floating Point", - "latency" => 10, - "units" => [ "SSE" ], - "mode" => "mode_E", -}, - -"Conv_FP2I" => { - "reg_req" => { "in" => [ "gp", "gp", "xmm", "none" ], "out" => [ "gp", "none" ] }, - "comment" => "construct Conv Floating Point -> Int", - "latency" => 10, - "units" => [ "SSE" ], - "mode" => "mode_Iu", -}, - -"Conv_FP2FP" => { - "reg_req" => { "in" => [ "gp", "gp", "xmm", "none" ], "out" => [ "xmm", "none" ] }, - "comment" => "construct Conv Floating Point -> Floating Point", - "latency" => 8, - "units" => [ "SSE" ], - "mode" => "mode_E", -}, - -"CmpCMov" => { - "irn_flags" => "R", - "comment" => "construct Conditional Move: CMov(sel, a, b) == sel ? a : b", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp" ], "out" => [ "in_r4" ] }, - "latency" => 2, - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"PsiCondCMov" => { - "irn_flags" => "R", - "comment" => "check if Psi condition tree evaluates to true and move result accordingly", - "reg_req" => { "in" => [ "gp", "gp", "gp" ], "out" => [ "in_r3" ] }, - "latency" => 2, - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"xCmpCMov" => { - "irn_flags" => "R", - "comment" => "construct Conditional Move: SSE Compare + int CMov ", - "reg_req" => { "in" => [ "xmm", "xmm", "gp", "gp" ], "out" => [ "in_r4" ] }, - "latency" => 5, - "units" => [ "SSE" ], - "mode" => "mode_Iu", -}, - -"vfCmpCMov" => { - "irn_flags" => "R", - "comment" => "construct Conditional Move: x87 Compare + int CMov", - "reg_req" => { "in" => [ "vfp", "vfp", "gp", "gp" ], "out" => [ "in_r4" ] }, - "latency" => 10, - "units" => [ "VFP" ], - "mode" => "mode_Iu", -}, - -"CmpSet" => { - "irn_flags" => "R", - "comment" => "construct Set: Set(sel) == sel ? 1 : 0", - "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "eax ebx ecx edx" ] }, - "latency" => 2, - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"PsiCondSet" => { - "irn_flags" => "R", - "comment" => "check if Psi condition tree evaluates to true and set result accordingly", - "reg_req" => { "in" => [ "gp" ], "out" => [ "eax ebx ecx edx" ] }, - "latency" => 2, - "units" => [ "GP" ], - "mode" => "mode_Iu", -}, - -"xCmpSet" => { - "irn_flags" => "R", - "comment" => "construct Set: SSE Compare + int Set", - "reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "eax ebx ecx edx" ] }, - "latency" => 5, - "units" => [ "SSE" ], - "mode" => "mode_Iu", -}, - -"vfCmpSet" => { - "irn_flags" => "R", - "comment" => "construct Set: x87 Compare + int Set", - "reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "eax ebx ecx edx" ] }, - "latency" => 10, - "units" => [ "VFP" ], - "mode" => "mode_Iu", -}, - -"vfCMov" => { - "irn_flags" => "R", - "comment" => "construct x87 Conditional Move: vfCMov(sel, a, b) = sel ? a : b", - "reg_req" => { "in" => [ "vfp", "vfp", "vfp", "vfp" ], "out" => [ "vfp" ] }, - "latency" => 10, - "units" => [ "VFP" ], - "mode" => "mode_E", +Conv_I2I => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "gp", "none" ] }, + ins => [ "base", "index", "mem", "val" ], + am => "source,unary", + units => [ "GP" ], + latency => 1, + attr => "ir_mode *smaller_mode", + init_attr => "attr->ls_mode = smaller_mode;", + mode => $mode_gp, +}, + +Conv_I2I8Bit => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx" ], out => [ "gp", "none" ] }, + ins => [ "base", "index", "mem", "val" ], + am => "source,unary", + units => [ "GP" ], + latency => 1, + attr => "ir_mode *smaller_mode", + init_attr => "attr->ls_mode = smaller_mode;", + mode => $mode_gp, +}, + +Conv_I2FP => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "xmm", "none" ] }, + ins => [ "base", "index", "mem", "val" ], + am => "source,unary", + latency => 10, + units => [ "SSE" ], + mode => $mode_xmm, +}, + +Conv_FP2I => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm" ], out => [ "gp", "none" ] }, + ins => [ "base", "index", "mem", "val" ], + am => "source,unary", + latency => 10, + units => [ "SSE" ], + mode => $mode_gp, +}, + +Conv_FP2FP => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm" ], out => [ "xmm", "none" ] }, + ins => [ "base", "index", "mem", "val" ], + am => "source,unary", + latency => 8, + units => [ "SSE" ], + mode => $mode_xmm, }, #----------------------------------------------------------# @@ -1296,260 +1854,267 @@ if (get_ia32_immop_type(node) == ia32_ImmNone) { # |_| |_|\___/ \__,_|\___||___/ # #----------------------------------------------------------# -"vfadd" => { - "irn_flags" => "R", - "comment" => "virtual fp Add: Add(a, b) = Add(b, a) = a + b", - "reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfmul" => { - "irn_flags" => "R", - "comment" => "virtual fp Mul: Mul(a, b) = Mul(b, a) = a * b", - "reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"l_vfmul" => { - "op_flags" => "C", - "cmp_attr" => "return 1;", - "comment" => "lowered virtual fp Mul: Mul(a, b) = Mul(b, a) = a * b", - "arity" => 2, -}, - -"vfsub" => { - "irn_flags" => "R", - "comment" => "virtual fp Sub: Sub(a, b) = a - b", - "reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"l_vfsub" => { - "cmp_attr" => "return 1;", - "comment" => "lowered virtual fp Sub: Sub(a, b) = a - b", - "arity" => 2, -}, - -"vfdiv" => { - "comment" => "virtual fp Div: Div(a, b) = a / b", - "reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "vfp" ] }, - "outs" => [ "res", "M" ], - "latency" => 20, - "units" => [ "VFP" ], -}, - -"l_vfdiv" => { - "cmp_attr" => "return 1;", - "comment" => "lowered virtual fp Div: Div(a, b) = a / b", - "outs" => [ "res", "M" ], - "arity" => 2, -}, - -"vfprem" => { - "comment" => "virtual fp Rem: Rem(a, b) = a - Q * b (Q is integer)", - "reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "vfp" ] }, - "latency" => 20, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"l_vfprem" => { - "cmp_attr" => "return 1;", - "comment" => "lowered virtual fp Rem: Rem(a, b) = a - Q * b (Q is integer)", - "arity" => 2, -}, - -"vfabs" => { - "irn_flags" => "R", - "comment" => "virtual fp Abs: Abs(a) = |a|", - "reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] }, - "latency" => 2, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfchs" => { - "irn_flags" => "R", - "comment" => "virtual fp Chs: Chs(a) = -a", - "reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] }, - "latency" => 2, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfsin" => { - "irn_flags" => "R", - "comment" => "virtual fp Sin: Sin(a) = sin(a)", - "reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] }, - "latency" => 150, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfcos" => { - "irn_flags" => "R", - "comment" => "virtual fp Cos: Cos(a) = cos(a)", - "reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] }, - "latency" => 150, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfsqrt" => { - "irn_flags" => "R", - "comment" => "virtual fp Sqrt: Sqrt(a) = a ^ 0.5", - "reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] }, - "latency" => 30, - "units" => [ "VFP" ], - "mode" => "mode_E", +# rematerialisation disabled for all float nodes for now, because the fpcw +# handler runs before spilling and we might end up with wrong fpcw then + +vfadd => { +# irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "vfp", "vfp", "fpcw" ], out => [ "vfp" ] }, + ins => [ "base", "index", "mem", "left", "right", "fpcw" ], + am => "source,binary", + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfmul => { +# irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "vfp", "vfp", "fpcw" ], out => [ "vfp" ] }, + ins => [ "base", "index", "mem", "left", "right", "fpcw" ], + am => "source,binary", + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfsub => { +# irn_flags => "R", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "vfp", "vfp", "fpcw" ], out => [ "vfp" ] }, + ins => [ "base", "index", "mem", "minuend", "subtrahend", "fpcw" ], + am => "source,binary", + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfdiv => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "vfp", "vfp", "fpcw" ], out => [ "vfp", "none" ] }, + ins => [ "base", "index", "mem", "dividend", "divisor", "fpcw" ], + am => "source,binary", + outs => [ "res", "M" ], + latency => 20, + units => [ "VFP" ], + attr_type => "ia32_x87_attr_t", +}, + +vfprem => { + reg_req => { in => [ "vfp", "vfp", "fpcw" ], out => [ "vfp" ] }, + ins => [ "left", "right", "fpcw" ], + latency => 20, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfabs => { + irn_flags => "R", + reg_req => { in => [ "vfp"], out => [ "vfp" ] }, + ins => [ "value" ], + latency => 2, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfchs => { + irn_flags => "R", + reg_req => { in => [ "vfp"], out => [ "vfp" ] }, + ins => [ "value" ], + latency => 2, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", }, # virtual Load and Store -"vfld" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "virtual fp Load: Load(ptr, mem) = LD ptr -> reg", - "reg_req" => { "in" => [ "gp", "gp", "none" ], "out" => [ "vfp", "none" ] }, - "outs" => [ "res", "M" ], - "latency" => 2, - "units" => [ "VFP" ], -}, - -"vfst" => { - "op_flags" => "L|F", - "state" => "exc_pinned", - "comment" => "virtual fp Store: Store(ptr, val, mem) = ST ptr,val", - "reg_req" => { "in" => [ "gp", "gp", "vfp", "none" ] }, - "latency" => 2, - "units" => [ "VFP" ], - "mode" => "mode_M", +vfld => { + irn_flags => "R", + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "vfp", "none", "none" ] }, + ins => [ "base", "index", "mem" ], + outs => [ "res", "M", "X_exc" ], + attr => "ir_mode *load_mode", + init_attr => "attr->attr.ls_mode = load_mode;", + latency => 2, + units => [ "VFP" ], + attr_type => "ia32_x87_attr_t", +}, + +vfst => { + irn_flags => "R", + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "vfp" ], out => [ "none", "none" ] }, + ins => [ "base", "index", "mem", "val" ], + outs => [ "M", "X_exc" ], + attr => "ir_mode *store_mode", + init_attr => "attr->attr.ls_mode = store_mode;", + latency => 2, + units => [ "VFP" ], + mode => "mode_M", + attr_type => "ia32_x87_attr_t", }, # Conversions -"vfild" => { - "comment" => "virtual fp integer Load: Load(ptr, mem) = iLD ptr -> reg", - "reg_req" => { "in" => [ "gp", "gp", "none" ], "out" => [ "vfp", "none" ] }, - "outs" => [ "res", "M" ], - "latency" => 4, - "units" => [ "VFP" ], +vfild => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "vfp", "none" ] }, + outs => [ "res", "M" ], + ins => [ "base", "index", "mem" ], + latency => 4, + units => [ "VFP" ], + attr_type => "ia32_x87_attr_t", }, -"l_vfild" => { - "cmp_attr" => "return 1;", - "comment" => "lowered virtual fp integer Load: Load(ptr, mem) = iLD ptr -> reg", - "outs" => [ "res", "M" ], - "arity" => 2, +vfist => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "vfp", "fpcw" ] }, + ins => [ "base", "index", "mem", "val", "fpcw" ], + latency => 4, + units => [ "VFP" ], + mode => "mode_M", + attr_type => "ia32_x87_attr_t", }, -"vfist" => { - "comment" => "virtual fp integer Store: Store(ptr, val, mem) = iST ptr,val", - "reg_req" => { "in" => [ "gp", "gp", "vfp", "none" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_M", -}, - -"l_vfist" => { - "cmp_attr" => "return 1;", - "comment" => "lowered virtual fp integer Store: Store(ptr, val, mem) = iST ptr,val", - "arity" => 3, - "mode" => "mode_M", +# SSE3 fisttp instruction +vfisttp => { + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "vfp" ], out => [ "in_r4", "none" ]}, + ins => [ "base", "index", "mem", "val" ], + outs => [ "res", "M" ], + latency => 4, + units => [ "VFP" ], + attr_type => "ia32_x87_attr_t", }, # constants -"vfldz" => { - "irn_flags" => "R", - "comment" => "virtual fp Load 0.0: Ld 0.0 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfld1" => { - "irn_flags" => "R", - "comment" => "virtual fp Load 1.0: Ld 1.0 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfldpi" => { - "irn_flags" => "R", - "comment" => "virtual fp Load pi: Ld pi -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfldln2" => { - "irn_flags" => "R", - "comment" => "virtual fp Load ln 2: Ld ln 2 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfldlg2" => { - "irn_flags" => "R", - "comment" => "virtual fp Load lg 2: Ld lg 2 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfldl2t" => { - "irn_flags" => "R", - "comment" => "virtual fp Load ld 10: Ld ld 10 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfldl2e" => { - "irn_flags" => "R", - "comment" => "virtual fp Load ld e: Ld ld e -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "latency" => 4, - "units" => [ "VFP" ], - "mode" => "mode_E", -}, - -"vfConst" => { - "op_flags" => "c", - "irn_flags" => "R", -# "init_attr" => " set_ia32_ls_mode(res, mode);", - "comment" => "represents a virtual floating point constant", - "reg_req" => { "out" => [ "vfp" ] }, - "latency" => 3, - "units" => [ "VFP" ], - "mode" => "mode_E", +vfldz => { + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfld1 => { + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfldpi => { + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfldln2 => { + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfldlg2 => { + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfldl2t => { + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", +}, + +vfldl2e => { + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + latency => 4, + units => [ "VFP" ], + mode => "mode_E", + attr_type => "ia32_x87_attr_t", }, # other -"vfCondJmp" => { - "op_flags" => "L|X|Y", - "comment" => "represents a virtual floating point compare", - "reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "none", "none", "eax" ] }, - "outs" => [ "false", "true", "temp_reg_eax" ], - "latency" => 10, - "units" => [ "VFP" ], +vFucomFnstsw => { +# we can't allow to rematerialize this node so we don't have +# accidently produce Phi(Fucom, Fucom(ins_permuted)) +# irn_flags => "R", + reg_req => { in => [ "vfp", "vfp" ], out => [ "eax" ] }, + ins => [ "left", "right" ], + outs => [ "flags" ], + attr => "int ins_permuted", + init_attr => "attr->attr.data.ins_permuted = ins_permuted;", + latency => 3, + units => [ "VFP" ], + attr_type => "ia32_x87_attr_t", + mode => $mode_gp +}, + +vFucomi => { + irn_flags => "R", + reg_req => { in => [ "vfp", "vfp" ], out => [ "eflags" ] }, + ins => [ "left", "right" ], + outs => [ "flags" ], + attr => "int ins_permuted", + init_attr => "attr->attr.data.ins_permuted = ins_permuted;", + latency => 3, + units => [ "VFP" ], + attr_type => "ia32_x87_attr_t", + mode => $mode_gp +}, + +vFtstFnstsw => { +# irn_flags => "R", + reg_req => { in => [ "vfp" ], out => [ "eax" ] }, + ins => [ "left" ], + outs => [ "flags" ], + attr => "int ins_permuted", + init_attr => "attr->attr.data.ins_permuted = ins_permuted;", + latency => 3, + units => [ "VFP" ], + attr_type => "ia32_x87_attr_t", + mode => $mode_gp +}, + +Sahf => { + irn_flags => "R", + reg_req => { in => [ "eax" ], out => [ "eflags" ] }, + ins => [ "val" ], + outs => [ "flags" ], + emit => '. sahf', + latency => 1, + units => [ "GP" ], + mode => $mode_flags, }, #------------------------------------------------------------------------# @@ -1560,360 +2125,485 @@ if (get_ia32_immop_type(node) == ia32_ImmNone) { # /_/\_\___//_/ |_| |_|\___/ \__,_|\__| |_| |_|\___/ \__,_|\___||___/ # #------------------------------------------------------------------------# -"fadd" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 Add: Add(a, b) = Add(b, a) = a + b", - "reg_req" => { }, - "emit" => '. fadd%XM %x87_binop', -}, - -"faddp" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 Add: Add(a, b) = Add(b, a) = a + b", - "reg_req" => { }, - "emit" => '. faddp %x87_binop', -}, - -"fmul" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Mul: Mul(a, b) = Mul(b, a) = a + b", - "reg_req" => { }, - "emit" => '. fmul%XM %x87_binop', -}, - -"fmulp" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Mul: Mul(a, b) = Mul(b, a) = a + b", - "reg_req" => { }, - "emit" => '. fmulp %x87_binop',, -}, - -"fsub" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Sub: Sub(a, b) = a - b", - "reg_req" => { }, - "emit" => '. fsub%XM %x87_binop', -}, - -"fsubp" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Sub: Sub(a, b) = a - b", - "reg_req" => { }, - "emit" => '. fsubp %x87_binop', -}, - -"fsubr" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "irn_flags" => "R", - "comment" => "x87 fp SubR: SubR(a, b) = b - a", - "reg_req" => { }, - "emit" => '. fsubr%XM %x87_binop', -}, - -"fsubrp" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "irn_flags" => "R", - "comment" => "x87 fp SubR: SubR(a, b) = b - a", - "reg_req" => { }, - "emit" => '. fsubrp %x87_binop', -}, - -"fprem" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Rem: Rem(a, b) = a - Q * b (Q is integer)", - "reg_req" => { }, - "emit" => '. fprem1', +# Note: gas is strangely buggy: fdivrp and fdivp as well as fsubrp and fsubp +# are swapped, we work this around in the emitter... + +fadd => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fadd%XM %x87_binop', + latency => 4, + attr_type => "ia32_x87_attr_t", +}, + +faddp => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. faddp%XM %x87_binop', + latency => 4, + attr_type => "ia32_x87_attr_t", +}, + +fmul => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fmul%XM %x87_binop', + latency => 4, + attr_type => "ia32_x87_attr_t", +}, + +fmulp => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fmulp%XM %x87_binop',, + latency => 4, + attr_type => "ia32_x87_attr_t", +}, + +fsub => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fsub%XM %x87_binop', + latency => 4, + attr_type => "ia32_x87_attr_t", +}, + +fsubp => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, +# see note about gas bugs + emit => '. fsubrp%XM %x87_binop', + latency => 4, + attr_type => "ia32_x87_attr_t", +}, + +fsubr => { + state => "exc_pinned", + rd_constructor => "NONE", + irn_flags => "R", + reg_req => { }, + emit => '. fsubr%XM %x87_binop', + latency => 4, + attr_type => "ia32_x87_attr_t", +}, + +fsubrp => { + state => "exc_pinned", + rd_constructor => "NONE", + irn_flags => "R", + reg_req => { }, +# see note about gas bugs + emit => '. fsubp%XM %x87_binop', + latency => 4, + attr_type => "ia32_x87_attr_t", +}, + +fprem => { + rd_constructor => "NONE", + reg_req => { }, + emit => '. fprem1', + latency => 20, + attr_type => "ia32_x87_attr_t", }, # this node is just here, to keep the simulator running # we can omit this when a fprem simulation function exists -"fpremp" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Rem: Rem(a, b) = a - Q * b (Q is integer)", - "reg_req" => { }, - "emit" => '. fprem1', -}, - -"fdiv" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Div: Div(a, b) = a / b", - "reg_req" => { }, - "emit" => '. fdiv%XM %x87_binop', -}, - -"fdivp" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Div: Div(a, b) = a / b", - "reg_req" => { }, - "emit" => '. fdivp %x87_binop', -}, - -"fdivr" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp DivR: DivR(a, b) = b / a", - "reg_req" => { }, - "emit" => '. fdivr%XM %x87_binop', -}, - -"fdivrp" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp DivR: DivR(a, b) = b / a", - "reg_req" => { }, - "emit" => '. fdivrp %x87_binop', -}, - -"fabs" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Abs: Abs(a) = |a|", - "reg_req" => { }, - "emit" => '. fabs', -}, - -"fchs" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Chs: Chs(a) = -a", - "reg_req" => { }, - "emit" => '. fchs', -}, - -"fsin" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Sin: Sin(a) = sin(a)", - "reg_req" => { }, - "emit" => '. fsin', -}, - -"fcos" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Cos: Cos(a) = cos(a)", - "reg_req" => { }, - "emit" => '. fcos', -}, - -"fsqrt" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp Sqrt: Sqrt(a) = a ^ 0.5", - "reg_req" => { }, - "emit" => '. fsqrt $', +fpremp => { + rd_constructor => "NONE", + reg_req => { }, + emit => '. fprem1\n'. + '. fstp %X0', + latency => 20, + attr_type => "ia32_x87_attr_t", +}, + +fdiv => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fdiv%XM %x87_binop', + latency => 20, + attr_type => "ia32_x87_attr_t", +}, + +fdivp => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, +# see note about gas bugs + emit => '. fdivrp%XM %x87_binop', + latency => 20, + attr_type => "ia32_x87_attr_t", +}, + +fdivr => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fdivr%XM %x87_binop', + latency => 20, + attr_type => "ia32_x87_attr_t", +}, + +fdivrp => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, +# see note about gas bugs + emit => '. fdivp%XM %x87_binop', + latency => 20, + attr_type => "ia32_x87_attr_t", +}, + +fabs => { + rd_constructor => "NONE", + reg_req => { }, + emit => '. fabs', + latency => 4, + attr_type => "ia32_x87_attr_t", +}, + +fchs => { + op_flags => "R|K", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fchs', + latency => 4, + attr_type => "ia32_x87_attr_t", }, # x87 Load and Store -"fld" => { - "rd_constructor" => "NONE", - "op_flags" => "R|L|F", - "state" => "exc_pinned", - "comment" => "x87 fp Load: Load(ptr, mem) = LD ptr -> reg", - "reg_req" => { }, - "emit" => '. fld%XM %AM', -}, - -"fst" => { - "rd_constructor" => "NONE", - "op_flags" => "R|L|F", - "state" => "exc_pinned", - "comment" => "x87 fp Store: Store(ptr, val, mem) = ST ptr,val", - "reg_req" => { }, - "emit" => '. fst%XM %AM', - "mode" => "mode_M", -}, - -"fstp" => { - "rd_constructor" => "NONE", - "op_flags" => "R|L|F", - "state" => "exc_pinned", - "comment" => "x87 fp Store: Store(ptr, val, mem) = ST ptr,val", - "reg_req" => { }, - "emit" => '. fstp%XM %AM', - "mode" => "mode_M", +fld => { + rd_constructor => "NONE", + op_flags => "R|L|F", + state => "exc_pinned", + reg_req => { }, + emit => '. fld%XM %AM', + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fst => { + rd_constructor => "NONE", + op_flags => "R|L|F", + state => "exc_pinned", + reg_req => { }, + emit => '. fst%XM %AM', + mode => "mode_M", + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fstp => { + rd_constructor => "NONE", + op_flags => "R|L|F", + state => "exc_pinned", + reg_req => { }, + emit => '. fstp%XM %AM', + mode => "mode_M", + attr_type => "ia32_x87_attr_t", + latency => 2, }, # Conversions -"fild" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp integer Load: Load(ptr, mem) = iLD ptr -> reg", - "reg_req" => { }, - "emit" => '. fild%XM %AM', +fild => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fild%M %AM', + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fist => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fist%M %AM', + mode => "mode_M", + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fistp => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fistp%M %AM', + mode => "mode_M", + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +# SSE3 firsttp instruction +fisttp => { + state => "exc_pinned", + rd_constructor => "NONE", + reg_req => { }, + emit => '. fisttp%M %AM', + mode => "mode_M", + attr_type => "ia32_x87_attr_t", + latency => 2, }, -"fist" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp integer Store: Store(ptr, val, mem) = iST ptr,val", - "reg_req" => { }, - "emit" => '. fist%M %AM', - "mode" => "mode_M", -}, +# constants -"fistp" => { - "op_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "x87 fp integer Store: Store(ptr, val, mem) = iST ptr,val", - "reg_req" => { }, - "emit" => '. fistp%M %AM', - "mode" => "mode_M", +fldz => { + op_flags => "R|c|K", + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + emit => '. fldz', + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fld1 => { + op_flags => "R|c|K", + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + emit => '. fld1', + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fldpi => { + op_flags => "R|c|K", + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + emit => '. fldpi', + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fldln2 => { + op_flags => "R|c|K", + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + emit => '. fldln2', + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fldlg2 => { + op_flags => "R|c|K", + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + emit => '. fldlg2', + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fldl2t => { + op_flags => "R|c|K", + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + emit => '. fldll2t', + attr_type => "ia32_x87_attr_t", + latency => 2, +}, + +fldl2e => { + op_flags => "R|c|K", + irn_flags => "R", + reg_req => { out => [ "vfp" ] }, + emit => '. fldl2e', + attr_type => "ia32_x87_attr_t", + latency => 2, }, -# constants +# fxch, fpush, fpop +# Note that it is NEVER allowed to do CSE on these nodes +# Moreover, note the virtual register requierements! -"fldz" => { - "op_flags" => "R|c", - "irn_flags" => "R", - "comment" => "x87 fp Load 0.0: Ld 0.0 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "emit" => '. fldz', +fxch => { + op_flags => "R|K", + reg_req => { }, + cmp_attr => "return 1;", + emit => '. fxch %X0', + attr_type => "ia32_x87_attr_t", + mode => "mode_ANY", + latency => 1, +}, + +fpush => { + op_flags => "R|K", + reg_req => {}, + cmp_attr => "return 1;", + emit => '. fld %X0', + attr_type => "ia32_x87_attr_t", + mode => "mode_ANY", + latency => 1, +}, + +fpushCopy => { + reg_req => { in => [ "vfp"], out => [ "vfp" ] }, + cmp_attr => "return 1;", + emit => '. fld %X0', + attr_type => "ia32_x87_attr_t", + latency => 1, +}, + +fpop => { + op_flags => "K", + reg_req => { }, + cmp_attr => "return 1;", + emit => '. fstp %X0', + attr_type => "ia32_x87_attr_t", + mode => "mode_ANY", + latency => 1, +}, + +ffreep => { + op_flags => "K", + reg_req => { }, + cmp_attr => "return 1;", + emit => '. ffreep %X0', + attr_type => "ia32_x87_attr_t", + mode => "mode_ANY", + latency => 1, +}, + +emms => { + op_flags => "K", + reg_req => { }, + cmp_attr => "return 1;", + emit => '. emms', + attr_type => "ia32_x87_attr_t", + mode => "mode_ANY", + latency => 3, +}, + +femms => { + op_flags => "K", + reg_req => { }, + cmp_attr => "return 1;", + emit => '. femms', + attr_type => "ia32_x87_attr_t", + mode => "mode_ANY", + latency => 3, }, -"fld1" => { - "op_flags" => "R|c", - "irn_flags" => "R", - "comment" => "x87 fp Load 1.0: Ld 1.0 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "emit" => '. fld1', -}, +# compare -"fldpi" => { - "op_flags" => "R|c", - "irn_flags" => "R", - "comment" => "x87 fp Load pi: Ld pi -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "emit" => '. fldpi', +FucomFnstsw => { + reg_req => { }, + emit => ". fucom %X1\n". + ". fnstsw %%ax", + attr_type => "ia32_x87_attr_t", + latency => 2, }, -"fldln2" => { - "op_flags" => "R|c", - "irn_flags" => "R", - "comment" => "x87 fp Load ln 2: Ld ln 2 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "emit" => '. fldln2', +FucompFnstsw => { + reg_req => { }, + emit => ". fucomp %X1\n". + ". fnstsw %%ax", + attr_type => "ia32_x87_attr_t", + latency => 2, }, -"fldlg2" => { - "op_flags" => "R|c", - "irn_flags" => "R", - "comment" => "x87 fp Load lg 2: Ld lg 2 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "emit" => '. fldlg2', +FucomppFnstsw => { + reg_req => { }, + emit => ". fucompp\n". + ". fnstsw %%ax", + attr_type => "ia32_x87_attr_t", + latency => 2, }, -"fldl2t" => { - "op_flags" => "R|c", - "irn_flags" => "R", - "comment" => "x87 fp Load ld 10: Ld ld 10 -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "emit" => '. fldll2t', +Fucomi => { + reg_req => { }, + emit => '. fucomi %X1', + attr_type => "ia32_x87_attr_t", + latency => 1, }, -"fldl2e" => { - "op_flags" => "R|c", - "irn_flags" => "R", - "comment" => "x87 fp Load ld e: Ld ld e -> reg", - "reg_req" => { "out" => [ "vfp" ] }, - "emit" => '. fldl2e', +Fucompi => { + reg_req => { }, + emit => '. fucompi %X1', + attr_type => "ia32_x87_attr_t", + latency => 1, }, -"fldConst" => { - "op_flags" => "R|c", - "irn_flags" => "R", - "rd_constructor" => "NONE", - "comment" => "represents a x87 constant", - "reg_req" => { "out" => [ "vfp" ] }, - "emit" => '. fld $%C', +FtstFnstsw => { + reg_req => { }, + emit => ". ftst\n". + ". fnstsw %%ax", + attr_type => "ia32_x87_attr_t", + latency => 2, }, -# fxch, fpush, fpop -# Note that it is NEVER allowed to do CSE on these nodes -# Moreover, note the virtual register requierements! -"fxch" => { - "op_flags" => "R|K", - "comment" => "x87 stack exchange", - "reg_req" => { }, - "cmp_attr" => "return 1;", - "emit" => '. fxch %X1', -}, +# -------------------------------------------------------------------------------- # +# ____ ____ _____ _ _ # +# / ___/ ___|| ____| __ _____ ___| |_ ___ _ __ _ __ ___ __| | ___ ___ # +# \___ \___ \| _| \ \ / / _ \/ __| __/ _ \| '__| | '_ \ / _ \ / _` |/ _ \/ __| # +# ___) |__) | |___ \ V / __/ (__| || (_) | | | | | | (_) | (_| | __/\__ \ # +# |____/____/|_____| \_/ \___|\___|\__\___/|_| |_| |_|\___/ \__,_|\___||___/ # +# # +# -------------------------------------------------------------------------------- # -"fpush" => { - "op_flags" => "R|K", - "comment" => "x87 stack push", - "reg_req" => {}, - "cmp_attr" => "return 1;", - "emit" => '. fld %X1', -}, -"fpushCopy" => { - "op_flags" => "R", - "comment" => "x87 stack push", - "reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] }, - "cmp_attr" => "return 1;", - "emit" => '. fld %X1', -}, +# Spilling and reloading of SSE registers, hardcoded, not generated # -"fpop" => { - "op_flags" => "R|K", - "comment" => "x87 stack pop", - "reg_req" => { }, - "cmp_attr" => "return 1;", - "emit" => '. fstp %X1', +xxLoad => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none" ], out => [ "xmm", "none" ] }, + emit => '. movdqu %D0, %AM', + outs => [ "res", "M" ], + units => [ "SSE" ], + latency => 1, }, -# compare - -"fcomJmp" => { - "op_flags" => "L|X|Y", - "comment" => "floating point compare", - "reg_req" => { }, +xxStore => { + op_flags => "L|F", + state => "exc_pinned", + reg_req => { in => [ "gp", "gp", "none", "xmm" ] }, + ins => [ "base", "index", "mem", "val" ], + emit => '. movdqu %binop', + units => [ "SSE" ], + latency => 1, + mode => "mode_M", }, -"fcompJmp" => { - "op_flags" => "L|X|Y", - "comment" => "floating point compare and pop", - "reg_req" => { }, -}, +); # end of %nodes -"fcomppJmp" => { - "op_flags" => "L|X|Y", - "comment" => "floating point compare and pop twice", - "reg_req" => { }, -}, +# Include the generated SIMD node specification written by the SIMD optimization +$my_script_name = dirname($myname) . "/../ia32/ia32_simd_spec.pl"; +unless ($return = do $my_script_name) { + warn "couldn't parse $my_script_name: $@" if $@; + warn "couldn't do $my_script_name: $!" unless defined $return; + warn "couldn't run $my_script_name" unless $return; +} -"fcomrJmp" => { - "op_flags" => "L|X|Y", - "comment" => "floating point compare reverse", - "reg_req" => { }, -}, +# Transform some attributes +foreach my $op (keys(%nodes)) { + my $node = $nodes{$op}; + my $op_attr_init = $node->{op_attr_init}; -"fcomrpJmp" => { - "op_flags" => "L|X|Y", - "comment" => "floating point compare reverse and pop", - "reg_req" => { }, -}, + if(defined($op_attr_init)) { + $op_attr_init .= "\n\t"; + } else { + $op_attr_init = ""; + } -"fcomrppJmp" => { - "op_flags" => "L|X|Y", - "comment" => "floating point compare reverse and pop twice", - "reg_req" => { }, -}, + if(!defined($node->{latency})) { + if($op =~ m/^l_/) { + $node->{latency} = 0; + } else { + die("Latency missing for op $op"); + } + } + $op_attr_init .= "attr->latency = ".$node->{latency} . ";"; -); # end of %nodes + $node->{op_attr_init} = $op_attr_init; +} + +print "";