X-Git-Url: http://nsz.repo.hu/git/?a=blobdiff_plain;f=ir%2Fbe%2Fia32%2Fia32_spec.pl;h=f830e75738206abc0ba57dcec364be23b7d325f6;hb=4be314ae49c26d8b9084200a570bc4a1285bef50;hp=372c59362ec8ce71804177270d5104a163ae3e82;hpb=180f84a1ee29292ad83809121983bf5ab4a2dc42;p=libfirm diff --git a/ir/be/ia32/ia32_spec.pl b/ir/be/ia32/ia32_spec.pl index 372c59362..f830e7573 100644 --- a/ir/be/ia32/ia32_spec.pl +++ b/ir/be/ia32/ia32_spec.pl @@ -28,10 +28,15 @@ $arch = "ia32"; # 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" +# 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", # }, # @@ -51,6 +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: @@ -144,14 +151,14 @@ $arch = "ia32"; { 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 => "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" } @@ -197,6 +204,7 @@ $arch = "ia32"; SB1 => "${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);", @@ -211,10 +219,8 @@ $arch = "ia32"; XXM => "${arch}_emit_xmm_mode_suffix(node);", XSD => "${arch}_emit_xmm_mode_suffix_s(node);", AM => "${arch}_emit_am(node);", - unop3 => "${arch}_emit_unop(node, 3);", - unop4 => "${arch}_emit_unop(node, 4);", - unop5 => "${arch}_emit_unop(node, 5);", - DAM0 => "${arch}_emit_am_or_dest_register(node, 0);", + 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 => "${arch}_emit_cmp_suffix_node(node, 0);", @@ -246,22 +252,10 @@ sub ia32_custom_init_attr { } if(defined($node->{am})) { my $am = $node->{am}; - if($am eq "full,binary") { - $res .= "\tset_ia32_am_support(res, ia32_am_Full, ia32_am_binary);"; - } elsif($am eq "full,unary") { - $res .= "\tset_ia32_am_support(res, ia32_am_Full, ia32_am_unary);"; - } elsif($am eq "source,unary") { - $res .= "\tset_ia32_am_support(res, ia32_am_Source, ia32_am_unary);"; + if($am eq "source,unary") { + $res .= "\tset_ia32_am_support(res, ia32_am_unary);"; } elsif($am eq "source,binary") { - $res .= "\tset_ia32_am_support(res, ia32_am_Source, ia32_am_binary);"; - } elsif($am eq "dest,unary") { - $res .= "\tset_ia32_am_support(res, ia32_am_Dest, ia32_am_unary);"; - } elsif($am eq "dest,binary") { - $res .= "\tset_ia32_am_support(res, ia32_am_Dest, ia32_am_binary);"; - } elsif($am eq "dest,ternary") { - $res .= "\tset_ia32_am_support(res, ia32_am_Dest, ia32_am_ternary);"; - } elsif($am eq "source,ternary") { - $res .= "\tset_ia32_am_support(res, ia32_am_Source, ia32_am_ternary);"; + $res .= "\tset_ia32_am_support(res, ia32_am_binary);"; } elsif($am eq "none") { # nothing to do } else { @@ -279,44 +273,46 @@ sub ia32_custom_init_attr { $custom_init_attr_func = \&ia32_custom_init_attr; %init_attr = ( - ia32_attr_t => "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);", - ia32_x87_attr_t => - "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". - "\tinit_ia32_x87_attributes(res);", ia32_asm_attr_t => "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". "\tinit_ia32_x87_attributes(res);". "\tinit_ia32_asm_attributes(res);", - ia32_immediate_attr_t => + ia32_attr_t => + "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);", + ia32_condcode_attr_t => "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". - "\tinit_ia32_immediate_attributes(res, symconst, symconst_sign, offset);", + "\tinit_ia32_condcode_attributes(res, pnc);", ia32_copyb_attr_t => "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". "\tinit_ia32_copyb_attributes(res, size);", - ia32_condcode_attr_t => + ia32_immediate_attr_t => "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". - "\tinit_ia32_condcode_attributes(res, pnc);", + "\tinit_ia32_immediate_attributes(res, symconst, symconst_sign, offset);", + ia32_x87_attr_t => + "\tinit_ia32_attributes(res, flags, in_reqs, out_reqs, exec_units, n_res);\n". + "\tinit_ia32_x87_attributes(res);", ); %compare_attr = ( - ia32_attr_t => "ia32_compare_nodes_attr", - ia32_x87_attr_t => "ia32_compare_x87_attr", ia32_asm_attr_t => "ia32_compare_asm_attr", - ia32_immediate_attr_t => "ia32_compare_immediate_attr", - ia32_copyb_attr_t => "ia32_compare_copyb_attr", + ia32_attr_t => "ia32_compare_nodes_attr", ia32_condcode_attr_t => "ia32_compare_condcode_attr", + ia32_copyb_attr_t => "ia32_compare_copyb_attr", + ia32_immediate_attr_t => "ia32_compare_immediate_attr", + ia32_x87_attr_t => "ia32_compare_x87_attr", ); %operands = ( ); -$mode_xmm = "mode_E"; -$mode_gp = "mode_Iu"; -$mode_flags = "mode_Iu"; -$mode_fpcw = "mode_fpcw"; -$status_flags = [ "CF", "PF", "AF", "ZF", "SF", "OF" ]; -$fpcw_flags = [ "FP_IM", "FP_DM", "FP_ZM", "FP_OM", "FP_UM", "FP_PM", - "FP_PC0", "FP_PC1", "FP_RC0", "FP_RC1", "FP_X" ]; +$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 = ( @@ -327,6 +323,7 @@ Immediate => { 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, }, @@ -343,6 +340,7 @@ Asm => { modified_flags => 1, }, +# "allocates" a free register ProduceVal => { op_flags => "c", irn_flags => "R", @@ -375,7 +373,7 @@ Add => { ins => [ "base", "index", "mem", "left", "right" ], outs => [ "res", "flags", "M" ], emit => '. add%M %binop', - am => "full,binary", + am => "source,binary", units => [ "GP" ], latency => 1, mode => $mode_gp, @@ -413,7 +411,7 @@ Adc => { ins => [ "base", "index", "mem", "left", "right", "eflags" ], outs => [ "res", "flags", "M" ], emit => '. adc%M %binop', - am => "full,binary", + am => "source,binary", units => [ "GP" ], latency => 1, mode => $mode_gp, @@ -433,13 +431,13 @@ l_Adc => { Mul => { # we should not rematrialize this node. It produces 2 results and has - # very strict constrains + # very strict constraints state => "exc_pinned", reg_req => { in => [ "gp", "gp", "none", "eax", "gp" ], - out => [ "eax", "edx", "none" ] }, - ins => [ "base", "index", "mem", "val_high", "val_low" ], + out => [ "eax", "flags", "edx", "none" ] }, + ins => [ "base", "index", "mem", "left", "right" ], emit => '. mul%M %unop4', - outs => [ "res_low", "res_high", "M" ], + outs => [ "res_low", "flags", "res_high", "M" ], am => "source,binary", latency => 10, units => [ "GP" ], @@ -448,10 +446,10 @@ Mul => { l_Mul => { # we should not rematrialize this node. It produces 2 results and has - # very strict constrains + # very strict constraints op_flags => "C", cmp_attr => "return 1;", - outs => [ "EAX", "EDX", "M" ], + outs => [ "EAX", "flags", "EDX", "M" ], arity => 2 }, @@ -476,7 +474,7 @@ IMul1OP => { state => "exc_pinned", reg_req => { in => [ "gp", "gp", "none", "eax", "gp" ], out => [ "eax", "edx", "none" ] }, - ins => [ "base", "index", "mem", "val_high", "val_low" ], + ins => [ "base", "index", "mem", "left", "right" ], emit => '. imul%M %unop4', outs => [ "res_low", "res_high", "M" ], am => "source,binary", @@ -500,7 +498,7 @@ And => { ins => [ "base", "index", "mem", "left", "right" ], outs => [ "res", "flags", "M" ], op_modes => "commutative | am | immediate | mode_neutral", - am => "full,binary", + am => "source,binary", emit => '. and%M %binop', units => [ "GP" ], latency => 1, @@ -539,7 +537,7 @@ Or => { out => [ "in_r4 in_r5", "flags", "none" ] }, ins => [ "base", "index", "mem", "left", "right" ], outs => [ "res", "flags", "M" ], - am => "full,binary", + am => "source,binary", emit => '. or%M %binop', units => [ "GP" ], latency => 1, @@ -578,7 +576,7 @@ Xor => { out => [ "in_r4 in_r5", "flags", "none" ] }, ins => [ "base", "index", "mem", "left", "right" ], outs => [ "res", "flags", "M" ], - am => "full,binary", + am => "source,binary", emit => '. xor%M %binop', units => [ "GP" ], latency => 1, @@ -617,9 +615,9 @@ Sub => { state => "exc_pinned", reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ], out => [ "in_r4", "flags", "none" ] }, - ins => [ "base", "index", "mem", "left", "right" ], + ins => [ "base", "index", "mem", "minuend", "subtrahend" ], outs => [ "res", "flags", "M" ], - am => "full,binary", + am => "source,binary", emit => '. sub%M %binop', units => [ "GP" ], latency => 1, @@ -631,7 +629,7 @@ SubMem => { irn_flags => "R", state => "exc_pinned", reg_req => { in => [ "gp", "gp", "none", "gp" ], out => [ "none" ] }, - ins => [ "base", "index", "mem", "val" ], + ins => [ "base", "index", "mem", "subtrahend" ], emit => '. sub%M %SI3, %AM', units => [ "GP" ], latency => 1, @@ -643,7 +641,7 @@ SubMem8Bit => { irn_flags => "R", state => "exc_pinned", reg_req => { in => [ "gp", "gp", "none", "eax ebx ecx edx" ], out => [ "none" ] }, - ins => [ "base", "index", "mem", "val" ], + ins => [ "base", "index", "mem", "subtrahend" ], emit => '. sub%M %SB3, %AM', units => [ "GP" ], latency => 1, @@ -655,9 +653,9 @@ Sbb => { 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" ], + ins => [ "base", "index", "mem", "minuend", "subtrahend", "eflags" ], outs => [ "res", "flags", "M" ], - am => "full,binary", + am => "source,binary", emit => '. sbb%M %binop', units => [ "GP" ], latency => 1, @@ -667,23 +665,23 @@ Sbb => { l_Sub => { reg_req => { in => [ "none", "none" ], out => [ "none" ] }, - ins => [ "left", "right" ], + ins => [ "minuend", "subtrahend" ], }, l_Sbb => { reg_req => { in => [ "none", "none", "none" ], out => [ "none" ] }, - ins => [ "left", "right", "eflags" ], + ins => [ "minuend", "subtrahend", "eflags" ], }, IDiv => { op_flags => "F|L", state => "exc_pinned", - reg_req => { in => [ "gp", "gp", "none", "eax", "edx", "gp" ], + reg_req => { in => [ "gp", "gp", "none", "gp", "eax", "edx" ], out => [ "eax", "flags", "none", "edx", "none" ] }, - ins => [ "base", "index", "mem", "left_low", "left_high", "right" ], + ins => [ "base", "index", "mem", "divisor", "dividend_low", "dividend_high" ], outs => [ "div_res", "flags", "M", "mod_res", "X_exc" ], - am => "source,ternary", - emit => ". idiv%M %unop5", + am => "source,unary", + emit => ". idiv%M %unop3", latency => 25, units => [ "GP" ], modified_flags => $status_flags @@ -692,12 +690,12 @@ IDiv => { Div => { op_flags => "F|L", state => "exc_pinned", - reg_req => { in => [ "gp", "gp", "none", "eax", "edx", "gp" ], + reg_req => { in => [ "gp", "gp", "none", "gp", "eax", "edx" ], out => [ "eax", "flags", "none", "edx", "none" ] }, - ins => [ "base", "index", "mem", "left_low", "left_high", "right" ], + ins => [ "base", "index", "mem", "divisor", "dividend_low", "dividend_high" ], outs => [ "div_res", "flags", "M", "mod_res", "X_exc" ], - am => "source,ternary", - emit => ". div%M %unop5", + am => "source,unary", + emit => ". div%M %unop3", latency => 25, units => [ "GP" ], modified_flags => $status_flags @@ -913,7 +911,7 @@ NegMem => { Minus64Bit => { irn_flags => "R", - reg_req => { in => [ "gp", "gp" ], out => [ "in_r1", "gp" ] }, + reg_req => { in => [ "gp", "gp" ], out => [ "in_r1", "in_r2" ] }, outs => [ "low_res", "high_res" ], units => [ "GP" ], latency => 3, @@ -925,12 +923,13 @@ 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 => [ "OF", "SF", "ZF", "AF", "PF" ] + modified_flags => $status_flags_wo_cf }, IncMem => { @@ -942,19 +941,20 @@ IncMem => { units => [ "GP" ], mode => "mode_M", latency => 1, - modified_flags => [ "OF", "SF", "ZF", "AF", "PF" ] + 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 => [ "OF", "SF", "ZF", "AF", "PF" ] + modified_flags => $status_flags_wo_cf }, DecMem => { @@ -966,7 +966,7 @@ DecMem => { units => [ "GP" ], mode => "mode_M", latency => 1, - modified_flags => [ "OF", "SF", "ZF", "AF", "PF" ] + modified_flags => $status_flags_wo_cf }, Not => { @@ -1017,9 +1017,10 @@ Stc => { Cmp => { irn_flags => "R", state => "exc_pinned", - reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ] , out => [ "flags" ] }, + reg_req => { in => [ "gp", "gp", "none", "gp", "gp" ], + out => [ "flags", "none", "none" ] }, ins => [ "base", "index", "mem", "left", "right" ], - outs => [ "eflags" ], + outs => [ "eflags", "unused", "M" ], am => "source,binary", emit => '. cmp%M %binop', attr => "int ins_permuted, int cmp_unsigned", @@ -1145,7 +1146,7 @@ SwitchJmp => { reg_req => { in => [ "gp" ], out => [ "none" ] }, mode => "mode_T", attr_type => "ia32_condcode_attr_t", - attr => "pn_Cmp pnc", + attr => "long pnc", latency => 3, units => [ "BRANCH" ], modified_flags => $status_flags, @@ -1174,9 +1175,18 @@ Const => { 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", + op_flags => "c|NB", irn_flags => "I", reg_req => { out => [ "gp_UKNWN" ] }, units => [], @@ -1187,7 +1197,7 @@ Unknown_GP => { Unknown_VFP => { state => "pinned", - op_flags => "c", + op_flags => "c|NB", irn_flags => "I", reg_req => { out => [ "vfp_UKNWN" ] }, units => [], @@ -1199,18 +1209,18 @@ Unknown_VFP => { Unknown_XMM => { state => "pinned", - op_flags => "c", + op_flags => "c|NB", irn_flags => "I", reg_req => { out => [ "xmm_UKNWN" ] }, units => [], emit => "", latency => 0, - mode => "mode_E" + mode => $mode_xmm }, NoReg_GP => { state => "pinned", - op_flags => "c", + op_flags => "c|NB|NI", irn_flags => "I", reg_req => { out => [ "gp_NOREG" ] }, units => [], @@ -1221,7 +1231,7 @@ NoReg_GP => { NoReg_VFP => { state => "pinned", - op_flags => "c", + op_flags => "c|NB|NI", irn_flags => "I", reg_req => { out => [ "vfp_NOREG" ] }, units => [], @@ -1233,7 +1243,7 @@ NoReg_VFP => { NoReg_XMM => { state => "pinned", - op_flags => "c", + op_flags => "c|NB|NI", irn_flags => "I", reg_req => { out => [ "xmm_NOREG" ] }, units => [], @@ -1289,7 +1299,7 @@ FnstCWNOP => { Cltd => { # we should not rematrialize this node. It has very strict constraints. reg_req => { in => [ "eax", "edx" ], out => [ "edx" ] }, - ins => [ "val", "globbered" ], + ins => [ "val", "clobbered" ], emit => '. cltd', latency => 1, mode => $mode_gp, @@ -1312,21 +1322,6 @@ Load => { units => [ "GP" ], }, -l_Load => { - op_flags => "L|F", - cmp_attr => "return 1;", - outs => [ "res", "M" ], - arity => 2, -}, - -l_Store => { - op_flags => "L|F", - cmp_attr => "return 1;", - state => "exc_pinned", - arity => 3, - mode => "mode_M", -}, - Store => { op_flags => "L|F", state => "exc_pinned", @@ -1366,22 +1361,31 @@ Lea => { Push => { state => "exc_pinned", - reg_req => { in => [ "gp", "gp", "none", "esp", "gp" ], out => [ "esp", "none" ] }, + reg_req => { in => [ "gp", "gp", "none", "gp", "esp" ], out => [ "esp", "none" ] }, ins => [ "base", "index", "mem", "val", "stack" ], - emit => '. push%M %unop4', + emit => '. push%M %unop3', outs => [ "stack:I|S", "M" ], - am => "source,binary", + am => "source,unary", latency => 2, units => [ "GP" ], }, Pop => { state => "exc_pinned", - reg_req => { in => [ "gp", "gp", "none", "esp" ], out => [ "gp", "none", "none", "esp" ] }, - emit => '. pop%M %DAM0', + 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" ], - am => "dest,unary", + outs => [ "unused0", "M", "unused1", "stack:I|S" ], + emit => '. pop%M %AM', latency => 3, # Pop is more expensive than Push on Athlon units => [ "GP" ], }, @@ -1395,7 +1399,7 @@ Enter => { }, Leave => { - reg_req => { in => [ "esp", "ebp" ], out => [ "ebp", "esp" ] }, + reg_req => { in => [ "ebp" ], out => [ "ebp", "esp" ] }, emit => '. leave', outs => [ "frame:I", "stack:I|S" ], latency => 3, @@ -1410,7 +1414,7 @@ AddSP => { am => "source,binary", emit => '. addl %binop', latency => 1, - outs => [ "stack:S", "M" ], + outs => [ "stack:I|S", "M" ], units => [ "GP" ], modified_flags => $status_flags }, @@ -1429,6 +1433,14 @@ SubSP => { 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" ] }, @@ -1436,6 +1448,17 @@ LdTls => { 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 +}, #-----------------------------------------------------------------------------# # _____ _____ ______ __ _ _ _ # @@ -1446,13 +1469,73 @@ LdTls => { # |_____/_____/|______| |_| |_|\___/ \__,_|\__| |_| |_|\___/ \__,_|\___||___/ # #-----------------------------------------------------------------------------# +# produces a 0/+0.0 xZero => { irn_flags => "R", reg_req => { out => [ "xmm" ] }, emit => '. xorp%XSD %D0, %D0', latency => 3, units => [ "SSE" ], - mode => "mode_E", + 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 }, # commutative operations @@ -1466,7 +1549,7 @@ xAdd => { emit => '. add%XXM %binop', latency => 4, units => [ "SSE" ], - mode => "mode_E", + mode => $mode_xmm }, xMul => { @@ -1478,7 +1561,7 @@ xMul => { emit => '. mul%XXM %binop', latency => 4, units => [ "SSE" ], - mode => "mode_E", + mode => $mode_xmm }, xMax => { @@ -1490,7 +1573,7 @@ xMax => { emit => '. max%XXM %binop', latency => 2, units => [ "SSE" ], - mode => "mode_E", + mode => $mode_xmm }, xMin => { @@ -1502,7 +1585,7 @@ xMin => { emit => '. min%XXM %binop', latency => 2, units => [ "SSE" ], - mode => "mode_E", + mode => $mode_xmm }, xAnd => { @@ -1514,7 +1597,7 @@ xAnd => { emit => '. andp%XSD %binop', latency => 3, units => [ "SSE" ], - mode => "mode_E", + mode => $mode_xmm }, xOr => { @@ -1526,7 +1609,7 @@ xOr => { emit => '. orp%XSD %binop', latency => 3, units => [ "SSE" ], - mode => "mode_E", + mode => $mode_xmm }, xXor => { @@ -1538,7 +1621,7 @@ xXor => { emit => '. xorp%XSD %binop', latency => 3, units => [ "SSE" ], - mode => "mode_E", + mode => $mode_xmm }, # not commutative operations @@ -1552,26 +1635,26 @@ xAndNot => { emit => '. andnp%XSD %binop', latency => 3, units => [ "SSE" ], - mode => "mode_E", + 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", "left", "right" ], + ins => [ "base", "index", "mem", "minuend", "subtrahend" ], am => "source,binary", emit => '. sub%XXM %binop', latency => 4, units => [ "SSE" ], - mode => "mode_E", + 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", "left", "right" ], + ins => [ "base", "index", "mem", "dividend", "divisor" ], am => "source,binary", outs => [ "res", "M" ], emit => '. div%XXM %binop', @@ -1734,7 +1817,7 @@ Conv_I2FP => { am => "source,unary", latency => 10, units => [ "SSE" ], - mode => "mode_E", + mode => $mode_xmm, }, Conv_FP2I => { @@ -1754,7 +1837,7 @@ Conv_FP2FP => { am => "source,unary", latency => 8, units => [ "SSE" ], - mode => "mode_E", + mode => $mode_xmm, }, #----------------------------------------------------------# @@ -1802,7 +1885,7 @@ vfsub => { # irn_flags => "R", state => "exc_pinned", reg_req => { in => [ "gp", "gp", "none", "vfp", "vfp", "fpcw" ], out => [ "vfp" ] }, - ins => [ "base", "index", "mem", "left", "right", "fpcw" ], + ins => [ "base", "index", "mem", "minuend", "subtrahend", "fpcw" ], am => "source,binary", latency => 4, units => [ "VFP" ], @@ -1813,7 +1896,7 @@ vfsub => { vfdiv => { state => "exc_pinned", reg_req => { in => [ "gp", "gp", "none", "vfp", "vfp", "fpcw" ], out => [ "vfp", "none" ] }, - ins => [ "base", "index", "mem", "left", "right", "fpcw" ], + ins => [ "base", "index", "mem", "dividend", "divisor", "fpcw" ], am => "source,binary", outs => [ "res", "M" ], latency => 20, @@ -1893,12 +1976,6 @@ vfild => { attr_type => "ia32_x87_attr_t", }, -l_vfild => { - cmp_attr => "return 1;", - outs => [ "res", "M" ], - arity => 2, -}, - vfist => { state => "exc_pinned", reg_req => { in => [ "gp", "gp", "none", "vfp", "fpcw" ] }, @@ -1909,11 +1986,15 @@ vfist => { attr_type => "ia32_x87_attr_t", }, -l_vfist => { - cmp_attr => "return 1;", +# SSE3 fisttp instruction +vfisttp => { state => "exc_pinned", - arity => 3, - mode => "mode_M", + 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", }, @@ -2262,6 +2343,17 @@ fistp => { 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, +}, + # constants fldz => {