From 85ad85b52ea35056887e642ecf13ff7b3a406d25 Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@gmail.com>
Date: Mon, 13 Mar 2017 21:58:54 +0000
Subject: [PATCH] [AVX-512] Fix another case where we are copying from a mask
 register using AH/BH/CH/DH with fastisel.

Fixes PR32256. Still planning to do an audit for other possible cases.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@297678 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86FastISel.cpp |  3 +-
 test/CodeGen/X86/pr32256.ll    | 67 ++++++++++++++++++++++++++++++++++
 2 files changed, 69 insertions(+), 1 deletion(-)
 create mode 100644 test/CodeGen/X86/pr32256.ll

diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index 6ffff041929..4eec15779a1 100644
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -1751,7 +1751,8 @@ bool X86FastISel::X86SelectBranch(const Instruction *I) {
   // In case OpReg is a K register, COPY to a GPR
   if (MRI.getRegClass(OpReg) == &X86::VK1RegClass) {
     unsigned KOpReg = OpReg;
-    OpReg = createResultReg(&X86::GR8RegClass);
+    OpReg = createResultReg(Subtarget->is64Bit() ? &X86::GR8RegClass
+                                                 : &X86::GR8_ABCD_LRegClass);
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
             TII.get(TargetOpcode::COPY), OpReg)
         .addReg(KOpReg);
diff --git a/test/CodeGen/X86/pr32256.ll b/test/CodeGen/X86/pr32256.ll
new file mode 100644
index 00000000000..a6ec02d8069
--- /dev/null
+++ b/test/CodeGen/X86/pr32256.ll
@@ -0,0 +1,67 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -fast-isel -mtriple=i686-unknown-unknown -O0 -mcpu=skx | FileCheck %s
+
+@c = external global i8, align 1
+
+; Function Attrs: noinline nounwind
+define void @_Z1av() {
+; CHECK-LABEL: _Z1av:
+; CHECK:       # BB#0: # %entry
+; CHECK-NEXT:    pushl %ebx
+; CHECK-NEXT:  .Lcfi0:
+; CHECK-NEXT:    .cfi_def_cfa_offset 8
+; CHECK-NEXT:    subl $8, %esp
+; CHECK-NEXT:  .Lcfi1:
+; CHECK-NEXT:    .cfi_def_cfa_offset 16
+; CHECK-NEXT:  .Lcfi2:
+; CHECK-NEXT:    .cfi_offset %ebx, -8
+; CHECK-NEXT:    xorl %eax, %eax
+; CHECK-NEXT:    movb %al, %cl
+; CHECK-NEXT:    movb $1, %dl
+; CHECK-NEXT:    movb c, %ch
+; CHECK-NEXT:    # implicit-def: %EAX
+; CHECK-NEXT:    movb %ch, %al
+; CHECK-NEXT:    andl $1, %eax
+; CHECK-NEXT:    kmovw %eax, %k0
+; CHECK-NEXT:    kmovq %k0, %k1
+; CHECK-NEXT:    kxnorw %k0, %k0, %k2
+; CHECK-NEXT:    kshiftrw $15, %k2, %k2
+; CHECK-NEXT:    kxorw %k2, %k0, %k0
+; CHECK-NEXT:    kmovb %k0, %ebx
+; CHECK-NEXT:    testb $1, %bl
+; CHECK-NEXT:    kmovb %ecx, %k0
+; CHECK-NEXT:    kmovw %k1, {{[0-9]+}}(%esp) # 2-byte Spill
+; CHECK-NEXT:    movb %dl, {{[0-9]+}}(%esp) # 1-byte Spill
+; CHECK-NEXT:    kmovw %k0, (%esp) # 2-byte Spill
+; CHECK-NEXT:    jne .LBB0_1
+; CHECK-NEXT:    jmp .LBB0_2
+; CHECK-NEXT:  .LBB0_1: # %land.rhs
+; CHECK-NEXT:    xorl %eax, %eax
+; CHECK-NEXT:    movb %al, %cl
+; CHECK-NEXT:    kmovb %ecx, %k0
+; CHECK-NEXT:    kmovw %k0, (%esp) # 2-byte Spill
+; CHECK-NEXT:    jmp .LBB0_2
+; CHECK-NEXT:  .LBB0_2: # %land.end
+; CHECK-NEXT:    kmovw (%esp), %k0 # 2-byte Reload
+; CHECK-NEXT:    kmovb %k0, %eax
+; CHECK-NEXT:    andb $1, %al
+; CHECK-NEXT:    movb %al, {{[0-9]+}}(%esp)
+; CHECK-NEXT:    addl $8, %esp
+; CHECK-NEXT:    popl %ebx
+; CHECK-NEXT:    retl
+entry:
+  %b = alloca i8, align 1
+  %0 = load i8, i8* @c, align 1
+  %tobool = trunc i8 %0 to i1
+  %lnot = xor i1 %tobool, true
+  br i1 %lnot, label %land.rhs, label %land.end
+
+land.rhs:                                         ; preds = %entry
+  br label %land.end
+
+land.end:                                         ; preds = %land.rhs, %entry
+  %1 = phi i1 [ false, %entry ], [ false, %land.rhs ]
+  %conv = zext i1 %1 to i8
+  store i8 %conv, i8* %b, align 1
+  ret void
+}
-- 
2.50.1