main-tot

Continuous build of the latest main branch by Github

Release 16.2.0

Changes in this release:

• Allow gl_FragDepth identifier to be redeclared
• Replace 'GL_TASK(MESH)_SHADER_NV' with 'GL_TASK(MESH)_SHADER_EXT'
• Scalarize swizzled MeshEXT stores
• Support GL_EXT_long_vector
• Add missing preamble #define for GL_EXT_nontemporal_keyword
• Support GL_NV_push_constant_bank
• Support GL_EXT_descriptor_heap

Release 16.1.0

Changes in this release:

• Avoid emitting OpCapability RuntimeDescriptorArray when unnecessary
• Improve compilation speed when debug infomation is enabled
• Fix Debug Info for Retargeted Builtins
• Support GL_EXT_shader_invocation_reorder
• Add checks to coopMatMulAdd
• Implement stringify operator
• Add ES support for depth layout qualifier
• Add debug info for hitObjectNV
• Emit a DebugGlobalVariable instead of DebugLocalVariable for rayQueryEXT
• Add debug info for constant variable
• Improve debug line to point declaration
• Fix bugs in buffer reference alignment
• Reject string operands in binary and select ops
• Support GL_EXT_shader_64bit_indexing
• Support GLSL_EXT_uniform_buffer_unsized_array
• Add semantic check for cooperative vector loads/stores
• Improve the debug info name of opaque (sampler) types
• Support IO mapping of combined samplers and acceleration structures
• Fix bug in debug info for bool types inside SSBO/UBO
• Fix bug in debug info for struct member names
• Add methods for entry point and invert-y to C interface

Release 16.0.0

Breaking changes:

• Remove spirv-remapper (ported to SPIRV-Tools as canonicalize-ids optimization pass)

Other changes:

• Emit error for in/out structs containing booleans
• Fix contradictory layout link errors
• Improve error checks for float8 math
• Fix bug in cooperative matrix load/store handling of array slices
• Update memory semantic validation rules to match update to Vulkan specification
• Simplify glslang grammar for identifier lists
• Fix erroneous parse errors in variable declarations using existing struct names
• Fix type conversions for createTensorLayout/ViewNV
• Fix OpAccessChain signed index issue for small unsigned integers
• Support GLSL_QCOM_cooperative_matrix_conversion
• Allow spec constant matrix use
• Fix inconsistent labeling of capability strings
• Fix crash on unnamed function parameters under relaxed rules
• Fix exported member functions in ShaderLang.h

Release 15.4.0

Changes in this release:

• Support GL_NV_gpu_shader5/GL_ARB_gpu_shader5
• Remove BUILD_SHARED_LIBS option (defer to built-in variable)
• Remap discard to OpDemoteToHelperInvocation
• Fix typos and improve readability of help message
• Support GLSL_QCOM_tile_shading
• Emit correct capability for linear swept sphere
• Support GL_EXT_float8_e5m2_e4m3
• Move E_SPV_ARM_cooperative_matrix_layouts into GLSL.ext.ARM.h
• Support GL_ARM_tensors

Release 15.3.0

Changes in this release:

• Fix CMake find_package on Windows when building shared libraries
• Fix crash calling coopMatLoadTensorNV on an array element
• Avoid reusing array types with stride when no stride is requested
• Fix non-ASCII comments
• Support GL_EXT_bfloat16
• Fix several issues with link-time implicit array size merging
• Support DebugSourceContinued
• LiveTraverser no longer traverses switch cases that are statically not accessed
• Implement optional cross-stage check for inputs with no matching outputs

Release 15.2.0

Changes in this release:

• Fix find_package on Windows when BUILD_SHARED_LIBS=ON
• Emit error if using in/out with struct pointer
• Emit SPV_EXT_opacity_micromap if GL extension is present
• Support GL_NV_linear_swept_spheres
• Support GLSL_EXT_nontemporal_keyword
• Support GL_NV_cluster_acceleration_structure
• Support GL_NV_cooperative_vector
• Check SparseTextureOffset non-const parameters
• Support GL_EXT_texture_offset_non_const
• Revert cross-stage check for missing outputs
• Support EXT_integer_dot_product
• Add support for OpTypeRayQueryKHR and OpTypeAccelerationStructureKHR to SPVRemapper

Release 15.1.0

Changes in this release:

• Add Vulkan 1.4 target and client
• Improve conversion of uniform block to push constant
• Improve cross stage error reporting by reporting proper stager rather than "unkwown stage"
• Add warning if forward declaration uses layout qualifiers
• Implement GLSL_NV_cooperative_matrix2
• Emit OpModfStruct instead of depracated OpModf
• Add link-time cross stage optimization
• Add column to DebugLexicalBlock
• Propagate errors from symbol table initialization
• Fix nonsemantic debuginfo line attribution for cooperative matrix

Release 15.0.0

Breaking changes in this release:

• Explicitly export all symbols that are part of the public API and hide other symbols by default

Other changes in this release:

• Allow building glslang without the SPIR-V backend using the new ENABLE_SPIRV build option
• Add setResourceSetBinding method to the API
• Add interface to get the GLSL IO mapper and resolver
• Allow compute derivative modes when the workgroup dimensions are spec constants
• Improve debug location of branch/return instructions
• Silence preprocessor '#' error reporting in inactive #if/#ifdef/#elif/#else blocks
• Apply GLSL memory decorations to top-level OpVariable
• Move definition of GLSLANG_EXPORT to visibility.h
• Merge ancillary libraries into main glslang library and stub originals
• Add public setSourceFile and addSourceText methods to TShader class
• Add type checks for hitObjectNV
• Add optimizerAllowExpandedIDBound to SpvOptions
• Add SpvTools.h back to public headers
• Add cross-stage check for missing outputs
• Fix HLSL offsets for non-buffers
• Add types and functions for IO mapping to API
• Add function to set preprocessed code to API
• Add set/get version functions to API
• Expose setGlobalUniform functions to API
• Don't emit debug instructions before an OpPhi
• Add command-line and API option to enable reporting column location for compiler errors
• Improve location aliasing checks
• Support constant expression calculated by matrixCompMult
• Fix crash caused by atomicCounter() use without arguments
• Fix multi-line function call line numbers
• Add line info to OpDebugDeclare for function parameters
• Fix HLSL OpDebugFunction file name
• Fix duplicate decorations
• Enable compilation of glslang without thread support for WASI

Release 14.3.0

Changes in this release:

• Generate vector constructions more efficiently when sizes match
• Skip identity conversions for 8-bit and 16-bit types
• Add cmake aliases for public libraries
• Support ARM extended matrix layout
• Emit debug info for buffer references
• Add support for OpExtInstWithForwardRefsKHR
• Generate SPV_EXT_replicated_compisites when requested by pragma
• Reuse loads generated for repeated function arguments
• Fix gl_HitT alias of gl_RayTmax
• Fix some cases where invalid SPIR-V was being generated when using separate samplers
• Add back layoutLocation to public API