<ul>
<li><p><code>OCL_SIMD_WIDTH</code> <code>(8 or 16)</code>. Change the number of lanes per hardware thread</p></li>
-<li><p><code>OCL_OUTPUT_GEN_IR</code> <code>(0 or 1)</code>. Output Gen ISA code into stdout</p></li>
+<li><p><code>OCL_OUTPUT_GEN_IR</code> <code>(0 or 1)</code>. Output Gen IR (scalar intermediate
+representation) code</p></li>
<li><p><code>OCL_OUTPUT_LLVM</code> <code>(0 or 1)</code>. Output LLVM code after the lowering passes</p></li>
<li><p><code>OCL_OUTPUT_LLVM_BEFORE_EXTRA_PASS</code> <code>(0 or 1)</code>. Output LLVM code before the
lowering passes</p></li>
<li><a href="doc/c++_simulator.html">C++ OpenCL simulator</a></li>
</ul>
-<p>Ben Segovia (<a href="mailto:benjamin.segovia@intel.com">benjamin.segovia@intel.com</a>)</p>
+<p>Ben Segovia (<a href="mailto:benjamin.segovia@intel.com">benjamin.segovia@intel.com</a>)</p>
<h2>LLVM front-end</h2>
-<p>The code is defined in <code>src/llvm/</code>. We used the PTX ABI and the OpenCL profile
+<p>The code is defined in <code>src/llvm</code>. We used the PTX ABI and the OpenCL profile
to compile the code. Therefore, a good part of the job is already done. However,
many things must be implemented:</p>
<ul>
<li><p>Lowering down of various intrinsics like <code>llvm.memcpy</code></p></li>
<li><p>Implementation of most of the OpenCL built-ins (<code>native_cos</code>, <code>native_sin</code>,
-atomic operations, barriers...)</p></li>
+<code>mad</code>, atomic operations, barriers...)</p></li>
<li><p>Lowering down of int16 / int8 / float16 / char16 / char8 / char4 loads and
stores into the supported loads and stores</p></li>
<li><p>Support for constant buffers declared in the OpenCL source file</p></li>
<h2>Gen IR</h2>
-<p>The code is defined in <code>src/ir/</code>. Main things to do are:</p>
+<p>The code is defined in <code>src/ir</code>. Main things to do are:</p>
<ul>
<li><p>Bringing support for doubles</p></li>
<li><p>Finishing the handling of function arguments (see the <a href="gen_ir.html">IR
description</a> for more details)</p></li>
<li><p>Adding support for constant data per unit</p></li>
-<li><p>Adding support for linking IR units together</p></li>
+<li><p>Adding support for linking IR units together. OpenCL indeed allows to create
+programs from several sources</p></li>
</ul>
<h2>Backend</h2>
<p>The code is defined in <code>src/backend</code>. Main things to do are:</p>
+
+<ul>
+<li><p>Bringing backend support for the missing instructions described above
+(native<em>sin, native</em>cos, barriers, samples...)</p></li>
+<li><p>Implementing support for doubles</p></li>
+<li><p>Implementing register spilling (see the <a href="./compiler_backend.html">compiler backend
+description</a> for more details)</p></li>
+<li><p>Implementing proper instruction selection. A "simple" tree matching algorithm
+should provide good results for Gen</p></li>
+<li><p>Implementing the instruction scheduling pass</p></li>
+</ul>
+
+<h2>General plumbering</h2>
+
+<p>I tried to keep the code clean, well, as far as C++ can be really clean. There
+are some header cleaning steps required though, in particular in the backend
+code.</p>
+
+<p>The context used in the IR code generation (see <code>src/ir/context.*pp</code>) should be
+split up and cleaned up too.</p>
+
+<p>I also purely and simply copied and pasted the Gen ISA disassembler from Mesa.
+This leads to code duplication. Also some messages used by OpenCL (untyped reads
+and writes) are not properly decoded yet.</p>
+
+<p>There are some quick and dirty hacks also like the use of function call <code>system</code>
+(...). This should be cleanly replaced by popen and stuff. I also directly
+called the LLVM compiler executable instead of using Clang library. All of this
+should be improved and cleaned up. Track "XXX" comments in the code.</p>
+
+<p>Parts of the code leaks memory when exceptions are used. There are some pointers
+to track and replace with std::unique_ptr. Note that we also add a custom memory
+debugger that nicely complements (i.e. it is fast) Valgrind.</p>
LLVM front-end
--------------
-The code is defined in `src/llvm/`. We used the PTX ABI and the OpenCL profile
+The code is defined in `src/llvm`. We used the PTX ABI and the OpenCL profile
to compile the code. Therefore, a good part of the job is already done. However,
many things must be implemented:
Gen IR
------
-The code is defined in `src/ir/`. Main things to do are:
+The code is defined in `src/ir`. Main things to do are:
- Bringing support for doubles
- Implementing the instruction scheduling pass
+General plumbering
+------------------
+
+I tried to keep the code clean, well, as far as C++ can be really clean. There
+are some header cleaning steps required though, in particular in the backend
+code.
+
+The context used in the IR code generation (see `src/ir/context.*pp`) should be
+split up and cleaned up too.
+
+I also purely and simply copied and pasted the Gen ISA disassembler from Mesa.
+This leads to code duplication. Also some messages used by OpenCL (untyped reads
+and writes) are not properly decoded yet.
+
+There are some quick and dirty hacks also like the use of function call `system`
+(...). This should be cleanly replaced by popen and stuff. I also directly
+called the LLVM compiler executable instead of using Clang library. All of this
+should be improved and cleaned up. Track "XXX" comments in the code.
+
+Parts of the code leaks memory when exceptions are used. There are some pointers
+to track and replace with std::unique_ptr. Note that we also add a custom memory
+debugger that nicely complements (i.e. it is fast) Valgrind.
+
projects / contrib / beignet.git / commitdiff