ClrxAsmInvoke - CLRX/CLRX-mirror GitHub Wiki

CLRadeonExtender Assembler

This is CLRadeonExtender assembler. This assembler can assemble code for all Radeon GPU's that based on the GCN1.0/1.1/1.2/1.4 (AMD VEGA) architecture and it can generate AMD Catalyst OpenCL binaries and the GalliumCompute OpenCL binaries. It is compatible with GNU assembler and support the almost GNU assembler's pseudo-operations (directives) including macros and repetitions.

Invoking an assembler

The clrxasm can be invoked in following way:

clrxasm [-63Swam?] [-D SYM[=VALUE]] [-I PATH] [-o OUTFILE] [-b BINFORMAT] [-g GPUDEVICE] [-A ARCH] [-t VERSION] [--defsym=SYM[=VALUE]] [--includePath=PATH] [--output OUTFILE] [--binaryFormat=BINFORMAT] [--64bit] [--gpuType=GPUDEVICE] [--arch=ARCH] [--driverVersion=VERSION] [--llvmVersion=VERSION] [--newROCmBinFormat] [--forceAddSymbols] [--noWarnings] [--alternate] [--buggyFPLit] [--oldModParam] [--noMacroCase] [--wave32] [--policy=VERSION] [--help] [--usage] [--version] [file...]

Input

The assembler read source from many files. If no input file specified an assembler will read source from standard input.

Program options

Following options clrxasm can recognize:

  • -D SYMBOL[=VALUE], --defsym=SYMBOL[=VALUE]

    Define symbol. Value is optional and if it is not given then assembler set 0 by default. This option can be occurred many times to defining many symbols.

  • -I PATH, --includePath=PATH

    Add an include path to search path list. Assembler begins search from current directory and follows to next include paths. This option can be occurred many times to adding many include paths.

  • -o FILENAME, --output=FILENAME

    Set output file name. By default assembler write output to the 'a.out' file.

  • -b BINFORMAT, --binaryFormat=BINFORMAT

    Set output binary format. This settings can be overriden in source code. Assembler accepts following formats: 'amd', 'amdcl2', 'gallium', 'rocm', 'rawcode'.

  • -6, --64bit

    Enable generating of the 64-bit binaries (only for AMD catalyst format).

  • -g GPUDEVICE, --gpuType=GPUDEVICE

    Choose device type. Device type name is case-insensitive. List of supported GPUs: Baffin, Bonaire, CapeVerde, Carrizo, Dummy, Ellesmere, Fiji, GFX700, GFX701, GFX801, GFX802, GFX803, GFX804, GFX900, GFX901, GFX902, GFX903, GFX904, GFX905, GFX906, GFX907, Goose, Hainan, Hawaii, Horse, Iceland, Kalindi, Mullins, Oland, Pitcairn, Polaris10, Polaris11, Polaris12, Polaris20, Polaris21, Polaris22, Raven, Spectre, Spooky, Stoney, Tahiti, Tonga, Topaz, Vega10, Vega11, Vega12, Vega20, GFX1000, GFX1010, GFX1011, Navi10.

  • -A ARCH, --arch=ARCH

    Choose device architecture. Architecture name is case-insensitive. List of supported architectures: SI, VI, CI, VEGA, VEGA20, Navi, NaviDL, GFX6, GFX7, GFX8, GFX9, GFX906, GFX10, GFX1011, GCN1.0, GCN1.1, GCN1.2, GCN1.4, GCN1.4.1, GCN1.5, GCN1.5.1.

  • -t VERSION, --driverVersion=VERSION

    Choose AMD Catalyst OpenCL driver version. Version can retrieved from clinfo program that display field 'Driver version' where version is. Version is number in that form: MajorVersion*100 + MinorVersion.

  • --llvmVersion=VERSION

    Choose LLVM compiler version. Version can be retrieved from clinfo program that display field Version. Version is number in that form: MajorVersion*100 + MinorVersion.

  • --newROCmBinFormat

    Enable new ROCm binary format.

  • -S, --forceAddSymbols

    Add all non-local symbols to binaries. By default any assembler does not add any symbols to keep compatibility with original format.

  • -w, --noWarnings

    Do not print all warnings.

  • -a, --alternate

    Enable alternate macro syntax.

  • --buggyFPLit

    Choose old and buggy floating point literals rules (to 0.1.2 version) for compatibility.

  • --oldModParam

    Choose old modifier parametrization that accepts only 0 and 1 values (to 0.1.5 version) for compatibility.

  • -m, --noMacroCase

    Do not ignore letter's case in macro names (by default is ignored).

  • -3, --wave32

    Set wavefront size as 32 elements (apply only for GFX10 devices).

  • --policy=VERSION

    Set CLRX policy version.

  • -?, --help

    Print help and list of the options.

  • --usage

    Print usage for this program

  • --version

    Print version

Environment

Following environment variables impacts on assembler work:

  • CLRX_AMDOCL_PATH

    Path to AMDOCL (AMD OpenCL implementation) shared library (libamdocl32.so, libamdocl64.so, amdocl.dll or amdocl64.dll).

  • CLRX_MESAOCL_PATH

    Path to Mesa3D Gallium OpenCL (libMesaOpenCL.so or libOpenCL.so) shared library.

  • CLRX_LLVMCONFIG_PATH

    Path to llvm-config program.

Output

The assembler generates single output binary. If no output specified an assembler will generate a.out binary file. clrxasm returns 0 if succeeded, otherwise it returns 1 and prints the error messages to stderr.

Sample usage

Following sample usages:

  • clrxasm -o output.clo source.clrx

    Assemble source code 'source.clrx' to binary output 'output.clo'.

  • clrxasm -bamdcl2 -6 -t240500 -o output.clo source.clrx

    Assemble source code 'source.clrx' to AMD OpenCL 2.0 binary format output 'output.clo' with 64-bits and for driver version 240500.