Commit Graph

306 Commits

Author SHA1 Message Date
florian
53c39d5420 * fixes peephole optimizer problems with -O3 on x86-64
git-svn-id: trunk@21288 -
2012-05-13 19:19:04 +00:00
Jonas Maebe
edd42aa42a * moved subsetref/reg and bit_set/test support from cgobj to hlcgobj for
future use by high level code generator targets
   o this in turn required that all a_load*_loc* methods are called via
     hlcg rather than via cg, since a location can be a subsetref/reg and
     and those are no longer handled in tcg
   o that then required moving several force_location_* routines into
     thlcg because they use a_load_loc*, but did not take tdef size
     parameters (which are required by the thlcg a_load_loc* routines)
   o the only practical consequence is that from now on, you have to
     use hlcg.location_force_mem/reg() (fpureg not yet) and
     hlcg.gen_load_loc_cgpara() instead of the removed versions from ncgutil,
     and hlcg.a_load*loc*() instead of cg.a_load*loc* if a subsetref/reg
     might be involved

git-svn-id: trunk@21287 -
2012-05-13 12:33:10 +00:00
Jonas Maebe
85a3fd3357 + ossinttype/osuinttype defs that correspond to OS_SINT/OS_INT for use in
the high level code generator

git-svn-id: trunk@21279 -
2012-05-12 16:03:15 +00:00
Jonas Maebe
834026bfb5 * synchronised with trunk up to r21067
git-svn-id: branches/jvmbackend@21068 -
2012-04-26 21:24:20 +00:00
pierre
bf1690458a Partial fix for x86_64 peephole optimization
git-svn-id: trunk@21022 -
2012-04-24 15:38:07 +00:00
florian
fc673340fe + implemented x86-64 mov optimization to test register usage tracking framework
git-svn-id: trunk@20892 -
2012-04-15 20:30:45 +00:00
Jonas Maebe
aee5380ae0 * merged trunk up to r20882
o support for the new codepage-aware ansistrings in the jvm branch
   o empty ansistrings are now always represented by a nil pointer rather than
     by an empty string, because an empty string also has a code page which
     can confuse code (although this will make ansistrings harder to use
     in Java code)
   o more string helpers code shared between the general and jvm rtl
   o support for indexbyte/word in the jvm rtl (warning: first parameter
     is an open array rather than an untyped parameter there, so
     indexchar(pcharvar^,10,0) will be equivalent to
     indexchar[pcharvar^],10,0) there, which is different from what is
     intended; changing it to an untyped parameter wouldn't help though)
   o default() support is not yet complete
   o calling fpcres is currently broken due to limitations in
     sysutils.executeprocess() regarding handling unix quoting and
     the compiler using the same command lines for scripts and directly
     calling external programs
   o compiling the Java compiler currently requires adding ALLOW_WARNINGS=1
     to the make command line

git-svn-id: branches/jvmbackend@20887 -
2012-04-15 15:54:10 +00:00
florian
9e105677db * optimize mov/and sequences on x86-64
git-svn-id: trunk@20883 -
2012-04-15 12:58:02 +00:00
florian
d24b70d22b * ported part of the i386 peep hole optimizer to x86-64 to see if it is possible to create a shared x86 peep hole optimizer
git-svn-id: trunk@20874 -
2012-04-14 19:08:02 +00:00
florian
2f5ce095ce * RefsHaveIndexReg -> cpurefshaveindexreg
* cpurefshaveindexreg defined properly in fpcdefs.inc

git-svn-id: trunk@20504 -
2012-03-10 19:43:52 +00:00
florian
fc569e224d * handle all operand combinations of xadd, resolves #21146
git-svn-id: trunk@20177 -
2012-01-25 21:26:59 +00:00
sergei
cbad0abab5 + Platform-specific exception support for x86_64-win64. Enable by cycling with OPT=-dTEST_WIN64_SEH.
git-svn-id: trunk@20098 -
2012-01-16 23:13:31 +00:00
sergei
fde1695961 * tcgx86_64.g_proc_exit: restore stack pointer using value that was calculated and stored in tcgx86.g_proc_entry, instead of recalculating it again.
git-svn-id: trunk@20078 -
2012-01-14 18:04:45 +00:00
sergei
fffa4a35a0 * x86_64: Don't waste time initializing/finalizing saved_standard_registers and saved_mm_registers arrays on every procedure. Initialize once, reinitialize if target system happens to change, and let finalization happen implicitly on shutdown.
git-svn-id: trunk@19672 -
2011-11-23 20:06:38 +00:00
sergei
6a3fe72de9 + Support .rva directive in AT&T reader. Put it into base class because it generally applies to all targets with COFF output, but enabled for Windows targets only (others need additional testing).
+ Support .seh_handlerdata directive in Win64.

git-svn-id: trunk@19546 -
2011-10-25 15:18:47 +00:00
sergei
e11c880b1e x86 assembler improvements:
* Don't generate rex.w for "CALL|JMP|LCALL|LJMP regmem", they are 64-bit by default.
* LCALL,LJMP flagged as calljump instructions.
* LCALL,LJMP encode only far jumps and don't accept register operands.
* GAS writer: fixed writing rip-relative operands of calljump instructions.
+ test.

git-svn-id: trunk@19413 -
2011-10-08 11:34:04 +00:00
sergei
a463fbc578 * Reset unwind_info.flags between procedures. Writing non-zero flags without handler ends up in corrupt unwind info.
git-svn-id: trunk@19403 -
2011-10-07 14:46:56 +00:00
sergei
85f7914906 * Don't generate .seh_endprologue if SEH directives are present in the text of (pure assembler) procedure, as it results in duplicate .seh_endprologue.
* Added checks for .seh_endprologue presence and correct position.

git-svn-id: trunk@19388 -
2011-10-06 03:29:35 +00:00
sergei
0db44ae108 + Support SEH directives in x86_64 AT&T asmreader.
git-svn-id: trunk@19366 -
2011-10-04 12:26:41 +00:00
sergei
9eb451756b + Support .seh_handler directive
git-svn-id: trunk@19365 -
2011-10-04 12:19:24 +00:00
sergei
4c21beb48e win64 unwinding, misc fixes:
* Insert seh_endprologue directive after the last prologue instruction, not before it
* Omit seh_stackalloc for zero bytes
* (For now) ignore SEH directives in NASM and MASM writers, instead of failing with internal error.

git-svn-id: trunk@19246 -
2011-09-26 13:55:21 +00:00
sergei
d94460e3b3 * Made the x86-specific scalefactor optimization of tvecnode available on x86_64 as well, by moving update_reference_reg_mul method from ti386vecnode to newly introduced tx86vecnode.
git-svn-id: trunk@19245 -
2011-09-26 12:58:59 +00:00
sergei
b997094755 + Generate unwind bytecode for function prologues on win64.
* For now placed actual processing of unwind info under {$ifdef TEST_WIN64_UNWIND}, because in the current state it doesn't add much value.

git-svn-id: trunk@19200 -
2011-09-23 21:22:25 +00:00
sergei
1365467b0d + Support classes for creating unwind bytecode for Win64.
git-svn-id: trunk@19071 -
2011-09-15 15:35:08 +00:00
florian
e8dafe4dde * fixed change information for a lot of sse instructions
git-svn-id: trunk@18787 -
2011-08-20 12:28:49 +00:00
Jonas Maebe
28740dce2d - removed extra "fordefinition" parameter again from tprocdef.mangledname(),
since the definition-specific adorning of JVM mangled names is Jasmin-
    specific, and such code has no place in symdef
  * moved code to adorn JVM mangled names for Jasmin definitions to agjasmin

git-svn-id: branches/jvmbackend@18346 -
2011-08-20 07:49:31 +00:00
Jonas Maebe
b837694207 * factored out releasing an unused return value into
release_unused_return_value_cpu(), so it can be cleanly
    overridden by cpu-specific code

git-svn-id: branches/jvmbackend@18322 -
2011-08-20 07:45:44 +00:00
Jonas Maebe
71deda6f50 + added interface to ncgutil.gen_load_loc_cgpara() to hlcgobj + generic
implementation (without loc_©mmregister support)
  * moved ncgutil.gen_load_return_value() to hlcgobj, and factored out
    architecture-specific behaviour to load an uninitialised function result
    into a virtual method (+ JVM-specific implementation of that method),
    gen_load_uninitialized_function_result()
  + added hlcgx86 unit and thlcgx86 type to override the
    thlcgobj.gen_load_uninitialized_function_result() method for x87
    function results; the i386 and x86_64 units now instantiate thlcgx86
    instead of thlcg2ll
  * moved calling of ncgutil.gen_load_loc_cgpara() from ncgcal also to hlcgobj
  -> returning function results works for JVM

git-svn-id: branches/jvmbackend@18317 -
2011-08-20 07:38:16 +00:00
Jonas Maebe
0ee702b3a2 * tprocdef.mangledname now gets an extra boolean parameter indicating
whether the mangled name is for defining a symbol, or for referencing
    it later (e.g. for a call or load of its address). The reason is that
    on the JVM both cases are different.
  + jvmdef unit to encode types according to the JVM rules
  + tprocdef.jvmmangledname() to encode a procdef's JVM mangled name
    (the common part of defining/referencing it; tprocdef.mangledname
     afterwards adorns it as required)

git-svn-id: branches/jvmbackend@18288 -
2011-08-20 07:22:00 +00:00
Jonas Maebe
72e9cfee24 * create/destroy also the high level code generator for all architectures,
so it can be used in generic code

git-svn-id: branches/jvmbackend@18280 -
2011-08-20 07:21:16 +00:00
florian
c011949765 + iretq for x86_64
git-svn-id: trunk@18273 -
2011-08-19 12:56:26 +00:00
florian
8308b46a94 + support for assembler instructions with four operands
+ support for insertq, resolves #19910

git-svn-id: trunk@18206 -
2011-08-14 16:46:35 +00:00
florian
ff5f311b34 - removed no more used constants
git-svn-id: trunk@18199 -
2011-08-13 20:54:01 +00:00
sergei
0231863fce + Added missing PMULLD instruction, part of Mantis #19910
git-svn-id: trunk@18106 -
2011-08-06 06:59:33 +00:00
sergei
7d99f95c45 * Always create a section before emitting data to current_asmdata.asmlists[al_typedconsts]. Without it, such data ends up in sections created elsewhere, creating very non-obvious dependencies on other parts of compiler.
git-svn-id: trunk@17816 -
2011-06-24 02:05:56 +00:00
sergei
d4b2998e07 * Another attempt to fix x86_64 stack frame calculation, tested better this time.
git-svn-id: trunk@17709 -
2011-06-10 03:41:33 +00:00
sergei
c40b8d92c7 * Revert r17695 because it breaks cycling.
git-svn-id: trunk@17698 -
2011-06-08 19:19:24 +00:00
sergei
318a55b96c * x86_64-win64: fixed stack calculations so that 32-byte spilling area is only included once and only for non-leaf procedures (it was included twice with -O- and 3 times with -O2, wasting large amounts of stack memory).
git-svn-id: trunk@17695 -
2011-06-08 14:46:22 +00:00
sergei
354d0520b7 + x86 assembler: fixed MOVABS instruction (it is a x86_64-only subset of MOV with 8-byte immediates/offsets) and same-form encodings of MOV instruction.
git-svn-id: trunk@17666 -
2011-06-05 17:32:18 +00:00
sergei
9608c4eabe * Interface wrappers must be global when linking smart, Mantis #19462. Other targets already have this condition in place.
git-svn-id: trunk@17631 -
2011-06-02 10:49:28 +00:00
joost
45bc3fcae0 * Enabled safecall on x86_64-linux
git-svn-id: trunk@17629 -
2011-06-02 10:43:18 +00:00
sergei
d89aeeadaf x86 assembler:
* Adjust rip-relative offsets using values known from calcsize, instead of reinterpreting the tail of instruction. This is simpler and remains the same regardless of particular codes used in opcode table.
* Reduced maxinfolen to 8.
- 'jmp imm64' and 64-bit versions of 'in' do not exist, removed.

git-svn-id: trunk@17546 -
2011-05-23 20:43:50 +00:00
sergei
504e0c6816 x86 assembler fixes:
* mkx86ins.pp: 'regmem' operand type in x86ins.dat must be converted to OT_RM_GPR, not OT_REGMEM. The latter does not restrict register type, allowing to use e.g. xmm registers in place of regular ones. 
* Finished 'movd' and 'movq', added some tests for them to tasm2.pp.

git-svn-id: trunk@17515 -
2011-05-20 20:39:35 +00:00
sergei
08e895cf7c * Fixed handling of 'movq' instruction by assembler reader, resolves #18205.
git-svn-id: trunk@17497 -
2011-05-19 06:44:24 +00:00
sergei
1aa5c4ac0e * Fixed wrong ModRM byte in movq encoding, causing regression of test/cg/tcalext6 in r17490.
git-svn-id: trunk@17493 -
2011-05-18 14:55:26 +00:00
sergei
9e8a31193b x86 assembler:
* Optimized the opcode representation of movq and remaining 3DNow instructions
* Disallow immediates not fitting in 32 bits (Mantis #14685) + test
* Disallow push/pop with 32-bit operands in x86_64 + test

git-svn-id: trunk@17490 -
2011-05-17 20:06:12 +00:00
sergei
c28197bac3 * PSADBW second operand is xmmrm, not xmmreg
+ A couple of tests related to the recent assembler changes

git-svn-id: trunk@17460 -
2011-05-14 18:59:06 +00:00
sergei
958954890e A big update of x86 instruction table, part 2:
- Removed remaining address-size control codes \300 and \301
- Also removed codes \323, they are no longer necessary once REX is being written on the first literal.

git-svn-id: trunk@17457 -
2011-05-14 16:36:42 +00:00
sergei
3b979fef6d * Re-commit r17437 after more testing and fixing aasmcpu.pp in r17449.
git-svn-id: trunk@17452 -
2011-05-14 11:04:52 +00:00
sergei
b257231203 * Revert r17437, it breaks builds with -O2 and builds on i386 (although -O- on x86_64 is ok).
git-svn-id: trunk@17439 -
2011-05-12 23:53:18 +00:00