Releases: KhronosGroup/MoltenVK
Releases · KhronosGroup/MoltenVK
Release for Vulkan SDK 1.3.204
- Add
MVK_HIDE_VULKAN_SYMBOLS
env var to support building MoltenVK with static Vulkan linkage symbols hidden. - Improve accuracy of
VkPhysicalDeviceLimits::timestampPeriod
. - Do not use
MTLEvent
forVkSemaphore
under Rosetta2. - On Apple Silicon (iOS, tvOS & macOS), set
VkPhysicalDeviceProperties::deviceID
from GPU capabilities. - Support compiling MSL 2.4 in runtime pipelines and
MoltenVKShaderConverterTool
. - Fix issue where MSL 2.3 only available on Apple Silicon, even on macOS.
- Fix memory leak of dummy
MTLTexture
in render subpasses that use no attachments. - Fix Metal object retain-release errors in assignment operators.
- Fix use of GPU counter sets on older versions of iOS running on the simulator.
mvk::getShaderOutputs()
inSPRIVReflection.h
support flattening nested structures.- Replaced ASL logging levels with
MVKConfigLogLevel
. MoltenVKShaderConverter
tool support loading tessellation shader files.MoltenVKShaderConverter
tool update to MSL 2.4 by default.- Upgrade to Vulkan 1.3 headers.
- Update
VK_MVK_MOLTENVK_SPEC_VERSION
to version33
. - Update copyright notices to year 2022.
- Update to latest SPIRV-Cross:
- MSL: Add 64 bit support for
OpSwitch
. - MSL: Don't output depth and stencil values with explicit early fragment tests.
- MSL: Fix incorrect MSL
gl_TessCoord
variable type. - MSL: Allow copy array from
UniformConstant
storage. - MSL: Handle signed
atomic min/max
. - Fix
normalize
onhalf3/half2
. - MSL: Handle constant construct of block-like array types.
- MSL: Rethink how opaque descriptors are passed to leaf functions.
- Rework how loop iteration counts are validated.
- Clamp vector element access to vector size.
- Handle aliased names in spec constants.
- Add support for
LocalSizeId
.
- MSL: Add 64 bit support for
Release for Vulkan SDK 1.2.198
- Set maximum point primitive size based on GPU vendor ID, which can now be as large as
511
. - Improved checks for timestamp GPU counter support on older devices.
- Fix incorrect validation error on multilayer
VkImage
marked for rendering, when multilayered-rendering
is not supported on platform, but app doesn't actually attempt to render to multiple layers. - Fix dynamic pipeline state such as
vkCmdSetDepthBias()
sometimes ignoring pipeline dynamic
state flags when called beforevkCmdBindPipeline()
. - Fix incorrect validation of multilayer-rendering validation when
MVKImageView
does not actually perform multilayer-rendering. - Fix issue where dynamic pipeline flags were sometimes read from previously bound pipeline.
- Fix issue where correct base layer and mipmap for attachment was sometimes not being used.
- Add
MTLFence
between Metal encoders and timestamp stage counter BLIT encoder
to ensure previous work is finished before being timestamped. - Fix issue with
BC1_RGB
compressed format where incorrect alpha value returned. - Add
vkGetMTLCommandQueueMVK()
function to expose underlyingMTLCommandQueue
object. - Add
vkSetWorkgroupSizeMVK()
function. - Add unsupported
VkApplicationInfo::apiVersion
warning. - Update to latest SPIRV-Cross version:
- MSL: Add support for
OpSpecConstantOp
opsOpQuantizeToF16
andOpSRem
. - MSL: Return fragment function value even when last SPIR-V Op is discard (
OpKill
). - MSL: Fix location and component variable matching between shader stages.
- MSL: Fix type redirection when struct members are reordered to align with offsets.
- MSL: Remove over-zealous check for struct packing compatibility.
- MSL: Correctly emit user(
clip/cullN
) for clip/cull builtins in tess output struct. - MSL: Don't output depth and stencil values with explicit early fragment tests.
- MSL: Track location component to match
vecsize
between shader stages. - MSL: Selectively enable
fast-math
in MSL code to match Vulkan CTS results. - MSL: Honor
DecorationNoContraction
when compiling usingfast-math
. - MSL: Honor infinities in
OpQuantizeToF16
when compiling usingfast-math
. - MSL: Support synthetic functions in function constants.
- MSL: Improve handling of
INT_MIN/INT64_MIN
literals. - MSL: Consolidate
spvQuantizeToF16()
functions into a single template function. - MSL: Use
vec<T, n>
in templateSpvHalfTypeSelector
for functionspvQuantizeToF16()
. - MSL: Support more usecases for unpacked vectors.
- MSL: Workaround compiler crashes when using
threadgroup bool
. - MSL: Handle non-thread storage class in
Modf/Frexp
pointer versions. - Improve handling of
INT_MIN/INT64_MIN
literals. - Per spec, support undefined behavior for out-of-bounds swizzles.
- Correctly reflect declared buffer size for out of order members.
- Separate (partially) the tracking of depth images from depth compare ops.
- MSL: Add support for
Release for Vulkan SDK 1.2.189
- Changes to how
VkSemaphore
is supported:- Revert to prefer
MTLEvent
forVkSemaphore
, except on NVIDIA, where emulation on CPU is preferred. - Set default value of the
MVK_ALLOW_METAL_FENCES
environment variable to0 (false)
,
- Revert to prefer
- Vulkan timestamp query pools use Metal GPU counters when available.
- Support resolving attachments with formats that Metal does not natively resolve.
- Support stencil-only partial attachment clearing.
- Support alpha-to-coverage without a color attachment.
- Fix issue where swapchain images were acquired out of order under heavy load.
- Fix issue with
vkCmdBlitImage()
from compressed textures. - Fix incorrect translation of clear color values on Apple Silicon.
- Fix swizzle of depth and stencil values into RGBA (
float4
) variable in shaders. - Fix occasional crash when swizzling used but shader swizzling not enabled.
- Fix pipeline barriers not working inside self-dependent subpasses on Apple GPUs.
- Fix GPU race condition when clearing a renderpass input attachment on Apple GPUs.
- Disable
VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT
for
VK_FORMAT_E5B9G9R9_UFLOAT_PACK32
on macOS Apple Silicon. - Fix execution order of
encodeBindings
forbufferBindings
. - Properly ignore non-null pipeline creation pointers that should be ignored.
- Add queue and queue family indices to
MTLCommandBuffer
label. - Update
VK_MVK_MOLTENVK_SPEC_VERSION
to32
. - Update to latest SPIRV-Cross version:
- MSL: Adjust
gl_SampleMaskIn
for sample-shading and/or fixed sample mask. - MSL: Remove redundant path for
SampleMask
. - MSL: Fix setting
SPIRVCrossDecorationInterpolantComponentExpr
decoration. - MSL: Support row-major transpose when storing matrix from constant RHS matrix.
- MSL: Fix casting in constant expressions with different sizes.
- MSL: Fix duplicate
gl_Position
outputs whengl_Position
defined but unused. - MSL: Simplify
spvSubgroupBallot()
. - MSL: Consider that function/private variables can be block-like.
- Track temporary access for
OpArrayLength
result.
- MSL: Adjust
Release for Vulkan SDK 1.2.182
- Add support for extensions:
VK_KHR_imageless_framebuffer
- Advertise Vulkan extension functions only from enabled extensions.
- Make
vkGetPastPresentationTimingGOOGLE()
queuing behavior compliant with Vulkan spec. - Expose
vkGetIOSurfaceMVK()
andvkUseIOSurfaceMVK()
without requiring Objective-C. - Default MoltenVK build using
C++17
and compiler optimization setting-O2
. - API fix to remove
#include "SPIRVReflection.h"
fromSPIRVToMSLConverter.h
header. - Support Xcode 12.5 build settings, build warnings, and SDK change to availability of
[MTLDevice supportsBCTextureCompression]
on Mac Catalyst. - Add support for new AMD devices supporting 32 lanes.
- Improve handling of sampler border color and mirror edge clamp.
- Improve cache hits when matching
SPIRVToMSLConversionConfiguration
structs to each other
to find a cached shader, by only considering resources from the current shader stage. - Rename
kMVKShaderStageMax
tokMVKShaderStageCount
. - Fix crash when requesting
MTLCommandBuffer
logs in runtime debug mode on older OS versions. - Fix synchronization issue with locking
MTLArgumentEncoder
for Metal Argument Buffers. - Fix race condition on submission fence during device loss.
- Fix crash due to incorrect number of attachments when clearing.
- Fix crash using memoryless storage for input attachments on Apple Silicon.
- Fix issue where M1 GPU does not support reusing Metal visibility buffer offsets
across separate render encoders within a single Metal command buffer (Vulkan submit). - On command buffer submission failure, if
MVKConfiguration::resumeLostDevice
enabled, do not release
waits onVkDevice
, and do not returnVK_ERROR_DEVICE_LOST
, unlessVkPhysicalDevice
is also lost. - Fix inconsistent handling of linear attachment decisions on Apple Silicon.
- Fix small memory leak during swapchain creation.
- Fix stencil clear incorrectly using using the depth
loadOp
, when stencil is smaller than render area. - Reorganize coherent texture flushing on memory map and unmap`.
- Fix issues where data in temporary internal buffers are discarded while in use.
- Protect against crash when retrieving
MTLTexture
whenVkImage
has noVkDeviceMemory
bound. - Adjust some
VkPhysicalDeviceLimits
values for Vulkan and Metal compliance. - Fix internal reference from
SPIRV_CROSS_NAMESPACE_OVERRIDE
toSPIRV_CROSS_NAMESPACE
. - Add label strings to
MTLCommandBuffers
, based on use type, for GPU Capture debugging. - Add
Scripts/runcts
script as a convenience for running Vulkan CTS tests. - Support Xcode 13 SDK APIs and build settings.
- Update dependency libraries to match Vulkan SDK 1.2.182.
- Update to latest SPIRV-Cross version:
- MSL: Handle array of IO variable with Component decoration.
- MSL: Handle array with component when we cannot rely on
user()
attrib. - MSL: Improve handling of split tessellation access chains.
- MSL: Always enable support for base vertex/index on iOS.
Release for Vulkan SDK 1.2.176
- Add beta support for using Metal argument buffers for shader resources on macOS, by setting
MVK_CONFIG_USE_METAL_ARGUMENT_BUFFERS
environment variable (disabled by default). This dramatically
expands the number of resources that can be submitted to a pipeline stage, per the Vulkan
VK_EXT_descriptor_indexing
extension. Currently available on macOS 11.0 (Big Sur) or later,
and on earlier macOS versions on Intel GPU's. - Add support for
HDR10
colorspace viaVK_COLOR_SPACE_HDR10_HLG_EXT
andVK_COLOR_SPACE_HDR10_ST2084_EXT
. - Always explicitly set
CAMetalLayer
colorspace property based on Vulkan parameters,
and don't rely on Metal default values. - Add
MVKConfiguration::resumeLostDevice
andMVK_CONFIG_RESUME_LOST_DEVICE
env var,
to allowVkDevice
to resume after non-fatalVK_ERROR_DEVICE_LOST
error. MVKDescriptorPool
pools its descriptor sets.- Enable
MVKConfiguration::preallocateDescriptors
andMVK_CONFIG_PREALLOCATE_DESCRIPTORS
environment variable by default to preallocate descriptors when aVkDescriptorPool
is created. - Avoid use of Metal renderpass load and store actions on memoryless attachments.
- Fix memory leak on swapchain destruction.
- Fix
MVKPhysicalDevice::getSurfaceFormats()
returningVK_FORMAT_UNDEFINED
. - Fix memory leak where swapchains and images were not destroyed due to a retention loop.
- Fix advertising single-texel alignment capability for texel buffers.
- Fix time and space inefficiencies caused by missed shader cache lookup hits in
MVKShaderLibraryCache
. - Log enhanced command buffer errors in debug mode.
- Ensure queue submission message logging occurs before submission object is destroyed.
- Remove project qualifiers from references to
SPIRV-Cross
header files. - Only perform automatic GPU capture on queue identified by
MVKConfiguration::defaultGPUCaptureScopeQueueIndex
anddefaultGPUCaptureScopeQueueFamilyIndex
. - Introduce
MVKConfigLogLevelBits
,MVKConfigAutoGPUCaptureScopeBits
, andMVKConfigTraceVulkanCallsBits
enums to specify config API values in a Vulkan-friendly manner, while automatically documenting the same values for env vars. - Add
MVKConfiguration::apiVersionToAdvertise
andMVK_CONFIG_API_VERSION_TO_ADVERTISE
env var to configure MoltenVK to advertise a particular Vulkan version. - Add
MVKConfiguration::advertiseExtensions
andMVK_CONFIG_ADVERTISE_EXTENSIONS
env var to configure MoltenVK to force advertising support for no, or minimally few, Vulkan extensions. - Remove the
Hologram
andAPI-Samples
demo apps, and remove
LunarG/VulkanSamples
as a dependency library. - Add
NDEBUG
macro to all Release builds to removeassert()
calls. - Update
VK_MVK_MOLTENVK_SPEC_VERSION
to31
. - Update dependency libraries to match Vulkan SDK 1.2.176.
- Update to latest SPIRV-Cross version:
- MSL: Support long
ulong
types in buffers in 2.3+. - MSL: Support padding Metal argument buffer entries based on argument index.
- Add interfaces to aid with LTO-style optimization
- Handle edge cases in
OpCopyMemory
.
- MSL: Support long
Release for Vulkan SDK 1.2.170
- Advertise support for
shaderInt64
feature. - Support fast math on MSL compiler via
MVKConfiguration::fastMathEnabled
configuration
setting andMVK_CONFIG_FAST_MATH_ENABLED
environment variable (enabled by default). - Support compiling MSL with position invariance if indicated in SPIRV shader.
vkGetMoltenVKConfigurationMVK()
andvkSetMoltenVKConfigurationMVK()
functions
can now be used with aVkInstance
from another Vulkan layer, or with aVK_NULL_HANDLE VkInstance
.MVKConfiguration
extended to cover all MoltenVK environment variables.- Report accurate value of 8 for
VkPhysicalDeviceLimits::maxBoundDescriptorSets
. - Advertise macOS M1 GPU as
VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU
. - For Vulkan semaphores, prefer using
MTLFence
overMTLEvent
. - Support
immmutableSamplers
with sampler arrays. - Query pools write to dedicated temporary internal buffer to span multiple render passes
and support larger query counts. - Fixes for race conditions in CTS multithread tests, guarding
MTLDevice
on cross-thread syncing. - Set Metal buffer alignment to 256 on non-Apple Silicon iOS/tvOS simulators.
MVKMTLBufferAllocation
: Support device-local temp buffers for tessellation,
indirect multiview, or occlusion queries buffer usage.MVKPixelFormats
: EnableRenderTarget
usage for linear textures on Apple GPUs.MVKRenderPass
: Use a non-trivial granularity for TBDR GPUs.- Don't use barriers in render passes on Apple GPUs.
MVKGraphicsPipeline
: Fix color write mask withRGB9E5
RTs.MVKImagePlane
: When sync'ing, create the texture if it doesn't exist.MVKRenderPass
: Don't useLoad/Store
actions for memoryless.MVKRenderPass
: Only useMTLStorageModeMemoryless
where available.MVKDeviceMemory
: Don't consider Memoryless host-accessible on macOS/tvOS.MVKCmdResolveImage
fix incorrectly changing first resolve layer's src/dst base layer.MVKPhysicalDevice
: Require Mac family 2 for render without attachments.MVKPhysicalDevice
: Disable SIMD-group permutation for Mac family 1.MVKPhysicalDevice
: Clamp maximum buffer range to 4 GiB - 1.MVKMTLBufferAllocation
: Mark temp buffers as volatile.SPIRVReflection
: Multiple entry point support forgetShaderOutputs()
.- Remove obsolete
MVKVector
, which was long ago replaced withMVKSmallVector
. - Remove official support for submitting shaders directly containing MSL source code or compiled MSL code.
MSL shaders may still be directly submitted, and may work, but it is not officially supported at this time.
Future versions of MoltenVK may support direct MSL submission again. - Remove
ONLY_ACTIVE_ARCH
from Debug builds. MoltenVK Debug builds now build for all platform architectures. - Add ability to automatically capture first GPU frame by setting
MVKConfiguration::autoGPUCaptureScope
(or environment variableMVK_CONFIG_AUTO_GPU_CAPTURE_SCOPE
) to2
. - Support GitHub Actions for CI builds on pull requests.
- Support pre-built binary artifacts from GitHub Actions CI builds.
Makefile
andfetchDependencies
supportxcpretty
, if available.- Remove support for Travis-CI.
- Update
VK_MVK_MOLTENVK_SPEC_VERSION
to30
. - Update copyright notices to year 2021.
- Update Xcode build settings check to Xcode 12.4.
- Support legacy building on Xcode 11.7.
- Update dependency libraries to match Vulkan SDK 1.2.170.
- Update to latest SPIRV-Cross version:
- MSL: Gracefully assign automatic input locations to builtin attributes.
- MSL: Refactor out location consumption count computation.
- MSL: Always return
[[position]]
when required. - MSL: Fix some edge cases in input attachment handling.
- MSL: Fix various bugs with framebuffer fetch on macOS and argument buffers.
- MSL: Always use input_attachment_index for framebuffer fetch binding.
- MSL: Make sure initialized output builtins are considered active.
- MSL: Always enable Outputs in vertex stages.
- MSL: Only consider builtin variables and flatten builtin arrays if they are part of IO interface.
- MSL: Handle load and store to TessLevel array in TESC.
- MSL: Fixup type when using tessellation levels in TESC functions.
- MSL: Fix automatic assign of builtin attributes in tessellation.
- GLSL/MSL: Handle more complicated tessellation Output initializers.
- MSL: Emit correct initializer for tessellation control points.
- MSL: Handle initializers for tess levels.
- MSL: Fix initializer for tess level outputs.
- MSL: Fix broken reserved identifier handling for entry points.
- MSL: Handle reserved identifiers for entry point.
- Fix pathological complexity explosion for certain shaders.
- Improve handling of complex variable initialization.
- Expose position invariance.
Release for Vulkan SDK 1.2.162
- Add support for extensions:
VK_KHR_sampler_mirror_clamp_to_edge
(iOS)VK_KHR_timeline_semaphore
VK_EXT_descriptor_indexing
(initial release limited to Metal Tier 1: 96/128 textures, 16 samplers)VK_EXT_post_depth_coverage
(macOS)VK_EXT_private_data
VK_EXT_subgroup_size_control
VK_EXT_texture_compression_astc_hdr
VK_AMD_shader_image_load_store
(macOS)VK_IMG_format_pvrtc
(macOS)
- Support the Mac Catalyst platform for iOS apps on macOS 11.0+,
under bothx86_64
andarm64
architectures. - Re-enable
MTLEvent
-based semaphores and fix severalMVKSync
issues. - Use
VK_KHR_image_format_list
to disableMTLTextureUsagePixelFormatView
if only swizzles orsRGB
conversion will be used for image views, improving
performance on iOS by allowing Metal to use lossless texture compression. - Handle device loss.
- Fix crash in
vkUpdateDescriptorSets()
when copying inline block descriptors. - Fix Metal validation error when unused elements in an array of sampler are not populated by descriptors.
- Fix crashes in
vkUseIOSurfaceMVK()
on chroma sampling and double releasing ofMTLTexture
. - Fix potential drawable present race conditions.
- Fix crash in
vkUpdateDescriptorSets()
when copying inline block descriptors. - Fix Metal validation error when unused elements in an array of sampler are not populated by descriptors.
- Move Metal drawable presentation from
MTLCommandBuffer
toMTLDrawable
to improve performance and reduce blocking. - Allow binding descriptor set using layout different than it was created with.
- Report
VkPhysicalDeviceLimits::maxPerStageDescriptorStorageImages
as Metal limit of8
. - Increase per-stage texture count to
96
for A11 SoC's and above. - Use variable descriptor count when determining descriptor binding count.
- Support setting sizes of SPIR-V unsized arrays.
- Support
MTLStorageTypeMemoryless
for Apple Silicon on Mac. - Support Apple GPU pixel formats with Apple Silicon on Mac.
- Allow linear images on Apple GPUs in Blit and Clear commands.
- Apple family 7 GPUs (A14) on iOS support multisample layered rendering,
as well as sampler border colors and the mirror clamp to edge sampler address mode. - Set fill mode, depth bias, viewport, and scissor states before clearing attachments
- Disable culling for the duration of
vkCmdClearAttachments()
MVKDevice
increase minimum OS for shared-storage textures.MVKDevice
set properties for Apple Silicon GPUs on macOS.MVKQueue
only create oneGPUCaptureScope
per queue.MVKSamplerYcbcrConversion
always make sure there is one plane.MVKPhysicalDevice
correct fragment input component limit.MVKPhysicalDevice
set max visibility buffer size to 256 kiB where supported.MVKPhysicalDevice
expose support for interpolation functions.MVKPhysicalDevice
enable 3D compressed textures on iOS/tvOS, and forbid ETC2 and EAC
3D textures on all platforms. Apple GPUs do not support 3D for those.MVKPhysicalDevice
enable Apple family 7 features on iOS.MVKPhysicalDevice
remove need to callinitGPUInfoProperties()
twice.MVKPhysicalDevice
correct max descriptor set resources.MVKPhysicalDevice
reduce maximum point size to 64.MVKPhysicalDevice
enable strictLines for Intel and NVIDIA.MVKPhysicalDevice
enableshaderResourceMinLod
on iOS.MVKPhysicalDevice
set correct subgroup properties.MVKPhysicalDevice
enable texture swizzle on all Apple GPUs.MVKPipelineLayout
only set configuration result if validating.MVKPipeline
fix calculation of atomic image buffer addresses.MVKPipeline
disable rasterization if culling both sides and don't try to add a fragment shader.MVKPipeline
shorten vertex attribute format's length whenstride
<size
.MVKGraphicsPipeline
handleminSampleShading
.MVKImage
always use texel buffers for linear images inMTLHeaps
.MVKImage
make sure plane heap offsets are properly aligned.MVKImage
always set the depth plane when rendering to a 3D image.MVKImage
avoid swizzling storage and/or attachment image views.MVKImage
avoid texel buffer for atomics if view format list
exists and does not include eitherR32_UINT
orR32_SINT
format.MVKImageView
always ignore transfer usages.MVKBufferView
avoid triggeringbytesPerRow
validation warning.MVKDescriptorSetLayout
speed up lookup of descriptor index and streamline binding access.MVKCmdBlitImage
use layered draws when possible.MVKCmdBlitImage
add 0.5 to layer index before interpolating.MVKCmdBlitImage
support depth/stencil blits with inversion and scaling.MVKSwapchain
allow images whose size doesn't match theCAMetalLayer
.MVKPixelFormats
add 0.5 ULP to clear values for normalized formats.MVKCommandEncoder
don't setrenderTargetArrayLength
for mixed 2D/3D renders.MVKExtensions
add missing tvOS case for unsupported extensions.MVKDescriptorSetLayout
sort and hold bindings by binding number.MVKDescriptor
simplify subclass implementations.MVKComputePipeline
override max threads per threadgroup.MVKGPUCapture
make sure theMTLCaptureScope
has only one reference.MoltenShaderConverter
fix tvOS build to support tvOS 9.0.- Support building external libraries and dylibs against sanitizers.
- Clarify documentation on mapping limitations for host-coherent image memory on macOS.
- Add validation policy for MoltenVK development to
README.md
document. - Update Xcode build settings check to Xcode 12.2.
- Update
VK_MVK_MOLTENVK_SPEC_VERSION
to29
. - Update dependency libraries to match Vulkan SDK 1.2.162.
- Update to latest SPIRV-Cross version:
- MSL: Support
SPV_EXT_demote_to_helper_invocation
for MSL 2.3. - MSL: Support run-time sized image and sampler arrays
(GL_EXT_nonuniform_qualifier/SPV_EXT_descriptor_indexing
). - MSL: Support atomic access to images from argument buffers.
- MSL: Add missing interlock handling to atomic image buffers.
- MSL: Fix calculation of atomic image buffer address.
- MSL: Support querying and modifying generated combined sampler suffix.
- MSL: Don't use a bitcast for tessellation levels in tesc shaders.
- MSL: Handle Offset and Grad operands for 1D-as-2D textures.
- MSL: Don't remove periods from swizzle buffer index exprs.
- MSL: Correct definitions of subgroup ballot mask variables.
- MSL: Extract global variables from subgroup ballot operations.
- MSL: Mask ballots passed to ballot bit ops.
- MSL: Don't mask off inactive bits in ballot masks.
- MSL: Support vectors with
OpGroupNonUniformAllEqual
. - MSL: Cast broadcast booleans to
ushort
. - MSL: Do not use
component::x
gather fordepth2d
textures. - MSL: For 2.1+, don't disable rasterization for vertex writes.
- MSL: Allow post-depth coverage on Mac in MSL 2.3.
- MSL: Allow framebuffer fetch on Mac in MSL 2.3.
- MSL: Allow Bias and Grad arguments with comparison on Mac in MSL 2.3.
- MSL: Support pull-model interpolation on MSL 2.3+
- MSL: Expand subgroup support.
- MSL: Adjust
FragCoord
for sample-rate shading. - MSL: Expose some more features on iOS:
min_lod_clamp()
,simd_is_helper_thread()
,
barycentric_coord
, andprimitive_id
. - MSL: Don't add fixup hooks for builtin variables if they're unused.
- MSL: Don't try to use
[[thread_index_in_simdgroup]]
in vertex shaders. - Handle case where block is loop header, continue AND break block.
- MSL: Added
fmin3
andfmax3
library functions to the illegal name list. - Added Metal keyword:
level
. - Parser: Don't assume
OpTypePointer
will always take aSPIRType
. - Add MIT dual license for the SPIRV-Cross API and CLI.
- MSL: Support
Release for Vulkan 1.1 Support and Vulkan SDK 1.2.154
Note: This release contains changes to library paths and framework linking options.
- Add support for Vulkan 1.1, including:
- The
vkEnumerateInstanceVersion()
function - The
vkGetDeviceQueue2()
function - Protected memory (non-functional)
- A feature struct for
VK_KHR_shader_draw_parameters
- All extensions that were promoted to core in Vulkan 1.1
- The
- Add support for extensions:
VK_KHR_portability_subset
VK_KHR_create_renderpass2
VK_KHR_depth_stencil_resolve
VK_KHR_external_fence
(non-functional groundwork for future extensions,
including support for GCD and Mach semaphores)VK_KHR_external_fence_capabilities
(non-functional groundwork for future
extensions, including support for GCD and Mach semaphores)VK_KHR_external_semaphore
(non-functional groundwork for future
MTLSharedEvent
Vulkan extension)VK_KHR_external_semaphore_capabilities
(non-functional groundwork for
futureMTLSharedEvent
Vulkan extension)VK_KHR_multiview
VK_KHR_shader_subgroup_extended_types
- Remove support for obsolete
VK_EXTX_portability_subset
extension. - Redesign build and linking options that leverage newer framework technology:
- Add comprehensive support for multi-platform, multi-architecture
XCFrameworks
. - Build fat single-platform, multi-architecture
dylibs
. - Add support for Apple Silicon builds for macOS and Simulators.
- Remove support for distinct legacy frameworks and static libraries.
- Remove support for fat libraries and frameworks that span device and simulators.
- Combine
MoltenVKSPIRVToMSLConverter
andMoltenVKGLSLToSPIRVConverter
frameworks
into a singleMoltenVKShaderConverter
framework.
- Add comprehensive support for multi-platform, multi-architecture
- Support copying between 3D and 2D images.
- Support clearing 3D images.
- Support linear images and buffer views in shared memory on macOS 10.15.5+.
- Support
VK_IMAGE_CREATE_EXTENDED_USAGE_BIT
. - Support multiple aliased images for dedicated image memory allocations.
- Improve performance of tessellation control pipeline stage by processing multiple
patches per workgroup. vkCmdBindDescriptorSets
orderpDynamicOffsets
by descriptor binding number
within each descriptor set.vkCmdCopyImage
on macOS flush non-coherent image memory before copy operation.- Process multiple patches per workgroup in a tessellation control shader.
- Initialize tessellation related variables conditionally in indirect draws.
- Fix build errors on Simulator not supporting
MTLDrawable
present time options. - Fix
dynamicOffsets
ordering based on binding index. - Ensure the base texture is created when creating a view texture.
- Fix features, alignment, and image properties of chroma subsampled formats.
- Disable
variableMultisampleRate
feature. - Permit renderpass to have no attachments.
- Clear linear images using a compute shader.
- Reject attempts to create compressed linear images.
- Always use a texel buffer for atomic linear storage images.
- Use deferred store actions instead of tracking multi-pass draws.
- Use
MTLRenderPassDescriptor renderTargetWidth
andrenderTargetHeight
on macOS 10.15 and newer. - Use a 2D non-arrayed view for 2D non-arrayed attachments.
- Use arrayed textures for layered subpass input.
- Use layered rendering when clearing or resolving images whenever possible,
and takerenderArea
into account. - Streamline image copying when the entire source is being copied to the entire destination.
- Don't set
renderTargetArrayLength
on devices that don't support it. - Don't attempt to clear unused attachments and fix vertex count for clearing multiple layers.
- Don't encode commands that draw zero vertices.
- Ignore
addressModeW
on samplers with unnormalized coordinates. - Flip GPU when swapchain is created, not when device object is created,
to optimize this condition if it is not needed. - Flush source image of a transfer on macOS.
- Fix offset for texel buffers and heaps for multi-planar images.
- Fix memory leak when object create fails.
- Fix several use-after-freed errors.
- Fix applying divisors to instanced vertex attributes.
- Fix issue in accessing
MTLTexture
for swapchain image planes. - Fix occasional crash on vertex stage indirect draws.
- Fix reported image format properties for 1D images as 2D.
VK_EXT_vertex_attribute_divisor
fix issues with zero divisors.- Re-add support for bitcode generation on iOS and tvOS.
fetchDependencies
builds serially by default for better build script integration.- Combine
MoltenVKSPIRVToMSLConverter
andMoltenVKGLSLToSPIRVConverter
frameworks into a singleMoltenVKShaderConverter
framework. - Fix Metal validation error when occlusion query and renderpass are in separate
Vulkan command buffers. - Fix a crash on starting a query outside a render pass.
- Fix Metal validation error that forbids rendering to an image of zero size.
- Fix issue where descriptor pool releasing descriptor sets it doesn't own.
- Fix issue where load/store actions not set in renderpass that includes compute stage.
- Prevent accidental
setColorStoreAction
for non-color attachments. - Fix image copy destination extent of compressed images.
- Fix image resolve sub-region extent.
- Fix owner of merged
MVKShaderLibraries
. - Fix rounding of buffer sizing when
VK_WHOLE_SIZE
is used. - Fix detection of indexed draw calls.
vkSetMTLTextureMVK()
Fix crash if incomingMTLTexture
does not have an associatedIOSurface
.- Update dependency libraries to match Vulkan SDK 1.2.154.
- Update to latest SPIRV-Cross version:
- MSL: Fix handling of matrices and structs in the output control point array.
- MSL: Enclose args when convert
distance(a,b)
toabs(a-b)
. - MSL: Fix multiview view index calculation with a non-zero base instance.
- MSL: Don't set the layer for multiview if the device doesn't support it.
- MSL: Fix
OpCompositeInsert
andOpVectorInsertDynamic
. - MSL: Support layered input attachments.
- Ensure that we use primary alias type when emitting flattened members.
- Only rewrite type aliases for the base type.
- Overhaul how we deal with reserved identifiers.
- Implement a simple evaluator of specialization constants.
- Deal with case where a selection construct conditionally merges/breaks.
- Allow
flip_vert_y
in all relevant stages. - Clean up conditional branch codegen.
Release for Vulkan SDK 1.2.148
- Add support for extensions:
VK_KHR_driver_properties
VK_KHR_sampler_ycbcr_conversion
VK_EXT_image_robustness
VK_EXT_robustness2
- Add native support for
VK_FORMAT_D16_UNORM
on iOS 13.0+ and tvOS 13.0+. - Add support for tvOS platform.
- Add support for iOS Simulator and tvOS Simulator build destinations.
- Cube demo runs on tvOS, iOS Simulator, and tvOS Simulator.
- MoltenVK Xcode schemes support building fat Platform+Simulator binaries.
Makefile
supports building fat Platform+Simulator binaries, plus Debug builds.fetchDependencies
script supports platform build selection, plus parallel builds.vkCmdBlitImage()
returns error if scaling or inverting to linear image on macOS.- Support
VK_FORMAT_A2B10G10R10_UNORM_PACK32
as a surface format. - Support
VkPipelineMultisampleStateCreateInfo::pSampleMask
. - Support
VkPhysicalDeviceSubgroupProperties
. - Fix memory layout of inline uniform blocks.
- Fix issue where mapped host-coherent device memory not updated from image contents on macOS.
- Fix image memory sizing and offsets.
- Fix small memory leak when setting swapchain color space.
- Fix new and unexpected App Store failure on newly deprecated color space values.
- Fix intermittent concurrent shader specialization race condition.
- Fix offsets when flushing buffer data to GPU.
- Fix issue where expected buffer-sizes buffer not bound to Metal compute encoder.
- Ensure fragment shader inputs to have as many components as vertex shader outputs.
- Include vertex attribute size when testing whether attribute offset exceeds stride.
- Add support for
USCALED/SSCALED
vertex formats. - Add host-coherent texel buffer caching just for buffer views.
- Include MoltenVK Git revision hash in
VkPhysicalDeviceProperties::pipelineCacheUUID
. - Add
MVKPhysicalDeviceMetalFeatures::vertexStrideAlignment
to track Metal vertex binding stride alignment. - Add
MVKPhysicalDeviceMetalFeatures::indirectTessellationDrawing
to track if indirect tessellation drawing is supported. - Remove use of
@available()
directive as it was causing issues in some build environments. - Pass pipeline sample mask, if present, to SPIRV-Cross.
- Refactor MoltenVK Xcode build architectures.
- Demo
API-Samples generateSPIRVShaders
no longer buildsMoltenVKShaderController
tool. - Update
VK_MVK_MOLTENVK_SPEC_VERSION
to27
. - Update dependency libraries to match Vulkan SDK 1.2.148.
- Update to latest SPIRV-Cross version:
- MSL: Add support for processing more than one patch per workgroup.
- MSL: Workaround broken scalar access chain behavior in MSL LLVM IR / AIR.
- MSL: Do not emit swizzled writes in packing fixups.
- MSL: Ensure
OpStore
source operands are marked for inclusion in function arguments. - MSL: Enabling setting an additional fixed
sampleMask
in fragment shaders. - MSL: Remove obsolete
MSLVertexAttr
andMSLShaderInput
members. - MSL: Fix up input variables' vector lengths in all stages.
- MSL: Improve handling of array types in buffer objects.
- MSL: Deal with loading non-value-type arrays.
- MSL: Deal with array load-store in buffer-block structs.
- MSL: Use input attachment index directly for resource index fallback.
- Fix missing switch cases in
Y'CbCr
conversion. - Implement context-sensitive expression read tracking.
Release for Vulkan SDK 1.2.141.1
- Fix issue in reporting properties of substitutable
VkFormats
. - Fix vertex attribute offset adjustments when vertex buffer stride is zero.
- Update
fetchDependencies
script to use pre-builtspirv-tools
files by default. - Update
maxVertexInputBindingStride
andmaxVertexInputAttributeOffset
to minimum Vulkan values. - Numerous documentation typo corrections.
- Update
VK_MVK_MOLTENVK_SPEC_VERSION
to26
. - Update Travis CI to Xcode 11.5.