From 2671f80b2d151c9a87a4a93860f8a6b98db57aad Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Wed, 14 Mar 2018 17:57:19 +0000
Subject: [PATCH] [X86] Add back fast-isel code for handling i8 shifts.

I removed this in r316797 because the coverage report showed no coverage and I thought it should have been handled by the auto generated table. I now see that there is code that bypasses the table if the shift amount is out of bounds.

This adds back the code. We'll codegen out of bounds i8 shifts to effectively (amount & 0x1f). The 0x1f is a strange quirk of x86 that shift amounts are always masked to 5-bits(except 64-bits). So if the masked value is still out bounds the result will be 0.

Fixes PR36731.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@327540 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86FastISel.cpp      | 21 ++++++++++++++-------
 test/CodeGen/X86/fast-isel-shift.ll | 12 ++++++++++++
 2 files changed, 26 insertions(+), 7 deletions(-)

diff --git a/lib/Target/X86/X86FastISel.cpp b/lib/Target/X86/X86FastISel.cpp
index 7e88304ae19..943022585a1 100644
--- a/lib/Target/X86/X86FastISel.cpp
+++ b/lib/Target/X86/X86FastISel.cpp
@@ -1789,9 +1789,16 @@ bool X86FastISel::X86SelectBranch(const Instruction *I) {
 bool X86FastISel::X86SelectShift(const Instruction *I) {
   unsigned CReg = 0, OpReg = 0;
   const TargetRegisterClass *RC = nullptr;
-  assert(!I->getType()->isIntegerTy(8) &&
-         "i8 shifts should be handled by autogenerated table");
-  if (I->getType()->isIntegerTy(16)) {
+  if (I->getType()->isIntegerTy(8)) {
+    CReg = X86::CL;
+    RC = &X86::GR8RegClass;
+    switch (I->getOpcode()) {
+    case Instruction::LShr: OpReg = X86::SHR8rCL; break;
+    case Instruction::AShr: OpReg = X86::SAR8rCL; break;
+    case Instruction::Shl:  OpReg = X86::SHL8rCL; break;
+    default: return false;
+    }
+  } else if (I->getType()->isIntegerTy(16)) {
     CReg = X86::CX;
     RC = &X86::GR16RegClass;
     switch (I->getOpcode()) {
@@ -1836,10 +1843,10 @@ bool X86FastISel::X86SelectShift(const Instruction *I) {
 
   // The shift instruction uses X86::CL. If we defined a super-register
   // of X86::CL, emit a subreg KILL to precisely describe what we're doing here.
-  assert(CReg != X86::CL && "CReg should be a super register of CL");
-  BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
-          TII.get(TargetOpcode::KILL), X86::CL)
-    .addReg(CReg, RegState::Kill);
+  if (CReg != X86::CL)
+    BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
+            TII.get(TargetOpcode::KILL), X86::CL)
+      .addReg(CReg, RegState::Kill);
 
   unsigned ResultReg = createResultReg(RC);
   BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(OpReg), ResultReg)
diff --git a/test/CodeGen/X86/fast-isel-shift.ll b/test/CodeGen/X86/fast-isel-shift.ll
index 3699b7ba4bf..4dc56f351f5 100644
--- a/test/CodeGen/X86/fast-isel-shift.ll
+++ b/test/CodeGen/X86/fast-isel-shift.ll
@@ -381,3 +381,15 @@ define i64 @ashr_imm4_i64(i64 %a) {
   %c = ashr i64 %a, 4
   ret i64 %c
 }
+
+; Make sure we don't crash on out of bounds i8 shifts.
+define i8 @PR36731(i8 %a) {
+; CHECK-LABEL: PR36731:
+; CHECK:       ## %bb.0:
+; CHECK-NEXT:    movb $255, %cl
+; CHECK-NEXT:    shlb %cl, %dil
+; CHECK-NEXT:    movl %edi, %eax
+; CHECK-NEXT:    retq
+  %b = shl i8 %a, -1
+  ret i8 %b
+}
-- 
2.11.0