Home » Gadgets » Compiler kit: LLVM 5.0 expands Ryzen and AMDGPU support

Compiler kit: LLVM 5.0 expands Ryzen and AMDGPU support

Thursday, September 14th, 2017 | Gadgets

LLVM 5.0 supports other schedulers in various architectures, including the Zen scheduler in AMD's Ryzen processors. Also the schedulers in Intel's CPUs Sandy Bridge and Silvermont as well as AMD's Jaguar processors are better supported. There is also new information for Cavion's 64-bit ARMv8 processor ThunderX2. In addition, optimized code can be generated for the Goldmont processors from Intel, as well as the processors ARMv8-M, ARMv8.1-A, ARMv8.2-A, and ARMv8.3-A. With the support of Gfx9 in AMDGPU, the driver for AMD Vega graphics cards can be compiled in the Linux 3D library Mesa.

The new library Binary format comes from the support and brings the function identify_magic as well as the file_magic structure. In addition, the new library implements structure and type definitions for Dwarf, Elf, Coff, Wasm and MachO. With the also new tool llvm-dll-tool, import libraries can be generated from GNU Style Definition Files. Optimization suggestions in Yaml format can be converted to HTML with the Python tool opt-viewer.py.
New Weak value handles
The value handle WeakVH sets itself to zero as soon as the value to which it points is deleted. The ValueHandle previously managed under the name WeakVH was renamed WeakTrackingVH. The tool for manipulating Microsoft's Program Database format (PDB) has also been given a new name: llvm-pdbdump is now called llvm-pdbutil. With the Debugger LLDB, hardware breakpoints can be set in the future. There are also suggestions for troubleshooting assembler.
LLVM's compiler frontend Clang 5.0 completely supports C ++ 17 when the -std = c ++ 17 parameter is set. In addition the Koroutinen TS are supported, they were still marked as experimental in version 4.0. In addition, additional C ++ keywords and constructs were added in the completion. Open CL also has additional features. Replacing the basic block vectorizer with the superword-level parallelism vectorizer also proceeds: The LLVMAddBBVectorizePass option has been patterned.

Polly lags better
The linkers Lld and Polly also receive numerous improvements. Polly compiled Android – and is doing all the robustness tests. Also Ffmpeg and the LLVM Testsuite can be compiled with Polly. In addition, Polly uses the optimization and diagnostic API to generate suggestions in Yaml format.

Related

Packstations: Amazon wants DHL to compet

Amazon wants to strengthen its delivery services and make the receipt of goods

Sinead O'connor Peter Gabriel

Sinead O'connor Peter Gabriel Sinead O'connor Peter Gabriel Vlogs Sinead O'Connor Mandinka Peter

Gabriel mourns Barcelona victims

Two days after the terrorist attack, Sigmar Gabriel (SPD) visited the crime scene

Papas tracks: Mick Schumacher drives leg

August 14, 2017 – 10:31 am Mick Schumacher has become accustomed to the