suyu/dynarmic
26
1
Fork 1
Code Issues Pull requests Projects Releases Packages Wiki Activity
1309 commits 2 branches 0 tags 34 MiB
Commit graph

81 commits

Author SHA1 Message Date
Lioncash
d65b056eba Simplify fallback case for EmitVectorSetElement64() 2020-04-22 20:46:21 +01:00
Lioncash
46cb0d813b emit_x64_vector: Append 'v' prefix onto movq in AVX path 
This is something I missed when adding in the AVX broadcast code.
 2020-04-22 20:46:21 +01:00
Lioncash
f939bd0228 emit_x64_vector{_floating_point}: Add helper alias for sizing arrays relative to vector width 
Avoids needing to remember to specify the proper size of the arrays, all
that's needed is to specify the type of the array and the size will
automatically be deduced from it. This helps prevent potential oversized
or undersized arrays from being specified.
 2020-04-22 20:46:21 +01:00
Lioncash
7797bc2fb2 emit_x64_vector: Use non-scratch Use* variants of registers within EmitVectorUnsignedAbsoluteDifference() 
In some cases, a register isn't modified, depending on the branch taken,
so we can signify this by using the non-scratch variants in certain
cases.
 2020-04-22 20:46:20 +01:00
MerryMage
9dba273a8c A64: Implement SADDLP 2020-04-22 20:46:19 +01:00
MerryMage
70ff2d73b5 A64: Implement UADDLP 2020-04-22 20:46:19 +01:00
MerryMage
5563bbbd79 A64: Implement EXT 2020-04-22 20:46:19 +01:00
Lioncash
35026a6ce3 emit_x64_vector: Vectorize fallback path for EmitVectorMaxU32() 2020-04-22 20:46:19 +01:00
Lioncash
0bee648b4f emit_x64_vector: Deduplicate a bit of code in EmitVectorSetElement{8, 32, 64} functions 
Given both branches are the same, we can hoist out the common code.
 2020-04-22 20:46:18 +01:00
Lioncash
b6e223fc58 emit_x64_vector: Deduplicate a bit of code within EmitVectorGetElement8() 
Given both branches use the same destination register size, we can hoist
the common code out.
 2020-04-22 20:46:18 +01:00
Lioncash
cf188448d4 emit_x64_vector: Vectorize fallback case in EmitVectorMultiply64() 
Gets rid of the need to perform a fallback.
 2020-04-22 20:46:18 +01:00
Lioncash
954deff2d4 emit_x64_vector: Add break to final case in EmitVectorRoundingHalvingAddUnsigned() 
This doesn't alter behavior but does make the code better if anything
else is ever added to this function in the future.
 2020-04-22 20:46:18 +01:00
Lioncash
bc718c5b28 ir: Add opcodes for performing rounding halving adds 2020-04-22 20:46:18 +01:00
Lioncash
054549da35 emit_x64_vector: Simplify AVX-512 codepath in EmitVectorMultiply64 
I realized I introduced a helper for simple AVX operation emitting, so
use that instead of writing it all out long-form.
 2020-04-22 20:46:18 +01:00
Lioncash
6de5ed96e5 emit_x64_vector: Emit VPMULLQ in EmitVectorMultiply64 on AVX-512{DQ, VL} capable CPUs 
Shortens code-gen down to a single instruction in the 64-bit path.
 2020-04-22 20:46:18 +01:00
Lioncash
1e10017f4b ir: Add opcodes for signed absolute differences 2020-04-22 20:46:17 +01:00
Lioncash
44a5f8095a ir: Add opcodes for performing vector halving subtracts 2020-04-22 20:46:17 +01:00
Lioncash
27a6d5f6ce emit_x64_vector: Use VPOPCNTB in EmitVectorPopulationCount() if AVX-512 BITALG is available 2020-04-22 20:46:17 +01:00
Lioncash
089096948a ir: Add opcodes for performing halving adds 2020-04-22 20:46:17 +01:00
Lioncash
3d00dd63b4 emit_x64_vector: Emit VPMINSQ and VPMINUQ for 64-bit vector min operations if AVX-512VL is available 2020-04-22 20:46:17 +01:00
Lioncash
b97b71b8aa emit_x64_vector: Emit VPMAXSQ and VPMAXUQ for 64-bit vector max operations if AVX-512VL is available 2020-04-22 20:46:17 +01:00
Lioncash
0f067b7330 emit_x64_vector: Emit VPABSQ in EmitVectorAbs() for the 64-bit case if AVX-512VL is available 2020-04-22 20:46:17 +01:00
Lioncash
d4ee878cbd emit_x64_vector: Use VPSRAQ in EmitVectorArithmeticShiftRight64() if AVX-512VL is available 2020-04-22 20:46:17 +01:00
Lioncash
51e4f1d9db emit_x64_vector: Vectorize fallback path of EmitVectorMaxS32() 2020-04-22 20:46:17 +01:00
Lioncash
c692ccdd6d emit_x64_vector: Vectorize fallback path of EmitVectorMaxS8() 2020-04-22 20:46:17 +01:00
Lioncash
b194313d8c emit_x64_vector: Vectorize fallback path in EmitVectorMinU32() 2020-04-22 20:46:17 +01:00
Lioncash
7ceda6d919 emit_x64_vector: Vectorize fallback path in EmitVectorMinU16() 2020-04-22 20:46:17 +01:00
Lioncash
cda85a1da0 emit_x64_vector: Vectorize fallback path in EmitVectorMinS32() 2020-04-22 20:46:17 +01:00
Lioncash
6e08eed210 emit_x64_vector: Vectorize fallback path in EmitVectorMinS8() 2020-04-22 20:46:17 +01:00
Lioncash
0fb6dce689 emit_x64_vector: Remove unnecessary if constexpr expression in LogicalVShift 
This can simply be merged with the previous one.
 2020-04-22 20:46:17 +01:00
Lioncash
5b71b1337b emit_x64_vector: Avoid left shift of negative value in LogicalVShift 
Now that we handle the signed variants, we also have to be careful about left shifts with negative values,
as this is considered undefined behavior.
 2020-04-22 20:46:17 +01:00
Lioncash
21974ee57e backend_x64/ir: Amend generic LogicalVShift() template to also handle signed variants 
Also adds IR opcodes to dispatch said variants
 2020-04-22 20:46:17 +01:00
Lioncash
af28e89a13 emit_x64_vector: Vectorize fallback path in EmitVectorMaxU16() 2020-04-22 20:46:17 +01:00
Lioncash
0d20423ad5 emit_x64_vector: Vectorize non-SSE4.1 fallback path for VectorMultiply32() 2020-04-22 20:46:17 +01:00
Lioncash
d70ee7c0d1 emit_x64_vector: Use VBPROADCAST where applicable and available 
Uses the instruction that does what it says in its name if available. Allows avoiding the use
of a scratch register in EmitVectorBroadcast8() and EmitVectorBroadcastLower8()'s SSSE3 path.
 2020-04-22 20:46:17 +01:00
Lioncash
26d77c6f09 ir: Add opcodes for performing vector deinterleaving 2020-04-22 20:46:17 +01:00
Lioncash
87ca63699f emit_x64_vector: Emit PMAXUD in EmitVectorMaxU32 on SSE4.1-capable CPUs 2020-04-22 20:46:17 +01:00
Lioncash
f17702f608 emit_x64_vector: Emit PMINUD in EmitVectorMinU32 on SSE4.1-capable CPUs 2020-04-22 20:46:17 +01:00
Lioncash
596a8dd1dd emit_x64_vector: Emit PMINSD in EmitVectorMinS32 on SSE4.1-capable CPUs 
Provides a better alternative to a fallback operation.
 2020-04-22 20:46:17 +01:00
Lioncash
75fd4eaaaa emit_x64_vector: Get rid of some magic numbers in loop bounds 2020-04-22 20:46:17 +01:00
Lioncash
7b80ac25eb emit_x64_vector: Generify variable shift functions 2020-04-22 20:46:17 +01:00
Lioncash
64b1f2d468 ir: Add opcode for reversing bits in a vector 2020-04-22 20:46:15 +01:00
Lioncash
e33dcce14a ir: Add opcodes for performing vector absolute values 2020-04-22 20:46:15 +01:00
MerryMage
3472f371df IR: Implement VectorExtract, VectorExtractLower IR instructions 2020-04-22 20:46:15 +01:00
Lioncash
ad5cf584ce ir: Add opcodes for performing vector unsigned absolute differences 2020-04-22 20:46:15 +01:00
Lioncash
701f43d61e IR: Add opcodes for interleaving upper-order bytes/halfwords/words/doublewords 
I should have added this when I introduced the functions for interleaving
low-order equivalents for consistency in the interface.
 2020-04-22 20:46:15 +01:00
Lioncash
b8d5765f9b emit_x64_vector: Fix typo in VectorShuffleImpl 
This is supposed to be pshufd, not pshufw (which only allows a 64-bit operand)
 2020-04-22 20:46:15 +01:00
Lioncash
6b0010c940 ir: Add IR opcodes for emitting vector shuffles 
This uses the ARM terminology for sizes (Halfword -> 2 bytes, Word -> 4 bytes)
as opposed to the x86 terminology of (Word -> 2 bytes, Double word -> 4 bytes)
 2020-04-22 20:46:15 +01:00
MerryMage
147284427b A64: Implement USHL 2020-04-22 20:46:15 +01:00
MerryMage
710d09471b IR: Add IR instruction ZeroVector 2020-04-22 20:46:15 +01:00