mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-11-24 11:42:57 +01:00
d654e7d40c
Enable enableMultipleCopyHints() on X86. Original Patch by @jonpa: While enabling the mischeduler for SystemZ, it was discovered that for some reason a test needed one extra seemingly needless COPY (test/CodeGen/SystemZ/call-03.ll). The handling for that is resulted in this patch, which improves the register coalescing by providing not just one copy hint, but a sorted list of copy hints. On SystemZ, this gives ~12500 less register moves on SPEC, as well as marginally less spilling. Instead of improving just the SystemZ backend, the improvement has been implemented in common-code (calculateSpillWeightAndHint(). This gives a lot of test failures, but since this should be a general improvement I hope that the involved targets will help and review the test updates. Differential Revision: https://reviews.llvm.org/D38128 llvm-svn: 342578
102 lines
2.5 KiB
LLVM
102 lines
2.5 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s --check-prefix=LNX
|
|
; RUN: llc < %s -mtriple=x86_64-win32 | FileCheck %s --check-prefix=WIN
|
|
|
|
; Reuse the flags value from the add instructions instead of emitting separate
|
|
; testl instructions.
|
|
|
|
; Use the flags on the add.
|
|
|
|
define i32 @test1(i32* %x, i32 %y, i32 %a, i32 %b) nounwind {
|
|
; LNX-LABEL: test1:
|
|
; LNX: # %bb.0:
|
|
; LNX-NEXT: movl %edx, %eax
|
|
; LNX-NEXT: addl (%rdi), %esi
|
|
; LNX-NEXT: cmovnsl %ecx, %eax
|
|
; LNX-NEXT: retq
|
|
;
|
|
; WIN-LABEL: test1:
|
|
; WIN: # %bb.0:
|
|
; WIN-NEXT: movl %r8d, %eax
|
|
; WIN-NEXT: addl (%rcx), %edx
|
|
; WIN-NEXT: cmovnsl %r9d, %eax
|
|
; WIN-NEXT: retq
|
|
%tmp2 = load i32, i32* %x, align 4 ; <i32> [#uses=1]
|
|
%tmp4 = add i32 %tmp2, %y ; <i32> [#uses=1]
|
|
%tmp5 = icmp slt i32 %tmp4, 0 ; <i1> [#uses=1]
|
|
%tmp.0 = select i1 %tmp5, i32 %a, i32 %b ; <i32> [#uses=1]
|
|
ret i32 %tmp.0
|
|
}
|
|
|
|
declare void @foo(i32)
|
|
|
|
; Don't use the flags result of the and here, since the and has no
|
|
; other use. A simple test is better.
|
|
|
|
define void @test2(i32 %x) nounwind {
|
|
; LNX-LABEL: test2:
|
|
; LNX: # %bb.0:
|
|
; LNX-NEXT: testb $16, %dil
|
|
; LNX-NEXT: jne .LBB1_2
|
|
; LNX-NEXT: # %bb.1: # %true
|
|
; LNX-NEXT: pushq %rax
|
|
; LNX-NEXT: callq foo
|
|
; LNX-NEXT: popq %rax
|
|
; LNX-NEXT: .LBB1_2: # %false
|
|
; LNX-NEXT: retq
|
|
;
|
|
; WIN-LABEL: test2:
|
|
; WIN: # %bb.0:
|
|
; WIN-NEXT: subq $40, %rsp
|
|
; WIN-NEXT: testb $16, %cl
|
|
; WIN-NEXT: jne .LBB1_2
|
|
; WIN-NEXT: # %bb.1: # %true
|
|
; WIN-NEXT: callq foo
|
|
; WIN-NEXT: .LBB1_2: # %false
|
|
; WIN-NEXT: addq $40, %rsp
|
|
; WIN-NEXT: retq
|
|
%y = and i32 %x, 16
|
|
%t = icmp eq i32 %y, 0
|
|
br i1 %t, label %true, label %false
|
|
true:
|
|
call void @foo(i32 %x)
|
|
ret void
|
|
false:
|
|
ret void
|
|
}
|
|
|
|
; Do use the flags result of the and here, since the and has another use.
|
|
|
|
define void @test3(i32 %x) nounwind {
|
|
; LNX-LABEL: test3:
|
|
; LNX: # %bb.0:
|
|
; LNX-NEXT: andl $16, %edi
|
|
; LNX-NEXT: jne .LBB2_2
|
|
; LNX-NEXT: # %bb.1: # %true
|
|
; LNX-NEXT: pushq %rax
|
|
; LNX-NEXT: callq foo
|
|
; LNX-NEXT: popq %rax
|
|
; LNX-NEXT: .LBB2_2: # %false
|
|
; LNX-NEXT: retq
|
|
;
|
|
; WIN-LABEL: test3:
|
|
; WIN: # %bb.0:
|
|
; WIN-NEXT: subq $40, %rsp
|
|
; WIN-NEXT: andl $16, %ecx
|
|
; WIN-NEXT: jne .LBB2_2
|
|
; WIN-NEXT: # %bb.1: # %true
|
|
; WIN-NEXT: callq foo
|
|
; WIN-NEXT: .LBB2_2: # %false
|
|
; WIN-NEXT: addq $40, %rsp
|
|
; WIN-NEXT: retq
|
|
%y = and i32 %x, 16
|
|
%t = icmp eq i32 %y, 0
|
|
br i1 %t, label %true, label %false
|
|
true:
|
|
call void @foo(i32 %y)
|
|
ret void
|
|
false:
|
|
ret void
|
|
}
|
|
|