Mehdi Amini 5eed637b34 Improve DAG combine pass on certain IR vector patterns ... Loading 2 2x32-bit float vectors into the bottom half of a 256-bit vector produced suboptimal code in AVX2 mode with certain IR combinations. In particular, the IR optimizer folded 2f32 + 2f32 -> 4f32, 4f32 + 4f32 (undef) -> 8f32 into a 2f32 + 2f32 -> 8f32, which seems more canonical, but then mysteriously generated rather bad code; the movq/movhpd combination didn't match. The problem lay in the BUILD_VECTOR optimization path. The 2f32 inputs would get promoted to 4f32 by the type legalizer, eventually resulting in a BUILD_VECTOR on two 4f32 into an 8f32. The BUILD_VECTOR then, recognizing these were both half the output size, concatted them and then produced a shuffle. However, the resulting concat + shuffle was more complex than it should be; in the case where the upper half of the output is undef, we probably want to generate shuffle + concat instead. This enhancement causes the vector_shuffle combine step to recognize this suboptimal pattern and correct it. I included it there instead of in BUILD_VECTOR in case the same suboptimal pattern occurs for other reasons. This results in the optimizer correctly producing the optimal movq + movhpd sequence for all three variations on this IR, even with AVX2. I've included a test case. Radar link: rdar://problem/19287012 Fix for PR 21943. From: Fiona Glaser <fglaser@apple.com> git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@226360 91177308-0d34-0410-b5e6-96231b3b80d8		2015-01-17 01:35:56 +00:00
..
Analysis	[PM] Port domtree to the new pass manager (at last).	2015-01-14 10:19:28 +00:00
Assembler	IR: Allow 16-bits for column info	2015-01-16 17:33:08 +00:00
Bindings	IR: Move MDLocation into place	2015-01-14 22:27:36 +00:00
Bitcode	Revert r226242 - Revert Revert Don't create new comdats in CodeGen	2015-01-16 08:38:45 +00:00
BugPoint	IR: Move MDLocation into place	2015-01-14 22:27:36 +00:00
CodeGen	Improve DAG combine pass on certain IR vector patterns	2015-01-17 01:35:56 +00:00
DebugInfo	IR: Move MDLocation into place	2015-01-14 22:27:36 +00:00
ExecutionEngine	[MCJIT] Remove a few redundant MCJIT tests, and drop the extraneous datalayout	2015-01-08 18:52:15 +00:00
Feature	IR: Move MDLocation into place	2015-01-14 22:27:36 +00:00
FileCheck
Instrumentation	IR: Move MDLocation into place	2015-01-14 22:27:36 +00:00
Integer
JitListener	IR: Move MDLocation into place	2015-01-14 22:27:36 +00:00
Linker	IR: Move MDLocation into place	2015-01-14 22:27:36 +00:00
LTO	Put this test's input in the Inputs directory where it belongs, rather than	2015-01-12 08:50:47 +00:00
MC	Revert r226242 - Revert Revert Don't create new comdats in CodeGen	2015-01-16 08:38:45 +00:00
Object	Fix edge case when Start overflowed in 32 bit mode	2015-01-15 23:50:44 +00:00
Other	[PM] Port TargetLibraryInfo to the new pass manager, provided by the	2015-01-15 11:39:46 +00:00
SymbolRewriter	Transform: add SymbolRewriter pass	2014-11-07 21:32:08 +00:00
TableGen	[AVX512] Added intrinsics for VPCMPEQB and VPCMPEQW.	2014-09-30 11:32:22 +00:00
tools	Change the test case for llvm-objdump’s -archive-headers option to not check the size	2015-01-16 23:29:07 +00:00
Transforms	Fix Reassociate handling of constant in presence of undef float	2015-01-16 03:00:58 +00:00
Unit
Verifier	Rename llvm.recoverframeallocation to llvm.framerecover	2015-01-13 01:51:34 +00:00
YAMLParser
.clang-format
CMakeLists.txt	Revert r224149, llvm-dsymutil was already here.	2014-12-12 21:25:07 +00:00
lit.cfg	llvm/test/lit.cfg: have_ld_plugin_support(): Use decode() for stdout.	2015-01-05 14:18:04 +00:00
lit.site.cfg.in	[lit] Make config.llvm_lib_dir available on cmake, too.	2014-12-30 03:24:11 +00:00
Makefile	[lit] Make config.llvm_lib_dir available on cmake, too.	2014-12-30 03:24:11 +00:00
Makefile.tests
TestRunner.sh