a6d204ec68
Make F16C feature flag imply AVX rather than just checking both at the patterns.
...
llvm-svn: 190775
2013-09-16 04:29:58 +00:00
8eb45a4ef6
Add the remaining Intel SHA instructions
...
Also assembly/disassembly tests, and for sha256rnds2, aliases with an explicit
xmm0 dependency.
llvm-svn: 190754
2013-09-14 15:03:21 +00:00
3fe264d625
Adds support for Atom Silvermont (SLM) - -march=slm
...
Implements Instruction scheduler latencies for Silvermont,
using latencies from the Intel Silvermont Optimization Guide.
Auto detects SLM.
Turns on post RA scheduler when generating code for SLM.
llvm-svn: 190717
2013-09-13 19:23:28 +00:00
1650175de6
Partial support for Intel SHA Extensions (sha1rnds4)
...
Add basic assembly/disassembly support for the first Intel SHA
instruction 'sha1rnds4'. Also includes feature flag, and test cases.
Support for the remaining instructions will follow in a separate patch.
llvm-svn: 190611
2013-09-12 15:51:31 +00:00
8952974e29
AVX-512: implemented extractelement with variable index.
...
Added parsing of mask register and "zeroing" semantic, like {%k1} {z}.
llvm-svn: 190595
2013-09-12 08:55:00 +00:00
adbb9a121f
Add neverHasSideEffects=1 on a couple move instructions.
...
llvm-svn: 190259
2013-09-08 00:50:45 +00:00
9a5ed9c3bd
AVX-512: added SQRT, VRSQRT14, VCOMISS, VUCOMISS, VRCP14, VPABS
...
llvm-svn: 189472
2013-08-28 11:21:58 +00:00
93eeb47d49
AVX-512: added conversion instructions.
...
llvm-svn: 189349
2013-08-27 13:54:04 +00:00
0a2b6290f1
AVX-512: Added shuffle instructions -
...
VPSHUFD, VPERMILPS, VMOVDDUP, VMOVLHPS, VMOVHLPS, VSHUFPS, VALIGN
single and double forms.
llvm-svn: 189215
2013-08-26 12:45:35 +00:00
540d582594
AVX-512: Added more patterns for VMOVSS, VMOVSD, VMOVD, VMOVQ
...
llvm-svn: 188786
2013-08-20 11:00:29 +00:00
fd2b389263
Move AVX and non-AVX replication inside a couple multiclasses to avoid repeating each instruction for both individually.
...
llvm-svn: 188743
2013-08-20 04:24:14 +00:00
3ce8dbbac2
AVX-512: Added VMOVD, VMOVQ, VMOVSS, VMOVSD instructions.
...
llvm-svn: 188637
2013-08-18 13:08:57 +00:00
5bc180c14f
X86: Turn fp selects into mask operations.
...
double test(double a, double b, double c, double d) { return a<b ? c : d; }
before:
_test:
ucomisd %xmm0, %xmm1
ja LBB0_2
movaps %xmm3, %xmm2
LBB0_2:
movaps %xmm2, %xmm0
after:
_test:
cmpltsd %xmm1, %xmm0
andpd %xmm0, %xmm2
andnpd %xmm3, %xmm0
orpd %xmm2, %xmm0
Small speedup on Benchmarks/SmallPT
llvm-svn: 187706
2013-08-04 12:05:16 +00:00
cd46691728
AVX-512 set: added VEXTRACTPS instruction
...
llvm-svn: 187705
2013-08-04 10:46:07 +00:00
67b05fc0b3
Added INSERT and EXTRACT intructions from AVX-512 ISA.
...
All insertf*/extractf* functions replaced with insert/extract since we have insertf and inserti forms.
Added lowering for INSERT_VECTOR_ELT / EXTRACT_VECTOR_ELT for 512-bit vectors.
Added lowering for EXTRACT/INSERT subvector for 512-bit vectors.
Added a test.
llvm-svn: 187491
2013-07-31 11:35:14 +00:00
efd67d4612
Changed register names (and pointer keywords) to be lower case when using Intel X86 assembler syntax.
...
Patch by Richard Mitton.
llvm-svn: 187476
2013-07-31 02:47:52 +00:00
6030a65039
Remove some errant space charcters in mnemonic strings.
...
llvm-svn: 186932
2013-07-23 06:45:34 +00:00
61da939a17
More Intel syntax alias fixes.
...
llvm-svn: 186814
2013-07-22 09:58:07 +00:00
03db790dc6
Change %xmm0 to XMM0 in Intel side of asm strings for PBLENDVB.
...
llvm-svn: 186812
2013-07-22 09:22:49 +00:00
89703c06f2
Removed PackedDouble domain from scalar instructions. Added more formats for the scalar stuff.
...
llvm-svn: 183626
2013-06-09 07:37:10 +00:00
00b20cc924
[PATCH] Fix VGATHER* operand constraints
...
Add earlyclobber constaints to prevent input register being allocated as
the output register because, according to Intel spec [1], "If any pair
of the index, mask, or destination registers are the same, this
instruction results a UD fault."
---
[1] http://software.intel.com/sites/default/files/319433-014.pdf
llvm-svn: 183327
2013-06-05 18:12:26 +00:00
fad029202f
Removed SSEPacked domain from all forms (AVX, SSE, signed, unsigned) scalar compare instructions, like COMISS, COMISD.
...
No functional changes.
llvm-svn: 182371
2013-05-21 12:04:22 +00:00
9264c95400
Corrected Atom latencies for SSE SQRT instructions.
...
llvm-svn: 181346
2013-05-07 19:57:34 +00:00
817c1d92b4
Put VMOVPQIto64rr in the VRPDI class.
...
Patch by Joshua Magee.
llvm-svn: 180842
2013-05-01 13:00:16 +00:00
aec90531f9
X86: Now that we have a canonical form for vector integer abs, match it into pabs.
...
llvm-svn: 180600
2013-04-26 12:05:21 +00:00
e440d476ee
Annotate the remaining x86 instructions with SchedRW lists.
...
Now all x86 instructions that have itinerary classes also have SchedRW
lists. This is required before the new scheduling models can be used.
There are still unannotated instructions remaining, but they don't have
itinerary classes either.
llvm-svn: 178051
2013-03-26 18:24:22 +00:00
4d39e81fb8
Remove IIC_DEFAULT from X86Schedule.td
...
All the instructions tagged with IIC_DEFAULT had nothing in common, and
we already have a NoItineraries class to represent untagged
instructions.
llvm-svn: 177937
2013-03-25 23:12:41 +00:00
712f674880
Model prefetches and barriers as loads.
...
It's not yet clear if these instructions need a more careful model.
llvm-svn: 177599
2013-03-20 23:09:53 +00:00
5b535c965e
Add a catch-all WriteSystem SchedWrite type.
...
This is used for all the expensive system instructions.
llvm-svn: 177598
2013-03-20 23:09:50 +00:00
cd4ebb7639
Annotate the remaining SSE MOV instructions.
...
llvm-svn: 177592
2013-03-20 22:37:16 +00:00
c6dc70d865
Annotate SSE horizontal and integer instructions.
...
llvm-svn: 177591
2013-03-20 22:37:13 +00:00
7a8bb72a3a
Add some missing SSE annotations.
...
llvm-svn: 177540
2013-03-20 16:56:39 +00:00
3a546156c7
Annotate various null idioms with SchedRW lists.
...
llvm-svn: 177461
2013-03-19 23:23:31 +00:00
24aac1dc92
Annotate SSE float conversions with SchedRW lists.
...
llvm-svn: 177460
2013-03-19 23:23:29 +00:00
a5158c8f0a
Add SchedRW annotations to most of X86InstrSSE.td.
...
We hitch a ride with the existing OpndItins class that was used to add
instruction itinerary classes in the many multiclasses in this file.
Use the link provided by the X86FoldableSchedWrite.Folded to find the
right SchedWrite for folded loads.
llvm-svn: 177326
2013-03-18 22:01:35 +00:00
adfa5eaf8c
Unaligned loads should use the VMOVUPS opcode.
...
llvm-svn: 177130
2013-03-14 23:49:44 +00:00
8fb09f0abb
Fix inconsistent usage of PALIGN and PALIGNR when referring to the same instruction.
...
llvm-svn: 173667
2013-01-28 06:48:25 +00:00
c7e6feee42
Combine AVX and SSE forms of MOVSS and MOVSD into the same multiclasses so they get instantiated together.
...
llvm-svn: 172704
2013-01-17 06:59:42 +00:00
0d2c29e807
Simplify nested strconcats in X86 td files since strconcat can take more than 2 arguments.
...
llvm-svn: 172379
2013-01-14 07:46:34 +00:00
4c69a05d2d
Create a single multiclass for SSE and AVX version of MOVL/MOVH. Prevents needing to specify everything twice. No functional change intended
...
llvm-svn: 172378
2013-01-14 07:26:58 +00:00
bcd14a0f26
X86: Add patterns for X86ISD::VSEXT in registers.
...
Those can occur when something between the sextload and the store is on the same
chain and blocks isel. Fixes PR14887.
llvm-svn: 172353
2013-01-13 11:37:04 +00:00
bd62d64cbf
Remove unnecessary # tokens at the beginning and end of defm names.
...
llvm-svn: 171694
2013-01-07 05:04:39 +00:00
4f1c7256f9
Fix suffix handling for parsing and printing of cvtsi2ss, cvtsi2sd, cvtss2si, cvttss2si, cvtsd2si, and cvttsd2si to match gas behavior.
...
cvtsi2* should parse with an 'l' or 'q' suffix or no suffix at all. No suffix should be treated the same as 'l' suffix. Printing should always print a suffix. Previously we didn't parse or print an 'l' suffix.
cvtt*2si/cvt*2si should parse with an 'l' or 'q' suffix or not suffix at all. No suffix should use the destination register size to choose encoding. Printing should not print a suffix.
Original 'l' suffix issue with cvtsi2* pointed out by Michael Kuperstein.
llvm-svn: 171668
2013-01-06 20:39:29 +00:00
9791afb182
Merge SSE and AVX instruction definitions for scalar forms of SQRT, RSQRT, and RCP.
...
llvm-svn: 171356
2013-01-02 08:00:39 +00:00
4bc5c4e152
Merge SSE and AVX instruction definitions for PSHUFD/PSHUFHW/PSHUFLW.
...
llvm-svn: 171355
2013-01-02 07:27:49 +00:00
db1a84c84a
Revert 171351. It broke MC/X86/x86-32-avx.s.
...
llvm-svn: 171352
2013-01-02 01:35:11 +00:00
86d0cdb82f
Merge SSE and AVX instruction definitions for scalar forms of SQRT, RSQRT, and RCP.
...
llvm-svn: 171351
2013-01-01 20:53:20 +00:00
12ed9cd6ae
Remove unused argument from a multiclass.
...
llvm-svn: 171340
2013-01-01 03:42:44 +00:00
2edafc059d
Merge intrinsic instruction definitions for SSE and AVX versions of RCPPS and RSQRTPS.
...
llvm-svn: 171339
2013-01-01 03:30:21 +00:00
d04dbec6c9
Remove 2 unused multiclasses.
...
llvm-svn: 171338
2013-01-01 02:02:45 +00:00
7cc4f322cf
Merge AVX/SSE instruction definitions for SQRTPS/PD, RSQRTPS, RCPPS. No funcitonal change intended.
...
llvm-svn: 171337
2013-01-01 00:11:07 +00:00
c2521cd309
Use packed instead of scalar itineraries for SSE1/2 SQRTPS/PD, RCPPS, and RSQRTPS. VEX-encoded forms already use packed.
...
llvm-svn: 171336
2012-12-31 23:49:05 +00:00
fe82eb6bcd
Remove intrinsic specific instructions for (V)SQRTPS/PD. Instead lower to target-independent ISD nodes and use the existing patterns for those.
...
llvm-svn: 171237
2012-12-29 18:18:20 +00:00
6b27251a76
Remove intrinsic specific instructions for SSE/SSE2/AVX floating point max/min instructions. Lower them to target specific nodes and use those patterns instead. This also allows them to be commuted if UnsafeFPMath is enabled.
...
llvm-svn: 171227
2012-12-29 16:44:25 +00:00
ab2e6842cc
Merge basic_sse12_fp_binop_p_int and basic_sse12_fp_binop_p_y_int multiclasses.
...
llvm-svn: 171171
2012-12-27 22:53:47 +00:00
e2eec3c52b
Merge basic_sse12_fp_binop_p and basic_sse12_fp_binop_p_y multiclasses.
...
llvm-svn: 171166
2012-12-27 18:51:50 +00:00
757f3fc394
Add hasSideEffects=0 to some forms of ROUND, RCP, and RSQRT.
...
llvm-svn: 171143
2012-12-27 07:16:08 +00:00
09ce4b9efe
Move single letter 'P' prefix out of multiclass now that tablegen allows defm to start with #NAME. This makes instruction names more searchable again.
...
llvm-svn: 171141
2012-12-27 06:34:54 +00:00
1b8c0750ee
Mark all the _REV instructions as not having side effects. They aren't really emitted by the backend, but it reduces the number of instructions in the output files with unmodelled side effects to make auditing easier.
...
llvm-svn: 171118
2012-12-26 21:30:22 +00:00
18f2675e9b
Remove a special conditional setting of neverHasSideEffects if the instruction didn't have a pattern. This was leftover from when tablegen used to complain if things were already inferred from patterns.
...
llvm-svn: 171117
2012-12-26 21:04:30 +00:00
24f316e4db
Merge still more SSE/AVX instruction definitions.
...
llvm-svn: 171103
2012-12-26 07:54:43 +00:00
af629e2700
Merge more SSE/AVX instruction definitions.
...
llvm-svn: 171102
2012-12-26 07:20:35 +00:00
65fe30450d
Fix 80 column violation.
...
llvm-svn: 171097
2012-12-26 06:15:53 +00:00
f4d0fe8fcd
Fix class name in comment.
...
llvm-svn: 171096
2012-12-26 06:15:09 +00:00
59747c4dbd
Merge SSE/AVX PCMPEQ/PCMPGT instruction definitions.
...
llvm-svn: 171095
2012-12-26 06:14:15 +00:00
8a48677586
Remove 'v' from mnemonic to fix asm matching failures.
...
llvm-svn: 171093
2012-12-26 06:02:15 +00:00
b4ef0fa3a1
Use an additional multiclass to merge the 128/256-bit SSE/AVX instruction definitions for a bunch of SSE2 integer arithmetic instructions.
...
llvm-svn: 171092
2012-12-26 05:49:15 +00:00
a2594dd5f0
Use an additional multiclass to merge the 128/256-bit SSE/AVX instruction definitions for PAND/POR/PXOR/PANDN
...
llvm-svn: 171087
2012-12-26 04:36:03 +00:00
97730a0d6a
Merge an AVX/SSE 256-bit and 128-bit multiclass.
...
llvm-svn: 171086
2012-12-26 03:56:47 +00:00
8b59746390
Mark VANDNPD/VANDNPDS as not commutable.
...
llvm-svn: 171085
2012-12-26 03:48:10 +00:00
4669d18893
X86: Match the SSE/AVX min/max vector ops using a custom node instead of intrinsics
...
This is very mechanical, no functionality change. Preparation for PR14667.
llvm-svn: 170898
2012-12-21 14:04:55 +00:00
14a4af0e66
Optimized load + SIGN_EXTEND patterns in the X86 backend.
...
llvm-svn: 170506
2012-12-19 07:50:20 +00:00
b16ccde7a4
X86: Add a couple of target-specific dag combines that turn VSELECTS into psubus if possible.
...
We match the pattern "x >= y ? x-y : 0" into "subus x, y" and two special cases
if y is a constant. DAGCombiner canonicalizes those so we first have to undo the
canonicalization for those cases. The pattern occurs in gzip when the loop
vectorizer is enabled. Part of PR14613.
llvm-svn: 170273
2012-12-15 16:47:44 +00:00
216bcd522b
Remove intrinsic specific instructions for (V)MOVQUmr with patterns pointing to the normal instructions.
...
llvm-svn: 169482
2012-12-06 07:31:16 +00:00
922f10aec4
Mark MOVDQ(A/U)rm as ReMaterializable. Mark all MOVDQ(A/U) instructions as neverHasSideEffects.
...
llvm-svn: 169477
2012-12-06 06:49:16 +00:00
cd3c1c4a16
Simplified BLEND pattern matching for shuffles.
...
Generate VPBLENDD for AVX2 and VPBLENDW for v16i16 type on AVX2.
llvm-svn: 169366
2012-12-05 09:24:57 +00:00
70601ba6f9
Use roundps/pd for llvm.ceil, llvm.trunc, llvm.rint, and llvm.nearbyint of vector types.
...
llvm-svn: 168141
2012-11-16 06:37:56 +00:00
9268c94b15
Cleanup pcmp(e/i)str(m/i) instruction definitions and load folding support.
...
llvm-svn: 167652
2012-11-10 01:23:36 +00:00
ec47090b1e
Remove tailing whitespaces
...
llvm-svn: 167445
2012-11-06 08:06:35 +00:00
6b223a4f06
X86 SSE: update rsqrtss and rcpss to use two source operands and
...
the first source operand is tied to the destination operand.
This is to accurately model the corresponding instructions where the upper
bits are unmodified.
rdar://12558838
PR14221
llvm-svn: 167064
2012-10-30 23:53:59 +00:00
ad0b69fe3e
Fix PR14204
...
- Add missing pattern on X86ISD::VZEXT from VR256 to VR256 when AVX2 is enabled.
llvm-svn: 166947
2012-10-29 17:57:12 +00:00
c5af149e70
Add custom conversion from v2u32 to v2f32 in 32-bit mode
...
- As there's no 64-bit GPRs in 32-bit mode, a custom conversion from v2u32 to
v2f32 is added to improve the efficiency of the code generated.
llvm-svn: 166545
2012-10-24 04:09:32 +00:00
1be96bb5ce
Enable lowering ZERO_EXTEND/ANY_EXTEND to PMOVZX from SSE4.1
...
llvm-svn: 166486
2012-10-23 17:34:00 +00:00
e999b865dd
Add support for FP_ROUND from v2f64 to v2f32
...
- Due to the current matching vector elements constraints in
ISD::FP_ROUND, rounding from v2f64 to v4f32 (after legalization from
v2f32) is scalarized. Add a customized v2f32 widening to convert it
into a target-specific X86ISD::VFPROUND to work around this
constraints.
llvm-svn: 165631
2012-10-10 16:53:28 +00:00
3f23c1a8b9
Remove code for setting the VEX L-bit as a function of operand size from the code emitters and the disassembler table builder. Fix a couple instructions that were still missing VEX_L.
...
llvm-svn: 164204
2012-09-19 06:37:45 +00:00
a73be890a1
Add explicit VEX_L tags to all 256-bit instructions. This will allow us to remove code from the code emitters that examined operands to set the L-bit.
...
llvm-svn: 164202
2012-09-19 06:06:34 +00:00
37521aa89c
The PMOVZXWD family of functions had patterns extends narrow vector types to wide vector types.
...
It had patterns for zext-loading and extending. This commit adds patterns for loading a wide type, performing a bitcast,
and extending. This is an odd pattern, but it is commonly used when writing code with intrinsics.
rdar://11897677
llvm-svn: 163995
2012-09-16 07:39:07 +00:00
400f7ef871
Enhance PR11334 fix to support extload from v2f32/v4f32
...
- Fix an remaining issue of PR11674 as well
llvm-svn: 163528
2012-09-10 18:33:51 +00:00
4ed79bd7d7
Add instruction selection for ffloor of vectors when SSE4.1 or AVX is enabled.
...
llvm-svn: 163473
2012-09-08 17:42:27 +00:00
f3e4aa8cdd
Use iPTR instead of i32 for extract_subvector/insert_subvector index in lowering and patterns. This makes it consistent with the incoming DAG nodes from the DAG builder.
...
llvm-svn: 163293
2012-09-06 06:09:01 +00:00
daa5ed1e0a
Add patterns for converting stores of subvector_extracts of lower 128-bits of a 256-bit vector to VMOVAPSmr/VMOVUPSmr.
...
llvm-svn: 163292
2012-09-06 05:15:01 +00:00
81f06df699
Remove some of the patterns added in r163196. Increasing the complexity on insert_subvector into undef accomplishes the same thing.
...
llvm-svn: 163198
2012-09-05 07:26:35 +00:00
f7c87d6eea
Add patterns for integer forms of VINSERTF128/VINSERTI128 folded with loads. Also add patterns to turn subvector inserts with loads to index 0 of an undef into VMOVAPS.
...
llvm-svn: 163196
2012-09-05 06:58:39 +00:00
2db2353b21
Convert vextracti128/vextractf128 intrinsics to extract_subvector at DAG build time. Similar was previously done for vinserti128/vinsertf128. Add patterns for folding these extract_subvectors with stores.
...
llvm-svn: 163192
2012-09-05 05:48:09 +00:00
d6cc4062be
Typos
...
llvm-svn: 163053
2012-09-01 06:33:50 +00:00
969f3913dd
Clean up AddedComplexity further after adding UseSSEx
...
llvm-svn: 162973
2012-08-31 03:01:35 +00:00
e423e865fe
X86: Fix encoding of 'movd %xmm0, %rax'
...
The assembly string for the VMOVPQIto64rr instruction incorrectly lacked the 'v'
prefix, resulting in mis-assembly of the vanilla movd instruction.
llvm-svn: 162963
2012-08-31 00:30:30 +00:00
bbd10792c2
Introduce 'UseSSEx' to force SSE legacy encoding
...
- Add 'UseSSEx' to force SSE legacy insn not being selected when AVX is
enabled.
As the penalty of inter-mixing SSE and AVX instructions, we need
prevent SSE legacy insn from being generated except explicitly
specified through some intrinsics. For patterns supported by both
SSE and AVX, so far, we force AVX insn will be tried first relying on
AddedComplexity or position in td file. It's error-prone and
introduces bugs accidentally.
'UseSSEx' is disabled when AVX is turned on. For SSE insns inherited
by AVX, we need this predicate to force VEX encoding or SSE legacy
encoding only.
For insns not inherited by AVX, we still use the previous predicates,
i.e. 'HasSSEx'. So far, these insns fall into the following
categories:
* SSE insns with MMX operands
* SSE insns with GPR/MEM operands only (xFENCE, PREFETCH, CLFLUSH,
CRC, and etc.)
* SSE4A insns.
* MMX insns.
* x87 insns added by SSE.
2 test cases are modified:
- test/CodeGen/X86/fast-isel-x86-64.ll
AVX code generation is different from SSE one. 'vcvtsi2sdq' cannot be
selected by fast-isel due to complicated pattern and fast-isel
fallback to materialize it from constant pool.
- test/CodeGen/X86/widen_load-1.ll
AVX code generation is different from SSE one after fixing SSE/AVX
inter-mixing. Exec-domain fixing prefers 'vmovapd' instead of
'vmovaps'.
llvm-svn: 162919
2012-08-30 16:54:46 +00:00
cc56718038
The commutative flag is already correctly set within the multiclass. If we set
...
it here, then a 'register-memory' version would wrongly get the commutative
flag.
<rdar://problem/12180135>
llvm-svn: 162741
2012-08-28 07:36:46 +00:00
72f51c3986
Convert V_SETALLONES/AVX_SETALLONES/AVX2_SETALLONES to Post-RA pseudos.
...
llvm-svn: 162740
2012-08-28 07:30:47 +00:00
bd509eea4a
Merge AVX_SET0PSY/AVX_SET0PDY/AVX2_SET0 into a single post-RA pseudo.
...
llvm-svn: 162738
2012-08-28 07:05:28 +00:00
89d6b29d16
More missing mayLoad flags on AVX multiclasses.
...
llvm-svn: 162714
2012-08-28 00:02:01 +00:00
5af2fed5f2
Don't allow vextractf128 to be folded with unaligned stores. We don't fold unaligned loads so shouldn't fold unaligned stores as it can cause an alignment fault to occur.
...
llvm-svn: 162658
2012-08-27 07:19:59 +00:00
6d44554cd4
Fold some patterns into instruction definitons so tablegen can infer flags removing the need for an explicit 'neverHasSideEffects = 1'
...
llvm-svn: 162656
2012-08-27 07:04:50 +00:00
f7828f91ee
Add HasAVX1Only predicate and use it for patterns that have an AVX1 instruction and an AVX2 instruction rather than relying on AddedComplexity.
...
llvm-svn: 162654
2012-08-27 06:08:57 +00:00
3d91b43ad2
Add missing mayLoad flags to a large class of AVX *_Int instructions.
...
llvm-svn: 162622
2012-08-24 23:29:07 +00:00
d3511235d1
Remove some spurious mayLoad = 0 flags.
...
They were inserted to silence TableGen's warning about
redundant properties. That warning is now gone.
llvm-svn: 162517
2012-08-24 00:31:20 +00:00
178250ad87
When unsafe math is used, we can use commutative FMAX and FMIN. In some cases
...
this allows for better code generation.
Added a new DAGCombine transformation to convert FMAX and FMIN to FMANC and
FMINC, which are commutative.
For example:
movaps %xmm0, %xmm1
movsd LC(%rip), %xmm0
minsd %xmm1, %xmm0
becomes:
minsd LC(%rip), %xmm0
llvm-svn: 162187
2012-08-19 13:06:16 +00:00
34107b9177
fix PR11334
...
- FP_EXTEND only support extending from vectors with matching elements.
This results in the scalarization of extending to v2f64 from v2f32,
which will be legalized to v4f32 not matching with v2f64.
- add X86-specific VFPEXT supproting extending from v4f32 to v2f64.
- add BUILD_VECTOR lowering helper to recover back the original
extending from v4f32 to v2f64.
- test case is enhanced to include different vector width.
llvm-svn: 161894
2012-08-14 21:24:47 +00:00
ab47fe4e16
Implement proper handling for pcmpistri/pcmpestri intrinsics. Requires custom handling in DAGISelToDAG due to limitations in TableGen's implicit def handling. Fixes PR11305.
...
llvm-svn: 161318
2012-08-06 06:22:36 +00:00
6d0408d3a5
Remove custom inserter for MWAIT. It doesn't do anything that couldn't be represented in a pattern.
...
llvm-svn: 161306
2012-08-05 00:36:57 +00:00
4059145396
X86: mark GATHER instructios as mayLoad
...
llvm-svn: 161143
2012-08-01 23:28:59 +00:00
14eac5dda8
Give VCVTTPD2DQ priority over CVTTPD2DQ.
...
llvm-svn: 160942
2012-07-30 02:20:32 +00:00
f881d385da
Fix patterns for CVTTPS2DQ to specify SSE2 instead of SSE1.
...
llvm-svn: 160941
2012-07-30 02:14:02 +00:00
415b3586d0
Fix up patterns for VCVTSS2SD. Specifically give it priority over SSE form. Add an OptForSpeed to explicitly pair up with an OptForSize that was already on another pattern.
...
llvm-svn: 160939
2012-07-30 01:38:57 +00:00
28402efcb6
Fix load types on intrinsic forms of SS2SD and SD2SS AVX/SSE convert instruction patterns.
...
llvm-svn: 160938
2012-07-29 23:26:34 +00:00
b6767f3acd
Move more SSE/AVX convert instruction patterns into their definitions.
...
llvm-svn: 160937
2012-07-29 22:30:06 +00:00
fc93281c07
Fold patterns for some of the SSE/AVX convert instructions into their instruction definitions.
...
llvm-svn: 160922
2012-07-28 18:59:19 +00:00
024797b9a2
Mark some of the SSE/AVX convert instructions as mayLoad/neverHasSideEffects.
...
llvm-svn: 160921
2012-07-28 18:36:39 +00:00
44f9b5343d
Make CVTSS2SI instruction definition consistent with CVTSD2SI.
...
llvm-svn: 160914
2012-07-28 08:28:23 +00:00
1c1aef07b8
Fix up memory load types for SSE scalar convert intrinsic patterns.
...
llvm-svn: 160913
2012-07-28 07:59:59 +00:00
77cd55b4ee
Remove the last mentions of sub_ss and sub_sd from patterns.
...
I'll remove these two sub-register indexes shortly.
llvm-svn: 160831
2012-07-26 23:03:08 +00:00
b96d0b4e08
Eliminate sub_ss, sub_sd from broadcast patterns.
...
The (COPY_TO_REGCLASS GR32:$src, VR128) pattern looks odd, but
copyPhysReg does the right thing with it. (The old pattern would
eventually produce the same cross-class copy).
llvm-svn: 160830
2012-07-26 22:59:06 +00:00
206b825f5c
Eliminate more sub_ss / sub_sd patterns.
...
This gets rid of some more INSERT_SUBREG - IMPLICIT_DEF patterns,
simplifying the emitted code a bit.
llvm-svn: 160820
2012-07-26 22:30:18 +00:00
75d17b0577
Eliminate some SUBREG_TO_REG patterns with sub_ss and sub_sd.
...
The SUBREG_TO_REG instruction has magic semantics asserting that the
source value was defined by an instruction that cleared the high half of
the register. Those semantics are never actually exploited for xmm
registers.
llvm-svn: 160818
2012-07-26 22:03:21 +00:00
ceee4a9d0c
Eliminate a batch of uses of sub_ss and sub_sd in the X86 target.
...
These idempotent sub-register indices don't do anything --- They simply
map XMM registers to themselves. They no longer affect register classes
either since the SubRegClasses field has been removed from Target.td.
This patch replaces XMM->XMM EXTRACT_SUBREG and INSERT_SUBREG patterns
with COPY_TO_REGCLASS patterns which simply become COPY instructions.
The number of IMPLICIT_DEF instructions before register allocation is
reduced, and that is the cause of the test case changes.
llvm-svn: 160816
2012-07-26 21:40:42 +00:00
c7690ac7ac
Make l/q suffixes on AVX forms of scalar convert instructions consistent with their non-AVX forms.
...
llvm-svn: 160775
2012-07-26 07:48:28 +00:00
4c12245b3a
The vbroadcast family of instructions has 'fallback patterns' in case where the
...
load source operand is used by multiple nodes. The v2i64 broadcast was emulated
by shuffling the two lower i32 elements to the upper two.
We had a bug in the immediate used for the broadcast.
Replacing 0 to 0x44.
0x44 means [01|00|01|00] which corresponds to the correct lane.
Patch by Michael Kuperstein.
llvm-svn: 160430
2012-07-18 08:14:48 +00:00
01deb5f2df
Make x86 asm parser to check for xmm vs ymm for index register in gather instructions. Also fix Intel syntax for gather instructions to use 'DWORD PTR' or 'QWORD PTR' to match gas.
...
llvm-svn: 160420
2012-07-18 04:11:12 +00:00
ee3552f88d
Rename VBROADCASTSDrm into VBROADCASTSDYrm to match the naming convention.
...
Allow the folding of vbroadcastRR to vbroadcastRM, where the memory operand is a spill slot.
PR12782.
Together with Michael Kuperstein <michael.m.kuperstein@intel.com>
llvm-svn: 160230
2012-07-15 12:26:30 +00:00
b3bac4908e
Mark VINSERTI128rm as MayLoad=1. Fixes PR13348.
...
llvm-svn: 160162
2012-07-13 05:46:28 +00:00
f7755df776
Update GATHER instructions to support 2 read-write operands. Patch from myself and Manman Ren.
...
llvm-svn: 160110
2012-07-12 06:52:41 +00:00
be41e2daa6
Reverse assembler/disassembler operand order for gather instructions.
...
llvm-svn: 159983
2012-07-10 06:38:33 +00:00
85c938f44f
Remove extra space.
...
llvm-svn: 159647
2012-07-03 06:48:58 +00:00
f067f9aa51
Change i128mem/i256mem to f128mem/f256mem on some floating point vector instructions.
...
llvm-svn: 159646
2012-07-03 06:11:06 +00:00
676dcd8c39
Add aliases for pblendvb, blendvpd, and blendvps instructions with the implicit xmm0 operand specified. Fixes PR13252.
...
llvm-svn: 159644
2012-07-03 05:49:45 +00:00
9af899fa88
Optimization of shuffle node that can fit to the register form of VBROADCAST instruction on AVX2.
...
llvm-svn: 159504
2012-07-01 06:12:26 +00:00
98a5bf24a9
X86: add more GATHER intrinsics in LLVM
...
Corrected type for index of llvm.x86.avx2.gather.d.pd.256
from 256-bit to 128-bit.
Corrected types for src|dst|mask of llvm.x86.avx2.gather.q.ps.256
from 256-bit to 128-bit.
Support the following intrinsics:
llvm.x86.avx2.gather.d.q, llvm.x86.avx2.gather.q.q
llvm.x86.avx2.gather.d.q.256, llvm.x86.avx2.gather.q.q.256
llvm.x86.avx2.gather.d.d, llvm.x86.avx2.gather.q.d
llvm.x86.avx2.gather.d.d.256, llvm.x86.avx2.gather.q.d.256
llvm-svn: 159402
2012-06-29 00:54:20 +00:00
a09820414a
X86: add GATHER intrinsics (AVX2) in LLVM
...
Support the following intrinsics:
llvm.x86.avx2.gather.d.pd, llvm.x86.avx2.gather.q.pd
llvm.x86.avx2.gather.d.pd.256, llvm.x86.avx2.gather.q.pd.256
llvm.x86.avx2.gather.d.ps, llvm.x86.avx2.gather.q.ps
llvm.x86.avx2.gather.d.ps.256, llvm.x86.avx2.gather.q.ps.256
Modified Disassembler to handle VSIB addressing mode.
llvm-svn: 159221
2012-06-26 19:47:59 +00:00
94bf0f3855
Remove some duplicate instructions that exist only to given different mnemonics for the assembler. Use InstAlias instead.
...
llvm-svn: 159184
2012-06-26 04:12:49 +00:00
357de815b4
Add SSE2 predicate to CVTPS2PD instructions. Doesn't matter much because there are no patterns in the instruction.
...
llvm-svn: 159127
2012-06-25 06:51:42 +00:00
b6eb513c68
Remove codegen only instruction in favor of one that has the same definition. Make some pattern operands more explicit about types.
...
llvm-svn: 159126
2012-06-25 06:16:00 +00:00
fd5e6e7db1
Remove intrinsic specific instructions for (V)CVTPS2DQ and replace with patterns.
...
llvm-svn: 159109
2012-06-24 07:07:16 +00:00
b925230fb1
Remove intrinsic specific instructions for (V)CVTPS2DQ and replace with patterns.
...
llvm-svn: 159108
2012-06-24 06:55:37 +00:00
f48ec7a708
Fix build failures from r159106.
...
llvm-svn: 159107
2012-06-24 06:08:31 +00:00
bab2b89944
Remove intrinsic specific instructions for CVTPD2PS and replace with just patterns.
...
llvm-svn: 159106
2012-06-24 05:44:31 +00:00
3cee08ce7d
Remove intrinsic specific instructions for CVTPD2DQ. Replace with patterns.
...
llvm-svn: 159105
2012-06-24 05:33:24 +00:00
a899cc15f1
Remove intrinsic specific instructions for (V)CVTDQ2PS. Use a Pat instead instead.
...
llvm-svn: 159090
2012-06-23 22:33:14 +00:00
7e9415220a
Make CVTDQ2PS instruction use SSE2 predicate instead of SSE1. No functional change because there are no patterns in the instructions. Also fix a typo in a comment.
...
llvm-svn: 159087
2012-06-23 20:52:45 +00:00
24e3418215
Move CVTPD2DQ to use SSE2 predicate instead of SSE3. Move DQ2PD and PD2DQ to the SSE2 section of the file.
...
llvm-svn: 159086
2012-06-23 20:15:42 +00:00
8c03ea79c4
Use correct memory types for (V)CVTDQ2PD instructions.
...
llvm-svn: 159075
2012-06-23 08:30:27 +00:00
431f1e7192
Remove intrinsic specific instructions for 128-bit (V)CVTDQ2PD. Replace with intrinsic patterns. Mem forms omitted because the load size is only 64-bits.
...
llvm-svn: 159070
2012-06-23 04:23:36 +00:00
21d04fc118
Add predicate check around some patterns.
...
llvm-svn: 158797
2012-06-20 07:30:23 +00:00
3b662a6279
Add predicate check around some patterns.
...
llvm-svn: 158795
2012-06-20 07:01:11 +00:00
390edb0d91
*no need to pollute Intel syntax with bonus mnemonics; operand size is explicitly specified
...
llvm-svn: 158603
2012-06-16 17:19:49 +00:00
bf2409e8aa
Mark several instructions SSE2 instead of SSE3 as they should be.
...
llvm-svn: 158049
2012-06-06 06:45:27 +00:00
a0396e4583
X86: Rename the CLMUL target feature to PCLMUL.
...
It was renamed in gcc/gas a while ago and causes all kinds of
confusion because it was named differently in llvm and clang.
llvm-svn: 157745
2012-05-31 14:34:17 +00:00
c1ac05dad5
Add intrinsic for pclmulqdq instruction.
...
llvm-svn: 157731
2012-05-31 04:37:40 +00:00
ef479ea854
Add intrinsics, code gen, assembler and disassembler support for the SSE4a extrq and insertq instructions.
...
This required light surgery on the assembler and disassembler
because the instructions use an uncommon encoding. They are
the only two instructions in x86 that use register operands
and two immediates.
llvm-svn: 157634
2012-05-29 19:05:25 +00:00
7daf897678
Remove 256-bit AVX non-temporal store intrinsics. Similar was previously done for 128-bit.
...
llvm-svn: 156375
2012-05-08 06:58:15 +00:00
dbb98b4917
Fix some issues in the f16c instructions.
...
llvm-svn: 156287
2012-05-07 06:00:15 +00:00
d4e1894ec1
Add SSE4A MOVNTSS/MOVNTSD instructions.
...
llvm-svn: 156281
2012-05-07 05:36:19 +00:00
810734b7f4
AVX: Add additional vbroadcast replacement sequences for integers.
...
Remove the v2f64 patterns because it does not match any vbroadcast
instruction.
llvm-svn: 155461
2012-04-24 18:09:59 +00:00
aa3ff8da00
AVX: We lower VECTOR_SHUFFLE and BUILD_VECTOR nodes into vbroadcast instructions
...
using the pattern (vbroadcast (i32load src)). In some cases, after we generate
this pattern new users are added to the load node, which prevent the selection
of the blend pattern. This commit provides fallback patterns which perform
in-vector broadcast (using in-vector vbroadcast in AVX2 and pshufd on AVX1).
llvm-svn: 155437
2012-04-24 11:07:03 +00:00
8d7e56c409
ZERO_EXTEND/SIGN_EXTEND/TRUNCATE optimization for AVX2
...
llvm-svn: 155309
2012-04-22 09:39:03 +00:00
4badeb3f0d
Replace vpermd/vpermps intrinic patterns with custom lowering to target specific nodes.
...
llvm-svn: 154801
2012-04-16 07:13:00 +00:00
c0075aa7ff
Flip the arguments when converting vpermd/vpermps intrinsics into instructions. The intrinsic has the mask as the last operand, but the instruction has it as the second.
...
llvm-svn: 154797
2012-04-16 06:26:15 +00:00
b86fa404d3
Merge vpermps/vpermd and vpermpd/vpermq SD nodes.
...
llvm-svn: 154782
2012-04-16 00:41:45 +00:00
bfc9a5f7d3
Remove AVX2 vpermq and vpermpd intrinsics. These can now be handled with normal shuffle vectors.
...
llvm-svn: 154778
2012-04-15 22:43:31 +00:00
42bcd04ee3
Fix PR12529. The Vxx family of instructions are only supported by AVX.
...
Use non-vex instructions for SSE4.
llvm-svn: 154770
2012-04-15 19:36:44 +00:00
779a72b49e
Added VPERM optimization for AVX2 shuffles
...
llvm-svn: 154761
2012-04-15 11:18:59 +00:00
d0271b27cb
Fix 128-bit ptest intrinsics to take v2i64 instead of v4f32 since these are integer instructions.
...
llvm-svn: 154580
2012-04-12 07:23:00 +00:00
9bc178ac5c
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
65ada95b84
Temporarily revert this patch to see if it brings the buildbots back.
...
llvm-svn: 154425
2012-04-10 19:33:16 +00:00
f934f91709
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
d024cef233
Turn avx2 vinserti128 intrinsic calls into INSERT_SUBVECTOR DAG nodes and remove patterns for selecting the intrinsic. Similar was already done for avx1.
...
llvm-svn: 154272
2012-04-07 22:32:29 +00:00
aa9aab5ad2
Move vinsertf128 patterns near the instruction definitions. Add AddedComplexity to AVX2 vextracti128 patterns to give them priority over the integer versions of vextractf128 patterns.
...
llvm-svn: 154268
2012-04-07 21:57:43 +00:00
7629d63bc4
Add support for AVX enhanced comparison predicates. Patch from Kay Tiong Khoo.
...
llvm-svn: 153935
2012-04-03 05:20:24 +00:00
4106917355
[avx] Add patterns for combining vextractf128 + vmovaps/vmovups/vmobdqu to
...
vextractf128 with 128-bit mem dest.
Combines
vextractf128 $0, %ymm0, %xmm0
vmovaps %xmm0, (%rdi)
to
vextractf128 $0, %ymm0, (%rdi)
rdar://11082570
llvm-svn: 153139
2012-03-20 21:43:40 +00:00
0158ae2e5b
[avx] Add the AddedComplexity to the VINSERTI128 avx2 patterns to give
...
precedence over the VINSERTF128 avx1 patterns.
llvm-svn: 153114
2012-03-20 19:45:07 +00:00
93d5427c69
Whitespace.
...
llvm-svn: 153105
2012-03-20 18:38:33 +00:00
5a6011267a
[avx] Move the vextractf128 patterns closer to the vextractf128 def. Remove
...
whitespace from test case. No functional change intended.
llvm-svn: 153103
2012-03-20 18:24:55 +00:00
07a4cb9382
[avx] Adjust the VINSERTF128rm pattern to allow for unaligned loads.
...
This results in things such as
vmovups 16(%rdi), %xmm0
vinsertf128 $1, %xmm0, %ymm0, %ymm0
to be combined to
vinsertf128 $1, 16(%rdi), %ymm0, %ymm0
rdar://11076953
llvm-svn: 153092
2012-03-20 17:08:51 +00:00
b9b73170e3
[avx] Add patterns for VINSERTF128rm.
...
This results in things such as
vmovaps -96(%rbx), %xmm1
vinsertf128 $1, %xmm1, %ymm0, %ymm0
to be combined to
vinsertf128 $1, -96(%rbx), %ymm0, %ymm0
rdar://10643481
llvm-svn: 152762
2012-03-15 00:45:30 +00:00
57c8e7f364
*fix typo in comment; test of commit access
...
llvm-svn: 152507
2012-03-10 21:29:49 +00:00
a281afc676
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
a49ef92a76
This patch adds instruction latencies for the SSE instructions
...
to the instruction scheduler for the Intel Atom.
llvm-svn: 151590
2012-02-27 23:35:03 +00:00
682c76b7d4
Turn avx insert intrinsic calls into INSERT_SUBVECTOR DAG nodes and remove duplicate patterns for selecting the intrinsics
...
llvm-svn: 151342
2012-02-24 03:51:49 +00:00
e1d619691b
some comment fix for X86 and ARM
...
llvm-svn: 150902
2012-02-19 02:03:36 +00:00
b22310fda6
Emacs-tag and some comment fix for all ARM, CellSPU, Hexagon, MBlaze, MSP430, PPC, PTX, Sparc, X86, XCore.
...
llvm-svn: 150878
2012-02-18 12:03:15 +00:00
ba172d2d59
Remove the last of the old vector_shuffle patterns from X86 isel.
...
llvm-svn: 150795
2012-02-17 07:02:34 +00:00
cfad98f745
Move old movl vector_shuffle patterns. Not needed anymore since vector_shuffles shouldn't reach isel.
...
llvm-svn: 150462
2012-02-14 08:14:53 +00:00
8b19d78808
Still more vector_shuffle pattern removal.
...
llvm-svn: 150365
2012-02-13 07:23:41 +00:00
74650add0e
Remove more vector_shuffle patterns for unpack. These should be target specific nodes when they get to isel.
...
llvm-svn: 150363
2012-02-13 05:48:49 +00:00
6d471c9e49
Recommit r150328. Previous test failures should be fixed by r150360.
...
llvm-svn: 150362
2012-02-13 05:10:10 +00:00
0826c17d00
Revert r150328, "Remove more vector_shuffle patterns."
...
It caused 3 failures on pre-penryn and non-x86(generic) hosts.
llvm-svn: 150357
2012-02-13 00:10:15 +00:00
e24c94af81
Remove more vector_shuffle patterns.
...
llvm-svn: 150328
2012-02-12 08:14:35 +00:00
d40d9eb2b3
Remove more vector_shuffle patterns.
...
llvm-svn: 150321
2012-02-12 01:07:34 +00:00
330ca97700
Remove more vector_shuffle patterns.
...
llvm-svn: 150314
2012-02-11 23:31:01 +00:00
981c6cf7b3
Remove some patterns for matching vector_shuffle instructions since vector_shuffles should be custom lowered before isel.
...
llvm-svn: 150299
2012-02-11 07:43:35 +00:00
Craig Topper