From 450c33b212860a3d7e2be7aa23efba40665430bb Mon Sep 17 00:00:00 2001 From: Juergen Ributzka Date: Thu, 31 Jul 2014 00:11:11 +0000 Subject: [PATCH] [FastISel][AArch64 and X86] Don't emit stores for UNDEF arguments during function call lowering. UNDEF arguments are not ment to be touched - especially for the webkit_js calling convention. This fix reproduces the already existing behavior of SelectionDAG in FastISel. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@214366 91177308-0d34-0410-b5e6-96231b3b80d8 --- lib/Target/AArch64/AArch64FastISel.cpp | 4 + lib/Target/X86/X86FastISel.cpp | 5 ++ test/CodeGen/X86/patchpoint-webkit_jscc.ll | 88 ++++++++++++++++++++++ test/CodeGen/X86/patchpoint.ll | 57 +------------- 4 files changed, 99 insertions(+), 55 deletions(-) create mode 100644 test/CodeGen/X86/patchpoint-webkit_jscc.ll diff --git a/lib/Target/AArch64/AArch64FastISel.cpp b/lib/Target/AArch64/AArch64FastISel.cpp index 371accb5829..d80f34ca2d3 100644 --- a/lib/Target/AArch64/AArch64FastISel.cpp +++ b/lib/Target/AArch64/AArch64FastISel.cpp @@ -1280,6 +1280,10 @@ bool AArch64FastISel::ProcessCallArgs(CallLoweringInfo &CLI, } else { assert(VA.isMemLoc() && "Assuming store on stack."); + // Don't emit stores for undef values. + if (isa(ArgVal)) + continue; + // Need to store on the stack. unsigned ArgSize = (ArgVT.getSizeInBits() + 7) / 8; diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp index 52e65670a50..0938eda9dad 100644 --- a/lib/Target/X86/X86FastISel.cpp +++ b/lib/Target/X86/X86FastISel.cpp @@ -2831,6 +2831,11 @@ bool X86FastISel::FastLowerCall(CallLoweringInfo &CLI) { OutRegs.push_back(VA.getLocReg()); } else { assert(VA.isMemLoc()); + + // Don't emit stores for undef values. + if (isa(ArgVal)) + continue; + unsigned LocMemOffset = VA.getLocMemOffset(); X86AddressMode AM; AM.Base.Reg = RegInfo->getStackRegister(); diff --git a/test/CodeGen/X86/patchpoint-webkit_jscc.ll b/test/CodeGen/X86/patchpoint-webkit_jscc.ll new file mode 100644 index 00000000000..03f662ceadc --- /dev/null +++ b/test/CodeGen/X86/patchpoint-webkit_jscc.ll @@ -0,0 +1,88 @@ +; RUN: llc -mtriple=x86_64-apple-darwin -mcpu=corei7 -disable-fp-elim < %s | FileCheck %s +; RUN: llc -mtriple=x86_64-apple-darwin -mcpu=corei7 -disable-fp-elim -fast-isel -fast-isel-abort < %s | FileCheck %s --check-prefix=FAST + +; Test the webkit_jscc calling convention. +; One argument will be passed in register, the other will be pushed on the stack. +; Return value in $rax. +define void @jscall_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) { +entry: +; CHECK-LABEL: jscall_patchpoint_codegen: +; CHECK: Ltmp +; CHECK: movq %r{{.+}}, (%rsp) +; CHECK: movq %r{{.+}}, %rax +; CHECK: Ltmp +; CHECK-NEXT: movabsq $-559038736, %r11 +; CHECK-NEXT: callq *%r11 +; CHECK: movq %rax, (%rsp) +; CHECK: callq +; FAST-LABEL: jscall_patchpoint_codegen: +; FAST: Ltmp +; FAST: movq %r{{.+}}, (%rsp) +; FAST: movq %r{{.+}}, %rax +; FAST: Ltmp +; FAST-NEXT: movabsq $-559038736, %r11 +; FAST-NEXT: callq *%r11 +; FAST: movq %rax, (%rsp) +; FAST: callq + %resolveCall2 = inttoptr i64 -559038736 to i8* + %result = tail call webkit_jscc i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 5, i32 15, i8* %resolveCall2, i32 2, i64 %p4, i64 %p2) + %resolveCall3 = inttoptr i64 -559038737 to i8* + tail call webkit_jscc void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 6, i32 15, i8* %resolveCall3, i32 2, i64 %p4, i64 %result) + ret void +} + +; Test if the arguments are properly aligned and that we don't store undef arguments. +define i64 @jscall_patchpoint_codegen2(i64 %callee) { +entry: +; CHECK-LABEL: jscall_patchpoint_codegen2: +; CHECK: Ltmp +; CHECK: movq $6, 24(%rsp) +; CHECK-NEXT: movl $4, 16(%rsp) +; CHECK-NEXT: movq $2, (%rsp) +; CHECK: Ltmp +; CHECK-NEXT: movabsq $-559038736, %r11 +; CHECK-NEXT: callq *%r11 +; FAST-LABEL: jscall_patchpoint_codegen2: +; FAST: Ltmp +; FAST: movq $2, (%rsp) +; FAST-NEXT: movl $4, 16(%rsp) +; FAST-NEXT: movq $6, 24(%rsp) +; FAST: Ltmp +; FAST-NEXT: movabsq $-559038736, %r11 +; FAST-NEXT: callq *%r11 + %call = inttoptr i64 -559038736 to i8* + %result = call webkit_jscc i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 7, i32 15, i8* %call, i32 6, i64 %callee, i64 2, i64 undef, i32 4, i32 undef, i64 6) + ret i64 %result +} + +; Test if the arguments are properly aligned and that we don't store undef arguments. +define i64 @jscall_patchpoint_codegen3(i64 %callee) { +entry: +; CHECK-LABEL: jscall_patchpoint_codegen3: +; CHECK: Ltmp +; CHECK: movq $10, 48(%rsp) +; CHECK-NEXT: movl $8, 36(%rsp) +; CHECK-NEXT: movq $6, 24(%rsp) +; CHECK-NEXT: movl $4, 16(%rsp) +; CHECK-NEXT: movq $2, (%rsp) +; CHECK: Ltmp +; CHECK-NEXT: movabsq $-559038736, %r11 +; CHECK-NEXT: callq *%r11 +; FAST-LABEL: jscall_patchpoint_codegen3: +; FAST: Ltmp +; FAST: movq $2, (%rsp) +; FAST-NEXT: movl $4, 16(%rsp) +; FAST-NEXT: movq $6, 24(%rsp) +; FAST-NEXT: movl $8, 36(%rsp) +; FAST-NEXT: movq $10, 48(%rsp) +; FAST: Ltmp +; FAST-NEXT: movabsq $-559038736, %r11 +; FAST-NEXT: callq *%r11 + %call = inttoptr i64 -559038736 to i8* + %result = call webkit_jscc i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 7, i32 15, i8* %call, i32 10, i64 %callee, i64 2, i64 undef, i32 4, i32 undef, i64 6, i32 undef, i32 8, i32 undef, i64 10) + ret i64 %result +} + +declare void @llvm.experimental.patchpoint.void(i64, i32, i8*, i32, ...) +declare i64 @llvm.experimental.patchpoint.i64(i64, i32, i8*, i32, ...) + diff --git a/test/CodeGen/X86/patchpoint.ll b/test/CodeGen/X86/patchpoint.ll index 62b12732ded..d988195e21b 100644 --- a/test/CodeGen/X86/patchpoint.ll +++ b/test/CodeGen/X86/patchpoint.ll @@ -1,4 +1,5 @@ -; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7 -disable-fp-elim | FileCheck %s +; RUN: llc -mtriple=x86_64-apple-darwin -mcpu=corei7 -disable-fp-elim < %s | FileCheck %s +; RUN: llc -mtriple=x86_64-apple-darwin -mcpu=corei7 -disable-fp-elim -fast-isel -fast-isel-abort < %s | FileCheck %s ; Trivial patchpoint codegen ; @@ -38,60 +39,6 @@ entry: ret void } -; Test the webkit_jscc calling convention. -; One argument will be passed in register, the other will be pushed on the stack. -; Return value in $rax. -define void @jscall_patchpoint_codegen(i64 %p1, i64 %p2, i64 %p3, i64 %p4) { -entry: -; CHECK-LABEL: jscall_patchpoint_codegen: -; CHECK: Ltmp -; CHECK: movq %r{{.+}}, (%rsp) -; CHECK: movq %r{{.+}}, %rax -; CHECK: Ltmp -; CHECK-NEXT: movabsq $-559038736, %r11 -; CHECK-NEXT: callq *%r11 -; CHECK: movq %rax, (%rsp) -; CHECK: callq - %resolveCall2 = inttoptr i64 -559038736 to i8* - %result = tail call webkit_jscc i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 5, i32 15, i8* %resolveCall2, i32 2, i64 %p4, i64 %p2) - %resolveCall3 = inttoptr i64 -559038737 to i8* - tail call webkit_jscc void (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.void(i64 6, i32 15, i8* %resolveCall3, i32 2, i64 %p4, i64 %result) - ret void -} - -; Test if the arguments are properly aligned and that we don't store undef arguments. -define i64 @jscall_patchpoint_codegen2(i64 %callee) { -entry: -; CHECK-LABEL: jscall_patchpoint_codegen2: -; CHECK: Ltmp -; CHECK: movq $6, 24(%rsp) -; CHECK-NEXT: movl $4, 16(%rsp) -; CHECK-NEXT: movq $2, (%rsp) -; CHECK: Ltmp -; CHECK-NEXT: movabsq $-559038736, %r11 -; CHECK-NEXT: callq *%r11 - %call = inttoptr i64 -559038736 to i8* - %result = call webkit_jscc i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 7, i32 15, i8* %call, i32 6, i64 %callee, i64 2, i64 undef, i32 4, i32 undef, i64 6) - ret i64 %result -} - -; Test if the arguments are properly aligned and that we don't store undef arguments. -define i64 @jscall_patchpoint_codegen3(i64 %callee) { -entry: -; CHECK-LABEL: jscall_patchpoint_codegen3: -; CHECK: Ltmp -; CHECK: movq $10, 48(%rsp) -; CHECK-NEXT: movl $8, 36(%rsp) -; CHECK-NEXT: movq $6, 24(%rsp) -; CHECK-NEXT: movl $4, 16(%rsp) -; CHECK-NEXT: movq $2, (%rsp) -; CHECK: Ltmp -; CHECK-NEXT: movabsq $-559038736, %r11 -; CHECK-NEXT: callq *%r11 - %call = inttoptr i64 -559038736 to i8* - %result = call webkit_jscc i64 (i64, i32, i8*, i32, ...)* @llvm.experimental.patchpoint.i64(i64 7, i32 15, i8* %call, i32 10, i64 %callee, i64 2, i64 undef, i32 4, i32 undef, i64 6, i32 undef, i32 8, i32 undef, i64 10) - ret i64 %result -} ; Test patchpoints reusing the same TargetConstant. ; Assertion failed: (CI.getNumArgOperands() >= NumArgs + 4) -- 2.34.1