635 lines
21 KiB
YAML
635 lines
21 KiB
YAML
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
|
|
# RUN: llc -mtriple=aarch64-unknown-unknown -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s
|
|
|
|
--- |
|
|
define void @ldrxrox_breg_oreg(i64* %addr) { ret void }
|
|
define void @ldrdrox_breg_oreg(i64* %addr) { ret void }
|
|
define void @more_than_one_use(i64* %addr) { ret void }
|
|
define void @ldrxrox_shl(i64* %addr) { ret void }
|
|
define void @ldrdrox_shl(i64* %addr) { ret void }
|
|
define void @ldrxrox_mul_rhs(i64* %addr) { ret void }
|
|
define void @ldrdrox_mul_rhs(i64* %addr) { ret void }
|
|
define void @ldrxrox_mul_lhs(i64* %addr) { ret void }
|
|
define void @ldrdrox_mul_lhs(i64* %addr) { ret void }
|
|
define void @mul_not_pow_2(i64* %addr) { ret void }
|
|
define void @mul_wrong_pow_2(i64* %addr) { ret void }
|
|
define void @more_than_one_use_shl_1(i64* %addr) { ret void }
|
|
define void @more_than_one_use_shl_2(i64* %addr) { ret void }
|
|
define void @more_than_one_use_shl_lsl_fast(i64* %addr) #1 { ret void }
|
|
define void @more_than_one_use_shl_lsl_slow(i64* %addr) { ret void }
|
|
define void @more_than_one_use_shl_minsize(i64* %addr) #0 { ret void }
|
|
define void @ldrwrox(i64* %addr) { ret void }
|
|
define void @ldrsrox(i64* %addr) { ret void }
|
|
define void @ldrhrox(i64* %addr) { ret void }
|
|
define void @ldbbrox(i64* %addr) { ret void }
|
|
define void @ldrqrox(i64* %addr) { ret void }
|
|
attributes #0 = { optsize minsize }
|
|
attributes #1 = { "target-features"="+lsl-fast" }
|
|
...
|
|
|
|
---
|
|
name: ldrxrox_breg_oreg
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1
|
|
|
|
; CHECK-LABEL: name: ldrxrox_breg_oreg
|
|
; CHECK: liveins: $x0, $x1
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[LDRXroX:%[0-9]+]]:gpr64 = LDRXroX [[COPY]], [[COPY1]], 0, 0 :: (load 8 from %ir.addr)
|
|
; CHECK: $x0 = COPY [[LDRXroX]]
|
|
; CHECK: RET_ReallyLR implicit $x0
|
|
%0:gpr(p0) = COPY $x0
|
|
%1:gpr(s64) = COPY $x1
|
|
%2:gpr(p0) = G_PTR_ADD %0, %1
|
|
%4:gpr(s64) = G_LOAD %2(p0) :: (load 8 from %ir.addr)
|
|
$x0 = COPY %4(s64)
|
|
RET_ReallyLR implicit $x0
|
|
...
|
|
|
|
---
|
|
name: ldrdrox_breg_oreg
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $d0, $x1
|
|
; CHECK-LABEL: name: ldrdrox_breg_oreg
|
|
; CHECK: liveins: $d0, $x1
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64sp = COPY $d0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[LDRDroX:%[0-9]+]]:fpr64 = LDRDroX [[COPY]], [[COPY1]], 0, 0 :: (load 8 from %ir.addr)
|
|
; CHECK: $d0 = COPY [[LDRDroX]]
|
|
; CHECK: RET_ReallyLR implicit $d0
|
|
%0:gpr(p0) = COPY $d0
|
|
%1:gpr(s64) = COPY $x1
|
|
%2:gpr(p0) = G_PTR_ADD %0, %1
|
|
%4:fpr(s64) = G_LOAD %2(p0) :: (load 8 from %ir.addr)
|
|
$d0 = COPY %4(s64)
|
|
RET_ReallyLR implicit $d0
|
|
...
|
|
---
|
|
name: more_than_one_use
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1
|
|
; This shouldn't be folded, since we reuse the result of the G_PTR_ADD outside
|
|
; the G_LOAD
|
|
; CHECK-LABEL: name: more_than_one_use
|
|
; CHECK: liveins: $x0, $x1
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[ADDXrr:%[0-9]+]]:gpr64common = ADDXrr [[COPY]], [[COPY1]]
|
|
; CHECK: [[LDRXui:%[0-9]+]]:gpr64 = LDRXui [[ADDXrr]], 0 :: (load 8 from %ir.addr)
|
|
; CHECK: [[COPY2:%[0-9]+]]:gpr64 = COPY [[ADDXrr]]
|
|
; CHECK: [[ADDXrr1:%[0-9]+]]:gpr64 = ADDXrr [[COPY2]], [[LDRXui]]
|
|
; CHECK: $x0 = COPY [[ADDXrr1]]
|
|
; CHECK: RET_ReallyLR implicit $x0
|
|
%0:gpr(p0) = COPY $x0
|
|
%1:gpr(s64) = COPY $x1
|
|
%2:gpr(p0) = G_PTR_ADD %0, %1
|
|
%4:gpr(s64) = G_LOAD %2(p0) :: (load 8 from %ir.addr)
|
|
%5:gpr(s64) = G_PTRTOINT %2
|
|
%6:gpr(s64) = G_ADD %5, %4
|
|
$x0 = COPY %6(s64)
|
|
RET_ReallyLR implicit $x0
|
|
|
|
...
|
|
---
|
|
name: ldrxrox_shl
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1, $x2
|
|
; CHECK-LABEL: name: ldrxrox_shl
|
|
; CHECK: liveins: $x0, $x1, $x2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRXroX:%[0-9]+]]:gpr64 = LDRXroX [[COPY1]], [[COPY]], 0, 1 :: (load 8 from %ir.addr)
|
|
; CHECK: $x2 = COPY [[LDRXroX]]
|
|
; CHECK: RET_ReallyLR implicit $x2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 3
|
|
%2:gpr(s64) = G_SHL %0, %1(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
$x2 = COPY %5(s64)
|
|
RET_ReallyLR implicit $x2
|
|
|
|
...
|
|
---
|
|
name: ldrdrox_shl
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1, $d2
|
|
; CHECK-LABEL: name: ldrdrox_shl
|
|
; CHECK: liveins: $x0, $x1, $d2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRDroX:%[0-9]+]]:fpr64 = LDRDroX [[COPY1]], [[COPY]], 0, 1 :: (load 8 from %ir.addr)
|
|
; CHECK: $d2 = COPY [[LDRDroX]]
|
|
; CHECK: RET_ReallyLR implicit $d2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 3
|
|
%2:gpr(s64) = G_SHL %0, %1(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:fpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
$d2 = COPY %5(s64)
|
|
RET_ReallyLR implicit $d2
|
|
|
|
...
|
|
---
|
|
name: ldrxrox_mul_rhs
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1, $x2
|
|
; CHECK-LABEL: name: ldrxrox_mul_rhs
|
|
; CHECK: liveins: $x0, $x1, $x2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRXroX:%[0-9]+]]:gpr64 = LDRXroX [[COPY1]], [[COPY]], 0, 1 :: (load 8 from %ir.addr)
|
|
; CHECK: $x2 = COPY [[LDRXroX]]
|
|
; CHECK: RET_ReallyLR implicit $x2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 8
|
|
%2:gpr(s64) = G_MUL %0, %1(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
$x2 = COPY %5(s64)
|
|
RET_ReallyLR implicit $x2
|
|
|
|
...
|
|
---
|
|
name: ldrdrox_mul_rhs
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1, $d2
|
|
; CHECK-LABEL: name: ldrdrox_mul_rhs
|
|
; CHECK: liveins: $x0, $x1, $d2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRDroX:%[0-9]+]]:fpr64 = LDRDroX [[COPY1]], [[COPY]], 0, 1 :: (load 8 from %ir.addr)
|
|
; CHECK: $d2 = COPY [[LDRDroX]]
|
|
; CHECK: RET_ReallyLR implicit $d2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 8
|
|
%2:gpr(s64) = G_MUL %0, %1(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:fpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
$d2 = COPY %5(s64)
|
|
RET_ReallyLR implicit $d2
|
|
|
|
...
|
|
---
|
|
name: ldrxrox_mul_lhs
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1, $x2
|
|
; CHECK-LABEL: name: ldrxrox_mul_lhs
|
|
; CHECK: liveins: $x0, $x1, $x2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRXroX:%[0-9]+]]:gpr64 = LDRXroX [[COPY1]], [[COPY]], 0, 1 :: (load 8 from %ir.addr)
|
|
; CHECK: $x2 = COPY [[LDRXroX]]
|
|
; CHECK: RET_ReallyLR implicit $x2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 8
|
|
%2:gpr(s64) = G_MUL %1, %0(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
$x2 = COPY %5(s64)
|
|
RET_ReallyLR implicit $x2
|
|
|
|
...
|
|
---
|
|
name: ldrdrox_mul_lhs
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1, $d2
|
|
; CHECK-LABEL: name: ldrdrox_mul_lhs
|
|
; CHECK: liveins: $x0, $x1, $d2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRDroX:%[0-9]+]]:fpr64 = LDRDroX [[COPY1]], [[COPY]], 0, 1 :: (load 8 from %ir.addr)
|
|
; CHECK: $d2 = COPY [[LDRDroX]]
|
|
; CHECK: RET_ReallyLR implicit $d2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 8
|
|
%2:gpr(s64) = G_MUL %1, %0(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:fpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
$d2 = COPY %5(s64)
|
|
RET_ReallyLR implicit $d2
|
|
|
|
...
|
|
---
|
|
name: mul_not_pow_2
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
; Show that we don't get a shifted load from a mul when we don't have a
|
|
; power of 2. (The bit isn't set on the load.)
|
|
liveins: $x0, $x1, $d2
|
|
; CHECK-LABEL: name: mul_not_pow_2
|
|
; CHECK: liveins: $x0, $x1, $d2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[MOVi32imm:%[0-9]+]]:gpr32 = MOVi32imm 7
|
|
; CHECK: [[SUBREG_TO_REG:%[0-9]+]]:gpr64 = SUBREG_TO_REG 0, [[MOVi32imm]], %subreg.sub_32
|
|
; CHECK: [[MADDXrrr:%[0-9]+]]:gpr64 = MADDXrrr [[SUBREG_TO_REG]], [[COPY]], $xzr
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRDroX:%[0-9]+]]:fpr64 = LDRDroX [[COPY1]], [[MADDXrrr]], 0, 0 :: (load 8 from %ir.addr)
|
|
; CHECK: $d2 = COPY [[LDRDroX]]
|
|
; CHECK: RET_ReallyLR implicit $d2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 7
|
|
%2:gpr(s64) = G_MUL %1, %0(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:fpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
$d2 = COPY %5(s64)
|
|
RET_ReallyLR implicit $d2
|
|
|
|
...
|
|
---
|
|
name: mul_wrong_pow_2
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
; Show that we don't get a shifted load from a mul when we don't have
|
|
; the right power of 2. (The bit isn't set on the load.)
|
|
liveins: $x0, $x1, $d2
|
|
; CHECK-LABEL: name: mul_wrong_pow_2
|
|
; CHECK: liveins: $x0, $x1, $d2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[MOVi32imm:%[0-9]+]]:gpr32 = MOVi32imm 16
|
|
; CHECK: [[SUBREG_TO_REG:%[0-9]+]]:gpr64 = SUBREG_TO_REG 0, [[MOVi32imm]], %subreg.sub_32
|
|
; CHECK: [[MADDXrrr:%[0-9]+]]:gpr64 = MADDXrrr [[SUBREG_TO_REG]], [[COPY]], $xzr
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRDroX:%[0-9]+]]:fpr64 = LDRDroX [[COPY1]], [[MADDXrrr]], 0, 0 :: (load 8 from %ir.addr)
|
|
; CHECK: $d2 = COPY [[LDRDroX]]
|
|
; CHECK: RET_ReallyLR implicit $d2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 16
|
|
%2:gpr(s64) = G_MUL %1, %0(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:fpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
$d2 = COPY %5(s64)
|
|
RET_ReallyLR implicit $d2
|
|
|
|
...
|
|
---
|
|
name: more_than_one_use_shl_1
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
; Show that we can still fall back to the register-register addressing
|
|
; mode when we fail to pull in the shift.
|
|
liveins: $x0, $x1, $x2
|
|
; CHECK-LABEL: name: more_than_one_use_shl_1
|
|
; CHECK: liveins: $x0, $x1, $x2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[UBFMXri:%[0-9]+]]:gpr64common = UBFMXri [[COPY]], 61, 60
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRXroX:%[0-9]+]]:gpr64 = LDRXroX [[COPY1]], [[UBFMXri]], 0, 0 :: (load 8 from %ir.addr)
|
|
; CHECK: [[ADDXri:%[0-9]+]]:gpr64common = ADDXri [[UBFMXri]], 3, 0
|
|
; CHECK: [[ADDXrr:%[0-9]+]]:gpr64 = ADDXrr [[LDRXroX]], [[ADDXri]]
|
|
; CHECK: $x2 = COPY [[ADDXrr]]
|
|
; CHECK: RET_ReallyLR implicit $x2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 3
|
|
%2:gpr(s64) = G_SHL %0, %1(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
%6:gpr(s64) = G_ADD %2, %1
|
|
%7:gpr(s64) = G_ADD %5, %6
|
|
$x2 = COPY %7(s64)
|
|
RET_ReallyLR implicit $x2
|
|
|
|
...
|
|
---
|
|
name: more_than_one_use_shl_2
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
; Show that when the GEP is used outside a memory op, we don't do any
|
|
; folding at all.
|
|
liveins: $x0, $x1, $x2
|
|
; CHECK-LABEL: name: more_than_one_use_shl_2
|
|
; CHECK: liveins: $x0, $x1, $x2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[UBFMXri:%[0-9]+]]:gpr64common = UBFMXri [[COPY]], 61, 60
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[ADDXrr:%[0-9]+]]:gpr64common = ADDXrr [[COPY1]], [[UBFMXri]]
|
|
; CHECK: [[LDRXui:%[0-9]+]]:gpr64 = LDRXui [[ADDXrr]], 0 :: (load 8 from %ir.addr)
|
|
; CHECK: [[ADDXri:%[0-9]+]]:gpr64common = ADDXri [[UBFMXri]], 3, 0
|
|
; CHECK: [[ADDXrr1:%[0-9]+]]:gpr64 = ADDXrr [[LDRXui]], [[ADDXri]]
|
|
; CHECK: [[COPY2:%[0-9]+]]:gpr64 = COPY [[ADDXrr]]
|
|
; CHECK: [[ADDXrr2:%[0-9]+]]:gpr64 = ADDXrr [[COPY2]], [[ADDXrr1]]
|
|
; CHECK: $x2 = COPY [[ADDXrr2]]
|
|
; CHECK: RET_ReallyLR implicit $x2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 3
|
|
%2:gpr(s64) = G_SHL %0, %1(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
%6:gpr(s64) = G_ADD %2, %1
|
|
%7:gpr(s64) = G_ADD %5, %6
|
|
%8:gpr(s64) = G_PTRTOINT %4
|
|
%9:gpr(s64) = G_ADD %8, %7
|
|
$x2 = COPY %9(s64)
|
|
RET_ReallyLR implicit $x2
|
|
|
|
...
|
|
---
|
|
name: more_than_one_use_shl_lsl_fast
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
; Show that when we have a fastpath for shift-left, we perform the folding
|
|
; if it has more than one use.
|
|
liveins: $x0, $x1, $x2
|
|
; CHECK-LABEL: name: more_than_one_use_shl_lsl_fast
|
|
; CHECK: liveins: $x0, $x1, $x2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64sp = COPY $x1
|
|
; CHECK: [[LDRXroX:%[0-9]+]]:gpr64 = LDRXroX [[COPY1]], [[COPY]], 0, 1 :: (load 8 from %ir.addr)
|
|
; CHECK: [[LDRXroX1:%[0-9]+]]:gpr64 = LDRXroX [[COPY1]], [[COPY]], 0, 1 :: (load 8 from %ir.addr)
|
|
; CHECK: [[ADDXrr:%[0-9]+]]:gpr64 = ADDXrr [[LDRXroX]], [[LDRXroX1]]
|
|
; CHECK: $x2 = COPY [[ADDXrr]]
|
|
; CHECK: RET_ReallyLR implicit $x2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 3
|
|
%2:gpr(s64) = G_SHL %0, %1(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
%6:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
%7:gpr(s64) = G_ADD %5, %6
|
|
$x2 = COPY %7(s64)
|
|
RET_ReallyLR implicit $x2
|
|
|
|
...
|
|
---
|
|
name: more_than_one_use_shl_lsl_slow
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
; Show that we don't fold into multiple memory ops when we don't have a
|
|
; fastpath for shift-left.
|
|
liveins: $x0, $x1, $x2
|
|
; CHECK-LABEL: name: more_than_one_use_shl_lsl_slow
|
|
; CHECK: liveins: $x0, $x1, $x2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[ADDXrs:%[0-9]+]]:gpr64common = ADDXrs [[COPY1]], [[COPY]], 3
|
|
; CHECK: [[LDRXui:%[0-9]+]]:gpr64 = LDRXui [[ADDXrs]], 0 :: (load 8 from %ir.addr)
|
|
; CHECK: [[LDRXui1:%[0-9]+]]:gpr64 = LDRXui [[ADDXrs]], 0 :: (load 8 from %ir.addr)
|
|
; CHECK: [[ADDXrr:%[0-9]+]]:gpr64 = ADDXrr [[LDRXui]], [[LDRXui1]]
|
|
; CHECK: $x2 = COPY [[ADDXrr]]
|
|
; CHECK: RET_ReallyLR implicit $x2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 3
|
|
%2:gpr(s64) = G_SHL %0, %1(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
%6:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
%7:gpr(s64) = G_ADD %5, %6
|
|
$x2 = COPY %7(s64)
|
|
RET_ReallyLR implicit $x2
|
|
|
|
...
|
|
---
|
|
name: more_than_one_use_shl_minsize
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
; Show that when we're optimizing for size, we'll do the folding no matter
|
|
; what.
|
|
liveins: $x0, $x1, $x2
|
|
; CHECK-LABEL: name: more_than_one_use_shl_minsize
|
|
; CHECK: liveins: $x0, $x1, $x2
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64 = COPY $x0
|
|
; CHECK: [[UBFMXri:%[0-9]+]]:gpr64common = UBFMXri [[COPY]], 61, 60
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64common = COPY $x1
|
|
; CHECK: [[COPY2:%[0-9]+]]:gpr64 = COPY [[COPY1]]
|
|
; CHECK: [[ADDXrs:%[0-9]+]]:gpr64 = ADDXrs [[COPY2]], [[COPY]], 3
|
|
; CHECK: [[LDRXroX:%[0-9]+]]:gpr64 = LDRXroX [[COPY1]], [[COPY]], 0, 1 :: (load 8 from %ir.addr)
|
|
; CHECK: [[ADDXri:%[0-9]+]]:gpr64common = ADDXri [[UBFMXri]], 3, 0
|
|
; CHECK: [[ADDXrr:%[0-9]+]]:gpr64 = ADDXrr [[LDRXroX]], [[ADDXri]]
|
|
; CHECK: [[ADDXrr1:%[0-9]+]]:gpr64 = ADDXrr [[ADDXrs]], [[ADDXrr]]
|
|
; CHECK: $x2 = COPY [[ADDXrr1]]
|
|
; CHECK: RET_ReallyLR implicit $x2
|
|
%0:gpr(s64) = COPY $x0
|
|
%1:gpr(s64) = G_CONSTANT i64 3
|
|
%2:gpr(s64) = G_SHL %0, %1(s64)
|
|
%3:gpr(p0) = COPY $x1
|
|
%4:gpr(p0) = G_PTR_ADD %3, %2
|
|
%5:gpr(s64) = G_LOAD %4(p0) :: (load 8 from %ir.addr)
|
|
%6:gpr(s64) = G_ADD %2, %1
|
|
%7:gpr(s64) = G_ADD %5, %6
|
|
%8:gpr(s64) = G_PTRTOINT %4
|
|
%9:gpr(s64) = G_ADD %8, %7
|
|
$x2 = COPY %9(s64)
|
|
RET_ReallyLR implicit $x2
|
|
...
|
|
---
|
|
name: ldrwrox
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1
|
|
; CHECK-LABEL: name: ldrwrox
|
|
; CHECK: liveins: $x0, $x1
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[LDRWroX:%[0-9]+]]:gpr32 = LDRWroX [[COPY]], [[COPY1]], 0, 0 :: (load 4 from %ir.addr)
|
|
; CHECK: $w2 = COPY [[LDRWroX]]
|
|
; CHECK: RET_ReallyLR implicit $w2
|
|
%0:gpr(p0) = COPY $x0
|
|
%1:gpr(s64) = COPY $x1
|
|
%2:gpr(p0) = G_PTR_ADD %0, %1
|
|
%4:gpr(s32) = G_LOAD %2(p0) :: (load 4 from %ir.addr)
|
|
$w2 = COPY %4(s32)
|
|
RET_ReallyLR implicit $w2
|
|
...
|
|
---
|
|
name: ldrsrox
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $d0, $x1
|
|
; CHECK-LABEL: name: ldrsrox
|
|
; CHECK: liveins: $d0, $x1
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64sp = COPY $d0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[LDRSroX:%[0-9]+]]:fpr32 = LDRSroX [[COPY]], [[COPY1]], 0, 0 :: (load 4 from %ir.addr)
|
|
; CHECK: $s2 = COPY [[LDRSroX]]
|
|
; CHECK: RET_ReallyLR implicit $h2
|
|
%0:gpr(p0) = COPY $d0
|
|
%1:gpr(s64) = COPY $x1
|
|
%2:gpr(p0) = G_PTR_ADD %0, %1
|
|
%4:fpr(s32) = G_LOAD %2(p0) :: (load 4 from %ir.addr)
|
|
$s2 = COPY %4(s32)
|
|
RET_ReallyLR implicit $h2
|
|
...
|
|
---
|
|
name: ldrhrox
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1
|
|
; CHECK-LABEL: name: ldrhrox
|
|
; CHECK: liveins: $x0, $x1
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[LDRHroX:%[0-9]+]]:fpr16 = LDRHroX [[COPY]], [[COPY1]], 0, 0 :: (load 2 from %ir.addr)
|
|
; CHECK: $h2 = COPY [[LDRHroX]]
|
|
; CHECK: RET_ReallyLR implicit $h2
|
|
%0:gpr(p0) = COPY $x0
|
|
%1:gpr(s64) = COPY $x1
|
|
%2:gpr(p0) = G_PTR_ADD %0, %1
|
|
%4:fpr(s16) = G_LOAD %2(p0) :: (load 2 from %ir.addr)
|
|
$h2 = COPY %4(s16)
|
|
RET_ReallyLR implicit $h2
|
|
...
|
|
---
|
|
name: ldbbrox
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $x0, $x1
|
|
; CHECK-LABEL: name: ldbbrox
|
|
; CHECK: liveins: $x0, $x1
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64sp = COPY $x0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[LDRBBroX:%[0-9]+]]:gpr32 = LDRBBroX [[COPY]], [[COPY1]], 0, 0 :: (load 1 from %ir.addr)
|
|
; CHECK: $w2 = COPY [[LDRBBroX]]
|
|
; CHECK: RET_ReallyLR implicit $w2
|
|
%0:gpr(p0) = COPY $x0
|
|
%1:gpr(s64) = COPY $x1
|
|
%2:gpr(p0) = G_PTR_ADD %0, %1
|
|
%4:gpr(s32) = G_LOAD %2(p0) :: (load 1 from %ir.addr)
|
|
$w2 = COPY %4(s32)
|
|
RET_ReallyLR implicit $w2
|
|
...
|
|
---
|
|
name: ldrqrox
|
|
alignment: 4
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $d0, $x1
|
|
; CHECK-LABEL: name: ldrqrox
|
|
; CHECK: liveins: $d0, $x1
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr64sp = COPY $d0
|
|
; CHECK: [[COPY1:%[0-9]+]]:gpr64 = COPY $x1
|
|
; CHECK: [[LDRQroX:%[0-9]+]]:fpr128 = LDRQroX [[COPY]], [[COPY1]], 0, 0 :: (load 16 from %ir.addr)
|
|
; CHECK: $q0 = COPY [[LDRQroX]]
|
|
; CHECK: RET_ReallyLR implicit $q0
|
|
%0:gpr(p0) = COPY $d0
|
|
%1:gpr(s64) = COPY $x1
|
|
%2:gpr(p0) = G_PTR_ADD %0, %1
|
|
%4:fpr(<2 x s64>) = G_LOAD %2(p0) :: (load 16 from %ir.addr)
|
|
$q0 = COPY %4(<2 x s64>)
|
|
RET_ReallyLR implicit $q0
|