Commit graph

234 commits

Author SHA1 Message Date
MerryMage
179a3388f9 block_of_code: Provide an alloc function to allocate space in the code block 2016-12-12 07:49:18 +00:00
Lioncash
f467589346 emit_x64: Remove unnecessary casts 2016-12-05 20:30:19 +00:00
Lioncash
a37631c010 emit_x64: Use reinterpret_cast for pointer casts 2016-12-05 20:30:19 +00:00
Lioncash
fafa845f64 emit_x64: Make GetBasicBlock() const qualified 2016-12-05 12:46:22 +00:00
Lioncash
6a16edc0fb emit_x64: Move implementations into the cpp file
Prevents needing to rebuild everything including the emitter if any
details ever change.
2016-12-05 12:46:22 +00:00
Lioncash
282029f60a emit_x64: Forward declare BlockOfCode 2016-12-05 12:46:22 +00:00
Lioncash
6898b74c78 emit_x64: Get rid of indirect includes 2016-12-05 12:46:22 +00:00
MerryMage
54d051977f emit_x64: Use movdqa instead of movaps in EmitPackedSubU8
While movaps and movdqa are behaviourly equivalent, using movaps may incur
a domain crossing penalty on some microarchitectures. This is because
movaps is an instruction in the floating-point domain while the following
instructions are in the integer domain.
2016-12-05 01:00:51 +00:00
MerryMage
52e1445f43 Implement USUB8 2016-12-05 00:29:15 +00:00
MerryMage
5c1aab1666 Implement CLZ
Includes tests
2016-12-04 22:56:33 +00:00
MerryMage
1a1646d962 Implement UADD8 2016-12-04 20:52:33 +00:00
MerryMage
7cad6949e7 IR: Implement new pseudo-operation GetGEFromOp 2016-12-04 20:52:06 +00:00
MerryMage
25f21b5371 emit_x64: Inline nzcv computation into EmitFPCompare32 and EmitFPCompare64 2016-12-04 11:43:31 +00:00
MerryMage
cede5e442a emit_x64: Use xorps/xorpd when argument to TransferToFP32/TransferToFP64 is an immediate zero 2016-12-03 11:41:10 +00:00
MerryMage
e166965f3e Implement VCMP 2016-12-03 11:41:09 +00:00
MerryMage
f2fe376fc6 Support 64-bit immediates 2016-12-03 11:29:50 +00:00
Mat M
95f34c683c reg_alloc: Remove unnecessary breaks after returns (#54) 2016-12-02 19:14:44 +00:00
Mat M
de1f831d79 microinstruction: Make use_count private (#53)
Makes the operation a part of the direct interface.
2016-11-30 21:51:06 +00:00
MerryMage
3621a925b2 reg_alloc: Register allocator related constraints belong with the rest of the register allocator
HostLocToReg64 contained two DEBUG_ASSERTs invloving constraints that
really belonged to the register allocator.

The register allocator prevents allocation of RSP and R15 because those
are reserved for the stack pointer and the state pointer respectively.
2016-11-30 19:42:41 +00:00
MerryMage
5f11b4f50e HostLoc: R15 is a GPR 2016-11-30 18:38:03 +00:00
Sebastian Valle
14eb70d7e4 VFP: Fixed the VCVT behavior when converting from unsigned 32-bit values. (#51)
Use a 64-bit register to hold the values so that we don't end up interpreting them as signed values.
2016-11-27 23:25:50 +00:00
Merry
0ff8c375af Implement UHSUB8 and UHSUB16 (#48) 2016-11-26 18:27:21 +00:00
Merry
cb17f9a3ed Implement SHADD8 and SHADD16 (#47) 2016-11-26 18:12:29 +00:00
MerryMage
c0c1bb1094 Implemented UHADD16 2016-11-26 11:28:20 +00:00
Sebastian Valle
4d44474ad4 Implemented the ARM UHADD8 instruction. (#45)
The x64 implementation uses the SSSE3 instruction PSHUFB.
A non-SSE fallback is provided in case the CPU doesn't support it.
2016-11-25 20:32:22 +00:00
MerryMage
b6f7b8babd ir: Implement GetGEFlags, SetGEFlags 2016-11-23 19:44:27 +00:00
Mat M
6a2174ebfa Add missing explicit specifiers (#27) 2016-09-07 12:08:48 +01:00
Mat M
6e0f27a500 types: Add helpers for determining single and doubleword extension registers (#26) 2016-09-07 12:08:35 +01:00
Mat M
5bc9ce544f arm_types: Move into arm folder (#25) 2016-09-06 00:52:33 +01:00
MerryMage
1f61a3d7bc jitstate: Rename fields s/guest_FPSCR/FPSCR/ 2016-09-05 14:42:21 +01:00
Mat M
6d53bb6d7e arm_types: Split out LocationDescriptor (#20)
This isn't really an ARM-specific type, since it's used to indicate a
Block location.
2016-09-05 11:54:09 +01:00
Mat M
858796a029 Eliminate variable shadowing warnings with MSVC (#17) 2016-09-04 11:30:57 +01:00
Mat M
7f9a0c3c38 Remove unnecessary explicit includes (#16) 2016-09-03 21:48:03 +01:00
Mat M
26db11cd71 reg_alloc: Use a strongly-typed enum for representing OpArg type (#15) 2016-09-03 18:30:03 +01:00
Mat M
05b189bc26 arm_types: Specialize std::hash for LocationDescriptor (#14)
Same thing, but with the benefit of working with anything that uses
std::hash by default.
2016-09-03 12:48:47 +01:00
Mat M
8c4df46580 FPSCR: Make value constructor explicit (#13)
Maintains consistency between the PSR helper.
2016-09-03 12:48:31 +01:00
Mat M
6ec651498d arm: Add PSR helper type (#3) 2016-09-02 17:34:33 +01:00
Mat M
1e781d911a reg_alloc: const correctness (#8) 2016-09-02 17:30:01 +01:00
MerryMage
711b3e29d3 interface: Allow ClearCache to be called at any time 2016-09-02 10:59:33 +01:00
Mat M
fb6d838bd9 dynarmic: Remove poison_memory ClearCache parameter (#1)
Unused since the switch to Xbyak
2016-09-01 09:47:09 +01:00
MerryMage
4321e829d1 callbacks: Add user_arg argument to InterpreterFallback 2016-09-01 02:00:08 +01:00
MerryMage
3b5c43b427 Optimization: Read page-table directly for memory access 2016-09-01 00:58:02 +01:00
MerryMage
57169ec093 abi: Implement ABI_PushCallerSaveRegistersAndAdjustStack and ABI_PopCallerSaveRegistersAndAdjustStack 2016-09-01 00:57:22 +01:00
MerryMage
702e181b35 backend_x64/abi: Reversing XMM list leads to incorrect ordering 2016-08-31 23:06:49 +01:00
MerryMage
b10c438e8e jitstate: Remove code argument from ResetRSB 2016-08-31 21:57:33 +01:00
Lioncash
ea6a4e82b5 block_of_code: Make CallFunction accept function pointers only 2016-08-31 21:51:44 +01:00
Lioncash
37d64f0c86 hostloc: Simplify static_assert 2016-08-28 22:10:23 +01:00
Lioncash
f2bf795876 intrusive_list: Interface changes
- Remove the root pointer from iterators.

This is unnecessary, since the only way to get a valid iterator is
either from a node itself (it transiently becomes an iterator via the
underlying interface), or through the iterator interface for the list.
This should also result in better code generation, as each increment or
decrement of an iterator is now branchless.

- Remove iterator_to

This is actually a pretty dangerous function, since it would immediately
create an iterator into the list using the given item, even if it's not
actually part of the list. This was only left around due to lack of
type handling around constructors.

- Add other overloads for erase() and remove()

Now handles iterators, pointers, and references.
2016-08-28 20:56:40 +01:00
MerryMage
7912a79fa5 emit_x64: align before emitting blocks 2016-08-27 11:04:43 +01:00
MerryMage
41c8dabf0b block_of_code: nop should probably default to a size of 1 2016-08-27 10:57:48 +01:00
MerryMage
dca3b2f079 Implement VMRS and VMSR 2016-08-26 22:47:54 +01:00
MerryMage
814348371e emit_x64: EmitX64::Emit: block.Location() returns by value 2016-08-26 19:43:29 +01:00
MerryMage
4f6ea715b2 emit_x64: EmitX64::Emit doesn't need descriptor argument 2016-08-26 19:14:25 +01:00
Lioncash
32c24d2cb3 Use 'false' instead of '0' in asserts 2016-08-26 18:52:08 +01:00
MerryMage
ba31f43672 reg_alloc: UseDefOpArgXmm: default value for argument desired_location should be any_xmm, not any_gpr 2016-08-26 18:50:08 +01:00
MerryMage
7fedf04e79 reg_alloc: Deduplicate constants in RegAlloc::HostCall that were already defined by abi.h 2016-08-26 18:43:50 +01:00
MerryMage
59a8e14d1c reg_alloc: Correct OpArg::setBit for Reg 2016-08-26 15:23:38 +01:00
MerryMage
065c53ebfc emit_x64: Make ZeroIfNaN64 branchless 2016-08-26 15:23:08 +01:00
MerryMage
9901ed0f51 block_of_code: Optimize nops 2016-08-26 13:46:19 +01:00
Lioncash
0102951bdd Convert formatting over to fmtlib 2016-08-26 13:13:19 +01:00
MerryMage
ed3a686d1d Implement public header files 2016-08-26 00:44:50 +01:00
MerryMage
656d4f7252 emit_x64: inhibit_emission is obsolete
Not used anymore; unused ever since intrusive lists were introduced.
2016-08-25 23:24:16 +01:00
MerryMage
4322c0907c microinstruction: Rename FindUseWithOpcode to GetAssociatedPseudoOperation, encapsulate associated variables 2016-08-25 21:08:47 +01:00
MerryMage
922d1fd198 Merge branch 'xbyak' 2016-08-25 16:54:48 +01:00
MerryMage
d04b9eaa81 backend_x64/block_of_code: Reset labels when ClearCache() is called 2016-08-25 16:18:18 +01:00
MerryMage
e32812cd00 Port x64 backend to xbyak 2016-08-25 16:18:17 +01:00
Lioncash
0e12fb6a56 basic_block: Move all variables behind a public interface 2016-08-25 16:14:37 +01:00
MerryMage
13908c5a58 reg_alloc: Insert braces around DEBUG_ASSERT
DEBUG_ASSERT becomes an empty statement in release-mode; an if statement
with an empty statement produces a compiler warning.
2016-08-25 13:09:18 +01:00
MerryMage
dc26afbd7e translate_arm: Translate more than one conditional instruction in a block 2016-08-25 13:05:33 +01:00
Lioncash
1395baefa9 interface: Return register files by const reference
Prevents unnecessary copies where they aren't particularly required.
2016-08-25 12:51:41 +01:00
MerryMage
22cca5ff72 emit_x64: Actually advance RSB pointer 2016-08-24 23:19:47 +01:00
Lioncash
eba3a06d80 frontend: Introduce FPSCR register helper class
Encapsulates all of the FPSCR state.
2016-08-24 20:51:14 +01:00
MerryMage
b5a86889cd Implement VCVT 2016-08-23 22:20:04 +01:00
MerryMage
c8b2f63c93 get_set_elimination_pass: Eliminate unnecessary gets/sets of extended registers 2016-08-23 15:57:20 +01:00
MerryMage
8d1b9f32ca Standardize indentation of switch statments 2016-08-23 12:19:27 +01:00
MerryMage
72250b119f backend_x64/block_of_code: Add more floating point constants
* MFloatPositiveZero32
* MFloatPositiveZero64
* MFloatMinS32
* MFloatMaxS32
* MFloatMinU32
* MFloatMaxU32
2016-08-22 15:54:19 +01:00
Lioncash
1abe881921 basic_block: Add proxy member functions for the instruction list
Currently basic block kind of acts like a 'dumb struct' which makes things
a little more verbose to write (as opposed to keeping it all in one place,
I guess). It's also a little wonky conceptually, considering a block is
composed of instructions (i.e. 'contains' them).

So providing accessors that make it act more like a container can make working
with algorithms a little nicer. It also makes the API a little more
defined.

Ideally, the list would be only available through a function, but
currently, the pool allocator is exposed, which seems somewhat odd,
considering the block itself should manage its overall allocations
(with placement new, and regular new), rather than putting that
sanitizing directly on the IR emitter (it should just care about emission,
not block state). However, recontaining that can be followed up with,
as it's very trivial to do.
2016-08-22 13:44:56 +01:00
Lioncash
86f803da04 reg_alloc: Use Inst's HasUses() function where applicable 2016-08-20 21:26:09 +01:00
Lioncash
36a0ad5bc2 reg_alloc: const correctness for ValueLocation() 2016-08-19 19:33:57 +01:00
MerryMage
2d6a86e43c Remove <cassert> 2016-08-19 01:53:24 +01:00
MerryMage
269160ef0d emit_x64: Clear RSB-related caches when ClearCache() is called 2016-08-18 18:18:44 +01:00
MerryMage
1a3f3ac435 emit_x64: Correct behaviour of PackedOperation for immediate argument
x64 MOVD_xmm does not support an immediate oparg
2016-08-18 18:17:17 +01:00
MerryMage
b2de47954b EmitX64: Emit correct cycle count on cond failure 2016-08-18 18:16:18 +01:00
Lioncash
841098a0bc ir: separate components out a little more 2016-08-17 20:46:21 +01:00
Lioncash
cbd99e4367 jitstate: Use std::array's fill() in ResetRSB
Performs the equivalent behavior
2016-08-17 10:10:43 +01:00
Lioncash
74ee92ee38 jitstate: const correctness
GetReturnFromRunCodeAddress is const qualified, so this can accept a const
pointer. This also allows for making the constructor accept a const
pointer as well.
2016-08-17 10:10:43 +01:00
Lioncash
439619c827 reg_alloc: Make GetRegLoc return by const reference
Considering a HostLocInfo instance houses a std::vector, every time this
function is called can cause a potential heap allocation.

This can be somewhat unnecessary because this function is only used to query
for information we already have.

Considering this is used by several other internal query functions such as
IsRegisterOccupied, IsRegisterAllocated, and IsLastUse, this will result
in better codegen (returning an address is just 3 instructions excluding
the ret instruction for returning, meanwhile heap alloc can be 60+).

This also renames the function to have the same name as its non-const
counterpart, since overloading will just select the correct function
instead of putting that onus on the developer.
2016-08-17 10:08:08 +01:00
MerryMage
0ebb572e2d Optimization: Make RSB a ring buffer instead of a stack 2016-08-15 15:48:22 +01:00
MerryMage
7d7ac0af71 Optimization: Make SVC use RSB 2016-08-15 15:02:08 +01:00
MerryMage
6c45619aa1 Optimization: Implement terminal LinkBlockFast 2016-08-15 14:33:17 +01:00
MerryMage
624e84fa09 Optimization: Tweak RSB 2016-08-15 14:08:06 +01:00
MerryMage
070298b948 Optimization: bugfix! Return Stack Buffer location hash calculation was incorrect 2016-08-15 13:21:58 +01:00
MerryMage
e164ede4dc TranslateArm: Implement MRS, MSR (imm), MSR (reg) 2016-08-15 11:50:49 +01:00
MerryMage
8fc21f481a RegAlloc: Handle case when def is unused 2016-08-13 01:55:03 +01:00
MerryMage
d43d97b990 EmitX64/EmitPushRSB: Assert that patch location is of correct size 2016-08-13 00:52:31 +01:00
MerryMage
960d14d18e Optimization: Implement Return Stack Buffer 2016-08-13 00:10:23 +01:00
bunnei
8e68e6fdd9 TranslateArm: Implement QADD16/QSUB16/UQADD16/UQSUB16. 2016-08-12 19:00:44 +01:00
bunnei
4b09c0d032 TranslateArm: Implement QADD8 and UQADD8. 2016-08-12 19:00:44 +01:00
bunnei
127fbe99cb TranslateArm: Implement QSUB8. 2016-08-12 19:00:44 +01:00
bunnei
86fe29c6d2 TranslateArm: Implement UQSUB8. 2016-08-12 19:00:44 +01:00