1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-02-01 05:01:59 +01:00
llvm-mirror/test/CodeGen/AArch64/GlobalISel/legalize-cmpxchg.mir
Amara Emerson 8eb83e55cc [AArch64][GlobalISel] Make s8 and s16 G_CONSTANTs legal.
We sometimes get poor code size because constants of types < 32b are legalized
as 32 bit G_CONSTANTs with a truncate to fit. This works but means that the
localizer can no longer sink them (although it's possible to extend it to do so).

On AArch64 however s8 and s16 constants can be selected in the same way as s32
constants, with a mov pseudo into a W register. If we make s8 and s16 constants
legal then we can avoid unnecessary truncates, they can be CSE'd, and the
localizer can sink them as normal.

There is a caveat: if the user of a smaller constant has to widen the sources,
we end up with an anyext of the smaller typed G_CONSTANT. This can cause
regressions because of the additional extend and missed pattern matching. To
remedy this, there's a new artifact combiner to generate the wider G_CONSTANT
if it's legal for the target.

Differential Revision: https://reviews.llvm.org/D63587

llvm-svn: 364075
2019-06-21 16:43:50 +00:00

92 lines
3.2 KiB
YAML

# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
# RUN: llc -mtriple=aarch64-- -mattr=+lse -run-pass=legalizer -verify-machineinstrs %s -o - | FileCheck %s
--- |
target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
define void @cmpxchg_i8(i8* %addr) { ret void }
define void @cmpxchg_i16(i16* %addr) { ret void }
define void @cmpxchg_i32(i32* %addr) { ret void }
define void @cmpxchg_i64(i64* %addr) { ret void }
...
---
name: cmpxchg_i8
body: |
bb.0:
liveins: $x0
; CHECK-LABEL: name: cmpxchg_i8
; CHECK: [[COPY:%[0-9]+]]:_(p0) = COPY $x0
; CHECK: [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 0
; CHECK: [[C1:%[0-9]+]]:_(s8) = G_CONSTANT i8 1
; CHECK: [[ATOMIC_CMPXCHG:%[0-9]+]]:_(s8) = G_ATOMIC_CMPXCHG [[COPY]](p0), [[C]], [[C1]] :: (load store monotonic 1 on %ir.addr)
; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ATOMIC_CMPXCHG]](s8)
; CHECK: $w0 = COPY [[ANYEXT]](s32)
%0:_(p0) = COPY $x0
%1:_(s8) = G_CONSTANT i8 0
%2:_(s8) = G_CONSTANT i8 1
%3:_(s8) = G_ATOMIC_CMPXCHG %0, %1, %2 :: (load store monotonic 1 on %ir.addr)
%4:_(s32) = G_ANYEXT %3
$w0 = COPY %4(s32)
...
---
name: cmpxchg_i16
body: |
bb.0:
liveins: $x0
; CHECK-LABEL: name: cmpxchg_i16
; CHECK: [[COPY:%[0-9]+]]:_(p0) = COPY $x0
; CHECK: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 0
; CHECK: [[C1:%[0-9]+]]:_(s16) = G_CONSTANT i16 1
; CHECK: [[ATOMIC_CMPXCHG:%[0-9]+]]:_(s16) = G_ATOMIC_CMPXCHG [[COPY]](p0), [[C]], [[C1]] :: (load store monotonic 2 on %ir.addr)
; CHECK: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[ATOMIC_CMPXCHG]](s16)
; CHECK: $w0 = COPY [[ANYEXT]](s32)
%0:_(p0) = COPY $x0
%1:_(s16) = G_CONSTANT i16 0
%2:_(s16) = G_CONSTANT i16 1
%3:_(s16) = G_ATOMIC_CMPXCHG %0, %1, %2 :: (load store monotonic 2 on %ir.addr)
%4:_(s32) = G_ANYEXT %3
$w0 = COPY %4(s32)
...
---
name: cmpxchg_i32
body: |
bb.0:
liveins: $x0
; CHECK-LABEL: name: cmpxchg_i32
; CHECK: [[COPY:%[0-9]+]]:_(p0) = COPY $x0
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
; CHECK: [[ATOMIC_CMPXCHG:%[0-9]+]]:_(s32) = G_ATOMIC_CMPXCHG [[COPY]](p0), [[C]], [[C1]] :: (load store monotonic 4 on %ir.addr)
; CHECK: $w0 = COPY [[ATOMIC_CMPXCHG]](s32)
%0:_(p0) = COPY $x0
%1:_(s32) = G_CONSTANT i32 0
%2:_(s32) = G_CONSTANT i32 1
%3:_(s32) = G_ATOMIC_CMPXCHG %0, %1, %2 :: (load store monotonic 4 on %ir.addr)
$w0 = COPY %3(s32)
...
---
name: cmpxchg_i64
body: |
bb.0:
liveins: $x0
; CHECK-LABEL: name: cmpxchg_i64
; CHECK: [[COPY:%[0-9]+]]:_(p0) = COPY $x0
; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 1
; CHECK: [[ATOMIC_CMPXCHG:%[0-9]+]]:_(s64) = G_ATOMIC_CMPXCHG [[COPY]](p0), [[C]], [[C1]] :: (load store monotonic 8 on %ir.addr)
; CHECK: $x0 = COPY [[ATOMIC_CMPXCHG]](s64)
%0:_(p0) = COPY $x0
%1:_(s64) = G_CONSTANT i64 0
%2:_(s64) = G_CONSTANT i64 1
%3:_(s64) = G_ATOMIC_CMPXCHG %0, %1, %2 :: (load store monotonic 8 on %ir.addr)
$x0 = COPY %3(s64)
...