RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-22 02:33:06 +01:00
Code Issues Projects Releases Wiki Activity

[AArch64][GlobalISel] Legalize ctpop s128

Browse Source 
Differential revision: https://reviews.llvm.org/D106494
This commit is contained in:
Jon Roelofs 2021-07-21 13:04:06 -07:00
parent 8bfb4cf745
commit 18108c50ec
4 changed files with 69 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
lib/CodeGen/GlobalISel/LegalizerHelper.cpp
View File

@ -5622,7 +5622,15 @@ LegalizerHelper::narrowScalarCTPOP(MachineInstr &MI, unsigned TypeIdx,
auto LoCTPOP = MIRBuilder.buildCTPOP(DstTy, UnmergeSrc.getReg(0));
auto HiCTPOP = MIRBuilder.buildCTPOP(DstTy, UnmergeSrc.getReg(1));
MIRBuilder.buildAdd(DstReg, HiCTPOP, LoCTPOP);
LLT CountTy = LLT::scalar(Log2_64_Ceil(SrcTy.getSizeInBits()));
if (CountTy.getSizeInBits() < DstTy.getSizeInBits()) {
LoCTPOP = MIRBuilder.buildTrunc(CountTy, LoCTPOP);
HiCTPOP = MIRBuilder.buildTrunc(CountTy, HiCTPOP);
auto Add = MIRBuilder.buildAdd(CountTy, HiCTPOP, LoCTPOP);
MIRBuilder.buildZExt(DstReg, Add);
} else
MIRBuilder.buildAdd(DstReg, HiCTPOP, LoCTPOP);
MI.eraseFromParent();
return Legalized;

7
lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
View File

@ -764,7 +764,6 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
getActionDefinitionsBuilder({G_SBFX, G_UBFX})
.customFor({{s32, s32}, {s64, s64}});
// TODO: Custom legalization for s128
// TODO: Use generic lowering when custom lowering is not possible.
auto always = [=](const LegalityQuery &Q) { return true; };
getActionDefinitionsBuilder(G_CTPOP)
@ -775,6 +774,7 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
.maxScalarEltSameAsIf(always, 1, 0)
.customFor({{s32, s32},
{s64, s64},
{s128, s128},
{v2s64, v2s64},
{v2s32, v2s32},
{v4s32, v4s32},
@ -1151,8 +1151,7 @@ bool AArch64LegalizerInfo::legalizeCTPOP(MachineInstr &MI,
// v8s16,v4s32,v2s64 -> v16i8
LLT VTy = Size == 128 ? LLT::fixed_vector(16, 8) : LLT::fixed_vector(8, 8);
if (Ty.isScalar()) {
// TODO: Handle s128.
assert((Size == 32 || Size == 64) && "Expected only 32 or 64 bit scalars!");
assert((Size == 32 || Size == 64 || Size == 128) && "Expected only 32, 64, or 128 bit scalars!");
if (Size == 32) {
Val = MIRBuilder.buildZExt(LLT::scalar(64), Val).getReg(0);
}
@ -1198,7 +1197,7 @@ bool AArch64LegalizerInfo::legalizeCTPOP(MachineInstr &MI,
}
// Post-conditioning.
if (Ty.isScalar() && Size == 64)
if (Ty.isScalar() && (Size == 64 || Size == 128))
MIRBuilder.buildZExt(Dst, UADD);
else
UADD->getOperand(0).setReg(Dst);

24
test/CodeGen/AArch64/GlobalISel/legalize-ctpop.mir
View File

@ -78,6 +78,30 @@ body: |
$x0 = COPY %ctpop(s64)
RET_ReallyLR implicit $x0
...
---
name: s128_lower
tracksRegLiveness: true
body: |
bb.0:
liveins: $q0
; CHECK-LABEL: name: s128_lower
; CHECK: liveins: $q0
; CHECK: %copy:_(s128) = COPY $q0
; CHECK: [[BITCAST:%[0-9]+]]:_(<16 x s8>) = G_BITCAST %copy(s128)
; CHECK: [[CTPOP:%[0-9]+]]:_(<16 x s8>) = G_CTPOP [[BITCAST]](<16 x s8>)
; CHECK: [[INT:%[0-9]+]]:_(s32) = G_INTRINSIC intrinsic(@llvm.aarch64.neon.uaddlv), [[CTPOP]](<16 x s8>)
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
; CHECK: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[INT]](s32), [[C]](s32)
; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 0
; CHECK: %ctpop:_(s128) = G_MERGE_VALUES [[MV]](s64), [[C1]](s64)
; CHECK: $q0 = COPY %ctpop(s128)
; CHECK: RET_ReallyLR implicit $q0
%copy:_(s128) = COPY $q0
%ctpop:_(s128) = G_CTPOP %copy(s128)
$q0 = COPY %ctpop(s128)
RET_ReallyLR implicit $q0
...
---
name: widen_s16

58
test/CodeGen/AArch64/popcount.ll
View File

@ -5,15 +5,12 @@
define i8 @popcount128(i128* nocapture nonnull readonly %0) {
; CHECK-LABEL: popcount128:
; CHECK: // %bb.0: // %Entry
; CHECK-NEXT: ldr x8, [x0, #8]
; CHECK-NEXT: ldr d1, [x0]
; CHECK-NEXT: // implicit-def: $q0
; CHECK-NEXT: mov v0.16b, v1.16b
; CHECK-NEXT: mov v0.d[1], x8
; CHECK-NEXT: ldr q0, [x0]
; CHECK-NEXT: cnt v0.16b, v0.16b
; CHECK-NEXT: uaddlv h1, v0.16b
; CHECK-NEXT: // implicit-def: $q0
; CHECK-NEXT: mov v0.16b, v1.16b
; CHECK-NEXT: // kill: def $s0 killed $s0 killed $q0
; CHECK-NEXT: fmov w0, s0
; CHECK-NEXT: ret
Entry:
@ -30,27 +27,35 @@ declare i128 @llvm.ctpop.i128(i128)
define i16 @popcount256(i256* nocapture nonnull readonly %0) {
; CHECK-LABEL: popcount256:
; CHECK: // %bb.0: // %Entry
; CHECK-NEXT: ldr x8, [x0, #8]
; CHECK-NEXT: ldr x9, [x0, #24]
; CHECK-NEXT: ldr d1, [x0, #16]
; CHECK-NEXT: ldr x11, [x0]
; CHECK-NEXT: ldr x10, [x0, #8]
; CHECK-NEXT: ldr x9, [x0, #16]
; CHECK-NEXT: ldr x8, [x0, #24]
; CHECK-NEXT: // implicit-def: $q0
; CHECK-NEXT: mov v0.16b, v1.16b
; CHECK-NEXT: mov v0.d[1], x9
; CHECK-NEXT: mov v0.d[0], x11
; CHECK-NEXT: mov v0.d[1], x10
; CHECK-NEXT: // implicit-def: $q1
; CHECK-NEXT: mov v1.d[0], x9
; CHECK-NEXT: mov v1.d[1], x8
; CHECK-NEXT: cnt v0.16b, v0.16b
; CHECK-NEXT: uaddlv h1, v0.16b
; CHECK-NEXT: uaddlv h2, v0.16b
; CHECK-NEXT: // implicit-def: $q0
; CHECK-NEXT: mov v0.16b, v1.16b
; CHECK-NEXT: mov v0.16b, v2.16b
; CHECK-NEXT: // kill: def $s0 killed $s0 killed $q0
; CHECK-NEXT: cnt v1.16b, v1.16b
; CHECK-NEXT: uaddlv h2, v1.16b
; CHECK-NEXT: // implicit-def: $q1
; CHECK-NEXT: mov v1.16b, v2.16b
; CHECK-NEXT: // kill: def $s1 killed $s1 killed $q1
; CHECK-NEXT: fmov w8, s1
; CHECK-NEXT: fmov w9, s0
; CHECK-NEXT: ldr d1, [x0]
; CHECK-NEXT: // implicit-def: $q0
; CHECK-NEXT: mov v0.16b, v1.16b
; CHECK-NEXT: mov v0.d[1], x8
; CHECK-NEXT: cnt v0.16b, v0.16b
; CHECK-NEXT: uaddlv h1, v0.16b
; CHECK-NEXT: // implicit-def: $q0
; CHECK-NEXT: mov v0.16b, v1.16b
; CHECK-NEXT: fmov w8, s0
; CHECK-NEXT: add w0, w8, w9
; CHECK-NEXT: add w8, w8, w9
; CHECK-NEXT: // implicit-def: $w9
; CHECK-NEXT: // kill: def $x8 killed $w8
; CHECK-NEXT: // kill: def $x9 killed $w9
; CHECK-NEXT: bfi x8, x9, #32, #32
; CHECK-NEXT: and x8, x8, #0xff
; CHECK-NEXT: mov w0, w8
; CHECK-NEXT: ret
Entry:
%1 = load i256, i256* %0, align 16
@ -66,16 +71,19 @@ define <1 x i128> @popcount1x128(<1 x i128> %0) {
; CHECK-LABEL: popcount1x128:
; CHECK: // %bb.0: // %Entry
; CHECK-NEXT: // implicit-def: $q0
; CHECK-NEXT: fmov d0, x0
; CHECK-NEXT: mov v0.d[0], x0
; CHECK-NEXT: mov v0.d[1], x1
; CHECK-NEXT: cnt v0.16b, v0.16b
; CHECK-NEXT: uaddlv h1, v0.16b
; CHECK-NEXT: // implicit-def: $q0
; CHECK-NEXT: mov v0.16b, v1.16b
; CHECK-NEXT: // kill: def $s0 killed $s0 killed $q0
; CHECK-NEXT: fmov w0, s0
; CHECK-NEXT: mov w8, wzr
; CHECK-NEXT: // kill: def $x0 killed $w0
; CHECK-NEXT: movi v0.2d, #0000000000000000
; CHECK-NEXT: mov x1, v0.d[1]
; CHECK-NEXT: // kill: def $x8 killed $w8
; CHECK-NEXT: bfi x0, x8, #32, #32
; CHECK-NEXT: mov x1, xzr
; CHECK-NEXT: ret
Entry:
%1 = tail call <1 x i128> @llvm.ctpop.v1.i128(<1 x i128> %0)