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-fsqrt.mir

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

479 lines
26 KiB
Plaintext
Raw Normal View History

AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
AMDGPU/GlobalISel: Remove -global-isel-abort=0 from some tests
2020-03-16 04:58:26 +08:00
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer %s -o - | FileCheck -check-prefix=SI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -run-pass=legalizer %s -o - | FileCheck -check-prefix=VI %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=gfx900 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX9 %s
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
---
name: test_fsqrt_s32
body: |
bb.0:
liveins: $vgpr0
; SI-LABEL: name: test_fsqrt_s32
; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
AMDGPU/GlobalISel: Fix broken tests llvm-svn: 353559
2019-02-09 03:59:39 +08:00
; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[COPY]]
; SI: $vgpr0 = COPY [[FSQRT]](s32)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; VI-LABEL: name: test_fsqrt_s32
; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
AMDGPU/GlobalISel: Fix broken tests llvm-svn: 353559
2019-02-09 03:59:39 +08:00
; VI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[COPY]]
; VI: $vgpr0 = COPY [[FSQRT]](s32)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; GFX9-LABEL: name: test_fsqrt_s32
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
AMDGPU/GlobalISel: Fix broken tests llvm-svn: 353559
2019-02-09 03:59:39 +08:00
; GFX9: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[COPY]]
; GFX9: $vgpr0 = COPY [[FSQRT]](s32)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
%0:_(s32) = COPY $vgpr0
%1:_(s32) = G_FSQRT %0
AMDGPU/GlobalISel: Fix broken tests llvm-svn: 353559
2019-02-09 03:59:39 +08:00
$vgpr0 = COPY %1
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
...
AMDGPU/GlobalISel: Fix broken tests llvm-svn: 353559
2019-02-09 03:59:39 +08:00
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
---
name: test_fsqrt_s64
body: |
bb.0:
liveins: $vgpr0
; SI-LABEL: name: test_fsqrt_s64
; SI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
AMDGPU/GlobalISel: Fix broken tests llvm-svn: 353559
2019-02-09 03:59:39 +08:00
; SI: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[COPY]]
; SI: $vgpr0_vgpr1 = COPY [[FSQRT]](s64)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; VI-LABEL: name: test_fsqrt_s64
; VI: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
AMDGPU/GlobalISel: Fix broken tests llvm-svn: 353559
2019-02-09 03:59:39 +08:00
; VI: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[COPY]]
; VI: $vgpr0_vgpr1 = COPY [[FSQRT]](s64)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; GFX9-LABEL: name: test_fsqrt_s64
; GFX9: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
AMDGPU/GlobalISel: Fix broken tests llvm-svn: 353559
2019-02-09 03:59:39 +08:00
; GFX9: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[COPY]]
; GFX9: $vgpr0_vgpr1 = COPY [[FSQRT]](s64)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
%0:_(s64) = COPY $vgpr0_vgpr1
%1:_(s64) = G_FSQRT %0
AMDGPU/GlobalISel: Fix broken tests llvm-svn: 353559
2019-02-09 03:59:39 +08:00
$vgpr0_vgpr1 = COPY %1
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
...
---
name: test_fsqrt_s16
body: |
bb.0:
liveins: $vgpr0
; SI-LABEL: name: test_fsqrt_s16
; SI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; SI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32)
; SI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FPTRUNC]](s16)
; SI: $vgpr0 = COPY [[ANYEXT]](s32)
; VI-LABEL: name: test_fsqrt_s16
; VI: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; VI: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; VI: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT]](s16)
; VI: $vgpr0 = COPY [[ANYEXT]](s32)
; GFX9-LABEL: name: test_fsqrt_s16
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT]](s16)
; GFX9: $vgpr0 = COPY [[ANYEXT]](s32)
%0:_(s32) = COPY $vgpr0
%1:_(s16) = G_TRUNC %0
%2:_(s16) = G_FSQRT %1
%3:_(s32) = G_ANYEXT %2
$vgpr0 = COPY %3
...
---
name: test_fsqrt_v2s32
body: |
bb.0:
liveins: $vgpr0_vgpr1
; SI-LABEL: name: test_fsqrt_v2s32
; SI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32)
; SI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
; VI-LABEL: name: test_fsqrt_v2s32
; VI: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
; VI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
; VI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32)
; VI: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
; GFX9-LABEL: name: test_fsqrt_v2s32
; GFX9: [[COPY:%[0-9]+]]:_(<2 x s32>) = COPY $vgpr0_vgpr1
; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<2 x s32>)
; GFX9: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
; GFX9: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32)
; GFX9: $vgpr0_vgpr1 = COPY [[BUILD_VECTOR]](<2 x s32>)
%0:_(<2 x s32>) = COPY $vgpr0_vgpr1
%1:_(<2 x s32>) = G_FSQRT %0
$vgpr0_vgpr1 = COPY %1
...
---
name: test_fsqrt_v3s32
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2
; SI-LABEL: name: test_fsqrt_v3s32
; SI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
; SI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
; SI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[UV2]]
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32), [[FSQRT2]](s32)
; SI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
; VI-LABEL: name: test_fsqrt_v3s32
; VI: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
; VI: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
; VI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
; VI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
; VI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[UV2]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32), [[FSQRT2]](s32)
; VI: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
; GFX9-LABEL: name: test_fsqrt_v3s32
; GFX9: [[COPY:%[0-9]+]]:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
; GFX9: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<3 x s32>)
; GFX9: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[UV]]
; GFX9: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[UV1]]
; GFX9: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[UV2]]
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[FSQRT]](s32), [[FSQRT1]](s32), [[FSQRT2]](s32)
; GFX9: $vgpr0_vgpr1_vgpr2 = COPY [[BUILD_VECTOR]](<3 x s32>)
%0:_(<3 x s32>) = COPY $vgpr0_vgpr1_vgpr2
%1:_(<3 x s32>) = G_FSQRT %0
$vgpr0_vgpr1_vgpr2 = COPY %1
...
---
name: test_fsqrt_v2s64
body: |
bb.0:
liveins: $vgpr0_vgpr1_vgpr2_vgpr3
; SI-LABEL: name: test_fsqrt_v2s64
; SI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; SI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; SI: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[UV]]
; SI: [[FSQRT1:%[0-9]+]]:_(s64) = G_FSQRT [[UV1]]
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FSQRT]](s64), [[FSQRT1]](s64)
; SI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
; VI-LABEL: name: test_fsqrt_v2s64
; VI: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; VI: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; VI: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[UV]]
; VI: [[FSQRT1:%[0-9]+]]:_(s64) = G_FSQRT [[UV1]]
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FSQRT]](s64), [[FSQRT1]](s64)
; VI: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
; GFX9-LABEL: name: test_fsqrt_v2s64
; GFX9: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
; GFX9: [[UV:%[0-9]+]]:_(s64), [[UV1:%[0-9]+]]:_(s64) = G_UNMERGE_VALUES [[COPY]](<2 x s64>)
; GFX9: [[FSQRT:%[0-9]+]]:_(s64) = G_FSQRT [[UV]]
; GFX9: [[FSQRT1:%[0-9]+]]:_(s64) = G_FSQRT [[UV1]]
; GFX9: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[FSQRT]](s64), [[FSQRT1]](s64)
; GFX9: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[BUILD_VECTOR]](<2 x s64>)
%0:_(<2 x s64>) = COPY $vgpr0_vgpr1_vgpr2_vgpr3
%1:_(<2 x s64>) = G_FSQRT %0
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...
---
name: test_fsqrt_v2s16
body: |
bb.0:
liveins: $vgpr0
; SI-LABEL: name: test_fsqrt_v2s16
; SI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
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: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32)
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: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT1]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT1]](s32)
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: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; SI: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; SI: $vgpr0 = COPY [[BITCAST1]](<2 x s16>)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; VI-LABEL: name: test_fsqrt_v2s16
; VI: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
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: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
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 [[FSQRT]](s16)
; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FSQRT1]](s16)
; VI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; VI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; VI: [[BITCAST1:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; VI: $vgpr0 = COPY [[BITCAST1]](<2 x s16>)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; GFX9-LABEL: name: test_fsqrt_v2s16
; GFX9: [[COPY:%[0-9]+]]:_(<2 x s16>) = COPY $vgpr0
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
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[COPY]](<2 x s16>)
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
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
; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT]](s16)
; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT1]](s16)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[ANYEXT1]](s32)
; GFX9: $vgpr0 = COPY [[BUILD_VECTOR_TRUNC]](<2 x s16>)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
%0:_(<2 x s16>) = COPY $vgpr0
%1:_(<2 x s16>) = G_FSQRT %0
$vgpr0 = COPY %1
...
---
name: test_fsqrt_v3s16
body: |
bb.0:
; SI-LABEL: name: test_fsqrt_v3s16
GlobalISel: Implement moreElementsVector for implicit_def llvm-svn: 353754
2019-02-12 06:00:39 +08:00
; SI: [[DEF:%[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: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
; SI: [[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>)
; SI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
; SI: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>)
; SI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<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: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
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: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<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: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32)
[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: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT1]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT1]](s32)
[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: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; SI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT2]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT2]](s32)
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: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; SI: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC2]](s16)
; SI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C]](s32)
; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
; SI: [[BITCAST3:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
GlobalISel: Merge FewerElements for G_BUILD_VECTOR/G_CONCAT_VECTORS This switches from using G_EXTRACT in odd cases to widen with undef and unmerge.
2020-08-04 06:37:29 +08:00
; SI: [[DEF2:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
; SI: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST2]](<2 x s16>), [[BITCAST3]](<2 x s16>), [[DEF2]](<2 x s16>)
; SI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS1]](<6 x s16>), 0
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: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0
; SI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>)
; SI: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV6]](<2 x s16>)
; SI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
; SI: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>)
; SI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
; SI: [[COPY:%[0-9]+]]:_(s32) = COPY [[BITCAST4]](s32)
; SI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; SI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST5]](s32)
; SI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
; SI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s32>)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; VI-LABEL: name: test_fsqrt_v3s16
GlobalISel: Implement moreElementsVector for implicit_def llvm-svn: 353754
2019-02-12 06:00:39 +08:00
; VI: [[DEF:%[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: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<12 x s16>) = G_CONCAT_VECTORS [[DEF]](<4 x s16>), [[DEF1]](<4 x s16>), [[DEF1]](<4 x s16>)
; VI: [[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>)
; VI: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
; VI: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>)
; VI: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<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: 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: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<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)
; VI: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
; VI: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
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 [[FSQRT]](s16)
; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FSQRT1]](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: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FSQRT2]](s16)
; VI: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[C1]], [[C]](s32)
; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
; VI: [[BITCAST3:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
GlobalISel: Merge FewerElements for G_BUILD_VECTOR/G_CONCAT_VECTORS This switches from using G_EXTRACT in odd cases to widen with undef and unmerge.
2020-08-04 06:37:29 +08:00
; VI: [[DEF2:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
; VI: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BITCAST2]](<2 x s16>), [[BITCAST3]](<2 x s16>), [[DEF2]](<2 x s16>)
; VI: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS1]](<6 x s16>), 0
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: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0
; VI: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>)
; VI: [[BITCAST4:%[0-9]+]]:_(s32) = G_BITCAST [[UV6]](<2 x s16>)
; VI: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST4]], [[C]](s32)
; VI: [[BITCAST5:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>)
; VI: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST5]], [[C]](s32)
; VI: [[COPY:%[0-9]+]]:_(s32) = COPY [[BITCAST4]](s32)
; VI: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; VI: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST5]](s32)
; VI: [[BUILD_VECTOR:%[0-9]+]]:_(<3 x s32>) = G_BUILD_VECTOR [[COPY]](s32), [[COPY1]](s32), [[COPY2]](s32)
; VI: S_NOP 0, implicit [[BUILD_VECTOR]](<3 x s32>)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; GFX9-LABEL: name: test_fsqrt_v3s16
GlobalISel: Implement moreElementsVector for implicit_def llvm-svn: 353754
2019-02-12 06:00:39 +08:00
; GFX9: [[DEF:%[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: [[DEF1:%[0-9]+]]:_(<4 x s16>) = G_IMPLICIT_DEF
; 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>)
; GFX9: [[INSERT:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[UV]](<3 x s16>), 0
; GFX9: [[UV4:%[0-9]+]]:_(<2 x s16>), [[UV5:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT]](<4 x s16>)
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV4]](<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
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
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: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV5]](<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
; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
; GFX9: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
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
; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT]](s16)
; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT1]](s16)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[ANYEXT1]](s32)
; GFX9: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT2]](s16)
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: [[DEF2:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT2]](s32), [[DEF2]](s32)
GlobalISel: Merge FewerElements for G_BUILD_VECTOR/G_CONCAT_VECTORS This switches from using G_EXTRACT in odd cases to widen with undef and unmerge.
2020-08-04 06:37:29 +08:00
; GFX9: [[DEF3:%[0-9]+]]:_(<2 x s16>) = G_IMPLICIT_DEF
; GFX9: [[CONCAT_VECTORS1:%[0-9]+]]:_(<6 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<2 x s16>), [[DEF3]](<2 x s16>)
; GFX9: [[EXTRACT:%[0-9]+]]:_(<3 x s16>) = G_EXTRACT [[CONCAT_VECTORS1]](<6 x s16>), 0
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: [[INSERT1:%[0-9]+]]:_(<4 x s16>) = G_INSERT [[DEF1]], [[EXTRACT]](<3 x s16>), 0
; GFX9: [[UV6:%[0-9]+]]:_(<2 x s16>), [[UV7:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[INSERT1]](<4 x s16>)
; GFX9: [[BITCAST2:%[0-9]+]]:_(s32) = G_BITCAST [[UV6]](<2 x s16>)
; GFX9: [[LSHR2:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST2]], [[C]](s32)
; GFX9: [[BITCAST3:%[0-9]+]]:_(s32) = G_BITCAST [[UV7]](<2 x s16>)
; GFX9: [[LSHR3:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST3]], [[C]](s32)
; GFX9: [[COPY:%[0-9]+]]:_(s32) = COPY [[BITCAST2]](s32)
; GFX9: [[COPY1:%[0-9]+]]:_(s32) = COPY [[LSHR2]](s32)
; GFX9: [[COPY2:%[0-9]+]]:_(s32) = COPY [[BITCAST3]](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: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
%0:_(<3 x s16>) = G_IMPLICIT_DEF
%1:_(<3 x s16>) = G_FSQRT %0
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
%2:_(<3 x s32>) = G_ANYEXT %1
S_NOP 0, implicit %2
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
...
---
name: test_fsqrt_v4s16
body: |
bb.0:
liveins: $vgpr0_vgpr1
; SI-LABEL: name: test_fsqrt_v4s16
; SI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
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: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; SI: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; SI: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; SI: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; SI: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; SI: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; SI: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; SI: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; SI: [[FPEXT:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC]](s16)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; SI: [[FSQRT:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT]]
; SI: [[FPTRUNC:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT]](s32)
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: [[FPEXT1:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC1]](s16)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; SI: [[FSQRT1:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT1]]
; SI: [[FPTRUNC1:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT1]](s32)
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: [[FPEXT2:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC2]](s16)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; SI: [[FSQRT2:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT2]]
; SI: [[FPTRUNC2:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT2]](s32)
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: [[FPEXT3:%[0-9]+]]:_(s32) = G_FPEXT [[TRUNC3]](s16)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; SI: [[FSQRT3:%[0-9]+]]:_(s32) = G_FSQRT [[FPEXT3]]
; SI: [[FPTRUNC3:%[0-9]+]]:_(s16) = G_FPTRUNC [[FSQRT3]](s32)
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: [[ZEXT:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC]](s16)
; SI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC1]](s16)
; SI: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[ZEXT1]], [[C]](s32)
; SI: [[OR:%[0-9]+]]:_(s32) = G_OR [[ZEXT]], [[SHL]]
; SI: [[BITCAST2:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR]](s32)
; SI: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC2]](s16)
; SI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[FPTRUNC3]](s16)
; SI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
; SI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
; SI: [[BITCAST3:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; SI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST2]](<2 x s16>), [[BITCAST3]](<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: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; VI-LABEL: name: test_fsqrt_v4s16
; VI: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
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: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; VI: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
; VI: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
; VI: [[FSQRT3:%[0-9]+]]:_(s16) = G_FSQRT [[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 [[FSQRT]](s16)
; VI: [[ZEXT1:%[0-9]+]]:_(s32) = G_ZEXT [[FSQRT1]](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: [[ZEXT2:%[0-9]+]]:_(s32) = G_ZEXT [[FSQRT2]](s16)
; VI: [[ZEXT3:%[0-9]+]]:_(s32) = G_ZEXT [[FSQRT3]](s16)
; VI: [[SHL1:%[0-9]+]]:_(s32) = G_SHL [[ZEXT3]], [[C]](s32)
; VI: [[OR1:%[0-9]+]]:_(s32) = G_OR [[ZEXT2]], [[SHL1]]
; VI: [[BITCAST3:%[0-9]+]]:_(<2 x s16>) = G_BITCAST [[OR1]](s32)
; VI: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BITCAST2]](<2 x s16>), [[BITCAST3]](<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: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
; GFX9-LABEL: name: test_fsqrt_v4s16
; GFX9: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $vgpr0_vgpr1
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
; GFX9: [[UV:%[0-9]+]]:_(<2 x s16>), [[UV1:%[0-9]+]]:_(<2 x s16>) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
; GFX9: [[BITCAST:%[0-9]+]]:_(s32) = G_BITCAST [[UV]](<2 x s16>)
; GFX9: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST]](s32)
; GFX9: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
; GFX9: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST]], [[C]](s32)
; GFX9: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR]](s32)
; GFX9: [[BITCAST1:%[0-9]+]]:_(s32) = G_BITCAST [[UV1]](<2 x s16>)
; GFX9: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[BITCAST1]](s32)
; GFX9: [[LSHR1:%[0-9]+]]:_(s32) = G_LSHR [[BITCAST1]], [[C]](s32)
; GFX9: [[TRUNC3:%[0-9]+]]:_(s16) = G_TRUNC [[LSHR1]](s32)
; GFX9: [[FSQRT:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC]]
; GFX9: [[FSQRT1:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC1]]
; GFX9: [[FSQRT2:%[0-9]+]]:_(s16) = G_FSQRT [[TRUNC2]]
; GFX9: [[FSQRT3:%[0-9]+]]:_(s16) = G_FSQRT [[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
; GFX9: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT]](s16)
; GFX9: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT1]](s16)
; GFX9: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT]](s32), [[ANYEXT1]](s32)
; GFX9: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT2]](s16)
; GFX9: [[ANYEXT3:%[0-9]+]]:_(s32) = G_ANYEXT [[FSQRT3]](s16)
; GFX9: [[BUILD_VECTOR_TRUNC1:%[0-9]+]]:_(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[ANYEXT2]](s32), [[ANYEXT3]](s32)
; GFX9: [[CONCAT_VECTORS:%[0-9]+]]:_(<4 x s16>) = G_CONCAT_VECTORS [[BUILD_VECTOR_TRUNC]](<2 x s16>), [[BUILD_VECTOR_TRUNC1]](<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
; GFX9: $vgpr0_vgpr1 = COPY [[CONCAT_VECTORS]](<4 x s16>)
AMDGPU/GlobalISel: Legalize fsqrt llvm-svn: 353438
2019-02-08 02:14:39 +08:00
%0:_(<4 x s16>) = COPY $vgpr0_vgpr1
%1:_(<4 x s16>) = G_FSQRT %0
$vgpr0_vgpr1 = COPY %1
...