address16_op flags into a single postop enum.
* x86id.re (yasm_x86__finalie_insn): Set new enum rather than flags.
* x86bc.c: Use new combined enum.
svn path=/trunk/yasm/; revision=1268
While we're here, add 64-bit register versions of SSE2 instructions movmskps,
pextrw, pinsrw, and pmovmskb that are documented by Intel but not AMD.
Add test for above.
svn path=/trunk/yasm/; revision=1265
directive for GAS (it's generated by GCC).
* bytecode.c (bytecode_align): Update for full align implementation.
(yasm_bc_create_align): Likewise.
(bc_align_finalize): New.
(bc_align_resolve, bc_align_tobytes): Real implementation.
* bytecode.h (yasm_bc_create_align): Update to match.
* arch.h (get_fill, yasm_arch_get_fill): New, to get NOP fill patterns.
* x86arch.c (x86_get_fill): Implement.
* lc3barch.c (lc3b_get_fill): Likewise (probably buggy, there's no real NOP).
* bytecode.c (bytecode_org): New org bytecode.
(bc_org_callback, bc_org_destroy, bc_org_print, bc_org_resolve)
(bc_org_tobytes, yasm_bc_create_org): Implement.
* gas-parser.h (yasm_parser_gas): Add code_section flag to indicate when to
use code fill vs. data fill.
* gas-parser.c: Initialize flag.
* gas-bison.y: Update flag in various places. Generate org bytecode.
Call gas_parser_align to generate align bytecode.
(gas_parser_align): Generate align bytecode.
* gas-bison.y: Implement .zero directive.
* gas-token.re: Likewise.
* align32, align64: New tests to test align directive NOP generation.
svn path=/trunk/yasm/; revision=1263
these cases seems to be a bug. If you get a linker error about ADDR32, it
means you aren't using RIP-relative instructions. Note this means to access
an array you need to do:
lea rax, [var wrt rip] ; generates RIP-relative insn and REL32 reloc
mov rcx, [rax+rbx] ; rbx is index
and not:
mov rcx, [var+rbx] ; generates ADDR32 reloc
At least when trying to build a DLL (the ADDR32 reloc fails the DLL link).
When building statically, ADDR32 should work okay and thus the latter form
can be used.
svn path=/trunk/yasm/; revision=1257
output better by generating relocs directly to the symbol being relocated
rather than to the section. Use a new coff_objfmt->win64 flag to
conditionalize this rather than just COFF_MACHINE_AMD64.
(coff_objfmt): New win64 flag.
(coff_objfmt_create, win32_objfmt_create, win64_objfmt_create): Initialize
flag.
(coff_objfmt_output): Turn on outputting all symbols in win64 mode so they
can be referenced by relocs. This isn't quite correct: we should only turn
on the symbols that are actually used by relocs, but having them there
doesn't hurt linking; it only exposes all of the internal symbol names.
With these changes, yasm output matches the new ML64 output except for a
very few cases:
- ML64 generates REL32 relocs when referencing objects in the same .text
section. I cannot see how this is necessary because call instructions
don't generate REL32 relocs! I currently do not plan on fixing this unless
it causes a problem.
- ML64 generates ADDR32 relocs instead of REL32 relocs when loading a
32-bit register with the address of an object. I will probably try to fix
this.
Extended test case for this a bit.
svn path=/trunk/yasm/; revision=1256
defining the mov forms for GAS. While I'm here, fix movq so it also
supports the 64-bit move registers (per Intel's spec, AMD has it under movd)
and copy the MMX/SSE2 versions of movq into mov so they're visible to the
GAS parser (and only the GAS parser).
Add a whole bunch of testcases to test movd and movq in both 32 bit and 64
bit modes for both GAS and NASM parsers.
svn path=/trunk/yasm/; revision=1255
- Add elf32 as an alias for -f elf.
- Add elf64 as an alias for -f elf -m amd64.
Note the old command lines still work.
Add a testcase for win64 (includes masm -> yasm mapping, look at
win64-dataref.masm and win64-dataref.asm files respectively).
svn path=/trunk/yasm/; revision=1252
instructions in Win64 to always be REL32 regardless of whether they're
RIP relative or not. I don't understand this behavior, but it matches how
ML64 generates relocs and unbreaks linking. The handling of this case must
be handled at a higher level somehow (either at the linker or the compiler).
Note that this REL32 generation behavior of ML64 happens only with the
latest version (VC8); the 2003 SP1 SDK ML64 doesn't do this.
* bytecode.c (yasm_bc_is_data): New supporting function.
* bytecode.h (yasm_bc_is_data): Prototype.
Testcase pending.
Reported by and much debugging support contributed by:
Brian Gladman <brg@gladman.plus.com>. Thanks!
svn path=/trunk/yasm/; revision=1251
parse_check_segreg, parse_check_insn, parse_check_prefix, and
parse_check_targetmod. This will allow for future improvements to
identifier handling in the various parsers.
svn path=/trunk/yasm/; revision=1249
(yasm_x86__finalize_insn): Ignore special suffix value 0x80 when matching
in info, but use strict matching. This unbreaks jmp/call broken in previous
commit.
(yasm_x86__parse_check_insn): Initialize data[3] and for FLDT and FSTPT, set
special suffix value 0x80.
svn path=/trunk/yasm/; revision=1247
interrupts, conditional movs/sets, and a few other minor things. What's
left: string insns, loop insns, other jmp/call forms, protection control,
SSE/SSE2, and odds and ends (like prefixes-as-instructions).
Add an incomplete FP testcase.
svn path=/trunk/yasm/; revision=1246
errors when using enter in 64-bit mode.
* x86arch.h (yasm_x86__expr_checkea): Update prototype.
* x86bc.c (x86_bc_insn_resolve, x86_bc_insn_tobytes): Pass flag to above.
* x86id.re: Implement mul, imul, div, idiv, enter, leave for GAS mode.
Add tests for above.
svn path=/trunk/yasm/; revision=1245
This work is being done under contract with a company that has requested
to remain unnamed at the present time.
* bc-int.h (yasm_effaddr): Add strong flag to indicate if the effective
address is definitely an effective address; GAS does not use [] to designate
effective addresses so it's otherwise impossible to tell the difference
between "expr(,1)" and just "expr" (important for the relative jump
instructions).
* bytecode.h (yasm_ea_set_strong): New function to set the strong flag.
* bytecode.c (yasm_ea_set_strong): Implementation.
* x86bc.c (yasm_x86__ea_create_reg): Initialize strong flag.
* arch.h (yasm_insn_operand): Add deref flag to indicate use of "*foo" in
GAS syntax.
* arch.c (yasm_operand_create_reg, yasm_operand_create_segreg)
(yasm_operand_create_mem, yasm_operand_create_imm): Set deref flag to 0.
* gas: GAS syntax lexer and parser. Not all directives are implemented yet
(some will require additional core bytecodes).
* elf-objfmt.c (elf_objfmt_section_switch): Add support for GAS-style
section flags.
* x86arch.h (yasm_arch_x86): Add parser setting.
* x86arch.c (x86_create): Check for gas parser and initialize setting.
* x86bc.c (yasm_x86__ea_create_expr): Transform val+RIP to val wrt RIP when
using the GAS parser (this is how GAS interprets "expr(%rip)").
* x86id.re: Too many changes to enumerate in detail. Add new modifiers for
GAS suffixes. Start using them in a couple instructions. Split check_id
into subfunctions (still one entry point at present).
(yasm_x86__finalize_insn): Support new modifiers, reverse operands, derefs.
* yasm.c (main): Change all undef to extern when using GAS parser (this is
default GAS behavior).
svn path=/trunk/yasm/; revision=1239
represent a register group (e.g. indexed registers).
(yasm_arch_reggroup_get_reg): New function to get a specific register from
a register group and index.
(yasm_arch_module): Add module version of yasm_arch_reggroup_get_reg().
* lc3barch.c (lc3b_reggroup_get_reg): Implement.
(yasm_lc3b_LTX_arch): Point to implementation.
* x86arch.c (x86_reggroup_get_reg, yasm_x86_LTX_arch): Likewise.
svn path=/trunk/yasm/; revision=1234
undefined symbols extern if unused rather than causing undef errors.
* symrec.c (yasm_symtab_parser_finalize): Implement.
(symtab_finalize_info): New (more data to pass to
(symtab_parser_finalize_checksym): Update finalize helper.
* yasm.c (main): Update call to yasm_symtab_parser_finalize().
svn path=/trunk/yasm/; revision=1232
(yasm_arch_create): Add parser and error parameters; now the arch is given
the keyword of the parser in use. The error parameter allows the caller to
find out whether it was the machine name or the parser name that was in
error.
(yasm_arch_module): Change create definition to match yasm_arch_create().
(yasm_arch_create_error): New error typedef for yasm_arch_create() errors.
* lc3barch.c (lc3b_create): Update to match new yasm_arch_create().
* x86arch.c (x86_create): Likewise.
* yasm.c (main): Use new yasm_arch_create() and handle the two kinds of
errors it can now generate. Move parser creation up in the sequence so it
happens before the arch is created.
svn path=/trunk/yasm/; revision=1230
the object format default BITS setting. This makes it so [bits 64] is not
necessary to explicitly specify in the source file.
svn path=/trunk/yasm/; revision=1228
shortening to signed 8-bit immediate from a larger immediate size. This
yields much smaller code for many arithmetic instructions.
Noticed by: Brian Gladman <brg@gladman.plus.com>
svn path=/trunk/yasm/; revision=1227
sections. This breaks section numbering between the file section headers
and the section numbering used by symbols to reference sections.
While we're here, don't even try to number sections during parse... this
numbering is getting overwritten anyway.
Fixes#59 (reported by hkmaly@gmail.com).
svn path=/trunk/yasm/; revision=1226
platforms (notably Win64), unsigned long is not big enough to hold a
pointer.
* Makefile.am, configure.ac: Use ax_create_stdint_h to get us uintptr_t.
* ax_create_stdint_h.m4: Implementation of ax_create_stdint_h autoconf
macro from http://ac-archive.sourceforge.net/guidod/ax_create_stdint_h.html
svn path=/trunk/yasm/; revision=1223
works (was broken for >1 byte files).
(yasm_bc_tobytes): Fix handling of bytecodes that are larger than provided
buffer.
Patch by: Stephen Polkowski <stephen@centtech.com>
svn path=/trunk/yasm/; revision=1219
Peter Johnson