Introduction

The Gallium llvmpipe driver is a software rasterizer that uses LLVM to do runtime code generation. Shaders, point/line/triangle rasterization and vertex processing are implemented with LLVM IR which is translated to x86 or x86-64 machine code. Also, the driver is multithreaded to take advantage of multiple CPU cores (up to 8 at this time). It's the fastest software rasterizer for Mesa.

Requirements

• An x86 or amd64 processor; 64-bit mode recommended.

  Support for SSE2 is strongly encouraged. Support for SSSE3 and SSE4.1 will yield the most efficient code. The fewer features the CPU has the more likely is that you run into underperforming, buggy, or incomplete code.

  See /proc/cpuinfo to know what your CPU supports.

• LLVM: version 2.9 recommended; 2.6 or later required.

  NOTE: LLVM 2.8 and earlier will not work on systems that support the Intel AVX extensions (e.g. Sandybridge). LLVM's code generator will fail when trying to emit AVX instructions. This was fixed in LLVM 2.9.

  For Linux, on a recent Debian based distribution do:

       aptitude install llvm-dev
  

  For a RPM-based distribution do:

       yum install llvm-devel
  

  For Windows you will need to build LLVM from source with MSVC or MINGW (either natively or through cross compilers) and CMake, and set the LLVM environment variable to the directory you installed it to. LLVM will be statically linked, so when building on MSVC it needs to be built with a matching CRT as Mesa, and you'll need to pass -DLLVM_USE_CRT_RELEASE=MTd for debug and checked builds, -DLLVM_USE_CRT_RELEASE=MTd for profile and release builds. You can build only the x86 target by passing -DLLVM_TARGETS_TO_BUILD=X86 to cmake.

• scons (optional)

Building

To build everything on Linux invoke scons as:

  scons build=debug libgl-xlib

Alternatively, you can build it with GNU make, if you prefer, by invoking it as

  make linux-llvm

but the rest of these instructions assume that scons is used. For Windows the procedure is similar except the target:

  scons build=debug libgl-gdi

Using

On Linux, building will create a drop-in alternative for libGL.so into

  build/foo/gallium/targets/libgl-xlib/libGL.so

or

  lib/gallium/libGL.so

To use it set the LD_LIBRARY_PATH environment variable accordingly. For performance evaluation pass debug=no to scons, and use the corresponding lib directory without the "-debug" suffix. On Windows, building will create a drop-in alternative for opengl32.dll. To use it put it in the same directory as the application. It can also be used by replacing the native ICD driver, but it's quite an advanced usage, so if you need to ask, don't even try it.

Profiling

To profile llvmpipe you should build as

  scons build=profile <same-as-before>

This will ensure that frame pointers are used both in C and JIT functions, and that no tail call optimizations are done by gcc.

Linux perf integration

On Linux, it is possible to have symbol resolution of JIT code with Linux perf:

	perf record -g /my/application
	perf report

When run inside Linux perf, llvmpipe will create a /tmp/perf-XXXXX.map file with symbol address table. It also dumps assembly code to /tmp/perf-XXXXX.map.asm, which can be used by the bin/perf-annotate-jit script to produce disassembly of the generated code annotated with the samples.

You can obtain a call graph via Gprof2Dot.

Unit testing

Building will also create several unit tests in build/linux-???-debug/gallium/drivers/llvmpipe:

• lp_test_blend: blending
• lp_test_conv: SIMD vector conversion
• lp_test_format: pixel unpacking/packing

Some of this tests can output results and benchmarks to a tab-separated-file for posterior analysis, e.g.:

  build/linux-x86_64-debug/gallium/drivers/llvmpipe/lp_test_blend -o blend.tsv

Development Notes

• When looking to this code by the first time start in lp_state_fs.c, and then skim through the lp_bld_* functions called in there, and the comments at the top of the lp_bld_*.c functions.
• The driver-independent parts of the LLVM / Gallium code are found in src/gallium/auxiliary/gallivm/. The filenames and function prefixes need to be renamed from "lp_bld_" to something else though.
• We use LLVM-C bindings for now. They are not documented, but follow the C++ interfaces very closely, and appear to be complete enough for code generation. See this stand-alone example. See the llvm-c/Core.h file for reference.

Recommended Reading

• Rasterization

  • Triangle Scan Conversion using 2D Homogeneous Coordinates
  • Rasterization on Larrabee (DevMaster copy)
  • Rasterization using half-space functions
  • Advanced Rasterization
  • Optimizing Software Occlusion Culling

• Texture sampling

  • Perspective Texture Mapping
  • Texturing As In Unreal
  • Run-Time MIP-Map Filtering
  • Will "brilinear" filtering persist?
  • Trilinear filtering
  • Texture Swizzling

• SIMD

  • Whole-Function Vectorization

• Optimization

  • Optimizing Pixomatic For Modern x86 Processors
  • Intel 64 and IA-32 Architectures Optimization Reference Manual
  • Software optimization resources
  • Intel Intrinsics Guide

• LLVM

  • LLVM Language Reference Manual
  • The secret of LLVM C bindings

• Misc

  • WARP Architecture and Performance