[llvm-commits] [llvm] r41017 - /llvm/trunk/lib/Target/X86/X86InstrSSE.td
Author: void
Date: Sat Aug 11 04:52:53 2007
New Revision: 41017
URL: http://llvm.org/viewvc/llvm-project?rev=41017&view=rev
Log:
64-bit SSSE3 ops that use MMX registers don't require 16-byte alignment.
Make a 'memop' pattern just for them.
Modified:
llvm/trunk/lib/Target/X86/X86InstrSSE.td
Modified: llvm/trunk/lib/Target/X86/X86InstrSSE.td
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrSSE.td?rev=41017&r1=41016&r2=41017&view=diff
==
--- llvm/trunk/lib/Target/X86/X86InstrSSE.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrSSE.td Sat Aug 11 04:52:53 2007
@@ -128,14 +128,25 @@
def memopfsf64 : PatFrag<(ops node:$ptr), (f64 (memop node:$ptr))>;
def memopv4f32 : PatFrag<(ops node:$ptr), (v4f32 (memop node:$ptr))>;
def memopv2f64 : PatFrag<(ops node:$ptr), (v2f64 (memop node:$ptr))>;
-def memopv8i8 : PatFrag<(ops node:$ptr), (v8i8 (memop node:$ptr))>;
-def memopv16i8 : PatFrag<(ops node:$ptr), (v16i8 (memop node:$ptr))>;
-def memopv4i16 : PatFrag<(ops node:$ptr), (v4i16 (memop node:$ptr))>;
-def memopv8i16 : PatFrag<(ops node:$ptr), (v8i16 (memop node:$ptr))>;
-def memopv2i32 : PatFrag<(ops node:$ptr), (v2i32 (memop node:$ptr))>;
def memopv4i32 : PatFrag<(ops node:$ptr), (v4i32 (memop node:$ptr))>;
def memopv2i64 : PatFrag<(ops node:$ptr), (v2i64 (memop node:$ptr))>;
+// SSSE3 uses MMX registers for some instructions. They aren't aligned on a
+// 16-byte boundary.
+def memop64 : PatFrag<(ops node:$ptr), (ld node:$ptr), [{
+ if (LoadSDNode *LD = dyn_cast(N))
+return LD->getExtensionType() == ISD::NON_EXTLOAD &&
+ LD->getAddressingMode() == ISD::UNINDEXED &&
+ LD->getAlignment() >= 8;
+ return false;
+}]>;
+
+def memopv8i8 : PatFrag<(ops node:$ptr), (v8i8 (memop64 node:$ptr))>;
+def memopv16i8 : PatFrag<(ops node:$ptr), (v16i8 (memop64 node:$ptr))>;
+def memopv4i16 : PatFrag<(ops node:$ptr), (v4i16 (memop64 node:$ptr))>;
+def memopv8i16 : PatFrag<(ops node:$ptr), (v8i16 (memop64 node:$ptr))>;
+def memopv2i32 : PatFrag<(ops node:$ptr), (v2i32 (memop64 node:$ptr))>;
+
def bc_v4f32 : PatFrag<(ops node:$in), (v4f32 (bitconvert node:$in))>;
def bc_v2f64 : PatFrag<(ops node:$in), (v2f64 (bitconvert node:$in))>;
def bc_v16i8 : PatFrag<(ops node:$in), (v16i8 (bitconvert node:$in))>;
@@ -2284,10 +2295,14 @@
// SSSE3 Instructions
//===--===//
-// SSE3 Instruction Templates:
+// SSSE3 Instruction Templates:
//
// SS38I - SSSE3 instructions with T8 prefix.
// SS3AI - SSSE3 instructions with TA prefix.
+//
+// Note: SSSE3 instructions have 64-bit and 128-bit versions. The 64-bit
version
+// uses the MMX registers. We put those instructions here because they better
+// fit into the SSSE3 instruction category rather than the MMX category.
class SS38I o, Format F, dag outs, dag ins, string asm,
list pattern>
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r41018 - /llvm/trunk/lib/ExecutionEngine/ExecutionEngine.cpp
Author: reid
Date: Sat Aug 11 10:57:56 2007
New Revision: 41018
URL: http://llvm.org/viewvc/llvm-project?rev=41018&view=rev
Log:
Fix a comment typo noticed by Sandro Magi.
Modified:
llvm/trunk/lib/ExecutionEngine/ExecutionEngine.cpp
Modified: llvm/trunk/lib/ExecutionEngine/ExecutionEngine.cpp
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/ExecutionEngine/ExecutionEngine.cpp?rev=41018&r1=41017&r2=41018&view=diff
==
--- llvm/trunk/lib/ExecutionEngine/ExecutionEngine.cpp (original)
+++ llvm/trunk/lib/ExecutionEngine/ExecutionEngine.cpp Sat Aug 11 10:57:56 2007
@@ -336,7 +336,7 @@
/// This function converts a Constant* into a GenericValue. The interesting
/// part is if C is a ConstantExpr.
-/// @brief Get a GenericValue for a Constnat*
+/// @brief Get a GenericValue for a Constant*
GenericValue ExecutionEngine::getConstantValue(const Constant *C) {
// If its undefined, return the garbage.
if (isa(C))
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r41019 - /llvm/trunk/utils/findmisopt
Author: reid Date: Sat Aug 11 11:11:22 2007 New Revision: 41019 URL: http://llvm.org/viewvc/llvm-project?rev=41019&view=rev Log: Now that llvm-ld can accept - as input from stdin, use this feature to extract the pass arguments that it supports. Modified: llvm/trunk/utils/findmisopt Modified: llvm/trunk/utils/findmisopt URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/utils/findmisopt?rev=41019&r1=41018&r2=41019&view=diff == --- llvm/trunk/utils/findmisopt (original) +++ llvm/trunk/utils/findmisopt Sat Aug 11 11:11:22 2007 @@ -95,8 +95,8 @@ # Define the list of optimizations to run. This comprises the same set of # optimizations that opt -std-compile-opts and gccld run, in the same order. opt_switches=`llvm-as < /dev/null -o - | opt -std-compile-opts -disable-output -debug-pass=Arguments 2>&1 | sed 's/Pass Arguments: //'` -gccld_switches="-internalize -ipsccp -globalopt -constmerge -deadargelim -inline -prune-eh -globalopt -globaldce -argpromotion -instcombine -predsimplify -scalarrepl -globalsmodref-aa -licm -load-vn -gcse -dse -instcombine -simplifycfg -verify" -all_switches="$opt_switches $gccld_switches" +ld_switches=`llvm-as < /dev/null -o - | llvm-ld - -std-link-opts -disable-output -debug-pass=Arguments 2>&1 | sed 's/Pass Arguments: //'` +all_switches="$opt_switches $ld_switches" echo "Passes : $all_switches" # Current set of switches is empty ___ llvm-commits mailing list llvm-commits@cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r41020 - /llvm/trunk/lib/Target/X86/README.txt
Author: lattner
Date: Sat Aug 11 13:16:46 2007
New Revision: 41020
URL: http://llvm.org/viewvc/llvm-project?rev=41020&view=rev
Log:
With evan's explicit flag representation, hopefully we will finally be
able to 3-addressify away stuff like this:
movl%ecx, %eax
decl%eax
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=41020&r1=41019&r2=41020&view=diff
==
--- llvm/trunk/lib/Target/X86/README.txt (original)
+++ llvm/trunk/lib/Target/X86/README.txt Sat Aug 11 13:16:46 2007
@@ -959,7 +959,7 @@
//===-===//
We need to teach the codegen to convert two-address INC instructions to LEA
-when the flags are dead. For example, on X86-64, compile:
+when the flags are dead (likewise dec). For example, on X86-64, compile:
int foo(int A, int B) {
return A+1;
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r41021 - /llvm/trunk/lib/Target/X86/README.txt
Author: lattner
Date: Sat Aug 11 13:19:07 2007
New Revision: 41021
URL: http://llvm.org/viewvc/llvm-project?rev=41021&view=rev
Log:
expand 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=41021&r1=41020&r2=41021&view=diff
==
--- llvm/trunk/lib/Target/X86/README.txt (original)
+++ llvm/trunk/lib/Target/X86/README.txt Sat Aug 11 13:19:07 2007
@@ -559,7 +559,25 @@
//===-===//
-Implement CTTZ, CTLZ with bsf and bsr.
+Implement CTTZ, CTLZ with bsf and bsr. GCC produces:
+
+int ctz_(unsigned X) { return __builtin_ctz(X); }
+int clz_(unsigned X) { return __builtin_clz(X); }
+int ffs_(unsigned X) { return __builtin_ffs(X); }
+
+_ctz_:
+bsfl4(%esp), %eax
+ret
+_clz_:
+bsrl4(%esp), %eax
+xorl$31, %eax
+ret
+_ffs_:
+movl$-1, %edx
+bsfl4(%esp), %eax
+cmove %edx, %eax
+addl$1, %eax
+ret
//===-===//
@@ -1077,8 +1095,9 @@
define <4 x float> @test1(<4 x float> %V) {
entry:
-%tmp8 = shufflevector <4 x float> %V, <4 x float> undef, <4 x i32> <
i32 3, i32 2, i32 1, i32 0 > ; <<4 x float>> [#uses=1]
-%add = add <4 x float> %tmp8, %V; <<4 x float>>
[#uses=1]
+%tmp8 = shufflevector <4 x float> %V, <4 x float> undef,
+<4 x i32> < i32 3, i32 2, i32 1, i32 0
>
+%add = add <4 x float> %tmp8, %V
ret <4 x float> %add
}
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r41022 - in /llvm/trunk: lib/Transforms/Scalar/InstructionCombining.cpp test/Transforms/InstCombine/load2.ll
Author: lattner
Date: Sat Aug 11 13:48:48 2007
New Revision: 41022
URL: http://llvm.org/viewvc/llvm-project?rev=41022&view=rev
Log:
Transform a load from an undef/zero global into an undef/global even if we
have complex pointer manipulation going on. This allows us to compile
stuff like this:
__m128i foo(__m128i x){
static const unsigned int c_0[4] = { 0, 0, 0, 0 };
__m128i v_Zero = _mm_loadu_si128((__m128i*)c_0);
x = _mm_unpacklo_epi8(x, v_Zero);
return x;
}
into:
_foo:
xorps %xmm1, %xmm1
punpcklbw %xmm1, %xmm0
ret
Added:
llvm/trunk/test/Transforms/InstCombine/load2.ll
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=41022&r1=41021&r2=41022&view=diff
==
--- llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/InstructionCombining.cpp Sat Aug 11
13:48:48 2007
@@ -8782,6 +8782,28 @@
return false;
}
+/// GetUnderlyingObject - Trace through a series of getelementptrs and bitcasts
+/// until we find the underlying object a pointer is referring to or something
+/// we don't understand. Note that the returned pointer may be offset from the
+/// input, because we ignore GEP indices.
+static Value *GetUnderlyingObject(Value *Ptr) {
+ while (1) {
+if (ConstantExpr *CE = dyn_cast(Ptr)) {
+ if (CE->getOpcode() == Instruction::BitCast ||
+ CE->getOpcode() == Instruction::GetElementPtr)
+Ptr = CE->getOperand(0);
+ else
+return Ptr;
+} else if (BitCastInst *BCI = dyn_cast(Ptr)) {
+ Ptr = BCI->getOperand(0);
+} else if (GetElementPtrInst *GEP = dyn_cast(Ptr)) {
+ Ptr = GEP->getOperand(0);
+} else {
+ return Ptr;
+}
+ }
+}
+
Instruction *InstCombiner::visitLoadInst(LoadInst &LI) {
Value *Op = LI.getOperand(0);
@@ -8860,6 +8882,17 @@
return Res;
}
}
+
+ // If this load comes from anywhere in a constant global, and if the global
+ // is all undef or zero, we know what it loads.
+ if (GlobalVariable *GV = dyn_cast(GetUnderlyingObject(Op))) {
+if (GV->isConstant() && GV->hasInitializer()) {
+ if (GV->getInitializer()->isNullValue())
+return ReplaceInstUsesWith(LI, Constant::getNullValue(LI.getType()));
+ else if (isa(GV->getInitializer()))
+return ReplaceInstUsesWith(LI, UndefValue::get(LI.getType()));
+}
+ }
if (Op->hasOneUse()) {
// Change select and PHI nodes to select values instead of addresses: this
Added: llvm/trunk/test/Transforms/InstCombine/load2.ll
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/load2.ll?rev=41022&view=auto
==
--- llvm/trunk/test/Transforms/InstCombine/load2.ll (added)
+++ llvm/trunk/test/Transforms/InstCombine/load2.ll Sat Aug 11 13:48:48 2007
@@ -0,0 +1,11 @@
+; RUN: llvm-as < %s | opt -instcombine | llvm-dis | not grep load
+
[EMAIL PROTECTED] = internal constant [4 x i32] zeroinitializer
+
+
+define <16 x i8> @foo(<2 x i64> %x) {
+entry:
+ %tmp = load <16 x i8> * bitcast ([4 x i32]* @GLOBAL to <16 x i8>*)
+ ret <16 x i8> %tmp
+}
+
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r41023 - /llvm/trunk/lib/Analysis/ConstantFolding.cpp
Author: lattner
Date: Sat Aug 11 18:49:01 2007
New Revision: 41023
URL: http://llvm.org/viewvc/llvm-project?rev=41023&view=rev
Log:
constant fold ptrtoint(inttoptr) with target data when available. This allows
us to fold the entry block of PR1602 to false instead of:
br i1 icmp eq (i32 and (i32 ptrtoint (void (%struct.S*)* inttoptr (i64
1 to void (%struct.S*)*) to i32), i32 1), i32 0), label %cond_next, label
%cond_true
Modified:
llvm/trunk/lib/Analysis/ConstantFolding.cpp
Modified: llvm/trunk/lib/Analysis/ConstantFolding.cpp
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Analysis/ConstantFolding.cpp?rev=41023&r1=41022&r2=41023&view=diff
==
--- llvm/trunk/lib/Analysis/ConstantFolding.cpp (original)
+++ llvm/trunk/lib/Analysis/ConstantFolding.cpp Sat Aug 11 18:49:01 2007
@@ -217,6 +217,23 @@
case Instruction::FCmp:
return ConstantExpr::getCompare(cast(I)->getPredicate(), Ops[0],
Ops[1]);
+ case Instruction::PtrToInt:
+// If the input is a inttoptr, eliminate the pair. This requires knowing
+// the width of a pointer, so it can't be done in ConstantExpr::getCast.
+if (ConstantExpr *CE = dyn_cast(Ops[0])) {
+ if (TD && CE->getOpcode() == Instruction::IntToPtr) {
+Constant *Input = CE->getOperand(0);
+unsigned InWidth = Input->getType()->getPrimitiveSizeInBits();
+Constant *Mask =
+ ConstantInt::get(APInt::getLowBitsSet(InWidth,
+TD->getPointerSizeInBits()));
+Input = ConstantExpr::getAnd(Input, Mask);
+// Do a zext or trunc to get to the dest size.
+return ConstantExpr::getIntegerCast(Input, I->getType(), false);
+ }
+}
+// FALL THROUGH.
+ case Instruction::IntToPtr:
case Instruction::Trunc:
case Instruction::ZExt:
case Instruction::SExt:
@@ -226,8 +243,6 @@
case Instruction::SIToFP:
case Instruction::FPToUI:
case Instruction::FPToSI:
- case Instruction::PtrToInt:
- case Instruction::IntToPtr:
case Instruction::BitCast:
return ConstantExpr::getCast(Opc, Ops[0], DestTy);
case Instruction::Select:
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r41024 - /llvm/trunk/lib/CodeGen/SimpleRegisterCoalescing.cpp
Author: evancheng
Date: Sat Aug 11 20:26:19 2007
New Revision: 41024
URL: http://llvm.org/viewvc/llvm-project?rev=41024&view=rev
Log:
No need to remove dead range from soon-to-be-dead live interval. Its val# may
be out of whack.
Modified:
llvm/trunk/lib/CodeGen/SimpleRegisterCoalescing.cpp
Modified: llvm/trunk/lib/CodeGen/SimpleRegisterCoalescing.cpp
URL:
http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SimpleRegisterCoalescing.cpp?rev=41024&r1=41023&r2=41024&view=diff
==
--- llvm/trunk/lib/CodeGen/SimpleRegisterCoalescing.cpp (original)
+++ llvm/trunk/lib/CodeGen/SimpleRegisterCoalescing.cpp Sat Aug 11 20:26:19 2007
@@ -316,9 +316,9 @@
}
if (isShorten || isDead) {
- // Shorten the live interval.
- LiveInterval &LiveInInt = (repSrcReg == DstInt.reg) ? DstInt : SrcInt;
- LiveInInt.removeRange(RemoveStart, RemoveEnd);
+ // Shorten the destination live interval.
+ if (repSrcReg == DstInt.reg)
+DstInt.removeRange(RemoveStart, RemoveEnd);
}
} else {
// Coalescing failed.
___
llvm-commits mailing list
llvm-commits@cs.uiuc.edu
http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
[llvm-commits] [llvm] r41028 - /llvm/trunk/lib/Transforms/Scalar/LoopIndexSplit.cpp
Author: reid Date: Sat Aug 11 23:45:36 2007 New Revision: 41028 URL: http://llvm.org/viewvc/llvm-project?rev=41028&view=rev Log: Remove unused variables. Modified: llvm/trunk/lib/Transforms/Scalar/LoopIndexSplit.cpp Modified: llvm/trunk/lib/Transforms/Scalar/LoopIndexSplit.cpp URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/LoopIndexSplit.cpp?rev=41028&r1=41027&r2=41028&view=diff == --- llvm/trunk/lib/Transforms/Scalar/LoopIndexSplit.cpp (original) +++ llvm/trunk/lib/Transforms/Scalar/LoopIndexSplit.cpp Sat Aug 11 23:45:36 2007 @@ -621,12 +621,12 @@ if (L->contains(ExitDest)) ExitDest = ExitInsn->getSuccessor(0); assert (!L->contains(ExitDest) && " Unable to find exit edge destination"); - BasicBlock *ExitSplitBlock = SplitEdge(ExitBlock, ExitDest, this); + SplitEdge(ExitBlock, ExitDest, this); //[*] Clone loop. Avoid true destination of split condition and //the blocks dominated by true destination. DenseMap ValueMap; - Loop *FalseLoop = CloneLoop(L, LPM, LI, ValueMap, this); + CloneLoop(L, LPM, LI, ValueMap, this); //[*] True loops exit edge enters False loop. //[*] Eliminate split condition's false branch from True loop. ___ llvm-commits mailing list llvm-commits@cs.uiuc.edu http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
