From 24c0aa9af62dab27aa856c5b3423878298ef49dd Mon Sep 17 00:00:00 2001
From: Roman Lebedev <lebedev.ri@gmail.com>
Date: Thu, 26 Jul 2018 17:34:28 +0000
Subject: [PATCH] [DAGCombine] optimizeSetCCOfSignedTruncationCheck(): handle
 ule,ugt CondCodes.

Summary:
A follow-up for D49266 / rL337166.

At least one of these cases is more canonical,
so we really do have to handle it.
https://godbolt.org/g/pkzP3X
https://rise4fun.com/Alive/pQyhZZ

We won't get to these cases with I1 being -1,
as that will be constant-folded to true or false.

I'm also not sure we actually hit the 'ule' case,
but i think the worst think that could happen is that being dead code.

Reviewers: spatel, craig.topper, RKSimon, javed.absar, efriedma

Reviewed By: spatel

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D49497

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@338044 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/CodeGen/SelectionDAG/TargetLowering.cpp        | 27 ++++++++++++++--------
 .../AArch64/lack-of-signed-truncation-check.ll     |  6 ++---
 .../CodeGen/X86/lack-of-signed-truncation-check.ll | 16 ++++++-------
 3 files changed, 28 insertions(+), 21 deletions(-)

diff --git a/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 75d0a6bd58e..fa867fcec36 100644
--- a/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -1864,14 +1864,6 @@ SDValue TargetLowering::simplifySetCCWithAnd(EVT VT, SDValue N0, SDValue N1,
 SDValue TargetLowering::optimizeSetCCOfSignedTruncationCheck(
     EVT SCCVT, SDValue N0, SDValue N1, ISD::CondCode Cond, DAGCombinerInfo &DCI,
     const SDLoc &DL) const {
-  ISD::CondCode NewCond;
-  if (Cond == ISD::CondCode::SETULT)
-    NewCond = ISD::CondCode::SETEQ;
-  else if (Cond == ISD::CondCode::SETUGE)
-    NewCond = ISD::CondCode::SETNE;
-  else
-    return SDValue();
-
   // We must be comparing with a constant.
   ConstantSDNode *C1;
   if (!(C1 = dyn_cast<ConstantSDNode>(N1)))
@@ -1891,7 +1883,24 @@ SDValue TargetLowering::optimizeSetCCOfSignedTruncationCheck(
 
   // Validate constants ...
 
-  const APInt &I1 = C1->getAPIntValue();
+  APInt I1 = C1->getAPIntValue();
+
+  ISD::CondCode NewCond;
+  if (Cond == ISD::CondCode::SETULT) {
+    NewCond = ISD::CondCode::SETEQ;
+  } else if (Cond == ISD::CondCode::SETULE) {
+    NewCond = ISD::CondCode::SETEQ;
+    // But need to 'canonicalize' the constant.
+    I1 += 1;
+  } else if (Cond == ISD::CondCode::SETUGT) {
+    NewCond = ISD::CondCode::SETNE;
+    // But need to 'canonicalize' the constant.
+    I1 += 1;
+  } else if (Cond == ISD::CondCode::SETUGE) {
+    NewCond = ISD::CondCode::SETNE;
+  } else
+    return SDValue();
+
   const APInt &I01 = C01->getAPIntValue();
   // Both of them must be power-of-two, and the constant from setcc is bigger.
   if (!(I1.ugt(I01) && I1.isPowerOf2() && I01.isPowerOf2()))
diff --git a/test/CodeGen/AArch64/lack-of-signed-truncation-check.ll b/test/CodeGen/AArch64/lack-of-signed-truncation-check.ll
index 9bfac1bdc05..be2dffde824 100644
--- a/test/CodeGen/AArch64/lack-of-signed-truncation-check.ll
+++ b/test/CodeGen/AArch64/lack-of-signed-truncation-check.ll
@@ -257,10 +257,10 @@ define i1 @add_ugecmp_i64_i8(i64 %x) nounwind {
 define i1 @add_ugtcmp_i16_i8(i16 %x) nounwind {
 ; CHECK-LABEL: add_ugtcmp_i16_i8:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    add w8, w0, #128 // =128
+; CHECK-NEXT:    sxtb w8, w0
 ; CHECK-NEXT:    and w8, w8, #0xffff
-; CHECK-NEXT:    cmp w8, #255 // =255
-; CHECK-NEXT:    cset w0, hi
+; CHECK-NEXT:    cmp w8, w0, uxth
+; CHECK-NEXT:    cset w0, ne
 ; CHECK-NEXT:    ret
   %tmp0 = add i16 %x, 128 ; 1U << (8-1)
   %tmp1 = icmp ugt i16 %tmp0, 255 ; (1U << 8) - 1
diff --git a/test/CodeGen/X86/lack-of-signed-truncation-check.ll b/test/CodeGen/X86/lack-of-signed-truncation-check.ll
index 813c4c58d4c..ac36087d002 100644
--- a/test/CodeGen/X86/lack-of-signed-truncation-check.ll
+++ b/test/CodeGen/X86/lack-of-signed-truncation-check.ll
@@ -422,19 +422,17 @@ define i1 @add_ugecmp_i64_i8(i64 %x) nounwind {
 define i1 @add_ugtcmp_i16_i8(i16 %x) nounwind {
 ; X86-LABEL: add_ugtcmp_i16_i8:
 ; X86:       # %bb.0:
-; X86-NEXT:    movl $128, %eax
-; X86-NEXT:    addl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movzwl %ax, %eax
-; X86-NEXT:    cmpl $255, %eax
-; X86-NEXT:    seta %al
+; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movsbl %al, %ecx
+; X86-NEXT:    cmpw %ax, %cx
+; X86-NEXT:    setne %al
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: add_ugtcmp_i16_i8:
 ; X64:       # %bb.0:
-; X64-NEXT:    subl $-128, %edi
-; X64-NEXT:    movzwl %di, %eax
-; X64-NEXT:    cmpl $255, %eax
-; X64-NEXT:    seta %al
+; X64-NEXT:    movsbl %dil, %eax
+; X64-NEXT:    cmpw %di, %ax
+; X64-NEXT:    setne %al
 ; X64-NEXT:    retq
   %tmp0 = add i16 %x, 128 ; 1U << (8-1)
   %tmp1 = icmp ugt i16 %tmp0, 255 ; (1U << 8) - 1
-- 
2.11.0