diff options
author | Chris Lattner <sabre@nondot.org> | 2009-04-30 00:48:50 +0000 |
---|---|---|
committer | Chris Lattner <sabre@nondot.org> | 2009-04-30 00:48:50 +0000 |
commit | 6c147293d63e836e3677eb933ff0513c7cad1e6a (patch) | |
tree | ee0747f3bcefcead033152b33d01a5d67c290460 | |
parent | 610079287ef50721548a0399c126076be3a9a508 (diff) | |
download | llvm-6c147293d63e836e3677eb933ff0513c7cad1e6a.tar.gz llvm-6c147293d63e836e3677eb933ff0513c7cad1e6a.tar.bz2 llvm-6c147293d63e836e3677eb933ff0513c7cad1e6a.tar.xz |
fix a regression handling indirect results: these need to be considered
memory operands otherwise the writebacks get lost when the inline asm
doesn't otherwise have side effects. This fixes rdar://6839427, though
clang really shouldn't generate these anymore.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@70455 91177308-0d34-0410-b5e6-96231b3b80d8
-rw-r--r-- | lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp | 21 | ||||
-rw-r--r-- | test/CodeGen/X86/2009-04-29-IndirectDestOperands.ll | 22 |
2 files changed, 35 insertions, 8 deletions
diff --git a/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp b/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp index 01d73b3ffb..71f5d81314 100644 --- a/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp +++ b/lib/CodeGen/SelectionDAG/SelectionDAGBuild.cpp @@ -5075,6 +5075,10 @@ hasInlineAsmMemConstraint(std::vector<InlineAsm::ConstraintInfo> &CInfos, if (CType == TargetLowering::C_Memory) return true; } + + // Indirect operand accesses access memory. + if (CI.isIndirect) + return true; } return false; @@ -5088,11 +5092,6 @@ void SelectionDAGLowering::visitInlineAsm(CallSite CS) { /// ConstraintOperands - Information about all of the constraints. std::vector<SDISelAsmOperandInfo> ConstraintOperands; - // We won't need to flush pending loads if this asm doesn't touch - // memory and is nonvolatile. - SDValue Chain = IA->hasSideEffects() ? getRoot() : DAG.getRoot(); - SDValue Flag; - std::set<unsigned> OutputRegs, InputRegs; // Do a prepass over the constraints, canonicalizing them, and building up the @@ -5101,10 +5100,15 @@ void SelectionDAGLowering::visitInlineAsm(CallSite CS) { ConstraintInfos = IA->ParseConstraints(); bool hasMemory = hasInlineAsmMemConstraint(ConstraintInfos, TLI); - // Flush pending loads if this touches memory (includes clobbering it). - // It's possible this is overly conservative. - if (hasMemory) + + SDValue Chain, Flag; + + // We won't need to flush pending loads if this asm doesn't touch + // memory and is nonvolatile. + if (hasMemory || IA->hasSideEffects()) Chain = getRoot(); + else + Chain = DAG.getRoot(); unsigned ArgNo = 0; // ArgNo - The argument of the CallInst. unsigned ResNo = 0; // ResNo - The result number of the next output. @@ -5482,6 +5486,7 @@ void SelectionDAGLowering::visitInlineAsm(CallSite CS) { SDValue OutVal = OutRegs.getCopyFromRegs(DAG, getCurDebugLoc(), Chain, &Flag); StoresToEmit.push_back(std::make_pair(OutVal, Ptr)); + } // Emit the non-flagged stores from the physregs. diff --git a/test/CodeGen/X86/2009-04-29-IndirectDestOperands.ll b/test/CodeGen/X86/2009-04-29-IndirectDestOperands.ll new file mode 100644 index 0000000000..fc31c0b416 --- /dev/null +++ b/test/CodeGen/X86/2009-04-29-IndirectDestOperands.ll @@ -0,0 +1,22 @@ +; RUN: llvm-as < %s | llc | grep {movl.*%ebx, 8(%esi)} +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" +target triple = "i386-apple-darwin9.0" + +define void @cpuid(i32* %data) nounwind { +entry: + %arrayidx = getelementptr i32* %data, i32 1 ; <i32*> [#uses=1] + %arrayidx2 = getelementptr i32* %data, i32 2 ; <i32*> [#uses=1] + %arrayidx4 = getelementptr i32* %data, i32 3 ; <i32*> [#uses=1] + %arrayidx6 = getelementptr i32* %data, i32 4 ; <i32*> [#uses=1] + %arrayidx8 = getelementptr i32* %data, i32 5 ; <i32*> [#uses=1] + %tmp9 = load i32* %arrayidx8 ; <i32> [#uses=1] + %arrayidx11 = getelementptr i32* %data, i32 6 ; <i32*> [#uses=1] + %tmp12 = load i32* %arrayidx11 ; <i32> [#uses=1] + %arrayidx14 = getelementptr i32* %data, i32 7 ; <i32*> [#uses=1] + %tmp15 = load i32* %arrayidx14 ; <i32> [#uses=1] + %arrayidx17 = getelementptr i32* %data, i32 8 ; <i32*> [#uses=1] + %tmp18 = load i32* %arrayidx17 ; <i32> [#uses=1] + %0 = call i32 asm "cpuid", "={ax},=*{bx},=*{cx},=*{dx},{ax},{bx},{cx},{dx},~{dirflag},~{fpsr},~{flags}"(i32* %arrayidx2, i32* %arrayidx4, i32* %arrayidx6, i32 %tmp9, i32 %tmp12, i32 %tmp15, i32 %tmp18) nounwind ; <i32> [#uses=1] + store i32 %0, i32* %arrayidx + ret void +} |