| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
| |
__builtin_bswap64 is now available both in 32 and 64-bit mode.
The DWARF bit is the numbering of registers in ia32/Asmexpand.ml.
|
| |
|
|
|
| |
- Avoid absolute addressing for labels, use RIP-relative addressing
- Different, RIP-relative implementation of jump tables
|
| |
|
|
|
|
| |
This trick was already implemented for 32-bit integer division and modulus. Here we extend it to the 64-bit case.
For 32-bit target processors, the runtime library must implement 64-bit multiply-high (signed and unsigned). Tentative implementations are provided for IA32 and PowerPC, but need testing.
|
| |
|
|
|
|
| |
Implement the 'shift right extended' trick, both in the generic implementation (backend/SplitLong) and in the IA32 port.
Note that now SelectDiv depends on SelectLong, and that some work was moved from SelectLong to SelectDiv.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This commit enriches the IA32 port so that it supports x86 processors in 64-bit mode as well as in 32-bit mode, depending on the value of Archi.ptr64, which itself is set from the configuration model.
To activate x86-64 bit support, configure with "x86_64-linux".
Main steps:
- Enrich Op.v and Asm.v with 64-bit operations
- SelectLong: in 64-bit mode, use 64-bit operations directly; in 32-bit mode, fall back on the old implementation based on pairs of 32-bit integers
- Conventions1: support x86-64 ABI in addition to the 32-bit ABI.
- Add support for the new 64-bit operations everywhere.
- runtime/x86_64: implementation of the supporting library appropriate for x86 in 64-bit mode
To do:
- More optimizations are possible on 64-bit integer arithmetic operations.
- Could add new chunks to load, say, an unsigned byte into a 64-bit long
(currently we load as a 32-bit int then zero-extend).
- Implements the wrong ABI for struct passing.
|
| |
|
|
|
|
| |
Avoids problems with overwritting the registe containing the
function address.
Bug 19779
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
architectures
The original Stacking pass and its proof hard-wire assumptions about the processor and the register allocation, namely that integer registers are 32 bit wide and that all stack slots have natural alignment 4, which precludes having stack slots of type Tlong. Those assumptions become false if the target processor has 64-bit integer registers.
This commit makes minimal adjustments to the Stacking pass so as to lift these assumptions:
- Stack slots of type Tlong (or more generally of natural alignment 8) are supported. For slots produced by register allocation, the alignment is validated a posteriori in Lineartyping. For slots produced by the calling conventions, alignment is proved as part of the "loc_argument_acceptable" property in Conventions1.
- The code generated by Stacking to save and restore used callee-save registers no longer assumes 32-bit integer registers. Actually, it supports any combination of sizes for registers.
- To support the new save/restore code, Bounds was changed to record the set of all callee-save registers used, rather than just the max index of callee-save registers used.
On CompCert's current 32-bit target architectures, the new Stacking pass should generate pretty much the same code as the old one, modulo minor differences in the layout of the stack frame. (E.g. padding could be introduced at different places.)
The bulk of this big commit is related to the proof of the Stacking phase. The old proof strategy was painful and not obviously adaptable to the new Stacking phase, so I rewrote Stackingproof entirely, using an approach inspired by separation logic. The new library common/Separation.v defines assertions about memory states that can be composed using a separating conjunction, just like pre- and post-conditions in separation logic. Those assertions are used in Stackingproof to describe the contents of the stack frames during the execution of the generated Mach code, and relate them with the Linear location maps.
As a further simplification, the callee-save/caller-save distinction is now defined in Conventions1 by a function is_callee_save: mreg -> bool, instead of lists of registers of either kind as before. This eliminates many boring classification lemmas from Conventions1. LTL and Lineartyping were adapted accordingly.
Finally, this commit introduces a new library called Decidableplus to prove some propositions by reflection as Boolean computations. It is used to further simplify the proofs in Conventions1.
|
| | |
|
| |
|
|
|
|
|
|
|
|
| |
The AST.ident type represents source-level identifiers as unique positive numbers. However, the mapping identifiers <-> AST.ident differs between runs of CompCert on different source files. This is problematic when we need to produce or recognize external functions and builtin functions with fixed names, for example:
* in $ARCH/Machregs.v to define the register conventions for builtin functions;
* in the VST program logic from Princeton to treat thread primitives specially.
So far, we used AST.ident_of_string to recover the ident associated with a string. However, this function is defined in OCaml and doesn't execute within Coq. This is a problem both for VST and for future executability of CompCert within Coq.
This commit replaces "ident" by "string" in the arguments of EF_external, EF_builtin, EF_inline_asm, EF_annot, and EF_annot_val. This provides stable names for externals and builtins, as needed. For inline asm and annotations, it's a matter of taste, but using strings feels more natural. EF_debug keeps using idents, since some kinds of EF_debug annotations talk about program variables.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Before, the back-end languages had distinct instructions
- Iannot for annotations, taking structured expressions (annot_arg)
as arguments, and producing no results'
- Ibuiltin for other builtins, using simple pseudoregs/locations/registers
as arguments and results.
This branch enriches Ibuiltin instructions so that they take structured
expressions (builtin_arg and builtin_res) as arguments and results.
This way,
- Annotations fit the general pattern of builtin functions,
so Iannot instructions are removed.
- EF_vload_global and EF_vstore_global become useless, as the
same optimization can be achieved by EF_vload/vstore taking
a structured argument of the "address of global" kind.
- Better code can be generated for builtin_memcpy between stack locations,
or volatile accesses to stack locations.
Finally, this commit also introduces a new kind of external function,
EF_debug, which is like EF_annot but produces no observable events.
It will be used later to transport debug info through the back-end,
without preventing optimizations.
|
| |
|
|
|
|
|
|
|
|
| |
- Treat clobbered registers as being destroyed by EF_inline_asm builtins
(which is the truth, semantically).
- To enable the above, represent clobbers as Coq strings rather than idents
and move register_by_name from Machregsaux.ml to Machregs.v.
- Side benefit: more efficient implementation of Machregsaux.name_of_register.
-# Please enter the commit message for your changes. Lines starting
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Support single-precision floats as first-class values
- Introduce chunks Many32, Many64 and types Tany32, Tany64 to
support saving and restoring registers without knowing
the exact types (int/single/float) of their contents, just
their sizes.
- Memory model: generalize the opaque encoding of pointers to
apply to any value, not just pointers, if chunks Many32/Many64
are selected.
- More properties of FP arithmetic proved.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2537 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
| |
|
|
|
|
| |
selection.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2451 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
| |
|
|
|
|
|
|
| |
that pop the x87 FP stack (var <- FP0). Otherwise, (void) f();
where f returns a float eventually produces a FP stack overflow.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2416 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
| |
|
|
|
|
|
| |
- Revised printing of intermediate RTL code.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2403 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
| |
|
|
|
|
|
| |
multiply-high and shifts.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2300 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
| |
|
|
|
|
|
|
| |
- added RTL type "Tsingle"
- ABI-compatible passing of single-precision floats on ARM and x86
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2260 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
| |
|
|
|
|
|
|
| |
for compatibility with earlier CompCert versions.
But don't use them in PackedStructs.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2216 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
| |
|
|
|
|
|
|
|
| |
Remove __builtin_{read,write}_reversed from IA32 and ARM ports.
Machregs: tighten destroyed_by_builtin
Packedstructs: use bswap if read/write-reversed not available.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2208 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
| |
|
|
|
|
|
|
|
| |
1- new register allocator (+ live range splitting, spilling&reloading, etc)
based on a posteriori validation using the Rideau-Leroy algorithm
2- support for 64-bit integer arithmetic (type "long long").
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2200 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
| |
|
|
|
|
|
|
| |
for no good reason.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@2140 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
|
|
- Reload temporaries are marked as destroyed (set to Vundef) across
operations in the semantics of LTL, LTLin, Linear and Mach,
allowing Asmgen to reuse them.
- Added IA32 port.
- Cleaned up float conversions and axiomatization of floats.
git-svn-id: https://yquem.inria.fr/compcert/svn/compcert/trunk@1499 fca1b0fc-160b-0410-b1d3-a4f43f01ea2e
|
