New feature and improvement
* LLPC compile time optimization
* Update Vulkan header to 1.1.106
* Enable shader cache to disk for Dawn of War3
* Performance tuning for game Thrones of Britannia
* Add option to use GpuHeapGartUswc for the pipeline uploader
New feature and improvement
* Performance optimization for Total:WarhammerII, Talos principle and Thrones of Britannia
Issue fix
* Corruption in multiple dxvk games
* CPU mapping problem for memory shared cross device
* Thrones of Britannia tearing regression
New feature and improvement
* Update Vulkan headers to 1.1.105
* Allow shared memory to be CPU visible
* Enable priority regardless of whether VK_EXT_memory_priority is enabled or not for all external queues
* Performance optimization for Total:WarhammerII
Issue fix
* Just Cause3 flickering
* Path of Exile incorrect rendering
New feature and improvement
* Enable VK_EXT_memory_priority extension
* Enable VK_EXT_memory_budget extension
* Enable skip fast clear eliminate optimization by default
* Reduce unnecessary L2 cache actions on GFX9
* Update pipeline ABI metadata note ID to 32 (from 13) to match HSA code objects
* Improve CPU-bound performance in Mad Max
Issue fix
* Abnormal timestamp on transfer queue
* ShaderDbg bugs
* Minor profiling validation and memory leak fixes
* Dead loop in converting a FP16 denorm back to normalized FP32
New feature and improvement
* Enable below extensions:
- VK_KHR_vulkan_memory_model
- VK_EXT_depth_clip_enable
- VK_KHR_depth_stencil_resolve
- VK_KHR_shader_float16_int8
- VK_EXT_debug_utils extension
- VK_EXT_transform_feedback
* Fall back to the internal shader cache for the case of VkPipelineCache miss
Issue fix
* dEQP-VK.spirv_assembly.instruction.graphics.float16.derivative_* and dEQP-VK.spirv_assembly.instruction*.float16.arithmetic_*.atan2* test failure
* A bug in LLPC ShaderCache Merge function which causes Dota2 stuttering and performance drop after recent game update
* Dynamic loop unroll crash
* Rise of the Tomb Rider game crash with gcc 7 build
* Witcher3-dxvk hang after loading screen
New feature and improvement
* [TransformFeedback] Fix the register setting for RasterizationStreamSelect capablitity
* [VK_KHR_shader_float16_int8] handling of OpVectorInsertDynamic
* Add supports for min and max stencil resolve using compute pipeline
Issue fix
* Memory leak in CmdBufferLogger
* vkmark corruption observed on Fiji + wayland
* Corruption issue caused by DPP operand order in atomic optimizer
New feature and improvement
* [RGP]Add functionality to enable instruction-level trace per-pipeline
* Add debug support to the CmdBufferLogger layer that allows for "single-stepping" of draws/dispatches
Issue fix
* Vulkan CTS OOM test case crashes and a wide range of many other potential PAL_NEW alloc failure crashes
* Broken path of offchip GS
* Transformfeedback multi-stream CTS failure on gfxip 8 and gfxip 9
* PRT issues with SDMA on gfxip 6-8
New feature and improvement
* Update Vulkan Headers to 1.1.97
* Enable the extensions under development (VK_EXT_DEBUG_UTILS, VK_KHR_SHADER_FLOAT16_INT8,VK_ EXT_TRANSFORM_FEEDBACK) through the environment variable AMDVLK_ENABLE_DEVELOPING_EXT
* Add proper handling of allocation failures
Issue fix
* Support for ECC GPR protection on Vega20
* A memory priority issue
* GFX7 indirect dispatch does not work correctly when submitted to a compute queue
2019-1-29 update
New feature and improvement
* Add Vega20 and Raven2 support
* Move VK_EXT_swapchain_colorspace to instance extension
* Report CPU clock speed to RGP
Issue fix
* Bugs in Release/acquire-based barrier
* Top pipelines identified by timeReport.py is mismatched with the dumped pipelines
New feature and improvement
* Expose YUV planes, allow applications to implement their own color conversion accessing each YUV planes
* Do not include symbols while building release driver
* Change the default WgpMode from wgp to cu
Issue fix
* Random VM fault caused by that the image descriptor and the fmask descriptor contain the same lower virtual address (as designed) but use different offsets in the suballocation
* dxvk ELEX corruption issue
* Performance regression introduced by changes which added support for the LOAD_INDEX path for handling pipeline binds
* A number of issues affecting many-core CPUs like Threadripper and Epyc
New feature and improvement
* Update Vulkan Headers to 1.1.96
* Add pipeline optimizer key to pipeline dump file
* Optimization for fully overwritten resolve
* Add atomicOp support for the variable pointer
Issue fix
* A performance regression with the Talos Principle
* A potential access violation
* A RGP regression
* Multi-process failure
New feature and improvement
* Merge tbuffer.store.i32 with tbuffer.store.v2i32 and tbuffer.store.v4i32, up to 5% performance gain for tessellation
* Enable the atomic optimizer
* Update Vulkan headers to 1.1.94
* Enable primitiveUnderestimation capability of VK_EXT_conservative_rasterization
Issue fix
* GPU hang when Vulkan API accesses the stencil aspect of VK_Format_S8_Uint image explicitly
* CTS dEQP-VK.spirv_assembly.instruction.graphics.16bit_storage.struct_mixed_types.uniform_geom failure
2018-12-4 update
New feature and improvement
* Enable VK_EXT_scalar_block_layout extension
* Enable VK_KHR_swapchain_mutable_format extension
* Enable on-chip GSVS ring on GFX9, seeing up to 10% performance gain
* Add dccBitsPerPixelThreshold setting
* Re-enable tracing of missing registers in thread trace
* Remove some useless settings
Issue fix
* Undefined reference to vk::OverrideDispatchTable_ND(vk::DispatchTable*)
* Errors with multiple monitors and direct display
* dEQP-VK.api.device_init.create_instance_device_intentional_alloc_fail test failure
* Crash issue on Raven caused by the implementation of VK_AMD_memory_overallocation_behavior extension
* dEQP-VK.glsl.atomic_operations.*_*signed64bit_* failure on gfx7
* ELEX crash with Steam Proton
* CopyTextureRegion failure when copying a MSAA Resource in D32_FLOAT or D16_UNORM to a R32_UINT format
* GFX9: CB_DCC_CONTROL register is programmed incorrectly
* Perf counter instance incorrect calculation
1. Clean the submodule
2. Unused variable is treated as error on Ubuntu 16.04 gcc-5.4, turn it off
3. Add "Driver installation instruction" into the release notes
2018-11-19 update
* Enable sparse support by default
* Enable VK_AMD_memory_overallocation_behavior extension
* Enable degenerate triangles for conservative rasterizations
* Add setting for exiting after compilation failure
* Add setting CsCuEnLimitMask for limiting CUs enabled for compute shaders
* Add the ability to dump debug stack traces
* Set default -log-file-dbgs to "" (meaning stderr): this brings amdllpc into line with other LLVM tools
* Can't clear a mutable format image
* CTS memory.pipeline_barrier.transfer_dst_storage_image tests fail
* MGPU asserts when creating graphics pipeline
* Wrong target device ID passed to PAL
* Dual source bend is being enabled when blending is disabled
* Regression dEQP-VK.wsi.wayland.swapchain.create.min_image_count test crashes
* Dependency on x11/xcb libraries even there is no x11/xcb surface created
* Assert in dEQP-VK.binding_model.descriptorset_random.*