Skip to content

Change Logs

Jump to bottom Edit New page
wuye9036 edited this page Jul 10, 2015 · 2 revisions

0.5.8 (Jul 10, 2015)

  • Ported code repository from Hg to Git
    • Separated resources and third parties from main repository
  • Upgrade LLVM to 3.6.1
  • Removed most MSVC 12.0 level 3 warnings.

0.5.7 (Dec 22, 2014)

  • Feature and improvements

    • New sample framework is available. Playback, interactive, benchmark and testing support are integrated to one framework
    • Improved UMD device
    • Developing register-file based shader compiler and runtime
    • Introduced C++ 11 features (lambda and for-each) to replace boost features
    • Simplified unit tests files for shader testing
    • Renamed robust_test to death_test
    • Refactor fixed_string to make it simpler
    • Ported sasl enums to C++ 11 enums
    • Remove unused WTL code
    • Added constants for indicating compilation configuration.

  • Bug fixing

    • Remove unused code for old vertex_shader_unit.
    • Fixed align_free crash while pointer is null.
    • Fixed an crash error in d3d11_swap_chain deconstructor while present has not been called before release.
    • Fixed Font crash while msyh.ttc is not existed on system
    • Fixed performance issue while converting rgba8 to bgra8
    • Now render_core returns immediately if color targets and DS target are all none

0.5.6 (Jun 05, 2014)

  • Feature and improvements

    • Rasterizer/Pixel shader/Sample rendering are working on quad(2x2)
    • Add two new vertex cache algorithm, and common vertex cache code was extracted to vertex_cache_impl
    • Remove old vertex_shader_unit. Now vertex shader always use host.
    • Add time stamp profiling into pipeline
    • New benchmark framework
    • Benchmark could generate detailed performance report as JSON format

  • Optimization: 27% for Sponza and 6.6% for complex mesh

    • geom_setup_engine are shared between draw calls
    • Add fetch3 to reduce fetch function called.
    • New vertex cache algorithm to reduce time of vertex gathering.
    • Eliminated unused malloc/free in geom_setup_engine.
    • For early-z test, depth-test-write are executed together.

  • Bug fix

    • Fixed interpolation error in lerp_n.
    • Fixed backend_input_pixel counter.
    • LoD could be shared in quad pixels in cpp shader.
    • Host will return null pointer when vertex shader is not set.
    • Fixed bug of early-z condition computation.

0.5.5 (May 06, 2014)

  • Building and deployment

    • Supported platforms changed:
      • vs2012 or higher;
      • gcc 4.7 or higher;
    • Support MinGW on Windows;
    • Support GCC on Linux;
    • Boost building parameters changed for cross platform support.
    • SCL macros are moved to CMake files.

  • Third-parties

    • LLVM upgraded to 3.4.
    • LLVM-boost workaround code is removed.

  • Optimization

    • Performance improved 60% - 70%
    • Removed duplicated triangle info computation.
    • Add quick test for clipper.
    • interpolation function works on quad to reduce multiply ops.
    • c++ pixel shader
      • has quad info to instead triangle info.
      • ddx/ddy is computed by subtraction directly.
      • since step-subtraction method for ddx/ddy is removed, related functions are removed also.
    • removed unused vso operators.
    • optimized derivative computation by edge.
    • Reused transformed verts pool in default_vertex_cache.
    • move front_face from vso to triangle_ctx.
    • vso viewport transform are executed parallel.

  • Refactor

    • Refactor file organization of shader/shader_regs/shader_regs_op
    • cube map implementation was changed.
    • interfaces of texture/sampler/surface have been changed.

0.5.4 (Dec 30, 2013)

  • Feature

    • ESM-based Shadow Map demo and stencil based mirror/reflect demo;
    • MinGW and MinGW64 were supported;
    • Depth and stencil can be cleaned separately;
    • Pipeline could works with DS target only;
    • Now pixel shader could overwrite depth(PS depth output);

  • Refactor

    • Texture is not an available resource for shader, but sampler holds texture and could be used in shader.
    • All third-party libraries is intergated as source code and CMake was added to them.
    • Map/Unmap is belongs to renderer but not resource for synchronization.
    • Due to C++11/boost.locale, implementation of multibyte-unicode conversation was improved.
    • Mask of swizzle/write mask is not uint32 any longer. elem_indexes is introduced thus mask is strong-typed.
    • Improve build scripts and CMake files for:
      • Executable of build/make/compiler toolchains detection is supported or improved.
      • Clean variable names of CMake.
      • DirectX SDK will not be used in VS2012 or later.
      • Report and interrupt immediately once error occurs.
      • Code of project configuring and building were refractored.
    • Cleanup code of texture input/output.

  • Bug fixes

    • Fixed a layout computation bug for "float4 main(): SV_Position"
    • Fixed count_bits functions.
    • Buffer allocation is 16-bytes aligned.
    • Fixed error that culled pixel rendered to frame buffer.
    • Fixed default Z-range in pipeline.
    • Fixed a bugs of GL swap chain.
    • Fixed a potential issues in member of __m128 settings.
    • Fixed bugs caused by incorrect cache in shims.
    • Fixes batch of bugs in shader compiler.
    • Fixed windows kit and d3d detection.

0.5.3 (Sep 10, 2013)

  • Refactor pipeline!
    • All commands (draw, clear, asynchoronous begin/end) were treated as commands. Every command has a immutable state object.
    • Old renderer_impl was removed.
    • render_core was added to execute command .
    • New renderer_impl is designed to set or get status from state object.
    • Renderers were refactored to async_renderer and sync_renderer. They are all based on renderer_impl and render_core
    • async_renderer holds an asynchronized queue for executing commands asynchronized.
    • Clipper is stateless between primitives.
    • Add new stage geom_setup_engine to process IA and clipping. Its output is primitives for rasterizer.
    • Add thread_context to pipeline. It will make code more readable when function is parallel called.
    • Interpolation method was changed to base of total derivative but not integral any more.
    • Replace "presenters" with swap chain.
  • Third-party and tool-chain changes
    • Upgraded LLVM to 3.3.
    • Only support VS2010 and later.
    • Remove dependencies of GDI+ and D3DX.
  • New features
    • Support performance counter.
    • Support early-z (per sample).
  • Bug fixes and misc
    • Compact Sponza scene file by collapse vertexes. It will make Sponza demo ~30% faster.
    • Fixed bugs of default_vso2reg.
    • Replace macro guard with pragma.
    • Now set_variable of cpp shader is async-safey.

0.5.2 (Feb 10, 2013)

  • 蛇年新春快乐！
  • Optimized performance of interpolator ~25% - 50%.
  • New architecture of shader is applied to Vertex Shader.
  • Visual Studio 2012 have been supported offically.
  • The minimum requirement of boost is 1.53.0 for now.
  • Remove some implementations from eflib when it is supported by boost.
  • Build problems caused by DX SDK environment are fixed.
  • dynamic_lib: Shared library will be free when reference count is 0.
  • Some bugs fixed.

0.5.1 (Dec 28, 2012)

  • The last update of SALVIA in 2012. Happy new year!
  • Optimized performance of parser and semantic analyzer. The performance is increased 70% in debug mode and about 150% in release mode.
  • *New architecture of code generation which support pixel shader much better than old. *
  • ABI of data transferred between pipeline and pixel shader was changed.
  • Code clean.
  • Explicit cast operator has been supported.
  • Bug fixes:
    • fast_roundi always returns 0 when /fp:fast is enabled on MSVC.
    • Add SEH guard to preventing JIT error stops the whole test.
    • Add cast_f2i.
    • All comments are in English for now.

0.3.6 (Sep 12, 2012)

  • Now API are called asynchronized.
  • Add inference for l-value and r-value.
  • Fixed a bug that input data will be modified if semantic variable have been assigned in vertex shader.
  • Fixed threadpool compilation error when Boost 1.51 is used.
  • Refactor code generation.
  • Now LLVM code generator is the only valid code generator.
  • Fixed some compilation error and run-time bugs.
  • Add "--clean" option to build_all.py to support clean build.

0.3.5 (Aug 21, 2012)

  • Add 28 intrinsic support to vertex shader.
  • Support anisotropic texture filter and added an demo.
  • Integrated FreeType2, added font drawing code and demo for it.
  • Re-organize documentations.

0.3.4 (Jul 17, 2012)

  • Sorry I am late!
  • Add 3 IMPORTANT samples!
    • Skinning Mesh
    • Multiple stream based vertex blending
    • Terrain Rendering based on Vertex Texture Fetching
  • Shader compiling was at least 11% faster than old one.
  • Support simple COLLADA model.
  • Context of semantic analysis was changed.
  • Refactor code about code generation.

0.3.3 (May 08, 2012)

  • All operators enabled for scalar, vector and matrix in Vertex Shader.
  • Intrinsics now support matrix parameters.
  • Add new intrinsics support to Vertex Shader: sin/cos/tan/asin/acos/atan/ceil/floor/log/log2/log10/rsqrt/exp2/ldexp.
  • Support vector and matrix per-element implicit cast.
  • Add system include and include search path.
  • Add missing LLVM files to fix compiling bugs.
  • Port eflib matrix and vector to generic version.
  • Refactor code generators and external intrinsics machanism.

0.3.2 (Apr 20, 2012)

  • Add intrinsics abs, any, all, asint, asuint, asfloat, distance, dst, exp, fmod, lerp, radians, degrees support to vertex shader and general for scalar and vector.
  • Add operators div, mod, bitwise and, bitwise or, logic and, logic or, comparison operators, shift operators support to vertex shader and general for scalar and vector.
  • Add include and virtual include to SASL. Didn't support system include search path yet.
  • Handling more errors in semantic.
  • Bool operators are no longer short-evaluation. Only logic bool operator for scalar in general is shortly.
  • Project host now depend on driver.
  • Multiple swizzle/write-mask now works.
  • Now defining variable in loop is no longer cause a stack overflow.
  • Implicit cast is executabe when it is needed in before call.

0.3.1 (Mar 30, 2012)

  • Add error handling for tokenizer and parser
  • Partially support error report for semantic analysis
  • Split sasl_compiler to sasl_driver and sasl_command.
    • sasl_driver is facade of compiler classes.
    • sasl_command is compiler as command line executable file .
  • Now jit test is based on driver.
  • Add document in Chinese.
  • Add CMake Project Group support.
  • Add emit_abs for SISD.
  • Fixed a false error report of 'no overloads of operator '#''
  • Fixed presenter name bug in ObjLoader when DX is enabled.
  • Fixed cmake generator error when vs2005/2008 used.

0.3 (Mar 06, 2012)

  • Add tex2D support to pixel shader.
  • Fixed a crash bug on ntx64 release without debugger.
  • Add constructors to builtin type.

0.2.6 (Mar 01, 2012)

  • Integrated Pixel Shader to Pipeline!
  • Fixed jittering on Sponza when grand band is enabled.
  • Default clip planes is reduced to 2 faces.
  • Scanline based rasterizer was removed, now Larrabee rasterizer is as default.

0.2.5 (Feb 24, 2012)

  • Pixel Shader now support for, while and do-while loop.
  • Add ddx and ddy support.
  • Add texture sampler prototype to pixel shader.
  • Optimized rasterizer.

0.2.4 (Feb 11, 2012)

  • Now pixel shader of SASL support if-statement via execution mask.
  • Fix a bug that cmake cannot be customized.
  • Fix ambiguous overloads bug caused by vector1.
  • Fix a bug of memory alignment error in jit test.
  • Add a prototype of SSA analyser, in future it is used for optimization.

0.2.3 (Jan 06, 2012)

  • Now DirectX depenencies is not necessary of SALVIA any more.
  • Visual C++ Express now could be used to compile SALVIA. (x86 32-bit only)
  • Added intrinsics 'dot', 'cross' and 'sqrt' support to pixel shader.
  • Fixed unaligned memory bug because asm generated by MSVC x86 is not keep the stack pointer 16 bytes aligned.

0.2.2 (Dec 25, 2011)

  • Added ALL-IN-ONE build script.
  • Finished prototype of pixel shader compiler.
    • Implemented memory layout of parameters and variables of pixel shader.
    • Added operators '+', '-', '*' support to scalar and vector on SIMD mode.
    • Added swizzle and write mask support to vector on SIMD mode.
  • Removed pre-built llvm libraries located in 3rd_party/llvm
  • Moved llvm source code from 3rd_party/src/llvm to 3rd_party/llvm
  • Renamed platform name and re-organized build directories.
  • Arranged documents.
  • Added project depenedency to fix test installation.
  • Now vertex output data is 16 bytes aligned.
  • Members of vertex output data now confirms C member alignment.

0.2.1 (Nov 30, 2011)

  • Added switch-case support to vertex shader
  • Integer and floating point now could be casted to boolean implicit.

0.2 (Nov 25, 2011)

  • Fixed unaligment memory access exception by SSE instruction while platform is x64 or SSE is enabled on x86-32.
  • New features have been added to vertex shader:
    • if-then-else, while, for statements.
    • break and continue statements
    • intrinsics support: mul, cross, sqrt and dot
    • SSE acceleration on x86 and x64
  • Refactored code generation in SASL.
  • Regression test framework and tests for Just-In-Time.

0.1.1 (Jul 11, 2011)

  • Add new demo Sponza.
  • Add 32-bit index support for wavefront obj.
  • Fixed a mipmap computation bug.
  • Fixed a culling bug.

0.1.0 (Jun 01, 2011)

  • Project configuration and management:
    • Code repository now changed to Mercurial.
    • Rename !SoftArt to SALVIA.
    • Logo of SALVIA is published.
    • Format of check in log now obey google issue tracker's demands.
    • Added unit test and auto test system.
    • Merge LLVM 2.9 source code to repo.
    • Fixed some compatibility bugs between LLVM and Boost.
    • Boost now requires 1.44 or later.
    • Add a tool for statistic code lines.
    • Documents were updated.
  • Graphis:
    • Add an OpenGL based presenter.
    • Optimize subdivision based rasterizer.
    • Changed namespace and organization of EFLIB.
    • Add centroid sampling support.
    • Add 2x and 4x MSAA support.
  • SASL:
    • Add a new combinator based parser to replace Boost.Spirit for reduce compiling time and object file size.
    • Add preprocessor based on Boost.Wave.
    • Add name mangling and overloading.
    • Finished shader ABI design.
    • Add JIT support.
    • Cross compilers is supported better.
    • Add CL styled compiler.
    • Add a reusable multi-targets compiler framework.
    • Add regression test to compiler.
    • Support following language features:
      • Scalar types, vector types, structures
      • Sequential statements.
      • Function declaration and definition.
      • Arithmentic operators and member operators(.)
      • Swizzle and mask.

0.0.1 (Aug 23, 2010)

  • Full render pipeline.
  • New Larrabee like rasterizer is enabled and optimized.
  • Optimized texture sampling code by SSE.
  • Added shader ABI specification.
  • Protype of parser, semantic analyzer and code generator of shader compiler works now.
Add a custom footer
Add a custom sidebar
Clone this wiki locally