mirror of
https://github.com/RPCS3/llvm-mirror.git
synced 2024-10-21 03:53:04 +02:00
83dddda4a1
In the ELFv2 ABI, we are not required to save all CR fields. If only one nonvolatile CR field is clobbered, use mfocrf instead of mfcr to selectively save the field, because mfocrf has short latency compares to mfcr. Thanks Nemanja's invaluable hint! Reviewers: nemanjai tjablin hfinkel kbarton http://reviews.llvm.org/D17749 llvm-svn: 266038
83 lines
2.3 KiB
LLVM
83 lines
2.3 KiB
LLVM
; RUN: llc -O0 -disable-fp-elim -mtriple=powerpc-unknown-linux-gnu -mcpu=g5 < %s | FileCheck %s -check-prefix=PPC32
|
|
; RUN: llc -O0 -mtriple=powerpc64-unknown-linux-gnu -mcpu=g5 < %s | FileCheck %s -check-prefix=PPC64
|
|
; RUN: llc -O0 -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s -check-prefix=PPC64-ELFv2
|
|
|
|
declare void @foo()
|
|
|
|
define i32 @test_cr2() nounwind uwtable {
|
|
entry:
|
|
%ret = alloca i32, align 4
|
|
%0 = call i32 asm sideeffect "\0A\09mtcr $4\0A\09cmpw 2,$2,$1\0A\09mfcr $0", "=r,r,r,r,r,~{cr2}"(i32 1, i32 2, i32 3, i32 0) nounwind
|
|
store i32 %0, i32* %ret, align 4
|
|
call void @foo()
|
|
%1 = load i32, i32* %ret, align 4
|
|
ret i32 %1
|
|
}
|
|
|
|
; PPC32: stw 31, -4(1)
|
|
; PPC32: stwu 1, -32(1)
|
|
; PPC32: mfcr 12
|
|
; PPC32-NEXT: stw 12, 24(31)
|
|
; PPC32: lwz 12, 24(31)
|
|
; PPC32-NEXT: mtocrf 32, 12
|
|
|
|
; PPC64: .cfi_startproc
|
|
; PPC64: mfcr 12
|
|
; PPC64: stw 12, 8(1)
|
|
; PPC64: stdu 1, -[[AMT:[0-9]+]](1)
|
|
; PPC64: .cfi_def_cfa_offset 128
|
|
; PPC64: .cfi_offset lr, 16
|
|
; PPC64: .cfi_offset cr2, 8
|
|
; PPC64: addi 1, 1, [[AMT]]
|
|
; PPC64: lwz 12, 8(1)
|
|
; PPC64: mtocrf 32, 12
|
|
; PPC64: .cfi_endproc
|
|
|
|
define i32 @test_cr234() nounwind {
|
|
entry:
|
|
%ret = alloca i32, align 4
|
|
%0 = call i32 asm sideeffect "\0A\09mtcr $4\0A\09cmpw 2,$2,$1\0A\09cmpw 3,$2,$2\0A\09cmpw 4,$2,$3\0A\09mfcr $0", "=r,r,r,r,r,~{cr2},~{cr3},~{cr4}"(i32 1, i32 2, i32 3, i32 0) nounwind
|
|
store i32 %0, i32* %ret, align 4
|
|
call void @foo()
|
|
%1 = load i32, i32* %ret, align 4
|
|
ret i32 %1
|
|
}
|
|
|
|
; PPC32: stw 31, -4(1)
|
|
; PPC32: stwu 1, -32(1)
|
|
; PPC32: mfcr 12
|
|
; PPC32-NEXT: stw 12, 24(31)
|
|
; PPC32: lwz 12, 24(31)
|
|
; PPC32-NEXT: mtocrf 32, 12
|
|
; PPC32-NEXT: mtocrf 16, 12
|
|
; PPC32-NEXT: mtocrf 8, 12
|
|
|
|
; PPC64: mfcr 12
|
|
; PPC64: stw 12, 8(1)
|
|
; PPC64: stdu 1, -[[AMT:[0-9]+]](1)
|
|
; PPC64: addi 1, 1, [[AMT]]
|
|
; PPC64: lwz 12, 8(1)
|
|
; PPC64: mtocrf 32, 12
|
|
; PPC64: mtocrf 16, 12
|
|
; PPC64: mtocrf 8, 12
|
|
|
|
; Generate mfocrf in prologue when we need to save 1 nonvolatile CR field
|
|
define void @cloberOneNvCrField() {
|
|
entry:
|
|
tail call void asm sideeffect "# clobbers", "~{cr2}"()
|
|
ret void
|
|
|
|
; PPC64-ELFv2-LABEL: @cloberOneNvCrField
|
|
; PPC64-ELFv2: mfocrf [[REG1:[0-9]+]], 32
|
|
}
|
|
|
|
; Generate mfcr in prologue when we need to save all nonvolatile CR field
|
|
define void @cloberAllNvCrField() {
|
|
entry:
|
|
tail call void asm sideeffect "# clobbers", "~{cr2},~{cr3},~{cr4}"()
|
|
ret void
|
|
|
|
; PPC64-ELFv2-LABEL: @cloberAllNvCrField
|
|
; PPC64-ELFv2: mfcr [[REG1:[0-9]+]]
|
|
}
|