/opt/chainguard/nemo-rl/shared/triton/backends/nvidia/bin/ptxas-blackwell

Usage  : ptxas-blackwell [options] <ptx file>,...

Options
=======

--Ofast-compile <0|min|mid|max>                     (-Ofc)
        Specify fast-compile level for device code, which controls the tradeoff between
        compilation speed and runtime performance by disabling certain optimizations
        at varying levels.
        Levels:
                'max': Focus only on the fastest compilation speed, disabling many
        optimizations.
                'mid': Balance compile time and runtime, disabling expensive optimizations.
                'min': More minimal impact on both compile time and runtime, minimizing
        some expensive optimizations.
                '0': Disables fast-compile.
        The option is disabled by default.
        Default value:  '0'.

--allow-expensive-optimizations <true|false>        (-allow-expensive-optimizations)
        Enable (disable) to allow compiler to perform expensive optimizations
        using maximum available resources (memory and compile-time).
        If unspecified default behavior is to enable this feature for optimization
        level >= O2.

--compile-as-tools-patch                            (-astoolspatch)
        Compile patch code for CUDA tools. For codes compiled with this mode, compiler
        sets maxrregcount to the minimum registers required by ABI.

--compile-only                                      (-c)
        Generate relocatable object.

--compiler-annotations                              (-annotate)
        Annotate compiler-internal information in binary output.

--def-load-cache                                    (-dlcm)
        Default cache modifier on global/generic load.

--def-store-cache                                   (-dscm)
        Default cache modifier on global/generic store.

--device-debug                                      (-g)
        Generate debug information for device code.

--device-function-maxrregcount <archmax/archmin/N>  (-func-maxrregcount)
        When compiling with -c (--compile-only) option, specify the maximum number
        of registers that device functions can use. This option is ignored for whole-program
        compilation and does not affect registers used by entry functions. For device
        functions, this option overrides the value specified by -maxrregcount option.
        If neither device-function-maxrregcount nor maxrregcount is specified, then
        no maximum is assumed.
        Note: Under certain situations, static device functions can safely inherit
        a higher register count from the caller entry function. In such cases, PTXAS
        may apply the higher count for compiling the static function.
        Value less than the minimum registers required by ABI will be bumped up by
        the compiler to ABI minimum limit.


--device-stack-protector <true|false>               (-device-stack-protector)
        Enable stack canaries in device code. Stack canaries make it more difficult
        to exploit certain types of memory safety bugs involving stack-local variables.
        The compiler uses heuristics to assess the risk of such a bug in each function.
        Only those functions which are deemed high-risk make use of a stack canary.
        Default value:  false.

--disable-optimizer-constants                       (-disable-optimizer-consts)
        Disable use of optimizer constant bank.

--disable-warnings                                  (-w)
        Inhibit all warning messages.

--dont-merge-basicblocks                            (-no-bb-merge)
        Normally, ptxas attempts to merge consecutive basic blocks as part of its
        optization process. However, for debuggable code this is very confusing.
        This option prevents basic block merging, at a slight perfomance cost.

--entry <entry function>,...                        (-e)
        Entry function name.

--extensible-whole-program                          (-ewp)
        Do extensible whole program mode.

--fmad <true|false>                                 (-fmad)
        Enables (disables) the contraction of floating-point multiplies and
        adds/subtracts into floating-point multiply-add operations (FMAD, FFMA,
        or DFMA).
        Default value:  true.

--force-load-cache                                  (-flcm)
        Force specified cache modifier on global/generic load.

--force-store-cache                                 (-fscm)
        Force specified cache modifier on global/generic store.

--g-tensor-memory-access-check                      (-g-tmem-access-check)
        Enable tensor memory access checks for tcgen05 operations.

--generate-line-info                                (-lineinfo)
        Generate line-number information for device code.

--gpu-name <gpu name>                               (-arch)
        Specify name of NVIDIA GPU to generate code for. This option also takes virtual
        compute architectures, in which case code generation is suppressed. This
        can be used for parsing only.
        Allowed values for this option:  'compute_100','compute_100a','compute_100f',
        'compute_101','compute_101a','compute_101f','compute_103','compute_103a',
        'compute_103f','compute_120','compute_120a','compute_120f','compute_121',
        'compute_121a','compute_121f','compute_50','compute_52','compute_53','compute_60',
        'compute_61','compute_62','compute_70','compute_72','compute_75','compute_80',
        'compute_86','compute_87','compute_89','compute_90','compute_90a','lto_100',
        'lto_100a','lto_100f','lto_101','lto_101a','lto_101f','lto_103','lto_103a',
        'lto_103f','lto_120','lto_120a','lto_120f','lto_121','lto_121a','lto_121f',
        'lto_50','lto_52','lto_53','lto_60','lto_61','lto_62','lto_70','lto_72',
        'lto_75','lto_80','lto_86','lto_87','lto_89','lto_90','lto_90a','sm_100',
        'sm_100a','sm_100f','sm_101','sm_101a','sm_101f','sm_103','sm_103a','sm_103f',
        'sm_120','sm_120a','sm_120f','sm_121','sm_121a','sm_121f','sm_50','sm_52',
        'sm_53','sm_60','sm_61','sm_62','sm_70','sm_72','sm_75','sm_80','sm_86',
        'sm_87','sm_89','sm_90','sm_90a'.
        Default value:  'sm_52'.

--help                                              (-h)
        Print this help information on this tool.

--input-as-string <ptx string>,...                  (-ias)
        This option allows ptx modules to be passed directly as strings instead of
        via files. It can be used for simple runtime support, or when it is somehow
        not desired to pass the ptx string via the file system.

--legacy-bar-warp-wide-behavior                     (-legacy-bar-warp-wide-behavior)
        This option will cause PTXAS to retain legacy behavior of PTX bar instructionswhere
        if any thread in the warp executes bar instruction it is considered as if
        entire warp has executed it.This option is ignored for programs with .target
        sm_70 or higher.This is a deprecated option and it will be removed in future
        release and is not recommended.

--machine <bits>                                    (-m)
        Specify 64-bit host architecture.

        Allowed values for this option:  64.
        Default value:  64.

--make-errors-visible-at-exit                       (-make-errors-visible-at-exit)
        Generate required instructions at exit point to make memory faults and errors
        visible at exit.

--maxntid <Comma separated list of numbers>,...     (-maxntid)
        Specify maxntid configuration. Note this option is ignored if -maxrregcount
        option is used. For entry functions already using PTX directive .maxntid,
        this option will be ignored.

--maxrregcount <archmax/archmin/N>                  (-maxrregcount)
        Specify the maximum amount of registers that GPU functions can use. Until
        a function- specific limit, a higher value will generally increase the performance
        of individual GPU threads that execute this function. However, because thread
        registers are allocated from a global register pool on each GPU, a higher
        value of this option will also reduce the maximum thread block size, thereby
        reducing the amount of thread parallelism. Hence, a good maxrregcount value
        is the result of a trade-off.
        If this option is not specified, then no maximum is assumed.
        Value less than the minimum registers required by ABI will be bumped up by
        the compiler to ABI minimum limit.
        User program may not be able to make use of all registers as some registers
        are reserved by compiler.

--minnctapersm <N>                                  (-minnctapersm)
        Specify the minimum number of CTAs to be mapped to a single multiprocessor.
        Thisoption is ignored if -maxrregcount option is used. For entry functions
        already usingPTX directive .minnctapersm, this option will be ignored.

--opt-level <N>                                     (-O)
        Specify optimization level.
        Default value:  3.

--optimize-float-atomics                            (-opt-fp-atomics)
        This option enables optimizations on floating point atomics that may affect
        FP precision.

--options-file <file>,...                           (-optf)
        Include command line options from specified file.

--output-file <file>                                (-o)
        Specify name of output file.
        Default value:  'elf.o'.

--override-directive-values                         (-override-directive-values)
        Specify that values provided with the PTX directives are overriden by the
        corresponding option values. Effective only for -minnctapersm, -maxntid and
        -maxrregcount options.

--position-independent-code <true|false>            (-pic)
        Generate position independent code. This mode is enabled by default for whole
        program compilation.
        For other compilation modes, this is disabled by default.
        Default value:  false.

--preserve-relocs                                   (-preserve-relocs)
        This option will make PTXAS to generate relocatable references for variables
        and preserve relocations generated for them in linked executable.

--register-usage-level <0..10>                      (-regUsageLevel)
        Controls the aggressiveness of optimizations that affect register usage.
        ([0..10], default = 5) Higher values aggressively optimize the source program,
        trading off additional register usage for potential improvements in the generated
        code. Lower values inhibit optimizations that aggressively increase register
        usage. This option can work in conjunction with -maxrregcount and CUDA launch
        bounds. This is a BETA feature for advanced users and there is no guarantee
        that the implementation stays consistent between ptxas releases.
        Default value:  5.

--return-at-end                                     (-ret-end)
        Normally, ptxas optimizes return instructions at the end of the program.
        However, for debuggable code this causes problems setting breakpoint at the
        end. This option prevents ptxas from optimizing this last return instruction.

--sp-bounds-check                                   (-sp-bounds-check)
        Generate stack-pointer bounds-checking code sequence. This option is turned
        on automatically when device-debug (-g) or opt-level(-O) 0 is specified.

--split-compile <N>                                 (-split-compile)
        Specify the maximum amount of concurrent threads to be utilized when running
        compiler optimizations. If <N> is 1, this option is ignored. If <N> is 0,
        the number of threads will be the number of CPUs on the machine. This option
        will have minimal (if any) impact on performance of the compiled binary.

--suppress-async-bulk-multicast-advisory-warning    (-suppress-async-bulk-multicast-advisory-warning)
        Suppress advisory warning for .multicast::cluster.

--suppress-debug-info                               (-suppress-debug-info)
        Do not generate debug information sections in final output object file.
        Note: This option is ignored if used without --device-debug or --generate-line-info
        option.

--suppress-double-demote-warning                    (-suppress-double-demote-warning)
        Suppress the warning that is otherwise emitted when a double precision instruction
        is encountered in PTX that is targeted for an SM version that does not have
        double precision support.

--suppress-stack-size-warning                       (-suppress-stack-size-warning)
        Suppress the warning that otherwise is printed when stack size cannot be
        determined.

--verbose                                           (-v)
        Enable verbose mode which prints code generation statistics.

--version                                           (-V)
        Print version information on this tool.

--warn-on-double-precision-use                      (-warn-double-usage)
        Warning if double(s) are used in an instruction.

--warn-on-local-memory-usage                        (-warn-lmem-usage)
        Warning if local memory is used.

--warn-on-spills                                    (-warn-spills)
        Warning if registers are spilled to local memory.

--warning-as-error                                  (-Werror)
        Make all warnings into errors.