| 1 |
Revision history for SPIRV-Tools |
| 2 |
|
| 3 |
v2020.6 2020-09-24 |
| 4 |
- Start SPIRV-Tools v2020.6 |
| 5 |
|
| 6 |
v2020.5 2020-09-22 |
| 7 |
- General |
| 8 |
- Enable building with BUILD_SHARED_LIBS=1 (#3490) |
| 9 |
- Avoid using /MP4 for clang on windows. (#3662) |
| 10 |
- Fix compiler error on macOS with XCode12. (#3836) |
| 11 |
- Optimizer |
| 12 |
- Preserve OpenCL.DebugInfo.100 through private-to-local pass (#3571) |
| 13 |
- Preserve debug info in scalar replacement pass (#3461) |
| 14 |
- Debug info preservation in loop-unroll pass (#3548) |
| 15 |
- Preserve debug info in dead-insert-elim pass (#3652) |
| 16 |
- Improve non-semantic instruction handling in the optimizer (#3693) |
| 17 |
- Let ADCE pass check DebugScope (#3703) |
| 18 |
- Add undef for inlined void function (#3720) |
| 19 |
- Fix SSA-rewrite to remove DebugDeclare for variables without loads (#3719) |
| 20 |
- Handle DebugScope in compact-ids pass (#3724) |
| 21 |
- Add buffer oob check to bindless instrumentation (#3800) |
| 22 |
- Validator |
| 23 |
- Update OpenCL capabilities validation (#3149) |
| 24 |
- Validator support for non-semantic clspv reflection (#3618) |
| 25 |
- OpenCL.DebugInfo.100 DebugTypeArray with variable size (#3549) |
| 26 |
- Only validation locations for appropriate execution models (#3656) |
| 27 |
- Validate more OpenCL.DebugInfo.100 instructions (#3684) |
| 28 |
- Allow DebugTypeTemplate for Type operand (#3702) |
| 29 |
- spirv-val: Add Vulkan VUID labels to BuiltIn (#3756) |
| 30 |
- Allow SPV_KHR_8bit_storage extension. (#3780) |
| 31 |
- Validate SPIRV Version number when parsing binary header (#3834) |
| 32 |
- Reduce |
| 33 |
- Support reducing a specific function (#3774) |
| 34 |
- Fuzz |
| 35 |
- adds TransformationReplaceCopyObjectWithStoreLoad (#3567) |
| 36 |
- adds TransformationReplaceCopyMemoryWithLoadStore (#3575) |
| 37 |
- adds TransformationReplaceLoadStoreWithCopyMemory (#3586) |
| 38 |
- Implement the OpOuterProduct linear algebra case (#3617) |
| 39 |
- Pass to replace int operands with ints of opposite signedness (#3612) |
| 40 |
- TransformationMoveInstructionDown (#3477) |
| 41 |
- Add TransformationMakeVectorOperationDynamic (#3597) |
| 42 |
- TransformationReplaceAddSubMulWithCarryingExtended (#3598) |
| 43 |
- FuzzerPassPropagateInstructionsUp (#3478) |
| 44 |
- add FuzzerPassAddCompositeInserts (#3606) |
| 45 |
- Add inline function transformation (#3517) |
| 46 |
- Transformation to replace the use of an irrelevant id (#3697) |
| 47 |
- Add SPIRV_FUZZ_PROTOC_COMMAND (#3789) |
| 48 |
- Add TransformationDuplicateRegionWithSelection (#3773) |
| 49 |
- Transformation to flatten conditional branch (#3667) |
| 50 |
- Handle OpPhis in TransformationInlineFunction (#3833) |
| 51 |
- Create synonym of int constant using a loop (#3790) |
| 52 |
- Support dead blocks in TransformationAddSynonym (#3832) |
| 53 |
- Linker |
| 54 |
|
| 55 |
v2020.4 2020-07-22 |
| 56 |
- General |
| 57 |
- Changed variable names to be more descriptive (#3433) |
| 58 |
- Add support to GPU-AV instrumentation for Task and Mesh shaders (#3512) |
| 59 |
- Permit Simple and GLSL450 memory model in WEBGPU_0 (#3463) |
| 60 |
- Support SPV_KHR_terminate_invocation (#3568) |
| 61 |
- Optimizer |
| 62 |
- Preserving debug information in optimizations |
| 63 |
(#3389,#3420,#3425,#3356,#3459,#3444,#3492,#3451,#3497i,#3498,#3542) |
| 64 |
- Eliminate branches with condition of OpConstantNull (#3438) |
| 65 |
- Use structured order to unroll loops. (#3443) |
| 66 |
- Updated desc_sroa to support flattening structures (#3448) |
| 67 |
- Support OpCompositeExtract pattern in desc_sroa (#3456) |
| 68 |
- Fix ADCE pass bug for mulitple entries (#3470) |
| 69 |
- Sink pointer instructions in merge return (#3569) |
| 70 |
- Validator |
| 71 |
- Validate location assignments (#3308) |
| 72 |
- Fix reachability in the validator (#3541) |
| 73 |
- Reduce |
| 74 |
- Fuzz |
| 75 |
- Add support for OpSpecConstant* (#3373) |
| 76 |
- Add replace linear algebra instruction transformation (#3402) |
| 77 |
- Implement vector shuffle fuzzer pass (#3412) |
| 78 |
- Swap operands in OpBranchConditional (#3423) |
| 79 |
- Permute OpPhi instruction operands (#3421) |
| 80 |
- Add FuzzerPassAddCopyMemoryInstructions (#3391) |
| 81 |
- TransformationInvertComparisonOperator (#3475) |
| 82 |
- Add variables with workgroup storage class (#3485) |
| 83 |
- Add image sample unused components transformation (#3439) |
| 84 |
- TransformationReplaceParameterWithGlobal (#3434) |
| 85 |
- Support adding dead break from back-edge block (#3519) |
| 86 |
- Fuzzer pass to interchange zero-like constants (#3524) |
| 87 |
- Linker |
| 88 |
|
| 89 |
v2020.3 2020-05-27 |
| 90 |
- General |
| 91 |
- Prevent Effcee from installing things when building spirv-tools with testing enabled (#3256) |
| 92 |
- Update acorn version (#3294) |
| 93 |
- If SPIRV-Headers is in our tree, include it as subproject (#3299) |
| 94 |
- allow cross compiling for Windows Store, UWP, etc. (#3330) |
| 95 |
- Optimizer |
| 96 |
- Remove deprecated interfaces from instrument passes (#3361) |
| 97 |
- Preserve debug info in inline pass (#3349) |
| 98 |
- Handle more cases in dead member elim (#3289) |
| 99 |
- Preserve debug info in eliminate-dead-functions (#3251) |
| 100 |
- Fix Struct CFG analysis for single block loop (#3293) |
| 101 |
- Add tests for recently added command line option (#3297) |
| 102 |
- Consider sampled images as read-only storage (#3295) |
| 103 |
- Allow various validation options to be passed to spirv-opt (#3314) |
| 104 |
- Add debug information analysis (#3305) |
| 105 |
- Preserve debug info for wrap-opkill (#3331) |
| 106 |
- refactor inlining pass (#3328) |
| 107 |
- Add unrolling to performance passes (#3082) |
| 108 |
- Validator |
| 109 |
- Add validation support for ImageGatherBiasLodAMD (#3363) |
| 110 |
- Validate ShaderCallKHR memory scope (#3332) |
| 111 |
- Validate Buffer and BufferBlock apply only to struct types (#3259) |
| 112 |
- Reduce |
| 113 |
- increase default step limit (#3327) |
| 114 |
- Remove unused uniforms and similar (#3321) |
| 115 |
- Fuzz |
| 116 |
- Add support for StorageBuffer (#3348) |
| 117 |
- Add validator options (#3254) |
| 118 |
- Limit adding of new variables to 'basic' types (#3257) |
| 119 |
- Transformation to add OpConstantNull (#3273) |
| 120 |
- Handling of more fuzzing opportunities (#3277, #3280, #3281, #3290, #3292) |
| 121 |
- Respect rules for OpSampledImage (#3287) |
| 122 |
- Do not outline regions that produce pointer outputs (#3291) |
| 123 |
- Linker |
| 124 |
|
| 125 |
v2020.2 2020-03-26 |
| 126 |
- General: |
| 127 |
- Support extended instructions in the vscode language server |
| 128 |
- Make spvOpcodeString part of the public API (#3174) |
| 129 |
- Added guide to writing a spirv-fuzz fuzzer pass (#3190) |
| 130 |
- Add support for KHR_ray_{query,tracing} extensions (#3235) |
| 131 |
- Optimizer |
| 132 |
- Debug Printf support (#3215) |
| 133 |
- Add data structure for DebugScope, DebugDeclare in spirv-opt (#3183) |
| 134 |
- Fix identification of Vulkan images and buffers (#3253) |
| 135 |
- Validator |
| 136 |
- Add support for SPV_AMD_shader_image_load_store_lod (#3186) |
| 137 |
- Add validation rules for OpenCL.DebugInfo.100 extension (#3133) |
| 138 |
- Adding WebGPU specific Workgroup scope rule (#3204) |
| 139 |
- Disallow phis of images, samplers and sampled images (#3246) |
| 140 |
- Reduce |
| 141 |
- Fuzz |
| 142 |
- Fuzzer passes to add local and global variables (#3175) |
| 143 |
- Add fuzzer passes to add loads/stores (#3176) |
| 144 |
- Fuzzer pass to add function calls (#3178) |
| 145 |
- Fuzzer pass that adds access chains (#3182) |
| 146 |
- Fuzzer pass to add equation instructions (#3202) |
| 147 |
- Add swap commutable operands transformation (#3205) |
| 148 |
- Add fuzzer pass to permute function parameters (#3212) |
| 149 |
- Allow OpPhi operand to be replaced with a composite synonym (#3221) |
| 150 |
- Linker |
| 151 |
|
| 152 |
v2020.1 2020-02-03 |
| 153 |
- General: |
| 154 |
- Add support for SPV_KHR_non_semantic_info (#3110) |
| 155 |
- Support OpenCL.DebugInfo.100 extended instruction set (#3080) |
| 156 |
- Added support for Vulkan 1.2 |
| 157 |
- Add API function to better handle getting the necessary environment (#3142) |
| 158 |
- Clarify mapping of target env to SPIR-V version (#3150) |
| 159 |
- Implement constant folding for many transcendentals (#3166) |
| 160 |
- Optimizer |
| 161 |
- Change default version for CreatInstBindlessCheckPass to 2 (#3096, #3119) |
| 162 |
- Better handling of OpLine on merge blocks (#3130) |
| 163 |
- Use placeholder switch instead of placeholder loop in MergeReturn pass. (#3151) |
| 164 |
- Handle TimeAMD in AmdExtensionToKhrPass. (#3168) |
| 165 |
- Validator |
| 166 |
- Fix structured exit validation (#3141) |
| 167 |
- Reduce |
| 168 |
- Fuzz |
| 169 |
- Fuzzer pass to merge blocks (#3097) |
| 170 |
- Transformation to add a new function to a module (#3114) |
| 171 |
- Add fuzzer pass to perform module donation (#3117) |
| 172 |
- Fuzzer passes to create and branch to new dead blocks (#3135) |
| 173 |
- Fuzzer pass to add composite types (#3171) |
| 174 |
- Linker: |
| 175 |
- Remove names and decorations of imported symbols (#3081) |
| 176 |
|
| 177 |
v2019.5 2019-12-11 |
| 178 |
- General: |
| 179 |
- Export SPIRV-Tools targets on installation |
| 180 |
- SPIRV-Tools support for SPIR-V 1.5 (#2865) |
| 181 |
- Add WebGPU SPIR-V Assembler in JavaScript. (#2876) |
| 182 |
- Add Bazel build configuration. (#2891) |
| 183 |
- Add support for building with emscripten (#2948) |
| 184 |
- Update SPIR-V binary header test for SPIR-V 1.5 (#2967) |
| 185 |
- Add fuzzer for spirv-as call path (#2976) |
| 186 |
- Improved CMake install step. (#2963) |
| 187 |
- Add fuzzer for spirv-dis call path (#2977) |
| 188 |
- Ensure timestamp does not vary with timezone. (#2982) |
| 189 |
- Add a vscode extension for SPIR-V disassembly files (#2987) |
| 190 |
- Add iOS as a supported platform (#3001) |
| 191 |
- utils/vscode: Add SPIR-V language server support |
| 192 |
- Respect CMAKE_INSTALL_LIBDIR in installed CMake files (#3054) |
| 193 |
- Permit the debug instructions in WebGPU SPIR-V (#3063) |
| 194 |
- Add support for Fuchsia. (#3062) |
| 195 |
- Optimizer |
| 196 |
- Add descriptor array scalar replacement (#2742) |
| 197 |
- Add pass to wrap OpKill in a function call (#2790) |
| 198 |
- Fold FMix during constant folding. (#2818) |
| 199 |
- Add pass to replace AMD shader ballot extension (#2811) |
| 200 |
- Add pass to make Float32 operation relax precision (#2808) |
| 201 |
- Add pass to make relax precision operation Float16 (#2808) |
| 202 |
- Add pass to replace uses of 3 AMD extensions (#2814) |
| 203 |
- Fold Min, Max, and Clamp instructions. (#2836) |
| 204 |
- Better handling of OpKill in continues (#2842,#2922,#2933) |
| 205 |
- Enable OpTypeCooperativeMatrix specialization (#2927) |
| 206 |
- Support constant-folding UConvert and SConvert (#2960) |
| 207 |
- Update Offset to ConstOffset bitmask if operand is constant. (#3024) |
| 208 |
- Improve RegisterSizePasses (#3059) |
| 209 |
- Folding: perform add and sub on mismatched integer types (#3084) |
| 210 |
- Graphics robust access: use signed clamp (#3073) |
| 211 |
Fixes: |
| 212 |
- Instrument: Fix version 2 output record write for tess eval shaders. (#2782) |
| 213 |
- Instrument: Add support for Buffer Device Address extension (#2792) |
| 214 |
- Fix check for changed binary in API call. (#2798) |
| 215 |
- For WebGPU<->Vulkan optimization, set correct execution environment (#2834) |
| 216 |
- Handle OpConstantNull in copy-prop-arrays. (#2870) |
| 217 |
- Use OpReturn* in wrap-opkill (#2886) |
| 218 |
- Validator |
| 219 |
- Add generic builtin validation of target (#2843) |
| 220 |
- Extra resource interface validation (#2864) |
| 221 |
- Adding valilidation checks for OpEntryPoint duplicate names and execution mode (#2862) |
| 222 |
- Relaxed bitcast with pointers (#2878) |
| 223 |
- Validate physical storage buffer restrictions (#2930) |
| 224 |
- Add SPV_KHR_shader_clock validation (#2879, #3013) |
| 225 |
- Validate that selections are structured (#2962) |
| 226 |
- Disallow use of OpCompositeExtract/OpCompositeInsert with no indices (#2980) |
| 227 |
- Check that derivatives operate on 32-bit values (#2983) |
| 228 |
- Validate array stride does not cause overlap (#3028) |
| 229 |
- Validate nested constructs (#3068) |
| 230 |
Fixes: |
| 231 |
- Fix validation of constant matrices (#2794) |
| 232 |
- Update "remquor" validation |
| 233 |
- Only allow previously declared forward refs in structs (#2920) |
| 234 |
- Reduce |
| 235 |
- Remove relaxed precision decorations (#2797) |
| 236 |
- Reduce/fuzz: improve command line args (#2932) |
| 237 |
- Improve remove unref instr pass (#2945) |
| 238 |
Fixes: |
| 239 |
- Fuzz |
| 240 |
- Fix add-dead-break and add-dead-continue passes to respect dominance (#2838) |
| 241 |
- Add fuzzer pass to copy objects (#2853) |
| 242 |
- Add fuzzer pass to replace ids with synonyms (#2857) |
| 243 |
- Allow validation during spirv-fuzz replay (#2873) |
| 244 |
- Employ the "swarm testing" idea in spirv-fuzz (#2890) |
| 245 |
- reduce/fuzz: improve command line args (#2932) |
| 246 |
- option to convert shader into a form that renders red (#2934) |
| 247 |
- Add fuzzer pass to change selection controls (#2944) |
| 248 |
- add transformation and pass to construct composites (#2941) |
| 249 |
- Add fuzzer pass to change loop controls (#2949) |
| 250 |
- Add fuzzer pass to change function controls (#2951) |
| 251 |
- Add fuzzer pass to add NoContraction decorations (#2950) |
| 252 |
- Add missing functionality for matrix composites (#2974) |
| 253 |
- Fuzzer pass to adjust memory access operands (#2968) |
| 254 |
- Transformation to extract from a composite object (#2991) |
| 255 |
- Vector shuffle transformation (#3015) |
| 256 |
- Improve debugging facilities (#3074) |
| 257 |
- Function outlining fuzzer pass (#3078) |
| 258 |
|
| 259 |
|
| 260 |
v2019.4 2019-08-08 |
| 261 |
- General: |
| 262 |
- Memory model support for SPIR-V 1.4 |
| 263 |
- Add new spirv-fuzz tool |
| 264 |
- Add option for base branch in check_code_format.sh |
| 265 |
- Removed MarkV and Stats code. (#2576) |
| 266 |
- Instrument: Add version 2 of record formats (#2630) |
| 267 |
- Linker: Better type comparison for OpTypeArray and OpTypeForwardPointer (#2580) |
| 268 |
- Optimizer |
| 269 |
- Bindless Validation: Instrument descriptor-based loads and stores (#2583) |
| 270 |
- Better folding for OpSpecConstantOp (#2585, #2614) |
| 271 |
- Add in individual flags for Vulkan <-> WebGPU passes (#2615) |
| 272 |
- Handle nested breaks from switches. (#2624) |
| 273 |
- Optimizer: Handle array type with OpSpecConstantOp length (#2652) |
| 274 |
- Perform merge return with single return in loop. (#2714) |
| 275 |
- Add --preserve-bindings and --preserve-spec-constants (#2693) |
| 276 |
- Remove Common Uniform Elimination Pass (#2731) |
| 277 |
- Allow ray tracing shaders in inst bindle check pass. (#2733) |
| 278 |
- Add pass to inject code for robust-buffer-access semantics (#2771) |
| 279 |
- Treat access chain indexes as signed in SROA (#2776) |
| 280 |
- Handle RelaxedPrecision in SROA (#2788) |
| 281 |
- Add descriptor array scalar replacement (#2742) |
| 282 |
Fixes: |
| 283 |
- Handle decorations better in some optimizations (#2716) |
| 284 |
- Change the order branches are simplified in dead branch elim (#2728) |
| 285 |
- Fix bug in merge return (#2734) |
| 286 |
- SSA rewriter: Don't use trivial phis (#2757) |
| 287 |
- Record correct dominators in merge return (#2760) |
| 288 |
- Process OpDecorateId in ADCE (#2761) |
| 289 |
- Fix check for unreachable blocks in merge-return (#2762) |
| 290 |
- Handle out-of-bounds scalar replacements. (#2767) |
| 291 |
- Don't move debug or decorations when folding (#2772) |
| 292 |
- Protect against out-of-bounds references when folding OpCompositeExtract (#2774) |
| 293 |
- Validator |
| 294 |
- Validate loop merge (#2579) |
| 295 |
- Validate construct exits (#2459) |
| 296 |
- Validate OpenCL memory and addressing model environment rules (#2589) |
| 297 |
- Validate OpenCL environment rules for OpTypeImage (#2606) |
| 298 |
- Allow breaks to switch merge from nested construct (#2604) |
| 299 |
- Validate OpenCL environment rules for OpImageWrite (#2619) |
| 300 |
- Allow arrays of out per-primitive builtins for mesh shaders (#2617) |
| 301 |
- Validate OpenCL rules for ImageRead and OpImageSampleExplicitLod (#2643) |
| 302 |
- Add validation for SPV_EXT_fragment_shader_interlock (#2650) |
| 303 |
- Add builtin validation for SPV_NV_shader_sm_builtins (#2656) |
| 304 |
- Add validation for Subgroup builtins (#2637) |
| 305 |
- Validate variable initializer type (#2668) |
| 306 |
- Disallow stores to UBOs (#2651)A |
| 307 |
- Validate Volatile memory semantics bit (#2672) |
| 308 |
- Basic validation for Component decorations (#2679) |
| 309 |
- Validate that in OpenGL env block variables have Binding (#2685) |
| 310 |
- Validate usage of 8- and 16-bit types with only storage capabilities (#2704) |
| 311 |
- Add validation for SPV_EXT_demote_to_helper_invocation (#2707) |
| 312 |
- Extra small storage validation (#2732) |
| 313 |
- For Vulkan, disallow structures containing opaque types (#2546) |
| 314 |
- Validate storage class OpenCL environment rules for atomics (#2750) |
| 315 |
- Update OpControlBarriers rules for WebGPU (#2769) |
| 316 |
- Update OpMemoryBarriers rules for WebGPU (#2775) |
| 317 |
- Update WebGPU validation rules of OpAtomic*s (#2777) |
| 318 |
Fixes: |
| 319 |
- Disallow merge targeting block with OpLoopMerge (#2610) |
| 320 |
- Update vloadn and vstoren validation to match the OpenCL Extended |
| 321 |
Instruction Set Specification (#2599) |
| 322 |
- Update memory scope rules for WebGPU (#2725) |
| 323 |
- Allow LOD ops in compute shaders with derivative group execution modes (#2752) |
| 324 |
- Reduce |
| 325 |
Fixes: |
| 326 |
|
| 327 |
v2019.3 2019-05-14 |
| 328 |
- General: |
| 329 |
- Require Python 3 since Python 2 will out of service soon. |
| 330 |
- Add a continuous test that does memory checks using the address sanitizer. |
| 331 |
- Fix the build files so the SPIRV_USE_SANITIZER=address build works. |
| 332 |
- Packaging top of tree build artifacts again. |
| 333 |
- Added support for SPIR-V 1.4. (#2550) |
| 334 |
- Optimizer |
| 335 |
- Remove duplicates from list of interface IDs in OpEntryPoint instruction (#2449) |
| 336 |
- Bindless Validation: Descriptor Initialization Check (#2419) |
| 337 |
- Add option to validate after each pass (#2462) |
| 338 |
- Add legalization pass to fix mismatched pointer (#2430, #2535) |
| 339 |
- Add error messages when the input contains unknown instructions. (#2487) |
| 340 |
- Add pass to convert from WebGPU Spir-V to Vulkan Spir-V and back. (#2495) |
| 341 |
Fixes: |
| 342 |
- #2412: Dead memeber elimination should not change input and output variables. |
| 343 |
- #2405: Fix OpDot folding of half float vectors. |
| 344 |
- #2391: Dead branch elim should not fold away back edges. |
| 345 |
- #2441: Removing decorations when doing constant propagation. |
| 346 |
- #2455: Maintain inst to block mapping in merge return. |
| 347 |
- #2453: Fix merge return in the face of breaks. |
| 348 |
- #2456: Handle dead infinite loops in DCE. |
| 349 |
- #2458: Handle variable pointer in some optimizations. |
| 350 |
- #2452: Fix dead branch elimination to handle unreachable blocks better. |
| 351 |
- #2528: Fix undefined bit shift in sroa. |
| 352 |
- #2539: Change implementation of post order CFG traversal. |
| 353 |
- Validator |
| 354 |
- Add validation of storage classes for WebGPU (#2446) |
| 355 |
- Add validation for ExecutionMode in WebGPU (#2443) |
| 356 |
- Implement WebGPU specific CFG validation (#2386) |
| 357 |
- Allow NonWritable to target struct members. (#2420) |
| 358 |
- Allow storage type mismatch for parameter in relaxed addressing mode. |
| 359 |
- Allow non memory objects as parameter in relaxed addressing mode. |
| 360 |
- Disallow nested Blocks and buffer blocks (#2410). |
| 361 |
- Add validation for SPV_NV_cooperative_matrix (#2404) |
| 362 |
- Add --strip-atomic-counter-memory (#2413) |
| 363 |
- Check OpSampledImage is only passed into valid instructions (#2467) |
| 364 |
- Handle function decls in Structured CFG analysis (#2474) |
| 365 |
- Validate that OpUnreacahble is not statically reachable (#2473) |
| 366 |
- Add pass to generate needed initializers for WebGPU (#2481) |
| 367 |
- Allow images without format for OpenCL. (#2470) |
| 368 |
- Remove unreachable block validation (#2525) |
| 369 |
- Reduce runtime of array layout checks (#2534) |
| 370 |
- Add validation specific to OpExecutionModeId (#2536) |
| 371 |
- Validate sign of int types. (#2549) |
| 372 |
- VK_KHR_uniform_buffer_standard_layout validation (#2562) |
| 373 |
Fixes: |
| 374 |
- #2439: Add missing DepthGreater case to Fragment only check. |
| 375 |
- #2168: Disallow BufferBlock on StorageBuffer variables for Vulkan. |
| 376 |
- #2408: Restrict and Aliased decorations cannot be applied to the same id. |
| 377 |
- #2447: Improve function call parameter check. |
| 378 |
- Reduce |
| 379 |
- Add Pass to remove unreferenced blocks. (#2398) |
| 380 |
- Allows passing options to the validator. (#2401) |
| 381 |
- Improve reducer algorithm and other changes (#2472) |
| 382 |
- Add Pass to remove selections (#2485) |
| 383 |
- Add passes to simplify branches (#2507) |
| 384 |
Fixes: |
| 385 |
- #2478: fix loop to selection pass for loops with combined header/continue block |
| 386 |
|
| 387 |
v2019.2 2019-02-20 |
| 388 |
- General: |
| 389 |
- Support SPV_EXT_physical_storage_buffer |
| 390 |
- A number of memory leak have been fixed. |
| 391 |
- Removed use of deprecated Google test macro: |
| 392 |
- Changed the BUILD.gn to only build tests in Chromium. |
| 393 |
- Optimizer |
| 394 |
- Upgrade memory model improvments for modf and frexp. |
| 395 |
- Add a new pass to move loads closer to their uses: code sinking. |
| 396 |
- Invalidating the type manager now invalidates the constnat manager. |
| 397 |
- Expand instrumentation pass for bindless bounds checking to runtime-sized descriptor arrays. |
| 398 |
- Add a new pass that removes members from structs that are not used: dead member elimination. |
| 399 |
Fixes: |
| 400 |
- #2292: Remove undefined behaviour when folding bit shifts. |
| 401 |
- #2294: Fixes for instrumentation code. |
| 402 |
- #2293: Fix overflow when folding -INT_MIN. |
| 403 |
- #2374: Don't merge unreachable blocks when merging blocks. |
| 404 |
- Validator |
| 405 |
- Support SPV_KHR_no_integer_wrap and related decorations. |
| 406 |
- Validate Vulkan rules for OpTypeRuntimeArray. |
| 407 |
- Validate NonWritable decoration. |
| 408 |
- Many WebGPU specific validation rules were added. |
| 409 |
- Validate variable pointer related function call rules. |
| 410 |
- Better error messages. |
| 411 |
Fixes: |
| 412 |
- #2307: Check forwards references in OpTypeArray. |
| 413 |
- #2315, #2303: Fixed the layout check for relaxed layout. |
| 414 |
- #1628: Emit an error when an OpSwitch target is not an OpLabel. |
| 415 |
- Reduce |
| 416 |
- Added more documentation for spirv-reduce. |
| 417 |
- Add ability to remove OpPhi instructions. |
| 418 |
- Add ability to merge two basic blocks. |
| 419 |
- Add ability to remove unused functions and unused basic blocks. |
| 420 |
Fixes: |
| 421 |
|
| 422 |
v2019.1 2019-01-07 |
| 423 |
- General: |
| 424 |
- Created a new tool called spirv-reduce. |
| 425 |
- Add cmake option to turn off SPIRV_TIMER_ENABLED (#2103) |
| 426 |
- New optimization pass to update the memory model from GLSL450 to VulkanKHR. |
| 427 |
- Recognize OpTypeAccelerationStructureNV as a type instruction and ray tracing storage classes. |
| 428 |
- Fix GCC8 build. |
| 429 |
- Add --target-env flag to spirv-opt. |
| 430 |
- Add --webgpu-mode flag to run optimizations for webgpu. |
| 431 |
- The output disassembled line number stead of byte offset in validation errors. (#2091) |
| 432 |
- Optimizer |
| 433 |
- Added the instrumentation passes for bindless validation. |
| 434 |
- Added passes to help preserve OpLine information (#2027) |
| 435 |
- Add basic support for EXT_fragment_invocation_density (#2100) |
| 436 |
- Fix invalid OpPhi generated by merge-return. (#2172) |
| 437 |
- Constant and type manager have been turned into analysies. (#2251) |
| 438 |
Fixes: |
| 439 |
- #2018: Don't inline functions with a return in a structured CFG contstruct. |
| 440 |
- #2047: Fix bug in folding when volatile stores are present. |
| 441 |
- #2053: Fix check for when folding floating pointer values is allowed. |
| 442 |
- #2130: Don't inline recursive functions. |
| 443 |
- #2202: Handle multiple edges between two basic blocks in SSA-rewriter. |
| 444 |
- #2205: Don't unswitch a latch condition during loop unswitch. |
| 445 |
- #2245: Don't fold branch in loop unswitch. Run dead branch elimination to fold them. |
| 446 |
- #2204: Fix eliminate common uniform to place OpPhi instructions correctly. |
| 447 |
- #2247: Fix type mismatches caused by scalar replacement. |
| 448 |
- #2248: Fix missing OpPhi after merge return. |
| 449 |
- #2211: After merge return, fix invalid continue target. |
| 450 |
- #2210: Fix loop invariant code motion to not place code between merge instruction and branch. |
| 451 |
- #2258: Handle CompositeInsert with no indices in VDCE. |
| 452 |
- #2261: Have replace load size handle extact with no index. |
| 453 |
- Validator |
| 454 |
- Changed the naming convention of outputing ids with names in diagnostic messages. |
| 455 |
- Added validation rules for UniformConstant variables in Vulkan. |
| 456 |
- #1949: Validate uniform variable type in Vulkan |
| 457 |
- Ensure for OpVariable that result type and storage class operand agree (#2052) |
| 458 |
- Validator: Support VK_EXT_scalar_block_layout |
| 459 |
- Added Vulkan memory model semantics validation |
| 460 |
- Added validation checkes spefic to WebGPU environment. |
| 461 |
- Add support for VK_EXT_Transform_feedback capabilities (#2088) |
| 462 |
- Add validation for OpArrayLength. (#2117) |
| 463 |
- Ensure that function parameter's type is not void (#2118) |
| 464 |
- Validate pointer variables (#2111) |
| 465 |
- Add check for QueueFamilyKHMR memory scope (#2144) |
| 466 |
- Validate PushConstants annotation and type (#2140) |
| 467 |
- Allow Float16/Int8 for Vulkan 1.0 (#2153) |
| 468 |
- Check binding annotations in resource variables (#2151, #2167) |
| 469 |
- Validate OpForwardPointer (#2156) |
| 470 |
- Validate operation for OpSpecConstantOp (#2260) |
| 471 |
Fixes: |
| 472 |
- #2049: Allow InstanceId for NV ray tracing |
| 473 |
- Reduce |
| 474 |
- Initial commit wit a few passes to reduce test cases. |
| 475 |
- Validation is run after each reduction step. |
| 476 |
Fixes: |
| 477 |
|
| 478 |
|
| 479 |
v2018.6 2018-11-07 |
| 480 |
- General: |
| 481 |
- Added support for the Nvidia Turing and ray tracing extensions. |
| 482 |
- Make C++11 the CXX standard in CMakeLists.txt. |
| 483 |
- Enabled a parallel build for MSVC. |
| 484 |
- Enable pre-compiled headers for MSVC. |
| 485 |
- Added a code of conduct. |
| 486 |
- EFFCEE and RE2 are now required when build the tests. |
| 487 |
- Optimizer |
| 488 |
- Unrolling loops marked for unrolling in the legalization passes. |
| 489 |
- Improved the compile time of loop unrolling. |
| 490 |
- Changee merge-return to create a placeholder loop around the function. |
| 491 |
- Small improvement to merge-blocks to allow it to merge more often. |
| 492 |
- Enforce an upper bound for the ids, and add option to set it. |
| 493 |
- #1966: Report error if there are unreachable block before running merge return |
| 494 |
Fixes: |
| 495 |
- #1917: Allow 0 (meaning unlimited) as a parameter to --scalar-replacement |
| 496 |
- #1915: Improve handling of group decorations. |
| 497 |
- #1942: Fix incorrect uses of the constant manager. Avoids type mismatches in generated code. |
| 498 |
- #1997: Fix dead branch elimination when there is a loop in folded selection. |
| 499 |
- #1991: Fixes legality check in if-conversion. |
| 500 |
- #1987: Add nullptr check to array copy propagation. |
| 501 |
- #1984: Better handling of OpUnreachable in ADCE. |
| 502 |
- #1983: Run merge return on reachable functions only. |
| 503 |
- #1956: Handled atomic operations in ADCE. |
| 504 |
- #1963: Fold integer divisions by 0 to 0. |
| 505 |
- #2019: Handle MemberDecorateStringGOOGLE in ADCE and strip reflect. |
| 506 |
- Validator |
| 507 |
- Added validation for OpGroupNonUniformBallotBitCount. |
| 508 |
- Added validation for the Vulkan memory model. |
| 509 |
- Added support for VK_KHR_shader_atddomic_int64. |
| 510 |
- Added validation for execution modes. |
| 511 |
- Added validation for runtime array layouts. |
| 512 |
- Added validation for 8-bit storage. |
| 513 |
- Added validation of OpPhi instructions with pointer result type. |
| 514 |
- Added checks for the Vulkan memory model. |
| 515 |
- Validate MakeTexelAvailableKHR and MakeTexelVisibleKHR |
| 516 |
- Allow atomic function pointer for OpenCL. |
| 517 |
- FPRounding mode checks were implemented. |
| 518 |
- Added validation for the id bound with an option to set the max id bound. |
| 519 |
Fixes: |
| 520 |
- #1882: Improve the validation of decorations to reduce memory usage. |
| 521 |
- #1891: Fix an potential infinite loop in dead-branch-elimination. |
| 522 |
- #1405: Validate the storage class of boolean objects. |
| 523 |
- #1880: Identify arrays of type void as invalid. |
| 524 |
- #487: Validate OpImageTexelPointer. |
| 525 |
- #1922: Validate OpPhi instructions are at the start of a block correctly. |
| 526 |
- #1923: Validate function scope variable are at the start of the entry block. |
| 527 |
|
| 528 |
v2018.5 2018-09-07 |
| 529 |
- General: |
| 530 |
- Support SPV_KHR_vulkan_memory_model |
| 531 |
- Update Dim capabilities, to match SPIR-V 1.3 Rev 4 |
| 532 |
- Automated build bots no run tests for the VS2013 case |
| 533 |
- Support Chromium GN build |
| 534 |
- Use Kokoro bots: |
| 535 |
- Disable Travis-CI bots |
| 536 |
- Disable AppVeyor VisualStudio Release builds. Keep VS 2017 Debug build |
| 537 |
- Don't check export symbols on OSX (Darwin): some installations don't have 'objdump' |
| 538 |
- Reorganize source files and namespaces |
| 539 |
- Fixes for ClangTidy, and whitespace (passes 'git cl presumit --all -uf') |
| 540 |
- Fix unused param compile warnings/errors when Effcee not present |
| 541 |
- Avoid including time headers when timer functionality is disabled |
| 542 |
- Avoid too-stringent warnings flags for Clang on Windows |
| 543 |
- Internal refactoring |
| 544 |
- Add hooks for automated fuzzing |
| 545 |
- Add testing of command line executables |
| 546 |
- #1688: Use binary mode on stdin; fixes "spirv-dis <foo.spv" on Windows |
| 547 |
- Optimizer |
| 548 |
- The optimizer validates the module before it begins |
| 549 |
- Add API to register passes by string name |
| 550 |
- Fold a vector shuffle feeding a vector shuffle |
| 551 |
- Add -combine-access-chains transform |
| 552 |
- Refactor how IRContext is handled by passes |
| 553 |
- Improve bookkeeping for instruction result type and result id |
| 554 |
- Fix over-duplication of decorations |
| 555 |
- Fix handling of exits from selections in dead-branch elimination, and dead code |
| 556 |
elimination. |
| 557 |
- Fix handling of certain kinds of flow control in merge-return |
| 558 |
Fixes: |
| 559 |
- #1721: Fix size bug when folding vector shuffles |
| 560 |
- #1722: Fix size infinite loop when folding vector shuffles |
| 561 |
- #1724: Fix finding a constant of a specific type |
| 562 |
- #1727: Dead branch elim: Reorder blocks if needed to satisfy dominance rule |
| 563 |
- #1729: Handle VariablePointers cases in various optimizations |
| 564 |
- #1731: Fix vector shuffle with literal id indicating undef value |
| 565 |
- #1736: Fix handling of decorations and phis in merge-return |
| 566 |
- #1787: Fix handling of decorations related to access chains |
| 567 |
- #1865: Avoid leaking memory for SPIR-V constant values |
| 568 |
- Validator |
| 569 |
- Improve error messages |
| 570 |
- Avoid platform-dependent traversal ordering, to ensure consistent messages |
| 571 |
- Use libspirv::Instruction where possible |
| 572 |
- Add option to skip all block layout checks |
| 573 |
- Validate all type IDs |
| 574 |
- Validate uses of OpFunction |
| 575 |
- Validate uses of OpTypeFunction |
| 576 |
- Disallow a struct containing its own type https://crbug.com/874372 |
| 577 |
- #1685: Vulkan permits non-monotonic offsets for block members |
| 578 |
- #1697: Enforce block layout rules even when using relaxed block layout option |
| 579 |
- #1719: Fix line number for vector shuffle valiation error |
| 580 |
- #1789: Avoid assertion failure when validating some functions |
| 581 |
- #1800: Fix validation of OpCopyMemorySized |
| 582 |
- #1822: Stop enforcing struct member offset montonicity |
| 583 |
- #1831: Disallow void members in structs |
| 584 |
|
| 585 |
v2018.4 2018-07-08 |
| 586 |
- General: |
| 587 |
- Support SPV_KHR_8bit_storage |
| 588 |
- Add gclient and presubmit configurations |
| 589 |
- Enable Kokoro build bots (#1625) |
| 590 |
- Group tests into fewer executables, reduces load on CI |
| 591 |
- Port test script to Python 3 |
| 592 |
- Symbol export tests respect SPIRV_SKIP_TESTS |
| 593 |
- #1596: Operand lookup succeeds if enabled by a capability |
| 594 |
- #1624: Instruction lookup succeeds if enabled by a capability |
| 595 |
- Refactoring namespaces: |
| 596 |
- #1678: Change libspirv to spvtools |
| 597 |
- Code in source/utils moved into spvtools::utils |
| 598 |
- Code in source/comp moved into spvtools::comp |
| 599 |
- Optimizer: |
| 600 |
- Remove insert-extract-elim pass. Use simplification pass instead. |
| 601 |
- Preserve instruction-to-block mapping in most passes, to reduce runtime. |
| 602 |
- Small vector optimization for operands |
| 603 |
- Add pass to move Private variables to Function. Increase opportunity to optimize. |
| 604 |
- Fixes: |
| 605 |
#1120: Check static uses of entry point interfaces |
| 606 |
#1372: Avoid merging some structs, to preserve names for reflection |
| 607 |
#1577: Scalar replacement uses only undecorated types. |
| 608 |
#1578: Fix handling of forward-pointer types, and types embedding pointers |
| 609 |
to themselves. |
| 610 |
#1591: Inliner: Callee variable with initializer should have a store at the call site. |
| 611 |
#1634: Fix crash: Use type id in vector type lookup |
| 612 |
#1649: Fix assert in compact-ids pass |
| 613 |
Fix constant folder: ensure it uses the right type |
| 614 |
#1659: Folding rules added to IRContext. Avoids leak. |
| 615 |
- Validator |
| 616 |
- Add work-in-progress WebGPU environment. Disallows OpUndef |
| 617 |
- #670, #1581: Improve error messages; disassemble instruction |
| 618 |
- #491: Check structured switches |
| 619 |
- #937: Check layout rules for Block and BufferBlock in Uniform, StorageBuffer, PushConstant |
| 620 |
- #1281: Check invalid branches into structured constructs |
| 621 |
- #1522: Disallow array-of-arrays with DescriptorSets |
| 622 |
- #1577: Allow duplicate pointer types. |
| 623 |
- #1581: Better messages: output ID names along with numbers in more cases. |
| 624 |
- #1597: Check Vulkan 1.1 capabilities |
| 625 |
- #1618: Check invalid exit from structured case construct |
| 626 |
- #1622: Run IdPass before DataRulesPass |
| 627 |
- #1632: Reduce test time by artificially lowering limits in limit test |
| 628 |
- #1638: Block-decorated structs member order must respect offset order |
| 629 |
- #1657: Improve CFG validation diagnostics |
| 630 |
- Khronos SPIR-V #337: GLSL.std.450 Refract instruction Eta param can be any float scalar. |
| 631 |
- #1606: PushConstant Blocks follow storage-buffer layout rules |
| 632 |
- #1664: Check layout of StorageBuffer variables with Block decoration, using storage buffer |
| 633 |
rules |
| 634 |
- #1666: Layout validation should permit {vec3; float} packing |
| 635 |
- #1637, #1668: Layout validation uses RowMajor, ArrayStride, MatrixStride properly |
| 636 |
- Linker |
| 637 |
- Avoid buffer overrun when creating OpModuleProcessed |
| 638 |
|
| 639 |
v2018.3 2018-05-25 |
| 640 |
- General: |
| 641 |
- Support SPV_EXT_descriptor_indexing |
| 642 |
- Support SPV_GOOGLE_decorate_string |
| 643 |
- Support SPV_GOOGLE_hlsl_functionality1 |
| 644 |
- Support SPV_NV_shader_subgroup_partitioned |
| 645 |
- Use "unified1" grammar from SPIRV-Headers |
| 646 |
- Simplify support for new extensions. Assembler, disassembler, and simple validation |
| 647 |
support is automatic if new tokens are introduced with appropriate extension |
| 648 |
attributes in the "unified1" SPIR-V core grammar. |
| 649 |
- Disassembler: Emit more digits on floating point, to reliably reproduce all |
| 650 |
significand bits. (Use std::max_digits10 instead of std::digits10) |
| 651 |
- Fix compilation for old XCode versions: Explicit construction required for std::set. |
| 652 |
- Optimizer: |
| 653 |
- Add --strip-reflect |
| 654 |
- Add --time-report |
| 655 |
- Add --loop-fission |
| 656 |
- Add lop fusion. |
| 657 |
- Add loop peeling pass and internal utility. |
| 658 |
- Improve optimizer runtime. |
| 659 |
- Merge-return now works with structured control flow. |
| 660 |
- New (faster) SSA rewriter to convert local loads and stores to SSA IDs and phis. |
| 661 |
Can replace load/store elimination passes. |
| 662 |
- Fix instruction folding case: insertion that feeds and extract, when the extract |
| 663 |
remains. |
| 664 |
- Fold OpDot. |
| 665 |
- Fold OpFNegate. |
| 666 |
- Fold multply and divide of same value. |
| 667 |
- Fold FClamp feeding a compare. |
| 668 |
- Fold OpLoad feeding an extract, to reduce excessive copying. (#1547) |
| 669 |
- Fold Fmix feeding an extract. |
| 670 |
- Use simplification pass instead of insert-extract elimination. |
| 671 |
- Constant fold OpVectorTimesScalar. |
| 672 |
- Copy propagate arrays, in simple cases. |
| 673 |
- Aggressive dead code elimination: Can remove more instructions, e.g. derivatives. |
| 674 |
- Aggressive dead code elimination: Remove Workgroup variables that are written but not read. |
| 675 |
- Better handling of OpImageTexelPointer |
| 676 |
- Initial utilities for scalar evolution. |
| 677 |
- Add Vector dead code elimination. |
| 678 |
- Each pass can only run once. |
| 679 |
- Allow code hosting in if-conversion. |
| 680 |
- Add external interface for adding a PassToken, so external code can make their own |
| 681 |
passes. |
| 682 |
- Fixes: |
| 683 |
#1404: Don't optimize away the compute compute workgroup size constant. |
| 684 |
#1407: Remove a bad assertion |
| 685 |
#1456: Fix bug in SSA rewriter related to variables updated in loops. |
| 686 |
#1487: Fix long runtime in Dead insertion elimination: Don't revist select phi nodes. |
| 687 |
#1492: Aggressive dead code elimination can remove OpDecorateStringGOOGLE. |
| 688 |
#1527: Fix inlining of functions having OpKill and OpUnreachable. |
| 689 |
#1559: Fix assert failure in reduce-load-size pass. |
| 690 |
#1556: Aggressive dead code elimination: Fix handling of OpCopyMemory. |
| 691 |
- Validator: |
| 692 |
- Check Vulkan built-in variables |
| 693 |
- Check Vulkan-specific atomic result type rule. |
| 694 |
- Relax control barrier check for SPIR-V 1.3. Fixes #1427 |
| 695 |
- Check OpPhi. |
| 696 |
- Check OpMemoryModel. |
| 697 |
- Stop checking sizes derived from spec-constants. |
| 698 |
- Re-enable checks for OpUConvert. |
| 699 |
- Vulkan: Fix check for PrimitiveId: Permit as Input in fragment shader. |
| 700 |
- Validate binary version for the given target environment. |
| 701 |
- Add tests for OpBranch checks. |
| 702 |
- Vulkan 1.1: Check scope for non-uniform subgroup operations. |
| 703 |
- Fix checks for SPV_AMD_gpu_shader_int16. |
| 704 |
- Fix logical layout check for OpDecorateId. |
| 705 |
- Fix checks for ViewportIndex & Layer for Vulkan and SPV_EXT_shader_viewport_index_layer. |
| 706 |
- Fixes: |
| 707 |
#1470: Vulkan: Don't restrict WorkgroupSize to Input storage class. |
| 708 |
#1469: Vulkan: Permit Subgroup memory scope for Vulkan 1.1. |
| 709 |
#1472: Per-vertex variable validation fixes. |
| 710 |
#1483: Valdiate barrier execution scopes for Vulkan 1.1. |
| 711 |
- Fixes: |
| 712 |
#898: Linker properly removes FuncParamAttr from imported symbols. |
| 713 |
#924, #1174: Fix handling of decoration groups in optimizer, linker. |
| 714 |
|
| 715 |
v2018.2 2018-03-07 |
| 716 |
- General: |
| 717 |
- Support SPIR-V 1.3 and Vulkan 1.1. |
| 718 |
- Default target environment is now SPIR-V 1.3. For command-line tools, |
| 719 |
use the --target-env option to override the default. Examples: |
| 720 |
# Generate a SPIR-V 1.0 binary instead of SPIR-V 1.3 |
| 721 |
spirv-as --target-env spv1.0 a.spvasm -o a.spv |
| 722 |
spirv-as --target-env vulkan1.0 a.spvasm -o a.spv |
| 723 |
# Validate as Vulkan 1.0 |
| 724 |
spirv-val --target-env vulkan1.0 a.spv |
| 725 |
- Support SPV_GOOGLE_decorate_string and SPV_GOOGLE_hlsl_functionality1 |
| 726 |
- Fixes: |
| 727 |
- Fix Android.mk build. Compilation was failing due to missing definitions of |
| 728 |
SpvCapabilityFloat16ImageAMD and other enumerated values. |
| 729 |
- Optimizer: Avoid generating duplicate names when merging types. |
| 730 |
- #1375: Validator: SPV_AMD_gpu_shaer_half_float implicitly allows declaration |
| 731 |
of the 16-bit floating point type. |
| 732 |
- #1376: Optimizer: Avoid folding half-precision float. |
| 733 |
|
| 734 |
v2018.1 2018-03-02 |
| 735 |
- General: |
| 736 |
- Support Visual Studio 2013 again. (Continue support for VS 2015 and VS 2017.) |
| 737 |
- Support building SPIRV-Tools as a shared library. |
| 738 |
- Improve the HLSL legalization optimization recipe. #1311 |
| 739 |
- Optimizer: |
| 740 |
- General speedups. |
| 741 |
- Remove generic dead code elimination functionality from transforms: |
| 742 |
--eliminate-local-single-block |
| 743 |
--eliminate-local-single-store |
| 744 |
--eliminate-local-multi-store |
| 745 |
To recover the previous behaviour, a recipe using those transforms should now |
| 746 |
also invoke the --eliminate-dead-code-aggressive transform. |
| 747 |
- Improve folding, including coverage for floating point, OpSelect, and arithmetic |
| 748 |
with non-trivial constant operands. |
| 749 |
- Add loop-invariant code motion pass. |
| 750 |
- Add loop-unrolling pass, for honouring unroll hits. |
| 751 |
- Add loop-unswitch pass. |
| 752 |
- Add instruction simplification pass. |
| 753 |
- Aggressive dead code elimination: Understands capability hierarchy when finding |
| 754 |
instructions it can eliminate (combinators). (PR #1268) |
| 755 |
- CCP can now fold floating point arithmetic. #1311 |
| 756 |
- Validator: |
| 757 |
- Validate barrier instructions. |
| 758 |
- Check Vulkan-specific rules for atomics. |
| 759 |
- Check Vulkan prohibition of Location or Component decorations on BuiltIn variables. |
| 760 |
- Linker: |
| 761 |
- Add --verify-ids option |
| 762 |
- Add option to allow a resulting module to be partially linked. |
| 763 |
- Handle OpModuleProcessed (instructions in SPIR-V layout section 7c) |
| 764 |
- Fixes: |
| 765 |
- #1265: Optimizer: Fix use-after free bug in if-conversion. (Fix object lifecycle bug |
| 766 |
in type manager.) |
| 767 |
- #1282: Fix new warnings found by GCC 8.0.1. |
| 768 |
- #1285: Optimizer: Fix random failures during inlining. (Dangling references in DefUseManager) |
| 769 |
- #1295: Optimizer: Fix incorrect handling of Phi nodes in CCP. |
| 770 |
- #1300: Fix CCP: avoid bad CCP transitions and unsettled values. |
| 771 |
- #1304: Avoid static-duration variables of class type (with constructors). |
| 772 |
- #1323: Fix folding of an insert composite feeding a composite extract. |
| 773 |
- #1339: Fix CCP: Handle OpConstantNull boolean values as conditions. |
| 774 |
- #1341: DCEInst: Keep atomic instructions (and some others with side effects). |
| 775 |
- #1354: Don't fold integer division. |
| 776 |
- #1357: Support OpConstantNull in folding. |
| 777 |
- #1361: CCP: Fix handling of non-constant module-scope values |
| 778 |
|
| 779 |
v2018.0 2018-02-02 |
| 780 |
- General |
| 781 |
- VisualStudio 2013 is no longer supported. VisualStudio 2015 is supported. |
| 782 |
- Use "include/unified1" directory from SPIRV-Headers. Requires recent SPIRV-Headers source. |
| 783 |
- Disassembler: spirv-dis adds --color option to force color disassembly. |
| 784 |
- Optimizer: |
| 785 |
- Add pass to eliminate dead insertions. |
| 786 |
- Aggressive dead code elimination now removes OpSwitch constructs. |
| 787 |
- Block merging occurs in more cases. |
| 788 |
- Add driver workaround transform: replace OpUnreachable with harmless branch to merge. |
| 789 |
- Improve instruction folding framework. |
| 790 |
- Add loop analysis. |
| 791 |
- Add scalar replacement of aggregates to size-optimization recipe. |
| 792 |
- Add pass to replace instructions invalid for a shader stage, with a harmless value. |
| 793 |
This changes the semantics of the program! Not for general use! |
| 794 |
- Rearragne and add passes to performance-optimization recipe, to produce better results. |
| 795 |
- Validator: |
| 796 |
- Validate OpenCL extended instructions. |
| 797 |
- Shaders can't perform atomics on floats. |
| 798 |
- Validate memory semantics values in atomics. |
| 799 |
- Validate instruction-adjacency constraints, e.g. OpPhi predecessors, merge instructions |
| 800 |
immediately precede branches. |
| 801 |
- Fixes: |
| 802 |
- PR 1198: Optimizer: Fix CCP in presence of matrix constants. |
| 803 |
- #1199: Optimizer: Fix CCP: don't propagate spec constants. |
| 804 |
- #1203: Optimizer: Fix common uniform elim bug introduced by refactoring. |
| 805 |
- #1210: Optimizer: Aggressive dead code elimination: Fix 'break' identification. |
| 806 |
- #1212: Optimizer: Aggressive dead code elimination: Was skipping too many instructions. |
| 807 |
- #1214: Optimizer: Aggressive dead code elimination: Fix infinite loop. |
| 808 |
- #1228: Optimizer: Fix CCP: Handling of varying Phi nodes; was resulting in infinite loop. |
| 809 |
- #1245: Optimizer: Dead branch elimination: Avoid a null pointer dereference. |
| 810 |
- #1250: Optimizer: Dead branch elimination: Avoid spuriously reporting a change. |
| 811 |
|
| 812 |
v2017.3 2018-01-12 |
| 813 |
- General: |
| 814 |
- Support DebugInfo extended instruction set, targeted at OpenCL environments. |
| 815 |
See the SPIR-V Registry. |
| 816 |
- Generate a SPIRV-Tools.pc file for pkg-config. |
| 817 |
- Optimizer: |
| 818 |
- Progress for legalization of code generated from HLSL (issue #1118): |
| 819 |
- Add --legalize-hlsl option to run transforms used to transform intermediate |
| 820 |
code generated by HLSL to SPIR-V for Vulkan compilers. Those compilers |
| 821 |
normally run these transforms automatically. This option is used for developing |
| 822 |
those transforms. |
| 823 |
- Add Private-to-Function variable conversion for modules with logical |
| 824 |
addressing. |
| 825 |
- Add --ccp: SSA Conditional Constant Propagation (CCP) |
| 826 |
- Add --print-all to show disassembly for each optimization pass. |
| 827 |
- Internal: Add loop descriptors and post-order tree iterator. |
| 828 |
- Generalized dead branch elimination |
| 829 |
- Aggressive dead code elimination (ADCE) now removes dead functions and |
| 830 |
module-scope variables. |
| 831 |
- Vector extract/insert elimination now optimizes through some cases of |
| 832 |
VectorShuffle, and GLSL.std.450 Mix extended instruction. |
| 833 |
- Validator: |
| 834 |
- Add validation for GLSL.std.450 extended instruction set. |
| 835 |
- Check out of bounds composite accesses, where that's statically computable. |
| 836 |
Fixes #1112. |
| 837 |
- Check upper bits of literal numbers that aren't a multiple of 32-bits wide. |
| 838 |
- More validation of primitive instructions |
| 839 |
- Add optional "relaxed" checking logical addressing mode to permit some |
| 840 |
cases of pointer-to-pointer. Contributes to HLSL legalization (issue #1118). |
| 841 |
- Fixes: |
| 842 |
#1100: Validator: Image operand Sample can be used with OpImageSparseFetch, |
| 843 |
OpImageSparseRead. |
| 844 |
#1108: Remove duplicates transform was incorrectly removing non-duplicate |
| 845 |
decorations. |
| 846 |
#1111: Optimizer's type manager could reference deleted memory. |
| 847 |
#1112: Fix decoration equality check, e.g. it is now symmetric. |
| 848 |
#1129: Validator now disallows Dim=SupbassData for OpImageSparseRead. |
| 849 |
#1143: Fix CCP: Was generating incorrect code for loops. |
| 850 |
#1153: Fix CCP crash. |
| 851 |
#1154: Optimizer's internal instruction-to-block mappings were sometimes |
| 852 |
inconsistent. |
| 853 |
#1159: Fix CCP infinite loop. |
| 854 |
#1168: Fix dead branch elimination intermittently generating incorrect code. |
| 855 |
Fixes https://github.com/KhronosGroup/glslang/issues/1205 |
| 856 |
#1186: Fix validation of PackDouble2x32 and UnpackDouble2x32 |
| 857 |
|
| 858 |
v2017.2 2017-12-15 |
| 859 |
- General: |
| 860 |
- Support OpenCL 1.2, 2.0 target environments, including embedded profiles |
| 861 |
- Add CONTRIBUTING.md |
| 862 |
- Fix exit status code for spirv-link |
| 863 |
- Disassember: Enable emitting ANSI colour codes to a string |
| 864 |
- Library avoids polluting global namespace. The libraries can export C and C++ |
| 865 |
symbols starting with "spv", or in a C++ namespace. Add a test for this. |
| 866 |
- Linux release builds include debug information, for easier profiling |
| 867 |
- Build bots no longer test VisualStudio 2013 |
| 868 |
- Testing dependency RE2 requires VisualStudio 2015 or later |
| 869 |
- Build bots check code formatting |
| 870 |
- Optimizer: |
| 871 |
- Add --skip-validation to spirv-opt |
| 872 |
- Add dominance tree analysis |
| 873 |
- Add generic value propagation engine |
| 874 |
- Add global redundancy elimination within a function |
| 875 |
- Add scalar replacement of function-scope variables of composite type |
| 876 |
- Aggressive dead code elimination: Remove empty loops |
| 877 |
- Killing an instruction notifies the IRContext |
| 878 |
- IRContext::KillInst deletes the instruction |
| 879 |
- Move CFG analysis to IRContext |
| 880 |
- Add constant manager |
| 881 |
- Fix: Don't consider derivative instructions as combinators. |
| 882 |
- Fix: Don't delete an instruction twice in local dead-code-elimination |
| 883 |
- Fix: Don't consider derivative instructions as combinators. |
| 884 |
- Validator: |
| 885 |
- Finish checking of image instructions (Section 3.32.10) |
| 886 |
- Check sparse image instructions |
| 887 |
- Check OpTypeImage, OpTypeSampleImage |
| 888 |
- Check composite instructions (Section 3.32.12) |
| 889 |
- Check atomic instructions (Section 3.32.18) |
| 890 |
- Check OpEmitStreamVertex, OpEndStreamPrimitive instructions |
| 891 |
- Re-enable validation of OpCopyObject |
| 892 |
- OpKill, image ImplicitLod and QueryLod instructions can only be used in Fragment |
| 893 |
shaders. |
| 894 |
- Fixes for image instruction validation: |
| 895 |
- Lod image operand only usable with ExplicitLod and OpImageFetch |
| 896 |
- ExplicitLod Lod image operand must be float scalar |
| 897 |
- OpImageFectch Lod image operand must be int scalar |
| 898 |
- OpImageGather component operand must be 32-bits (integer scalar) |
| 899 |
- OpImageQuerySizeLod Lod must be integer scalar |
| 900 |
- Fixes: |
| 901 |
#622: Remove names and decorations when inlining |
| 902 |
#989: Aggressive dead code elim: Don't optimize away live breaks from a loop |
| 903 |
#991: Fix validation of SPV_AMD_shader_ballot |
| 904 |
#1004: Use after free of an instruction, in remove-duplicates transform |
| 905 |
#1007: OpImageRead not required to return 4-component vector |
| 906 |
#1009: OpImageRead can return scalar int/float types |
| 907 |
#1011: OpImageWrite should allow scalar int/float texel types |
| 908 |
#1012: Fix validat Dref type check |
| 909 |
#1017: Load-store elimination considers variable initializations |
| 910 |
#1034: Fix Windows debug build: operator< should be a weak ordering |
| 911 |
#1083: Inlining: Set parent (function) for each inlined basic block. |
| 912 |
#1075: Aggressive dead code elimination: Was leaving dangling references to |
| 913 |
removed blocks. |
| 914 |
|
| 915 |
v2017.1 2017-11-23 |
| 916 |
- Update README with details on the [email protected] mailing list. |
| 917 |
- General: |
| 918 |
- Automatically deploy built artifacts to GitHub Releases |
| 919 |
- Add a Linker (module combiner). Under development. |
| 920 |
- Add Android.mk for Android NDK builds. |
| 921 |
- Add the 'effcee' library as an optional dependency for use in tests. |
| 922 |
Eventually it will be a required dependency, once downstream projects have |
| 923 |
a chance to adjust. Requires 're2' library. |
| 924 |
- Avoid static-duration variables of class type (with constructors). |
| 925 |
- Hack around bugs in gcc-4.8.1 template handling |
| 926 |
- Faster opcode lookup |
| 927 |
- Validator: |
| 928 |
- Recognize extensions listed on SPIR-V registry, |
| 929 |
through #25 SPV_AMD_shader_fragment_mask |
| 930 |
- Validator issues an info message when it sees an unrecognized extension. |
| 931 |
- Type check basic arithmetic operations |
| 932 |
- Type check carry/extended arithmetic operations |
| 933 |
- Type check vector arithmetic operations |
| 934 |
- Type check Relational and Logical instructions |
| 935 |
- Type check Bit instructions |
| 936 |
- Check type uniqueness rules |
| 937 |
- Check conversion instructions |
| 938 |
- Check image instructions |
| 939 |
- Check derivative instructions |
| 940 |
- Check OpVectorShuffle |
| 941 |
- Check OpBranchConditional |
| 942 |
- OpModuleProcessed is only allowed after debug names section and before annotations |
| 943 |
section. |
| 944 |
- Checks the right kind of return is called for each function (void or non-void). |
| 945 |
- Add option to relax type check when storing structs (--relax-store-struct) |
| 946 |
- Optimizer: |
| 947 |
- Refactoring internal representation of the module, including: |
| 948 |
- IRContext: owns a module and manages analyses |
| 949 |
- Instructions are owned by intrusive lists, and have unique IDs |
| 950 |
- BasicBlock owns its instruction list. |
| 951 |
- DefUseManager: change representation of uses, for faster processing |
| 952 |
on large modules. |
| 953 |
- Add high level recipes: -O, -Os, and -Oconfig |
| 954 |
Recipes for -O and -Os are under development. |
| 955 |
- Add eliminate-dead-function transform |
| 956 |
- Add strength reduction transform: For now, convert multiply by power of 2 |
| 957 |
to a bit shift. |
| 958 |
- Add CFG cleanup transform |
| 959 |
- Add removal of dead module-scope variables |
| 960 |
- Add merge-return transform for modules without structured control flow |
| 961 |
- Add redundancy elimination within a basic block (local value numbering) |
| 962 |
- Extract-insert elimination: |
| 963 |
- Recognize the case where the first instruction in the sequence is an |
| 964 |
OpCompositeConstruct or OpConstantComposite |
| 965 |
- Handle some cases of nested structs |
| 966 |
- Dead branch elimination now can eliminate entire selection constructs |
| 967 |
when all arms are dead. |
| 968 |
- Compressing codec: |
| 969 |
- Updated algorithm to 1.01, 1.02, 1.03 |
| 970 |
- Not built by default. Use -DSPIRV_BUILD_COMPRESSION=ON to build. |
| 971 |
- Codec can be parameterized by a customized model. |
| 972 |
- Fixes: |
| 973 |
#728: Fix decoration of inlined functions |
| 974 |
#798: spirv-as should fail when given unrecognized long option |
| 975 |
#800: Inliner: Fix inlining function into header of multi-block loop |
| 976 |
#824: Eliminate-local-multi-store: Fix a crash |
| 977 |
#826: Elimiante-local-multi-store: Fix a crash |
| 978 |
#827: Fix crash when compact-ids transform runs before another transform. |
| 979 |
#834: Add Cmake option to build the compressing codec. Off by default. |
| 980 |
#911: Fix classification of Line and NoLine instructions |
| 981 |
|
| 982 |
v2017.0 2017-09-01 |
| 983 |
- Update README to describe that assembler, disassembler, and binary parser support |
| 984 |
are based on grammar files from the SPIRV-Headers repository. |
| 985 |
|
| 986 |
v2016.7 2017-09-01 |
| 987 |
- Add SPIR-V 1.2 |
| 988 |
- OpenCL 2.2 support is now based on SPIR-V 1.2 |
| 989 |
- Support AMD extensions in assembler, disassembler: |
| 990 |
SPV_AMD_gcn_shader |
| 991 |
SPV_AMD_shader_ballot |
| 992 |
SPV_AMD_shader_explicit_vertex_parameter |
| 993 |
SPV_AMD_shader_trinary_minmax |
| 994 |
SPV_AMD_gpu_shader_half_float |
| 995 |
SPV_AMD_texture_gather_bias_lod |
| 996 |
SPV_AMD_gpu_shader_int16 |
| 997 |
- Optimizer: Add support for: |
| 998 |
- Inline all function calls in entry points. |
| 999 |
- Flatten decoration groups. Fixes #602 |
| 1000 |
- Id compaction (minimize Id bound). Fixes #624 |
| 1001 |
- Eliminate redundant composite insert followed by extract |
| 1002 |
- Simplify access chains to local variables |
| 1003 |
- Eliminate local variables with a single store, if possible |
| 1004 |
- Eliminate local variables with a several stores, if possible |
| 1005 |
- Eliminate loads and stores in same block to local variables |
| 1006 |
- Eliminate redundant insert/extract to composite values |
| 1007 |
- Aggressive dead instruction elimination |
| 1008 |
- Eliminate dead branches |
| 1009 |
- Merge blocks when the second can only be preceded by the first |
| 1010 |
- Eliminate ommon uniform loads |
| 1011 |
- Assembler: Add option to preserve numeric ids. Fixes #625 |
| 1012 |
- Add build target spirv-tools-vimsyntax to generate spvasm.vim, a SPIR-V |
| 1013 |
assembly syntax file for Vim. |
| 1014 |
- Version string: Allow overriding of wall clock timestamp with contents |
| 1015 |
of environment variable SOURCE_DATE_EPOCH. |
| 1016 |
- Validator implements relaxed rules for SPV_KHR_16bit_storage. |
| 1017 |
- CMake installation rules use GNUInstallDirs. For example, libraries |
| 1018 |
will be installed into a lib64 directory if that's the norm for the |
| 1019 |
current system. |
| 1020 |
- Fixes: |
| 1021 |
#500: Parameterize validator limit checks |
| 1022 |
#508: Support compilation under CYGWIN |
| 1023 |
#517: Fix validation when continue (or case) contstruct is also the head of a |
| 1024 |
nested control construct. |
| 1025 |
#551: If a merge block is reachable, it must be *strictly* dominated by its |
| 1026 |
header. |
| 1027 |
#548: Validator: Error when the reserved OpImageSparseSampleProj* opcodes |
| 1028 |
are used. |
| 1029 |
#611: spvtools::Optimizer was failing to save the module to the output |
| 1030 |
binary vector when all passes succeded without changes. |
| 1031 |
#629: The inline-entry-points-all optimization could generate invalidly |
| 1032 |
structured code when the inlined function had early returns. |
| 1033 |
#697: Optimizer's Instruction::ForEachInId method was skipping semantics-id |
| 1034 |
and scope-id. |
| 1035 |
#755: Inliner: Fix inlining of callee with single Return appearing before |
| 1036 |
the end of the function. |
| 1037 |
#776: Fix dead branch elimination in presence of complex but dead control |
| 1038 |
flow. |
| 1039 |
#781: SPV_KHR_variable_pointers allows duplicate pointer types |
| 1040 |
#782: Inliner: Fix remapping of non-label forward references in callee |
| 1041 |
#787: Inliner: Fix remapping of inlined entry block when called from |
| 1042 |
single block loop. |
| 1043 |
#790: Inliner: Fix remapping of inlined entry block when callee has |
| 1044 |
multiple returns. |
| 1045 |
|
| 1046 |
v2016.6 2016-12-13 |
| 1047 |
- Published the C++ interface for assembling, disassembling, validation, and |
| 1048 |
optimization. |
| 1049 |
- Support SPV_KHR_shader_draw_parameters in assembler, disassembler, parser. |
| 1050 |
- Validator: |
| 1051 |
- Add validator API accepting raw binary words |
| 1052 |
- Increased coverage: |
| 1053 |
- Checks "Data rules" in Universal Validation Rules, section 2.16.1 |
| 1054 |
- WIP: Universal Limits. |
| 1055 |
- The minimum mandated upper bounds are checked. |
| 1056 |
- TODO: Parameterize the validator to allow larger limits accepted by |
| 1057 |
a more than minimally capable implementation. |
| 1058 |
- OpSampledImage checks |
| 1059 |
- OpConstantComposite checks |
| 1060 |
- Id bound check |
| 1061 |
- Disasssembler: |
| 1062 |
- Generates friendly GLSL-based names for more builtin variables |
| 1063 |
- Generates friendly names for numeric OpConstant values |
| 1064 |
- Vendor tool info extracted from SPIR-V XML registry file. |
| 1065 |
- Fixes issues: |
| 1066 |
#429: Validator: Allow OpTypeForwardPointer and OpTypeStruct to reference |
| 1067 |
undefined IDs |
| 1068 |
#482: Validator: OpVariable initializer can be an ID of a module-scope variable |
| 1069 |
|
| 1070 |
v2016.5 2016-09-16 |
| 1071 |
- Support SPV_KHR_shader_ballot in assembler, disassembler, parser. |
| 1072 |
- Disassembler: Generate friendly names for built-in variables. |
| 1073 |
- Partial fixes: |
| 1074 |
#359: Add Emacs helper for automatically diassembling/assembling a SPIR-V |
| 1075 |
binary on file load/save. |
| 1076 |
- Fixes: |
| 1077 |
#414: Validator: Allow OpUndef for composite constants |
| 1078 |
#415: Validator: Phi can use its own value in some cases. |
| 1079 |
|
| 1080 |
v2016.4 2016-09-01 |
| 1081 |
- Relicensed under Apache 2.0 |
| 1082 |
- Add optimization passes (in API and spirv-opt command) |
| 1083 |
- Fold spec constants defined with OpSpecConstantOp and |
| 1084 |
OpSpecConstantComposite to normal constants with fixed value(s). |
| 1085 |
- Fixes issues: |
| 1086 |
#318: Relicensed under Apache 2.0 |
| 1087 |
|
| 1088 |
v2016.3 2016-08-24 |
| 1089 |
- Add target environment enums for OpenCL 2.1, OpenCL 2.2, |
| 1090 |
OpenGL 4.0, OpenGL 4.1, OpenGL 4.2, OpenGL 4.3, OpenGL 4.5. |
| 1091 |
- Add spirv-cfg, an experimental tool to dump the control flow graph |
| 1092 |
as a GraphiViz "dot" graph |
| 1093 |
- Add optimization pass: Eliminate dead constants. |
| 1094 |
- Add spirv-lesspipe.sh filter utility |
| 1095 |
- Fixes issues: |
| 1096 |
#288: Check def-use dominance rules for OpPhi (variable,parent) operands |
| 1097 |
#339: Allow OpUndef in types-constants-global-vars section, as required |
| 1098 |
by SPIR-V 1.0 Rev7, 1.1 Rev 3. |
| 1099 |
#340: Avoid race on mkdir during build |
| 1100 |
#365: Relax PointSize, ClipDistance, CullDistance capability check in all |
| 1101 |
environments not just Vulkan 1.0. |
| 1102 |
|
| 1103 |
v2016.2 2016-08-05 |
| 1104 |
- Validator is incomplete |
| 1105 |
- Checks ID use block is dominated by definition block |
| 1106 |
- Add optimization passes (in API and spirv-opt command) |
| 1107 |
- Strip debug info instructions |
| 1108 |
- Freeze spec constant to their default values |
| 1109 |
- Allow INotEqual as operation for OpSpecConstantOp |
| 1110 |
- Fixes bugs: |
| 1111 |
#270: validator: crash when continue construct is unreachable |
| 1112 |
#279: validator: infinite loop when analyzing some degenerate control |
| 1113 |
flow graphs |
| 1114 |
#286: validator: don't incorrectly generate def-use error for |
| 1115 |
(variable,parent) parameters to OpPhi |
| 1116 |
#290: disassembler: never generate bare % for an identifier |
| 1117 |
#295: validator: def-use dominance check should ignore unreachable uses |
| 1118 |
#276: validator: allow unreachable continue constructs |
| 1119 |
#297: validator: allow an unreachable block to branch to a reachable |
| 1120 |
merge block |
| 1121 |
|
| 1122 |
v2016.1 2016-07-19 |
| 1123 |
- Fix https://github.com/KhronosGroup/SPIRV-Tools/issues/261 |
| 1124 |
Turn off ClipDistance and CullDistance capability checks for Vulkan. |
| 1125 |
- The disassembler can emit friendly names based on debug info (OpName |
| 1126 |
instructions), and will infer somewhat friendly names for most types. |
| 1127 |
This is turned on by default for the spirv-dis command line tool. |
| 1128 |
- Updated to support SPIR-V 1.1 rev 2 |
| 1129 |
- Input StorageClass, Sampled1D capability, and SampledBuffer capability |
| 1130 |
do not require Shader capability anymore. |
| 1131 |
|
| 1132 |
v2016.0 2016-07-04 |
| 1133 |
|
| 1134 |
- Adds v<year>.<index> versioning, with "-dev" indicating |
| 1135 |
work in progress. The intent is to more easly report |
| 1136 |
and summarize functionality when SPIRV-Tools is incorporated |
| 1137 |
in downstream projects. |
| 1138 |
|
| 1139 |
- Summary of functionality (See the README.md for more): |
| 1140 |
- Supports SPIR-V 1.1 Rev 1 |
| 1141 |
- Supports SPIR-V 1.0 Rev 5 |
| 1142 |
- Supports GLSL std450 extended instructions 1.0 Rev 3 |
| 1143 |
- Supports OpenCL extended instructions 1.0 Rev 2 |
| 1144 |
- Assembler, disassembler are complete |
| 1145 |
- Supports floating point widths of 16, 32, 64 bits |
| 1146 |
- Supports integer widths up to 64 bits |
| 1147 |
- Validator is incomplete |
| 1148 |
- Checks capability requirements in most cases |
| 1149 |
- Checks module layout constraints |
| 1150 |
- Checks ID use-definition ordering constraints, |
| 1151 |
ignoring control flow |
| 1152 |
- Checks some control flow graph rules |
| 1153 |
- Optimizer is introduced, with few available transforms. |
| 1154 |
- Supported on Linux, OSX, Android, Windows |
| 1155 |
|
| 1156 |
- Fixes bugs: |
| 1157 |
- #143: OpenCL pow and pown arguments |