16b7cc69ef
On SystemZ, the outgoing argument area which is big enough for all calls in the function is created once during the prolog, as opposed to adjusting the stack around each call. The call-sequence instructions are therefore not really useful any more than to compute the maximum call frame size, which has so far been done by PEI, but can just as well be done at an earlier point. This patch removes the mapping of the CallFrameSetupOpcode and CallFrameDestroyOpcode and instead computes the MaxCallFrameSize directly after instruction selection and then removes the ADJCALLSTACK pseudos. This removes the confusing pseudos and also avoids the problem of having to keep the call frame size accurate when creating new MBBs. This fixes #76618 which exposed the need to maintain the call frame size when splitting blocks (which was not done).
328 lines
8.1 KiB
YAML
328 lines
8.1 KiB
YAML
# RUN: llc -mtriple=s390x-linux-gnu -start-before=prologepilog %s -o - -mcpu=z14 \
|
|
# RUN: -verify-machineinstrs 2>&1 | FileCheck %s
|
|
# REQUIRES: asserts
|
|
#
|
|
# Test that redundant frame addressing anchor points are removed by
|
|
# MachineLateInstrsCleanup.
|
|
|
|
--- |
|
|
define void @fun1() { ret void }
|
|
define void @fun2() { ret void }
|
|
define void @fun3() { ret void }
|
|
define void @fun4() { ret void }
|
|
define void @fun5() { ret void }
|
|
define void @fun6() { ret void }
|
|
define void @fun7() { ret void }
|
|
define void @fun8() { ret void }
|
|
|
|
declare i32 @foo()
|
|
|
|
@ptr = external dso_local local_unnamed_addr global ptr
|
|
---
|
|
|
|
# Test elimination of redundant LAYs in successor blocks.
|
|
# CHECK-LABEL: fun1:
|
|
# CHECK: lay %r1, 4096(%r15)
|
|
# CHECK: # %bb.1:
|
|
# CHECK-NOT: lay
|
|
# CHECK: .LBB0_2:
|
|
# CHECK-NOT: lay
|
|
---
|
|
name: fun1
|
|
tracksRegLiveness: true
|
|
stack:
|
|
- { id: 0, size: 5000 }
|
|
- { id: 1, size: 2500 }
|
|
- { id: 2, size: 2500 }
|
|
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0 (%ir-block.0):
|
|
liveins: $f16d
|
|
successors: %bb.2(0x00000001), %bb.1(0x7fffffff)
|
|
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.1, 0, $noreg
|
|
CHIMux undef $r0l, 3, implicit-def $cc
|
|
BRC 14, 8, %bb.2, implicit killed $cc
|
|
J %bb.1
|
|
|
|
bb.1:
|
|
liveins: $f16d
|
|
VST64 renamable $f16d, %stack.2, 0, $noreg
|
|
J %bb.2
|
|
|
|
bb.2:
|
|
liveins: $f16d
|
|
VST64 renamable $f16d, %stack.1, 0, $noreg
|
|
Return
|
|
...
|
|
|
|
# In this function the LAY in bb.1 will have a different offset, so the first
|
|
# LAY in bb.2 must remain.
|
|
# CHECK-LABEL: fun2:
|
|
# CHECK: lay %r1, 4096(%r15)
|
|
# CHECK: # %bb.1:
|
|
# CHECK: lay %r1, 8192(%r15)
|
|
# CHECK: .LBB1_2:
|
|
# CHECK: lay %r1, 4096(%r15)
|
|
# CHECK-NOT: lay
|
|
---
|
|
name: fun2
|
|
tracksRegLiveness: true
|
|
stack:
|
|
- { id: 0, size: 5000 }
|
|
- { id: 1, size: 5000 }
|
|
- { id: 2, size: 2500 }
|
|
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0 (%ir-block.0):
|
|
liveins: $f16d
|
|
successors: %bb.2(0x00000001), %bb.1(0x7fffffff)
|
|
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.1, 0, $noreg
|
|
CHIMux undef $r0l, 3, implicit-def $cc
|
|
BRC 14, 8, %bb.2, implicit killed $cc
|
|
J %bb.1
|
|
|
|
bb.1:
|
|
liveins: $f16d
|
|
VST64 renamable $f16d, %stack.2, 0, $noreg
|
|
J %bb.2
|
|
|
|
bb.2:
|
|
liveins: $f16d
|
|
VST64 renamable $f16d, %stack.1, 0, $noreg
|
|
VST64 renamable $f16d, %stack.1, 0, $noreg
|
|
Return
|
|
...
|
|
|
|
# Test case with a loop (with room for improvement: since %r1 is not clobbered
|
|
# inside the loop only the first LAY is needed).
|
|
# CHECK-LABEL: fun3:
|
|
# CHECK: lay %r1, 4096(%r15)
|
|
# CHECK: .LBB2_1:
|
|
# CHECK: lay %r1, 4096(%r15)
|
|
# CHECK: .LBB2_2:
|
|
# CHECK-NOT: lay %r1, 4096(%r15)
|
|
---
|
|
name: fun3
|
|
tracksRegLiveness: true
|
|
stack:
|
|
- { id: 0, size: 5000 }
|
|
- { id: 1, size: 2500 }
|
|
- { id: 2, size: 2500 }
|
|
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0 (%ir-block.0):
|
|
liveins: $f16d
|
|
successors: %bb.2(0x00000001), %bb.1(0x7fffffff)
|
|
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.1, 0, $noreg
|
|
CHIMux undef $r0l, 3, implicit-def $cc
|
|
BRC 14, 8, %bb.2, implicit killed $cc
|
|
J %bb.1
|
|
|
|
bb.1:
|
|
liveins: $f16d
|
|
successors: %bb.2(0x00000001), %bb.1(0x7fffffff)
|
|
|
|
VST64 renamable $f16d, %stack.2, 0, $noreg
|
|
CHIMux undef $r0l, 3, implicit-def $cc
|
|
BRC 14, 8, %bb.1, implicit killed $cc
|
|
J %bb.2
|
|
|
|
bb.2:
|
|
liveins: $f16d
|
|
VST64 renamable $f16d, %stack.1, 0, $noreg
|
|
Return
|
|
...
|
|
|
|
# Test case with a call which clobbers r1: the second LAY after the call is needed.
|
|
# CHECK-LABEL: fun4:
|
|
# CHECK: lay %r1, 4096(%r15)
|
|
# CHECK: brasl
|
|
# CHECK: lay %r1, 4096(%r15)
|
|
---
|
|
name: fun4
|
|
tracksRegLiveness: true
|
|
frameInfo:
|
|
adjustsStack: true
|
|
stack:
|
|
- { id: 0, size: 5000 }
|
|
- { id: 1, size: 2500 }
|
|
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0 (%ir-block.0):
|
|
liveins: $f16d
|
|
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.1, 0, $noreg
|
|
CallBRASL @foo, csr_systemz_elf, implicit-def dead $r14d, implicit-def dead $cc, implicit $fpc, implicit-def $r2l
|
|
$f17d = IMPLICIT_DEF
|
|
VST64 renamable $f17d, %stack.1, 0, $noreg
|
|
Return
|
|
...
|
|
|
|
# Test case where index reg is loaded instead of using an LAY. Only one LGHI is needed.
|
|
# CHECK-LABEL: fun5:
|
|
# CHECK: lghi %r1, 4096
|
|
# CHECK-NOT: lghi
|
|
---
|
|
name: fun5
|
|
tracksRegLiveness: true
|
|
stack:
|
|
- { id: 0, size: 5000 }
|
|
- { id: 1, size: 2500 }
|
|
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0 (%ir-block.0):
|
|
liveins: $f16d
|
|
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
VST64 renamable $f16d, %stack.0, 0, $noreg
|
|
$f0q = nofpexcept LXEB %stack.1, 0, $noreg, implicit $fpc
|
|
$f1q = nofpexcept LXEB %stack.1, 0, $noreg, implicit $fpc
|
|
Return
|
|
...
|
|
|
|
# Test where the constant is a Global. Only one LARL is needed.
|
|
# CHECK-LABEL: fun6:
|
|
# CHECK: larl %r1, ptr
|
|
# CHECK-NOT: larl
|
|
---
|
|
name: fun6
|
|
alignment: 16
|
|
tracksRegLiveness: true
|
|
tracksDebugUserValues: true
|
|
frameInfo:
|
|
maxAlignment: 1
|
|
maxCallFrameSize: 0
|
|
fixedStack:
|
|
- { id: 0, offset: -160, size: 8, alignment: 8 }
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
successors: %bb.2(0x30000000), %bb.1(0x50000000)
|
|
|
|
renamable $r1d = LARL @ptr
|
|
CGHSI killed renamable $r1d, 0, 0, implicit-def $cc :: (volatile dereferenceable load (s64) from @ptr)
|
|
BRC 14, 8, %bb.2, implicit killed $cc
|
|
J %bb.1
|
|
|
|
bb.1:
|
|
renamable $r1d = LARL @ptr
|
|
MVGHI killed renamable $r1d, 0, 0
|
|
|
|
bb.2:
|
|
Return
|
|
|
|
...
|
|
|
|
# Load of an invariant location (GOT). Only one LGRL is needed.
|
|
# CHECK-LABEL: fun7:
|
|
# CHECK: lgrl %r1, ptr
|
|
# CHECK-NOT: lgrl
|
|
---
|
|
name: fun7
|
|
alignment: 16
|
|
tracksRegLiveness: true
|
|
tracksDebugUserValues: true
|
|
frameInfo:
|
|
maxAlignment: 1
|
|
maxCallFrameSize: 0
|
|
fixedStack:
|
|
- { id: 0, offset: -160, size: 8, alignment: 8 }
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
successors: %bb.2(0x30000000), %bb.1(0x50000000)
|
|
|
|
renamable $r1d = LGRL @ptr :: (load (s64) from got)
|
|
CGHSI killed renamable $r1d, 0, 0, implicit-def $cc :: (volatile dereferenceable load (s64) from @ptr)
|
|
BRC 14, 8, %bb.2, implicit killed $cc
|
|
J %bb.1
|
|
|
|
bb.1:
|
|
renamable $r1d = LGRL @ptr :: (load (s64) from got)
|
|
MVGHI killed renamable $r1d, 0, 0
|
|
|
|
bb.2:
|
|
Return
|
|
|
|
...
|
|
|
|
# Load from constant pool. Only one LARL is needed.
|
|
# CHECK-LABEL: fun8:
|
|
# CHECK: larl %r1, .LCPI7_0
|
|
# CHECK-NOT: larl
|
|
---
|
|
name: fun8
|
|
alignment: 16
|
|
tracksRegLiveness: true
|
|
tracksDebugUserValues: true
|
|
liveins:
|
|
- { reg: '$f0s' }
|
|
frameInfo:
|
|
maxAlignment: 1
|
|
maxCallFrameSize: 0
|
|
fixedStack:
|
|
- { id: 0, offset: -160, size: 8, alignment: 8 }
|
|
constants:
|
|
- id: 0
|
|
value: float 0x43E0000000000000
|
|
alignment: 4
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0 (%ir-block.0):
|
|
successors: %bb.1, %bb.2
|
|
liveins: $f0s
|
|
|
|
renamable $r1d = LARL %const.0
|
|
renamable $f1s = LE killed renamable $r1d, 0, $noreg :: (load (s32) from constant-pool)
|
|
nofpexcept CEBR renamable $f0s, renamable $f1s, implicit-def $cc, implicit $fpc
|
|
BRC 15, 11, %bb.2, implicit killed $cc
|
|
|
|
bb.1:
|
|
liveins: $f0s
|
|
|
|
J %bb.3
|
|
|
|
bb.2 (%ir-block.0):
|
|
liveins: $f0s, $f1s
|
|
|
|
renamable $r1d = LARL %const.0
|
|
renamable $f1s = LE killed renamable $r1d, 0, $noreg :: (load (s32) from constant-pool)
|
|
|
|
bb.3 (%ir-block.0):
|
|
liveins: $r2d
|
|
|
|
Return
|
|
|
|
...
|