maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-smax.mir

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

558 lines
30 KiB
Plaintext
Raw Normal View History

AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
AMDGPU/GlobalISel: Use more accurate legality rules for merge/unmerge Most notably, we were incorrectly reporting <3 x s16> as a legal type for these. Make sure these aren't legal to help make progress on fixing the artifact combiner and vector legalizer rules. Unfortunately, this means spreading the -global-isel-abort=0 hack, although this doesn't change the legalizer result in any situation.
2020-08-23 06:00:08 +08:00
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=SI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=VI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer -global-isel-abort=0 %s -o - | FileCheck -check-prefix=GFX9 %s
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
---
name: test_smax_s32
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; SI-LABEL: name: test_smax_s32
; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[COPY]], [[COPY1]]
; SI: $vgpr0 = COPY [[SMAX]](s32)
; VI-LABEL: name: test_smax_s32
; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; VI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[COPY]], [[COPY1]]
; VI: $vgpr0 = COPY [[SMAX]](s32)
; GFX9-LABEL: name: test_smax_s32
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[COPY]], [[COPY1]]
; GFX9: $vgpr0 = COPY [[SMAX]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
%2:_(s32) = G_SMAX %0, %1
$vgpr0 = COPY %2
...
AMDGPU/GlobalISel: add legalize tests for s64 max/min
2020-02-25 07:21:00 +08:00
---
name: test_smax_s64
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_smax_s64
; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
AMDGPU/GlobalISel: Legalize s64 min/max by lowering Reviewers: arsenm, rampitec Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, rovka, dstuttard, tpr, t-tye, hiraditya, kerbowa, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D75108
2020-02-25 18:31:37 +08:00
; SI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(sgt), [[COPY]](s64), [[COPY1]]
; SI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
; SI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
AMDGPU/GlobalISel: add legalize tests for s64 max/min
2020-02-25 07:21:00 +08:00
; VI-LABEL: name: test_smax_s64
; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
AMDGPU/GlobalISel: Legalize s64 min/max by lowering Reviewers: arsenm, rampitec Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, rovka, dstuttard, tpr, t-tye, hiraditya, kerbowa, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D75108
2020-02-25 18:31:37 +08:00
; VI: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(sgt), [[COPY]](s64), [[COPY1]]
; VI: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
; VI: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
AMDGPU/GlobalISel: add legalize tests for s64 max/min
2020-02-25 07:21:00 +08:00
; GFX9-LABEL: name: test_smax_s64
; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(s64) = COPY $vgpr2_vgpr3
AMDGPU/GlobalISel: Legalize s64 min/max by lowering Reviewers: arsenm, rampitec Subscribers: kzhuravl, jvesely, wdng, nhaehnle, yaxunl, rovka, dstuttard, tpr, t-tye, hiraditya, kerbowa, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D75108
2020-02-25 18:31:37 +08:00
; GFX9: [[ICMP:%[0-9]+]]:_(s1) = G_ICMP intpred(sgt), [[COPY]](s64), [[COPY1]]
; GFX9: [[SELECT:%[0-9]+]]:_(s64) = G_SELECT [[ICMP]](s1), [[COPY]], [[COPY1]]
; GFX9: $vgpr0_vgpr1 = COPY [[SELECT]](s64)
AMDGPU/GlobalISel: add legalize tests for s64 max/min
2020-02-25 07:21:00 +08:00
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s64) = COPY $vgpr2_vgpr3
%2:_(s64) = G_SMAX %0, %1
$vgpr0_vgpr1 = COPY %2
...
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
---
name: test_smax_s16
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; SI-LABEL: name: test_smax_s16
; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 16
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 16
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]]
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SMAX]](s32)
; SI: $vgpr0 = COPY [[COPY4]](s32)
; VI-LABEL: name: test_smax_s16
; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC]], [[TRUNC1]]
; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX]](s16)
; VI: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-LABEL: name: test_smax_s16
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; GFX9: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC]], [[TRUNC1]]
; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX]](s16)
; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
%2:_(s16) = G_TRUNC %0
%3:_(s16) = G_TRUNC %1
%4:_(s16) = G_SMAX %2, %3
%5:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
...
---
name: test_smax_s8
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; SI-LABEL: name: test_smax_s8
; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 8
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 8
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]]
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SMAX]](s32)
; SI: $vgpr0 = COPY [[COPY4]](s32)
; VI-LABEL: name: test_smax_s8
; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
GlobalISel: Implement moreElementsVector for G_UNMERGE_VALUES sources This is necessary for handling <3 x s16> on AMDGPU, assuming this should be handled as 2 separate legalization actions. The alternative would be for fewerElementsVector to handle 3->2. llvm-svn: 369547
2019-08-22 00:59:10 +08:00
; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
AMDGPU/GlobalISel: Make 16-bit constants legal This is mostly for the benefit of patterns which use 16-bit constants. llvm-svn: 370921
2019-09-05 00:19:45 +08:00
; VI: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
; VI: [[SHL:%[0-9]+]]:_(s16) = G_SHL [[TRUNC]], [[C]](s16)
; VI: [[ASHR:%[0-9]+]]:_(s16) = G_ASHR [[SHL]], [[C]](s16)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s32)
; VI: [[SHL1:%[0-9]+]]:_(s16) = G_SHL [[TRUNC1]], [[C]](s16)
; VI: [[ASHR1:%[0-9]+]]:_(s16) = G_ASHR [[SHL1]], [[C]](s16)
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[ASHR]], [[ASHR1]]
; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX]](s16)
; VI: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-LABEL: name: test_smax_s8
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
; GFX9: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 8
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG]](s32)
; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
; GFX9: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 8
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[SEXT_INREG1]](s32)
; GFX9: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC]], [[TRUNC1]]
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX]](s16)
; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
%2:_(s8) = G_TRUNC %0
%3:_(s8) = G_TRUNC %1
%4:_(s8) = G_SMAX %2, %3
%5:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
...
---
name: test_smax_s17
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; SI-LABEL: name: test_smax_s17
; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 17
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 17
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]]
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SMAX]](s32)
; SI: $vgpr0 = COPY [[COPY4]](s32)
; VI-LABEL: name: test_smax_s17
; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; VI: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 17
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; VI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; VI: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 17
; VI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]]
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; VI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SMAX]](s32)
; VI: $vgpr0 = COPY [[COPY4]](s32)
; GFX9-LABEL: name: test_smax_s17
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY $vgpr1
; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[COPY]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; GFX9: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 17
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; GFX9: [[COPY3:%[0-9]+]]:_(s32) = COPY [[COPY1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; GFX9: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 17
; GFX9: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]]
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; GFX9: [[COPY4:%[0-9]+]]:_(s32) = COPY [[SMAX]](s32)
; GFX9: $vgpr0 = COPY [[COPY4]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s32) = COPY $vgpr1
%2:_(s17) = G_TRUNC %0
%3:_(s17) = G_TRUNC %1
%4:_(s17) = G_SMAX %2, %3
%5:_(s32) = G_ANYEXT %4
$vgpr0 = COPY %5
...
---
name: test_smax_v2s32
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_smax_v2s32
; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
; SI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[UV]], [[UV2]]
; SI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[UV1]], [[UV3]]
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SMAX]](s32), [[SMAX1]](s32)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
; VI-LABEL: name: test_smax_v2s32
; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
; VI: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
; VI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[UV]], [[UV2]]
; VI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[UV1]], [[UV3]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SMAX]](s32), [[SMAX1]](s32)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
; GFX9-LABEL: name: test_smax_v2s32
; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr2_vgpr3
; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
; GFX9: [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<2 x s32>)
; GFX9: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[UV]], [[UV2]]
; GFX9: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[UV1]], [[UV3]]
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[SMAX]](s32), [[SMAX1]](s32)
; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = COPY $vgpr2_vgpr3
%2:_(<2 x s32>) = G_SMAX %0, %1
$vgpr0_vgpr1 = COPY %2
...
---
name: test_smax_v3s32
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2, $vgpr3_vgpr4_vgpr5
; SI-LABEL: name: test_smax_v3s32
; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
; SI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
; SI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[UV]], [[UV3]]
; SI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[UV1]], [[UV4]]
; SI: [[SMAX2:%[0-9]+]]:_(s32) = G_SMAX [[UV2]], [[UV5]]
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SMAX]](s32), [[SMAX1]](s32), [[SMAX2]](s32)
; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
; VI-LABEL: name: test_smax_v3s32
; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
; VI: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
; VI: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
; VI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[UV]], [[UV3]]
; VI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[UV1]], [[UV4]]
; VI: [[SMAX2:%[0-9]+]]:_(s32) = G_SMAX [[UV2]], [[UV5]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SMAX]](s32), [[SMAX1]](s32), [[SMAX2]](s32)
; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
; GFX9-LABEL: name: test_smax_v3s32
; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
; GFX9: [[COPY1:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
; GFX9: [[UV3:%[0-9]+]]:_(s32), [[UV4:%[0-9]+]]:_(s32), [[UV5:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY1]](<3 x s32>)
; GFX9: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[UV]], [[UV3]]
; GFX9: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[UV1]], [[UV4]]
; GFX9: [[SMAX2:%[0-9]+]]:_(s32) = G_SMAX [[UV2]], [[UV5]]
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[SMAX]](s32), [[SMAX1]](s32), [[SMAX2]](s32)
; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
%0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
%1:_(<3 x s32>) = COPY $vgpr3_vgpr4_vgpr5
%2:_(<3 x s32>) = G_SMAX %0, %1
$vgpr0_vgpr1_vgpr2 = COPY %2
...
---
name: test_smax_v2s16
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; SI-LABEL: name: test_smax_v2s16
; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; SI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
GlobalISel: Lower scalarizing unmerge of a vector to shifts AMDGPU sometimes has legal s16 and <2 x s16> operations, but all registers are really 32-bit. An unmerge destination really should ben widened to a 32-bit register. If widening a scalarizing vector with a target size that matches the vector size, bitcast to integer and extract the relevant bits with shifts. I'm not sure if this is the right place for this. This could arguably be part of widenScalar for the result. I also have a growing feeling that we're missing a bitcast legalize action. llvm-svn: 367604
2019-08-02 03:10:05 +08:00
; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 16
GlobalISel: Lower scalarizing unmerge of a vector to shifts AMDGPU sometimes has legal s16 and <2 x s16> operations, but all registers are really 32-bit. An unmerge destination really should ben widened to a 32-bit register. If widening a scalarizing vector with a target size that matches the vector size, bitcast to integer and extract the relevant bits with shifts. I'm not sure if this is the right place for this. This could arguably be part of widenScalar for the result. I also have a growing feeling that we're missing a bitcast legalize action. llvm-svn: 367604
2019-08-02 03:10:05 +08:00
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 16
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]]
GlobalISel: Lower scalarizing unmerge of a vector to shifts AMDGPU sometimes has legal s16 and <2 x s16> operations, but all registers are really 32-bit. An unmerge destination really should ben widened to a 32-bit register. If widening a scalarizing vector with a target size that matches the vector size, bitcast to integer and extract the relevant bits with shifts. I'm not sure if this is the right place for this. This could arguably be part of widenScalar for the result. I also have a growing feeling that we're missing a bitcast legalize action. llvm-svn: 367604
2019-08-02 03:10:05 +08:00
; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 16
GlobalISel: Lower scalarizing unmerge of a vector to shifts AMDGPU sometimes has legal s16 and <2 x s16> operations, but all registers are really 32-bit. An unmerge destination really should ben widened to a 32-bit register. If widening a scalarizing vector with a target size that matches the vector size, bitcast to integer and extract the relevant bits with shifts. I'm not sure if this is the right place for this. This could arguably be part of widenScalar for the result. I also have a growing feeling that we're missing a bitcast legalize action. llvm-svn: 367604
2019-08-02 03:10:05 +08:00
; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY5]], 16
; SI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG2]], [[SEXT_INREG3]]
AMDGPU/GlobalISel: Don't use legal v2s16 G_BUILD_VECTOR If we have s_pack_* instructions, legalize this to G_BUILD_VECTOR_TRUNC from s32 elements. This is closer to how how the s_pack_* instructions really behave. If we don't have s_pack_ instructions, expand this by creating a merge to s32 and bitcasting. This expands to the expected bit operations. I think this eventually should go in a new bitcast legalize action type in LegalizerHelper. We already directly emit the shift operations in RegBankSelect for the vector case. This could possibly be cleaned up, but I also may want to defer doing this expansion to selection anyway. I'll see about that when I try to actually match VOP3P instructions. This breaks the selection of the build_vector since tablegen doesn't know how to match G_BUILD_VECTOR_TRUNC yet, so just xfail it for now.
2020-01-05 04:35:26 +08:00
; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[SMAX]](s32)
; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY6]], [[C1]]
; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[SMAX1]](s32)
; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY7]], [[C1]]
; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
; SI: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; SI: $vgpr0 = COPY [[BITCAST2]](<2 x s16>)
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; VI-LABEL: name: test_smax_v2s16
; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; VI: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
GlobalISel: Lower scalarizing unmerge of a vector to shifts AMDGPU sometimes has legal s16 and <2 x s16> operations, but all registers are really 32-bit. An unmerge destination really should ben widened to a 32-bit register. If widening a scalarizing vector with a target size that matches the vector size, bitcast to integer and extract the relevant bits with shifts. I'm not sure if this is the right place for this. This could arguably be part of widenScalar for the result. I also have a growing feeling that we're missing a bitcast legalize action. llvm-svn: 367604
2019-08-02 03:10:05 +08:00
; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[COPY1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC]], [[TRUNC2]]
; VI: [[SMAX1:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC1]], [[TRUNC3]]
AMDGPU/GlobalISel: Don't use legal v2s16 G_BUILD_VECTOR If we have s_pack_* instructions, legalize this to G_BUILD_VECTOR_TRUNC from s32 elements. This is closer to how how the s_pack_* instructions really behave. If we don't have s_pack_ instructions, expand this by creating a merge to s32 and bitcasting. This expands to the expected bit operations. I think this eventually should go in a new bitcast legalize action type in LegalizerHelper. We already directly emit the shift operations in RegBankSelect for the vector case. This could possibly be cleaned up, but I also may want to defer doing this expansion to selection anyway. I'll see about that when I try to actually match VOP3P instructions. This breaks the selection of the build_vector since tablegen doesn't know how to match G_BUILD_VECTOR_TRUNC yet, so just xfail it for now.
2020-01-05 04:35:26 +08:00
; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[SMAX]](s16)
; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[SMAX1]](s16)
; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; VI: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; VI: $vgpr0 = COPY [[BITCAST2]](<2 x s16>)
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; GFX9-LABEL: name: test_smax_v2s16
; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
; GFX9: [[COPY1:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr1
; GFX9: [[SMAX:%[0-9]+]]:_(<2 x s16>) = G_SMAX [[COPY]], [[COPY1]]
; GFX9: $vgpr0 = COPY [[SMAX]](<2 x s16>)
%0:_(<2 x s16>) = COPY $vgpr0
%1:_(<2 x s16>) = COPY $vgpr1
%2:_(<2 x s16>) = G_SMAX %0, %1
$vgpr0 = COPY %2
...
---
name: test_smax_v3s16
body: |
bb.0:
liveins: $vgpr0, $vgpr1
; SI-LABEL: name: test_smax_v3s16
; SI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; SI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
GlobalISel: Reimplement moreElementsVectorDst Use pad with undef and unmerge with unused results. This is annoyingly similar to several other places in LegalizerHelper, but they're all slightly different.
2020-02-03 06:42:02 +08:00
; SI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF]](<4 x s16>)
; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF2]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[COPY:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY]], 16
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY1]], 16
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]]
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 16
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 16
; SI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG2]], [[SEXT_INREG3]]
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 16
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY5]], 16
; SI: [[SMAX2:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG4]], [[SEXT_INREG5]]
GlobalISel: Use unmerge when copying wide vectors to result registers Avoid using G_EXTRACT and move towards a more consistent vector legalization strategy.
2020-07-28 22:15:42 +08:00
; SI: [[DEF3:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
; SI: [[DEF4:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
AMDGPU/GlobalISel: Don't use legal v2s16 G_BUILD_VECTOR If we have s_pack_* instructions, legalize this to G_BUILD_VECTOR_TRUNC from s32 elements. This is closer to how how the s_pack_* instructions really behave. If we don't have s_pack_ instructions, expand this by creating a merge to s32 and bitcasting. This expands to the expected bit operations. I think this eventually should go in a new bitcast legalize action type in LegalizerHelper. We already directly emit the shift operations in RegBankSelect for the vector case. This could possibly be cleaned up, but I also may want to defer doing this expansion to selection anyway. I'll see about that when I try to actually match VOP3P instructions. This breaks the selection of the build_vector since tablegen doesn't know how to match G_BUILD_VECTOR_TRUNC yet, so just xfail it for now.
2020-01-05 04:35:26 +08:00
; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[SMAX]](s32)
; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[SMAX1]](s32)
; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[SMAX2]](s32)
GlobalISel: Use unmerge when copying wide vectors to result registers Avoid using G_EXTRACT and move towards a more consistent vector legalization strategy.
2020-07-28 22:15:42 +08:00
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY6]](s32), [[COPY7]](s32), [[COPY8]](s32)
AMDGPU/GlobalISel: Fix using unlegalizable values in tests Implicit uses of non-register value types places impossible to satisfy constraints on the legalizer / artifact combiner. These prevent writing sensible legalize rules for the artifacts without triggering infinite loops in the legalizer. The verifier really needs to enforce this, but I'm not sure what the exact conditions would look like yet.
2020-08-23 05:24:47 +08:00
; SI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s32>)
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; VI-LABEL: name: test_smax_v3s16
; VI: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; VI: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
GlobalISel: Reimplement moreElementsVectorDst Use pad with undef and unmerge with unused results. This is annoyingly similar to several other places in LegalizerHelper, but they're all slightly different.
2020-02-03 06:42:02 +08:00
; VI: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF]](<4 x s16>)
; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[DEF2]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
[GlobalISel] Legalizer: Retry combining illegal artifacts as long as there new artifacts Summary: Currently, Legalizer aborts if it’s unable to legalize artifacts. However, it’s possible to combine them after processing the rest of the instruction because the legalization is likely to generate more artifacts that allow ArtifactCombiner to combine away them. Instead, move illegal artifacts to another list called RetryList and wait until all of the instruction in InstList are legalized. After that, check if there is any new artifacts and try to combine them again if that’s the case. If not, abort. The idea is similar to D59339, but the approach is a bit different. This patch fixes the issue described above, but the legalizer still may be unable to handle some cases depending on when to legalize artifacts. So, in the long run, we probably need a different legalization strategy that handles this dependency in a better way. Reviewers: dsanders, aditya_nandakumar, qcolombet, arsenm, aemerson, paquette Reviewed By: dsanders Subscribers: jvesely, wdng, nhaehnle, rovka, javed.absar, hiraditya, Petar.Avramovic, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65894 llvm-svn: 369805
2019-08-24 04:30:35 +08:00
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC]], [[TRUNC3]]
; VI: [[SMAX1:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC1]], [[TRUNC4]]
; VI: [[SMAX2:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC2]], [[TRUNC5]]
GlobalISel: Use unmerge when copying wide vectors to result registers Avoid using G_EXTRACT and move towards a more consistent vector legalization strategy.
2020-07-28 22:15:42 +08:00
; VI: [[DEF3:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
; VI: [[DEF4:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX]](s16)
; VI: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX1]](s16)
; VI: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[SMAX2]](s16)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[ANYEXT]](s32), [[ANYEXT1]](s32), [[ANYEXT2]](s32)
AMDGPU/GlobalISel: Fix using unlegalizable values in tests Implicit uses of non-register value types places impossible to satisfy constraints on the legalizer / artifact combiner. These prevent writing sensible legalize rules for the artifacts without triggering infinite loops in the legalizer. The verifier really needs to enforce this, but I'm not sure what the exact conditions would look like yet.
2020-08-23 05:24:47 +08:00
; VI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s32>)
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; GFX9-LABEL: name: test_smax_v3s16
; GFX9: [[DEF:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; GFX9: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
GlobalISel: Reimplement moreElementsVectorDst Use pad with undef and unmerge with unused results. This is annoyingly similar to several other places in LegalizerHelper, but they're all slightly different.
2020-02-03 06:42:02 +08:00
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
; GFX9: [[UV:%[0-9]+]]:_(<3 x s16>), [[UV1:%[0-9]+]]:_(<3 x s16>), [[UV2:%[0-9]+]]:_(<3 x s16>), [[UV3:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS]](<12 x s16>)
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; GFX9: [[DEF2:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
GlobalISel: Reimplement moreElementsVectorDst Use pad with undef and unmerge with unused results. This is annoyingly similar to several other places in LegalizerHelper, but they're all slightly different.
2020-02-03 06:42:02 +08:00
; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF2]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
; GFX9: [[UV4:%[0-9]+]]:_(<3 x s16>), [[UV5:%[0-9]+]]:_(<3 x s16>), [[UV6:%[0-9]+]]:_(<3 x s16>), [[UV7:%[0-9]+]]:_(<3 x s16>) = G_UNMERGE_VALUES [[CONCAT_VECTORS1]](<12 x s16>)
; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
; GFX9: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV4]](<3 x s16>), 0
; GFX9: [[UV8:%[0-9]+]]:_(<2 x s16>), [[UV9:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>)
; GFX9: [[UV10:%[0-9]+]]:_(<2 x s16>), [[UV11:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>)
; GFX9: [[SMAX:%[0-9]+]]:_(<2 x s16>) = G_SMAX [[UV8]], [[UV10]]
; GFX9: [[SMAX1:%[0-9]+]]:_(<2 x s16>) = G_SMAX [[UV9]], [[UV11]]
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[SMAX]](<2 x s16>)
AMDGPU/GlobalISel: Fix using unlegalizable values in tests Implicit uses of non-register value types places impossible to satisfy constraints on the legalizer / artifact combiner. These prevent writing sensible legalize rules for the artifacts without triggering infinite loops in the legalizer. The verifier really needs to enforce this, but I'm not sure what the exact conditions would look like yet.
2020-08-23 05:24:47 +08:00
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
GlobalISel: Artifact combine unmerge of unmerge Unmerges have the same fundamental problem as G_TRUNC, and G_TRUNC could be implemented in terms of G_UNMERGE_VALUES. Reducing the number of elements in unmerge results ends up producing the original unmerge type profile, so the artifact combiner needs to eliminate the intermediate illegal registers. This avoids infinite looping in the legalizer in a future change. Assuming an unmerge has each result unmerged the same way, this ends up producing a new unmerge of the source for every definition. I'm not sure if the artifact combiner should either insert temporary merges here and erase the original merge, or if the combiner should look at uses from defs rather than defs from uses for unmerges. In a few cases this regresses from using 16-bit shifts for 8-bit values to using 32-bit shifts, but I think these can be legalized later (the other legalization rules don't try very hard to use 16-bit shifts either).
2020-08-20 20:55:08 +08:00
; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[SMAX1]](<2 x s16>)
AMDGPU/GlobalISel: Fix using unlegalizable values in tests Implicit uses of non-register value types places impossible to satisfy constraints on the legalizer / artifact combiner. These prevent writing sensible legalize rules for the artifacts without triggering infinite loops in the legalizer. The verifier really needs to enforce this, but I'm not sure what the exact conditions would look like yet.
2020-08-23 05:24:47 +08:00
; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
; GFX9: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s32>)
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
%0:_(<3 x s16>) = G_IMPLICIT_DEF
%1:_(<3 x s16>) = G_IMPLICIT_DEF
%2:_(<3 x s16>) = G_SMAX %0, %1
AMDGPU/GlobalISel: Fix using unlegalizable values in tests Implicit uses of non-register value types places impossible to satisfy constraints on the legalizer / artifact combiner. These prevent writing sensible legalize rules for the artifacts without triggering infinite loops in the legalizer. The verifier really needs to enforce this, but I'm not sure what the exact conditions would look like yet.
2020-08-23 05:24:47 +08:00
%3:_(<3 x s32>) = G_ANYEXT %2
S_NOP 0, implicit %3
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
...
---
name: test_smax_v4s16
body: |
bb.0:
liveins: $vgpr0_vgpr1, $vgpr2_vgpr3
; SI-LABEL: name: test_smax_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; SI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; SI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; SI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; SI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY2]], 16
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; SI: [[COPY3:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG1:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY3]], 16
; SI: [[SMAX:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]]
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; SI: [[COPY4:%[0-9]+]]:_(s32) = COPY [[LSHR]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG2:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY4]], 16
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; SI: [[COPY5:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG3:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY5]], 16
; SI: [[SMAX1:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG2]], [[SEXT_INREG3]]
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; SI: [[COPY6:%[0-9]+]]:_(s32) = COPY [[BITCAST1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG4:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY6]], 16
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; SI: [[COPY7:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG5:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY7]], 16
; SI: [[SMAX2:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG4]], [[SEXT_INREG5]]
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; SI: [[COPY8:%[0-9]+]]:_(s32) = COPY [[LSHR1]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG6:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY8]], 16
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; SI: [[COPY9:%[0-9]+]]:_(s32) = COPY [[LSHR3]](s32)
AMDGPU/GlobalISel: Legalize G_SEXT_INREG Split the VALU 64-bit case in RegBankSelect.
2020-02-05 05:06:34 +08:00
; SI: [[SEXT_INREG7:%[0-9]+]]:_(s32) = G_SEXT_INREG [[COPY9]], 16
; SI: [[SMAX3:%[0-9]+]]:_(s32) = G_SMAX [[SEXT_INREG6]], [[SEXT_INREG7]]
AMDGPU/GlobalISel: Don't use legal v2s16 G_BUILD_VECTOR If we have s_pack_* instructions, legalize this to G_BUILD_VECTOR_TRUNC from s32 elements. This is closer to how how the s_pack_* instructions really behave. If we don't have s_pack_ instructions, expand this by creating a merge to s32 and bitcasting. This expands to the expected bit operations. I think this eventually should go in a new bitcast legalize action type in LegalizerHelper. We already directly emit the shift operations in RegBankSelect for the vector case. This could possibly be cleaned up, but I also may want to defer doing this expansion to selection anyway. I'll see about that when I try to actually match VOP3P instructions. This breaks the selection of the build_vector since tablegen doesn't know how to match G_BUILD_VECTOR_TRUNC yet, so just xfail it for now.
2020-01-05 04:35:26 +08:00
; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 65535
; SI: [[COPY10:%[0-9]+]]:_(s32) = COPY [[SMAX]](s32)
; SI: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY10]], [[C1]]
; SI: [[COPY11:%[0-9]+]]:_(s32) = COPY [[SMAX1]](s32)
; SI: [[AND1:%[0-9]+]]:_(s32) = G_AND [[COPY11]], [[C1]]
; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[AND1]], [[C]](s32)
; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[AND]], [[SHL]]
; SI: [[BITCAST4:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; SI: [[COPY12:%[0-9]+]]:_(s32) = COPY [[SMAX2]](s32)
; SI: [[AND2:%[0-9]+]]:_(s32) = G_AND [[COPY12]], [[C1]]
; SI: [[COPY13:%[0-9]+]]:_(s32) = COPY [[SMAX3]](s32)
; SI: [[AND3:%[0-9]+]]:_(s32) = G_AND [[COPY13]], [[C1]]
; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[AND3]], [[C]](s32)
; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[AND2]], [[SHL1]]
; SI: [[BITCAST5:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST4]](<2 x s16>), [[BITCAST5]](<2 x s16>)
GlobalISel: Implement fewerElementsVector for G_BUILD_VECTOR Turn it into a G_CONCAT_VECTORS of G_BUILD_VECTOR. llvm-svn: 374252
2019-10-10 06:44:43 +08:00
; SI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; VI-LABEL: name: test_smax_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; VI: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
GlobalISel: Partially implement fewerElementsVector G_UNMERGE_VALUES Odd sized vectors aren't handled yet. llvm-svn: 368713
2019-08-14 00:26:28 +08:00
; VI: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; VI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; VI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; VI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; VI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; VI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; VI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; VI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; VI: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; VI: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV2]](<2 x s16>)
; VI: [[TRUNC4:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST2]](s32)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; VI: [[TRUNC5:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR2]](s32)
; VI: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV3]](<2 x s16>)
; VI: [[TRUNC6:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST3]](s32)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; VI: [[TRUNC7:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR3]](s32)
; VI: [[SMAX:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC]], [[TRUNC4]]
; VI: [[SMAX1:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC1]], [[TRUNC5]]
; VI: [[SMAX2:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC2]], [[TRUNC6]]
; VI: [[SMAX3:%[0-9]+]]:_(s16) = G_SMAX [[TRUNC3]], [[TRUNC7]]
AMDGPU/GlobalISel: Don't use legal v2s16 G_BUILD_VECTOR If we have s_pack_* instructions, legalize this to G_BUILD_VECTOR_TRUNC from s32 elements. This is closer to how how the s_pack_* instructions really behave. If we don't have s_pack_ instructions, expand this by creating a merge to s32 and bitcasting. This expands to the expected bit operations. I think this eventually should go in a new bitcast legalize action type in LegalizerHelper. We already directly emit the shift operations in RegBankSelect for the vector case. This could possibly be cleaned up, but I also may want to defer doing this expansion to selection anyway. I'll see about that when I try to actually match VOP3P instructions. This breaks the selection of the build_vector since tablegen doesn't know how to match G_BUILD_VECTOR_TRUNC yet, so just xfail it for now.
2020-01-05 04:35:26 +08:00
; VI: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[SMAX]](s16)
; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[SMAX1]](s16)
; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; VI: [[BITCAST4:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; VI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[SMAX2]](s16)
; VI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[SMAX3]](s16)
; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
; VI: [[BITCAST5:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST4]](<2 x s16>), [[BITCAST5]](<2 x s16>)
GlobalISel: Implement fewerElementsVector for G_BUILD_VECTOR Turn it into a G_CONCAT_VECTORS of G_BUILD_VECTOR. llvm-svn: 374252
2019-10-10 06:44:43 +08:00
; VI: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
AMDGPU/GlobalISel: Legality for integer min/max llvm-svn: 361519
2019-05-24 01:58:48 +08:00
; GFX9-LABEL: name: test_smax_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
; GFX9: [[COPY1:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr2_vgpr3
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[UV2:%[0-9]+]]:_(<2 x s16>), [[UV3:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY1]](<4 x s16>)
; GFX9: [[SMAX:%[0-9]+]]:_(<2 x s16>) = G_SMAX [[UV]], [[UV2]]
; GFX9: [[SMAX1:%[0-9]+]]:_(<2 x s16>) = G_SMAX [[UV1]], [[UV3]]
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[SMAX]](<2 x s16>), [[SMAX1]](<2 x s16>)
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = COPY $vgpr2_vgpr3
%2:_(<4 x s16>) = G_SMAX %0, %1
$vgpr0_vgpr1 = COPY %2
...