Arnold Schwaighofer 5f0d9dbdf4 X86 cost model: Adjust cost for custom lowered vector multiplies ... This matters for example in following matrix multiply: int **mmult(int rows, int cols, int **m1, int **m2, int **m3) { int i, j, k, val; for (i=0; i<rows; i++) { for (j=0; j<cols; j++) { val = 0; for (k=0; k<cols; k++) { val += m1[i][k] * m2[k][j]; } m3[i][j] = val; } } return(m3); } Taken from the test-suite benchmark Shootout. We estimate the cost of the multiply to be 2 while we generate 9 instructions for it and end up being quite a bit slower than the scalar version (48% on my machine). Also, properly differentiate between avx1 and avx2. On avx-1 we still split the vector into 2 128bits and handle the subvector muls like above with 9 instructions. Only on avx-2 will we have a cost of 9 for v4i64. I changed the test case in test/Transforms/LoopVectorize/X86/avx1.ll to use an add instead of a mul because with a mul we now no longer vectorize. I did verify that the mul would be indeed more expensive when vectorized with 3 kernels: for (i ...) r += a[i] * 3; for (i ...) m1[i] = m1[i] * 3; // This matches the test case in avx1.ll and a matrix multiply. In each case the vectorized version was considerably slower. radar://13304919 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176403 91177308-0d34-0410-b5e6-96231b3b80d8		2013-03-02 04:02:52 +00:00
..
Analysis	X86 cost model: Adjust cost for custom lowered vector multiplies	2013-03-02 04:02:52 +00:00
Archive
Assembler	Modify the LLVM assembly output so that it uses references to represent function attributes.	2013-02-20 07:21:42 +00:00
Bindings/Ocaml	Use the attribute group reference instead of the attribute directly.	2013-02-20 07:48:23 +00:00
Bitcode	Unify clang/llvm attributes for asan/tsan/msan (LLVM part)	2013-02-26 06:58:09 +00:00
BugPoint
CodeGen	Revert "Rewrite a test to count emitted instructions without using -stats"	2013-03-02 00:53:20 +00:00
DebugInfo	Debug Info: for static member variables, always put AT_MIPS_linkage_name to the	2013-02-27 23:21:02 +00:00
ExecutionEngine	Corrections for XFAIL armv5 tests	2013-02-28 10:05:10 +00:00
Feature	Modify the LLVM assembly output so that it uses references to represent function attributes.	2013-02-20 07:21:42 +00:00
FileCheck	Canonicalize line endings to Linux style also when the --strict-whitespace flag is in use. This flag is supposed to affect horizontal whitespaces only.	2013-02-06 20:40:38 +00:00
Instrumentation	[msan] Implement sanitize_memory attribute.	2013-02-28 11:25:14 +00:00
Integer
JitListener	Add support for source and line information to IntelJITEventListener for object emitted by MCJIT.	2013-01-28 19:52:37 +00:00
Linker	Linker: correctly link in dbg.declare	2013-01-31 21:19:18 +00:00
MC	Rewrite a test to check actual output rather than intermediate implementation	2013-03-01 20:54:00 +00:00
Object	[llvm-readobj] Add ELF .dynamic table dumping.	2013-02-20 02:37:12 +00:00
Other	IR: Don't constant fold GEP bitcasts between different address spaces	2013-02-27 02:26:42 +00:00
Scripts	Set the +x bit on two batch scripts	2013-01-25 22:12:50 +00:00
TableGen	FileCheck-ify some grep tests	2013-01-25 22:11:46 +00:00
tools
Transforms	X86 cost model: Adjust cost for custom lowered vector multiplies	2013-03-02 04:02:52 +00:00
Unit
Verifier	[IR] Add 'Append' and 'AppendUnique' module flag behaviors.	2013-01-16 21:38:56 +00:00
YAMLParser
CMakeLists.txt	llvm/test/CMakeLists.txt: Add a dependency to llvm-rtdyld in check-llvm.	2013-01-25 23:24:07 +00:00
lit.cfg	Introduce llvm::sys::getProcessTriple() function.	2013-01-16 17:27:22 +00:00
lit.site.cfg.in	Introduce llvm::sys::getProcessTriple() function.	2013-01-16 17:27:22 +00:00
Makefile	No need to force-create clang-tools-extra lit.site.cfg	2013-03-01 19:58:58 +00:00
Makefile.tests
TestRunner.sh