llvm-svn: 16805

llvm-svn: 16804

llvm-svn: 16803

* Instead of handling dead functions specially, just nuke them.
* Be more aggressive about cleaning up after constification, in
  particular, handle getelementptr instructions and constantexprs.
* Be a little bit more structured about how we process globals.

*** Delete globals that are only stored to, and never read.  These are
    clearly not useful, so they should go.  This implements deadglobal.llx

This last one triggers quite a few times.  In particular, 2208 in the
external tests, 1865 of which are in 252.eon.  This shrinks eon from
1995094 to 1732341 bytes of bytecode.

llvm-svn: 16802

llvm-svn: 16801

llvm-svn: 16800

llvm-svn: 16799

llvm-svn: 16798

llvm-svn: 16797

simplifications of the resultant program to avoid making later passes
do it all.

This allows us to constify globals that just have the same constant that
they are initialized stored into them.

Suprisingly this comes up ALL of the freaking time, dozens of times in
SPEC, 30 times in vortex alone.

For example, on 256.bzip2, it allows us to constify these two globals:

%smallMode = internal global ubyte 0             ; <ubyte*> [#uses=8]
%verbosity = internal global int 0               ; <int*> [#uses=49]

Which (with later optimizations) results in the bytecode file shrinking
from 82286 to 69686 bytes!  Lets hear it for IPO :)

For the record, it's nuking lots of "if (verbosity > 2) { do lots of stuff }"
code.

llvm-svn: 16793

llvm-svn: 16791

llvm-svn: 16779

llvm-svn: 16774

llvm-svn: 16773

(PromoteAbstractToConcrete), and to use a set to avoid recomputation.
In particular, this set eliminates the potentially exponential cases
from this little recursive algorithm.

On a particularly nasty testcase, llvm-dis on the .bc file went from 34
minutes (which is when I killed it, it still hadn't finished) to 0.57s.
Remember kids, exponential algorithms are bad.

llvm-svn: 16772

llvm-svn: 16771

llvm-svn: 16770

llvm-svn: 16769

llvm-svn: 16768

the JIT had last night.

llvm-svn: 16766

llvm-svn: 16765

instruction.

Now, rather than emitting the following loop out of bisect:
.LBB_main_19:	; no_exit.0.i
	rlwinm r3, r2, 3, 0, 28
	lfdx f1, r3, r27
	addis r3, r30, ha16(.CPI_main_1-"L00000$pb")
	lfd f2, lo16(.CPI_main_1-"L00000$pb")(r3)
	fsub f2, f2, f1
	addis r3, r30, ha16(.CPI_main_1-"L00000$pb")
	lfd f4, lo16(.CPI_main_1-"L00000$pb")(r3)
	fcmpu cr0, f1, f4
	bge .LBB_main_64	; no_exit.0.i
.LBB_main_63:	; no_exit.0.i
	b .LBB_main_65	; no_exit.0.i
.LBB_main_64:	; no_exit.0.i
	fmr f2, f1
.LBB_main_65:	; no_exit.0.i
	addi r3, r2, 1
	rlwinm r3, r3, 3, 0, 28
	lfdx f1, r3, r27
	addis r3, r30, ha16(.CPI_main_1-"L00000$pb")
	lfd f4, lo16(.CPI_main_1-"L00000$pb")(r3)
	fsub f4, f4, f1
	addis r3, r30, ha16(.CPI_main_1-"L00000$pb")
	lfd f5, lo16(.CPI_main_1-"L00000$pb")(r3)
	fcmpu cr0, f1, f5
	bge .LBB_main_67	; no_exit.0.i
.LBB_main_66:	; no_exit.0.i
	b .LBB_main_68	; no_exit.0.i
.LBB_main_67:	; no_exit.0.i
	fmr f4, f1
.LBB_main_68:	; no_exit.0.i
	fadd f1, f2, f4
	addis r3, r30, ha16(.CPI_main_2-"L00000$pb")
	lfd f2, lo16(.CPI_main_2-"L00000$pb")(r3)
	fmul f1, f1, f2
	rlwinm r3, r2, 3, 0, 28
	lfdx f2, r3, r28
	fadd f4, f2, f1
	fcmpu cr0, f4, f0
	bgt .LBB_main_70	; no_exit.0.i
.LBB_main_69:	; no_exit.0.i
	b .LBB_main_71	; no_exit.0.i
.LBB_main_70:	; no_exit.0.i
	fmr f0, f4
.LBB_main_71:	; no_exit.0.i
	fsub f1, f2, f1
	addi r2, r2, -1
	fcmpu cr0, f1, f3
	blt .LBB_main_73	; no_exit.0.i
.LBB_main_72:	; no_exit.0.i
	b .LBB_main_74	; no_exit.0.i
.LBB_main_73:	; no_exit.0.i
	fmr f3, f1
.LBB_main_74:	; no_exit.0.i
	cmpwi cr0, r2, -1
	fmr f16, f0
	fmr f17, f3
	bgt .LBB_main_19	; no_exit.0.i

We emit this instead:
.LBB_main_19:	; no_exit.0.i
	rlwinm r3, r2, 3, 0, 28
	lfdx f1, r3, r27
	addis r3, r30, ha16(.CPI_main_1-"L00000$pb")
	lfd f2, lo16(.CPI_main_1-"L00000$pb")(r3)
	fsub f2, f2, f1
	fsel f1, f1, f1, f2
	addi r3, r2, 1
	rlwinm r3, r3, 3, 0, 28
	lfdx f2, r3, r27
	addis r3, r30, ha16(.CPI_main_1-"L00000$pb")
	lfd f4, lo16(.CPI_main_1-"L00000$pb")(r3)
	fsub f4, f4, f2
	fsel f2, f2, f2, f4
	fadd f1, f1, f2
	addis r3, r30, ha16(.CPI_main_2-"L00000$pb")
	lfd f2, lo16(.CPI_main_2-"L00000$pb")(r3)
	fmul f1, f1, f2
	rlwinm r3, r2, 3, 0, 28
	lfdx f2, r3, r28
	fadd f4, f2, f1
	fsub f5, f0, f4
	fsel f0, f5, f0, f4
	fsub f1, f2, f1
	addi r2, r2, -1
	fsub f2, f1, f3
	fsel f3, f2, f3, f1
	cmpwi cr0, r2, -1
	fmr f16, f0
	fmr f17, f3
	bgt .LBB_main_19	; no_exit.0.i

llvm-svn: 16764

t:
        mov %EDX, DWORD PTR [%ESP + 4]
        mov %ECX, 2
        mov %EAX, %EDX
        sar %EDX, 31
        idiv %ECX
        mov %EAX, %EDX
        ret

Generate:
t:
        mov %ECX, DWORD PTR [%ESP + 4]
***     mov %EAX, %ECX
        cdq
        and %ECX, 1
        xor %ECX, %EDX
        sub %ECX, %EDX
***     mov %EAX, %ECX
        ret

Note that the two marked moves are redundant, and should be eliminated by the
register allocator, but aren't.

Compare this to GCC, which generates:

t:
        mov     %eax, DWORD PTR [%esp+4]
        mov     %edx, %eax
        shr     %edx, 31
        lea     %ecx, [%edx+%eax]
        and     %ecx, -2
        sub     %eax, %ecx
        ret

or ICC 8.0, which generates:

t:
        movl      4(%esp), %ecx                                 #3.5
        movl      $-2147483647, %eax                            #3.25
        imull     %ecx                                          #3.25
        movl      %ecx, %eax                                    #3.25
        sarl      $31, %eax                                     #3.25
        addl      %ecx, %edx                                    #3.25
        subl      %edx, %eax                                    #3.25
        addl      %eax, %eax                                    #3.25
        negl      %eax                                          #3.25
        subl      %eax, %ecx                                    #3.25
        movl      %ecx, %eax                                    #3.25
        ret                                                     #3.25

We would be in great shape if not for the moves.

llvm-svn: 16763

Thanks to Jeff Cohen for pointing out my goof.

llvm-svn: 16762

s:   ;; X / 4
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %ECX, %EAX
        sar %ECX, 1
        shr %ECX, 30
        mov %EDX, %EAX
        add %EDX, %ECX
        sar %EAX, 2
        ret

When we really meant:

s:
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %ECX, %EAX
        sar %ECX, 1
        shr %ECX, 30
        add %EAX, %ECX
        sar %EAX, 2
        ret

Hey, this also reduces register pressure too :)

llvm-svn: 16761

instead of:

s:   ;; X / 2
        movl 4(%esp), %eax
        movl %eax, %ecx
        shrl $31, %ecx
        movl %eax, %edx
        addl %ecx, %edx
        sarl $1, %eax
        ret

t:   ;; X / -2
        movl 4(%esp), %eax
        movl %eax, %ecx
        shrl $31, %ecx
        movl %eax, %edx
        addl %ecx, %edx
        sarl $1, %eax
        negl %eax
        ret

Emit:

s:
        movl 4(%esp), %eax
        cmpl $-2147483648, %eax
        sbbl $-1, %eax
        sarl $1, %eax
        ret

t:
        movl 4(%esp), %eax
        cmpl $-2147483648, %eax
        sbbl $-1, %eax
        sarl $1, %eax
        negl %eax
        ret

llvm-svn: 16760

llvm-svn: 16759

an instruction if it can be hoisted to a common dominator of the block.
This implements: test/Regression/Transforms/TailDup/MergeTest.ll

llvm-svn: 16758

duplicated, even though the block it is in is duplicated.

llvm-svn: 16757

llvm-svn: 16756

llvm-svn: 16755

llvm-svn: 16728

llvm-svn: 16719

llvm-svn: 16714

of disagreeing constness.  This fixes
test/Regression/Linker/ConstantGlobals[123].ll

llvm-svn: 16692

llvm-svn: 16691

is a question.

llvm-svn: 16690

llvm-svn: 16686

llvm-svn: 16685

llvm-svn: 16682