RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-29 23:12:55 +01:00
Code Issues Projects Releases Wiki Activity
87,183 Commits 4 Branches 5 Tags 704 MiB
fc6e3e3272
Commit Graph

26 Commits

Author SHA1 Message Date
Nadav Rotem
fc6e3e3272 X86: Prefer using VPSHUFD over VPERMIL because it has better throughput. 
llvm-svn: 169624
 2012-12-07 19:01:13 +00:00
Manman Ren
e98ec5dd77 X86: when constructing VZEXT_LOAD from other loads, makes sure its output 
chain is correctly setup.

As an example, if the original load must happen before later stores, we need
to make sure the constructed VZEXT_LOAD is constrained to be before the stores.

rdar://12684358

llvm-svn: 167859
 2012-11-13 19:13:05 +00:00
Elena Demikhovsky
61924c155d This patch optimizes shuffle instruction - generates 2 instructions instead of 4. 
Since this specific shuffle is widely used in many workloads we have ~10% performance on them.

shufflevector <8 x float> %A, <8 x float> %B, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>

vmovaps (%rdx), %ymm0
vshufps $8, %ymm0, %ymm0, %ymm0
vmovaps (%rcx), %ymm1
vshufps $8, %ymm0, %ymm1, %ymm1
vunpcklps       %ymm0, %ymm1, %ymm0

vmovaps (%rcx), %ymm0
vmovsldup       (%rdx), %ymm1
vblendps        $85, %ymm0, %ymm1, %ymm0

llvm-svn: 163134
 2012-09-04 12:49:02 +00:00
Elena Demikhovsky
832f074a32 Shuffle optimization for AVX/AVX2. 
The current patch optimizes frequently used shuffle patterns and gives these instruction sequence reduction.
Before:
      vshufps $-35, %xmm1, %xmm0, %xmm2 ## xmm2 = xmm0[1,3],xmm1[1,3]
       vpermilps       $-40, %xmm2, %xmm2 ## xmm2 = xmm2[0,2,1,3]
       vextractf128    $1, %ymm1, %xmm1
       vextractf128    $1, %ymm0, %xmm0
       vshufps $-35, %xmm1, %xmm0, %xmm0 ## xmm0 = xmm0[1,3],xmm1[1,3]
       vpermilps       $-40, %xmm0, %xmm0 ## xmm0 = xmm0[0,2,1,3]
       vinsertf128     $1, %xmm0, %ymm2, %ymm0
After:
      vshufps $13, %ymm0, %ymm1, %ymm1 ## ymm1 = ymm1[1,3],ymm0[0,0],ymm1[5,7],ymm0[4,4]
      vshufps $13, %ymm0, %ymm0, %ymm0 ## ymm0 = ymm0[1,3,0,0,5,7,4,4]
      vunpcklps       %ymm1, %ymm0, %ymm0 ## ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]

llvm-svn: 159188
 2012-06-26 08:04:10 +00:00
Jakob Stoklund Olesen
9333a7fb3b Run ProcessImplicitDefs on SSA form where it can be much simpler. 
Implicitly defined virtual registers can simply have the <undef> bit set
on all uses, and copies can be turned into implicit defs recursively.

Physical registers are a bit trickier. We handle the common case where a
physreg def is used by a nearby instruction in the same basic block. For
more complicated cases, just leave the IMPLICIT_DEF instruction in.

llvm-svn: 159149
 2012-06-25 18:12:18 +00:00
Craig Topper
d63e429d68 Don't insert 128-bit UNDEF into 256-bit vectors. Just keep the 256-bit vector. Original patch by Elena Demikhovsky. Tweaked by me to allow possibility of covering more cases. 
llvm-svn: 158792
 2012-06-20 05:39:26 +00:00
Craig Topper
57d54b1831 Allow 256-bit shuffles to still be split even if only half of the shuffle comes from two 128-bit pieces. 
llvm-svn: 157175
 2012-05-21 06:40:16 +00:00
Craig Topper
6881f1067c Allow v16i16 and v32i8 shuffles to be rewritten as narrower shuffles. 
llvm-svn: 156156
 2012-05-04 04:44:49 +00:00
Nadav Rotem
c922b4f2a3 Reapply 154396 after fixing a test. 
Original message:
Modify the code that lowers shuffles to blends from using blendvXX to vblendXX.
blendV uses a register for the selection while Vblend uses an immediate.
On sandybridge they still have the same latency and execute on the same execution ports.

llvm-svn: 154483
 2012-04-11 06:40:27 +00:00
Eric Christopher
f8886e8f48 Temporarily revert this patch to see if it brings the buildbots back. 
llvm-svn: 154425
 2012-04-10 19:33:16 +00:00
Nadav Rotem
74f87a6bd8 Modify the code that lowers shuffles to blends from using blendvXX to vblendXX. 
blendv uses a register for the selection while vblend uses an immediate.
On sandybridge they still have the same latency and execute on the same execution ports.

llvm-svn: 154396
 2012-04-10 14:33:13 +00:00
Nadav Rotem
9f7f17826e Lower some x86 shuffle sequences to the vblend family of instructions. 
llvm-svn: 154313
 2012-04-09 08:33:21 +00:00
Craig Topper
40ac46c3d7 Test case for PR12413 
llvm-svn: 154172
 2012-04-06 14:38:25 +00:00
Chad Rosier
a10cf5e1b9 Fix a regression from r147481. 
Original commit message from r147481:
DAGCombine for transforming 128->256 casts into a vmovaps, rather
then a vxorps + vinsertf128 pair if the original vector came from a load.

Fix:
Unaligned loads need to generate a vmovups.
rdar://10974078

llvm-svn: 152366
 2012-03-09 02:00:48 +00:00
Craig Topper
a8a69356e1 Add instruction selection for 256-bit VPSHUFD and 128-bit VPERMILPS/VPERMILPD. 
llvm-svn: 149968
 2012-02-07 06:28:42 +00:00
Nadav Rotem
5c5681cf27 The type-legalizer often scalarizes code. One of the common patterns is extract-and-truncate. 
In this patch we optimize this pattern and convert the sequence into extract op of a narrow type.
This allows the BUILD_VECTOR dag optimizations to construct efficient shuffle operations in many cases.

llvm-svn: 149692
 2012-02-03 13:18:25 +00:00
Craig Topper
9a8c6c1633 Fix pattern for memory form of PSHUFD for use with FP vectors to remove bitcast to an integer vector that normal code wouldn't have. Also remove bitcasts from code that turns splat vector loads into a shuffle as it was making the broken pattern necessary. 
llvm-svn: 149232
 2012-01-30 07:50:31 +00:00
Nadav Rotem
c23e698b5c Transform: (EXTRACT_VECTOR_ELT( VECTOR_SHUFFLE )) -> EXTRACT_VECTOR_ELT. 
llvm-svn: 148337
 2012-01-17 21:44:01 +00:00
Nadav Rotem
c460c870b2 Fix 11769. 
In CanXFormVExtractWithShuffleIntoLoad we assumed that EXTRACT_VECTOR_ELT can be later handled by the DAGCombiner.
However, in some cases on AVX, the EXTRACT_VECTOR_ELT is legalized to EXTRACT_SUBVECTOR + EXTRACT_VECTOR_ELT, which
currently is not handled by the DAGCombiner. In this patch I added a check that we only extract from the XMM part.

llvm-svn: 148298
 2012-01-17 09:13:19 +00:00
Chad Rosier
b4f2ccc1e0 Cleanup test case by adding checks for test names. 
llvm-svn: 148166
 2012-01-14 01:46:51 +00:00
Elena Demikhovsky
beb66de0f9 Fixed a bug in LowerVECTOR_SHUFFLE caused assertion failure 
lc: X86ISelLowering.cpp:6480: llvm::SDValue llvm::X86TargetLowering::LowerVECTOR_SHUFFLE(llvm::SDValue, llvm::SelectionDAG&) const: Assertion `V1.getOpcode() != ISD::UNDEF&&  "Op 1 of shuffle should not be undef"' failed.
Added a test.

llvm-svn: 148044
 2012-01-12 20:33:10 +00:00
Chad Rosier
afcaa8f38a Enhance DAGCombine for transforming 128->256 casts into a vmovaps, rather 
then a vxorps + vinsertf128 pair if the original vector came from a load.
rdar://10594409

llvm-svn: 147481
 2012-01-03 21:05:52 +00:00
Elena Demikhovsky
9b4613ff14 Fixed a bug in LowerVECTOR_SHUFFLE and LowerBUILD_VECTOR. 
Matching MOVLP mask for AVX (265-bit vectors) was wrong.
The failure was detected by conformance tests.

llvm-svn: 147308
 2011-12-28 08:14:01 +00:00
Evan Cheng
77f0fb0296 Update test to something more sensible. 
llvm-svn: 146282
 2011-12-09 21:54:10 +00:00
Evan Cheng
ad8debd736 Add 256-bit variant vmovss and vmovsd patterns. rdar://10538417 
llvm-svn: 146196
 2011-12-08 22:30:45 +00:00
Eli Friedman
7188ba35cb Make sure the X86 backend doesn't explode on 128-bit shuffles in AVX mode. Fixes PR11102. 
llvm-svn: 141585
 2011-10-10 22:28:47 +00:00