Environment Variables ¶

select a compatibility (non-Core) profile with GL version 2.1.

select a compatibility (non-Core) profile with GL version 3.0.

select a Core+Forward Compatible profile with GL version 3.0.

select GL version 3.1 with GL_ARB_compatibility enabled per the driver default.

select GL version 3.1 with forward compatibility and GL_ARB_compatibility disabled.

select GL version 3.1 with GL_ARB_compatibility enabled.

override GL version to X.Y without changing the profile.

select a Core+Forward Compatible profile with GL version X.Y.

select a Compatibility profile with GL version X.Y.

changes the value returned by glGetString(GL_VERSION) for OpenGL

changes the value returned by glGetString(GL_SHADING_LANGUAGE_VERSION) . Valid values are integers, such as 130 . Mesa will not really implement all the features of the given language version if it’s higher than what’s normally reported. (for developers only)

if set, overrides the “executable” string used specifically for driconf option matching. This takes higher precedence over more general process name override (e.g. MESA_PROCESS_NAME).

if set to true , disables the on-disk shader cache. If set to false , enables the on-disk shader cache when it is disabled by default. Note that EGL_ANDROID_blob_cache is still enabled even if on-disk shader cache is disabled.

if set, determines the maximum size of the on-disk cache of compiled shader programs. Should be set to a number optionally followed by K , M , or G to specify a size in kilobytes, megabytes, or gigabytes. By default, gigabytes will be assumed. And if unset, a maximum size of 1GB will be used.

A separate cache might be created for each architecture that Mesa is installed for on your system. For example under the default settings you may end up with a 1GB cache for x86_64 and another 1GB cache for i386.

if set, determines the directory to be used for the on-disk cache of compiled shader programs. If this variable is not set, then the cache will be stored in $XDG_CACHE_HOME/mesa_shader_cache_db (if that variable is set), or else within .cache/mesa_shader_cache_db within the user’s home directory.

if set to true , keeps hit/miss statistics for the shader cache. These statistics are printed when the app terminates.

if set to 1, enables the single file Fossilize DB on-disk shader cache implementation instead of the default Mesa-DB cache implementation. This implementation allows for loading of precompiled cache DBs via MESA_DISK_CACHE_READ_ONLY_FOZ_DBS or MESA_DISK_CACHE_READ_ONLY_FOZ_DBS_DYNAMIC_LIST . This implementation does not support cache size limits via MESA_SHADER_CACHE_MAX_SIZE . If MESA_SHADER_CACHE_DIR is not set, the cache will be stored in $XDG_CACHE_HOME/mesa_shader_cache_sf (if that variable is set) or else within .cache/mesa_shader_cache_sf within the user’s home directory.

if set to 1, enables the multi file on-disk shader cache implementation instead of the default Mesa-DB cache implementation. This implementation increases the overall disk usage. If MESA_SHADER_CACHE_DIR is not set, the cache will be stored in $XDG_CACHE_HOME/mesa_shader_cache (if that variable is set) or else within .cache/mesa_shader_cache within the user’s home directory.

if set with MESA_DISK_CACHE_SINGLE_FILE enabled, references a string of comma separated file paths to read only Fossilize DB shader caches for loading at initialization. The file paths are relative to the cache directory and do not include suffixes, referencing both the cache DB and its index file. E.g. MESA_DISK_CACHE_SINGLE_FILE=filename1 refers to filename1.foz and filename1_idx.foz . A limit of 8 DBs can be loaded and this limit is shared with MESA_DISK_CACHE_READ_ONLY_FOZ_DBS_DYNAMIC_LIST .

Mesa-DB cache eviction algorithm calculates weighted score for the cache items. The weight is doubled based on the last access time of cache entry. By default period of weight doubling is set to one month. Period value is given in seconds.

if set with MESA_DISK_CACHE_SINGLE_FILE enabled, references a text file that contains a new-line separated list of read only Fossilize DB shader caches to load. The list file is modifiable at runtime to allow for loading read only caches after initialization unlike MESA_DISK_CACHE_READ_ONLY_FOZ_DBS . This variable takes an absolute path to the list file. The list file must exist at initialization for updating to occur. Cache files in the list take relative paths to the current cache directory like MESA_DISK_CACHE_READ_ONLY_FOZ_DBS . A limit of 8 DBs can be loaded and this limit is shared with MESA_DISK_CACHE_READ_ONLY_FOZ_DBS .

if set to 1, enables simultaneous use of RW and RW Fossilize DB caches. At first, data will be retrieved from the RO Fossilize cache. If data isn’t found in the RO cache, then it will be retrieved from the RW cache.

changes the Vulkan physical device version as returned in VkPhysicalDeviceProperties::apiVersion .

overrides the WSI present mode clients specify in VkSwapchainCreateInfoKHR::presentMode . Values can be fifo , relaxed , mailbox or immediate .

Forces all swapchains to be headless (no rendering will be display in the swapchain’s window).

causes the Vulkan driver to call abort() immediately after detecting a lost device. This is extremely useful when testing as it prevents the test suite from continuing on with a lost device.

for Vulkan drivers which support real timeline semaphores, this forces them to use a submit thread from the beginning, regardless of whether or not they ever see a wait-before-signal condition.

A comma-separated list of trace types used for offline analysis. The option names are equal to the file extension. Traces are dumped into /tmp . Captures can be triggered by pressing F1 with the application window focused (Currently X11 only) or via MESA_VK_TRACE_FRAME and MESA_VK_TRACE_TRIGGER .

File extension

Offline analysis tool

Supported drivers

Radeon Memory Visualizer

Radeon GPU Profiler

Radeon Raytracing Analyzer

ctxroll

Context rolls

Specifies a trigger file. Creating the file triggers the capture. (e.g. export MESA_VK_TRACE_TRIGGER=/tmp/trigger and then touch /tmp/trigger )

the default GPU is the one used by Wayland/Xorg or the one connected to a display. This variable allows to select a different GPU. It applies to OpenGL and Vulkan (in this case “select” means the GPU will be first in the reported physical devices list). The supported syntaxes are:

For Vulkan it’s possible to append ! , in which case only the selected GPU will be exposed to the application (e.g.: DRI_PRIME=1!).

lspci -nn | grep VGA can be used to know the PCIe bus or ids to use.

If set, overrides the global search-directories used when searching for drirc config files. The user-local one will still be used. Mostly useful for internal debugging.

NIR passes environment variables ¶

The following are only applicable for drivers that uses NIR, as they modify the behavior for the common NIR_PASS and NIR_PASS_V macros, that wrap calls to NIR lowering/optimizations.

a comma-separated list of debug options to apply to NIR shaders. Use NIR_DEBUG=help to print a list of available options.

Mesa Xlib driver environment variables ¶

The following are only applicable to the Mesa Xlib software driver. See the Xlib software driver page for details.

specifies the X visual and depth for RGB mode

Mesa WGL driver environment variables ¶

The following are only applicable to the Mesa WGL driver, which is in use on Windows.

if set to a positive value, specifies the number of MSAA samples to force when choosing the display configuration.

if set to 1, true or yes, disables Win32 error dialogs. Useful for automated test-runs.

if set to 1, true or yes, then the OpenGL implementation will default GL_BLACKHOLE_RENDER_INTEL to true, thus disabling any rendering.

a comma-separated list of named flags, which do various things:

annotate IR in assembly dumps

emit batch information. Can control in which frames batches get dumped using INTEL_DEBUG_BATCH_FRAME_* , where INTEL_DEBUG_BATCH_FRAME_START <= frame < INTEL_DEBUG_BATCH_FRAME_STOP

emit messages about blit operations

emit messages about the blorp operations (blits & clears)

emit messages about buffer objects

emit messages binding tables

flag all buffers to be captured by the kernel driver when generating an error stage after a GPU hang

emit messages about the clip unit (for old gens, includes the CLIP program)

use color in output

dump shader assembly for compute shaders

generate compute shader SIMD32 programs even if workgroup size doesn’t exceed the SIMD16 limit

Add semaphore wait before/after draw call count. INTEL_DEBUG_BKP_BEFORE_DRAW_COUNT or INTEL_DEBUG_BKP_AFTER_DRAW_COUNT can control draw call number. To make test wait forever, we need to set preempt_timeout_ms and i915.enable_hangcheck to zero.

emit messages about performance issues (same as perf )

dump shader assembly for fragment shaders

dump shader assembly for geometry shaders

print information about the driver’s heaps (ANV only)

print instruction hex dump with the disassembly

emit messages about the new L3 state during transitions

dump shader assembly for mesh shaders

don’t generate SIMD8 fragment shader

suppress generation of 16-wide fragment shaders. useful for debugging broken shaders

suppress generation of 32-wide fragment shaders. useful for debugging broken shaders

disable HW performance metric configuration, and anything related to i915-perf (useful when running on simulation)

disable instruction compaction

suppress generation of dual-object geometry shader code

disable fast clears

disable lossless color compression

dump shader assembly to files at each optimization pass and iteration that make progress

emit messages about PIPE_CONTROL instruction usage

emit messages about performance issues

emit messages about GL_AMD_performance_monitor

use performance counter symbols instead of the counter name (counter symbols are like variable names, it’s sometimes easier to work with when you have lots of metrics to collect)

mark all state dirty on each draw call

dump shader assembly for ray tracing shaders

emit messages about the strips & fans unit (for old gens, includes the SF program)

allow developer print traces added by brw_nir_printf to be printed out on the console

enable implementation of software 64bit floating point support

dump usage of sparse resources

force spilling of all registers in the scalar backend (useful to debug spilling code)

force spilling of all registers in the vec4 backend (useful to debug spilling code)

inserts a stall on the GPU after each draw/dispatch command to wait for it to finish before starting any new work.

emit batchbuffer usage statistics

after sending each batch, wait on the CPU for that batch to finish rendering

Insert sync NOP after each instruction. This is only valid for Gfx12+.

dump shader assembly for task shaders

dump shader assembly for tessellation control shaders

force usage of 8-patches tessellation control shaders (only for gfx 9-11)

dump shader assembly for tessellation evaluation shaders

emit messages about textures.

emit messages about URB setup

dump shader assembly for vertex shaders

dump shader assembly for fragment shaders (same as fs )

quiets the OpenCL warnings recommending use of Intel compute-runtime

a comma-separated list of enable/disable flags configuring the output produced by INTEL_DEBUG=bat (use with INTEL_DECODE=+color,-floats ) :

print colored output

try to decode floating point data in buffers

print additional custom information for instructions (usually pulling more information by inspecting memory)

print offsets of instructions

By default, only a standard set of GPU metrics are advertised. This reduces time to collect metrics and hides infrequently used metrics. To enable all metrics, set value to 1.

A comma-separated list of device probe override values. The basic format is <pci-id>[,<pci-id>,...] . The wildcard value of * will specify all known PCI IDs. If ! precedes a PCI ID, or the wildcard value, then the device support will be disabled. All numbers are interpreted in base 16, and a 0x prefix is optional. Values specified later take precedence, so the wildcard probably should only be used at the beginning.

Some examples :

Device 0x1234 would be forced on and 0xabcd would be disabled.

All devices are disabled since the wildcard appears later.

All devices disabled except 0x1234 which is forced on.

All devices are forced on, except 0x1234 which is disabled.

Support for device 0x1234 is forced on since the enable appears later.

If a device requires using INTEL_FORCE_PROBE to force it to load, then Mesa does not have full support for the device. It may have limited, or possibly no functionality within Mesa at this point. It is recommended to upgrade to a Mesa which does not require INTEL_FORCE_PROBE for the device as soon as it is available.

Collects GPU timestamps over common intervals, and generates a CSV report to show how long rendering took. The overhead of collection is limited to the flushing that is required at the interval boundaries for accurate timestamps. By default, timing data is sent to stderr . To direct output to a file:

INTEL_MEASURE=file=/tmp/measure.csv {workload}

To begin capturing timestamps at a particular frame:

INTEL_MEASURE=file=/tmp/measure.csv,start=15 {workload}

To capture only 23 frames:

INTEL_MEASURE=count=23 {workload}

To capture frames 15-37, stopping before frame 38:

INTEL_MEASURE=start=15,count=23 {workload}

Designate an asynchronous control file with:

INTEL_MEASURE=control=path/to/control.fifo {workload}

As the workload runs, enable capture for 5 frames with:

$ echo 5 > path/to/control.fifo

Enable unbounded capture:

$ echo -1 > path/to/control.fifo

and disable with:

$ echo 0 > path/to/control.fifo

Select the boundaries of each snapshot with:

Collects timings for every render (DEFAULT)

Collects timings when the render target changes

Collects timings when batches are submitted

Collects timings at frame boundaries

With INTEL_MEASURE=interval=5 , the duration of 5 events will be combined into a single record in the output. When possible, a single start and end event will be submitted to the GPU to minimize stalling. Combined events will not span batches, except in the case of INTEL_MEASURE=frame .

Collect CPU timestamps instead of GPU timestamps. Prints results immediately instead of waiting for GPU execution. Useful when used with interactive debug to know which frame, or where in frame, you are currently in.

INTEL_MEASURE=cpu {workload}

if set, determines the single DRM modifier reported back to (Vulkan) applications, in order to make selecting modifier deterministic between Vulkan driver and applications. The value can be one of the supported modifiers on a platform, but other values are also acceptable for debug purposes.

if set, determines the directory to be used for overriding shader assembly. The binaries with custom assembly should be placed in this folder and have a name formatted as sha1_of_assembly.bin . The SHA-1 of a shader assembly is printed when assembly is dumped via corresponding INTEL_DEBUG flag (e.g. vs for vertex shader). A binary could be generated from a dumped assembly by i965_asm . For INTEL_SHADER_ASM_READ_PATH to work it is necessary to enable dumping of corresponding shader stages via INTEL_DEBUG . It is advised to use nocompact flag of INTEL_DEBUG when dumping and overriding shader assemblies. The success of assembly override would be signified by “Successfully overrode shader with sha1 <SHA-1>” in stderr replacing the original assembly.

if set, determines the directory to be used for overriding shader assembly. The binaries with custom assembly should be placed in this folder and have a name formatted as sha1_of_assembly.bin . The SHA-1 of a shader assembly is printed when assembly is dumped via corresponding INTEL_DEBUG flag (e.g. vs for vertex shader). A binary could be generated from a dumped assembly by i965_asm . For INTEL_SHADER_ASM_READ_PATH to work it is necessary to enable dumping of corresponding shader stages via INTEL_DEBUG . It is advised to use nocompact flag of INTEL_DEBUG when dumping and overriding shader assemblies. The success of assembly override would be signified by “Successfully overrode shader with sha1 <SHA-1>” in stderr replacing the original assembly.

if set, determines the directory to which the compiled shaders will be dumped. They will be dumped as sha1_of_assembly.bin , where the sha1 values will be the same as can be found in the INTEL_DEBUG output, and can be used for INTEL_SHADER_ASM_READ_PATH input.

Unlike the text form of shader dumping, INTEL_DEBUG does not affect on the list of shaders to dump. All generated shaders are always dumped if INTEL_SHADER_BIN_DUMP_PATH variable is INTEL_SIMD_DEBUG ¶

a comma-separated list of named flags, which control simd dispatch widths:

allow generation of SIMD8 fragment shader

allow generation of SIMD16 fragment shader

allow generation of SIMD32 fragment shader

allow generation of SIMD8 compute shader

allow generation of SIMD16 compute shader

allow generation of SIMD32 compute shader

allow generation of SIMD8 task shader

allow generation of SIMD16 task shader

allow generation of SIMD32 task shader

allow generation of SIMD8 mesh shader

allow generation of SIMD16 mesh shader

allow generation of SIMD32 mesh shader

allow generation of SIMD8 ray-tracing shader

allow generation of SIMD16 ray-tracing shader

allow generation of SIMD32 ray-tracing shader

If none of widths for particular shader stage was specified, then all widths are allowed.

If defined to 0 or false , this will disable pipeline caching, forcing ANV to reparse and recompile any VkShaderModule (SPIRV) it is given.

If defined to 1 or true , this will prevent usage of self modifying command buffers to implement vkCmdExecuteCommands . As a result of this, it will also disable VK_KHR_performance_query .

If defined to 1 or true , this forces ANV to always do kernel relocations in command buffers. This should only have an effect on hardware that doesn’t support soft-pinning (Ivybridge, Haswell, Cherryview).

Specifies up to how many view shaders can be lowered to handle VK_KHR_multiview . Beyond this number, multiview is implemented using instanced rendering. If unspecified, the value default to ANV_SPARSE ¶

By default, the sparse resources feature is enabled. However, if set to 0, false, or no, it will be disabled. Platforms older than Tiger Lake do not support this feature.

On platforms supported by Xe KMD (Lunar Lake and newer) this parameter changes the implementation of sparse resources feature. For i915 there is no option, sparse resources is always implemented with TRTT.

when set to “list” prints the list of devices. when set to “vid:did” number from PCI device. That PCI device is selected as default. The default device is returned as the first device in vkEnumeratePhysicalDevices API. Using “vid:did!” will have the same effect as using the MESA_VK_DEVICE_SELECT_FORCE_DEFAULT_DEVICE variable.

when set to 1, the device identified as default will be the only one returned in vkEnumeratePhysicalDevices API.

EGL environment variables ¶

Mesa EGL supports different sets of environment variables. See the Mesa EGL page for the details.

Gallium environment variables ¶

draws various information on the screen, like framerate, CPU load, driver statistics, performance counters, etc. Set GALLIUM_HUD to help and run e.g. glxgears for more info.

sets the HUD update rate in seconds (float). Use zero to update every frame. The default period is 1/2 second.

toggle visibility via user specified signal. Especially useful to toggle HUD at specific points of application and disable for unencumbered viewing the rest of the time. For example, set GALLIUM_HUD_VISIBLE to false and GALLIUM_HUD_TOGGLE_SIGNAL to 10 ( SIGUSR1 ). Use kill -10 <pid> to toggle the HUD as desired.

Rotate the HUD by an integer number of degrees, the specified value must be a multiple of 90. Default is 0.

specifies an alternate search directory for pipe-loader which overrides the compile-time path based on the install location.

if non-zero, print all the Gallium environment variables which are used, and their current values.

If set, this variable will cause the Trace output to be written to the specified file. Paths may be relative or absolute; relative paths are relative to the working directory. For example, setting it to “trace.xml” will cause the trace to be written to a file of the same name in the working directory.

If enabled while Trace is active, this variable specifies that the threaded context should be traced for drivers which implement it. By default, the driver thread is traced, which will include any reordering of the command stream from threaded context.

If set while Trace is active, this variable specifies a filename to monitor. Once the file exists (e.g., from the user running ‘touch /path/to/file’), a single frame will be recorded into the trace output. Paths may be relative or absolute; relative paths are relative to the working directory.

controls debug output from the Mesa/Gallium state tracker. Setting to tgsi , for example, will print all the TGSI shaders. See src/mesa/state_tracker/st_debug.c for other options.

Override CPU capabilities for LLVMpipe and Softpipe, possible values for x86: nosse ssse3 sse4.1 CLOVER_DEVICE_VERSION_OVERRIDE ¶

overwrites the auto detected OpenCL version of a device. Possible values: CLOVER_DEVICE_CLC_VERSION_OVERRIDE ¶

overwrites the auto detected CLC version. Possible values: CLOVER_EXTRA_BUILD_OPTIONS ¶

allows specifying additional compiler and linker options. Specified options are appended after the options set by the OpenCL program in clBuildProgram .

allows specifying additional compiler options. Specified options are appended after the options set by the OpenCL program in clCompileProgram .

allows specifying additional linker options. Specified options are appended after the options set by the OpenCL program in clLinkProgram .

allows to enable experimental Clover NIR support with the iris driver if set to 1 or true.

overwrites the auto detected OpenCL version of all devices. Specified as major.minor .

a comma-separated list of drivers to enable CL on. An optional list of comma-separated integers can be passed per driver to specify which devices to enable. Examples:

Supported drivers (decent support with maybe a few conformance issues or bugs): iris , llvmpipe , nouveau , panfrost , radeonsi , Experimental drivers (unknown level of support, expect conformance issues or major bugs): RUSTICL_FEATURES ¶

a comma-separated list of features to enable. Those are disabled by default as they might not be stable enough or break OpenCL conformance.

a comma-separated list of debug channels to enable.

enabled.

Limits the amount of threads per dimension in a work-group. Useful for splitting up long running tasks to increase responsiveness or to simulate the lowering of huge global sizes for testing.

a comma-separated list of debug channels to enable.

a comma-separated list of named flags, which do various things:

Dump vertex shader assembly to stderr

Dump fragment shader assembly to stderr

Dump geometry shader assembly to stderr

Dump compute shader assembly to stderr

rasterization is disabled. For profiling purposes.

the Softpipe driver will try to use LLVM JIT for vertex shading processing.

an integer indicating how many threads to use for rendering. Zero turns off threading completely. The default value is the number of CPU cores present.

if set, disables logging to the vmware.log file. This is useful when using Valgrind because it otherwise crashes when initializing the host log feature.

See the driver code for other, lesser-used variables.

WGL environment variables ¶

to set a swap interval, equivalent to calling wglSwapIntervalEXT() in an application. If this environment variable is set, application calls to wglSwapIntervalEXT() will have no effect.

a comma-separated list of named flags, which do various things:

enable LLVM compiler backend

force all allocated buffers to be referenced in submissions

validate the LLVM IR before LLVM compiles the shader

dump fragment shader epilogs

add extra information in bo metadata to help tools (umr)

Enables DCC,FMASK,CMASK,HTILE in situations where the driver supports it but normally does not deem it beneficial.

enable GPU hangs detection and dump a report to $HOME/radv_dumps_<pid>_<time> if a GPU hang is detected

Print image info

show GPU-related information

Mark geometry-affecting outputs as invariant. This works around a common class of application bugs appearing as flickering.

dump internal meta shaders

disable dithering for alpha to coverage

disable primitive binning

disable shaders cache

disable compute queue

disable Delta Color Compression (DCC) on images

disable Delta Color Compression (DCC) on displayable images

do not check OOB access for dynamic descriptors

disable VK_EXT_shader_object

disable fast color/depthstencil clears

disable FMASK compression on MSAA images (GFX6-GFX10.3)

disable VK_EXT_graphics_pipeline_library

disable HIZ for depthstencil images

disable directly recording command buffers in GPU-visible memory

disable mesh shader support on GFX10.3+

disable NGG for GFX10 and GFX10.3

disable NGG culling on GPUs where it’s enabled by default (GFX10.3 only).

disable NGG GS for GFX10 and GFX10.3

skip executing vkCmdTraceRays and ray queries (RT extensions will still be advertised)

disable TC-compat CMASK for MSAA surfaces

disable UMR dumps during GPU hang detection (only with RADV_DEBUG = hang )

disable VRS for flat shading (only on GFX10.3+)

dump LLVM IR before any optimizations

dump vertex shader prologs

dump shaders

dump shader statistics

enable register shadowing

dump SPIR-V

split application-provided fused multiply-add in geometry stages

display info at startup

synchronize shaders after all draws/dispatches

initialize all memory allocated in VRAM as zero

allow to force per-pipeline vertex VRS rates on GFX10.3+. This is only forced for pipelines that don’t explicitly use VRS or flat shading. The supported values are 2x2, 1x2, 2x1 and 1x1. Only for testing purposes.

similar to RADV_FORCE_VRS but allow to configure from a file. If present, this supersedes RADV_FORCE_VRS .

a comma-separated list of named flags, which do various things:

enable the global BO list

enable wave32 for compute shaders (GFX10+)

enable DCC for MSAA images

upload shaders to invisible VRAM (might be useful for non-resizable BAR systems)

forces ray-tracing to be emulated in software on GFX10_3+ and enables rt extensions with older hardware.

enable wave32 for vertex/tess/geometry shaders (GFX10+)

enable local BOs

enable NGG culling on GPUs where it’s not enabled by default (GFX10.1 only).

cache per-stage NIR for graphics pipelines

disable optimizations that get enabled when all VRAM is CPU visible.

enable wave32 for pixel shaders (GFX10+)

enable wave32 for ray tracing shaders (GFX11+)

enable wave64 for ray tracing shaders (GFX10-10.3)

enable optimizations to move more driver internal objects to VRAM.

enable experimental transfer queue support (GFX9+, not yet spec compliant)

enable experimental video decoding support

set the SQTT/RGP buffer size in bytes (default value is 32MiB, the buffer is automatically resized if too small)

set the ray history buffer size when capturing RRA traces (default value is 100MiB, small buffers may result in incomplete traces)

decrease the resolution used for dumping the ray history resolution when capturing RRA traces. This allows for dumping every Nth invocation along each dispatch dimension.

choose the specific pstate to enter when using thread tracing or when acquiring the profiling lock for performance queries.

force GPU clocks to an arbitrary fixed level

force the shader clock to its minimum level

force the memory clock to its minimum level

force GPU clocks to their maximum level, this is the default value

a comma-separated list of named flags, which do various things:

validate the ACO IR at various points of compilation (enabled by default for debug/debugoptimized builds)

disable ACO IR validation in debug/debugoptimized builds

validate register assignment of ACO IR and catches many RA bugs

force emitting waitcnt states if there is something to wait for

force emitting waitcnt dependencies for debugging hazards on GFX10+

disable value numbering

disable various optimizations

disable pre-RA, ILP and VOPD instruction scheduling

disable ILP instruction scheduling

disable VOPD instruction scheduling

print information used to calculate some pipeline statistics

print liveness and register demand information before scheduling

Disable DCC fast clear

disable Delta Color Compression (DCC) on displayable images

Disable DCC for MSAA

Disable DPBB. Overrules the dpbb enable option.

Disable hardware based encoder color format conversion

Enable low latency encoding

Disable tiling

Disable MSAA compression

Disable Hyper-Z

Disable 2D tiling

Print driver information

Print texture info

Print compute info

Print virtual addresses when creating resources

Print vertex shaders

Print pixel shaders

Print geometry shaders

Print tessellation control shaders

Print tessellation evaluation shaders

Print compute shaders

Don’t print the LLVM IR

Don’t print NIR when printing shaders

Don’t print disassembled shaders

Print the LLVM IR before initial optimizations

Use Wave32 for vertex, tessellation, and geometry shaders.

Use Wave32 for pixel shaders.

Use Wave32 for computes shaders.

Use Wave64 for vertex, tessellation, and geometry shaders.

Use Wave64 for pixel shaders.

Use Wave64 for computes shaders.

Enable additional sanity checks on shader IR

Use old-style monolithic shaders compiled on demand

Disable compiling optimized shader variants.

Use ACO as shader compiler when possible

Disable GTT write combining

Check VM faults and dump debug info.

Force VMID reservation per context.

Disable NGG and use the legacy pipeline.

Always use NGG culling even when it can hurt.

Disable NGG culling.

Program WD/IA to switch on end-of-packet.

Disable out-of-order rasterization

Enable DPBB. Enable DPBB for gfx9 dGPU. Default enabled for gfx9 APU and >= gfx10.

add extra information in bo metadata to help tools (umr)

Enable sb backend for compute shaders

Don’t use optimized bytecode (just print the dumps)

Print optimization statistics for shaders

Print IR dumps after some optimization passes

Abort on errors instead of fallback

Use sb disassembler for shader dumps

Disable unsafe math optimizations

Enable NIR with SB optimizer

Print texture info

Enable experimental NIR shaders

Print compute info

Print virtual addresses when creating resources

Print driver information

Print fetch shaders

Print vertex shaders

Print geometry shaders

Print pixel shaders

Print compute shaders

Print tessellation control shaders

Print tessellation evaluation shaders

Don’t print the LLVM IR

Don’t print the TGSI

Don’t print disassembled shaders

Print the LLVM IR before initial optimizations

Enable additional sanity checks on shader IR

Disable compiling optimized shader variants.

Invoke SDMA tests and exit.

Invoke a CP VM fault test and exit.

Invoke a SDMA VM fault test and exit.

Invoke a shader VM fault test and exit.

Disable asynchronous DMA

Disable Hyper-Z

Disable handling of INVALIDATE_RANGE map flags

Disable 2D tiling

Disable tiling

Program WD/IA to switch on end-of-packet.

Use asynchronous DMA for all operations when possible.

Compile one shader variant at shader creation.

Disable GTT write combining

Check VM faults and dump debug info.

Enable unsafe math shader optimizations

a comma-separated list of named flags, which do various things:

Log all consumed nir instructions

Log created R600 IR

Log R600 IR to assembly code creation

Don’t log shader conversion errors

Log shader info (non-zero values)

Log register allocation and lookup

Log shader in and output

Log IR to assembly conversion

Log control flow instructions

Log register merge operations

Skip register merge step

Log texture ops

Log generic translation messages

a comma-separated list of named flags, which do various things:

Print hardware info (printed by default on debug builds

Log fragment program compilation

Log vertex program compilation

Log draw calls

Log SWTCL-specific info

Log rasterizer registers

Log vertex stream registers

Log basic info about textures

Log texture mipmap tree info

Log rasterizer

Log framebuffer

Log fast color clear info

Log HyperZ info

Log scissor info

Log MSAA resources

Use high quality anisotropic filtering

Disable tiling

Disable immediate mode

Disable shader optimizations

Disable fast color clear

Disable zbuffer compression

Disable hierarchical zbuffer

Disable AA compression and fast AA clear

Disable hardware accelerated Transform/Clip/Lighting

a comma-separated list of named flags, which do various things:

Trace work submitted to the GPU to files, using the agxdecode infrastructure. This produces a large volume of data, so should be used with caution. The traces are written to agxdecode.dump , but this can be overridden using AGXDECODE_DUMP_FILE .

Disable 16-bit floating point support. This may workaround application bugs in certain OpenGL ES applications originally written for desktops. If such applications are found in the wild, they should be fixed upstream (if possible) or added in the Mesa-wide driconf (if closed source).

In debug builds only: disable dirty tracking optimizations.

Disable write-combining (force all allocations to be write-through). This may be useful for diagnosing certain performance issues. Note imported buffers may still be write-combined.

a comma-separated list of named flags, which do various things:

Print shaders being compiled at various stages in the pipeline.

Print statistics about compiled shaders.

Disassemble in verbose mode, including additional information that may be useful for debugging.

Include even internal shaders (as produced for clears, blits, and such) when printing shaders. Without this flag, internal shaders are ignored by the shaders and shaderdb flags.

In debug builds only: skip internal intermediate representation validation.

Disable various backend optimizations.

a comma-separated list of named flags for the Rogue compiler, which do various things:

Print the input NIR to stdout.

Print the output of each NIR pass to stdout.

Print the input Rogue IR to stdout.

Print the output of each Rogue IR pass to stdout.

Includes additional details when printing Rogue IR.

Skips the compiler validation step.

Prints all the validation errors instead of stopping after the first.

if set to auto Rogue IR will be colorized if stdout is not a pipe. Color is forced off if set to off / 0 or on if set to on / 1 . Defaults to auto .

Other Gallium drivers have their own environment variables. These may change frequently so the source code should be consulted for details.

Vulkan loader environment variables ¶

These variable are handled by Khronos’ Vulkan loader , not by Mesa , but they are documented here as we reference them in other places in our docs.

Force the loader to use the specific driver JSON files. The value contains a list of delimited full path listings to driver JSON Manifest files and/or paths to folders containing driver JSON files.

See Vulkan loader docs on environment variables .