# $Id$
# This is the specification for the ia32 assembler Firm-operations
+$new_emit_syntax = 1;
+
# the cpu architecture (ia32, ia64, mips, sparc, ppc, ...)
$arch = "ia32";
-# this string marks the beginning of a comment in emit
-$comment_string = "/*";
-
-# the number of additional opcodes you want to register
-#$additional_opcodes = 0;
-
# The node description is done as a perl hash initializer with the
# following structure:
#
#
# outs: if a node defines more than one output, the names of the projections
# nodes having outs having automatically the mode mode_T
+# One can also annotate some flags for each out, additional to irn_flags.
+# They are separated from name with a colon ':', and concatenated by pipe '|'
+# Only I and S are available at the moment (same meaning as in irn_flags).
+# example: [ "frame:I", "stack:I|S", "M" ]
#
# comment: OPTIONAL comment for the node constructor
#
# 1 - caller save (register must be saved by the caller of a function)
# 2 - callee save (register must be saved by the called function)
# 4 - ignore (do not assign this register)
+# 8 - emitter can choose an arbitrary register of this class
+# 16 - the register is a virtual one
# 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" => "r11", "type" => 1 },
-# { "name" => "r12", "type" => 1 },
-# { "name" => "r13", "type" => 1 },
-# { "name" => "r14", "type" => 1 },
-# { "name" => "r15", "type" => 1 },
-# { "name" => "r16", "type" => 1 },
-# { "name" => "r17", "type" => 1 },
-# { "name" => "r18", "type" => 1 },
-# { "name" => "r19", "type" => 1 },
-# { "name" => "r20", "type" => 1 },
-# { "name" => "r21", "type" => 1 },
-# { "name" => "r22", "type" => 1 },
-# { "name" => "r23", "type" => 1 },
-# { "name" => "r24", "type" => 1 },
-# { "name" => "r25", "type" => 1 },
-# { "name" => "r26", "type" => 1 },
-# { "name" => "r27", "type" => 1 },
-# { "name" => "r28", "type" => 1 },
-# { "name" => "r29", "type" => 1 },
-# { "name" => "r30", "type" => 1 },
-# { "name" => "r31", "type" => 1 },
-# { "name" => "r32", "type" => 1 },
- { "name" => "ebp", "type" => 2 },
- { "name" => "esp", "type" => 4 },
- { "name" => "gp_NOREG", "type" => 6 }, # we need a dummy register for NoReg nodes
- { "name" => "gp_UKNWN", "type" => 6 }, # we need a dummy register for Unknown nodes
- { "mode" => "mode_P" }
- ],
- "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" => 6 }, # we need a dummy register for NoReg nodes
- { "name" => "xmm_UKNWN", "type" => 6 }, # we need a dummy register for Unknown nodes
- { "mode" => "mode_D" }
- ],
- "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" => 6 }, # we need a dummy register for NoReg nodes
- { "name" => "vfp_UKNWN", "type" => 6 }, # 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" }
- ]
+ "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" }
+ ],
+ "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" }
+ ],
+ "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" }
+ ],
+ "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" }
+ ],
+ "fp_cw" => [ # the floating point control word
+ { "name" => "fpcw", "type" => 0 },
+ { "mode" => "mode_Hu" },
+ ],
); # %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" ],
+); # %cpu
+
+%vliw = (
+ "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);",
+ "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);",
+);
+
#--------------------------------------------------#
# _ #
# (_) #
# |_| #
#--------------------------------------------------#
+$default_cmp_attr = "return ia32_compare_attr(attr_a, attr_b);";
+
%operands = (
);
"Add" => {
"irn_flags" => "R",
"comment" => "construct Add: Add(a, b) = Add(b, a) = a + b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. add %ia32_emit_binop /* Add(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. addl %binop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
-"AddC" => {
- "comment" => "construct Add with Carry: AddC(a, b) = Add(b, a) = a + b + carry",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
+"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" => '. adc %ia32_emit_binop /* AddC(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "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;\n",
+ "cmp_attr" => "return 1;",
"comment" => "construct lowered Add: Add(a, b) = Add(b, a) = a + b",
"arity" => 2,
},
-"l_AddC" => {
+"l_Adc" => {
"op_flags" => "C",
- "cmp_attr" => " return 1;\n",
- "comment" => "construct lowered Add with Carry: AddC(a, b) = Add(b, a) = a + b + carry",
+ "cmp_attr" => "return 1;",
+ "comment" => "construct lowered Add with Carry: Adc(a, b) = Adc(b, a) = a + b + carry",
"arity" => 2,
},
-"MulS" => {
+"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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "eax", "gp", "none" ], "out" => [ "eax", "edx" ] },
- "emit" => '. mul %ia32_emit_unop /* Mul(%A1, %A2) -> %D1 */',
+ "reg_req" => { "in" => [ "gp", "gp", "eax", "gp", "none" ], "out" => [ "eax", "edx", "none" ] },
+ "emit" => '. mull %unop',
"outs" => [ "EAX", "EDX", "M" ],
"latency" => 10,
+ "units" => [ "GP" ],
},
-"l_MulS" => {
+"l_Mul" => {
# we should not rematrialize this node. It produces 2 results and has
# very strict constrains
"op_flags" => "C",
- "cmp_attr" => " return 1;\n",
- "comment" => "construct lowered MulS: MulS(a, b) = MulS(b, a) = a * b",
+ "cmp_attr" => "return 1;",
+ "comment" => "construct lowered MulS: Mul(a, b) = Mul(b, a) = a * b",
"outs" => [ "EAX", "EDX", "M" ],
"arity" => 2
},
-"Mul" => {
+"IMul" => {
"irn_flags" => "R",
"comment" => "construct Mul: Mul(a, b) = Mul(b, a) = a * b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. imul %ia32_emit_binop /* Mul(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. imull %binop',
"latency" => 5,
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
-"l_Mul" => {
- "op_flags" => "C",
- "cmp_attr" => " return 1;\n",
- "comment" => "construct lowered Mul: Mul(a, b) = Mul(b, a) = a * b",
- "arity" => 2
-},
-
-# Mulh is an exception from the 4 INs with AM because the target is always EAX:EDX
-"Mulh" => {
- # we should not rematrialize this node. It produces 2 results and has
- # very strict constrains
- "comment" => "construct Mul: Mul(a, b) = Mul(b, a) = a * b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "eax", "gp", "none" ], "out" => [ "eax", "edx" ] },
- "emit" => '. imul %ia32_emit_unop /* Mulh(%A1, %A2) -> %D1 */',
+"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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. and %ia32_emit_binop /* And(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. andl %binop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
"Or" => {
"irn_flags" => "R",
"comment" => "construct Or: Or(a, b) = Or(b, a) = a OR b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. or %ia32_emit_binop /* Or(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. orl %binop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
-"Eor" => {
+"Xor" => {
"irn_flags" => "R",
- "comment" => "construct Eor: Eor(a, b) = Eor(b, a) = a EOR b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
+ "comment" => "construct Xor: Xor(a, b) = Xor(b, a) = a EOR b",
"reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. xor %ia32_emit_binop /* Xor(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. xorl %binop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
-"l_Eor" => {
+"l_Xor" => {
"op_flags" => "C",
- "cmp_attr" => " return 1;\n",
- "comment" => "construct lowered Eor: Eor(a, b) = Eor(b, a) = a EOR b",
+ "cmp_attr" => "return 1;",
+ "comment" => "construct lowered Xor: Xor(a, b) = Xor(b, a) = a XOR b",
"arity" => 2
},
-"Max" => {
- "irn_flags" => "R",
- "comment" => "construct Max: Max(a, b) = Max(b, a) = a > b ? a : b",
- "reg_req" => { "in" => [ "gp", "gp" ], "out" => [ "in_r1" ] },
- "emit" =>
-'2. cmp %S1, %S2 /* prepare Max (%S1 - %S2), (%A1, %A2) */
- if (mode_is_signed(get_irn_mode(n))) {
-4. cmovl %D1, %S2 /* %S1 is less %S2 */
- }
- else {
-4. cmovb %D1, %S2 /* %S1 is below %S2 */
- }
-',
- "latency" => 2,
-},
-
-"Min" => {
- "irn_flags" => "R",
- "comment" => "construct Min: Min(a, b) = Min(b, a) = a < b ? a : b",
- "reg_req" => { "in" => [ "gp", "gp" ], "out" => [ "in_r1" ] },
- "emit" =>
-'2. cmp %S1, %S2 /* prepare Min (%S1 - %S2), (%A1, %A2) */
- if (mode_is_signed(get_irn_mode(n))) {
-2. cmovg %D1, %S2 /* %S1 is greater %S2 */
- }
- else {
-2. cmova %D1, %S2, %D1 /* %S1 is above %S2 */
- }
-',
- "latency" => 2,
-},
-
# not commutative operations
"Sub" => {
"irn_flags" => "R",
"comment" => "construct Sub: Sub(a, b) = a - b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. sub %ia32_emit_binop /* Sub(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. subl %binop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
-"SubC" => {
+"Sbb" => {
"comment" => "construct Sub with Carry: SubC(a, b) = a - b - carry",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "in_r3 !in_r4" ] },
- "emit" => '. sbb %ia32_emit_binop /* SubC(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "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;\n",
+ "cmp_attr" => "return 1;",
"comment" => "construct lowered Sub: Sub(a, b) = a - b",
"arity" => 2,
},
-"l_SubC" => {
- "cmp_attr" => " return 1;\n",
+"l_Sbb" => {
+ "cmp_attr" => "return 1;",
"comment" => "construct lowered Sub with Carry: SubC(a, b) = a - b - carry",
"arity" => 2,
},
-"DivMod" => {
+"IDiv" => {
"op_flags" => "F|L",
"state" => "exc_pinned",
- "reg_req" => { "in" => [ "eax", "gp", "edx", "none" ], "out" => [ "eax", "edx" ] },
+ "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;",
- "cmp_attr" => " return attr_a->data.op_flav != attr_b->data.op_flav;\n",
- "emit" =>
-' if (mode_is_signed(get_ia32_res_mode(n))) {
-4. idiv %S2 /* signed DivMod(%S1, %S2) -> %D1, (%A1, %A2, %A3) */
- }
- else {
-4. div %S2 /* unsigned DivMod(%S1, %S2) -> %D1, (%A1, %A2, %A3) */
- }
-',
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx gp_NOREG", "none" ], "out" => [ "in_r3 !in_r4" ] },
- "emit" => '. shl %ia32_emit_binop /* Shl(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "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;\n",
+ "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)",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "ecx", "none" ], "out" => [ "!ecx in_r3 !in_r4" ] },
+ # 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(n) == ia32_ImmNone) {
- if (get_ia32_op_type(n) == ia32_AddrModeD) {
-4. shld %ia32_emit_am, %S4, %%cl /* ShlD(%A3, %A4, %A5) -> %D1 */
- }
- else {
-4. shld %S3, %S4, %%cl /* ShlD(%A3, %A4, %A5) -> %D1 */
- }
-}
-else {
- if (get_ia32_op_type(n) == ia32_AddrModeD) {
-4. shld %ia32_emit_am, %S4, %C /* ShlD(%A3, %A4, %A5) -> %D1 */
- }
- else {
-4. shld %S3, %S4, %C /* ShlD(%A3, %A4, %A5) -> %D1 */
- }
+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
+ }
}
',
- "outs" => [ "res", "M" ],
"latency" => 6,
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
"l_ShlD" => {
- "cmp_attr" => " return 1;\n",
+ "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
+ "arity" => 3,
},
"Shr" => {
"irn_flags" => "R",
"comment" => "construct Shr: Shr(a, b) = a >> b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx gp_NOREG", "none" ], "out" => [ "in_r3 !in_r4" ] },
- "emit" => '. shr %ia32_emit_binop /* Shr(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "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;\n",
+ "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)",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "ecx", "none" ], "out" => [ "!ecx in_r3 !in_r4" ] },
- "emit" =>
-'
-if (get_ia32_immop_type(n) == ia32_ImmNone) {
- if (get_ia32_op_type(n) == ia32_AddrModeD) {
-4. shrd %ia32_emit_am, %S4, %%cl /* ShrD(%A3, %A4, %A5) -> %D1 */
- }
- else {
-4. shrd %S3, %S4, %%cl /* ShrD(%A3, %A4, %A5) -> %D1 */
- }
-}
-else {
- if (get_ia32_op_type(n) == ia32_AddrModeD) {
-4. shrd %ia32_emit_am, %S4, %C /* ShrD(%A3, %A4, %A5) -> %D1 */
- }
- else {
-4. shrd %S3, %S4, %C /* ShrD(%A3, %A4, %A5) -> %D1 */
- }
+ # 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
+ }
}
',
- "outs" => [ "res", "M" ],
"latency" => 6,
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
"l_ShrD" => {
- "cmp_attr" => " return 1;\n",
+ "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
},
-"Shrs" => {
+"Sar" => {
"irn_flags" => "R",
"comment" => "construct Shrs: Shrs(a, b) = a >> b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx gp_NOREG", "none" ], "out" => [ "in_r3 !in_r4" ] },
- "emit" => '. sar %ia32_emit_binop /* Shrs(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx", "none" ], "out" => [ "in_r3 !in_r4" ] },
+ "emit" => '. sarl %binop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
-"l_Shrs" => {
- "cmp_attr" => " return 1;\n",
- "comment" => "construct lowered Shrs: Shrs(a, b) = a << b",
+"l_Sar" => {
+ "cmp_attr" => "return 1;",
+ "comment" => "construct lowered Sar: Sar(a, b) = a << b",
"arity" => 2
},
-"RotR" => {
+"Ror" => {
"irn_flags" => "R",
- "comment" => "construct RotR: RotR(a, b) = a ROTR b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "gp", "ecx", "none" ], "out" => [ "in_r3 !in_r4" ] },
- "emit" => '. ror %ia32_emit_binop /* RotR(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "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",
},
-"RotL" => {
+"Rol" => {
"irn_flags" => "R",
- "comment" => "construct RotL: RotL(a, b) = a ROTL b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
+ "comment" => "construct Rol: Rol(a, b) = a ROL b",
"reg_req" => { "in" => [ "gp", "gp", "gp", "ecx", "none" ], "out" => [ "in_r3 !in_r4" ] },
- "emit" => '. rol %ia32_emit_binop /* RotL(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. roll %binop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
# unary operations
-"Minus" => {
+"Neg" => {
"irn_flags" => "R",
"comment" => "construct Minus: Minus(a) = -a",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. neg %ia32_emit_unop /* Neg(%A1) -> %D1, (%A1) */',
- "outs" => [ "res", "M" ],
+ "emit" => '. negl %unop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
-"l_Minus" => {
- "cmp_attr" => " return 1;\n",
+"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++",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. inc %ia32_emit_unop /* Inc(%S1) -> %D1, (%A1) */',
- "outs" => [ "res", "M" ],
+ "emit" => '. incl %unop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
"Dec" => {
"irn_flags" => "R",
"comment" => "construct Decrement: Dec(a) = a--",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. dec %ia32_emit_unop /* Dec(%S1) -> %D1, (%A1) */',
- "outs" => [ "res", "M" ],
+ "emit" => '. decl %unop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
"Not" => {
"irn_flags" => "R",
"comment" => "construct Not: Not(a) = !a",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. not %ia32_emit_unop /* Not(%S1) -> %D1, (%A1) */',
- "outs" => [ "res", "M" ],
+ "emit" => '. notl %unop',
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
# other operations
"CondJmp" => {
"op_flags" => "L|X|Y",
"comment" => "construct conditional jump: CMP A, B && JMPxx LABEL",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"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" ] },
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"outs" => [ "false", "true" ],
"latency" => 3,
+ "units" => [ "BRANCH" ],
},
"CJmpAM" => {
"op_flags" => "L|X|Y",
"comment" => "construct conditional jump without CMP (replaces CondJmp): JMPxx LABEL",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp" ] },
+ "units" => [ "BRANCH" ],
},
"SwitchJmp" => {
"op_flags" => "L|X|Y",
"comment" => "construct switch",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp" ], "out" => [ "none" ] },
"latency" => 3,
+ "units" => [ "BRANCH" ],
},
"Const" => {
"op_flags" => "c",
"irn_flags" => "R",
"comment" => "represents an integer constant",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "none" ], "out" => [ "gp" ] },
+ "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"
},
-"Cdq" => {
+"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" => '. cdq /* sign extend EAX -> EDX:EAX, (%A1) */',
+ "emit" => '. cltd',
"outs" => [ "EAX", "EDX" ],
+ "units" => [ "GP" ],
},
# Load / Store
"op_flags" => "L|F",
"state" => "exc_pinned",
"comment" => "construct Load: Load(ptr, mem) = LD ptr -> reg",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "none" ], "out" => [ "gp" ] },
+ "reg_req" => { "in" => [ "gp", "gp", "none" ], "out" => [ "gp", "none" ] },
"latency" => 3,
- "emit" =>
-' if (get_mode_size_bits(get_ia32_ls_mode(n)) < 32) {
-4. mov%Mx %D1, %ia32_emit_am /* Load((%A1)) -> %D1 */
- }
- else {
-4. mov %D1, %ia32_emit_am /* Load((%A1)) -> %D1 */
- }
-',
+ "emit" => ". mov%SE%ME%.l %AM, %D1",
"outs" => [ "res", "M" ],
+ "units" => [ "GP" ],
},
"l_Load" => {
"op_flags" => "L|F",
- "cmp_attr" => " return 1;\n",
+ "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;\n",
+ "cmp_attr" => "return 1;",
"state" => "exc_pinned",
"comment" => "construct lowered Store: Store(ptr, val, mem) = ST ptr,val",
"arity" => 3,
- "outs" => [ "M" ],
+ "mode" => "mode_M",
},
"Store" => {
"op_flags" => "L|F",
"state" => "exc_pinned",
"comment" => "construct Store: Store(ptr, val, mem) = ST ptr,val",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "gp", "none" ] },
- "emit" => '. mov %ia32_emit_binop /* Store(%A3) -> (%A1) */',
- "outs" => [ "M" ],
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "eax ebx ecx edx gp_NOREG", "none" ] },
- "emit" => '. mov %ia32_emit_binop /* Store(%A3) -> (%A1) */',
- "outs" => [ "M" ],
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp" ], "out" => [ "in_r1" ] },
- "emit" => '. lea %D1, %ia32_emit_am /* LEA(%A1, %A2) */',
+ "emit" => '. leal %AM, %D1',
"latency" => 2,
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
"Push" => {
- # We don't set class modify_stack here (but we will do this on proj 0)
"comment" => "push on the stack",
- "reg_req" => { "in" => [ "gp", "gp", "gp", "esp", "none" ], "out" => [ "esp" ] },
- "emit" => '. push %ia32_emit_unop /* PUSH(%A1) */',
- "outs" => [ "stack", "M" ],
+ "reg_req" => { "in" => [ "gp", "gp", "gp", "esp", "none" ], "out" => [ "esp", "none" ] },
+ "emit" => '. pushl %unop',
+ "outs" => [ "stack:I|S", "M" ],
"latency" => 3,
+ "units" => [ "GP" ],
},
"Pop" => {
- # We don't set class modify stack here (but we will do this on proj 1)
"comment" => "pop a gp register from the stack",
- "reg_req" => { "in" => [ "gp", "gp", "esp", "none" ], "out" => [ "gp", "esp" ] },
- "emit" => '. pop %ia32_emit_unop /* POP(%A1) */',
- "outs" => [ "res", "stack", "M" ],
+ "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 /* Enter */',
- "outs" => [ "frame", "stack", "M" ],
+ "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 /* Leave */',
- "outs" => [ "frame", "stack", "M" ],
+ "emit" => '. leave',
+ "outs" => [ "frame:I", "stack:I|S" ],
"latency" => 3,
+ "units" => [ "GP" ],
},
"AddSP" => {
"irn_flags" => "I",
"comment" => "allocate space on stack",
- "reg_req" => { "in" => [ "esp", "gp" ], "out" => [ "esp", "none" ] },
- "outs" => [ "stack", "M" ],
+ "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" => [ "esp", "gp" ], "out" => [ "esp", "none" ] },
- "outs" => [ "stack", "M" ],
+ "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" ],
},
"xAdd" => {
"irn_flags" => "R",
"comment" => "construct SSE Add: Add(a, b) = Add(b, a) = a + b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. adds%M %ia32_emit_binop /* SSE Add(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. muls%M %ia32_emit_binop /* SSE Mul(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. maxs%M %ia32_emit_binop /* SSE Max(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. mins%M %ia32_emit_binop /* SSE Min(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. andp%M %ia32_emit_binop /* SSE And(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. andp%XXM %binop',
"latency" => 3,
+ "units" => [ "SSE" ],
+ "mode" => "mode_E",
},
"xOr" => {
"irn_flags" => "R",
"comment" => "construct SSE Or: Or(a, b) = a OR b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. orp%M %ia32_emit_binop /* SSE Or(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. orp%XXM %binop',
+ "units" => [ "SSE" ],
+ "mode" => "mode_E",
},
-"xEor" => {
+"xXor" => {
"irn_flags" => "R",
- "comment" => "construct SSE Eor: Eor(a, b) = a XOR b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
+ "comment" => "construct SSE Xor: Xor(a, b) = a XOR b",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. xorp%M %ia32_emit_binop /* SSE Xor(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. xorp%XXM %binop',
"latency" => 3,
+ "units" => [ "SSE" ],
+ "mode" => "mode_E",
},
# not commutative operations
"xAndNot" => {
"irn_flags" => "R",
"comment" => "construct SSE AndNot: AndNot(a, b) = a AND NOT b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3 !in_r4" ] },
- "emit" => '. andnp%M %ia32_emit_binop /* SSE AndNot(%A3, %A4) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. andnp%XXM %binop',
"latency" => 3,
+ "units" => [ "SSE" ],
+ "mode" => "mode_E",
},
"xSub" => {
"irn_flags" => "R",
"comment" => "construct SSE Sub: Sub(a, b) = a - b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3" ] },
- "emit" => '. subs%M %ia32_emit_binop /* SSE Sub(%A1, %A2) -> %D1 */',
- "outs" => [ "res", "M" ],
+ "emit" => '. sub%XXM %binop',
"latency" => 4,
+ "units" => [ "SSE" ],
+ "mode" => "mode_E",
},
"xDiv" => {
"irn_flags" => "R",
"comment" => "construct SSE Div: Div(a, b) = a / b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "xmm", "none" ], "out" => [ "in_r3 !in_r4" ] },
- "emit" => '. divs%M %ia32_emit_binop /* SSE Div(%A1, %A2) -> %D1 */',
"outs" => [ "res", "M" ],
+ "emit" => '. div%XXM %binop',
"latency" => 16,
+ "units" => [ "SSE" ],
},
# other operations
"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" ] },
- "outs" => [ "res", "M" ],
"latency" => 3,
+ "units" => [ "SSE" ],
+ "mode" => "mode_E",
},
"xCondJmp" => {
"op_flags" => "L|X|Y",
"comment" => "construct conditional jump: UCOMIS A, B && JMPxx LABEL",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "none" ], "out" => [ "xmm" ] },
- "emit" => '. movs%M %D1, %C /* Load fConst into register */',
+ "reg_req" => { "out" => [ "xmm" ] },
+ "emit" => '. mov%XXM $%C, %D1',
"latency" => 2,
+ "units" => [ "SSE" ],
+ "mode" => "mode_E",
},
# Load / Store
"op_flags" => "L|F",
"state" => "exc_pinned",
"comment" => "construct SSE Load: Load(ptr, mem) = LD ptr",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "gp", "none" ], "out" => [ "xmm" ] },
- "emit" => '. movs%M %D1, %ia32_emit_am /* Load((%A1)) -> %D1 */',
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "xmm", "none" ] },
- "emit" => '. movs%M %ia32_emit_binop /* Store(%S3) -> (%A1) */',
- "outs" => [ "M" ],
- "latency" => 2,
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "xmm", "none" ] },
- "emit" => '. movs%M %ia32_emit_am, %S2 /* store XMM0 onto stack */',
- "outs" => [ "M" ],
- "latency" => 2,
+ "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;\n",
+ "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;\n",
+ "cmp_attr" => "return 1;",
"arity" => 3,
},
"irn_flags" => "I",
"state" => "exc_pinned",
"comment" => "store ST0 onto stack",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "gp", "none" ] },
- "emit" => '. fstp %ia32_emit_am /* store ST0 onto stack */',
- "outs" => [ "M" ],
+ "reg_req" => { "in" => [ "gp", "gp", "none" ] },
+ "emit" => '. fstp%XM %AM',
"latency" => 4,
+ "units" => [ "SSE" ],
+ "mode" => "mode_M",
},
"SetST0" => {
"irn_flags" => "I",
"state" => "exc_pinned",
"comment" => "load ST0 from stack",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "none" ], "out" => [ "vf0", "none" ] },
- "emit" => '. fld %ia32_emit_am /* load ST0 from stack */',
+ "emit" => '. fld%M %AM',
"outs" => [ "res", "M" ],
"latency" => 2,
+ "units" => [ "SSE" ],
},
# CopyB
"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))",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "edi", "esi", "none" ], "out" => [ "edi", "esi", "none" ] },
"outs" => [ "DST", "SRC", "M" ],
+ "units" => [ "GP" ],
},
# Conversions
"Conv_I2I" => {
"reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "in_r3", "none" ] },
- "cmp_attr" => " return ia32_compare_conv_attr(attr_a, attr_b);\n",
"comment" => "construct Conv Int -> Int",
- "outs" => [ "res", "M" ],
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
"Conv_I2I8Bit" => {
"reg_req" => { "in" => [ "gp", "gp", "eax ebx ecx edx", "none" ], "out" => [ "in_r3", "none" ] },
- "cmp_attr" => " return ia32_compare_conv_attr(attr_a, attr_b);\n",
"comment" => "construct Conv Int -> Int",
- "outs" => [ "res", "M" ],
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
"Conv_I2FP" => {
"reg_req" => { "in" => [ "gp", "gp", "gp", "none" ], "out" => [ "xmm", "none" ] },
- "cmp_attr" => " return ia32_compare_conv_attr(attr_a, attr_b);\n",
"comment" => "construct Conv Int -> Floating Point",
- "outs" => [ "res", "M" ],
"latency" => 10,
+ "units" => [ "SSE" ],
+ "mode" => "mode_E",
},
"Conv_FP2I" => {
"reg_req" => { "in" => [ "gp", "gp", "xmm", "none" ], "out" => [ "gp", "none" ] },
- "cmp_attr" => " return ia32_compare_conv_attr(attr_a, attr_b);\n",
"comment" => "construct Conv Floating Point -> Int",
- "outs" => [ "res", "M" ],
"latency" => 10,
+ "units" => [ "SSE" ],
+ "mode" => "mode_Iu",
},
"Conv_FP2FP" => {
"reg_req" => { "in" => [ "gp", "gp", "xmm", "none" ], "out" => [ "xmm", "none" ] },
- "cmp_attr" => " return ia32_compare_conv_attr(attr_a, attr_b);\n",
"comment" => "construct Conv Floating Point -> Floating Point",
- "outs" => [ "res", "M" ],
"latency" => 8,
+ "units" => [ "SSE" ],
+ "mode" => "mode_E",
},
"CmpCMov" => {
"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" => {
"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" => {
"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" => {
"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", "none" ] },
- "outs" => [ "res", "M" ],
+ "reg_req" => { "in" => [ "gp", "gp", "gp", "gp", "none" ], "out" => [ "eax ebx ecx edx" ] },
"latency" => 2,
+ "units" => [ "GP" ],
+ "mode" => "mode_Iu",
},
"PsiCondSet" => {
"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", "none" ] },
- "outs" => [ "res", "M" ],
+ "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", "none" ] },
- "outs" => [ "res", "M" ],
+ "reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "eax ebx ecx edx" ] },
"latency" => 10,
+ "units" => [ "VFP" ],
+ "mode" => "mode_Iu",
},
"vfCMov" => {
"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",
},
#----------------------------------------------------------#
"vfadd" => {
"irn_flags" => "R",
"comment" => "virtual fp Add: Add(a, b) = Add(b, a) = a + b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "vfp" ] },
- "outs" => [ "res", "M" ],
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfmul" => {
"irn_flags" => "R",
"comment" => "virtual fp Mul: Mul(a, b) = Mul(b, a) = a * b",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "vfp" ] },
- "outs" => [ "res", "M" ],
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"l_vfmul" => {
"op_flags" => "C",
- "cmp_attr" => " return 1;\n",
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "vfp" ] },
- "outs" => [ "res", "M" ],
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"l_vfsub" => {
- "cmp_attr" => " return 1;\n",
+ "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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "vfp" ] },
"outs" => [ "res", "M" ],
"latency" => 20,
+ "units" => [ "VFP" ],
},
"l_vfdiv" => {
- "cmp_attr" => " return 1;\n",
+ "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,
},
"comment" => "virtual fp Abs: Abs(a) = |a|",
"reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] },
"latency" => 2,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfchs" => {
"comment" => "virtual fp Chs: Chs(a) = -a",
"reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] },
"latency" => 2,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfsin" => {
"comment" => "virtual fp Sin: Sin(a) = sin(a)",
"reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] },
"latency" => 150,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfcos" => {
"comment" => "virtual fp Cos: Cos(a) = cos(a)",
"reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] },
"latency" => 150,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfsqrt" => {
"comment" => "virtual fp Sqrt: Sqrt(a) = a ^ 0.5",
"reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] },
"latency" => 30,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
# virtual Load and Store
"op_flags" => "L|F",
"state" => "exc_pinned",
"comment" => "virtual fp Load: Load(ptr, mem) = LD ptr -> reg",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"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",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "vfp", "none" ] },
- "outs" => [ "M" ],
"latency" => 2,
+ "units" => [ "VFP" ],
+ "mode" => "mode_M",
},
# Conversions
"vfild" => {
"comment" => "virtual fp integer Load: Load(ptr, mem) = iLD ptr -> reg",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "none" ], "out" => [ "vfp", "none" ] },
"outs" => [ "res", "M" ],
"latency" => 4,
+ "units" => [ "VFP" ],
},
"l_vfild" => {
- "cmp_attr" => " return 1;\n",
+ "cmp_attr" => "return 1;",
"comment" => "lowered virtual fp integer Load: Load(ptr, mem) = iLD ptr -> reg",
"outs" => [ "res", "M" ],
"arity" => 2,
"vfist" => {
"comment" => "virtual fp integer Store: Store(ptr, val, mem) = iST ptr,val",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "vfp", "none" ] },
- "outs" => [ "M" ],
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_M",
},
"l_vfist" => {
- "cmp_attr" => " return 1;\n",
+ "cmp_attr" => "return 1;",
"comment" => "lowered virtual fp integer Store: Store(ptr, val, mem) = iST ptr,val",
- "outs" => [ "M" ],
"arity" => 3,
+ "mode" => "mode_M",
},
"comment" => "virtual fp Load 0.0: Ld 0.0 -> reg",
"reg_req" => { "out" => [ "vfp" ] },
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfld1" => {
"comment" => "virtual fp Load 1.0: Ld 1.0 -> reg",
"reg_req" => { "out" => [ "vfp" ] },
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfldpi" => {
"comment" => "virtual fp Load pi: Ld pi -> reg",
"reg_req" => { "out" => [ "vfp" ] },
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfldln2" => {
"comment" => "virtual fp Load ln 2: Ld ln 2 -> reg",
"reg_req" => { "out" => [ "vfp" ] },
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfldlg2" => {
"comment" => "virtual fp Load lg 2: Ld lg 2 -> reg",
"reg_req" => { "out" => [ "vfp" ] },
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfldl2t" => {
"comment" => "virtual fp Load ld 10: Ld ld 10 -> reg",
"reg_req" => { "out" => [ "vfp" ] },
"latency" => 4,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
"vfldl2e" => {
"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);",
+# "init_attr" => " set_ia32_ls_mode(res, mode);",
"comment" => "represents a virtual floating point constant",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "in" => [ "none" ], "out" => [ "vfp" ] },
+ "reg_req" => { "out" => [ "vfp" ] },
"latency" => 3,
+ "units" => [ "VFP" ],
+ "mode" => "mode_E",
},
# other
"vfCondJmp" => {
"op_flags" => "L|X|Y",
"comment" => "represents a virtual floating point compare",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { "in" => [ "gp", "gp", "vfp", "vfp", "none" ], "out" => [ "none", "none", "eax" ] },
"outs" => [ "false", "true", "temp_reg_eax" ],
"latency" => 10,
+ "units" => [ "VFP" ],
},
#------------------------------------------------------------------------#
"rd_constructor" => "NONE",
"comment" => "x87 Add: Add(a, b) = Add(b, a) = a + b",
"reg_req" => { },
- "emit" => '. fadd %ia32_emit_x87_binop /* x87 fadd(%A3, %A4) -> %D1 */',
+ "emit" => '. fadd%XM %x87_binop',
},
"faddp" => {
"rd_constructor" => "NONE",
"comment" => "x87 Add: Add(a, b) = Add(b, a) = a + b",
"reg_req" => { },
- "emit" => '. faddp %ia32_emit_x87_binop /* x87 fadd(%A3, %A4) -> %D1 */',
+ "emit" => '. faddp %x87_binop',
},
"fmul" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Mul: Mul(a, b) = Mul(b, a) = a + b",
"reg_req" => { },
- "emit" => '. fmul %ia32_emit_x87_binop /* x87 fmul(%A3, %A4) -> %D1 */',
+ "emit" => '. fmul%XM %x87_binop',
},
"fmulp" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Mul: Mul(a, b) = Mul(b, a) = a + b",
"reg_req" => { },
- "emit" => '. fmulp %ia32_emit_x87_binop /* x87 fmul(%A3, %A4) -> %D1 */',,
+ "emit" => '. fmulp %x87_binop',,
},
"fsub" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Sub: Sub(a, b) = a - b",
"reg_req" => { },
- "emit" => '. fsub %ia32_emit_x87_binop /* x87 fsub(%A3, %A4) -> %D1 */',
+ "emit" => '. fsub%XM %x87_binop',
},
"fsubp" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Sub: Sub(a, b) = a - b",
"reg_req" => { },
- "emit" => '. fsubp %ia32_emit_x87_binop /* x87 fsub(%A3, %A4) -> %D1 */',
+ "emit" => '. fsubp %x87_binop',
},
"fsubr" => {
"irn_flags" => "R",
"comment" => "x87 fp SubR: SubR(a, b) = b - a",
"reg_req" => { },
- "emit" => '. fsubr %ia32_emit_x87_binop /* x87 fsubr(%A3, %A4) -> %D1 */',
+ "emit" => '. fsubr%XM %x87_binop',
},
"fsubrp" => {
"irn_flags" => "R",
"comment" => "x87 fp SubR: SubR(a, b) = b - a",
"reg_req" => { },
- "emit" => '. fsubrp %ia32_emit_x87_binop /* x87 fsubr(%A3, %A4) -> %D1 */',
+ "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',
+},
+
+# 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" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Div: Div(a, b) = a / b",
"reg_req" => { },
- "emit" => '. fdiv %ia32_emit_x87_binop /* x87 fdiv(%A3, %A4) -> %D1 */',
+ "emit" => '. fdiv%XM %x87_binop',
},
"fdivp" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Div: Div(a, b) = a / b",
"reg_req" => { },
- "emit" => '. fdivp %ia32_emit_x87_binop /* x87 fdiv(%A3, %A4) -> %D1 */',
+ "emit" => '. fdivp %x87_binop',
},
"fdivr" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp DivR: DivR(a, b) = b / a",
"reg_req" => { },
- "emit" => '. fdivr %ia32_emit_x87_binop /* x87 fdivr(%A3, %A4) -> %D1 */',
+ "emit" => '. fdivr%XM %x87_binop',
},
"fdivrp" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp DivR: DivR(a, b) = b / a",
"reg_req" => { },
- "emit" => '. fdivrp %ia32_emit_x87_binop /* x87 fdivr(%A3, %A4) -> %D1 */',
+ "emit" => '. fdivrp %x87_binop',
},
"fabs" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Abs: Abs(a) = |a|",
"reg_req" => { },
- "emit" => '. fabs /* x87 fabs(%A1) -> %D1 */',
+ "emit" => '. fabs',
},
"fchs" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Chs: Chs(a) = -a",
"reg_req" => { },
- "emit" => '. fchs /* x87 fchs(%A1) -> %D1 */',
+ "emit" => '. fchs',
},
"fsin" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Sin: Sin(a) = sin(a)",
"reg_req" => { },
- "emit" => '. fsin /* x87 sin(%A1) -> %D1 */',
+ "emit" => '. fsin',
},
"fcos" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Cos: Cos(a) = cos(a)",
"reg_req" => { },
- "emit" => '. fcos /* x87 cos(%A1) -> %D1 */',
+ "emit" => '. fcos',
},
"fsqrt" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp Sqrt: Sqrt(a) = a ^ 0.5",
"reg_req" => { },
- "emit" => '. fsqrt $ /* x87 sqrt(%A1) -> %D1 */',
+ "emit" => '. fsqrt $',
},
# x87 Load and Store
"state" => "exc_pinned",
"comment" => "x87 fp Load: Load(ptr, mem) = LD ptr -> reg",
"reg_req" => { },
- "emit" => '. fld %ia32_emit_am /* Load((%A1)) -> %D1 */',
+ "emit" => '. fld%XM %AM',
},
"fst" => {
"state" => "exc_pinned",
"comment" => "x87 fp Store: Store(ptr, val, mem) = ST ptr,val",
"reg_req" => { },
- "emit" => '. fst %ia32_emit_am /* Store(%A3) -> (%A1) */',
+ "emit" => '. fst%XM %AM',
+ "mode" => "mode_M",
},
"fstp" => {
"state" => "exc_pinned",
"comment" => "x87 fp Store: Store(ptr, val, mem) = ST ptr,val",
"reg_req" => { },
- "emit" => '. fstp %ia32_emit_am /* Store(%A3) -> (%A1) and pop */',
+ "emit" => '. fstp%XM %AM',
+ "mode" => "mode_M",
},
# Conversions
"rd_constructor" => "NONE",
"comment" => "x87 fp integer Load: Load(ptr, mem) = iLD ptr -> reg",
"reg_req" => { },
- "emit" => '. fild %ia32_emit_am /* integer Load((%A1)) -> %D1 */',
+ "emit" => '. fild%XM %AM',
},
"fist" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp integer Store: Store(ptr, val, mem) = iST ptr,val",
"reg_req" => { },
- "emit" => '. fist %ia32_emit_am /* integer Store(%A3) -> (%A1) */',
+ "emit" => '. fist%M %AM',
+ "mode" => "mode_M",
},
"fistp" => {
"rd_constructor" => "NONE",
"comment" => "x87 fp integer Store: Store(ptr, val, mem) = iST ptr,val",
"reg_req" => { },
- "emit" => '. fistp %ia32_emit_am /* integer Store(%A3) -> (%A1) and pop */',
+ "emit" => '. fistp%M %AM',
+ "mode" => "mode_M",
},
# constants
"op_flags" => "R|c",
"irn_flags" => "R",
"comment" => "x87 fp Load 0.0: Ld 0.0 -> reg",
- "reg_req" => { },
- "emit" => '. fldz /* x87 0.0 -> %D1 */',
+ "reg_req" => { "out" => [ "vfp" ] },
+ "emit" => '. fldz',
},
"fld1" => {
"op_flags" => "R|c",
"irn_flags" => "R",
"comment" => "x87 fp Load 1.0: Ld 1.0 -> reg",
- "reg_req" => { },
- "emit" => '. fld1 /* x87 1.0 -> %D1 */',
+ "reg_req" => { "out" => [ "vfp" ] },
+ "emit" => '. fld1',
},
"fldpi" => {
"op_flags" => "R|c",
"irn_flags" => "R",
"comment" => "x87 fp Load pi: Ld pi -> reg",
- "reg_req" => { },
- "emit" => '. fldpi /* x87 pi -> %D1 */',
+ "reg_req" => { "out" => [ "vfp" ] },
+ "emit" => '. fldpi',
},
"fldln2" => {
"op_flags" => "R|c",
"irn_flags" => "R",
"comment" => "x87 fp Load ln 2: Ld ln 2 -> reg",
- "reg_req" => { },
- "emit" => '. fldln2 /* x87 ln(2) -> %D1 */',
+ "reg_req" => { "out" => [ "vfp" ] },
+ "emit" => '. fldln2',
},
"fldlg2" => {
"op_flags" => "R|c",
"irn_flags" => "R",
"comment" => "x87 fp Load lg 2: Ld lg 2 -> reg",
- "reg_req" => { },
- "emit" => '. fldlg2 /* x87 log(2) -> %D1 */',
+ "reg_req" => { "out" => [ "vfp" ] },
+ "emit" => '. fldlg2',
},
"fldl2t" => {
"op_flags" => "R|c",
"irn_flags" => "R",
"comment" => "x87 fp Load ld 10: Ld ld 10 -> reg",
- "reg_req" => { },
- "emit" => '. fldll2t /* x87 ld(10) -> %D1 */',
+ "reg_req" => { "out" => [ "vfp" ] },
+ "emit" => '. fldll2t',
},
"fldl2e" => {
"op_flags" => "R|c",
"irn_flags" => "R",
"comment" => "x87 fp Load ld e: Ld ld e -> reg",
- "reg_req" => { },
- "emit" => '. fldl2e /* x87 ld(e) -> %D1 */',
+ "reg_req" => { "out" => [ "vfp" ] },
+ "emit" => '. fldl2e',
},
"fldConst" => {
"irn_flags" => "R",
"rd_constructor" => "NONE",
"comment" => "represents a x87 constant",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
- "reg_req" => { "out" => [ "st" ] },
- "emit" => '. fld %ia32_emit_adr /* Load fConst into register -> %D1 */',
+ "reg_req" => { "out" => [ "vfp" ] },
+ "emit" => '. fld $%C',
},
# 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" => { "in" => [ "st"], "out" => [ "st" ] },
- "cmp_attr" => " return 1;\n",
- "emit" => '. fxch %X1 /* x87 swap %X1, %X3 */',
+ "reg_req" => { },
+ "cmp_attr" => "return 1;",
+ "emit" => '. fxch %X1',
},
"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" => [ "st"], "out" => [ "st" ] },
- "cmp_attr" => " return 1;\n",
- "emit" => '. fld %X1 /* x87 push %X1 */',
+ "reg_req" => { "in" => [ "vfp"], "out" => [ "vfp" ] },
+ "cmp_attr" => "return 1;",
+ "emit" => '. fld %X1',
},
"fpop" => {
"op_flags" => "R|K",
"comment" => "x87 stack pop",
- "reg_req" => { "out" => [ "st" ] },
- "cmp_attr" => " return 1;\n",
- "emit" => '. fstp %X1 /* x87 pop %X1 */',
+ "reg_req" => { },
+ "cmp_attr" => "return 1;",
+ "emit" => '. fstp %X1',
},
# compare
"fcomJmp" => {
"op_flags" => "L|X|Y",
"comment" => "floating point compare",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { },
},
"fcompJmp" => {
"op_flags" => "L|X|Y",
"comment" => "floating point compare and pop",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { },
},
"fcomppJmp" => {
"op_flags" => "L|X|Y",
"comment" => "floating point compare and pop twice",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { },
},
"fcomrJmp" => {
"op_flags" => "L|X|Y",
"comment" => "floating point compare reverse",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { },
},
"fcomrpJmp" => {
"op_flags" => "L|X|Y",
"comment" => "floating point compare reverse and pop",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { },
},
"fcomrppJmp" => {
"op_flags" => "L|X|Y",
"comment" => "floating point compare reverse and pop twice",
- "cmp_attr" => " return ia32_compare_immop_attr(attr_a, attr_b);\n",
"reg_req" => { },
},