maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
25,853 Commits 23 Branches 0 Tags 2.2 GiB
Commit Graph

242 Commits

Author SHA1 Message Date
Evan Cheng a0374e1bed Oops 
llvm-svn: 27989
 2006-04-27 05:44:50 +00:00
Evan Cheng 24eb3f4765 Bug fix: not updating NumIntRegs. 
llvm-svn: 27988
 2006-04-27 05:35:28 +00:00
Evan Cheng 48940d16b2 - Clean up formal argument lowering code. Prepare for vector pass by value work. 
- Fixed vararg support.

llvm-svn: 27985
 2006-04-27 01:32:22 +00:00
Evan Cheng 1c39903297 Fix fastcc failures. 
llvm-svn: 27980
 2006-04-26 18:21:31 +00:00
Evan Cheng e0bcfbe811 Switching over FORMAL_ARGUMENTS mechanism to lower call arguments. 
llvm-svn: 27975
 2006-04-26 01:20:17 +00:00
Evan Cheng a9467aab0a Separate LowerOperation() into multiple functions, one per opcode. 
llvm-svn: 27972
 2006-04-25 20:13:52 +00:00
Evan Cheng 5c2bfb069e Special case handling two wide build_vector(0, x). 
llvm-svn: 27961
 2006-04-24 22:58:52 +00:00
Evan Cheng b0461080e4 A little bit more build_vector enhancement for v8i16 cases. 
llvm-svn: 27959
 2006-04-24 18:01:45 +00:00
Evan Cheng b4f31dd1a8 MOVL shuffle (i.e. movd or movss / movsd from memory) of undef, V2 == V2 
llvm-svn: 27953
 2006-04-23 06:35:19 +00:00
Nate Begeman 4ca2ea5b43 JumpTable support! What this represents is working asm and jit support for 
x86 and ppc for 100% dense switch statements when relocations are non-PIC.
This support will be extended and enhanced in the coming days to support
PIC, and less dense forms of jump tables.

llvm-svn: 27947
 2006-04-22 18:53:45 +00:00
Evan Cheng e728efdfce Don't do all the lowering stuff for 2-wide build_vector's. Also, minor optimization for shuffle of undef. 
llvm-svn: 27946
 2006-04-22 08:34:05 +00:00
Evan Cheng 16ef94f4e8 Fix a performance regression. Use {p}shuf* when there are only two distinct elements in a build_vector. 
llvm-svn: 27945
 2006-04-22 06:21:46 +00:00
Evan Cheng 14215c36b6 Revamp build_vector lowering to take advantage of movss and movd instructions. 
movd always clear the top 96 bits and movss does so when it's loading the
value from memory.
The net result is codegen for 4-wide shuffles is much improved. It is near
optimal if one or more elements is a zero. e.g.

__m128i test(int a, int b) {
  return _mm_set_epi32(0, 0, b, a);
}

compiles to

_test:
	movd 8(%esp), %xmm1
	movd 4(%esp), %xmm0
	punpckldq %xmm1, %xmm0
	ret

compare to gcc:

_test:
	subl	$12, %esp
	movd	20(%esp), %xmm0
	movd	16(%esp), %xmm1
	punpckldq	%xmm0, %xmm1
	movq	%xmm1, %xmm0
	movhps	LC0, %xmm0
	addl	$12, %esp
	ret

or icc:

_test:
        movd      4(%esp), %xmm0                                #5.10
        movd      8(%esp), %xmm3                                #5.10
        xorl      %eax, %eax                                    #5.10
        movd      %eax, %xmm1                                   #5.10
        punpckldq %xmm1, %xmm0                                  #5.10
        movd      %eax, %xmm2                                   #5.10
        punpckldq %xmm2, %xmm3                                  #5.10
        punpckldq %xmm3, %xmm0                                  #5.10
        ret                                                     #5.10

There are still room for improvement, for example the FP variant of the above example:

__m128 test(float a, float b) {
  return _mm_set_ps(0.0, 0.0, b, a);
}

_test:
	movss 8(%esp), %xmm1
	movss 4(%esp), %xmm0
	unpcklps %xmm1, %xmm0
	xorps %xmm1, %xmm1
	movlhps %xmm1, %xmm0
	ret

The xorps and movlhps are unnecessary. This will require post legalizer optimization to handle.

llvm-svn: 27939
 2006-04-21 23:03:30 +00:00
Evan Cheng e8b5180044 Now generating perfect (I think) code for "vector set" with a single non-zero 
scalar value.

e.g.
        _mm_set_epi32(0, a, 0, 0);
==>
	movd 4(%esp), %xmm0
	pshufd $69, %xmm0, %xmm0

        _mm_set_epi8(0, 0, 0, 0, 0, a, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);
==>
	movzbw 4(%esp), %ax
	movzwl %ax, %eax
	pxor %xmm0, %xmm0
	pinsrw $5, %eax, %xmm0

llvm-svn: 27923
 2006-04-21 01:05:10 +00:00
Evan Cheng 60f0b8998e - Added support to turn "vector clear elements", e.g. pand V, <-1, -1, 0, -1> 
to a vector shuffle.
- VECTOR_SHUFFLE lowering change in preparation for more efficient codegen
of vector shuffle with zero (or any splat) vector.

llvm-svn: 27875
 2006-04-20 08:58:49 +00:00
Evan Cheng 15c264b753 Handle v2i64 BUILD_VECTOR custom lowering correctly. v2i64 is a legal type, 
but i64 is not. If possible, change a i64 op to a f64 (e.g. load, constant)
and then cast it back.

llvm-svn: 27849
 2006-04-20 00:11:39 +00:00
Evan Cheng 4a1b0d3292 isSplatMask() bug: first element can be an undef. 
llvm-svn: 27847
 2006-04-19 23:28:59 +00:00
Evan Cheng a3caaee503 - Added support to do aribitrary 4 wide shuffle with no more than three 
instructions.
- Fixed a commute vector_shuff bug.

llvm-svn: 27845
 2006-04-19 22:48:17 +00:00
Evan Cheng 7855e4d032 Commute vector_shuffle to match more movlhps, movlp{s|d} cases. 
llvm-svn: 27840
 2006-04-19 20:35:22 +00:00
Evan Cheng 5421206c4b Use movss to insert_vector_elt(v, s, 0). 
llvm-svn: 27782
 2006-04-17 22:45:49 +00:00
Evan Cheng 6e5e205841 Use two pinsrw to insert an element into v4i32 / v4f32 vector. 
llvm-svn: 27779
 2006-04-17 22:04:06 +00:00
Evan Cheng 5022b3426e Implement v8i16, v16i8 splat using unpckl + pshufd. 
llvm-svn: 27768
 2006-04-17 20:43:08 +00:00
Chris Lattner c070c621ac implement returns of a vector, testcase here: CodeGen/X86/vec_return.ll 
llvm-svn: 27767
 2006-04-17 20:32:50 +00:00
Evan Cheng b3b41c4f3d FP SETOLT, SETOLT, SETUGE, SETUGT conditions were implemented incorrectly 
llvm-svn: 27755
 2006-04-17 07:24:10 +00:00
Evan Cheng 6222cf2a36 Silly bug 
llvm-svn: 27719
 2006-04-15 05:37:34 +00:00
Evan Cheng 65bb720a8b Do not use movs{h|l}dup for a shuffle with a single non-undef node. 
llvm-svn: 27718
 2006-04-15 03:13:24 +00:00
Evan Cheng 5d247f81c1 Last few SSE3 intrinsics. 
llvm-svn: 27711
 2006-04-14 21:59:03 +00:00
Evan Cheng e4f97ccf7f X86 SSE2 supports v8i16 multiplication 
llvm-svn: 27644
 2006-04-13 05:10:25 +00:00
Evan Cheng 92232307d0 All "integer" logical ops (pand, por, pxor) are now promoted to v2i64. 
Clean up and fix various logical ops issues.

llvm-svn: 27633
 2006-04-12 21:21:57 +00:00
Evan Cheng e2157c6e41 Promote v4i32, v8i16, v16i8 load to v2i64 load. 
llvm-svn: 27612
 2006-04-12 17:12:36 +00:00
Evan Cheng 12ba3e23d0 Added support for _mm_move_ss and _mm_move_sd. 
llvm-svn: 27575
 2006-04-11 00:19:04 +00:00
Evan Cheng 617a6a812e Conditional move of vector types. 
llvm-svn: 27556
 2006-04-10 07:23:14 +00:00
Evan Cheng ac847268c5 Code clean up. 
llvm-svn: 27501
 2006-04-07 21:53:05 +00:00
Evan Cheng c995b45f67 - movlp{s|d} and movhp{s|d} support. 
- Normalize shuffle nodes so result vector lower half elements come from the
  first vector, the rest come from the second vector. (Except for the
  exceptions :-).
- Other minor fixes.

llvm-svn: 27474
 2006-04-06 23:23:56 +00:00
Evan Cheng 780382946e Support for comi / ucomi intrinsics. 
llvm-svn: 27444
 2006-04-05 23:38:46 +00:00
Evan Cheng f3b52c84ea Handle canonical form of e.g. 
vector_shuffle v1, v1, <0, 4, 1, 5, 2, 6, 3, 7>

This is turned into
vector_shuffle v1, <undef>, <0, 0, 1, 1, 2, 2, 3, 3>
by dag combiner.

It would match a {p}unpckl on x86.

llvm-svn: 27437
 2006-04-05 07:20:06 +00:00
Evan Cheng 6d196db40d Bogus assert 
llvm-svn: 27434
 2006-04-05 06:11:20 +00:00
Evan Cheng 2cf4232ced Fallthrough to expand if a VECTOR_SHUFFLE cannot be custom lowered. 
llvm-svn: 27433
 2006-04-05 06:09:26 +00:00
Evan Cheng 59a6355e82 Handle v8i16 shuffle that must be broken into a pair of pshufhw / pshuflw. 
llvm-svn: 27427
 2006-04-05 01:47:37 +00:00
Evan Cheng b64827e662 Use movlpd to: store lower f64 extracted from v2f64. 
Use movhpd to: store upper f64 extracted from v2f64.

llvm-svn: 27382
 2006-04-03 22:30:54 +00:00
Evan Cheng ebf1006d16 - More efficient extract_vector_elt with shuffle and movss, movsd, movd, etc. 
- Some bug fixes and naming inconsistency fixes.

llvm-svn: 27377
 2006-04-03 20:53:28 +00:00
Evan Cheng 5fd7c69473 Use a X86 target specific node X86ISD::PINSRW instead of a mal-formed 
INSERT_VECTOR_ELT to insert a 16-bit value in a 128-bit vector.

llvm-svn: 27314
 2006-03-31 21:55:24 +00:00
Evan Cheng cbffa4656b Add support to use pextrw and pinsrw to extract and insert a word element 
from a 128-bit vector.

llvm-svn: 27304
 2006-03-31 19:22:53 +00:00
Evan Cheng 1b0d294de0 Expand all INSERT_VECTOR_ELT (obviously bad) for now. 
llvm-svn: 27275
 2006-03-31 01:30:39 +00:00
Evan Cheng d9d0bbb5ac Typo 
llvm-svn: 27272
 2006-03-31 00:33:57 +00:00
Evan Cheng 99d7205fba Ok for vector_shuffle mask to contain undef elements. 
llvm-svn: 27271
 2006-03-31 00:30:29 +00:00
Evan Cheng 7e2ff11a42 Make sure all possible shuffles are matched. 
Use pshufd, pshuhw, and pshulw to shuffle v4f32 if shufps doesn't match.
Use shufps to shuffle v4f32 if pshufd, pshuhw, and pshulw don't match.

llvm-svn: 27259
 2006-03-30 19:54:57 +00:00
Evan Cheng b7fedffc78 - Added some SSE2 128-bit packed integer ops. 
- Added SSE2 128-bit integer pack with signed saturation ops.
- Added pshufhw and pshuflw ops.

llvm-svn: 27252
 2006-03-29 23:07:14 +00:00
Evan Cheng acc336475e Need to special case splat after all. Make the second operand of splat 
vector_shuffle undef.

llvm-svn: 27250
 2006-03-29 19:02:40 +00:00
Evan Cheng 500ec16578 - More shuffle related bug fixes. 
- Whenever possible use ops of the right packed types for vector shuffles /
  splats.

llvm-svn: 27246
 2006-03-29 03:04:49 +00:00
Evan Cheng da59b0d2a8 - Only use pshufd for v4i32 vector shuffles. 
- Other shuffle related fixes.

llvm-svn: 27244
 2006-03-29 01:30:51 +00:00
Evan Cheng 8160fd3d42 Fixing buggy code. 
llvm-svn: 27239
 2006-03-28 23:41:33 +00:00
Jim Laskey 457e54efc1 Added missing paren on behalf of Ramana Radhakrishnan. 
llvm-svn: 27223
 2006-03-28 10:17:11 +00:00
Evan Cheng 21e5476deb Missed X86::isUNPCKHMask 
llvm-svn: 27222
 2006-03-28 08:27:15 +00:00
Evan Cheng 1a194a5264 * Prefer using operation of matching types. e.g unpcklpd rather than movlhps. 
* Bug fixes.

llvm-svn: 27218
 2006-03-28 06:50:32 +00:00
Evan Cheng 2bc3280659 - Clean up / consoladate various shuffle masks. 
- Some misc. bug fixes.
- Use MOVHPDrm to load from m64 to upper half of a XMM register.

llvm-svn: 27210
 2006-03-28 02:43:26 +00:00
Evan Cheng 5df75889db Model unpack lower and interleave as vector_shuffle so we can lower the 
intrinsics as such.

llvm-svn: 27200
 2006-03-28 00:39:58 +00:00
Evan Cheng 9b9cc4fb39 Use pcmpeq to generate vector of all ones. 
llvm-svn: 27167
 2006-03-27 07:00:16 +00:00
Nate Begeman ed728c1291 SelectionDAGISel can now natively handle Switch instructions, in the same 
manner that the LowerSwitch LLVM to LLVM pass does: emitting a binary
search tree of basic blocks.  The new approach has several advantages:
it is faster, it generates significantly smaller code in many cases, and
it paves the way for implementing dense switch tables as a jump table by
handling switches directly in the instruction selector.

This functionality is currently only enabled on x86, but should be safe for
every target.  In anticipation of making it the default, the cfg is now
properly updated in the x86, ppc, and sparc select lowering code.

llvm-svn: 27156
 2006-03-27 01:32:24 +00:00
Evan Cheng ed6184aef2 Remove X86:isZeroVector, use ISD::isBuildVectorAllZeros instead; some fixes / cleanups 
llvm-svn: 27150
 2006-03-26 09:53:12 +00:00
Evan Cheng 2bc0941e2a Build arbitrary vector with more than 2 distinct scalar elements with a 
series of unpack and interleave ops.

llvm-svn: 27119
 2006-03-25 09:37:23 +00:00
Evan Cheng 6f7d31ea50 Added 128-bit packed integer subtraction. 
llvm-svn: 27096
 2006-03-25 01:33:37 +00:00
Evan Cheng e7ee6a5e32 Support for scalar to vector with zero extension. 
llvm-svn: 27091
 2006-03-24 23:15:12 +00:00
Evan Cheng 082c8785ef Handle BUILD_VECTOR with all zero elements. 
llvm-svn: 27056
 2006-03-24 07:29:27 +00:00
Chris Lattner f5efddf80b Gabor points out that we can't spell. :) 
llvm-svn: 27049
 2006-03-24 07:12:19 +00:00
Evan Cheng a91d8a5b43 All v2f64 shuffle cases can be handled. 
llvm-svn: 27044
 2006-03-24 06:40:32 +00:00
Evan Cheng 2595a687da More efficient v2f64 shuffle using movlhps, movhlps, unpckhpd, and unpcklpd. 
llvm-svn: 27040
 2006-03-24 02:58:06 +00:00
Evan Cheng d27fb3e85e Handle more shuffle cases with SHUFP* instructions. 
llvm-svn: 27024
 2006-03-24 01:18:28 +00:00
Evan Cheng f842ea57bb Typo 
llvm-svn: 26997
 2006-03-23 20:26:04 +00:00
Evan Cheng b9b0550dc6 Add 128-bit integer vector load and add (for testing). 
llvm-svn: 26967
 2006-03-23 01:57:24 +00:00
Evan Cheng 021bb7c956 Added a ValueType operand to isShuffleMaskLegal(). For now, x86 will not do 
64-bit vector shuffle.

llvm-svn: 26964
 2006-03-22 22:07:06 +00:00
Evan Cheng bc04722860 Some clean up. 
llvm-svn: 26957
 2006-03-22 19:22:18 +00:00
Evan Cheng d4e1557941 - Supposely movlhps is faster / better than unpcklpd. 
- Don't forget pshufd is only available with sse2.

llvm-svn: 26956
 2006-03-22 19:16:21 +00:00
Evan Cheng 68ad48bd1a - Implement X86ISelLowering::isShuffleMaskLegal(). We currently only support 
splat and PSHUFD cases.
- Clean up shuffle / splat matching code.

llvm-svn: 26954
 2006-03-22 18:59:22 +00:00
Evan Cheng 8fdbdf20cd - VECTOR_SHUFFLE of v4i32 / v4f32 with undef second vector always matches 
PSHUFD. We can make permutes entries which point to the undef pointing
  anything we want.
- Change some names to appease Chris.

llvm-svn: 26951
 2006-03-22 08:01:21 +00:00
Chris Lattner f5e36c8bc0 fix a warning 
llvm-svn: 26941
 2006-03-22 04:18:34 +00:00
Evan Cheng d097e67544 Some splat and shuffle support. 
llvm-svn: 26940
 2006-03-22 02:53:00 +00:00
Evan Cheng d5e905d762 - Use movaps to store 128-bit vector integers. 
- Each scalar to vector v8i16 and v16i8 is a any_extend followed by a movd.

llvm-svn: 26932
 2006-03-21 23:01:21 +00:00
Chris Lattner 00f4683bf6 These targets don't support EXTRACT_VECTOR_ELT, though, in time, X86 will. 
llvm-svn: 26930
 2006-03-21 20:51:05 +00:00
Chris Lattner 80b6bd2746 Add a build_vector node 
llvm-svn: 26895
 2006-03-20 06:18:01 +00:00
Chris Lattner f7b6e7212f rename these nodes 
llvm-svn: 26848
 2006-03-19 01:13:28 +00:00
Evan Cheng b09a56f3a4 Darwin should use _setjmp/_longjmp instead of setjmp/longjmp. 
llvm-svn: 26833
 2006-03-17 20:31:41 +00:00
Chris Lattner 388fc4d9fb Disable x86 fastcc from passing args in registers 
llvm-svn: 26824
 2006-03-17 17:27:47 +00:00
Chris Lattner 43798850f9 Parameterize the number of integer arguments to pass in registers 
llvm-svn: 26818
 2006-03-17 05:10:20 +00:00
Nate Begeman bb01d4f272 Remove BRTWOWAY* 
Make the PPC backend not dependent on BRTWOWAY_CC and make the branch
selector smarter about the code it generates, fixing a case in the
readme.

llvm-svn: 26814
 2006-03-17 01:40:33 +00:00
Evan Cheng f75555feb9 Bug fix: condition inverted. 
llvm-svn: 26804
 2006-03-16 22:02:48 +00:00
Evan Cheng 20931a798e Added a way for TargetLowering to specify what values can be used as the 
scale component of the target addressing mode.

llvm-svn: 26802
 2006-03-16 21:47:42 +00:00
Evan Cheng af598d2461 Add LSR hooks. 
llvm-svn: 26740
 2006-03-13 23:18:16 +00:00
Evan Cheng adc7093fc1 Use rep/stosl; and Count 0x3; rep/stosb for memset with 4 byte aligned dest. 
and variable value.
Similarly for memcpy.

llvm-svn: 26603
 2006-03-07 23:29:39 +00:00
Evan Cheng 30d7b70b73 Enable Dwarf debugging info. 
llvm-svn: 26581
 2006-03-07 02:02:57 +00:00
Chris Lattner 9c7f50376a Copysign needs to be expanded everywhere. Note that Alpha and IA64 should 
implement copysign as a native op if they have it.

llvm-svn: 26541
 2006-03-05 05:08:37 +00:00
Evan Cheng 6dc73297c3 MEMSET / MEMCPY lowering bugs: we can't issue a single WORD / DWORD version of 
rep/stos and rep/mov if the count is not a constant. We could do
  rep/stosl; and $count, 3; rep/stosb
For now, I will lower them to memset / memcpy calls. We will revisit this after
a little bit experiment.

Also need to take care of the trailing bytes even if the count is a constant.
Since the max. number of trailing bytes are 3, we will simply issue loads /
stores.

llvm-svn: 26517
 2006-03-04 02:48:56 +00:00
Evan Cheng 084a102b17 Typo 
llvm-svn: 26512
 2006-03-04 01:12:00 +00:00
Chris Lattner ad3c974a77 remove the read/write port/io intrinsics. 
llvm-svn: 26479
 2006-03-03 00:19:58 +00:00
Evan Cheng 1926427351 Vector op lowering. 
llvm-svn: 26438
 2006-03-01 01:11:20 +00:00
Evan Cheng 994700101e Added a common about the need for X86ISD::Wrapper. 
llvm-svn: 26372
 2006-02-25 09:55:19 +00:00
Evan Cheng e0ed6ec13f - Clean up the lowering and selection code of ConstantPool, GlobalAddress, 
and ExternalSymbol.
- Use C++ code (rather than tblgen'd selection code) to match the above
  mentioned leaf nodes. Do not mutate and nodes and do not record the
  selection in CodeGenMap. These nodes should be safe to duplicate. This is
  a performance win.

llvm-svn: 26335
 2006-02-23 20:41:18 +00:00
Evan Cheng 1f342c2884 PIC related bug fixes. 
1. Various asm printer bug.
2. Lowering bug. Now TargetGlobalAddress is wrapped in X86ISD::TGAWrapper.

llvm-svn: 26324
 2006-02-23 02:43:52 +00:00
Evan Cheng 73136dfecc - Added option -relocation-model to set relocation model. Valid values include static, pic, 
dynamic-no-pic, and default.
PPC and x86 default is dynamic-no-pic for Darwin, pic for others.
- Removed options -enable-pic and -ppc-static.

llvm-svn: 26315
 2006-02-22 20:19:42 +00:00
Evan Cheng 9e252e3bcf Added MMX, SSE1, and SSE2 vector instructions and some simple patterns. 
Fixed some existing bugs (wrong predicates, prefixes) at the same time.

llvm-svn: 26310
 2006-02-22 02:26:30 +00:00