diff options
| -rw-r--r-- | llvm/lib/Target/X86/X86VZeroUpper.cpp | 13 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/vzero-excess.ll | 7 |
2 files changed, 9 insertions, 11 deletions
diff --git a/llvm/lib/Target/X86/X86VZeroUpper.cpp b/llvm/lib/Target/X86/X86VZeroUpper.cpp index 4465c0f988d..944c76ae68a 100644 --- a/llvm/lib/Target/X86/X86VZeroUpper.cpp +++ b/llvm/lib/Target/X86/X86VZeroUpper.cpp @@ -188,14 +188,15 @@ void VZeroUpperInserter::processBasicBlock(MachineBasicBlock &MBB) { bool IsReturnFromX86INTR = IsX86INTR && MI->isReturn(); bool IsControlFlow = MI->isCall() || MI->isReturn(); - // Shortcut: don't need to check regular instructions in dirty state. - if ((!IsControlFlow || IsReturnFromX86INTR) && CurState == EXITS_DIRTY) + // An existing VZERO* instruction resets the state. + if (MI->getOpcode() == X86::VZEROALL || + MI->getOpcode() == X86::VZEROUPPER) { + CurState = EXITS_CLEAN; continue; + } - // Ignore existing VZERO* instructions. - // FIXME: The existence of these instructions should be used to modify the - // current state and/or used when deciding whether we need to create a VZU. - if (MI->getOpcode() == X86::VZEROALL || MI->getOpcode() == X86::VZEROUPPER) + // Shortcut: don't need to check regular instructions in dirty state. + if ((!IsControlFlow || IsReturnFromX86INTR) && CurState == EXITS_DIRTY) continue; if (hasYmmReg(MI)) { diff --git a/llvm/test/CodeGen/X86/vzero-excess.ll b/llvm/test/CodeGen/X86/vzero-excess.ll index 7537b8eefb9..0ed90741b61 100644 --- a/llvm/test/CodeGen/X86/vzero-excess.ll +++ b/llvm/test/CodeGen/X86/vzero-excess.ll @@ -1,7 +1,8 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s -; FIXME: The vzeroupper added by the VZeroUpperInserter pass is unnecessary in these tests. +; In the following 4 tests, the existing call to VZU/VZA ensures clean state before +; the call to the unknown, so we don't need to insert a second VZU at that point. define <4 x float> @zeroupper_v4f32(<8 x float> *%x, <8 x float> %y) nounwind { ; CHECK-LABEL: zeroupper_v4f32: @@ -11,7 +12,6 @@ define <4 x float> @zeroupper_v4f32(<8 x float> *%x, <8 x float> %y) nounwind { ; CHECK-NEXT: vmovups %ymm0, (%rsp) # 32-byte Spill ; CHECK-NEXT: movq %rdi, %rbx ; CHECK-NEXT: vzeroupper -; CHECK-NEXT: vzeroupper ; CHECK-NEXT: callq the_unknown ; CHECK-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload ; CHECK-NEXT: vaddps (%rbx), %ymm0, %ymm0 @@ -37,7 +37,6 @@ define <8 x float> @zeroupper_v8f32(<8 x float> %x) nounwind { ; CHECK-NEXT: subq $56, %rsp ; CHECK-NEXT: vmovups %ymm0, (%rsp) # 32-byte Spill ; CHECK-NEXT: vzeroupper -; CHECK-NEXT: vzeroupper ; CHECK-NEXT: callq the_unknown ; CHECK-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload ; CHECK-NEXT: addq $56, %rsp @@ -55,7 +54,6 @@ define <4 x float> @zeroall_v4f32(<8 x float> *%x, <8 x float> %y) nounwind { ; CHECK-NEXT: vmovups %ymm0, (%rsp) # 32-byte Spill ; CHECK-NEXT: movq %rdi, %rbx ; CHECK-NEXT: vzeroall -; CHECK-NEXT: vzeroupper ; CHECK-NEXT: callq the_unknown ; CHECK-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload ; CHECK-NEXT: vaddps (%rbx), %ymm0, %ymm0 @@ -81,7 +79,6 @@ define <8 x float> @zeroall_v8f32(<8 x float> %x) nounwind { ; CHECK-NEXT: subq $56, %rsp ; CHECK-NEXT: vmovups %ymm0, (%rsp) # 32-byte Spill ; CHECK-NEXT: vzeroall -; CHECK-NEXT: vzeroupper ; CHECK-NEXT: callq the_unknown ; CHECK-NEXT: vmovups (%rsp), %ymm0 # 32-byte Reload ; CHECK-NEXT: addq $56, %rsp |
