[llvm-commits] [llvm] r45384 - /llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
Author: resistor
Date: Fri Dec 28 01:42:12 2007
New Revision: 45384
URL: http://llvm.org/viewvc/llvm-project?rev=45384&view=rev
Log:
Repair a transform that Chris noticed a bug in. Thanks to Nicholas for
pointing out my stupid mistakes when writing this patch. :-)
Modified:
llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
Modified: llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp?rev=45384&r1=45383&r2=45384&view=diff
==
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Fri Dec 28
01:42:12 2007
@@ -4835,11 +4835,18 @@
if (ConstantInt *CI = dyn_cast(Op1)) {
Value *A, *B;
-#if 0
-// (icmp cond (sub A B) 0) -> (icmp cond A B)
-if (CI->isNullValue() && match(Op0, m_Sub(m_Value(A), m_Value(B
- return new ICmpInst(I.getPredicate(), A, B);
-#endif
+// (icmp cond (sub A B) 0) -> ...
+if (CI->isNullValue() && match(Op0, m_Sub(m_Value(A), m_Value(B {
+ // (icmp cond A B) if cond is signed or equality
+ if (CmpInst::isSigned(I.getPredicate()) || I.isEquality())
+return new ICmpInst(I.getPredicate(), A, B);
+ // (icmp ne A B) if cond is ugt
+ else if (I.getPredicate() == ICmpInst::ICMP_UGT)
+return new ICmpInst(ICmpInst::ICMP_NE, A, B);
+ // (icmp eq A B) if cond is ule
+ else if (I.getPredicate() == ICmpInst::ICMP_ULE)
+return new ICmpInst(ICmpInst::ICMP_EQ, A, B);
+}
switch (I.getPredicate()) {
default: break;
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
Re: [llvm-commits] [llvm] r45384 - /llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
On Dec 27, 2007, at 11:42 PM, Owen Anderson wrote:
> URL: http://llvm.org/viewvc/llvm-project?rev=45384&view=rev
> Log:
> Repair a transform that Chris noticed a bug in. Thanks to Nicholas
> for pointing out my stupid mistakes when writing this patch. :-)
Thanks Owen! A minor tweak:
> =
> =
> =
> =
> =
> =
> =
> =
> ==
> --- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp
> (original)
> +++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Fri
> Dec 28 01:42:12 2007
> @@ -4835,11 +4835,18 @@
> if (ConstantInt *CI = dyn_cast(Op1)) {
> Value *A, *B;
>
> +// (icmp cond (sub A B) 0) -> ...
> +if (CI->isNullValue() && match(Op0, m_Sub(m_Value(A),
> m_Value(B {
> + // (icmp cond A B) if cond is signed or equality
> + if (CmpInst::isSigned(I.getPredicate()) || I.isEquality())
> +return new ICmpInst(I.getPredicate(), A, B);
> + // (icmp ne A B) if cond is ugt
> + else if (I.getPredicate() == ICmpInst::ICMP_UGT)
> +return new ICmpInst(ICmpInst::ICMP_NE, A, B);
> + // (icmp eq A B) if cond is ule
> + else if (I.getPredicate() == ICmpInst::ICMP_ULE)
> +return new ICmpInst(ICmpInst::ICMP_EQ, A, B);
> +}
Why not handle UGE and ULT? If it is handled elsewhere, please add a
comment to that effect, thanks!
-Chris
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm-gcc-4.2] r45385 - in /llvm-gcc-4.2/trunk: configure configure.in
Author: lattner
Date: Fri Dec 28 13:18:11 2007
New Revision: 45385
URL: http://llvm.org/viewvc/llvm-project?rev=45385&view=rev
Log:
when gfortran is enabled, built libgfortran, even for
darwin targets.
Modified:
llvm-gcc-4.2/trunk/configure
llvm-gcc-4.2/trunk/configure.in
Modified: llvm-gcc-4.2/trunk/configure
URL:
http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/configure?rev=45385&r1=45384&r2=45385&view=diff
==
--- llvm-gcc-4.2/trunk/configure (original)
+++ llvm-gcc-4.2/trunk/configure Fri Dec 28 13:18:11 2007
@@ -1214,7 +1214,7 @@
noconfigdirs="$noconfigdirs target-libstdc++-v3 target-libssp"
# LLVM LOCAL begin
noconfigdirs="$noconfigdirs target-boehm-gc target-libffi"
-noconfigdirs="$noconfigdirs target-libgfortran target-libjava"
+noconfigdirs="$noconfigdirs target-libjava"
noconfigdirs="$noconfigdirs target-libmudflap target-zlib"
# LLVM LOCAL end
;;
@@ -1224,7 +1224,7 @@
noconfigdirs="$noconfigdirs sim target-rda"
# LLVM LOCAL begin
noconfigdirs="$noconfigdirs target-boehm-gc target-libffi"
-noconfigdirs="$noconfigdirs target-libgfortran target-libjava"
+noconfigdirs="$noconfigdirs target-libjava"
noconfigdirs="$noconfigdirs target-libmudflap target-libobjc"
noconfigdirs="$noconfigdirs target-libssp target-libstdc++-v3 target-zlib"
# LLVM LOCAL end
@@ -1235,7 +1235,7 @@
noconfigdirs="$noconfigdirs ${libgcj}"
# LLVM LOCAL begin
noconfigdirs="$noconfigdirs target-boehm-gc target-libffi"
-noconfigdirs="$noconfigdirs target-libgfortran target-libjava"
+noconfigdirs="$noconfigdirs target-libjava"
noconfigdirs="$noconfigdirs target-libmudflap target-libobjc"
noconfigdirs="$noconfigdirs target-libssp target-libstdc++-v3 target-zlib"
# LLVM LOCAL end
Modified: llvm-gcc-4.2/trunk/configure.in
URL:
http://llvm.org/viewvc/llvm-project/llvm-gcc-4.2/trunk/configure.in?rev=45385&r1=45384&r2=45385&view=diff
==
--- llvm-gcc-4.2/trunk/configure.in (original)
+++ llvm-gcc-4.2/trunk/configure.in Fri Dec 28 13:18:11 2007
@@ -374,7 +374,7 @@
noconfigdirs="$noconfigdirs target-libstdc++-v3 target-libssp"
# LLVM LOCAL begin
noconfigdirs="$noconfigdirs target-boehm-gc target-libffi"
-noconfigdirs="$noconfigdirs target-libgfortran target-libjava"
+noconfigdirs="$noconfigdirs target-libjava"
noconfigdirs="$noconfigdirs target-libmudflap target-zlib"
# LLVM LOCAL end
;;
@@ -384,7 +384,7 @@
noconfigdirs="$noconfigdirs sim target-rda"
# LLVM LOCAL begin
noconfigdirs="$noconfigdirs target-boehm-gc target-libffi"
-noconfigdirs="$noconfigdirs target-libgfortran target-libjava"
+noconfigdirs="$noconfigdirs target-libjava"
noconfigdirs="$noconfigdirs target-libmudflap target-libobjc"
noconfigdirs="$noconfigdirs target-libssp target-libstdc++-v3 target-zlib"
# LLVM LOCAL end
@@ -395,7 +395,7 @@
noconfigdirs="$noconfigdirs ${libgcj}"
# LLVM LOCAL begin
noconfigdirs="$noconfigdirs target-boehm-gc target-libffi"
-noconfigdirs="$noconfigdirs target-libgfortran target-libjava"
+noconfigdirs="$noconfigdirs target-libjava"
noconfigdirs="$noconfigdirs target-libmudflap target-libobjc"
noconfigdirs="$noconfigdirs target-libssp target-libstdc++-v3 target-zlib"
# LLVM LOCAL end
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45386 - /llvm/trunk/test/Transforms/InstCombine/2007-12-28-IcmpSub2.ll
Author: resistor
Date: Fri Dec 28 15:08:43 2007
New Revision: 45386
URL: http://llvm.org/viewvc/llvm-project?rev=45386&view=rev
Log:
Add a testcase for my recent InstCombine fix, written by Nicholas.
Added:
llvm/trunk/test/Transforms/InstCombine/2007-12-28-IcmpSub2.ll
Added: llvm/trunk/test/Transforms/InstCombine/2007-12-28-IcmpSub2.ll
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/2007-12-28-IcmpSub2.ll?rev=45386&view=auto
==
--- llvm/trunk/test/Transforms/InstCombine/2007-12-28-IcmpSub2.ll (added)
+++ llvm/trunk/test/Transforms/InstCombine/2007-12-28-IcmpSub2.ll Fri Dec 28
15:08:43 2007
@@ -0,0 +1,89 @@
+; RUN: llvm-as < %s | opt -mem2reg -instcombine | llvm-dis | grep "ret i32 1"
| count 8
+
+define i32 @test1() {
+entry:
+ %z = alloca i32
+ store i32 0, i32* %z
+ %tmp = load i32* %z
+ %sub = sub i32 %tmp, 1
+ %cmp = icmp ule i32 %sub, 0
+ %retval = select i1 %cmp, i32 0, i32 1
+ ret i32 %retval
+}
+
+define i32 @test2() {
+entry:
+ %z = alloca i32
+ store i32 0, i32* %z
+ %tmp = load i32* %z
+ %sub = sub i32 %tmp, 1
+ %cmp = icmp ugt i32 %sub, 0
+ %retval = select i1 %cmp, i32 1, i32 0
+ ret i32 %retval
+}
+
+define i32 @test3() {
+entry:
+ %z = alloca i32
+ store i32 0, i32* %z
+ %tmp = load i32* %z
+ %sub = sub i32 %tmp, 1
+ %cmp = icmp slt i32 %sub, 0
+ %retval = select i1 %cmp, i32 1, i32 0
+ ret i32 %retval
+}
+
+define i32 @test4() {
+entry:
+ %z = alloca i32
+ store i32 0, i32* %z
+ %tmp = load i32* %z
+ %sub = sub i32 %tmp, 1
+ %cmp = icmp sle i32 %sub, 0
+ %retval = select i1 %cmp, i32 1, i32 0
+ ret i32 %retval
+}
+
+define i32 @test5() {
+entry:
+ %z = alloca i32
+ store i32 0, i32* %z
+ %tmp = load i32* %z
+ %sub = sub i32 %tmp, 1
+ %cmp = icmp sge i32 %sub, 0
+ %retval = select i1 %cmp, i32 0, i32 1
+ ret i32 %retval
+}
+
+define i32 @test6() {
+entry:
+ %z = alloca i32
+ store i32 0, i32* %z
+ %tmp = load i32* %z
+ %sub = sub i32 %tmp, 1
+ %cmp = icmp sgt i32 %sub, 0
+ %retval = select i1 %cmp, i32 0, i32 1
+ ret i32 %retval
+}
+
+define i32 @test7() {
+entry:
+ %z = alloca i32
+ store i32 0, i32* %z
+ %tmp = load i32* %z
+ %sub = sub i32 %tmp, 1
+ %cmp = icmp eq i32 %sub, 0
+ %retval = select i1 %cmp, i32 0, i32 1
+ ret i32 %retval
+}
+
+define i32 @test8() {
+entry:
+ %z = alloca i32
+ store i32 0, i32* %z
+ %tmp = load i32* %z
+ %sub = sub i32 %tmp, 1
+ %cmp = icmp ne i32 %sub, 0
+ %retval = select i1 %cmp, i32 1, i32 0
+ ret i32 %retval
+}
\ No newline at end of file
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45387 - /llvm/trunk/lib/Target/X86/README.txt
Author: lattner
Date: Fri Dec 28 15:50:40 2007
New Revision: 45387
URL: http://llvm.org/viewvc/llvm-project?rev=45387&view=rev
Log:
add a note.
Modified:
llvm/trunk/lib/Target/X86/README.txt
Modified: llvm/trunk/lib/Target/X86/README.txt
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/README.txt?rev=45387&r1=45386&r2=45387&view=diff
==
--- llvm/trunk/lib/Target/X86/README.txt (original)
+++ llvm/trunk/lib/Target/X86/README.txt Fri Dec 28 15:50:40 2007
@@ -1585,3 +1585,35 @@
if the flags of the xor are dead.
//===-===//
+
+This testcase misses a read/modify/write opportunity (from PR1425):
+
+void vertical_decompose97iH1(int *b0, int *b1, int *b2, int width){
+int i;
+for(i=0; i>0;
+}
+
+We compile it down to:
+
+LBB1_2:# bb
+ movl(%esi,%edi,4), %ebx
+ addl(%ecx,%edi,4), %ebx
+ addl(%edx,%edi,4), %ebx
+ movl%ebx, (%ecx,%edi,4)
+ incl%edi
+ cmpl%eax, %edi
+ jne LBB1_2 # bb
+
+the inner loop should add to the memory location (%ecx,%edi,4), saving
+a mov. Something like:
+
+movl(%esi,%edi,4), %ebx
+addl(%edx,%edi,4), %ebx
+addl%ebx, (%ecx,%edi,4)
+
+Additionally, LSR should rewrite the exit condition of the loop to use
+a stride-4 IV, would would allow all the scales in the loop to go away.
+This would result in smaller code and more efficient microops.
+
+//===-===//
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45388 - /llvm/trunk/lib/Target/README.txt
Author: lattner
Date: Fri Dec 28 16:30:05 2007
New Revision: 45388
URL: http://llvm.org/viewvc/llvm-project?rev=45388&view=rev
Log:
add a note.
Modified:
llvm/trunk/lib/Target/README.txt
Modified: llvm/trunk/lib/Target/README.txt
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/README.txt?rev=45388&r1=45387&r2=45388&view=diff
==
--- llvm/trunk/lib/Target/README.txt (original)
+++ llvm/trunk/lib/Target/README.txt Fri Dec 28 16:30:05 2007
@@ -480,6 +480,38 @@
}
}
+//===-===//
+
+We should investigate an instruction sinking pass. Consider this silly
+example in pic mode:
+
+#include
+void foo(int x) {
+ assert(x);
+ //...
+}
+
+we compile this to:
+_foo:
+ subl$28, %esp
+ call"L1$pb"
+"L1$pb":
+ popl%eax
+ cmpl$0, 32(%esp)
+ je LBB1_2 # cond_true
+LBB1_1:# return
+ # ...
+ addl$28, %esp
+ ret
+LBB1_2:# cond_true
+...
+
+The PIC base computation (call+popl) is only used on one path through the
+code, but is currently always computed in the entry block. It would be
+better to sink the picbase computation down into the block for the
+assertion, as it is the only one that uses it. This happens for a lot of
+code with early outs.
+In this case, whole-function-isel would also handle this.
//===-===//
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45389 - /llvm/trunk/test/Transforms/SimplifyCFG/2007-12-21-Crash.ll
Author: dpatel
Date: Fri Dec 28 16:59:48 2007
New Revision: 45389
URL: http://llvm.org/viewvc/llvm-project?rev=45389&view=rev
Log:
Test -simplifycfg only.
Modified:
llvm/trunk/test/Transforms/SimplifyCFG/2007-12-21-Crash.ll
Modified: llvm/trunk/test/Transforms/SimplifyCFG/2007-12-21-Crash.ll
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/SimplifyCFG/2007-12-21-Crash.ll?rev=45389&r1=45388&r2=45389&view=diff
==
--- llvm/trunk/test/Transforms/SimplifyCFG/2007-12-21-Crash.ll (original)
+++ llvm/trunk/test/Transforms/SimplifyCFG/2007-12-21-Crash.ll Fri Dec 28
16:59:48 2007
@@ -1,58 +1,35 @@
-; RUN: llvm-as < %s | opt -std-compile-opts -disable-output
-
+;RUN: llvm-as < %s | opt -simplifycfg -disable-output
target datalayout =
"e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
-define i32 @bork() {
+define i32 @bork() nounwind {
entry:
- %retval = alloca i32; [#uses=2]
- %opt = alloca i32 ; [#uses=3]
- %undo = alloca i32 ; [#uses=3]
- %tmp = alloca i32 ; [#uses=3]
- %"alloca point" = bitcast i32 0 to i32 ; [#uses=0]
- store i32 0, i32* %undo, align 4
- br label %bb5
-
-bb:; preds = %bb5
- %tmp1 = load i32* %opt, align 4 ; [#uses=1]
- switch i32 %tmp1, label %bb4 [
-i32 102, label %bb3
-i32 110, label %bb2
- ]
+ br label %bb5.outer
-bb2: ; preds = %bb
- store i32 1, i32* %undo, align 4
- br label %bb3
+bb5.outer.loopexit:; preds = %bb5
+ br label %bb5.outer
-bb3: ; preds = %bb2, %bb
+bb5.outer: ; preds = %bb5.outer.loopexit, %entry
+ %undo.0.ph = phi i32 [ 0, %entry ], [ 1, %bb5.outer.loopexit ]
; [#uses=1]
br label %bb5
-bb4: ; preds = %bb
- store i32 258, i32* %tmp, align 4
- br label %bb13
-
-bb5: ; preds = %bb3, %entry
- %tmp6 = call i32 (...)* @foo( ) nounwind;
[#uses=1]
- store i32 %tmp6, i32* %opt, align 4
- %tmp7 = load i32* %opt, align 4 ; [#uses=1]
- %tmp8 = icmp ne i32 %tmp7, -1 ; [#uses=1]
- %tmp89 = zext i1 %tmp8 to i8; [#uses=1]
- %toBool = icmp ne i8 %tmp89, 0 ; [#uses=1]
- br i1 %toBool, label %bb, label %bb10
+bb5: ; preds = %bb5, %bb5.outer
+ %tmp6 = tail call i32 (...)* @foo( ) nounwind ;
[#uses=1]
+ switch i32 %tmp6, label %bb13 [
+i32 -1, label %bb10
+i32 102, label %bb5
+i32 110, label %bb5.outer.loopexit
+ ]
bb10: ; preds = %bb5
- %tmp11 = load i32* %undo, align 4 ; [#uses=1]
- %tmp12 = call i32 (...)* @bar( i32 %tmp11 ) nounwind;
[#uses=0]
- store i32 1, i32* %tmp, align 4
- br label %bb13
-
-bb13: ; preds = %bb10, %bb4
- %tmp14 = load i32* %tmp, align 4; [#uses=1]
- store i32 %tmp14, i32* %retval, align 4
- br label %return
-
-return:; preds = %bb13
- %retval15 = load i32* %retval ; [#uses=1]
- ret i32 %retval15
+ %tmp12 = tail call i32 (...)* @bar( i32 %undo.0.ph ) nounwind
; [#uses=0]
+ br label %UnifiedReturnBlock
+
+bb13: ; preds = %bb5
+ br label %UnifiedReturnBlock
+
+UnifiedReturnBlock:; preds = %bb13, %bb10
+ %UnifiedRetVal = phi i32 [ 1, %bb10 ], [ 258, %bb13 ] ;
[#uses=1]
+ ret i32 %UnifiedRetVal
}
declare i32 @foo(...)
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
Re: [llvm-commits] [llvm] r45305 - in /llvm/trunk: lib/Transforms/Utils/SimplifyCFG.cpp test/Transforms/SimplifyCFG/2007-12-21-Crash.ll
On Dec 22, 2007, at 10:45 AM, Chris Lattner wrote: > Please just make the testcase use -simplifycfg instead of -std- > compile- > opts. The contents of -std-compile-opts can change over time and we > want to make sure that this is testing the right thing. Thanks! OK. Done. - Devang ___ llvm-commits mailing list llvm-commits@cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45391 - /llvm/trunk/test/Transforms/InstCombine/deadcode.ll
Author: lattner
Date: Fri Dec 28 18:57:06 2007
New Revision: 45391
URL: http://llvm.org/viewvc/llvm-project?rev=45391&view=rev
Log:
upgrade this test
Modified:
llvm/trunk/test/Transforms/InstCombine/deadcode.ll
Modified: llvm/trunk/test/Transforms/InstCombine/deadcode.ll
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/deadcode.ll?rev=45391&r1=45390&r2=45391&view=diff
==
--- llvm/trunk/test/Transforms/InstCombine/deadcode.ll (original)
+++ llvm/trunk/test/Transforms/InstCombine/deadcode.ll Fri Dec 28 18:57:06 2007
@@ -1,13 +1,14 @@
-; RUN: llvm-upgrade < %s | llvm-as | opt -instcombine | llvm-dis | \
-; RUN: grep {ret i32 %A}
+; RUN: llvm-as < %s | opt -instcombine | llvm-dis | grep {ret i32 %A}
-int %test(int %A) {
- %X = or bool false, false
- br bool %X, label %T, label %C
-T:
- %B = add int %A, 1
- br label %C
-C:
- %C = phi int [%B, %T], [%A, %0]
- ret int %C
+define i32 @test(i32 %A) {
+ %X = or i1 false, false
+ br i1 %X, label %T, label %C
+
+T: ; preds = %0
+ %B = add i32 %A, 1
+ br label %C
+
+C: ; preds = %T, %0
+ %C.upgrd.1 = phi i32 [ %B, %T ], [ %A, %0 ]
+ ret i32 %C.upgrd.1
}
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45392 - in /llvm/trunk: lib/Transforms/Utils/Local.cpp test/Transforms/InstCombine/deadcode.ll
Author: lattner
Date: Fri Dec 28 18:59:12 2007
New Revision: 45392
URL: http://llvm.org/viewvc/llvm-project?rev=45392&view=rev
Log:
dead calls to llvm.stacksave can be deleted, even though they
have potential side-effects.
Modified:
llvm/trunk/lib/Transforms/Utils/Local.cpp
llvm/trunk/test/Transforms/InstCombine/deadcode.ll
Modified: llvm/trunk/lib/Transforms/Utils/Local.cpp
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Utils/Local.cpp?rev=45392&r1=45391&r2=45392&view=diff
==
--- llvm/trunk/lib/Transforms/Utils/Local.cpp (original)
+++ llvm/trunk/lib/Transforms/Utils/Local.cpp Fri Dec 28 18:59:12 2007
@@ -17,6 +17,7 @@
#include "llvm/DerivedTypes.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
+#include "llvm/IntrinsicInst.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
@@ -173,8 +174,16 @@
bool llvm::isInstructionTriviallyDead(Instruction *I) {
if (!I->use_empty() || isa(I)) return false;
- if (!I->mayWriteToMemory()) return true;
+ if (!I->mayWriteToMemory())
+return true;
+ // Special case intrinsics that "may write to memory" but can be deleted when
+ // dead.
+ if (IntrinsicInst *II = dyn_cast(I))
+// Safe to delete llvm.stacksave if dead.
+if (II->getIntrinsicID() == Intrinsic::stacksave)
+ return true;
+
return false;
}
Modified: llvm/trunk/test/Transforms/InstCombine/deadcode.ll
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/deadcode.ll?rev=45392&r1=45391&r2=45392&view=diff
==
--- llvm/trunk/test/Transforms/InstCombine/deadcode.ll (original)
+++ llvm/trunk/test/Transforms/InstCombine/deadcode.ll Fri Dec 28 18:59:12 2007
@@ -1,4 +1,5 @@
; RUN: llvm-as < %s | opt -instcombine | llvm-dis | grep {ret i32 %A}
+; RUN: llvm-as < %s | opt -die | llvm-dis | not grep call.*llvm.stacksave
define i32 @test(i32 %A) {
%X = or i1 false, false
@@ -12,3 +13,12 @@
%C.upgrd.1 = phi i32 [ %B, %T ], [ %A, %0 ]
ret i32 %C.upgrd.1
}
+
+define i32* @test2(i32 %width) {
+ %tmp = call i8* @llvm.stacksave( )
+%tmp14 = alloca i32, i32 %width
+ ret i32* %tmp14
+}
+
+declare i8* @llvm.stacksave()
+
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45393 - /llvm/trunk/lib/Target/README.txt
Author: lattner Date: Fri Dec 28 19:05:01 2007 New Revision: 45393 URL: http://llvm.org/viewvc/llvm-project?rev=45393&view=rev Log: expand note. Modified: llvm/trunk/lib/Target/README.txt Modified: llvm/trunk/lib/Target/README.txt URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/README.txt?rev=45393&r1=45392&r2=45393&view=diff == --- llvm/trunk/lib/Target/README.txt (original) +++ llvm/trunk/lib/Target/README.txt Fri Dec 28 19:05:01 2007 @@ -512,6 +512,10 @@ assertion, as it is the only one that uses it. This happens for a lot of code with early outs. +Another example is loads of arguments, which are usually emitted into the +entry block on targets like x86. If not used in all paths through a +function, they should be sunk into the ones that do. + In this case, whole-function-isel would also handle this. //===-===// ___ llvm-commits mailing list llvm-commits@cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45397 - /llvm/trunk/lib/Target/X86/README.txt
Author: lattner
Date: Fri Dec 28 23:51:58 2007
New Revision: 45397
URL: http://llvm.org/viewvc/llvm-project?rev=45397&view=rev
Log:
add a note
Modified:
llvm/trunk/lib/Target/X86/README.txt
Modified: llvm/trunk/lib/Target/X86/README.txt
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/README.txt?rev=45397&r1=45396&r2=45397&view=diff
==
--- llvm/trunk/lib/Target/X86/README.txt (original)
+++ llvm/trunk/lib/Target/X86/README.txt Fri Dec 28 23:51:58 2007
@@ -1612,8 +1612,48 @@
addl(%edx,%edi,4), %ebx
addl%ebx, (%ecx,%edi,4)
-Additionally, LSR should rewrite the exit condition of the loop to use
+Here is another interesting example:
+
+void vertical_compose97iH1(int *b0, int *b1, int *b2, int width){
+int i;
+for(i=0; i>0;
+}
+
+We miss the r/m/w opportunity here by using 2 subs instead of an add+sub[mem]:
+
+LBB9_2:# bb
+ movl(%ecx,%edi,4), %ebx
+ subl(%esi,%edi,4), %ebx
+ subl(%edx,%edi,4), %ebx
+ movl%ebx, (%ecx,%edi,4)
+ incl%edi
+ cmpl%eax, %edi
+ jne LBB9_2 # bb
+
+Additionally, LSR should rewrite the exit condition of these loops to use
a stride-4 IV, would would allow all the scales in the loop to go away.
This would result in smaller code and more efficient microops.
//===-===//
+
+We should be smarter about conversion from fpstack to XMM regs.
+
+double foo();
+void bar(double *P) { *P = foo(); }
+
+We compile that to:
+
+_bar:
+ subl$12, %esp
+ callL_foo$stub
+ fstpl (%esp)
+ movl16(%esp), %eax
+ movsd (%esp), %xmm0
+ movsd %xmm0, (%eax)
+ addl$12, %esp
+ ret
+
+for example. The magic to/from the stack is unneeded.
+
+//===-===//
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45398 - /llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
Author: lattner
Date: Sat Dec 29 00:26:16 2007
New Revision: 45398
URL: http://llvm.org/viewvc/llvm-project?rev=45398&view=rev
Log:
Delete a store whose input is a load from the same pointer:
x = load p
store x -> p
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=45398&r1=45397&r2=45398&view=diff
==
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Sat Dec 29 00:26:16 2007
@@ -4223,7 +4223,7 @@
if (CombineToPreIndexedLoadStore(N) || CombineToPostIndexedLoadStore(N))
return SDOperand(N, 0);
- // FIXME: is there such a think as a truncating indexed store?
+ // FIXME: is there such a thing as a truncating indexed store?
if (ST->isTruncatingStore() && ST->getAddressingMode() == ISD::UNINDEXED &&
MVT::isInteger(Value.getValueType())) {
// See if we can simplify the input to this truncstore with knowledge that
@@ -4243,6 +4243,17 @@
return SDOperand(N, 0);
}
+ // If this is a load followed by a store to the same location, then the store
+ // is dead/noop.
+ if (LoadSDNode *Ld = dyn_cast(Value)) {
+if (Chain.Val == Ld && Ld->getBasePtr() == Ptr &&
+ST->getAddressingMode() == ISD::UNINDEXED &&
+ST->getStoredVT() == Ld->getLoadedVT()) {
+ // The store is dead, remove it.
+ return Chain;
+}
+ }
+
return SDOperand();
}
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
Re: [llvm-commits] [llvm] r45398 - /llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
On Dec 28, 2007, at 10:26 PM, Chris Lattner wrote:
> Author: lattner
> Date: Sat Dec 29 00:26:16 2007
> New Revision: 45398
>
> URL: http://llvm.org/viewvc/llvm-project?rev=45398&view=rev
> Log:
> Delete a store whose input is a load from the same pointer:
> x = load p
> store x -> p
This is unsafe for volatile. I don't see a check?
>
> Modified:
>llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
>
> Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
> URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=45398&r1=45397&r2=45398&view=diff
>
> =
> =
> =
> =
> =
> =
> =
> =
> ==
> --- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
> +++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Sat Dec 29
> 00:26:16 2007
> @@ -4223,7 +4223,7 @@
> if (CombineToPreIndexedLoadStore(N) ||
> CombineToPostIndexedLoadStore(N))
> return SDOperand(N, 0);
>
> - // FIXME: is there such a think as a truncating indexed store?
> + // FIXME: is there such a thing as a truncating indexed store?
> if (ST->isTruncatingStore() && ST->getAddressingMode() ==
> ISD::UNINDEXED &&
> MVT::isInteger(Value.getValueType())) {
> // See if we can simplify the input to this truncstore with
> knowledge that
> @@ -4243,6 +4243,17 @@
> return SDOperand(N, 0);
> }
>
> + // If this is a load followed by a store to the same location,
> then the store
> + // is dead/noop.
> + if (LoadSDNode *Ld = dyn_cast(Value)) {
> +if (Chain.Val == Ld && Ld->getBasePtr() == Ptr &&
> +ST->getAddressingMode() == ISD::UNINDEXED &&
> +ST->getStoredVT() == Ld->getLoadedVT()) {
> + // The store is dead, remove it.
> + return Chain;
> +}
> + }
> +
> return SDOperand();
> }
>
>
>
> ___
> llvm-commits mailing list
> llvm-commits@cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45399 - in /llvm/trunk: lib/Target/X86/README.txt lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/fp-stack-ret-store.ll
Author: lattner
Date: Sat Dec 29 00:41:28 2007
New Revision: 45399
URL: http://llvm.org/viewvc/llvm-project?rev=45399&view=rev
Log:
avoid going through a stack slot to convert from fpstack to xmm reg
if we are just going to store it back anyway. This improves things
like:
double foo();
void bar(double *P) { *P = foo(); }
Added:
llvm/trunk/test/CodeGen/X86/fp-stack-ret-store.ll
Modified:
llvm/trunk/lib/Target/X86/README.txt
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
Modified: llvm/trunk/lib/Target/X86/README.txt
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/README.txt?rev=45399&r1=45398&r2=45399&view=diff
==
--- llvm/trunk/lib/Target/X86/README.txt (original)
+++ llvm/trunk/lib/Target/X86/README.txt Sat Dec 29 00:41:28 2007
@@ -1636,24 +1636,3 @@
This would result in smaller code and more efficient microops.
//===-===//
-
-We should be smarter about conversion from fpstack to XMM regs.
-
-double foo();
-void bar(double *P) { *P = foo(); }
-
-We compile that to:
-
-_bar:
- subl$12, %esp
- callL_foo$stub
- fstpl (%esp)
- movl16(%esp), %eax
- movsd (%esp), %xmm0
- movsd %xmm0, (%eax)
- addl$12, %esp
- ret
-
-for example. The magic to/from the stack is unneeded.
-
-//===-===//
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=45399&r1=45398&r2=45399&view=diff
==
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Sat Dec 29 00:41:28 2007
@@ -33,7 +33,6 @@
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SSARegMap.h"
#include "llvm/Support/MathExtras.h"
-#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/ADT/SmallSet.h"
@@ -812,7 +811,6 @@
CCState CCInfo(CallingConv, isVarArg, getTargetMachine(), RVLocs);
CCInfo.AnalyzeCallResult(TheCall, RetCC_X86);
-
SmallVector ResultVals;
// Copy all of the result registers out of their specified physreg.
@@ -838,17 +836,50 @@
// an XMM register.
if ((X86ScalarSSEf32 && RVLocs[0].getValVT() == MVT::f32) ||
(X86ScalarSSEf64 && RVLocs[0].getValVT() == MVT::f64)) {
+ SDOperand StoreLoc;
+ const Value *SrcVal = 0;
+ int SrcValOffset = 0;
+
+ // Determine where to store the value. If the call result is directly
+ // used by a store, see if we can store directly into the location. In
+ // this case, we'll end up producing a fst + movss[load] + movss[store]
to
+ // the same location, and the two movss's will be nuked as dead. This
+ // optimizes common things like "*D = atof(..)" to not need an
+ // intermediate stack slot.
+ if (SDOperand(TheCall, 0).hasOneUse() &&
+ SDOperand(TheCall, 1).hasOneUse()) {
+// Ok, we have one use of the value and one use of the chain. See if
+// they are the same node: a store.
+if (StoreSDNode *N = dyn_cast(*TheCall->use_begin())) {
+ if (N->getChain().Val == TheCall && N->getValue().Val == TheCall &&
+ !N->isVolatile() && !N->isTruncatingStore() &&
+ N->getAddressingMode() == ISD::UNINDEXED) {
+StoreLoc = N->getBasePtr();
+SrcVal = N->getSrcValue();
+SrcValOffset = N->getSrcValueOffset();
+ }
+}
+ }
+
+ // If we weren't able to optimize the result, just create a temporary
+ // stack slot.
+ if (StoreLoc.Val == 0) {
+MachineFunction &MF = DAG.getMachineFunction();
+int SSFI = MF.getFrameInfo()->CreateStackObject(8, 8);
+StoreLoc = DAG.getFrameIndex(SSFI, getPointerTy());
+ }
+
// FIXME: Currently the FST is flagged to the FP_GET_RESULT. This
// shouldn't be necessary except that RFP cannot be live across
- // multiple blocks. When stackifier is fixed, they can be uncoupled.
- MachineFunction &MF = DAG.getMachineFunction();
- int SSFI = MF.getFrameInfo()->CreateStackObject(8, 8);
- SDOperand StackSlot = DAG.getFrameIndex(SSFI, getPointerTy());
+ // multiple blocks (which could happen if a select gets lowered into
+ // multiple blocks and scheduled in between them). When stackifier is
+ // fixed, they can be uncoupled.
SDOperand Ops[] = {
-Chain, RetVal, StackSlot, DAG.getValueType(RVLocs[0].getValVT()),
InFlag
+Chain, RetVal, StoreLoc, DAG.getValueType(RVLocs[0].getValVT()), InFlag
};
Chain = DAG.getNode(X86ISD::FST, MVT::Other, Ops, 5);
- Re
[llvm-commits] [llvm] r45400 - /llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
Author: lattner Date: Sat Dec 29 00:55:23 2007 New Revision: 45400 URL: http://llvm.org/viewvc/llvm-project?rev=45400&view=rev Log: don't fold fp_round(fp_extend(load)) -> fp_round(extload) Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=45400&r1=45399&r2=45400&view=diff == --- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original) +++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Sat Dec 29 00:55:23 2007 @@ -3672,6 +3672,10 @@ ConstantFPSDNode *N0CFP = dyn_cast(N0); MVT::ValueType VT = N->getValueType(0); + // If this is fp_round(fpextend), don't fold it, allow ourselves to be folded. + if (N->hasOneUse() && (*N->use_begin())->getOpcode() == ISD::FP_ROUND) +return SDOperand(); + // fold (fp_extend c1fp) -> c1fp if (N0CFP && VT != MVT::ppcf128) return DAG.getNode(ISD::FP_EXTEND, VT, N0); ___ llvm-commits mailing list llvm-commits@cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45401 - in /llvm/trunk: lib/Target/X86/X86ISelLowering.cpp test/CodeGen/X86/fp-stack-ret-store.ll
Author: lattner
Date: Sat Dec 29 00:57:38 2007
New Revision: 45401
URL: http://llvm.org/viewvc/llvm-project?rev=45401&view=rev
Log:
Codegen:
as:
_bar:
pushl %esi
subl$8, %esp
movl16(%esp), %esi
callL_foo$stub
fstps (%esi)
addl$8, %esp
popl%esi
#FP_REG_KILL
ret
instead of:
_bar:
pushl %esi
subl$8, %esp
movl16(%esp), %esi
callL_foo$stub
fstpl (%esi)
cvtsd2ss(%esi), %xmm0
movss %xmm0, (%esi)
addl$8, %esp
popl%esi
#FP_REG_KILL
ret
Modified:
llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
llvm/trunk/test/CodeGen/X86/fp-stack-ret-store.ll
Modified: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelLowering.cpp?rev=45401&r1=45400&r2=45401&view=diff
==
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp Sat Dec 29 00:57:38 2007
@@ -839,6 +839,7 @@
SDOperand StoreLoc;
const Value *SrcVal = 0;
int SrcValOffset = 0;
+ MVT::ValueType RetStoreVT = RVLocs[0].getValVT();
// Determine where to store the value. If the call result is directly
// used by a store, see if we can store directly into the location. In
@@ -848,15 +849,34 @@
// intermediate stack slot.
if (SDOperand(TheCall, 0).hasOneUse() &&
SDOperand(TheCall, 1).hasOneUse()) {
+// In addition to direct uses, we also support a FP_ROUND that uses the
+// value, if it is directly stored somewhere.
+SDNode *User = *TheCall->use_begin();
+if (User->getOpcode() == ISD::FP_ROUND && User->hasOneUse())
+ User = *User->use_begin();
+
// Ok, we have one use of the value and one use of the chain. See if
// they are the same node: a store.
-if (StoreSDNode *N = dyn_cast(*TheCall->use_begin())) {
- if (N->getChain().Val == TheCall && N->getValue().Val == TheCall &&
+if (StoreSDNode *N = dyn_cast(User)) {
+ // Verify that the value being stored is either the call or a
+ // truncation of the call.
+ SDNode *StoreVal = N->getValue().Val;
+ if (StoreVal == TheCall)
+; // ok.
+ else if (StoreVal->getOpcode() == ISD::FP_ROUND &&
+ StoreVal->hasOneUse() &&
+ StoreVal->getOperand(0).Val == TheCall)
+; // ok.
+ else
+N = 0; // not ok.
+
+ if (N && N->getChain().Val == TheCall &&
!N->isVolatile() && !N->isTruncatingStore() &&
N->getAddressingMode() == ISD::UNINDEXED) {
StoreLoc = N->getBasePtr();
SrcVal = N->getSrcValue();
SrcValOffset = N->getSrcValueOffset();
+RetStoreVT = N->getValue().getValueType();
}
}
}
@@ -875,12 +895,17 @@
// multiple blocks and scheduled in between them). When stackifier is
// fixed, they can be uncoupled.
SDOperand Ops[] = {
-Chain, RetVal, StoreLoc, DAG.getValueType(RVLocs[0].getValVT()), InFlag
+Chain, RetVal, StoreLoc, DAG.getValueType(RetStoreVT), InFlag
};
Chain = DAG.getNode(X86ISD::FST, MVT::Other, Ops, 5);
- RetVal = DAG.getLoad(RVLocs[0].getValVT(), Chain,
+ RetVal = DAG.getLoad(RetStoreVT, Chain,
StoreLoc, SrcVal, SrcValOffset);
Chain = RetVal.getValue(1);
+
+ // If we optimized a truncate, then extend the result back to its desired
+ // type.
+ if (RVLocs[0].getValVT() != RetStoreVT)
+RetVal = DAG.getNode(ISD::FP_EXTEND, RVLocs[0].getValVT(), RetVal);
}
ResultVals.push_back(RetVal);
}
Modified: llvm/trunk/test/CodeGen/X86/fp-stack-ret-store.ll
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/fp-stack-ret-store.ll?rev=45401&r1=45400&r2=45401&view=diff
==
--- llvm/trunk/test/CodeGen/X86/fp-stack-ret-store.ll (original)
+++ llvm/trunk/test/CodeGen/X86/fp-stack-ret-store.ll Sat Dec 29 00:57:38 2007
@@ -13,3 +13,14 @@
}
declare double @foo(...)
+
+define void @bar2(float* %P) {
+entry:
+ %tmp = tail call double (...)* @foo2( ) ; [#uses=1]
+ %tmp1 = fptrunc double %tmp to float; [#uses=1]
+ store float %tmp1, float* %P, align 4
+ ret void
+}
+
+declare double @foo2(...)
+
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r45402 - /llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
Author: lattner
Date: Sat Dec 29 01:15:45 2007
New Revision: 45402
URL: http://llvm.org/viewvc/llvm-project?rev=45402&view=rev
Log:
make sure not to zap volatile stores, thanks a lot to Dale for noticing this!
Modified:
llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
Modified: llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp?rev=45402&r1=45401&r2=45402&view=diff
==
--- llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp Sat Dec 29 01:15:45 2007
@@ -4252,7 +4252,8 @@
if (LoadSDNode *Ld = dyn_cast(Value)) {
if (Chain.Val == Ld && Ld->getBasePtr() == Ptr &&
ST->getAddressingMode() == ISD::UNINDEXED &&
-ST->getStoredVT() == Ld->getLoadedVT()) {
+ST->getStoredVT() == Ld->getLoadedVT() &&
+!ST->isVolatile()) {
// The store is dead, remove it.
return Chain;
}
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
