Document the SLP infrastructure.
authorNadav Rotem <nrotem@apple.com>
Sun, 14 Apr 2013 07:42:25 +0000 (07:42 +0000)
committerNadav Rotem <nrotem@apple.com>
Sun, 14 Apr 2013 07:42:25 +0000 (07:42 +0000)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@179480 91177308-0d34-0410-b5e6-96231b3b80d8

docs/Vectorizers.rst

index fa0f49c3ddfd675a31bc0ea12421c5fbc191711a..42e03d297842dd8844a98cc86fde797e3026b929 100644 (file)
@@ -329,3 +329,33 @@ into vector operations.
   }
 
 
+.. _slp-vectorizer:
+
+The SLP Vectorizer
+==========================
+
+The SLP vectorizer (superword-level parallelism) is a new experimental 
+infrastructure for vectorizing code and rolling loops. 
+A major focus of the work on the SLP vectorizer is to make it fast and
+flexible. It is designed as a library that can be used by other passes.
+
+The SLP vectorizer is in early development stages but can already vectorize
+and accelerate many programs in the LLVM test suite.
+
+=======================   ============
+Benchmark Name              Gain
+=======================   ============
+Misc/flops-7               -32.70%
+Misc/matmul_f64_4x4        -23.23%
+Olden/power                -21.45%
+Misc/flops-4               -14.90%
+ASC_Sequoia/AMGmk          -13.85%
+TSVC/LoopRerolling-flt     -11.76%
+Misc/flops-6               -9.70%
+Misc/flops-5               -8.54%
+Misc/flops                 -8.12%
+TSVC/NodeSplitting-dbl     -6.96%
+Misc-C++/sphereflake       -6.74%
+Ptrdist/yacr2              -6.31%
+=======================   ============
+