Jim Grosbach 49af380e3b x86: Move bitcasts outside concat_vector. ... Consider the following: typedef unsigned short ushort4U __attribute__((ext_vector_type(4), aligned(2))); typedef unsigned short ushort4 __attribute__((ext_vector_type(4))); typedef unsigned short ushort8 __attribute__((ext_vector_type(8))); typedef int int4 __attribute__((ext_vector_type(4))); int4 __bbase_cvt_int(ushort4 v) { ushort8 a; a.lo = v; return _mm_cvtepu16_epi32(a); } This generates the, not unreasonable, IR: define <4 x i32> @foo0(double %v.coerce) nounwind ssp { %tmp = bitcast double %v.coerce to <4 x i16> %tmp1 = shufflevector <4 x i16> %tmp, <4 x i16> undef, <8 x i32> <i32 %0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef> %tmp2 = tail call <4 x i32> @llvm.x86.sse41.pmovzxwd(<8 x i16> %tmp1) ret <4 x i32> %tmp2 } The problem is when type legalization gets hold of the v4i16. It legalizes that by spilling to the stack, then doing a zero-extending load. Things go even more silly from there, ending up with something like: _foo0: movsd %xmm0, -8(%rsp) <== Spill to the stack. movq -8(%rsp), %xmm0 <== Reload it right back out. pmovzxwd %xmm0, %xmm1 <== Here's what we actually asked for. pblendw $1, %xmm1, %xmm0 <== We don't need this at all pmovzxwd %xmm0, %xmm0 <== We already did this ret The v8i8 to v8i16 zext intrinsic gives even worse results, with two table lookups via pshufb instructions(!!). To avoid all that, we can move the bitcasting until after we've formed the wider (legal) vector type. Then our normal codegen flows along nicely and we get the expected: _foo0: pmovzxwd %xmm0, %xmm0 ret rdar://15245794 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@192866 91177308-0d34-0410-b5e6-96231b3b80d8		2013-10-17 02:58:06 +00:00
..
Analysis	Remove the very substantial, largely unmaintained legacy PGO	2013-10-02 15:42:23 +00:00
Assembler	Change objectsize intrinsic to accept different address spaces.	2013-10-07 18:06:48 +00:00
Bindings/Ocaml
Bitcode
BugPoint
CodeGen	x86: Move bitcasts outside concat_vector.	2013-10-17 02:58:06 +00:00
DebugInfo	According to the dwarf standard pubnames and pubtypes for languages	2013-10-17 02:06:06 +00:00
ExecutionEngine	Fixing problems in lli's RemoteMemoryManager.	2013-10-11 22:47:10 +00:00
Feature
FileCheck	Really fix CHECK-LABEL and CHECK-DAG interaction. This actually just restores the initial implementation that was in r186162 but got lost in some subsequent refactoring. More explicit variable names and comments are present now to hopefully prevent repeat regression, as well as another test.	2013-10-11 18:38:36 +00:00
Instrumentation	[asan] Optimize accesses to global arrays with constant index	2013-10-16 14:06:14 +00:00
Integer
JitListener
Linker	Debug Info: In DIBuilder, the derived-from field of a DW_TAG_pointer_type	2013-10-05 01:43:03 +00:00
LTO	llvm/test/LTO should run also on cygwin.	2013-10-09 01:07:31 +00:00
MC	Re-commit r192758 - MC: quote tricky symbol names in asm output	2013-10-17 01:13:02 +00:00
Object	Create an atom with just the data that failed to disassemble.	2013-10-16 19:03:14 +00:00
Other
TableGen
tools	[Mips] Teach llvm-readobj to print MIPS-specific ELF program headers.	2013-10-07 08:58:27 +00:00
Transforms	SLPVectorizer: Don't vectorize volatile memory operations	2013-10-16 17:52:40 +00:00
Unit	[tests] Update to use lit_config and lit package, as appropriate.	2013-08-09 16:22:05 +00:00
Verifier	Add support for metadata representing .ident directives.	2013-10-16 01:49:05 +00:00
YAMLParser
CMakeLists.txt	Fixing lli-child-target build	2013-10-02 22:27:23 +00:00
lit.cfg	Remove the very substantial, largely unmaintained legacy PGO	2013-10-02 15:42:23 +00:00
lit.site.cfg.in
Makefile
Makefile.tests
TestRunner.sh