RPCS3/llvm
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm.git synced 2024-12-03 17:32:59 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[x86-64] allow mfence even with -mno-sse (PR23203)

Browse Source 
As shown in:
https://llvm.org/bugs/show_bug.cgi?id=23203
...we currently die because lowering believes that mfence is allowed without SSE2 on x86-64,
but the instruction def doesn't know that.

I don't know if allowing mfence without SSE is right, but if not, at least now it's consistently wrong. :)

Differential Revision: http://reviews.llvm.org/D17219



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@260828 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Sanjay Patel 2016-02-13 17:26:29 +00:00
parent 026259738d
commit 5fb1d32848
5 changed files with 42 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
lib/Target/X86/X86ISelLowering.cpp
View File

@ -19717,13 +19717,6 @@ X86TargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const {
}
}
static bool hasMFENCE(const X86Subtarget &Subtarget) {
// Use mfence if we have SSE2 or we're on x86-64 (even if we asked for
// no-sse2). There isn't any reason to disable it if the target processor
// supports it.
return Subtarget.hasSSE2() || Subtarget.is64Bit();
}
LoadInst *
X86TargetLowering::lowerIdempotentRMWIntoFencedLoad(AtomicRMWInst *AI) const {
unsigned NativeWidth = Subtarget.is64Bit() ? 64 : 32;
@ -19763,7 +19756,7 @@ X86TargetLowering::lowerIdempotentRMWIntoFencedLoad(AtomicRMWInst *AI) const {
// the IR level, so we must wrap it in an intrinsic.
return nullptr;
if (!hasMFENCE(Subtarget))
if (!Subtarget.hasMFence())
// FIXME: it might make sense to use a locked operation here but on a
// different cache-line to prevent cache-line bouncing. In practice it
// is probably a small win, and x86 processors without mfence are rare
@ -19794,7 +19787,7 @@ static SDValue LowerATOMIC_FENCE(SDValue Op, const X86Subtarget &Subtarget,
// The only fence that needs an instruction is a sequentially-consistent
// cross-thread fence.
if (FenceOrdering == SequentiallyConsistent && FenceScope == CrossThread) {
if (hasMFENCE(Subtarget))
if (Subtarget.hasMFence())
return DAG.getNode(X86ISD::MFENCE, dl, MVT::Other, Op.getOperand(0));
SDValue Chain = Op.getOperand(0);

1
lib/Target/X86/X86InstrInfo.td
View File

@ -845,6 +845,7 @@ def CallImmAddr : Predicate<"Subtarget->IsLegalToCallImmediateAddr(TM)">;
def FavorMemIndirectCall : Predicate<"!Subtarget->callRegIndirect()">;
def NotSlowIncDec : Predicate<"!Subtarget->slowIncDec()">;
def HasFastMem32 : Predicate<"!Subtarget->isUnalignedMem32Slow()">;
def HasMFence : Predicate<"Subtarget->hasMFence()">;
//===----------------------------------------------------------------------===//
// X86 Instruction Format Definitions.

4
lib/Target/X86/X86InstrSSE.td
View File

@ -3762,6 +3762,8 @@ def PAUSE : I<0x90, RawFrm, (outs), (ins),
let SchedRW = [WriteFence] in {
// Load, store, and memory fence
// TODO: As with mfence, we may want to ease the availablity of sfence/lfence
// to include any 64-bit target.
def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
"sfence", [(int_x86_sse_sfence)], IIC_SSE_SFENCE>,
PS, Requires<[HasSSE1]>;
@ -3770,7 +3772,7 @@ def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
TB, Requires<[HasSSE2]>;
def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
"mfence", [(int_x86_sse2_mfence)], IIC_SSE_MFENCE>,
TB, Requires<[HasSSE2]>;
TB, Requires<[HasMFence]>;
} // SchedRW
def : Pat<(X86SFence), (SFENCE)>;

5
lib/Target/X86/X86Subtarget.h
View File

@ -446,6 +446,11 @@ public:
bool isSLM() const { return X86ProcFamily == IntelSLM; }
bool useSoftFloat() const { return UseSoftFloat; }
/// Use mfence if we have SSE2 or we're on x86-64 (even if we asked for
/// no-sse2). There isn't any reason to disable it if the target processor
/// supports it.
bool hasMFence() const { return hasSSE2() || is64Bit(); }
const Triple &getTargetTriple() const { return TargetTriple; }
bool isTargetDarwin() const { return TargetTriple.isOSDarwin(); }

36
test/CodeGen/X86/mfence.ll
View File

@ -1,11 +1,37 @@
; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+sse2 | FileCheck %s
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i386-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X32
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=-sse2 | FileCheck %s --check-prefix=X64
; It doesn't matter if an x86-64 target has specified "no-sse2"; we still can use mfence.
define void @test() {
; CHECK-LABEL: test:
; CHECK: # BB#0:
; CHECK-NEXT: mfence
; CHECK-NEXT: retl
; X32-LABEL: test:
; X32: # BB#0:
; X32-NEXT: mfence
; X32-NEXT: retl
;
; X64-LABEL: test:
; X64: # BB#0:
; X64-NEXT: mfence
; X64-NEXT: retq
fence seq_cst
ret void
}
define i32 @fence(i32* %ptr) {
; X32-LABEL: fence:
; X32: # BB#0:
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-NEXT: mfence
; X32-NEXT: movl (%eax), %eax
; X32-NEXT: retl
;
; X64-LABEL: fence:
; X64: # BB#0:
; X64-NEXT: mfence
; X64-NEXT: movl (%rdi), %eax
; X64-NEXT: retq
%atomic = atomicrmw add i32* %ptr, i32 0 seq_cst
ret i32 %atomic
}