

@5230

3 years 
nmedfort 
Multithreading support for PabloAST / PabloCompiler?. Requires unique …



@5229

3 years 
cameron 
Distinguishing between StreamType? and StreamSetType?; fixing …



@5227

3 years 
nmedfort 
Rewrite of the CarryManager? to support noncarrycollapsing loops.



@5220

3 years 
cameron 
Automate instance creation



@5217

3 years 
nmedfort 
Merged PabloFunction? and PabloKernel? classes. Updated projects where …



@5212

3 years 
lindanl 
editd for GPU.



@5204

3 years 
nmedfort 
More 32bit fixes.



@5203

3 years 
nmedfort 
Fix for 32bit architectures.



@5202

3 years 
nmedfort 
Initial work on adding types to PabloAST and mutable Var objects.



@5192

3 years 
cameron 
CreateAtomicLoadAcquire?, CreateAtomicStoreRelease? with NVPTX overrides



@5185

3 years 
cameron 
Linear buffers; stdout_kernel; cachealigned alloca, restructured …



@5176

3 years 
cameron 
Engine deletion causes segfaults with LLVM 3.9 changes



@5175

3 years 
cameron 
Some tidy ups and changes to prepare for LLVM 3.9



@5165

3 years 
lindanl 
Add segment pipeline parallel strategy. Move ballot function to IDISA …



@5151

3 years 
lindanl 
Add NVPTX Arch to the framework. Fix directory bug in make check.



@5140

3 years 
cameron 
Bitstream and byte space character class kernels  initial checkin



@5129

3 years 
lindanl 
Bug fix.



@5128

3 years 
lindanl 
Add NVPTX to IDISA.



@5126

3 years 
cameron 
Add getStride to IDISA builders; allows Stride > BlockSize? on GPU



@5121

3 years 
cameron 
New IDISA bitblock functions: bitblock_set_bit, bitblock_mask_from



@5118

3 years 
cameron 
AVX builders should inherit from SSE2 not just SSE



@5117

3 years 
cameron 
bitblock_advance



@5116

3 years 
cameron 
bitblock_advance performance improvement



@5115

3 years 
cameron 
Use bitblock_add_with_carry in carry_manager; add AVX2 implementation



@5114

3 years 
cameron 
bitblock_add, bitblock_advance



@5108

3 years 
cameron 
AVX2 checking via getHostCPUFeatures



@5106

3 years 
cameron 
Use dynamic size_t determination



@5084

3 years 
cameron 
Hex output for integers



@5076

3 years 
cameron 
Updates for kernels with variable output length; stdout kernel



@5073

3 years 
cameron 
Module names include BitBlockTypeName?



@5059

3 years 
cameron 
Pablo Kernel progress



@5037

3 years 
nmedfort 
UnicodeSet? bug fix and compile warning cleanup.



@5033

3 years 
cameron 
Refactor: move grepspecific code out of toolchain



@5026

3 years 
cameron 
Updates for LLVM 3.8; hide unrelated command line options



@5014

3 years 
cameron 
Some updates for compatibility with LLVM 3.8



@5007

3 years 
cameron 
u8u16 progress



@5001

3 years 
nmedfort 
Symbol table work



@4999

3 years 
cameron 
Further hsimd_signmask options using sse2.pmovmskb.128; allow …



@4998

3 years 
cameron 
SSE/SSE2 overrides for hsimd_signmask; support BlockSize?=256 long …



@4997

3 years 
cameron 
Fix AVX2 packh/l



@4996

3 years 
cameron 
Duplicated packh/l code with SSE2 builder buggy; eliminate duplicate code



@4995

3 years 
nmedfort 
More work on symbol table; unexpected bug with 4KiBone page tests …



@4986

3 years 
nmedfort 
First attempt at dynamic segment size intergration.



@4984

3 years 
cameron 
Refactor IDISA, re, pablo toolchain components



@4977

3 years 
cameron 
Clean up



@4975

3 years 
cameron 
Improved packh/packl for i64



@4974

3 years 
nmedfort 
Added the kernel instance class; removed original mmap file access in …



@4968

3 years 
nmedfort 
Some fixes for threading and kernel builder.



@4959

3 years 
nmedfort 
Initial modifications to Pablo Compiler and Kernel Builder to support …



@4958

3 years 
lindanl 
optimize pack for block size of 64.



@4957

3 years 
cameron 
Alternative transposition strategies with AVX2



@4956

3 years 
cameron 
Merge in 512bit hsimd_signmask



@4955

3 years 
cameron 
Improved implementation of simd_pack on AVX2



@4954

3 years 
cameron 
Support for BlockSize?=512 on AVX2



@4944

3 years 
lindanl 
64 bit version.



@4943

3 years 
lindanl 
format register printing.



@4942

3 years 
lindanl 
Remove simdlib. Print register implemented in LLVM IR.



@4937

3 years 
nmedfort 
Check in of misc changes prior to symbol table work.



@4922

3 years 
nmedfort 
Incorporated a few common case boolean optimizations in the Simplifier.



@4907

4 years 
cameron 
Kernels in icgrep: initial checkin



@4903

4 years 
cameron 
movmsk_pd_256 is available with AVX as well as AVX2



@4902

4 years 
cameron 
SSE specialized IDISA functions



@4901

4 years 
cameron 
SSE specialized IDISA functions



@4900

4 years 
cameron 
Dynamic generation of s2p code



@4898

4 years 
cameron 
Add IRBuilder functionality to IDISA_Builder; eliminate separate mBuilder



@4895

4 years 
cameron 
hi_mask in s2p, lo_mask



@4894

4 years 
cameron 
simd_himask



@4893

4 years 
cameron 
Fix



@4892

4 years 
cameron 
Move AVX2 specific IDISA function implementations into subclassed builder



@4891

4 years 
cameron 
Initial checkin of s2p generation



@4881

4 years 
cameron 
Parallel long addition within icgrep improves performance on AVX2



@4879

4 years 
cameron 
Add esimd_bitspread



@4845

4 years 
cameron 
Improved shifts



@4843

4 years 
cameron 
Move genPrintRegister into IDISA builder; fix GCB defs



@4837

4 years 
cameron 
Encapsulation of bitblock and carry pack construction



@4827

4 years 
cameron 
Parameterize carry_manager to inherit BitBlockWidth? from idisa builder



@4821

4 years 
cameron 
Encapsulating allzeroes and allones initializers within IDISA builder



@4750

4 years 
cameron 
Clean ups for compilation with gcc 4.8



@4720

4 years 
cameron 
Implementatation of the Pablo Count operation



@4697

4 years 
cameron 
Further progress on Carry Manager with bit packing



@4665

4 years 
nmedfort 
Made pablo compiler reenterant through alternate compile method that …



@4662

4 years 
cameron 
Add bitblock_any to IDISA, return signmask of N fields as iN



@4655

4 years 
cameron 
Simplify of unit advance with IDISA dslli improves performance



@4653

4 years 
cameron 
packh, packl, signmask



@4652

4 years 
cameron 
mergeh and mergel IDISA functions



@4651

4 years 
cameron 
Initial library for implementing IDISA operations in LLVM IR
