[llvm] [CodeGenPrepare] Folding `urem` with loop invariant value as remainder (PR #96625)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Jul 25 11:17:39 PDT 2024
https://github.com/goldsteinn updated https://github.com/llvm/llvm-project/pull/96625
>From 53916468e6ce9c9cac158c49771d322526025c97 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Tue, 25 Jun 2024 19:50:51 +0800
Subject: [PATCH 1/4] [CodeGenPrepare][X86] Add tests for folding `urem` with
loop invariant value; NFC
---
llvm/test/CodeGen/X86/fold-loop-of-urem.ll | 1159 +++++++++++++++++
.../CodeGenPrepare/X86/fold-loop-of-urem.ll | 858 ++++++++++++
2 files changed, 2017 insertions(+)
create mode 100644 llvm/test/CodeGen/X86/fold-loop-of-urem.ll
create mode 100644 llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll
diff --git a/llvm/test/CodeGen/X86/fold-loop-of-urem.ll b/llvm/test/CodeGen/X86/fold-loop-of-urem.ll
new file mode 100644
index 0000000000000..aad2e0dd7bd24
--- /dev/null
+++ b/llvm/test/CodeGen/X86/fold-loop-of-urem.ll
@@ -0,0 +1,1159 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
+
+declare void @use.i32(i32)
+declare void @use.2xi64(<2 x i64>)
+declare void @do_stuff0()
+declare void @do_stuff1()
+declare i1 @get.i1()
+declare i32 @get.i32()
+
+define void @simple_urem_to_sel(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_to_sel:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB0_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB0_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: jne .LBB0_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB0_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_nested2(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_to_sel_nested2:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB1_8
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: jmp .LBB1_2
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB1_5: # %for.body1
+; CHECK-NEXT: # in Loop: Header=BB1_2 Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: .LBB1_6: # %for.body.tail
+; CHECK-NEXT: # in Loop: Header=BB1_2 Depth=1
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: je .LBB1_7
+; CHECK-NEXT: .LBB1_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: testb $1, %al
+; CHECK-NEXT: je .LBB1_6
+; CHECK-NEXT: # %bb.3: # %for.body0
+; CHECK-NEXT: # in Loop: Header=BB1_2 Depth=1
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: testb $1, %al
+; CHECK-NEXT: jne .LBB1_5
+; CHECK-NEXT: # %bb.4: # %for.body2
+; CHECK-NEXT: # in Loop: Header=BB1_2 Depth=1
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: testb $1, %al
+; CHECK-NEXT: jne .LBB1_5
+; CHECK-NEXT: jmp .LBB1_6
+; CHECK-NEXT: .LBB1_7:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB1_8: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body.tail ], [ 0, %entry ]
+ %cond0 = call i1 @get.i1()
+ br i1 %cond0, label %for.body0, label %for.body.tail
+for.body0:
+ %cond1 = call i1 @get.i1()
+ br i1 %cond1, label %for.body1, label %for.body2
+for.body2:
+ %cond2 = call i1 @get.i1()
+ br i1 %cond2, label %for.body1, label %for.body.tail
+for.body1:
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ br label %for.body.tail
+for.body.tail:
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_bad_incr3(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_fail_bad_incr3:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB2_9
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: jmp .LBB2_2
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB2_6: # %for.body1
+; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1
+; CHECK-NEXT: movl %ebp, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: .LBB2_7: # %for.body.tail
+; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: testb $1, %al
+; CHECK-NEXT: jne .LBB2_8
+; CHECK-NEXT: .LBB2_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: testb $1, %al
+; CHECK-NEXT: je .LBB2_5
+; CHECK-NEXT: # %bb.3: # %for.body0
+; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: movl %eax, %r14d
+; CHECK-NEXT: callq get.i32 at PLT
+; CHECK-NEXT: testb $1, %r14b
+; CHECK-NEXT: je .LBB2_7
+; CHECK-NEXT: # %bb.4: # in Loop: Header=BB2_2 Depth=1
+; CHECK-NEXT: movl %eax, %ebp
+; CHECK-NEXT: incl %ebp
+; CHECK-NEXT: jmp .LBB2_6
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB2_5: # %for.body2
+; CHECK-NEXT: # in Loop: Header=BB2_2 Depth=1
+; CHECK-NEXT: xorl %ebp, %ebp
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: testb $1, %al
+; CHECK-NEXT: jne .LBB2_6
+; CHECK-NEXT: jmp .LBB2_7
+; CHECK-NEXT: .LBB2_8:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB2_9: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %cond0 = call i1 @get.i1()
+ br i1 %cond0, label %for.body0, label %for.body2
+for.body0:
+ %cond1 = call i1 @get.i1()
+ %val = call i32 @get.i32()
+ %inc = add nuw i32 %val, 1
+ br i1 %cond1, label %for.body1, label %for.body.tail
+for.body2:
+ %cond2 = call i1 @get.i1()
+ br i1 %cond2, label %for.body1, label %for.body.tail
+for.body1:
+ %i.04 = phi i32 [ %inc, %for.body0], [ 0, %for.body2 ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ br label %for.body.tail
+for.body.tail:
+ %exitcond.not = call i1 @get.i1()
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_vec(<2 x i64> %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_to_sel_vec:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: subq $24, %rsp
+; CHECK-NEXT: pxor %xmm1, %xmm1
+; CHECK-NEXT: movdqa %xmm1, (%rsp) # 16-byte Spill
+; CHECK-NEXT: movq %xmm0, %rbx
+; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
+; CHECK-NEXT: movq %xmm0, %r14
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB3_1: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movdqa (%rsp), %xmm1 # 16-byte Reload
+; CHECK-NEXT: movdqa %xmm1, (%rsp) # 16-byte Spill
+; CHECK-NEXT: movq %xmm1, %rax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divq %rbx
+; CHECK-NEXT: movq %rdx, %xmm0
+; CHECK-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
+; CHECK-NEXT: movq %xmm1, %rax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divq %r14
+; CHECK-NEXT: movq %rdx, %xmm1
+; CHECK-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; CHECK-NEXT: callq use.2xi64 at PLT
+; CHECK-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload
+; CHECK-NEXT: psubq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; CHECK-NEXT: movdqa %xmm0, (%rsp) # 16-byte Spill
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: testb $1, %al
+; CHECK-NEXT: je .LBB3_1
+; CHECK-NEXT: # %bb.2: # %for.cond.cleanup
+; CHECK-NEXT: addq $24, %rsp
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: retq
+entry:
+ br label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi <2 x i64> [ %inc, %for.body ], [ zeroinitializer, %entry ]
+ %rem = urem <2 x i64> %i.04, %rem_amt
+ tail call void @use.2xi64(<2 x i64> %rem)
+ %inc = add nuw <2 x i64> %i.04, <i64 1, i64 1>
+ %exitcond.not = call i1 @get.i1()
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_bad_incr(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_fail_bad_incr:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB4_6
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: jmp .LBB4_2
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB4_4: # %for.body.tail
+; CHECK-NEXT: # in Loop: Header=BB4_2 Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %ebp, %r14d
+; CHECK-NEXT: je .LBB4_5
+; CHECK-NEXT: .LBB4_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: testb $1, %al
+; CHECK-NEXT: je .LBB4_4
+; CHECK-NEXT: # %bb.3: # %for.body0
+; CHECK-NEXT: # in Loop: Header=BB4_2 Depth=1
+; CHECK-NEXT: callq get.i32 at PLT
+; CHECK-NEXT: movl %eax, %r14d
+; CHECK-NEXT: jmp .LBB4_4
+; CHECK-NEXT: .LBB4_5:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB4_6: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.03 = phi i32 [ %inc, %for.body.tail ], [ 0, %entry ]
+ %cond0 = call i1 @get.i1()
+ br i1 %cond0, label %for.body0, label %for.body.tail
+for.body0:
+ %some_val = call i32 @get.i32()
+ br label %for.body.tail
+
+for.body.tail:
+ %i.04 = phi i32 [ %i.03, %for.body ], [ %some_val, %for.body0 ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_second_acc(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_to_sel_second_acc:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: cmpl $2, %edi
+; CHECK-NEXT: jb .LBB5_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r15
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: pushq %rax
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: movl $1, %r15d
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB5_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: addl $2, %r15d
+; CHECK-NEXT: cmpl %ebp, %r15d
+; CHECK-NEXT: jbe .LBB5_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: addq $8, %rsp
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB5_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp ult i32 %N, 2
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %i.05 = phi i32 [ %inc2, %for.body ], [ 1, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %inc2 = add nuw i32 %i.05, 2
+ %exitcond.not = icmp ugt i32 %inc2, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_srem(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_fail_srem:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB6_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB6_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: cltd
+; CHECK-NEXT: idivl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: jne .LBB6_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB6_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %rem = srem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_missing_nuw(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_fail_missing_nuw:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB7_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB7_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: jne .LBB7_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB7_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nsw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_bad_incr2(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_fail_bad_incr2:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB8_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB8_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: addl $2, %r14d
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: jne .LBB8_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB8_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 2
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_non_zero_entry4(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_non_zero_entry4:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB9_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: movl $4, %r14d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB9_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: jne .LBB9_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB9_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 4, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_skip_const_rem_amt(i32 %N) nounwind {
+; CHECK-LABEL: simple_urem_skip_const_rem_amt:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB10_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %edi, %ebx
+; CHECK-NEXT: addl $-4, %ebx
+; CHECK-NEXT: movl $4, %ebp
+; CHECK-NEXT: movl $2938661835, %r14d # imm = 0xAF286BCB
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB10_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %ebp, %eax
+; CHECK-NEXT: imulq %r14, %rax
+; CHECK-NEXT: shrq $32, %rax
+; CHECK-NEXT: movl %ebp, %ecx
+; CHECK-NEXT: subl %eax, %ecx
+; CHECK-NEXT: shrl %ecx
+; CHECK-NEXT: addl %eax, %ecx
+; CHECK-NEXT: shrl $4, %ecx
+; CHECK-NEXT: leal (%rcx,%rcx,8), %eax
+; CHECK-NEXT: leal (%rcx,%rax,2), %eax
+; CHECK-NEXT: movl %ebp, %edi
+; CHECK-NEXT: subl %eax, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %ebp
+; CHECK-NEXT: decl %ebx
+; CHECK-NEXT: jne .LBB10_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB10_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 4, %entry ]
+ %rem = urem i32 %i.04, 19
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_no_preheader_non_canonical(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_fail_no_preheader_non_canonical:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB11_1
+; CHECK-NEXT: # %bb.2: # %for.body1
+; CHECK-NEXT: movl $1, %r14d
+; CHECK-NEXT: jmp .LBB11_3
+; CHECK-NEXT: .LBB11_1:
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB11_3: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: jne .LBB11_3
+; CHECK-NEXT: # %bb.4: # %for.cond.cleanup
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.body0, label %for.body1
+
+for.cond.cleanup:
+ ret void
+
+for.body0:
+ br label %for.body
+
+for.body1:
+ br label %for.body
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %for.body0 ], [ 1, %for.body1 ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_multi_latch_non_canonical(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_multi_latch_non_canonical:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB12_6
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r15
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: pushq %rax
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: decl %ebp
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: jmp .LBB12_2
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB12_3: # %for.body.backedge
+; CHECK-NEXT: # in Loop: Header=BB12_2 Depth=1
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: .LBB12_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: callq get.i1 at PLT
+; CHECK-NEXT: movl %eax, %r15d
+; CHECK-NEXT: callq do_stuff0 at PLT
+; CHECK-NEXT: testb $1, %r15b
+; CHECK-NEXT: je .LBB12_3
+; CHECK-NEXT: # %bb.4: # %for.body0
+; CHECK-NEXT: # in Loop: Header=BB12_2 Depth=1
+; CHECK-NEXT: callq do_stuff1 at PLT
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: jne .LBB12_3
+; CHECK-NEXT: # %bb.5:
+; CHECK-NEXT: addq $8, %rsp
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB12_6: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ %inc, %for.body0 ], [ 0, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %cond = call i1 @get.i1()
+ call void @do_stuff0()
+ br i1 %cond, label %for.body0, label %for.body
+for.body0:
+ call void @do_stuff1()
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_bad_loop(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_fail_bad_loop:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: callq get.i32 at PLT
+; CHECK-NEXT: testl %eax, %eax
+; CHECK-NEXT: # implicit-def: $r14d
+; CHECK-NEXT: jne .LBB13_4
+; CHECK-NEXT: # %bb.1:
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: .LBB13_2: # %for.cond
+; CHECK-NEXT: cmpl %ebp, %r14d
+; CHECK-NEXT: jae .LBB13_5
+; CHECK-NEXT: # %bb.3: # %for.body
+; CHECK-NEXT: movl %r14d, %edi
+; CHECK-NEXT: xorl $1, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: .LBB13_4: # %halfway
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: jmp .LBB13_2
+; CHECK-NEXT: .LBB13_5: # %for.end
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: retq
+entry:
+ %call = call i32 @get.i32()
+ %tobool.not = icmp eq i32 %call, 0
+ br i1 %tobool.not, label %for.cond, label %halfway
+
+for.cond:
+ %i.0 = phi i32 [ %inc, %halfway ], [ 0, %entry ]
+ %cmp = icmp ult i32 %i.0, %N
+ br i1 %cmp, label %for.body, label %for.end
+
+for.body:
+ %xor = xor i32 %i.0, 1
+ call void @use.i32(i32 %xor)
+ br label %halfway
+
+halfway:
+ %i.1 = phi i32 [ poison, %entry ], [ %i.0, %for.body ]
+ %rem = urem i32 %i.1, %rem_amt
+ call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.1, 1
+ br label %for.cond
+
+for.end:
+ ret void
+}
+
+define void @simple_urem_fail_intermediate_inc(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_fail_intermediate_inc:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: testl %edi, %edi
+; CHECK-NEXT: je .LBB14_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %r15
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %r14d
+; CHECK-NEXT: negl %r14d
+; CHECK-NEXT: movl $1, %r15d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB14_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r15d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: leal 1(%r14,%r15), %eax
+; CHECK-NEXT: movl %r15d, %ecx
+; CHECK-NEXT: incl %ecx
+; CHECK-NEXT: cmpl $1, %eax
+; CHECK-NEXT: movl %ecx, %r15d
+; CHECK-NEXT: jne .LBB14_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
+; CHECK-NEXT: .LBB14_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %inc2 = add nuw i32 %i.04, 1
+ %rem = urem i32 %inc2, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @weird_loop(i64 %sub.ptr.div.i56) personality ptr null {
+; CHECK-LABEL: weird_loop:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB15_1: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: jmp .LBB15_1
+entry:
+ br label %for.preheader
+
+for.preheader:
+ %i57.0540.us = phi i64 [ 0, %entry ], [ %add74.us, %for.body ]
+ %add74.us = add nuw i64 %i57.0540.us, 1
+ br label %for.body
+
+for.body:
+ %rem.us = urem i64 %i57.0540.us, %sub.ptr.div.i56
+ br i1 false, label %for.preheader, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_to_sel_non_zero_start:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: cmpl $3, %edi
+; CHECK-NEXT: jb .LBB16_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: movl $2, %r14d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB16_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: jne .LBB16_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: .LBB16_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp ult i32 %N, 3
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 2, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_add(i32 %N, i32 %rem_amt_in) nounwind {
+; CHECK-LABEL: simple_urem_to_sel_non_zero_start_through_add:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: cmpl $3, %edi
+; CHECK-NEXT: jb .LBB17_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %r15
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %r14d
+; CHECK-NEXT: orl $16, %ebx
+; CHECK-NEXT: negl %r14d
+; CHECK-NEXT: movl $7, %r15d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB17_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r15d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: leal 1(%r14,%r15), %eax
+; CHECK-NEXT: movl %r15d, %ecx
+; CHECK-NEXT: incl %ecx
+; CHECK-NEXT: cmpl $5, %eax
+; CHECK-NEXT: movl %ecx, %r15d
+; CHECK-NEXT: jne .LBB17_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
+; CHECK-NEXT: .LBB17_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %rem_amt = or i32 %rem_amt_in, 16
+ %cmp3.not = icmp ult i32 %N, 3
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 2, %entry ]
+ %i_with_off = add nuw i32 %i.04, 5
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_add_fail_missing_nuw(i32 %N, i32 %rem_amt_in) nounwind {
+; CHECK-LABEL: simple_urem_to_sel_non_zero_start_through_add_fail_missing_nuw:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: cmpl $3, %edi
+; CHECK-NEXT: jb .LBB18_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %r15
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %r14d
+; CHECK-NEXT: orl $16, %ebx
+; CHECK-NEXT: negl %r14d
+; CHECK-NEXT: movl $7, %r15d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB18_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r15d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: leal 1(%r14,%r15), %eax
+; CHECK-NEXT: movl %r15d, %ecx
+; CHECK-NEXT: incl %ecx
+; CHECK-NEXT: cmpl $5, %eax
+; CHECK-NEXT: movl %ecx, %r15d
+; CHECK-NEXT: jne .LBB18_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
+; CHECK-NEXT: .LBB18_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %rem_amt = or i32 %rem_amt_in, 16
+ %cmp3.not = icmp ult i32 %N, 3
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 2, %entry ]
+ %i_with_off = add i32 %i.04, 5
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_add_fail_no_simplify_rem(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: simple_urem_to_sel_non_zero_start_through_add_fail_no_simplify_rem:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: cmpl $3, %edi
+; CHECK-NEXT: jb .LBB19_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %r15
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %r14d
+; CHECK-NEXT: negl %r14d
+; CHECK-NEXT: movl $7, %r15d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB19_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r15d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: leal 1(%r14,%r15), %eax
+; CHECK-NEXT: movl %r15d, %ecx
+; CHECK-NEXT: incl %ecx
+; CHECK-NEXT: cmpl $5, %eax
+; CHECK-NEXT: movl %ecx, %r15d
+; CHECK-NEXT: jne .LBB19_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
+; CHECK-NEXT: .LBB19_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp ult i32 %N, 3
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 2, %entry ]
+ %i_with_off = add nuw i32 %i.04, 5
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_sub(i32 %N, i32 %rem_amt, i32 %start) nounwind {
+; CHECK-LABEL: simple_urem_to_sel_non_zero_start_through_sub:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: subl %edx, %ebp
+; CHECK-NEXT: jbe .LBB20_3
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB20_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r14d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: jne .LBB20_2
+; CHECK-NEXT: .LBB20_3: # %for.cond.cleanup
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %rbp
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp ule i32 %N, %start
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ %start, %entry ]
+ %i_with_off = sub nuw i32 %i.04, %start
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_sub_no_simplfy(i32 %N, i32 %rem_amt, i32 %start) nounwind {
+; CHECK-LABEL: simple_urem_to_sel_non_zero_start_through_sub_no_simplfy:
+; CHECK: # %bb.0: # %entry
+; CHECK-NEXT: cmpl %edx, %edi
+; CHECK-NEXT: jbe .LBB21_4
+; CHECK-NEXT: # %bb.1: # %for.body.preheader
+; CHECK-NEXT: pushq %r15
+; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %rbx
+; CHECK-NEXT: movl %edx, %r15d
+; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: movl %edi, %r14d
+; CHECK-NEXT: negl %r14d
+; CHECK-NEXT: addl $-2, %r15d
+; CHECK-NEXT: .p2align 4, 0x90
+; CHECK-NEXT: .LBB21_2: # %for.body
+; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
+; CHECK-NEXT: movl %r15d, %eax
+; CHECK-NEXT: xorl %edx, %edx
+; CHECK-NEXT: divl %ebx
+; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: callq use.i32 at PLT
+; CHECK-NEXT: leal 1(%r14,%r15), %eax
+; CHECK-NEXT: movl %r15d, %ecx
+; CHECK-NEXT: incl %ecx
+; CHECK-NEXT: cmpl $-2, %eax
+; CHECK-NEXT: movl %ecx, %r15d
+; CHECK-NEXT: jne .LBB21_2
+; CHECK-NEXT: # %bb.3:
+; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
+; CHECK-NEXT: .LBB21_4: # %for.cond.cleanup
+; CHECK-NEXT: retq
+entry:
+ %cmp3.not = icmp ule i32 %N, %start
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ %start, %entry ]
+ %i_with_off = sub nuw i32 %i.04, 2
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll b/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll
new file mode 100644
index 0000000000000..a019679e65905
--- /dev/null
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll
@@ -0,0 +1,858 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
+; RUN: opt < %s -mtriple=x86_64-unknown-unknown --loop-simplify -codegenprepare -S | FileCheck %s
+
+declare void @use.i32(i32)
+declare void @use.2xi64(<2 x i64>)
+declare void @do_stuff0()
+declare void @do_stuff1()
+declare i1 @get.i1()
+declare i32 @get.i32()
+
+define void @simple_urem_to_sel(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0:[0-9]+]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_nested2(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel_nested2(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY_TAIL:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[COND0:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: br i1 [[COND0]], label %[[FOR_BODY0:.*]], label %[[FOR_BODY_TAIL]]
+; CHECK: [[FOR_BODY0]]:
+; CHECK-NEXT: [[COND1:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: br i1 [[COND1]], label %[[FOR_BODY1:.*]], label %[[FOR_BODY2:.*]]
+; CHECK: [[FOR_BODY2]]:
+; CHECK-NEXT: [[COND2:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: br i1 [[COND2]], label %[[FOR_BODY1]], label %[[FOR_BODY_TAIL]]
+; CHECK: [[FOR_BODY1]]:
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: br label %[[FOR_BODY_TAIL]]
+; CHECK: [[FOR_BODY_TAIL]]:
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body.tail ], [ 0, %entry ]
+ %cond0 = call i1 @get.i1()
+ br i1 %cond0, label %for.body0, label %for.body.tail
+for.body0:
+ %cond1 = call i1 @get.i1()
+ br i1 %cond1, label %for.body1, label %for.body2
+for.body2:
+ %cond2 = call i1 @get.i1()
+ br i1 %cond2, label %for.body1, label %for.body.tail
+for.body1:
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ br label %for.body.tail
+for.body.tail:
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_bad_incr3(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_fail_bad_incr3(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[COND0:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: br i1 [[COND0]], label %[[FOR_BODY0:.*]], label %[[FOR_BODY2:.*]]
+; CHECK: [[FOR_BODY0]]:
+; CHECK-NEXT: [[COND1:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: [[VAL:%.*]] = call i32 @get.i32()
+; CHECK-NEXT: [[INC:%.*]] = add nuw i32 [[VAL]], 1
+; CHECK-NEXT: br i1 [[COND1]], label %[[FOR_BODY1:.*]], label %[[FOR_BODY_TAIL:.*]]
+; CHECK: [[FOR_BODY2]]:
+; CHECK-NEXT: [[COND2:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: br i1 [[COND2]], label %[[FOR_BODY1]], label %[[FOR_BODY_TAIL]]
+; CHECK: [[FOR_BODY1]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC]], %[[FOR_BODY0]] ], [ 0, %[[FOR_BODY2]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: br label %[[FOR_BODY_TAIL]]
+; CHECK: [[FOR_BODY_TAIL]]:
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %cond0 = call i1 @get.i1()
+ br i1 %cond0, label %for.body0, label %for.body2
+for.body0:
+ %cond1 = call i1 @get.i1()
+ %val = call i32 @get.i32()
+ %inc = add nuw i32 %val, 1
+ br i1 %cond1, label %for.body1, label %for.body.tail
+for.body2:
+ %cond2 = call i1 @get.i1()
+ br i1 %cond2, label %for.body1, label %for.body.tail
+for.body1:
+ %i.04 = phi i32 [ %inc, %for.body0], [ 0, %for.body2 ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ br label %for.body.tail
+for.body.tail:
+ %exitcond.not = call i1 @get.i1()
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_vec(<2 x i64> %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel_vec(
+; CHECK-SAME: <2 x i64> [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP:.*]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi <2 x i64> [ [[INC:%.*]], %[[FOR_BODY]] ], [ zeroinitializer, %[[ENTRY]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem <2 x i64> [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.2xi64(<2 x i64> [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw <2 x i64> [[I_04]], <i64 1, i64 1>
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ br label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi <2 x i64> [ %inc, %for.body ], [ zeroinitializer, %entry ]
+ %rem = urem <2 x i64> %i.04, %rem_amt
+ tail call void @use.2xi64(<2 x i64> %rem)
+ %inc = add nuw <2 x i64> %i.04, <i64 1, i64 1>
+ %exitcond.not = call i1 @get.i1()
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_bad_incr(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_fail_bad_incr(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_03:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY_TAIL:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[COND0:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: br i1 [[COND0]], label %[[FOR_BODY0:.*]], label %[[FOR_BODY_TAIL]]
+; CHECK: [[FOR_BODY0]]:
+; CHECK-NEXT: [[SOME_VAL:%.*]] = call i32 @get.i32()
+; CHECK-NEXT: br label %[[FOR_BODY_TAIL]]
+; CHECK: [[FOR_BODY_TAIL]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[I_03]], %[[FOR_BODY]] ], [ [[SOME_VAL]], %[[FOR_BODY0]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.03 = phi i32 [ %inc, %for.body.tail ], [ 0, %entry ]
+ %cond0 = call i1 @get.i1()
+ br i1 %cond0, label %for.body0, label %for.body.tail
+for.body0:
+ %some_val = call i32 @get.i32()
+ br label %for.body.tail
+
+for.body.tail:
+ %i.04 = phi i32 [ %i.03, %for.body ], [ %some_val, %for.body0 ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_second_acc(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel_second_acc(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp ult i32 [[N]], 2
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[I_05:%.*]] = phi i32 [ [[INC2:%.*]], %[[FOR_BODY]] ], [ 1, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[INC2]] = add nuw i32 [[I_05]], 2
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp ugt i32 [[INC2]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp ult i32 %N, 2
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %i.05 = phi i32 [ %inc2, %for.body ], [ 1, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %inc2 = add nuw i32 %i.05, 2
+ %exitcond.not = icmp ugt i32 %inc2, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_srem(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_fail_srem(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = srem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %rem = srem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_missing_nuw(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_fail_missing_nuw(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nsw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nsw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_bad_incr2(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_fail_bad_incr2(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 2
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 2
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_non_zero_entry4(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_non_zero_entry4(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 4, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 4, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_skip_const_rem_amt(i32 %N) nounwind {
+; CHECK-LABEL: define void @simple_urem_skip_const_rem_amt(
+; CHECK-SAME: i32 [[N:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 4, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], 19
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 4, %entry ]
+ %rem = urem i32 %i.04, 19
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_no_preheader_non_canonical(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_fail_no_preheader_non_canonical(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_BODY_PREHEADER:.*]], label %[[FOR_BODY1:.*]]
+; CHECK: [[FOR_COND_CLEANUP:.*]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY1]]:
+; CHECK-NEXT: br label %[[FOR_BODY_PREHEADER]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: [[I_04_PH:%.*]] = phi i32 [ 1, %[[FOR_BODY1]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[I_04_PH]], %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.body0, label %for.body1
+
+for.cond.cleanup:
+ ret void
+
+for.body0:
+ br label %for.body
+
+for.body1:
+ br label %for.body
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %for.body0 ], [ 1, %for.body1 ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_multi_latch_non_canonical(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_multi_latch_non_canonical(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ 0, %[[FOR_BODY_PREHEADER]] ], [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[INC]], %[[FOR_BODY0:.*]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[COND:%.*]] = call i1 @get.i1()
+; CHECK-NEXT: call void @do_stuff0()
+; CHECK-NEXT: br i1 [[COND]], label %[[FOR_BODY0]], label %[[FOR_BODY]]
+; CHECK: [[FOR_BODY0]]:
+; CHECK-NEXT: call void @do_stuff1()
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ %inc, %for.body0 ], [ 0, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %cond = call i1 @get.i1()
+ call void @do_stuff0()
+ br i1 %cond, label %for.body0, label %for.body
+for.body0:
+ call void @do_stuff1()
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_fail_bad_loop(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_fail_bad_loop(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*]]:
+; CHECK-NEXT: [[CALL:%.*]] = call i32 @get.i32()
+; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[CALL]], 0
+; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label %[[FOR_COND:.*]], label %[[HALFWAY:.*]]
+; CHECK: [[FOR_COND]]:
+; CHECK-NEXT: [[I_0:%.*]] = phi i32 [ [[INC:%.*]], %[[HALFWAY]] ], [ 0, %[[ENTRY]] ]
+; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[I_0]], [[N]]
+; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY:.*]], label %[[FOR_END:.*]]
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[XOR:%.*]] = xor i32 [[I_0]], 1
+; CHECK-NEXT: call void @use.i32(i32 [[XOR]])
+; CHECK-NEXT: br label %[[HALFWAY]]
+; CHECK: [[HALFWAY]]:
+; CHECK-NEXT: [[I_1:%.*]] = phi i32 [ poison, %[[ENTRY]] ], [ [[I_0]], %[[FOR_BODY]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_1]], [[REM_AMT]]
+; CHECK-NEXT: call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_1]], 1
+; CHECK-NEXT: br label %[[FOR_COND]]
+; CHECK: [[FOR_END]]:
+; CHECK-NEXT: ret void
+;
+entry:
+ %call = call i32 @get.i32()
+ %tobool.not = icmp eq i32 %call, 0
+ br i1 %tobool.not, label %for.cond, label %halfway
+
+for.cond:
+ %i.0 = phi i32 [ %inc, %halfway ], [ 0, %entry ]
+ %cmp = icmp ult i32 %i.0, %N
+ br i1 %cmp, label %for.body, label %for.end
+
+for.body:
+ %xor = xor i32 %i.0, 1
+ call void @use.i32(i32 %xor)
+ br label %halfway
+
+halfway:
+ %i.1 = phi i32 [ poison, %entry ], [ %i.0, %for.body ]
+ %rem = urem i32 %i.1, %rem_amt
+ call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.1, 1
+ br label %for.cond
+
+for.end:
+ ret void
+}
+
+define void @simple_urem_fail_intermediate_inc(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_fail_intermediate_inc(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp eq i32 [[N]], 0
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[INC2:%.*]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[INC2]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp eq i32 %N, 0
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 0, %entry ]
+ %inc2 = add nuw i32 %i.04, 1
+ %rem = urem i32 %inc2, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @weird_loop(i64 %sub.ptr.div.i56) personality ptr null {
+; CHECK-LABEL: define void @weird_loop(
+; CHECK-SAME: i64 [[SUB_PTR_DIV_I56:%.*]]) personality ptr null {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[ADD74_US:%.*]] = add nuw i64 0, 1
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[REM_US:%.*]] = urem i64 0, [[SUB_PTR_DIV_I56]]
+; CHECK-NEXT: br label %[[FOR_BODY]]
+;
+entry:
+ br label %for.preheader
+
+for.preheader:
+ %i57.0540.us = phi i64 [ 0, %entry ], [ %add74.us, %for.body ]
+ %add74.us = add nuw i64 %i57.0540.us, 1
+ br label %for.body
+
+for.body:
+ %rem.us = urem i64 %i57.0540.us, %sub.ptr.div.i56
+ br i1 false, label %for.preheader, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel_non_zero_start(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp ult i32 [[N]], 3
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 2, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp ult i32 %N, 3
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 2, %entry ]
+ %rem = urem i32 %i.04, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_add(i32 %N, i32 %rem_amt_in) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel_non_zero_start_through_add(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT_IN:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[REM_AMT:%.*]] = or i32 [[REM_AMT_IN]], 16
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp ult i32 [[N]], 3
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 2, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[I_WITH_OFF:%.*]] = add nuw i32 [[I_04]], 5
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_WITH_OFF]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %rem_amt = or i32 %rem_amt_in, 16
+ %cmp3.not = icmp ult i32 %N, 3
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 2, %entry ]
+ %i_with_off = add nuw i32 %i.04, 5
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_add_fail_missing_nuw(i32 %N, i32 %rem_amt_in) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel_non_zero_start_through_add_fail_missing_nuw(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT_IN:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[REM_AMT:%.*]] = or i32 [[REM_AMT_IN]], 16
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp ult i32 [[N]], 3
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 2, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[I_WITH_OFF:%.*]] = add i32 [[I_04]], 5
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_WITH_OFF]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %rem_amt = or i32 %rem_amt_in, 16
+ %cmp3.not = icmp ult i32 %N, 3
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 2, %entry ]
+ %i_with_off = add i32 %i.04, 5
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_add_fail_no_simplify_rem(i32 %N, i32 %rem_amt) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel_non_zero_start_through_add_fail_no_simplify_rem(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp ult i32 [[N]], 3
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 2, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[I_WITH_OFF:%.*]] = add nuw i32 [[I_04]], 5
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_WITH_OFF]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp ult i32 %N, 3
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ 2, %entry ]
+ %i_with_off = add nuw i32 %i.04, 5
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_sub(i32 %N, i32 %rem_amt, i32 %start) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel_non_zero_start_through_sub(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]], i32 [[START:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp ule i32 [[N]], [[START]]
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[START]], %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[I_WITH_OFF:%.*]] = sub nuw i32 [[I_04]], [[START]]
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_WITH_OFF]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp ule i32 %N, %start
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ %start, %entry ]
+ %i_with_off = sub nuw i32 %i.04, %start
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
+
+define void @simple_urem_to_sel_non_zero_start_through_sub_no_simplfy(i32 %N, i32 %rem_amt, i32 %start) nounwind {
+; CHECK-LABEL: define void @simple_urem_to_sel_non_zero_start_through_sub_no_simplfy(
+; CHECK-SAME: i32 [[N:%.*]], i32 [[REM_AMT:%.*]], i32 [[START:%.*]]) #[[ATTR0]] {
+; CHECK-NEXT: [[ENTRY:.*:]]
+; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp ule i32 [[N]], [[START]]
+; CHECK-NEXT: br i1 [[CMP3_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]]
+; CHECK: [[FOR_BODY_PREHEADER]]:
+; CHECK-NEXT: br label %[[FOR_BODY:.*]]
+; CHECK: [[FOR_COND_CLEANUP]]:
+; CHECK-NEXT: ret void
+; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[START]], %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[I_WITH_OFF:%.*]] = sub nuw i32 [[I_04]], 2
+; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_WITH_OFF]], [[REM_AMT]]
+; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
+;
+entry:
+ %cmp3.not = icmp ule i32 %N, %start
+ br i1 %cmp3.not, label %for.cond.cleanup, label %for.body
+
+for.cond.cleanup:
+ ret void
+
+for.body:
+ %i.04 = phi i32 [ %inc, %for.body ], [ %start, %entry ]
+ %i_with_off = sub nuw i32 %i.04, 2
+ %rem = urem i32 %i_with_off, %rem_amt
+ tail call void @use.i32(i32 %rem)
+ %inc = add nuw i32 %i.04, 1
+ %exitcond.not = icmp eq i32 %inc, %N
+ br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
+}
>From 441fa07861c3af026fdf1c1cdcd1b485a5fbf28b Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Tue, 25 Jun 2024 19:51:39 +0800
Subject: [PATCH 2/4] [CodeGenPrepare] Folding `urem` with loop invariant value
```
for(i = Start; i < End; ++i)
Rem = (i nuw+ IncrLoopInvariant) u% RemAmtLoopInvariant;
```
->
```
Rem = (Start nuw+ IncrLoopInvariant) % RemAmtLoopInvariant;
for(i = Start; i < End; ++i, ++rem)
Rem = rem == RemAmtLoopInvariant ? 0 : Rem;
```
In its current state, only if `IncrLoopInvariant` and `Start` both
being zero.
Alive2 seemed unable to prove this (see:
https://alive2.llvm.org/ce/z/ATGDp3 which is clearly wrong but still
checks out...) so wrote an exhaustive test here:
https://godbolt.org/z/WYa561388
---
llvm/lib/CodeGen/CodeGenPrepare.cpp | 160 ++++++++++++++++++
llvm/test/CodeGen/X86/fold-loop-of-urem.ll | 106 +++++++-----
.../CodeGenPrepare/X86/fold-loop-of-urem.ll | 22 ++-
3 files changed, 240 insertions(+), 48 deletions(-)
diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp
index 22d0708f54786..e0232afa4d917 100644
--- a/llvm/lib/CodeGen/CodeGenPrepare.cpp
+++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp
@@ -471,6 +471,7 @@ class CodeGenPrepare {
bool replaceMathCmpWithIntrinsic(BinaryOperator *BO, Value *Arg0, Value *Arg1,
CmpInst *Cmp, Intrinsic::ID IID);
bool optimizeCmp(CmpInst *Cmp, ModifyDT &ModifiedDT);
+ bool optimizeRem(Instruction *Rem);
bool combineToUSubWithOverflow(CmpInst *Cmp, ModifyDT &ModifiedDT);
bool combineToUAddWithOverflow(CmpInst *Cmp, ModifyDT &ModifiedDT);
void verifyBFIUpdates(Function &F);
@@ -1974,6 +1975,160 @@ static bool foldFCmpToFPClassTest(CmpInst *Cmp, const TargetLowering &TLI,
return true;
}
+static bool isRemOfLoopIncrementWithLoopInvariant(
+ Value *Rem, const LoopInfo *LI, Value *&RemAmtOut,
+ std::optional<bool> &AddOrSubOut, Value *&AddOrSubOffsetOut,
+ PHINode *&LoopIncrPNOut) {
+ Value *Incr, *RemAmt;
+ if (!isa<Instruction>(Rem))
+ return false;
+ // NB: If RemAmt is a power of 2 it *should* have been transformed by now.
+ if (!match(Rem, m_URem(m_Value(Incr), m_Value(RemAmt))))
+ return false;
+
+ // Only trivially analyzable loops.
+ Loop *L = LI->getLoopFor(cast<Instruction>(Rem)->getParent());
+ if (L == nullptr || L->getLoopPreheader() == nullptr ||
+ L->getLoopLatch() == nullptr)
+ return false;
+
+ std::optional<bool> AddOrSub;
+ Value *AddOrSubOffset;
+ // Find out loop increment PHI.
+ PHINode *PN = dyn_cast<PHINode>(Incr);
+ if (PN != nullptr) {
+ AddOrSub = std::nullopt;
+ AddOrSubOffset = nullptr;
+ } else {
+ // Search through a NUW add/sub.
+ Value *V0, *V1;
+ if (match(Incr, m_NUWAddLike(m_Value(V0), m_Value(V1))))
+ AddOrSub = true;
+ else if (match(Incr, m_NUWSub(m_Value(V0), m_Value(V1))))
+ AddOrSub = false;
+ else
+ return false;
+
+ PN = dyn_cast<PHINode>(V0);
+ if (PN != nullptr) {
+ AddOrSubOffset = V1;
+ } else if (*AddOrSub) {
+ PN = dyn_cast<PHINode>(V1);
+ AddOrSubOffset = V0;
+ }
+ }
+
+ if (PN == nullptr)
+ return false;
+
+ // This isn't strictly necessary, what we really need is one increment and any
+ // amount of initial values all being the same.
+ if (PN->getNumIncomingValues() != 2)
+ return false;
+
+ // Only works if the remainder amount is a loop invaraint
+ if (!L->isLoopInvariant(RemAmt))
+ return false;
+
+ // Is the PHI a loop increment?
+ auto LoopIncrInfo = getIVIncrement(PN, LI);
+ if (!LoopIncrInfo.has_value())
+ return false;
+
+ // We need remainder_amount % increment_amount to be zero. Increment of one
+ // satisfies that without any special logic and is overwhelmingly the common
+ // case.
+ if (!match(LoopIncrInfo->second, m_One()))
+ return false;
+
+ // Need the increment to not overflow.
+ if (!match(LoopIncrInfo->first, m_NUWAdd(m_Value(), m_Value())))
+ return false;
+
+ if (PN->getBasicBlockIndex(L->getLoopLatch()) < 0 ||
+ PN->getBasicBlockIndex(L->getLoopPreheader()) < 0)
+ return false;
+
+ // Set output variables.
+ RemAmtOut = RemAmt;
+ LoopIncrPNOut = PN;
+ AddOrSubOut = AddOrSub;
+ AddOrSubOffsetOut = AddOrSubOffset;
+
+ return true;
+}
+
+// Try to transform:
+//
+// for(i = Start; i < End; ++i)
+// Rem = (i nuw+ IncrLoopInvariant) u% RemAmtLoopInvariant;
+//
+// ->
+//
+// Rem = (Start nuw+ IncrLoopInvariant) % RemAmtLoopInvariant;
+// for(i = Start; i < End; ++i, ++rem)
+// Rem = rem == RemAmtLoopInvariant ? 0 : Rem;
+//
+// Currently only implemented for `Start` and `IncrLoopInvariant` being zero.
+static bool foldURemOfLoopIncrement(Instruction *Rem, const LoopInfo *LI,
+ SmallSet<BasicBlock *, 32> &FreshBBs,
+ bool IsHuge) {
+ std::optional<bool> AddOrSub;
+ Value *AddOrSubOffset, *RemAmt;
+ PHINode *LoopIncrPN;
+ if (!isRemOfLoopIncrementWithLoopInvariant(Rem, LI, RemAmt, AddOrSub,
+ AddOrSubOffset, LoopIncrPN))
+ return false;
+
+ // Only non-constant remainder as the extra IV is is probably not profitable
+ // in that case. Further, since remainder amount is non-constant, only handle
+ // case where `IncrLoopInvariant` and `Start` are 0 to entirely eliminate the
+ // rem (as opposed to just hoisting it outside of the loop).
+ //
+ // Potential TODO: Should we have a check for how "nested" this remainder
+ // operation is? The new code runs every iteration so if the remainder is
+ // guarded behind unlikely conditions this might not be worth it.
+ if (AddOrSub.has_value() || match(RemAmt, m_ImmConstant()))
+ return false;
+ Loop *L = LI->getLoopFor(Rem->getParent());
+ if (!match(LoopIncrPN->getIncomingValueForBlock(L->getLoopPreheader()),
+ m_Zero()))
+ return false;
+
+ // Create new remainder with induction variable.
+ Type *Ty = Rem->getType();
+ IRBuilder<> Builder(Rem->getContext());
+
+ Builder.SetInsertPoint(LoopIncrPN);
+ PHINode *NewRem = Builder.CreatePHI(Ty, 2);
+
+ Builder.SetInsertPoint(cast<Instruction>(
+ LoopIncrPN->getIncomingValueForBlock(L->getLoopLatch())));
+ // `(add (urem x, y), 1)` is always nuw.
+ Value *RemAdd = Builder.CreateNUWAdd(NewRem, ConstantInt::get(Ty, 1));
+ Value *RemCmp = Builder.CreateICmp(ICmpInst::ICMP_EQ, RemAdd, RemAmt);
+ Value *RemSel =
+ Builder.CreateSelect(RemCmp, Constant::getNullValue(Ty), RemAdd);
+
+ NewRem->addIncoming(Constant::getNullValue(Ty), L->getLoopPreheader());
+ NewRem->addIncoming(RemSel, L->getLoopLatch());
+
+ // Insert all touched BBs.
+ FreshBBs.insert(LoopIncrPN->getParent());
+ FreshBBs.insert(L->getLoopLatch());
+ FreshBBs.insert(Rem->getParent());
+
+ replaceAllUsesWith(Rem, NewRem, FreshBBs, IsHuge);
+ Rem->eraseFromParent();
+ return true;
+}
+
+bool CodeGenPrepare::optimizeRem(Instruction *Rem) {
+ if (foldURemOfLoopIncrement(Rem, LI, FreshBBs, IsHugeFunc))
+ return true;
+ return false;
+}
+
bool CodeGenPrepare::optimizeCmp(CmpInst *Cmp, ModifyDT &ModifiedDT) {
if (sinkCmpExpression(Cmp, *TLI))
return true;
@@ -8360,6 +8515,11 @@ bool CodeGenPrepare::optimizeInst(Instruction *I, ModifyDT &ModifiedDT) {
if (optimizeCmp(Cmp, ModifiedDT))
return true;
+ if (match(I, m_URem(m_Value(), m_Value())) ||
+ match(I, m_SRem(m_Value(), m_Value())))
+ if (optimizeRem(I))
+ return true;
+
if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
LI->setMetadata(LLVMContext::MD_invariant_group, nullptr);
bool Modified = optimizeLoadExt(LI);
diff --git a/llvm/test/CodeGen/X86/fold-loop-of-urem.ll b/llvm/test/CodeGen/X86/fold-loop-of-urem.ll
index aad2e0dd7bd24..c2c1124b1ac11 100644
--- a/llvm/test/CodeGen/X86/fold-loop-of-urem.ll
+++ b/llvm/test/CodeGen/X86/fold-loop-of-urem.ll
@@ -15,25 +15,31 @@ define void @simple_urem_to_sel(i32 %N, i32 %rem_amt) nounwind {
; CHECK-NEXT: je .LBB0_4
; CHECK-NEXT: # %bb.1: # %for.body.preheader
; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r15
; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %r12
; CHECK-NEXT: pushq %rbx
; CHECK-NEXT: movl %esi, %ebx
; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: xorl %r15d, %r15d
; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: xorl %r12d, %r12d
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB0_2: # %for.body
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
-; CHECK-NEXT: movl %r14d, %eax
-; CHECK-NEXT: xorl %edx, %edx
-; CHECK-NEXT: divl %ebx
-; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: movl %r14d, %edi
; CHECK-NEXT: callq use.i32 at PLT
; CHECK-NEXT: incl %r14d
-; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: cmpl %ebx, %r14d
+; CHECK-NEXT: cmovel %r15d, %r14d
+; CHECK-NEXT: incl %r12d
+; CHECK-NEXT: cmpl %r12d, %ebp
; CHECK-NEXT: jne .LBB0_2
; CHECK-NEXT: # %bb.3:
; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r12
; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
; CHECK-NEXT: popq %rbp
; CHECK-NEXT: .LBB0_4: # %for.cond.cleanup
; CHECK-NEXT: retq
@@ -60,24 +66,28 @@ define void @simple_urem_to_sel_nested2(i32 %N, i32 %rem_amt) nounwind {
; CHECK-NEXT: je .LBB1_8
; CHECK-NEXT: # %bb.1: # %for.body.preheader
; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r15
; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %r12
; CHECK-NEXT: pushq %rbx
; CHECK-NEXT: movl %esi, %ebx
; CHECK-NEXT: movl %edi, %ebp
+; CHECK-NEXT: xorl %r15d, %r15d
; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: xorl %r12d, %r12d
; CHECK-NEXT: jmp .LBB1_2
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB1_5: # %for.body1
; CHECK-NEXT: # in Loop: Header=BB1_2 Depth=1
-; CHECK-NEXT: movl %r14d, %eax
-; CHECK-NEXT: xorl %edx, %edx
-; CHECK-NEXT: divl %ebx
-; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: movl %r14d, %edi
; CHECK-NEXT: callq use.i32 at PLT
; CHECK-NEXT: .LBB1_6: # %for.body.tail
; CHECK-NEXT: # in Loop: Header=BB1_2 Depth=1
; CHECK-NEXT: incl %r14d
-; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: cmpl %ebx, %r14d
+; CHECK-NEXT: cmovel %r15d, %r14d
+; CHECK-NEXT: incl %r12d
+; CHECK-NEXT: cmpl %r12d, %ebp
; CHECK-NEXT: je .LBB1_7
; CHECK-NEXT: .LBB1_2: # %for.body
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
@@ -97,7 +107,9 @@ define void @simple_urem_to_sel_nested2(i32 %N, i32 %rem_amt) nounwind {
; CHECK-NEXT: jmp .LBB1_6
; CHECK-NEXT: .LBB1_7:
; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r12
; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
; CHECK-NEXT: popq %rbp
; CHECK-NEXT: .LBB1_8: # %for.cond.cleanup
; CHECK-NEXT: retq
@@ -213,40 +225,36 @@ for.body.tail:
define void @simple_urem_to_sel_vec(<2 x i64> %rem_amt) nounwind {
; CHECK-LABEL: simple_urem_to_sel_vec:
; CHECK: # %bb.0: # %entry
-; CHECK-NEXT: pushq %r14
-; CHECK-NEXT: pushq %rbx
-; CHECK-NEXT: subq $24, %rsp
+; CHECK-NEXT: subq $56, %rsp
+; CHECK-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; CHECK-NEXT: pxor %xmm1, %xmm1
-; CHECK-NEXT: movdqa %xmm1, (%rsp) # 16-byte Spill
-; CHECK-NEXT: movq %xmm0, %rbx
-; CHECK-NEXT: pshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
-; CHECK-NEXT: movq %xmm0, %r14
+; CHECK-NEXT: pxor %xmm0, %xmm0
+; CHECK-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB3_1: # %for.body
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
-; CHECK-NEXT: movdqa (%rsp), %xmm1 # 16-byte Reload
; CHECK-NEXT: movdqa %xmm1, (%rsp) # 16-byte Spill
-; CHECK-NEXT: movq %xmm1, %rax
-; CHECK-NEXT: xorl %edx, %edx
-; CHECK-NEXT: divq %rbx
-; CHECK-NEXT: movq %rdx, %xmm0
-; CHECK-NEXT: pshufd {{.*#+}} xmm1 = xmm1[2,3,2,3]
-; CHECK-NEXT: movq %xmm1, %rax
-; CHECK-NEXT: xorl %edx, %edx
-; CHECK-NEXT: divq %r14
-; CHECK-NEXT: movq %rdx, %xmm1
-; CHECK-NEXT: punpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; CHECK-NEXT: movdqa %xmm1, %xmm0
; CHECK-NEXT: callq use.2xi64 at PLT
-; CHECK-NEXT: movdqa (%rsp), %xmm0 # 16-byte Reload
-; CHECK-NEXT: psubq {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
-; CHECK-NEXT: movdqa %xmm0, (%rsp) # 16-byte Spill
+; CHECK-NEXT: pcmpeqd %xmm1, %xmm1
+; CHECK-NEXT: movdqa (%rsp), %xmm2 # 16-byte Reload
+; CHECK-NEXT: psubq %xmm1, %xmm2
+; CHECK-NEXT: movdqa %xmm2, %xmm0
+; CHECK-NEXT: movdqa %xmm2, %xmm3
+; CHECK-NEXT: pcmpeqd {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
+; CHECK-NEXT: pshufd {{.*#+}} xmm2 = xmm0[1,0,3,2]
+; CHECK-NEXT: pand %xmm0, %xmm2
+; CHECK-NEXT: pandn %xmm3, %xmm2
+; CHECK-NEXT: movdqa %xmm2, (%rsp) # 16-byte Spill
+; CHECK-NEXT: movdqa {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
+; CHECK-NEXT: psubq %xmm1, %xmm0
+; CHECK-NEXT: movdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; CHECK-NEXT: callq get.i1 at PLT
; CHECK-NEXT: testb $1, %al
+; CHECK-NEXT: movdqa (%rsp), %xmm1 # 16-byte Reload
; CHECK-NEXT: je .LBB3_1
; CHECK-NEXT: # %bb.2: # %for.cond.cleanup
-; CHECK-NEXT: addq $24, %rsp
-; CHECK-NEXT: popq %rbx
-; CHECK-NEXT: popq %r14
+; CHECK-NEXT: addq $56, %rsp
; CHECK-NEXT: retq
entry:
br label %for.body
@@ -336,27 +344,33 @@ define void @simple_urem_to_sel_second_acc(i32 %N, i32 %rem_amt) nounwind {
; CHECK-NEXT: pushq %rbp
; CHECK-NEXT: pushq %r15
; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %r13
+; CHECK-NEXT: pushq %r12
; CHECK-NEXT: pushq %rbx
; CHECK-NEXT: pushq %rax
; CHECK-NEXT: movl %esi, %ebx
; CHECK-NEXT: movl %edi, %ebp
; CHECK-NEXT: movl $1, %r15d
+; CHECK-NEXT: xorl %r12d, %r12d
; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: xorl %r13d, %r13d
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB5_2: # %for.body
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
-; CHECK-NEXT: movl %r14d, %eax
-; CHECK-NEXT: xorl %edx, %edx
-; CHECK-NEXT: divl %ebx
-; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: movl %r14d, %edi
; CHECK-NEXT: callq use.i32 at PLT
; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %ebx, %r14d
+; CHECK-NEXT: cmovel %r12d, %r14d
+; CHECK-NEXT: incl %r13d
; CHECK-NEXT: addl $2, %r15d
; CHECK-NEXT: cmpl %ebp, %r15d
; CHECK-NEXT: jbe .LBB5_2
; CHECK-NEXT: # %bb.3:
; CHECK-NEXT: addq $8, %rsp
; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r12
+; CHECK-NEXT: popq %r13
; CHECK-NEXT: popq %r14
; CHECK-NEXT: popq %r15
; CHECK-NEXT: popq %rbp
@@ -676,23 +690,27 @@ define void @simple_urem_multi_latch_non_canonical(i32 %N, i32 %rem_amt) nounwin
; CHECK-NEXT: pushq %rbp
; CHECK-NEXT: pushq %r15
; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %r13
+; CHECK-NEXT: pushq %r12
; CHECK-NEXT: pushq %rbx
; CHECK-NEXT: pushq %rax
; CHECK-NEXT: movl %esi, %ebx
; CHECK-NEXT: movl %edi, %ebp
; CHECK-NEXT: decl %ebp
+; CHECK-NEXT: xorl %r12d, %r12d
; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: xorl %r13d, %r13d
; CHECK-NEXT: jmp .LBB12_2
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB12_3: # %for.body.backedge
; CHECK-NEXT: # in Loop: Header=BB12_2 Depth=1
; CHECK-NEXT: incl %r14d
+; CHECK-NEXT: cmpl %ebx, %r14d
+; CHECK-NEXT: cmovel %r12d, %r14d
+; CHECK-NEXT: incl %r13d
; CHECK-NEXT: .LBB12_2: # %for.body
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
-; CHECK-NEXT: movl %r14d, %eax
-; CHECK-NEXT: xorl %edx, %edx
-; CHECK-NEXT: divl %ebx
-; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: movl %r14d, %edi
; CHECK-NEXT: callq use.i32 at PLT
; CHECK-NEXT: callq get.i1 at PLT
; CHECK-NEXT: movl %eax, %r15d
@@ -702,11 +720,13 @@ define void @simple_urem_multi_latch_non_canonical(i32 %N, i32 %rem_amt) nounwin
; CHECK-NEXT: # %bb.4: # %for.body0
; CHECK-NEXT: # in Loop: Header=BB12_2 Depth=1
; CHECK-NEXT: callq do_stuff1 at PLT
-; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: cmpl %r13d, %ebp
; CHECK-NEXT: jne .LBB12_3
; CHECK-NEXT: # %bb.5:
; CHECK-NEXT: addq $8, %rsp
; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r12
+; CHECK-NEXT: popq %r13
; CHECK-NEXT: popq %r14
; CHECK-NEXT: popq %r15
; CHECK-NEXT: popq %rbp
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll b/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll
index a019679e65905..2f4652b425753 100644
--- a/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll
@@ -19,9 +19,12 @@ define void @simple_urem_to_sel(i32 %N, i32 %rem_amt) nounwind {
; CHECK: [[FOR_COND_CLEANUP]]:
; CHECK-NEXT: ret void
; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[REM:%.*]] = phi i32 [ 0, %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i32 [[REM]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
+; CHECK-NEXT: [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
@@ -53,7 +56,8 @@ define void @simple_urem_to_sel_nested2(i32 %N, i32 %rem_amt) nounwind {
; CHECK: [[FOR_COND_CLEANUP]]:
; CHECK-NEXT: ret void
; CHECK: [[FOR_BODY]]:
-; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY_TAIL:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT: [[REM:%.*]] = phi i32 [ 0, %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY_TAIL:.*]] ]
+; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY_TAIL]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
; CHECK-NEXT: [[COND0:%.*]] = call i1 @get.i1()
; CHECK-NEXT: br i1 [[COND0]], label %[[FOR_BODY0:.*]], label %[[FOR_BODY_TAIL]]
; CHECK: [[FOR_BODY0]]:
@@ -63,10 +67,12 @@ define void @simple_urem_to_sel_nested2(i32 %N, i32 %rem_amt) nounwind {
; CHECK-NEXT: [[COND2:%.*]] = call i1 @get.i1()
; CHECK-NEXT: br i1 [[COND2]], label %[[FOR_BODY1]], label %[[FOR_BODY_TAIL]]
; CHECK: [[FOR_BODY1]]:
-; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
; CHECK-NEXT: br label %[[FOR_BODY_TAIL]]
; CHECK: [[FOR_BODY_TAIL]]:
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i32 [[REM]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
+; CHECK-NEXT: [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
@@ -164,9 +170,12 @@ define void @simple_urem_to_sel_vec(<2 x i64> %rem_amt) nounwind {
; CHECK: [[FOR_COND_CLEANUP:.*]]:
; CHECK-NEXT: ret void
; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[REM:%.*]] = phi <2 x i64> [ zeroinitializer, %[[ENTRY]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT: [[I_04:%.*]] = phi <2 x i64> [ [[INC:%.*]], %[[FOR_BODY]] ], [ zeroinitializer, %[[ENTRY]] ]
-; CHECK-NEXT: [[REM:%.*]] = urem <2 x i64> [[I_04]], [[REM_AMT]]
; CHECK-NEXT: tail call void @use.2xi64(<2 x i64> [[REM]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw <2 x i64> [[REM]], <i64 1, i64 1>
+; CHECK-NEXT: [[TMP2:%.*]] = icmp eq <2 x i64> [[TMP1]], [[REM_AMT]]
+; CHECK-NEXT: [[TMP3]] = select <2 x i1> [[TMP2]], <2 x i64> zeroinitializer, <2 x i64> [[TMP1]]
; CHECK-NEXT: [[INC]] = add nuw <2 x i64> [[I_04]], <i64 1, i64 1>
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = call i1 @get.i1()
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
@@ -246,10 +255,13 @@ define void @simple_urem_to_sel_second_acc(i32 %N, i32 %rem_amt) nounwind {
; CHECK: [[FOR_COND_CLEANUP]]:
; CHECK-NEXT: ret void
; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[REM:%.*]] = phi i32 [ 0, %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 0, %[[FOR_BODY_PREHEADER]] ]
; CHECK-NEXT: [[I_05:%.*]] = phi i32 [ [[INC2:%.*]], %[[FOR_BODY]] ], [ 1, %[[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i32 [[REM]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
+; CHECK-NEXT: [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
; CHECK-NEXT: [[INC2]] = add nuw i32 [[I_05]], 2
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp ugt i32 [[INC2]], [[N]]
>From 336f2acd2be0aaece58bb56f549d98d51ac685fd Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Thu, 18 Jul 2024 20:01:51 +0800
Subject: [PATCH 3/4] more better comments
---
llvm/lib/CodeGen/CodeGenPrepare.cpp | 11 ++++++++---
1 file changed, 8 insertions(+), 3 deletions(-)
diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp
index e0232afa4d917..d3ed74fcb9dc7 100644
--- a/llvm/lib/CodeGen/CodeGenPrepare.cpp
+++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp
@@ -2000,7 +2000,7 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
AddOrSub = std::nullopt;
AddOrSubOffset = nullptr;
} else {
- // Search through a NUW add/sub.
+ // Search through a NUW add/sub on top of the loop increment.
Value *V0, *V1;
if (match(Incr, m_NUWAddLike(m_Value(V0), m_Value(V1))))
AddOrSub = true;
@@ -2045,6 +2045,7 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
if (!match(LoopIncrInfo->first, m_NUWAdd(m_Value(), m_Value())))
return false;
+ // Need unique loop preheader and latch.
if (PN->getBasicBlockIndex(L->getLoopLatch()) < 0 ||
PN->getBasicBlockIndex(L->getLoopPreheader()) < 0)
return false;
@@ -2080,12 +2081,16 @@ static bool foldURemOfLoopIncrement(Instruction *Rem, const LoopInfo *LI,
AddOrSubOffset, LoopIncrPN))
return false;
- // Only non-constant remainder as the extra IV is is probably not profitable
+ // Only non-constant remainder as the extra IV is probably not profitable
// in that case. Further, since remainder amount is non-constant, only handle
// case where `IncrLoopInvariant` and `Start` are 0 to entirely eliminate the
// rem (as opposed to just hoisting it outside of the loop).
//
- // Potential TODO: Should we have a check for how "nested" this remainder
+ // Potential TODO(1): `urem` of a const ends up as `mul` + `shift` + `add`. If
+ // we can rule out register pressure and ensure this `urem` is executed each
+ // iteration, its probably profitable to handle the const case as well.
+ //
+ // Potential TODO(2): Should we have a check for how "nested" this remainder
// operation is? The new code runs every iteration so if the remainder is
// guarded behind unlikely conditions this might not be worth it.
if (AddOrSub.has_value() || match(RemAmt, m_ImmConstant()))
>From e88b6f144e74b29f5f99801e8798b4bd41ae21bd Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Fri, 26 Jul 2024 01:30:11 +0800
Subject: [PATCH 4/4] handle add sub
---
llvm/lib/CodeGen/CodeGenPrepare.cpp | 92 ++++++++++++-------
llvm/test/CodeGen/X86/fold-loop-of-urem.ll | 32 +++++--
.../CodeGenPrepare/X86/fold-loop-of-urem.ll | 27 ++++--
3 files changed, 100 insertions(+), 51 deletions(-)
diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp
index d3ed74fcb9dc7..abf36c281eaed 100644
--- a/llvm/lib/CodeGen/CodeGenPrepare.cpp
+++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp
@@ -471,7 +471,7 @@ class CodeGenPrepare {
bool replaceMathCmpWithIntrinsic(BinaryOperator *BO, Value *Arg0, Value *Arg1,
CmpInst *Cmp, Intrinsic::ID IID);
bool optimizeCmp(CmpInst *Cmp, ModifyDT &ModifiedDT);
- bool optimizeRem(Instruction *Rem);
+ bool optimizeURem(Instruction *Rem);
bool combineToUSubWithOverflow(CmpInst *Cmp, ModifyDT &ModifiedDT);
bool combineToUAddWithOverflow(CmpInst *Cmp, ModifyDT &ModifiedDT);
void verifyBFIUpdates(Function &F);
@@ -1976,26 +1976,23 @@ static bool foldFCmpToFPClassTest(CmpInst *Cmp, const TargetLowering &TLI,
}
static bool isRemOfLoopIncrementWithLoopInvariant(
- Value *Rem, const LoopInfo *LI, Value *&RemAmtOut,
- std::optional<bool> &AddOrSubOut, Value *&AddOrSubOffsetOut,
- PHINode *&LoopIncrPNOut) {
+ Instruction *Rem, const LoopInfo *LI, Value *&RemAmtOut,
+ std::optional<bool> &AddOrSubOut, Value *&AddOrSubInstOut,
+ Value *&AddOrSubOffsetOut, PHINode *&LoopIncrPNOut) {
Value *Incr, *RemAmt;
- if (!isa<Instruction>(Rem))
- return false;
// NB: If RemAmt is a power of 2 it *should* have been transformed by now.
if (!match(Rem, m_URem(m_Value(Incr), m_Value(RemAmt))))
return false;
// Only trivially analyzable loops.
- Loop *L = LI->getLoopFor(cast<Instruction>(Rem)->getParent());
- if (L == nullptr || L->getLoopPreheader() == nullptr ||
- L->getLoopLatch() == nullptr)
+ Loop *L = LI->getLoopFor(Rem->getParent());
+ if (!L || !L->getLoopPreheader() || !L->getLoopLatch())
return false;
std::optional<bool> AddOrSub;
Value *AddOrSubOffset;
// Find out loop increment PHI.
- PHINode *PN = dyn_cast<PHINode>(Incr);
+ auto *PN = dyn_cast<PHINode>(Incr);
if (PN != nullptr) {
AddOrSub = std::nullopt;
AddOrSubOffset = nullptr;
@@ -2009,6 +2006,8 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
else
return false;
+ AddOrSubInstOut = Incr;
+
PN = dyn_cast<PHINode>(V0);
if (PN != nullptr) {
AddOrSubOffset = V1;
@@ -2018,7 +2017,7 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
}
}
- if (PN == nullptr)
+ if (!PN)
return false;
// This isn't strictly necessary, what we really need is one increment and any
@@ -2032,7 +2031,12 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
// Is the PHI a loop increment?
auto LoopIncrInfo = getIVIncrement(PN, LI);
- if (!LoopIncrInfo.has_value())
+ if (!LoopIncrInfo)
+ return false;
+
+ // getIVIncrement finds the loop at PN->getParent(). This might be a different
+ // loop from the loop with Rem->getParent().
+ if (L->getHeader() != PN->getParent())
return false;
// We need remainder_amount % increment_amount to be zero. Increment of one
@@ -2045,11 +2049,6 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
if (!match(LoopIncrInfo->first, m_NUWAdd(m_Value(), m_Value())))
return false;
- // Need unique loop preheader and latch.
- if (PN->getBasicBlockIndex(L->getLoopLatch()) < 0 ||
- PN->getBasicBlockIndex(L->getLoopPreheader()) < 0)
- return false;
-
// Set output variables.
RemAmtOut = RemAmt;
LoopIncrPNOut = PN;
@@ -2071,20 +2070,19 @@ static bool isRemOfLoopIncrementWithLoopInvariant(
// Rem = rem == RemAmtLoopInvariant ? 0 : Rem;
//
// Currently only implemented for `Start` and `IncrLoopInvariant` being zero.
-static bool foldURemOfLoopIncrement(Instruction *Rem, const LoopInfo *LI,
+static bool foldURemOfLoopIncrement(Instruction *Rem, const DataLayout *DL,
+ const LoopInfo *LI,
SmallSet<BasicBlock *, 32> &FreshBBs,
bool IsHuge) {
std::optional<bool> AddOrSub;
- Value *AddOrSubOffset, *RemAmt;
+ Value *AddOrSubOffset, *RemAmt, *AddOrSubInst;
PHINode *LoopIncrPN;
- if (!isRemOfLoopIncrementWithLoopInvariant(Rem, LI, RemAmt, AddOrSub,
- AddOrSubOffset, LoopIncrPN))
+ if (!isRemOfLoopIncrementWithLoopInvariant(
+ Rem, LI, RemAmt, AddOrSub, AddOrSubInst, AddOrSubOffset, LoopIncrPN))
return false;
// Only non-constant remainder as the extra IV is probably not profitable
- // in that case. Further, since remainder amount is non-constant, only handle
- // case where `IncrLoopInvariant` and `Start` are 0 to entirely eliminate the
- // rem (as opposed to just hoisting it outside of the loop).
+ // in that case.
//
// Potential TODO(1): `urem` of a const ends up as `mul` + `shift` + `add`. If
// we can rule out register pressure and ensure this `urem` is executed each
@@ -2093,12 +2091,37 @@ static bool foldURemOfLoopIncrement(Instruction *Rem, const LoopInfo *LI,
// Potential TODO(2): Should we have a check for how "nested" this remainder
// operation is? The new code runs every iteration so if the remainder is
// guarded behind unlikely conditions this might not be worth it.
- if (AddOrSub.has_value() || match(RemAmt, m_ImmConstant()))
+ if (match(RemAmt, m_ImmConstant()))
return false;
Loop *L = LI->getLoopFor(Rem->getParent());
- if (!match(LoopIncrPN->getIncomingValueForBlock(L->getLoopPreheader()),
- m_Zero()))
- return false;
+
+ // If we have add/sub create initial value for remainder.
+ // The logic here is:
+ // (urem (add/sub nuw Start, IncrLoopInvariant), RemAmtLoopInvariant
+ //
+ // Only proceed if the expression simplifies (otherwise we can't fully
+ // optimize out the urem).
+ Value *Start = LoopIncrPN->getIncomingValueForBlock(L->getLoopPreheader());
+ if (AddOrSub) {
+ assert(AddOrSubOffset && AddOrSubInst &&
+ "We found an add/sub but missing values");
+ // Without dom-condition/assumption cache we aren't likely to get much out
+ // of a context instruction.
+ const SimplifyQuery Q(*DL);
+ bool NSW = cast<OverflowingBinaryOperator>(AddOrSubInst)->hasNoSignedWrap();
+ if (*AddOrSub)
+ Start = simplifyAddInst(Start, AddOrSubOffset, /*IsNSW=*/NSW,
+ /*IsNUW=*/true, Q);
+ else
+ Start = simplifySubInst(Start, AddOrSubOffset, /*IsNSW=*/NSW,
+ /*IsNUW=*/true, Q);
+ if (!Start)
+ return false;
+
+ Start = simplifyURemInst(Start, RemAmt, Q);
+ if (!Start)
+ return false;
+ }
// Create new remainder with induction variable.
Type *Ty = Rem->getType();
@@ -2115,7 +2138,7 @@ static bool foldURemOfLoopIncrement(Instruction *Rem, const LoopInfo *LI,
Value *RemSel =
Builder.CreateSelect(RemCmp, Constant::getNullValue(Ty), RemAdd);
- NewRem->addIncoming(Constant::getNullValue(Ty), L->getLoopPreheader());
+ NewRem->addIncoming(Start, L->getLoopPreheader());
NewRem->addIncoming(RemSel, L->getLoopLatch());
// Insert all touched BBs.
@@ -2125,11 +2148,13 @@ static bool foldURemOfLoopIncrement(Instruction *Rem, const LoopInfo *LI,
replaceAllUsesWith(Rem, NewRem, FreshBBs, IsHuge);
Rem->eraseFromParent();
+ if (AddOrSubInst && AddOrSubInst->use_empty())
+ cast<Instruction>(AddOrSubInst)->eraseFromParent();
return true;
}
-bool CodeGenPrepare::optimizeRem(Instruction *Rem) {
- if (foldURemOfLoopIncrement(Rem, LI, FreshBBs, IsHugeFunc))
+bool CodeGenPrepare::optimizeURem(Instruction *Rem) {
+ if (foldURemOfLoopIncrement(Rem, DL, LI, FreshBBs, IsHugeFunc))
return true;
return false;
}
@@ -8520,9 +8545,8 @@ bool CodeGenPrepare::optimizeInst(Instruction *I, ModifyDT &ModifiedDT) {
if (optimizeCmp(Cmp, ModifiedDT))
return true;
- if (match(I, m_URem(m_Value(), m_Value())) ||
- match(I, m_SRem(m_Value(), m_Value())))
- if (optimizeRem(I))
+ if (match(I, m_URem(m_Value(), m_Value())))
+ if (optimizeURem(I))
return true;
if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
diff --git a/llvm/test/CodeGen/X86/fold-loop-of-urem.ll b/llvm/test/CodeGen/X86/fold-loop-of-urem.ll
index c2c1124b1ac11..b9fe4d7c79c7f 100644
--- a/llvm/test/CodeGen/X86/fold-loop-of-urem.ll
+++ b/llvm/test/CodeGen/X86/fold-loop-of-urem.ll
@@ -890,25 +890,31 @@ define void @simple_urem_to_sel_non_zero_start(i32 %N, i32 %rem_amt) nounwind {
; CHECK-NEXT: jb .LBB16_4
; CHECK-NEXT: # %bb.1: # %for.body.preheader
; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r15
; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %r12
; CHECK-NEXT: pushq %rbx
; CHECK-NEXT: movl %esi, %ebx
; CHECK-NEXT: movl %edi, %ebp
; CHECK-NEXT: movl $2, %r14d
+; CHECK-NEXT: xorl %r15d, %r15d
+; CHECK-NEXT: movl $2, %r12d
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB16_2: # %for.body
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
-; CHECK-NEXT: movl %r14d, %eax
-; CHECK-NEXT: xorl %edx, %edx
-; CHECK-NEXT: divl %ebx
-; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: movl %r14d, %edi
; CHECK-NEXT: callq use.i32 at PLT
; CHECK-NEXT: incl %r14d
-; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: cmpl %ebx, %r14d
+; CHECK-NEXT: cmovel %r15d, %r14d
+; CHECK-NEXT: incl %r12d
+; CHECK-NEXT: cmpl %r12d, %ebp
; CHECK-NEXT: jne .LBB16_2
; CHECK-NEXT: # %bb.3:
; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r12
; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
; CHECK-NEXT: popq %rbp
; CHECK-NEXT: .LBB16_4: # %for.cond.cleanup
; CHECK-NEXT: retq
@@ -1086,28 +1092,34 @@ define void @simple_urem_to_sel_non_zero_start_through_sub(i32 %N, i32 %rem_amt,
; CHECK-LABEL: simple_urem_to_sel_non_zero_start_through_sub:
; CHECK: # %bb.0: # %entry
; CHECK-NEXT: pushq %rbp
+; CHECK-NEXT: pushq %r15
; CHECK-NEXT: pushq %r14
+; CHECK-NEXT: pushq %r12
; CHECK-NEXT: pushq %rbx
; CHECK-NEXT: movl %edi, %ebp
; CHECK-NEXT: subl %edx, %ebp
; CHECK-NEXT: jbe .LBB20_3
; CHECK-NEXT: # %bb.1: # %for.body.preheader
; CHECK-NEXT: movl %esi, %ebx
+; CHECK-NEXT: xorl %r15d, %r15d
; CHECK-NEXT: xorl %r14d, %r14d
+; CHECK-NEXT: xorl %r12d, %r12d
; CHECK-NEXT: .p2align 4, 0x90
; CHECK-NEXT: .LBB20_2: # %for.body
; CHECK-NEXT: # =>This Inner Loop Header: Depth=1
-; CHECK-NEXT: movl %r14d, %eax
-; CHECK-NEXT: xorl %edx, %edx
-; CHECK-NEXT: divl %ebx
-; CHECK-NEXT: movl %edx, %edi
+; CHECK-NEXT: movl %r14d, %edi
; CHECK-NEXT: callq use.i32 at PLT
; CHECK-NEXT: incl %r14d
-; CHECK-NEXT: cmpl %r14d, %ebp
+; CHECK-NEXT: cmpl %ebx, %r14d
+; CHECK-NEXT: cmovel %r15d, %r14d
+; CHECK-NEXT: incl %r12d
+; CHECK-NEXT: cmpl %r12d, %ebp
; CHECK-NEXT: jne .LBB20_2
; CHECK-NEXT: .LBB20_3: # %for.cond.cleanup
; CHECK-NEXT: popq %rbx
+; CHECK-NEXT: popq %r12
; CHECK-NEXT: popq %r14
+; CHECK-NEXT: popq %r15
; CHECK-NEXT: popq %rbp
; CHECK-NEXT: retq
entry:
diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll b/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll
index 2f4652b425753..61bbe08e293b3 100644
--- a/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll
+++ b/llvm/test/Transforms/CodeGenPrepare/X86/fold-loop-of-urem.ll
@@ -398,9 +398,12 @@ define void @simple_urem_non_zero_entry4(i32 %N, i32 %rem_amt) nounwind {
; CHECK: [[FOR_COND_CLEANUP]]:
; CHECK-NEXT: ret void
; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[REM:%.*]] = phi i32 [ 4, %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 4, %[[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i32 [[REM]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
+; CHECK-NEXT: [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
@@ -469,9 +472,12 @@ define void @simple_urem_fail_no_preheader_non_canonical(i32 %N, i32 %rem_amt) n
; CHECK-NEXT: [[I_04_PH:%.*]] = phi i32 [ 1, %[[FOR_BODY1]] ], [ 0, %[[ENTRY]] ]
; CHECK-NEXT: br label %[[FOR_BODY:.*]]
; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[REM:%.*]] = phi i32 [ [[I_04_PH]], %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[I_04_PH]], %[[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i32 [[REM]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
+; CHECK-NEXT: [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
@@ -662,9 +668,12 @@ define void @simple_urem_to_sel_non_zero_start(i32 %N, i32 %rem_amt) nounwind {
; CHECK: [[FOR_COND_CLEANUP]]:
; CHECK-NEXT: ret void
; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[REM:%.*]] = phi i32 [ 2, %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 2, %[[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_04]], [[REM_AMT]]
; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i32 [[REM]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
+; CHECK-NEXT: [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
@@ -697,10 +706,12 @@ define void @simple_urem_to_sel_non_zero_start_through_add(i32 %N, i32 %rem_amt_
; CHECK: [[FOR_COND_CLEANUP]]:
; CHECK-NEXT: ret void
; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[REM:%.*]] = phi i32 [ 7, %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ 2, %[[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT: [[I_WITH_OFF:%.*]] = add nuw i32 [[I_04]], 5
-; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_WITH_OFF]], [[REM_AMT]]
; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i32 [[REM]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
+; CHECK-NEXT: [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
@@ -808,10 +819,12 @@ define void @simple_urem_to_sel_non_zero_start_through_sub(i32 %N, i32 %rem_amt,
; CHECK: [[FOR_COND_CLEANUP]]:
; CHECK-NEXT: ret void
; CHECK: [[FOR_BODY]]:
+; CHECK-NEXT: [[REM:%.*]] = phi i32 [ 0, %[[FOR_BODY_PREHEADER]] ], [ [[TMP3:%.*]], %[[FOR_BODY]] ]
; CHECK-NEXT: [[I_04:%.*]] = phi i32 [ [[INC:%.*]], %[[FOR_BODY]] ], [ [[START]], %[[FOR_BODY_PREHEADER]] ]
-; CHECK-NEXT: [[I_WITH_OFF:%.*]] = sub nuw i32 [[I_04]], [[START]]
-; CHECK-NEXT: [[REM:%.*]] = urem i32 [[I_WITH_OFF]], [[REM_AMT]]
; CHECK-NEXT: tail call void @use.i32(i32 [[REM]])
+; CHECK-NEXT: [[TMP1:%.*]] = add nuw i32 [[REM]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 [[TMP1]], [[REM_AMT]]
+; CHECK-NEXT: [[TMP3]] = select i1 [[TMP2]], i32 0, i32 [[TMP1]]
; CHECK-NEXT: [[INC]] = add nuw i32 [[I_04]], 1
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[N]]
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]]
More information about the llvm-commits
mailing list