src/gallium/drivers/llvmpipe/README

   1 LLVMPIPE -- a fork of softpipe that employs LLVM for code generation.
   2
   3
   4 Status
   5 ======
   6
   7 Done so far is:
   8
   9  - the whole fragment pipeline is code generated in a single function
  10
  11    - input interpolation
  12
  13    - depth testing
  14
  15    - texture sampling (not all state/formats are supported)
  16
  17    - fragment shader TGSI translation
  18      - same level of support as the TGSI SSE2 exec machine, with the exception
  19        we don't fallback to TGSI interpretation when an unsupported opcode is
  20        found, but just ignore it
  21      - done in SoA layout
  22      - input interpolation also code generated
  23
  24    - alpha testing
  25
  26    - blend (including logic ops)
  27      - both in SoA and AoS layouts, but only the former used for now
  28
  29  - code is generic
  30    - intermediates can be vectors of floats, ubytes, fixed point, etc, and of
  31      any width and length
  32    - not all operations are implemented for these types yet though
  33
  34 Most mesa/progs/demos/* work.
  35
  36 To do (probably by this order):
  37
  38  - code generate stipple and stencil testing
  39
  40  - translate the remaining bits of texture sampling state
  41
  42  - translate TGSI control flow instructions, and all other remaining opcodes
  43
  44  - integrate with the draw module for VS code generation
  45
  46  - code generate the triangle setup and rasterization
  47
  48
  49 Requirements
  50 ============
  51
  52  - Linux
  53
  54  - A x86 or amd64 processor.  64bit mode is preferred.
  55
  56    Support for sse2 is strongly encouraged.  Support for ssse3, and sse4.1 will
  57    yield the most efficient code.  The less features the CPU has the more
  58    likely is that you ran into underperforming, buggy, or incomplete code.
  59
  60    See /proc/cpuinfo to know what your CPU supports.
  61
  62  - LLVM 2.5 or greater. LLVM 2.6 is preferred.
  63
  64    On Debian based distributions do:
  65
  66      aptitude install llvm-dev
  67
  68    There is a typo in one of the llvm 2.5 headers, that may cause compilation
  69    errors. To fix it apply the change:
  70
  71      --- /usr/include/llvm-c/Core.h.orig        2009-08-10 15:38:54.000000000 +0100
  72      +++ /usr/include/llvm-c/Core.h     2009-08-10 15:38:25.000000000 +0100
  73      @@ -831,7 +831,7 @@
  74         template<typename T>
  75         inline T **unwrap(LLVMValueRef *Vals, unsigned Length) {
  76           #if DEBUG
  77      -    for (LLVMValueRef *I = Vals, E = Vals + Length; I != E; ++I)
  78      +    for (LLVMValueRef *I = Vals, *E = Vals + Length; I != E; ++I)
  79             cast<T>(*I);
  80           #endif
  81           return reinterpret_cast<T**>(Vals);
  82
  83  - scons (optional)
  84
  85  - udis86, http://udis86.sourceforge.net/ (optional):
  86
  87      git clone git://udis86.git.sourceforge.net/gitroot/udis86/udis86
  88      cd udis86
  89      ./autogen.sh
  90      ./configure --with-pic
  91      make
  92      sudo make install
  93
  94
  95 Building
  96 ========
  97
  98 To build everything invoke scons as:
  99
 100   scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=xlib dri=false -k
 101
 102 Alternatively, you can build it with GNU make, if you prefer, by invoking it as
 103
 104   make linux-llvm
 105
 106 but the rest of these instructions assume that scons is used.
 107
 108
 109 Using
 110 =====
 111
 112 Building will create a drop-in alternative for libGL.so. To use it set the
 113 environment variables:
 114
 115   export LD_LIBRARY_PATH=$PWD/build/linux-x86_64-debug/lib:$LD_LIBRARY_PATH
 116
 117 or
 118
 119   export LD_LIBRARY_PATH=$PWD/build/linux-x86-debug/lib:$LD_LIBRARY_PATH
 120
 121 For performance evaluation pass debug=no to scons, and use the corresponding
 122 lib directory without the "-debug" suffix.
 123
 124
 125 Unit testing
 126 ============
 127
 128 Building will also create several unit tests in
 129 build/linux-???-debug/gallium/drivers/llvmpipe:
 130
 131  - lp_test_blend: blending
 132  - lp_test_conv: SIMD vector conversion
 133  - lp_test_format: pixel unpacking/packing
 134
 135 Some of this tests can output results and benchmarks to a tab-separated-file
 136 for posterior analysis, e.g.:
 137
 138   build/linux-x86_64-debug/gallium/drivers/llvmpipe/lp_test_blend -o blend.tsv
 139
 140
 141 Development Notes
 142 =================
 143
 144 - When looking to this code by the first time start in lp_state_fs.c, and
 145   then skim through the lp_bld_* functions called in there, and the comments
 146   at the top of the lp_bld_*.c functions.
 147
 148 - All lp_bld_*.[ch] are isolated from the rest of the driver, and could/may be
 149   put in a stand-alone Gallium state -> LLVM IR translation module.
 150
 151 - We use LLVM-C bindings for now. They are not documented, but follow the C++
 152   interfaces very closely, and appear to be complete enough for code
 153   generation. See
 154   http://npcontemplation.blogspot.com/2008/06/secret-of-llvm-c-bindings.html
 155   for a stand-alone example.