llvm-6502/test
Adam Nemet 49f31255be [AVX512] Fix miscompile for unpack
r189189 implemented AVX512 unpack by essentially performing a 256-bit unpack
between the low and the high 256 bits of src1 into the low part of the
destination and another unpack of the low and high 256 bits of src2 into the
high part of the destination.

I don't think that's how unpack works.  AVX512 unpack simply has more 128-bit
lanes but other than it works the same way as AVX.  So in each 128-bit lane,
we're always interleaving certain parts of both operands rather different
parts of one of the operands.

E.g. for this:
__v16sf a = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15 };
__v16sf b = { 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31 };
__v16sf c = __builtin_shufflevector(a, b, 0, 8, 1, 9, 4, 12, 5, 13, 16,
	    			       	     24, 17, 25, 20, 28, 21, 29);

we generated punpcklps (notice how the elements of a and b are not interleaved
in the shuffle).  In turn, c was set to this:

  0 16 1 17 4 20 5 21 8 24 9 25 12 28 13 29

Obviously this should have just returned the mask vector of the shuffle
vector.

I mostly reverted this change and made sure the original AVX code worked
for 512-bit vectors as well.

Also updated the tests because they matched the logic from the code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@217602 91177308-0d34-0410-b5e6-96231b3b80d8
2014-09-11 16:51:10 +00:00
..
Analysis
Assembler
Bindings
Bitcode
BugPoint
CodeGen [AVX512] Fix miscompile for unpack 2014-09-11 16:51:10 +00:00
DebugInfo
ExecutionEngine
Feature
FileCheck
Instrumentation [asan-assembly-instrumentation] Added CFI directives to the generated instrumentation code. 2014-09-10 09:45:49 +00:00
Integer
JitListener
Linker Merge alignment of common GlobalValue. 2014-09-09 17:48:18 +00:00
LTO
MC Object: Add support for bigobj 2014-09-10 12:51:52 +00:00
Object
Other
TableGen
tools Remember to eraseFromParent after replaceAllUsesWith. 2014-09-10 19:39:41 +00:00
Transforms [AlignmentFromAssumptions] Don't crash just because the target is 32-bit 2014-09-11 08:40:17 +00:00
Unit
Verifier
YAMLParser
.clang-format
CMakeLists.txt Add LLVMgold target to test dependencies. 2014-09-10 22:20:49 +00:00
lit.cfg
lit.site.cfg.in
Makefile
Makefile.tests
TestRunner.sh