projects / oota-llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: a637e58) | patch
Implement v16i8 multiply with this code:
authorChris Lattner <sabre@nondot.org>
Tue, 18 Apr 2006 03:57:35 +0000 (03:57 +0000)
committerChris Lattner <sabre@nondot.org>
Tue, 18 Apr 2006 03:57:35 +0000 (03:57 +0000)
commit19a815238e55458e95f99b4dad31ed053c9f635c
treed7a1e0c59ae8e186db779fc4787641557f9e81f9tree | snapshot
parenta637e589187ae92302fc33645df903507b33d262commit | diff
Implement v16i8 multiply with this code:

        vmuloub v5, v3, v2
        vmuleub v2, v3, v2
        vperm v2, v2, v5, v4

This implements CodeGen/PowerPC/vec_mul.ll.  With this, v16i8 multiplies are
6.79x faster than before.

Overall, UnitTests/Vector/multiplies.c is now 2.45x faster with LLVM than with
GCC.

Remove the 'integer multiplies' todo from the README file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@27792 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/PowerPC/PPCISelLowering.cpp diff | blob | history
lib/Target/PowerPC/README_ALTIVEC.txt diff | blob | history
C/C++ LLVM-based compilers that forbids OOTA behaviors