RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-28 14:32:51 +01:00
Code Issues Projects Releases Wiki Activity
31,524 Commits 4 Branches 5 Tags 704 MiB
3553a131d0
Commit Graph

1721 Commits

Author SHA1 Message Date
Chris Lattner
3553a131d0 Implement InstCombine/vec_demanded_elts.ll:test2. This allows us to turn 
unsigned test(float f) {
 return _mm_cvtsi128_si32( (__m128i) _mm_set_ss( f*f ));
}

into:

_test:
        movss 4(%esp), %xmm0
        mulss %xmm0, %xmm0
        movd %xmm0, %eax
        ret

instead of:

_test:
        movss 4(%esp), %xmm0
        mulss %xmm0, %xmm0
        xorps %xmm1, %xmm1
        movss %xmm0, %xmm1
        movd %xmm1, %eax
        ret

GCC gets:

_test:
        subl    $28, %esp
        movss   32(%esp), %xmm0
        mulss   %xmm0, %xmm0
        xorps   %xmm1, %xmm1
        movss   %xmm0, %xmm1
        movaps  %xmm1, %xmm0
        movd    %xmm0, 12(%esp)
        movl    12(%esp), %eax
        addl    $28, %esp
        ret

llvm-svn: 36020
 2007-04-14 22:29:23 +00:00
Chris Lattner
6f64f54168 Implement PR1201 and test/Transforms/InstCombine/malloc-free-delete.ll 
llvm-svn: 35981
 2007-04-14 00:20:02 +00:00
Chris Lattner
b97ff21db2 use an accessor to simplify code. 
llvm-svn: 35979
 2007-04-14 00:17:39 +00:00
Chris Lattner
8477dd1722 Now that codegen prepare isn't defeating me, I can finally fix what I set 
out to do! :)

This fixes a problem where LSR would insert a bunch of code into each MBB
that uses a particular subexpression (e.g. IV+base+C).  The problem is that
this code cannot be CSE'd back together if inserted into different blocks.

This patch changes LSR to attempt to insert a single copy of this code and
share it, allowing codegenprepare to duplicate the code if it can be sunk
into various addressing modes.  On CodeGen/ARM/lsr-code-insertion.ll,
for example, this gives us code like:

        add r8, r0, r5
        str r6, [r8, #+4]
..
        ble LBB1_4      @cond_next
LBB1_3: @cond_true
        str r10, [r8, #+4]
LBB1_4: @cond_next
...
LBB1_5: @cond_true55
        ldr r6, LCPI1_1
        str r6, [r8, #+4]

instead of:

        add r10, r0, r6
        str r8, [r10, #+4]
...
        ble LBB1_4      @cond_next
LBB1_3: @cond_true
        add r8, r0, r6
        str r10, [r8, #+4]
LBB1_4: @cond_next
...
LBB1_5: @cond_true55
        add r8, r0, r6
        ldr r10, LCPI1_1
        str r10, [r8, #+4]

Besides being smaller and more efficient, this makes it immediately
obvious that it is profitable to predicate LBB1_3 now :)

llvm-svn: 35972
 2007-04-13 20:42:26 +00:00
Chris Lattner
bc03b6c341 Completely rewrite addressing-mode related sinking of code. In particular, 
this fixes problems where codegenprepare would sink expressions into load/stores
that are not valid, and fixes cases where it would miss important valid ones.

This fixes several serious codesize and perf issues, particularly on targets
with complex addressing modes like arm and x86.  For example, now we compile
CodeGen/X86/isel-sink.ll to:

_test:
        movl 8(%esp), %eax
        movl 4(%esp), %ecx
        cmpl $1233, %eax
        ja LBB1_2       #F
LBB1_1: #T
        movl $4, (%ecx,%eax,4)
        movl $141, %eax
        ret
LBB1_2: #F
        movl (%ecx,%eax,4), %eax
        ret

instead of:

_test:
        movl 8(%esp), %eax
        leal (,%eax,4), %ecx
        addl 4(%esp), %ecx
        cmpl $1233, %eax
        ja LBB1_2       #F
LBB1_1: #T
        movl $4, (%ecx)
        movl $141, %eax
        ret
LBB1_2: #F
        movl (%ecx), %eax
        ret

llvm-svn: 35970
 2007-04-13 20:30:56 +00:00
Chris Lattner
f7451ea3c2 Fix Transforms/ScalarRepl/union-pointer.ll 
llvm-svn: 35906
 2007-04-11 15:45:25 +00:00
Chris Lattner
27a80589de Turn stuff like: 
icmp slt i32 %X, 0              ; <i1>:0 [#uses=1]
        sext i1 %0 to i32               ; <i32>:1 [#uses=1]

into:

        %X.lobit = ashr i32 %X, 31              ; <i32> [#uses=1]

This implements InstCombine/icmp.ll:test[34]

llvm-svn: 35891
 2007-04-11 06:57:46 +00:00
Chris Lattner
b659c04f13 Simplify some comparisons to arithmetic, this implements: 
Transforms/InstCombine/icmp.ll

llvm-svn: 35890
 2007-04-11 06:53:04 +00:00
Chris Lattner
50a7c8f34e canonicalize (x <u 2147483648) -> (x >s -1) and (x >u 2147483647) -> (x <s 0) 
llvm-svn: 35886
 2007-04-11 06:12:58 +00:00
Chris Lattner
cbd4a7e79c fix a miscompilation of: 
define i32 @test(i32 %X) {
entry:
        %Y = and i32 %X, 4              ; <i32> [#uses=1]
        icmp eq i32 %Y, 0               ; <i1>:0 [#uses=1]
        sext i1 %0 to i32               ; <i32>:1 [#uses=1]
        ret i32 %1
}

by moving code out of commonIntCastTransforms into visitZExt.  Simplify the
APInt gymnastics in it etc.

llvm-svn: 35885
 2007-04-11 05:45:39 +00:00
Chris Lattner
c7c7a4712e fix a regression introduced by my last patch. 
llvm-svn: 35879
 2007-04-11 03:27:24 +00:00
Chris Lattner
4ea7a156ba Simplify SROA conversion to integer in some ways, make it more general in others. 
We now tolerate small amounts of undefined behavior, better emulating what
would happen if the transaction actually occurred in memory.  This fixes
SingleSource/UnitTests/2007-04-10-BitfieldTest.c on PPC, at least until
Devang gets a chance to fix the CFE from doing undefined things with bitfields :)

llvm-svn: 35875
 2007-04-11 00:57:54 +00:00
Chris Lattner
fe1860b138 Strengthen the boundary conditions of this fold, implementing 
InstCombine/set.ll:test25

llvm-svn: 35852
 2007-04-09 23:52:13 +00:00
Chris Lattner
78fffcb81b eliminate the last uses of some TLI methods. 
llvm-svn: 35844
 2007-04-09 23:29:07 +00:00
Chris Lattner
87c89cafb2 switch LSR to use isLegalAddressingMode instead of other simpler hooks 
llvm-svn: 35837
 2007-04-09 22:20:14 +00:00
Devang Patel
70205cceea Check _all_ PHINodes. 
llvm-svn: 35836
 2007-04-09 22:20:10 +00:00
Devang Patel
5392489e86 Insert new pre-header before new header. Original pre-header may 
happen to be an entry, in such case, it is not a good idea to
insert new block before entry.

Also fix typo in assertion check.

llvm-svn: 35833
 2007-04-09 21:40:43 +00:00
Devang Patel
cdea453adb Preserve canonical loop form. 
llvm-svn: 35829
 2007-04-09 20:19:46 +00:00
Devang Patel
9263a797b3 Do not create new pre-header. Reuse original pre-header. 
llvm-svn: 35825
 2007-04-09 19:04:21 +00:00
Devang Patel
e038420dc6 Simpler for() loops. 
llvm-svn: 35822
 2007-04-09 17:09:13 +00:00
Devang Patel
dd269ce747 Fix future bug. Of course, Chris spotted this. 
Handle Argument or Undef as an incoming PHI value.

llvm-svn: 35821
 2007-04-09 16:41:46 +00:00
Devang Patel
ba5018aaff More cosmetic changes. 
llvm-svn: 35820
 2007-04-09 16:21:29 +00:00
Devang Patel
f66f3dd962 Only cosmetic changes. Zero functionality Change. 
llvm-svn: 35819
 2007-04-09 16:11:48 +00:00
Chris Lattner
218d43af10 Fix PR1304 and Transforms/InstCombine/2007-04-08-SingleEltVectorCrash.ll 
llvm-svn: 35792
 2007-04-09 01:37:55 +00:00
Chris Lattner
b3d105a4f9 Eliminate useless insertelement instructions. This implements 
Transforms/InstCombine/vec_insertelt.ll and fixes PR1286.

We now compile the code from that bug into:

_foo:
        movl 4(%esp), %eax
        movdqa (%eax), %xmm0
        movl 8(%esp), %ecx
        psllw (%ecx), %xmm0
        movdqa %xmm0, (%eax)
        ret

instead of:

_foo:
        subl $4, %esp
        movl %ebp, (%esp)
        movl %esp, %ebp
        movl 12(%ebp), %eax
        movdqa (%eax), %xmm0
        #IMPLICIT_DEF %eax
        pinsrw $2, %eax, %xmm0
        xorl %ecx, %ecx
        pinsrw $3, %ecx, %xmm0
        pinsrw $4, %eax, %xmm0
        pinsrw $5, %ecx, %xmm0
        pinsrw $6, %eax, %xmm0
        pinsrw $7, %ecx, %xmm0
        movl 8(%ebp), %eax
        movdqa (%eax), %xmm1
        psllw %xmm0, %xmm1
        movdqa %xmm1, (%eax)
        movl %ebp, %esp
        popl %ebp
        ret

woo :)

llvm-svn: 35788
 2007-04-09 01:11:16 +00:00
Chris Lattner
1a1b798eb5 reenable this xform, whoops :) 
llvm-svn: 35765
 2007-04-08 08:01:49 +00:00
Chris Lattner
1760b42378 Fix regression on Instcombine/apint-or2.ll 
llvm-svn: 35763
 2007-04-08 07:55:22 +00:00
Chris Lattner
d435e0bfd2 Generalize the code that handles (A&B)|(A&C) to work where B/C are not constants. 
Add a new xform to simplify (A&B)|(~A&C).  THis implements InstCombine/or2.ll:test1

llvm-svn: 35760
 2007-04-08 07:47:01 +00:00
Nick Lewycky
e6cb3e2433 Add support for cast instructions. 
llvm-svn: 35734
 2007-04-07 15:48:32 +00:00
Owen Anderson
85b0e20f2a Completely purge DomSet. This is the (hopefully) final patch for PR1171. 
llvm-svn: 35731
 2007-04-07 07:17:27 +00:00
Nick Lewycky
3e77af40ff Support NE inequality in ValueRanges. 
llvm-svn: 35724
 2007-04-07 04:49:12 +00:00
Nick Lewycky
7fbec59fb4 Cleanup. Refactor out the applying of value ranges to its own method. 
llvm-svn: 35719
 2007-04-07 03:36:51 +00:00
Nick Lewycky
80cf96b3f8 Use TargetData to find the size of a type. 
llvm-svn: 35718
 2007-04-07 03:16:12 +00:00
Nick Lewycky
3ddf638983 Strengthen icmp snuggling by doing 'compare-or-equal-to' to 'compare' 
first and then range testing second.

llvm-svn: 35715
 2007-04-07 02:30:14 +00:00
Devang Patel
562df7f986 Add loop rotation pass. 
llvm-svn: 35714
 2007-04-07 01:25:15 +00:00
Chris Lattner
03c84be56b implement Transforms/InstCombine/malloc2.ll and PR1313 
llvm-svn: 35700
 2007-04-06 18:57:34 +00:00
Chris Lattner
997967979f Use a worklist-driven algorithm instead of a recursive one. 
llvm-svn: 35680
 2007-04-05 01:27:02 +00:00
Dale Johannesen
fb15913194 Prevent transformConstExprCastCall from generating conversions that assert 
elsewhere.

llvm-svn: 35668
 2007-04-04 19:16:42 +00:00
Jeff Cohen
01d4afe6da Fix 2007-04-04-BadFoldBitcastIntoMalloc.ll 
llvm-svn: 35665
 2007-04-04 16:58:57 +00:00
Duncan Sands
de998e6599 Fix comment. 
llvm-svn: 35655
 2007-04-04 06:42:45 +00:00
Chris Lattner
974e931689 Fix a bug I introduced with my patch yesterday which broke Qt (I converted 
some constant exprs to apints).

Thanks to Anton for tracking down a small testcase that triggered this!

llvm-svn: 35633
 2007-04-03 23:29:39 +00:00
Chris Lattner
bfe18d29f9 reinstate the previous two patches, with a bugfix :) 
ldecod now passes.

llvm-svn: 35626
 2007-04-03 17:43:25 +00:00
Evan Cheng
e30cfe8e9e Reverting back to 1.723. The last two commits broke JM (and possibily others) on ARM. 
llvm-svn: 35620
 2007-04-03 08:11:50 +00:00
Chris Lattner
4302c8bcdb split some code out into a helper function 
llvm-svn: 35615
 2007-04-03 05:11:24 +00:00
Chris Lattner
23d58bde29 Split a whole ton of code out of visitICmpInst into visitICmpInstWithInstAndIntCst. 
llvm-svn: 35614
 2007-04-03 04:46:52 +00:00
Chris Lattner
9b66a372c8 Fix PR1253 and xor2.ll:test[01] 
llvm-svn: 35612
 2007-04-03 01:47:41 +00:00
Chris Lattner
3df2382275 allow -1 strides to reuse "1" strides. 
llvm-svn: 35607
 2007-04-02 22:51:58 +00:00
Zhou Sheng
98c161c290 1. Make use of APInt operation instead of using ConstantExpr::getXXX. 
2. Use cheaper APInt methods.

llvm-svn: 35594
 2007-04-02 13:45:30 +00:00
Zhou Sheng
47311533a7 Use uint32_t for bitwidth instead of unsigned. 
llvm-svn: 35593
 2007-04-02 08:20:41 +00:00
Chris Lattner
d7c96e25e4 Pass the type of the store access, not the type of the store, into the 
target hook.  This allows us to codegen a loop as:

LBB1_1: @cond_next
        mov r2, #0
        str r2, [r0, +r3, lsl #2]
        add r3, r3, #1
        cmn r3, #1
        bne LBB1_1      @cond_next

instead of:

LBB1_1: @cond_next
        mov r2, #0
        str r2, [r0], #+4
        add r3, r3, #1
        cmn r3, #1
        bne LBB1_1      @cond_next

This looks the same, but has one fewer induction variable (and therefore,
one fewer register) live in the loop.

llvm-svn: 35592
 2007-04-02 06:34:44 +00:00