910630c1e4
This saves us some unnecessary copies. If the inputs to a G_SELECT are floating point, we should use fcsel rather than csel. Changes here are... - Teach selectCopy about s1-to-s1 copies across register banks. - AArch64RegisterBankInfo about G_SELECT in general. - Teach the instruction selector about the FCSEL instructions. Also add two tests: - select-select.mir to show that we get the expected FCSEL - regbank-select.mir (unfortunately named) to show the register banks on G_SELECT are properly preserved And update fast-isel-select.ll to show that we do the same thing as other instruction selectors in these cases. llvm-svn: 359940
67 lines
2.1 KiB
YAML
67 lines
2.1 KiB
YAML
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
|
|
# RUN: llc -mtriple=aarch64-apple-darwin -run-pass=instruction-select -verify-machineinstrs %s -o - | FileCheck %s
|
|
|
|
...
|
|
---
|
|
name: select_f32
|
|
alignment: 2
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $s0, $s1, $w0
|
|
|
|
; CHECK-LABEL: name: select_f32
|
|
; CHECK: liveins: $s0, $s1, $w0
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
|
|
; CHECK: [[COPY1:%[0-9]+]]:fpr32 = COPY $s0
|
|
; CHECK: [[COPY2:%[0-9]+]]:fpr32 = COPY $s1
|
|
; CHECK: [[COPY3:%[0-9]+]]:fpr32 = COPY [[COPY]]
|
|
; CHECK: [[COPY4:%[0-9]+]]:gpr32 = COPY [[COPY3]]
|
|
; CHECK: $wzr = ANDSWri [[COPY4]], 0, implicit-def $nzcv
|
|
; CHECK: [[FCSELSrrr:%[0-9]+]]:fpr32 = FCSELSrrr [[COPY1]], [[COPY2]], 1, implicit $nzcv
|
|
; CHECK: $s0 = COPY [[FCSELSrrr]]
|
|
; CHECK: RET_ReallyLR implicit $s0
|
|
%3:gpr(s32) = COPY $w0
|
|
%0:gpr(s1) = G_TRUNC %3(s32)
|
|
%1:fpr(s32) = COPY $s0
|
|
%2:fpr(s32) = COPY $s1
|
|
%5:fpr(s1) = COPY %0(s1)
|
|
%4:fpr(s32) = G_SELECT %5(s1), %1, %2
|
|
$s0 = COPY %4(s32)
|
|
RET_ReallyLR implicit $s0
|
|
|
|
...
|
|
---
|
|
name: select_f64
|
|
alignment: 2
|
|
legalized: true
|
|
regBankSelected: true
|
|
tracksRegLiveness: true
|
|
machineFunctionInfo: {}
|
|
body: |
|
|
bb.0:
|
|
liveins: $d0, $d1, $w0
|
|
|
|
; CHECK-LABEL: name: select_f64
|
|
; CHECK: liveins: $d0, $d1, $w0
|
|
; CHECK: [[COPY:%[0-9]+]]:gpr32 = COPY $w0
|
|
; CHECK: [[COPY1:%[0-9]+]]:fpr64 = COPY $d0
|
|
; CHECK: [[COPY2:%[0-9]+]]:fpr64 = COPY $d1
|
|
; CHECK: [[COPY3:%[0-9]+]]:fpr32 = COPY [[COPY]]
|
|
; CHECK: [[COPY4:%[0-9]+]]:gpr32 = COPY [[COPY3]]
|
|
; CHECK: $wzr = ANDSWri [[COPY4]], 0, implicit-def $nzcv
|
|
; CHECK: [[FCSELDrrr:%[0-9]+]]:fpr64 = FCSELDrrr [[COPY1]], [[COPY2]], 1, implicit $nzcv
|
|
; CHECK: $d0 = COPY [[FCSELDrrr]]
|
|
; CHECK: RET_ReallyLR implicit $d0
|
|
%3:gpr(s32) = COPY $w0
|
|
%0:gpr(s1) = G_TRUNC %3(s32)
|
|
%1:fpr(s64) = COPY $d0
|
|
%2:fpr(s64) = COPY $d1
|
|
%5:fpr(s1) = COPY %0(s1)
|
|
%4:fpr(s64) = G_SELECT %5(s1), %1, %2
|
|
$d0 = COPY %4(s64)
|
|
RET_ReallyLR implicit $d0
|