[llvm] r270718 - [x86, AVX] allow explicit calls to VZERO* to modify state in VZeroUpperInserter pass (PR27823)

[Sanjay Patel via llvm-commits]

Author: spatel
Date: Wed May 25 11:39:47 2016
New Revision: 270718

URL: http://llvm.org/viewvc/llvm-project?rev=270718&view=rev
Log:
[x86, AVX] allow explicit calls to VZERO* to modify state in VZeroUpperInserter pass (PR27823)

As noted in the review, there are still problems, so this doesn't the bug completely.

Differential Revision: http://reviews.llvm.org/D20529


Modified:
    llvm/trunk/lib/Target/X86/X86VZeroUpper.cpp
    llvm/trunk/test/CodeGen/X86/vzero-excess.ll

Modified: llvm/trunk/lib/Target/X86/X86VZeroUpper.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86VZeroUpper.cpp?rev=270718&r1=270717&r2=270718&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86VZeroUpper.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86VZeroUpper.cpp Wed May 25 11:39:47 2016
@@ -188,14 +188,15 @@ void VZeroUpperInserter::processBasicBlo
     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)) {

Modified: llvm/trunk/test/CodeGen/X86/vzero-excess.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/vzero-excess.ll?rev=270718&r1=270717&r2=270718&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/vzero-excess.ll (original)
+++ llvm/trunk/test/CodeGen/X86/vzero-excess.ll Wed May 25 11:39:47 2016
@@ -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
 ; 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
 ; 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 f
 ; 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 f
 ; 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