1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-23 11:13:28 +01:00

[ARM] Allow half types in ConstantPool

Change ARMConstantIslandPass to:
- accept f16 literals as litpool entries,
- if the litpool needs to be inserted in the middle of a big block, then we
  need to 4-byte align the next instruction in ARM mode.

Differential Revision: https://reviews.llvm.org/D42784

llvm-svn: 325012
This commit is contained in:
Sjoerd Meijer 2018-02-13 15:34:09 +00:00
parent ec0afdb0d7
commit db5ee9a2c0
3 changed files with 178 additions and 1 deletions

View File

@ -510,7 +510,6 @@ ARMConstantIslands::doInitialConstPlacement(std::vector<MachineInstr*> &CPEMIs)
const DataLayout &TD = MF->getDataLayout();
for (unsigned i = 0, e = CPs.size(); i != e; ++i) {
unsigned Size = TD.getTypeAllocSize(CPs[i].getType());
assert(Size >= 4 && "Too small constant pool entry");
unsigned Align = CPs[i].getAlignment();
assert(isPowerOf2_32(Align) && "Invalid alignment");
// Verify that all constant pool entries are a multiple of their alignment.
@ -820,6 +819,11 @@ initializeFunctionInfo(const std::vector<MachineInstr*> &CPEMIs) {
Scale = 4; // +-(offset_8*4)
NegOk = true;
break;
case ARM::VLDRH:
Bits = 8;
Scale = 2; // +-(offset_8*2)
NegOk = true;
break;
case ARM::tLDRHi:
Bits = 5;
@ -1421,6 +1425,10 @@ void ARMConstantIslands::createNewWater(unsigned CPUserIndex,
assert(!isThumb || getITInstrPredicate(*MI, PredReg) == ARMCC::AL));
NewMBB = splitBlockBeforeInstr(&*MI);
// 4 byte align the next block after the constant pool when the CPE is a
// 16-bit value in ARM mode, and 2 byte for Thumb.
NewMBB->setAlignment(isThumb ? 1 : 2);
}
/// handleConstantPoolUser - Analyze the specified user, checking to see if it

View File

@ -0,0 +1,89 @@
# RUN: llc -mtriple=arm-none-eabi -run-pass=arm-cp-islands %s -o - | FileCheck %s
--- |
; ModuleID = '<stdin>'
source_filename = "<stdin>"
target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
target triple = "arm-arm--eabi"
declare i32 @llvm.arm.space(i32, i32) #0
define dso_local i32 @ARM(i64* %LL, i32 %A.coerce) local_unnamed_addr #1 {
entry:
%S = alloca half, align 2
%tmp.0.extract.trunc = trunc i32 %A.coerce to i16
%0 = bitcast i16 %tmp.0.extract.trunc to half
store volatile half 0xH3C00, half* %S, align 2
store volatile i64 4242424242424242, i64* %LL, align 8
%1 = call i32 @llvm.arm.space(i32 8920, i32 undef)
%S.0.S.0.570 = load volatile half, half* %S, align 2
%add298 = fadd half %S.0.S.0.570, 0xH2E66
store volatile half %add298, half* %S, align 2
%2 = call i32 @llvm.arm.space(i32 1350, i32 undef)
%3 = bitcast half %add298 to i16
%tmp343.0.insert.ext = zext i16 %3 to i32
ret i32 %tmp343.0.insert.ext
}
attributes #0 = { nounwind }
attributes #1 = { minsize nounwind optsize "target-features"="+crc,+crypto,+dsp,+fp-armv8,+fullfp16,+hwdiv,+hwdiv-arm,+neon,+ras,+strict-align,-thumb-mode" }
...
---
name: ARM
alignment: 2
tracksRegLiveness: true
liveins:
- { reg: '$r0' }
frameInfo:
stackSize: 4
maxAlignment: 2
maxCallFrameSize: 0
stack:
- { id: 0, name: S, offset: -2, size: 2, alignment: 2, stack-id: 0, local-offset: -2 }
constants:
- id: 0
value: i32 1576323506
alignment: 4
- id: 1
value: i32 987766
alignment: 4
- id: 2
value: half 0xH2E66
alignment: 2
#CHECK: B %[[BB4:bb.[0-9]]]
#CHECK: bb.{{.}} (align 2):
#CHECK: successors:
#CHECK: CONSTPOOL_ENTRY {{.}}, %const.{{.}}, 4
#CHECK: bb.{{.}} (align 2):
#CHECK: successors:
#CHECK: CONSTPOOL_ENTRY {{.}}, %const.{{.}}, 4
#CHECK: bb.{{.}} (align 1):
#CHECK: successors:
#CHECK: CONSTPOOL_ENTRY {{.}}, %const.{{.}}, 2
#CHECK: [[BB4]].entry (align 2):
body: |
bb.0.entry:
liveins: $r0
$sp = frame-setup SUBri $sp, 4, 14, $noreg, $noreg
frame-setup CFI_INSTRUCTION def_cfa_offset 4
renamable $s0 = FCONSTH 112, 14, $noreg
renamable $r1 = LDRcp %const.0, 0, 14, $noreg :: (load 4 from constant-pool)
renamable $r2 = LDRcp %const.1, 0, 14, $noreg :: (load 4 from constant-pool)
VSTRH killed renamable $s0, $sp, 1, 14, $noreg :: (volatile store 2 into %ir.S)
STRi12 killed renamable $r2, renamable $r0, 4, 14, $noreg :: (volatile store 4 into %ir.LL + 4)
renamable $s0 = VLDRH %const.2, 0, 14, $noreg :: (load 2 from constant-pool)
STRi12 killed renamable $r1, killed renamable $r0, 0, 14, $noreg :: (volatile store 4 into %ir.LL, align 8)
dead renamable $r0 = SPACE 8920, undef renamable $r0
renamable $s2 = VLDRH $sp, 1, 14, $noreg :: (volatile dereferenceable load 2 from %ir.S)
renamable $s0 = VADDH killed renamable $s2, killed renamable $s0, 14, $noreg
VSTRH renamable $s0, $sp, 1, 14, $noreg :: (volatile store 2 into %ir.S)
renamable $r0 = VMOVRH killed renamable $s0, 14, $noreg
dead renamable $r1 = SPACE 1350, undef renamable $r0
$sp = ADDri $sp, 4, 14, $noreg, $noreg
MOVPCLR 14, $noreg, implicit killed $r0
...

View File

@ -0,0 +1,80 @@
# RUN: llc -mtriple=thumbv7-none-eabi -run-pass=arm-cp-islands %s -o - | FileCheck %s
--- |
; ModuleID = '<stdin>'
source_filename = "<stdin>"
target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
target triple = "thumbv7-arm--eabi"
declare i32 @llvm.arm.space(i32, i32) #0
define dso_local i32 @THUMB(i32 %A.coerce) local_unnamed_addr #1 {
entry:
%F = alloca float, align 4
%S = alloca half, align 2
%tmp.0.extract.trunc = trunc i32 %A.coerce to i16
%0 = bitcast i16 %tmp.0.extract.trunc to half
store volatile float 4.200000e+01, float* %F, align 4
store volatile half 0xH3C00, half* %S, align 2
%S.0.S.0.142 = load volatile half, half* %S, align 2
%1 = call i32 @llvm.arm.space(i32 1230, i32 undef)
%add42 = fadd half %S.0.S.0.142, 0xH2E66
store volatile half %add42, half* %S, align 2
%2 = call i32 @llvm.arm.space(i32 1330, i32 undef)
%S.0.S.0.119 = load volatile half, half* %S, align 2
%3 = bitcast half %add42 to i16
%tmp87.0.insert.ext = zext i16 %3 to i32
ret i32 %tmp87.0.insert.ext
}
attributes #0 = { nounwind }
attributes #1 = { minsize nounwind optsize "target-features"="+crc,+crypto,+dsp,+fp-armv8,+fullfp16,+hwdiv,+hwdiv-arm,+neon,+ras,+strict-align,+thumb-mode" }
...
---
name: THUMB
alignment: 1
tracksRegLiveness: true
frameInfo:
stackSize: 8
maxAlignment: 4
maxCallFrameSize: 0
stack:
- { id: 0, name: F, offset: -4, size: 4, alignment: 4, stack-id: 0, local-offset: -4 }
- { id: 1, name: S, offset: -6, size: 2, alignment: 2, stack-id: 0, local-offset: -6 }
constants:
- id: 0
value: i32 1109917696
alignment: 4
- id: 1
value: half 0xH2E66
alignment: 2
#CHECK: t2B %[[BB3:bb.[0-9]]]
#CHECK: bb.{{.}} (align 2):
#CHECK: successors:
#CHECK: CONSTPOOL_ENTRY 2, %const.{{.}}, 4
#CHECK: bb.{{.}} (align 1):
#CHECK: successors:
#CHECK: CONSTPOOL_ENTRY 3, %const.{{.}}, 2
#CHECK: [[BB3]].entry (align 1):
body: |
bb.0.entry:
$sp = frame-setup tSUBspi $sp, 2, 14, $noreg
frame-setup CFI_INSTRUCTION def_cfa_offset 8
renamable $r0 = tLDRpci %const.0, 14, $noreg :: (load 4 from constant-pool)
renamable $s0 = FCONSTH 112, 14, $noreg
tSTRspi killed renamable $r0, $sp, 1, 14, $noreg :: (volatile store 4 into %ir.F)
VSTRH killed renamable $s0, $sp, 1, 14, $noreg :: (volatile store 2 into %ir.S)
renamable $s2 = VLDRH $sp, 1, 14, $noreg :: (volatile dereferenceable load 2 from %ir.S)
renamable $s0 = VLDRH %const.1, 0, 14, $noreg :: (load 2 from constant-pool)
dead renamable $r0 = SPACE 1230, undef renamable $r0
renamable $s0 = VADDH killed renamable $s2, killed renamable $s0, 14, $noreg
VSTRH renamable $s0, $sp, 1, 14, $noreg :: (volatile store 2 into %ir.S)
renamable $r0 = VMOVRH killed renamable $s0, 14, $noreg
dead renamable $r1 = SPACE 1330, undef renamable $r0
dead renamable $s0 = VLDRH $sp, 1, 14, $noreg :: (volatile dereferenceable load 2 from %ir.S)
$sp = tADDspi $sp, 2, 14, $noreg
tBX_RET 14, $noreg, implicit killed $r0
...