2019-07-21 03:25:44 +08:00
|
|
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
|
|
|
; RUN: llc -mtriple=i686-unknown-linux-gnu < %s | FileCheck %s --check-prefixes=CHECK,X86
|
|
|
|
|
; RUN: llc -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s --check-prefixes=CHECK,X64
|
|
|
|
|
|
|
|
|
|
; On X86, division in expensive. BuildRemEqFold should therefore run even
|
|
|
|
|
; when optimizing for size. Only optimizing for minimum size retains a plain div.
|
|
|
|
|
|
|
|
|
|
define i32 @test_minsize(i32 %X) optsize minsize nounwind readnone {
|
|
|
|
|
; X86-LABEL: test_minsize:
|
|
|
|
|
; X86: # %bb.0:
|
|
|
|
|
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
|
|
|
|
|
; X86-NEXT: pushl $5
|
|
|
|
|
; X86-NEXT: popl %ecx
|
|
|
|
|
; X86-NEXT: cltd
|
|
|
|
|
; X86-NEXT: idivl %ecx
|
|
|
|
|
; X86-NEXT: testl %edx, %edx
|
|
|
|
|
; X86-NEXT: je .LBB0_1
|
|
|
|
|
; X86-NEXT: # %bb.2:
|
|
|
|
|
; X86-NEXT: pushl $-10
|
|
|
|
|
; X86-NEXT: popl %eax
|
|
|
|
|
; X86-NEXT: retl
|
|
|
|
|
; X86-NEXT: .LBB0_1:
|
|
|
|
|
; X86-NEXT: pushl $42
|
|
|
|
|
; X86-NEXT: popl %eax
|
|
|
|
|
; X86-NEXT: retl
|
|
|
|
|
;
|
|
|
|
|
; X64-LABEL: test_minsize:
|
|
|
|
|
; X64: # %bb.0:
|
|
|
|
|
; X64-NEXT: movl %edi, %eax
|
|
|
|
|
; X64-NEXT: pushq $5
|
|
|
|
|
; X64-NEXT: popq %rcx
|
|
|
|
|
; X64-NEXT: cltd
|
|
|
|
|
; X64-NEXT: idivl %ecx
|
|
|
|
|
; X64-NEXT: testl %edx, %edx
|
|
|
|
|
; X64-NEXT: pushq $42
|
|
|
|
|
; X64-NEXT: popq %rcx
|
|
|
|
|
; X64-NEXT: pushq $-10
|
|
|
|
|
; X64-NEXT: popq %rax
|
|
|
|
|
; X64-NEXT: cmovel %ecx, %eax
|
|
|
|
|
; X64-NEXT: retq
|
|
|
|
|
%rem = srem i32 %X, 5
|
|
|
|
|
%cmp = icmp eq i32 %rem, 0
|
|
|
|
|
%ret = select i1 %cmp, i32 42, i32 -10
|
|
|
|
|
ret i32 %ret
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
define i32 @test_optsize(i32 %X) optsize nounwind readnone {
|
|
|
|
|
; X86-LABEL: test_optsize:
|
|
|
|
|
; X86: # %bb.0:
|
[CodeGen][SelectionDAG] More efficient code for X % C == 0 (SREM case)
Summary:
This implements an optimization described in Hacker's Delight 10-17:
when `C` is constant, the result of `X % C == 0` can be computed
more cheaply without actually calculating the remainder.
The motivation is discussed here: https://bugs.llvm.org/show_bug.cgi?id=35479.
One huge caveat: this signed case is only valid for positive divisors.
While we can freely negate negative divisors, we can't negate `INT_MIN`,
so for now if `INT_MIN` is encountered, we bailout.
As a follow-up, it should be possible to handle that more gracefully
via extra `and`+`setcc`+`select`.
This passes llvm's test-suite, and from cursory(!) cross-examination
the folds (the assembly) match those of GCC, and manual checking via alive
did not reveal any issues (other than the `INT_MIN` case)
Reviewers: RKSimon, spatel, hermord, craig.topper, xbolva00
Reviewed By: RKSimon, xbolva00
Subscribers: xbolva00, thakis, javed.absar, hiraditya, dexonsmith, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D65366
llvm-svn: 368702
2019-08-13 22:57:37 +08:00
|
|
|
; X86-NEXT: imull $-858993459, {{[0-9]+}}(%esp), %eax # imm = 0xCCCCCCCD
|
|
|
|
|
; X86-NEXT: addl $429496729, %eax # imm = 0x19999999
|
|
|
|
|
; X86-NEXT: cmpl $858993459, %eax # imm = 0x33333333
|
2019-07-21 03:25:44 +08:00
|
|
|
; X86-NEXT: movl $42, %eax
|
[CodeGen][SelectionDAG] More efficient code for X % C == 0 (SREM case)
Summary:
This implements an optimization described in Hacker's Delight 10-17:
when `C` is constant, the result of `X % C == 0` can be computed
more cheaply without actually calculating the remainder.
The motivation is discussed here: https://bugs.llvm.org/show_bug.cgi?id=35479.
One huge caveat: this signed case is only valid for positive divisors.
While we can freely negate negative divisors, we can't negate `INT_MIN`,
so for now if `INT_MIN` is encountered, we bailout.
As a follow-up, it should be possible to handle that more gracefully
via extra `and`+`setcc`+`select`.
This passes llvm's test-suite, and from cursory(!) cross-examination
the folds (the assembly) match those of GCC, and manual checking via alive
did not reveal any issues (other than the `INT_MIN` case)
Reviewers: RKSimon, spatel, hermord, craig.topper, xbolva00
Reviewed By: RKSimon, xbolva00
Subscribers: xbolva00, thakis, javed.absar, hiraditya, dexonsmith, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D65366
llvm-svn: 368702
2019-08-13 22:57:37 +08:00
|
|
|
; X86-NEXT: jb .LBB1_2
|
2019-07-21 03:25:44 +08:00
|
|
|
; X86-NEXT: # %bb.1:
|
|
|
|
|
; X86-NEXT: movl $-10, %eax
|
|
|
|
|
; X86-NEXT: .LBB1_2:
|
|
|
|
|
; X86-NEXT: retl
|
|
|
|
|
;
|
|
|
|
|
; X64-LABEL: test_optsize:
|
|
|
|
|
; X64: # %bb.0:
|
[CodeGen][SelectionDAG] More efficient code for X % C == 0 (SREM case)
Summary:
This implements an optimization described in Hacker's Delight 10-17:
when `C` is constant, the result of `X % C == 0` can be computed
more cheaply without actually calculating the remainder.
The motivation is discussed here: https://bugs.llvm.org/show_bug.cgi?id=35479.
One huge caveat: this signed case is only valid for positive divisors.
While we can freely negate negative divisors, we can't negate `INT_MIN`,
so for now if `INT_MIN` is encountered, we bailout.
As a follow-up, it should be possible to handle that more gracefully
via extra `and`+`setcc`+`select`.
This passes llvm's test-suite, and from cursory(!) cross-examination
the folds (the assembly) match those of GCC, and manual checking via alive
did not reveal any issues (other than the `INT_MIN` case)
Reviewers: RKSimon, spatel, hermord, craig.topper, xbolva00
Reviewed By: RKSimon, xbolva00
Subscribers: xbolva00, thakis, javed.absar, hiraditya, dexonsmith, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D65366
llvm-svn: 368702
2019-08-13 22:57:37 +08:00
|
|
|
; X64-NEXT: imull $-858993459, %edi, %eax # imm = 0xCCCCCCCD
|
|
|
|
|
; X64-NEXT: addl $429496729, %eax # imm = 0x19999999
|
|
|
|
|
; X64-NEXT: cmpl $858993459, %eax # imm = 0x33333333
|
2019-07-21 03:25:44 +08:00
|
|
|
; X64-NEXT: movl $42, %ecx
|
|
|
|
|
; X64-NEXT: movl $-10, %eax
|
[CodeGen][SelectionDAG] More efficient code for X % C == 0 (SREM case)
Summary:
This implements an optimization described in Hacker's Delight 10-17:
when `C` is constant, the result of `X % C == 0` can be computed
more cheaply without actually calculating the remainder.
The motivation is discussed here: https://bugs.llvm.org/show_bug.cgi?id=35479.
One huge caveat: this signed case is only valid for positive divisors.
While we can freely negate negative divisors, we can't negate `INT_MIN`,
so for now if `INT_MIN` is encountered, we bailout.
As a follow-up, it should be possible to handle that more gracefully
via extra `and`+`setcc`+`select`.
This passes llvm's test-suite, and from cursory(!) cross-examination
the folds (the assembly) match those of GCC, and manual checking via alive
did not reveal any issues (other than the `INT_MIN` case)
Reviewers: RKSimon, spatel, hermord, craig.topper, xbolva00
Reviewed By: RKSimon, xbolva00
Subscribers: xbolva00, thakis, javed.absar, hiraditya, dexonsmith, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D65366
llvm-svn: 368702
2019-08-13 22:57:37 +08:00
|
|
|
; X64-NEXT: cmovbl %ecx, %eax
|
2019-07-21 03:25:44 +08:00
|
|
|
; X64-NEXT: retq
|
|
|
|
|
%rem = srem i32 %X, 5
|
|
|
|
|
%cmp = icmp eq i32 %rem, 0
|
|
|
|
|
%ret = select i1 %cmp, i32 42, i32 -10
|
|
|
|
|
ret i32 %ret
|
|
|
|
|
}
|