llvm-ve-rv 1.7.0

• Based on llvm-ve-1.15.0
• OpenMP target offloading for C and C++ with vhcall (VE to VH) or veo (VH to VE). (Sotoc offloading with NCC still available).
• More robust tail-predicated, outer-loop vectorization with RV.
• Preview feature: Automatic outer-loop vectorization (requires int64_t iteration variable loops).

Refer to the README for details, build & usage instructions.

llvm-ve-1.15.0

• [VE] Fix vmv intrinsic problem if a given scalar immediate value is from 64 to 127
• [VE] Fix vector register related problem caused by spill/restore of multiple vector registers with multiple vector lengths
• [VE] Change to expand arguments and return values to 64 bits width to follow ABI correctly

github_release_20200605

• [VE] Minor bug fixes
• [VE] Support disassembler (llvm-objdump)

llvm-ve-rv 1.6.0

• Auto-vectorization with vector math library calls (SLEEF).
• Vector codegen improvements (sqrt, shuffle sequences, ..).
• Merges github_release_20200515.
• Fixes OpenMP offloading for VEO versions > 7.

github_release_20200515: [VE] Update several vector instructions

• [VE] Optimize comparisons and min/max
• [VE] Optimize handling of float immediate values
• [VE] Support NND instruction for better performance
• [VE] Change clang to work with only "-target ve" option

github_release_20200413

1.12.0:

• [VE] Support (m)0 and (m)1 operands
• [VE] Fix OperationAction for UMULO/SMULO
• [VE] Implement all instructions experimentally and support reg+reg+imm format
• [VE] Implement disassembler

llvm-ve-rv 1.5.1

• Sotoc: bug fix (don't crash on nullptr FuncDecls when traversing the AST for target regions).
• Sotoc: use ncc as the default target compiler.

llvm-ve-rv 1.5.0

• C,C++ support.
• Automatic loop vectorization (LLVM Loop Vectorizer, SLP).
• Prototype feature: OpenMP target offloading from VH to VE for C code (by RWTH Aachen).
• Includes the Region Vectorizer for outer-loop and whole-function vectorization (pragma triggered).
• VEL intrinsics.

github_release_20200305

• Support load/store of __vm256 and __vm512
• Change __vm256 and __vm512 from <[48] x double> to <[48] x unsinged long>
• Improve codegen
• Merge upstream

github_release_20200129

Merge branch 'feature/merge-fix-mask' of ve-llvm/llvm-project into de…