From cc5dbbf759ae10da7378fc831d5d0cdeb11e814d Mon Sep 17 00:00:00 2001
From: Craig Topper <craig.topper@intel.com>
Date: Sat, 18 Aug 2018 05:52:42 +0000
Subject: [PATCH] [DAGCombiner] Allow divide by constant optimization on opaque
 constants.

Summary:
I believe this restores the behavior we had before r339147.

Fixes PR38622.

Reviewers: RKSimon, chandlerc, spatel

Reviewed By: chandlerc

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D50936

llvm-svn: 340120
---
 llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp |  4 +-
 llvm/test/CodeGen/X86/divide-by-constant.ll   | 78 +++++++++++++++++++
 2 files changed, 80 insertions(+), 2 deletions(-)

diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 453f506d7980..62ee5c30086d 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -3224,7 +3224,7 @@ SDValue DAGCombiner::visitSDIVLike(SDValue N0, SDValue N1, SDNode *N) {
   // alternate sequence.  Targets may check function attributes for size/speed
   // trade-offs.
   AttributeList Attr = DAG.getMachineFunction().getFunction().getAttributes();
-  if (isConstantOrConstantVector(N1, /*NoOpaques*/ true) &&
+  if (isConstantOrConstantVector(N1) &&
       !TLI.isIntDivCheap(N->getValueType(0), Attr))
     if (SDValue Op = BuildSDIV(N))
       return Op;
@@ -3316,7 +3316,7 @@ SDValue DAGCombiner::visitUDIVLike(SDValue N0, SDValue N1, SDNode *N) {
 
   // fold (udiv x, c) -> alternate
   AttributeList Attr = DAG.getMachineFunction().getFunction().getAttributes();
-  if (isConstantOrConstantVector(N1, /*NoOpaques*/ true) &&
+  if (isConstantOrConstantVector(N1) &&
       !TLI.isIntDivCheap(N->getValueType(0), Attr))
     if (SDValue Op = BuildUDIV(N))
       return Op;
diff --git a/llvm/test/CodeGen/X86/divide-by-constant.ll b/llvm/test/CodeGen/X86/divide-by-constant.ll
index cc2dc1b1d094..8f0bb1075bd1 100644
--- a/llvm/test/CodeGen/X86/divide-by-constant.ll
+++ b/llvm/test/CodeGen/X86/divide-by-constant.ll
@@ -330,3 +330,81 @@ entry:
 	%div = udiv i64 %rem, 7
 	ret i64 %div
 }
+
+define { i64, i32 } @PR38622(i64) {
+; X32-LABEL: PR38622:
+; X32:       # %bb.0:
+; X32-NEXT:    pushl %ebp
+; X32-NEXT:    .cfi_def_cfa_offset 8
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    .cfi_def_cfa_offset 12
+; X32-NEXT:    pushl %edi
+; X32-NEXT:    .cfi_def_cfa_offset 16
+; X32-NEXT:    pushl %esi
+; X32-NEXT:    .cfi_def_cfa_offset 20
+; X32-NEXT:    subl $12, %esp
+; X32-NEXT:    .cfi_def_cfa_offset 32
+; X32-NEXT:    .cfi_offset %esi, -20
+; X32-NEXT:    .cfi_offset %edi, -16
+; X32-NEXT:    .cfi_offset %ebx, -12
+; X32-NEXT:    .cfi_offset %ebp, -8
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X32-NEXT:    pushl $0
+; X32-NEXT:    .cfi_adjust_cfa_offset 4
+; X32-NEXT:    pushl $-294967296 # imm = 0xEE6B2800
+; X32-NEXT:    .cfi_adjust_cfa_offset 4
+; X32-NEXT:    pushl %ebp
+; X32-NEXT:    .cfi_adjust_cfa_offset 4
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    .cfi_adjust_cfa_offset 4
+; X32-NEXT:    calll __udivdi3
+; X32-NEXT:    addl $16, %esp
+; X32-NEXT:    .cfi_adjust_cfa_offset -16
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    pushl $0
+; X32-NEXT:    .cfi_adjust_cfa_offset 4
+; X32-NEXT:    pushl $-294967296 # imm = 0xEE6B2800
+; X32-NEXT:    .cfi_adjust_cfa_offset 4
+; X32-NEXT:    pushl %ebp
+; X32-NEXT:    .cfi_adjust_cfa_offset 4
+; X32-NEXT:    pushl %ebx
+; X32-NEXT:    .cfi_adjust_cfa_offset 4
+; X32-NEXT:    calll __umoddi3
+; X32-NEXT:    addl $16, %esp
+; X32-NEXT:    .cfi_adjust_cfa_offset -16
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %edi, %edx
+; X32-NEXT:    addl $12, %esp
+; X32-NEXT:    .cfi_def_cfa_offset 20
+; X32-NEXT:    popl %esi
+; X32-NEXT:    .cfi_def_cfa_offset 16
+; X32-NEXT:    popl %edi
+; X32-NEXT:    .cfi_def_cfa_offset 12
+; X32-NEXT:    popl %ebx
+; X32-NEXT:    .cfi_def_cfa_offset 8
+; X32-NEXT:    popl %ebp
+; X32-NEXT:    .cfi_def_cfa_offset 4
+; X32-NEXT:    retl
+;
+; X64-LABEL: PR38622:
+; X64:       # %bb.0:
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    shrq $11, %rax
+; X64-NEXT:    movabsq $4835703278458517, %rcx # imm = 0x112E0BE826D695
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    shrq $9, %rdx
+; X64-NEXT:    imull $-294967296, %edx, %eax # imm = 0xEE6B2800
+; X64-NEXT:    subl %eax, %edi
+; X64-NEXT:    movq %rdx, %rax
+; X64-NEXT:    movl %edi, %edx
+; X64-NEXT:    retq
+  %2 = udiv i64 %0, 4000000000
+  %3 = urem i64 %0, 4000000000
+  %4 = trunc i64 %3 to i32
+  %5 = insertvalue { i64, i32 } undef, i64 %2, 0
+  %6 = insertvalue { i64, i32 } %5, i32 %4, 1
+  ret { i64, i32 } %6
+}