29f88b93fd
Artifact combiner is not able to access individual elements after using LCMTy style merge/unmerge, extract and insert to change vector number of elements (pad with undef or split to sub-vector instructions). Use unmerge to individual elements instead and then merge elements into requested types. Change argument lowering for vectors and moreElementsVector to use buildPadVectorWithUndefElements and buildDeleteTrailingVectorElements. FewerElementsVector had a few helpers that had different behavior, introduce new helper for most of the opcodes. FewerElementsVector helper is more flexible since it can create leftover instruction smaller then requested type (useful in case target wants to avoid pad with undef and use fewer registers). If target does not want leftover of different type it should call more elements first. Some helpers were performing more elements first to have split without leftover. Opcodes that used this helper use clampMaxNumElementsStrict (does more elements first) in LegalizerInfo to avoid test changes. Fixes failures caused by failing to combine artifacts created during more/fewer elements vector. Differential Revision: https://reviews.llvm.org/D114198
201 lines
9.4 KiB
YAML
201 lines
9.4 KiB
YAML
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
|
|
# RUN: llc -march=amdgcn -run-pass=instruction-select -verify-machineinstrs -o - %s | FileCheck %s
|
|
---
|
|
name: extract512
|
|
legalized: true
|
|
regBankSelected: true
|
|
|
|
body: |
|
|
bb.0:
|
|
; CHECK-LABEL: name: extract512
|
|
; CHECK: [[DEF:%[0-9]+]]:sgpr_512 = IMPLICIT_DEF
|
|
; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub0
|
|
; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub1
|
|
; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub2
|
|
; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub3
|
|
; CHECK: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub4
|
|
; CHECK: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub5
|
|
; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub6
|
|
; CHECK: [[COPY7:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub7
|
|
; CHECK: [[COPY8:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub8
|
|
; CHECK: [[COPY9:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub9
|
|
; CHECK: [[COPY10:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub10
|
|
; CHECK: [[COPY11:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub11
|
|
; CHECK: [[COPY12:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub12
|
|
; CHECK: [[COPY13:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub13
|
|
; CHECK: [[COPY14:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub14
|
|
; CHECK: [[COPY15:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub15
|
|
; CHECK: $sgpr0 = COPY [[COPY]]
|
|
; CHECK: $sgpr1 = COPY [[COPY1]]
|
|
; CHECK: $sgpr2 = COPY [[COPY2]]
|
|
; CHECK: $sgpr3 = COPY [[COPY3]]
|
|
; CHECK: $sgpr4 = COPY [[COPY4]]
|
|
; CHECK: $sgpr5 = COPY [[COPY5]]
|
|
; CHECK: $sgpr6 = COPY [[COPY6]]
|
|
; CHECK: $sgpr7 = COPY [[COPY7]]
|
|
; CHECK: $sgpr8 = COPY [[COPY8]]
|
|
; CHECK: $sgpr9 = COPY [[COPY9]]
|
|
; CHECK: $sgpr10 = COPY [[COPY10]]
|
|
; CHECK: $sgpr11 = COPY [[COPY11]]
|
|
; CHECK: $sgpr12 = COPY [[COPY12]]
|
|
; CHECK: $sgpr13 = COPY [[COPY13]]
|
|
; CHECK: $sgpr14 = COPY [[COPY14]]
|
|
; CHECK: $sgpr15 = COPY [[COPY15]]
|
|
; CHECK: SI_RETURN_TO_EPILOG $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $sgpr14, $sgpr15
|
|
%0:sgpr(s512) = G_IMPLICIT_DEF
|
|
%1:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 0
|
|
%2:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 32
|
|
%3:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 64
|
|
%4:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 96
|
|
%5:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 128
|
|
%6:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 160
|
|
%7:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 192
|
|
%8:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 224
|
|
%9:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 256
|
|
%10:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 288
|
|
%11:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 320
|
|
%12:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 352
|
|
%13:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 384
|
|
%14:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 416
|
|
%15:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 448
|
|
%16:sgpr(s32) = G_EXTRACT %0:sgpr(s512), 480
|
|
$sgpr0 = COPY %1:sgpr(s32)
|
|
$sgpr1 = COPY %2:sgpr(s32)
|
|
$sgpr2 = COPY %3:sgpr(s32)
|
|
$sgpr3 = COPY %4:sgpr(s32)
|
|
$sgpr4 = COPY %5:sgpr(s32)
|
|
$sgpr5 = COPY %6:sgpr(s32)
|
|
$sgpr6 = COPY %7:sgpr(s32)
|
|
$sgpr7 = COPY %8:sgpr(s32)
|
|
$sgpr8 = COPY %9:sgpr(s32)
|
|
$sgpr9 = COPY %10:sgpr(s32)
|
|
$sgpr10 = COPY %11:sgpr(s32)
|
|
$sgpr11 = COPY %12:sgpr(s32)
|
|
$sgpr12 = COPY %13:sgpr(s32)
|
|
$sgpr13 = COPY %14:sgpr(s32)
|
|
$sgpr14 = COPY %15:sgpr(s32)
|
|
$sgpr15 = COPY %16:sgpr(s32)
|
|
SI_RETURN_TO_EPILOG $sgpr0, $sgpr1, $sgpr2, $sgpr3, $sgpr4, $sgpr5, $sgpr6, $sgpr7, $sgpr8, $sgpr9, $sgpr10, $sgpr11, $sgpr12, $sgpr13, $sgpr14, $sgpr15
|
|
...
|
|
|
|
---
|
|
name: extract_s_s32_s1024
|
|
legalized: true
|
|
regBankSelected: true
|
|
|
|
body: |
|
|
bb.0:
|
|
; CHECK-LABEL: name: extract_s_s32_s1024
|
|
; CHECK: [[DEF:%[0-9]+]]:sgpr_1024 = IMPLICIT_DEF
|
|
; CHECK: [[COPY:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub0
|
|
; CHECK: [[COPY1:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub1
|
|
; CHECK: [[COPY2:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub2
|
|
; CHECK: [[COPY3:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub3
|
|
; CHECK: [[COPY4:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub4
|
|
; CHECK: [[COPY5:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub5
|
|
; CHECK: [[COPY6:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub6
|
|
; CHECK: [[COPY7:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub7
|
|
; CHECK: [[COPY8:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub8
|
|
; CHECK: [[COPY9:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub9
|
|
; CHECK: [[COPY10:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub10
|
|
; CHECK: [[COPY11:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub11
|
|
; CHECK: [[COPY12:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub12
|
|
; CHECK: [[COPY13:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub13
|
|
; CHECK: [[COPY14:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub14
|
|
; CHECK: [[COPY15:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub15
|
|
; CHECK: [[COPY16:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub16
|
|
; CHECK: [[COPY17:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub17
|
|
; CHECK: [[COPY18:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub18
|
|
; CHECK: [[COPY19:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub19
|
|
; CHECK: [[COPY20:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub20
|
|
; CHECK: [[COPY21:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub21
|
|
; CHECK: [[COPY22:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub22
|
|
; CHECK: [[COPY23:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub23
|
|
; CHECK: [[COPY24:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub24
|
|
; CHECK: [[COPY25:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub25
|
|
; CHECK: [[COPY26:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub26
|
|
; CHECK: [[COPY27:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub27
|
|
; CHECK: [[COPY28:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub28
|
|
; CHECK: [[COPY29:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub29
|
|
; CHECK: [[COPY30:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub30
|
|
; CHECK: [[COPY31:%[0-9]+]]:sreg_32 = COPY [[DEF]].sub31
|
|
; CHECK: S_ENDPGM 0, implicit [[DEF]], implicit [[COPY]], implicit [[COPY1]], implicit [[COPY2]], implicit [[COPY3]], implicit [[COPY4]], implicit [[COPY5]], implicit [[COPY6]], implicit [[COPY7]], implicit [[COPY8]], implicit [[COPY9]], implicit [[COPY10]], implicit [[COPY11]], implicit [[COPY12]], implicit [[COPY13]], implicit [[COPY14]], implicit [[COPY15]], implicit [[COPY16]], implicit [[COPY17]], implicit [[COPY18]], implicit [[COPY19]], implicit [[COPY20]], implicit [[COPY21]], implicit [[COPY22]], implicit [[COPY23]], implicit [[COPY24]], implicit [[COPY25]], implicit [[COPY26]], implicit [[COPY27]], implicit [[COPY28]], implicit [[COPY29]], implicit [[COPY30]], implicit [[COPY31]]
|
|
%0:sgpr(s1024) = G_IMPLICIT_DEF
|
|
%1:sgpr(s32) = G_EXTRACT %0:sgpr, 0
|
|
%2:sgpr(s32) = G_EXTRACT %0:sgpr, 32
|
|
%3:sgpr(s32) = G_EXTRACT %0:sgpr, 64
|
|
%4:sgpr(s32) = G_EXTRACT %0:sgpr, 96
|
|
%5:sgpr(s32) = G_EXTRACT %0:sgpr, 128
|
|
%6:sgpr(s32) = G_EXTRACT %0:sgpr, 160
|
|
%7:sgpr(s32) = G_EXTRACT %0:sgpr, 192
|
|
%8:sgpr(s32) = G_EXTRACT %0:sgpr, 224
|
|
%9:sgpr(s32) = G_EXTRACT %0:sgpr, 256
|
|
%10:sgpr(s32) = G_EXTRACT %0:sgpr, 288
|
|
%11:sgpr(s32) = G_EXTRACT %0:sgpr, 320
|
|
%12:sgpr(s32) = G_EXTRACT %0:sgpr, 352
|
|
%13:sgpr(s32) = G_EXTRACT %0:sgpr, 384
|
|
%14:sgpr(s32) = G_EXTRACT %0:sgpr, 416
|
|
%15:sgpr(s32) = G_EXTRACT %0:sgpr, 448
|
|
%16:sgpr(s32) = G_EXTRACT %0:sgpr, 480
|
|
|
|
%17:sgpr(s32) = G_EXTRACT %0:sgpr, 512
|
|
%18:sgpr(s32) = G_EXTRACT %0:sgpr, 544
|
|
%19:sgpr(s32) = G_EXTRACT %0:sgpr, 576
|
|
%20:sgpr(s32) = G_EXTRACT %0:sgpr, 608
|
|
%21:sgpr(s32) = G_EXTRACT %0:sgpr, 640
|
|
%22:sgpr(s32) = G_EXTRACT %0:sgpr, 672
|
|
%23:sgpr(s32) = G_EXTRACT %0:sgpr, 704
|
|
%24:sgpr(s32) = G_EXTRACT %0:sgpr, 736
|
|
%25:sgpr(s32) = G_EXTRACT %0:sgpr, 768
|
|
%26:sgpr(s32) = G_EXTRACT %0:sgpr, 800
|
|
%27:sgpr(s32) = G_EXTRACT %0:sgpr, 832
|
|
%28:sgpr(s32) = G_EXTRACT %0:sgpr, 864
|
|
%29:sgpr(s32) = G_EXTRACT %0:sgpr, 896
|
|
%30:sgpr(s32) = G_EXTRACT %0:sgpr, 928
|
|
%31:sgpr(s32) = G_EXTRACT %0:sgpr, 960
|
|
%32:sgpr(s32) = G_EXTRACT %0:sgpr, 992
|
|
|
|
S_ENDPGM 0, implicit %0, implicit %1, implicit %2, implicit %3, implicit %4, implicit %5, implicit %6, implicit %7, implicit %8, implicit %9, implicit %10, implicit %11, implicit %12, implicit %13, implicit %14, implicit %15, implicit %16, implicit %17, implicit %18, implicit %19, implicit %20, implicit %21, implicit %22, implicit %23, implicit %24, implicit %25, implicit %26, implicit %27, implicit %28, implicit %29, implicit %30, implicit %31, implicit %32
|
|
...
|
|
|
|
# TODO: Handle offset 32
|
|
---
|
|
name: extract_sgpr_s64_from_s128
|
|
legalized: true
|
|
regBankSelected: true
|
|
|
|
body: |
|
|
bb.0:
|
|
; CHECK-LABEL: name: extract_sgpr_s64_from_s128
|
|
; CHECK: [[DEF:%[0-9]+]]:sgpr_128 = IMPLICIT_DEF
|
|
; CHECK: [[COPY:%[0-9]+]]:sreg_64 = COPY [[DEF]].sub0_sub1
|
|
; CHECK: [[COPY1:%[0-9]+]]:sreg_64 = COPY [[DEF]].sub2_sub3
|
|
; CHECK: S_ENDPGM 0, implicit [[COPY]], implicit [[COPY1]]
|
|
%0:sgpr(s128) = G_IMPLICIT_DEF
|
|
%1:sgpr(s64) = G_EXTRACT %0, 0
|
|
%2:sgpr(s64) = G_EXTRACT %0, 64
|
|
S_ENDPGM 0, implicit %1, implicit %2
|
|
|
|
...
|
|
|
|
---
|
|
name: extract_sgpr_s96_from_s128
|
|
legalized: true
|
|
regBankSelected: true
|
|
|
|
body: |
|
|
bb.0:
|
|
liveins: $sgpr0_sgpr1_sgpr2_sgpr3
|
|
; CHECK-LABEL: name: extract_sgpr_s96_from_s128
|
|
; CHECK: [[COPY:%[0-9]+]]:sgpr_128_with_sub1_sub2_sub3 = COPY $sgpr0_sgpr1_sgpr2_sgpr3
|
|
; CHECK: [[COPY1:%[0-9]+]]:sgpr_128_with_sub0_sub1_sub2 = COPY [[COPY]]
|
|
; CHECK: [[COPY2:%[0-9]+]]:sgpr_96 = COPY [[COPY1]].sub0_sub1_sub2
|
|
; CHECK: [[COPY3:%[0-9]+]]:sgpr_96 = COPY [[COPY]].sub1_sub2_sub3
|
|
; CHECK: S_ENDPGM 0, implicit [[COPY2]], implicit [[COPY3]]
|
|
%0:sgpr(s128) = COPY $sgpr0_sgpr1_sgpr2_sgpr3
|
|
%1:sgpr(s96) = G_EXTRACT %0, 0
|
|
%2:sgpr(s96) = G_EXTRACT %0, 32
|
|
S_ENDPGM 0, implicit %1, implicit %2
|
|
|
|
...
|